[BACK]Return to imach.c CVS log [TXT] [DIR] Up to [Local Repository] / imach / src

Diff for /imach/src/imach.c between versions 1.84 and 1.335

version 1.84, 2003/06/13 21:44:43 version 1.335, 2022/08/31 08:23:16
Line 1 Line 1
 /* $Id$  /* $Id$
   $State$    $State$
   $Log$    $Log$
   Revision 1.84  2003/06/13 21:44:43  brouard    Revision 1.335  2022/08/31 08:23:16  brouard
   * imach.c (Repository): Replace "freqsummary" at a correct    Summary: improvements...
   place. It differs from routine "prevalence" which may be called  
   many times. Probs is memory consuming and must be used with  
   parcimony.  
   Version 0.95a2 (should output exactly the same maximization than 0.8a2)  
   
   Revision 1.83  2003/06/10 13:39:11  lievre    Revision 1.334  2022/08/25 09:08:41  brouard
     Summary: In progress for quantitative
   
     Revision 1.333  2022/08/21 09:10:30  brouard
     * src/imach.c (Module): Version 0.99r33 A lot of changes in
     reassigning covariates: my first idea was that people will always
     use the first covariate V1 into the model but in fact they are
     producing data with many covariates and can use an equation model
     with some of the covariate; it means that in a model V2+V3 instead
     of codtabm(k,Tvaraff[j]) which calculates for combination k, for
     three covariates (V1, V2, V3) the value of Tvaraff[j], but in fact
     the equation model is restricted to two variables only (V2, V3)
     and the combination for V2 should be codtabm(k,1) instead of
     (codtabm(k,2), and the code should be
     codtabm(k,TnsdVar[Tvaraff[j]]. Many many changes have been
     made. All of these should be simplified once a day like we did in
     hpxij() for example by using precov[nres] which is computed in
     decoderesult for each nres of each resultline. Loop should be done
     on the equation model globally by distinguishing only product with
     age (which are changing with age) and no more on type of
     covariates, single dummies, single covariates.
   
     Revision 1.332  2022/08/21 09:06:25  brouard
     Summary: Version 0.99r33
   
     * src/imach.c (Module): Version 0.99r33 A lot of changes in
     reassigning covariates: my first idea was that people will always
     use the first covariate V1 into the model but in fact they are
     producing data with many covariates and can use an equation model
     with some of the covariate; it means that in a model V2+V3 instead
     of codtabm(k,Tvaraff[j]) which calculates for combination k, for
     three covariates (V1, V2, V3) the value of Tvaraff[j], but in fact
     the equation model is restricted to two variables only (V2, V3)
     and the combination for V2 should be codtabm(k,1) instead of
     (codtabm(k,2), and the code should be
     codtabm(k,TnsdVar[Tvaraff[j]]. Many many changes have been
     made. All of these should be simplified once a day like we did in
     hpxij() for example by using precov[nres] which is computed in
     decoderesult for each nres of each resultline. Loop should be done
     on the equation model globally by distinguishing only product with
     age (which are changing with age) and no more on type of
     covariates, single dummies, single covariates.
   
     Revision 1.331  2022/08/07 05:40:09  brouard
   *** empty log message ***    *** empty log message ***
   
   Revision 1.82  2003/06/05 15:57:20  brouard    Revision 1.330  2022/08/06 07:18:25  brouard
   Add log in  imach.c and  fullversion number is now printed.    Summary: last 0.99r31
   
 */    *  imach.c (Module): Version of imach using partly decoderesult to rebuild xpxij function
 /*  
    Interpolated Markov Chain  
   
   Short summary of the programme:    Revision 1.329  2022/08/03 17:29:54  brouard
       *  imach.c (Module): Many errors in graphs fixed with Vn*age covariates.
   This program computes Healthy Life Expectancies from  
   cross-longitudinal data. Cross-longitudinal data consist in: -1- a  
   first survey ("cross") where individuals from different ages are  
   interviewed on their health status or degree of disability (in the  
   case of a health survey which is our main interest) -2- at least a  
   second wave of interviews ("longitudinal") which measure each change  
   (if any) in individual health status.  Health expectancies are  
   computed from the time spent in each health state according to a  
   model. More health states you consider, more time is necessary to reach the  
   Maximum Likelihood of the parameters involved in the model.  The  
   simplest model is the multinomial logistic model where pij is the  
   probability to be observed in state j at the second wave  
   conditional to be observed in state i at the first wave. Therefore  
   the model is: log(pij/pii)= aij + bij*age+ cij*sex + etc , where  
   'age' is age and 'sex' is a covariate. If you want to have a more  
   complex model than "constant and age", you should modify the program  
   where the markup *Covariates have to be included here again* invites  
   you to do it.  More covariates you add, slower the  
   convergence.  
   
   The advantage of this computer programme, compared to a simple    Revision 1.328  2022/07/27 17:40:48  brouard
   multinomial logistic model, is clear when the delay between waves is not    Summary: valgrind bug fixed by initializing to zero DummyV as well as Tage
   identical for each individual. Also, if a individual missed an  
   intermediate interview, the information is lost, but taken into  
   account using an interpolation or extrapolation.    
   
   hPijx is the probability to be observed in state i at age x+h    Revision 1.327  2022/07/27 14:47:35  brouard
   conditional to the observed state i at age x. The delay 'h' can be    Summary: Still a problem for one-step probabilities in case of quantitative variables
   split into an exact number (nh*stepm) of unobserved intermediate  
   states. This elementary transition (by month, quarter,  
   semester or year) is modelled as a multinomial logistic.  The hPx  
   matrix is simply the matrix product of nh*stepm elementary matrices  
   and the contribution of each individual to the likelihood is simply  
   hPijx.  
   
   Also this programme outputs the covariance matrix of the parameters but also    Revision 1.326  2022/07/26 17:33:55  brouard
   of the life expectancies. It also computes the stable prevalence.     Summary: some test with nres=1
     
   Authors: Nicolas Brouard (brouard@ined.fr) and Agnès Lièvre (lievre@ined.fr).  
            Institut national d'études démographiques, Paris.  
   This software have been partly granted by Euro-REVES, a concerted action  
   from the European Union.  
   It is copyrighted identically to a GNU software product, ie programme and  
   software can be distributed freely for non commercial use. Latest version  
   can be accessed at http://euroreves.ined.fr/imach .  
   
   Help to debug: LD_PRELOAD=/usr/local/lib/libnjamd.so ./imach foo.imach    Revision 1.325  2022/07/25 14:27:23  brouard
   or better on gdb : set env LD_PRELOAD=/usr/local/lib/libnjamd.so    Summary: r30
     
   **********************************************************************/  
 /*  
   main  
   read parameterfile  
   read datafile  
   concatwav  
   freqsummary  
   if (mle >= 1)  
     mlikeli  
   print results files  
   if mle==1   
      computes hessian  
   read end of parameter file: agemin, agemax, bage, fage, estepm  
       begin-prev-date,...  
   open gnuplot file  
   open html file  
   stable prevalence  
    for age prevalim()  
   h Pij x  
   variance of p varprob  
   forecasting if prevfcast==1 prevforecast call prevalence()  
   health expectancies  
   Variance-covariance of DFLE  
   prevalence()  
    movingaverage()  
   varevsij()   
   if popbased==1 varevsij(,popbased)  
   total life expectancies  
   Variance of stable prevalence  
  end  
 */  
   
     * imach.c (Module): Error cptcovn instead of nsd in bmij (was
     coredumped, revealed by Feiuno, thank you.
   
     Revision 1.324  2022/07/23 17:44:26  brouard
     *** empty log message ***
   
      Revision 1.323  2022/07/22 12:30:08  brouard
 #include <math.h>    *  imach.c (Module): Output of Wald test in the htm file and not only in the log.
 #include <stdio.h>  
 #include <stdlib.h>  
 #include <unistd.h>  
   
 #define MAXLINE 256    Revision 1.322  2022/07/22 12:27:48  brouard
 #define GNUPLOTPROGRAM "gnuplot"    *  imach.c (Module): Output of Wald test in the htm file and not only in the log.
 /*#define GNUPLOTPROGRAM "..\\gp37mgw\\wgnuplot"*/  
 #define FILENAMELENGTH 80  
 /*#define DEBUG*/  
 #define windows  
 #define GLOCK_ERROR_NOPATH              -1      /* empty path */  
 #define GLOCK_ERROR_GETCWD              -2      /* cannot get cwd */  
   
 #define MAXPARM 30 /* Maximum number of parameters for the optimization */    Revision 1.321  2022/07/22 12:04:24  brouard
 #define NPARMAX 64 /* (nlstate+ndeath-1)*nlstate*ncovmodel */    Summary: r28
   
 #define NINTERVMAX 8    *  imach.c (Module): Output of Wald test in the htm file and not only in the log.
 #define NLSTATEMAX 8 /* Maximum number of live states (for func) */  
 #define NDEATHMAX 8 /* Maximum number of dead states (for func) */  
 #define NCOVMAX 8 /* Maximum number of covariates */  
 #define MAXN 20000  
 #define YEARM 12. /* Number of months per year */  
 #define AGESUP 130  
 #define AGEBASE 40  
 #ifdef windows  
 #define DIRSEPARATOR '\\'  
 #define ODIRSEPARATOR '/'  
 #else  
 #define DIRSEPARATOR '/'  
 #define ODIRSEPARATOR '\\'  
 #endif  
   
 /* $Id$ */    Revision 1.320  2022/06/02 05:10:11  brouard
 /* $State$ */    *** empty log message ***
   
 char version[]="Imach version 0.95a2, June 2003, INED-EUROREVES ";    Revision 1.319  2022/06/02 04:45:11  brouard
 char fullversion[]="$Revision$ $Date$";     * imach.c (Module): Adding the Wald tests from the log to the main
 int erreur; /* Error number */    htm for better display of the maximum likelihood estimators.
 int nvar;  
 int cptcovn=0, cptcovage=0, cptcoveff=0,cptcov;  
 int npar=NPARMAX;  
 int nlstate=2; /* Number of live states */  
 int ndeath=1; /* Number of dead states */  
 int ncovmodel, ncovcol;     /* Total number of covariables including constant a12*1 +b12*x ncovmodel=2 */  
 int popbased=0;  
   
 int *wav; /* Number of waves for this individuual 0 is possible */    Revision 1.318  2022/05/24 08:10:59  brouard
 int maxwav; /* Maxim number of waves */    * imach.c (Module): Some attempts to find a bug of wrong estimates
 int jmin, jmax; /* min, max spacing between 2 waves */    of confidencce intervals with product in the equation modelC
 int mle, weightopt;  
 int **mw; /* mw[mi][i] is number of the mi wave for this individual */  
 int **dh; /* dh[mi][i] is number of steps between mi,mi+1 for this individual */  
 int **bh; /* bh[mi][i] is the bias (+ or -) for this individual if the delay between  
            * wave mi and wave mi+1 is not an exact multiple of stepm. */  
 double jmean; /* Mean space between 2 waves */  
 double **oldm, **newm, **savm; /* Working pointers to matrices */  
 double **oldms, **newms, **savms; /* Fixed working pointers to matrices */  
 FILE *fic,*ficpar, *ficparo,*ficres,  *ficrespl, *ficrespij, *ficrest,*ficresf,*ficrespop;  
 FILE *ficlog, *ficrespow;  
 FILE *ficgp,*ficresprob,*ficpop, *ficresprobcov, *ficresprobcor;  
 FILE *ficresprobmorprev;  
 FILE *fichtm; /* Html File */  
 FILE *ficreseij;  
 char filerese[FILENAMELENGTH];  
 FILE  *ficresvij;  
 char fileresv[FILENAMELENGTH];  
 FILE  *ficresvpl;  
 char fileresvpl[FILENAMELENGTH];  
 char title[MAXLINE];  
 char optionfile[FILENAMELENGTH], datafile[FILENAMELENGTH],  filerespl[FILENAMELENGTH];  
 char optionfilext[10], optionfilefiname[FILENAMELENGTH], plotcmd[FILENAMELENGTH];  
   
 char fileres[FILENAMELENGTH], filerespij[FILENAMELENGTH], filereso[FILENAMELENGTH], rfileres[FILENAMELENGTH];    Revision 1.317  2022/05/15 15:06:23  brouard
 char filelog[FILENAMELENGTH]; /* Log file */    * imach.c (Module):  Some minor improvements
 char filerest[FILENAMELENGTH];  
 char fileregp[FILENAMELENGTH];  
 char popfile[FILENAMELENGTH];  
   
 char optionfilegnuplot[FILENAMELENGTH], optionfilehtm[FILENAMELENGTH];    Revision 1.316  2022/05/11 15:11:31  brouard
     Summary: r27
   
 #define NR_END 1    Revision 1.315  2022/05/11 15:06:32  brouard
 #define FREE_ARG char*    *** empty log message ***
 #define FTOL 1.0e-10  
   
 #define NRANSI     Revision 1.314  2022/04/13 17:43:09  brouard
 #define ITMAX 200     * imach.c (Module): Adding link to text data files
   
 #define TOL 2.0e-4     Revision 1.313  2022/04/11 15:57:42  brouard
     * imach.c (Module): Error in rewriting the 'r' file with yearsfproj or yearsbproj fixed
   
 #define CGOLD 0.3819660     Revision 1.312  2022/04/05 21:24:39  brouard
 #define ZEPS 1.0e-10     *** empty log message ***
 #define SHFT(a,b,c,d) (a)=(b);(b)=(c);(c)=(d);   
   
 #define GOLD 1.618034     Revision 1.311  2022/04/05 21:03:51  brouard
 #define GLIMIT 100.0     Summary: Fixed quantitative covariates
 #define TINY 1.0e-20   
   
 static double maxarg1,maxarg2;            Fixed covariates (dummy or quantitative)
 #define FMAX(a,b) (maxarg1=(a),maxarg2=(b),(maxarg1)>(maxarg2)? (maxarg1):(maxarg2))          with missing values have never been allowed but are ERRORS and
 #define FMIN(a,b) (maxarg1=(a),maxarg2=(b),(maxarg1)<(maxarg2)? (maxarg1):(maxarg2))          program quits. Standard deviations of fixed covariates were
             wrongly computed. Mean and standard deviations of time varying
 #define SIGN(a,b) ((b)>0.0 ? fabs(a) : -fabs(a))          covariates are still not computed.
 #define rint(a) floor(a+0.5)  
   
 static double sqrarg;    Revision 1.310  2022/03/17 08:45:53  brouard
 #define SQR(a) ((sqrarg=(a)) == 0.0 ? 0.0 :sqrarg*sqrarg)    Summary: 99r25
 #define SWAP(a,b) {temp=(a);(a)=(b);(b)=temp;}   
   
 int imx;     Improving detection of errors: result lines should be compatible with
 int stepm;    the model.
 /* Stepm, step in month: minimum step interpolation*/  
   
 int estepm;    Revision 1.309  2021/05/20 12:39:14  brouard
 /* Estepm, step in month to interpolate survival function in order to approximate Life Expectancy*/    Summary: Version 0.99r24
   
 int m,nb;    Revision 1.308  2021/03/31 13:11:57  brouard
 int *num, firstpass=0, lastpass=4,*cod, *ncodemax, *Tage;    Summary: Version 0.99r23
 double **agev,*moisnais, *annais, *moisdc, *andc,**mint, **anint;  
 double **pmmij, ***probs;  
 double dateintmean=0;  
   
 double *weight;  
 int **s; /* Status */  
 double *agedc, **covar, idx;  
 int **nbcode, *Tcode, *Tvar, **codtab, **Tvard, *Tprod, cptcovprod, *Tvaraff;  
   
 double ftol=FTOL; /* Tolerance for computing Max Likelihood */    * imach.c (Module): Still bugs in the result loop. Thank to Holly Benett
 double ftolhess; /* Tolerance for computing hessian */  
   
 /**************** split *************************/    Revision 1.307  2021/03/08 18:11:32  brouard
 static  int split( char *path, char *dirc, char *name, char *ext, char *finame )    Summary: 0.99r22 fixed bug on result:
 {  
   char  *ss;                            /* pointer */  
   int   l1, l2;                         /* length counters */  
   
   l1 = strlen(path );                   /* length of path */    Revision 1.306  2021/02/20 15:44:02  brouard
   if ( l1 == 0 ) return( GLOCK_ERROR_NOPATH );    Summary: Version 0.99r21
   ss= strrchr( path, DIRSEPARATOR );            /* find last / */  
   if ( ss == NULL ) {                   /* no directory, so use current */  
     /*if(strrchr(path, ODIRSEPARATOR )==NULL)  
       printf("Warning you should use %s as a separator\n",DIRSEPARATOR);*/  
     /* get current working directory */  
     /*    extern  char* getcwd ( char *buf , int len);*/  
     if ( getcwd( dirc, FILENAME_MAX ) == NULL ) {  
       return( GLOCK_ERROR_GETCWD );  
     }  
     strcpy( name, path );               /* we've got it */  
   } else {                              /* strip direcotry from path */  
     ss++;                               /* after this, the filename */  
     l2 = strlen( ss );                  /* length of filename */  
     if ( l2 == 0 ) return( GLOCK_ERROR_NOPATH );  
     strcpy( name, ss );         /* save file name */  
     strncpy( dirc, path, l1 - l2 );     /* now the directory */  
     dirc[l1-l2] = 0;                    /* add zero */  
   }  
   l1 = strlen( dirc );                  /* length of directory */  
 #ifdef windows  
   if ( dirc[l1-1] != '\\' ) { dirc[l1] = '\\'; dirc[l1+1] = 0; }  
 #else  
   if ( dirc[l1-1] != '/' ) { dirc[l1] = '/'; dirc[l1+1] = 0; }  
 #endif  
   ss = strrchr( name, '.' );            /* find last / */  
   ss++;  
   strcpy(ext,ss);                       /* save extension */  
   l1= strlen( name);  
   l2= strlen(ss)+1;  
   strncpy( finame, name, l1-l2);  
   finame[l1-l2]= 0;  
   return( 0 );                          /* we're done */  
 }  
   
     * imach.c (Module): Fix bug on quitting after result lines!
     (Module): Version 0.99r21
   
 /******************************************/    Revision 1.305  2021/02/20 15:28:30  brouard
     * imach.c (Module): Fix bug on quitting after result lines!
   
 void replace(char *s, char*t)    Revision 1.304  2021/02/12 11:34:20  brouard
 {    * imach.c (Module): The use of a Windows BOM (huge) file is now an error
   int i;  
   int lg=20;  
   i=0;  
   lg=strlen(t);  
   for(i=0; i<= lg; i++) {  
     (s[i] = t[i]);  
     if (t[i]== '\\') s[i]='/';  
   }  
 }  
   
 int nbocc(char *s, char occ)    Revision 1.303  2021/02/11 19:50:15  brouard
 {    *  (Module): imach.c Someone entered 'results:' instead of 'result:'. Now it is an error which is printed.
   int i,j=0;  
   int lg=20;  
   i=0;  
   lg=strlen(s);  
   for(i=0; i<= lg; i++) {  
   if  (s[i] == occ ) j++;  
   }  
   return j;  
 }  
   
 void cutv(char *u,char *v, char*t, char occ)    Revision 1.302  2020/02/22 21:00:05  brouard
 {    *  (Module): imach.c Update mle=-3 (for computing Life expectancy
   /* cuts string t into u and v where u is ended by char occ excluding it    and life table from the data without any state)
      and v is after occ excluding it too : ex cutv(u,v,"abcdef2ghi2j",2)  
      gives u="abcedf" and v="ghi2j" */  
   int i,lg,j,p=0;  
   i=0;  
   for(j=0; j<=strlen(t)-1; j++) {  
     if((t[j]!= occ) && (t[j+1]== occ)) p=j+1;  
   }  
   
   lg=strlen(t);    Revision 1.301  2019/06/04 13:51:20  brouard
   for(j=0; j<p; j++) {    Summary: Error in 'r'parameter file backcast yearsbproj instead of yearsfproj
     (u[j] = t[j]);  
   }  
      u[p]='\0';  
   
    for(j=0; j<= lg; j++) {    Revision 1.300  2019/05/22 19:09:45  brouard
     if (j>=(p+1))(v[j-p-1] = t[j]);    Summary: version 0.99r19 of May 2019
   }  
 }  
   
 /********************** nrerror ********************/    Revision 1.299  2019/05/22 18:37:08  brouard
     Summary: Cleaned 0.99r19
   
 void nrerror(char error_text[])    Revision 1.298  2019/05/22 18:19:56  brouard
 {    *** empty log message ***
   fprintf(stderr,"ERREUR ...\n");  
   fprintf(stderr,"%s\n",error_text);  
   exit(EXIT_FAILURE);  
 }  
 /*********************** vector *******************/  
 double *vector(int nl, int nh)  
 {  
   double *v;  
   v=(double *) malloc((size_t)((nh-nl+1+NR_END)*sizeof(double)));  
   if (!v) nrerror("allocation failure in vector");  
   return v-nl+NR_END;  
 }  
   
 /************************ free vector ******************/    Revision 1.297  2019/05/22 17:56:10  brouard
 void free_vector(double*v, int nl, int nh)    Summary: Fix bug by moving date2dmy and nhstepm which gaefin=-1
 {  
   free((FREE_ARG)(v+nl-NR_END));  
 }  
   
 /************************ivector *******************************/    Revision 1.296  2019/05/20 13:03:18  brouard
 char *cvector(long nl,long nh)    Summary: Projection syntax simplified
 {  
   char *v;  
   v=(char *) malloc((size_t)((nh-nl+1+NR_END)*sizeof(char)));  
   if (!v) nrerror("allocation failure in cvector");  
   return v-nl+NR_END;  
 }  
   
 /******************free ivector **************************/  
 void free_cvector(char *v, long nl, long nh)  
 {  
   free((FREE_ARG)(v+nl-NR_END));  
 }  
   
 /************************ivector *******************************/    We can now start projections, forward or backward, from the mean date
 int *ivector(long nl,long nh)    of inteviews up to or down to a number of years of projection:
 {    prevforecast=1 yearsfproj=15.3 mobil_average=0
   int *v;    or
   v=(int *) malloc((size_t)((nh-nl+1+NR_END)*sizeof(int)));    prevforecast=1 starting-proj-date=1/1/2007 final-proj-date=12/31/2017 mobil_average=0
   if (!v) nrerror("allocation failure in ivector");    or
   return v-nl+NR_END;    prevbackcast=1 yearsbproj=12.3 mobil_average=1
 }    or
     prevbackcast=1 starting-back-date=1/10/1999 final-back-date=1/1/1985 mobil_average=1
   
 /******************free ivector **************************/    Revision 1.295  2019/05/18 09:52:50  brouard
 void free_ivector(int *v, long nl, long nh)    Summary: doxygen tex bug
 {  
   free((FREE_ARG)(v+nl-NR_END));  
 }  
   
 /******************* imatrix *******************************/    Revision 1.294  2019/05/16 14:54:33  brouard
 int **imatrix(long nrl, long nrh, long ncl, long nch)     Summary: There was some wrong lines added
      /* allocate a int matrix with subscript range m[nrl..nrh][ncl..nch] */   
 {   
   long i, nrow=nrh-nrl+1,ncol=nch-ncl+1;   
   int **m;   
     
   /* allocate pointers to rows */   
   m=(int **) malloc((size_t)((nrow+NR_END)*sizeof(int*)));   
   if (!m) nrerror("allocation failure 1 in matrix()");   
   m += NR_END;   
   m -= nrl;   
     
     
   /* allocate rows and set pointers to them */   
   m[nrl]=(int *) malloc((size_t)((nrow*ncol+NR_END)*sizeof(int)));   
   if (!m[nrl]) nrerror("allocation failure 2 in matrix()");   
   m[nrl] += NR_END;   
   m[nrl] -= ncl;   
     
   for(i=nrl+1;i<=nrh;i++) m[i]=m[i-1]+ncol;   
     
   /* return pointer to array of pointers to rows */   
   return m;   
 }   
   
 /****************** free_imatrix *************************/    Revision 1.293  2019/05/09 15:17:34  brouard
 void free_imatrix(m,nrl,nrh,ncl,nch)    *** empty log message ***
       int **m;  
       long nch,ncl,nrh,nrl;   
      /* free an int matrix allocated by imatrix() */   
 {   
   free((FREE_ARG) (m[nrl]+ncl-NR_END));   
   free((FREE_ARG) (m+nrl-NR_END));   
 }   
   
 /******************* matrix *******************************/    Revision 1.292  2019/05/09 14:17:20  brouard
 double **matrix(long nrl, long nrh, long ncl, long nch)    Summary: Some updates
 {  
   long i, nrow=nrh-nrl+1, ncol=nch-ncl+1;  
   double **m;  
   
   m=(double **) malloc((size_t)((nrow+NR_END)*sizeof(double*)));    Revision 1.291  2019/05/09 13:44:18  brouard
   if (!m) nrerror("allocation failure 1 in matrix()");    Summary: Before ncovmax
   m += NR_END;  
   m -= nrl;  
   
   m[nrl]=(double *) malloc((size_t)((nrow*ncol+NR_END)*sizeof(double)));    Revision 1.290  2019/05/09 13:39:37  brouard
   if (!m[nrl]) nrerror("allocation failure 2 in matrix()");    Summary: 0.99r18 unlimited number of individuals
   m[nrl] += NR_END;  
   m[nrl] -= ncl;  
   
   for (i=nrl+1; i<=nrh; i++) m[i]=m[i-1]+ncol;    The number n which was limited to 20,000 cases is now unlimited, from firstobs to lastobs. If the number is too for the virtual memory, probably an error will occur.
   return m;  
   /* print *(*(m+1)+70) ou print m[1][70]; print m+1 or print &(m[1])   
    */  
 }  
   
 /*************************free matrix ************************/    Revision 1.289  2018/12/13 09:16:26  brouard
 void free_matrix(double **m, long nrl, long nrh, long ncl, long nch)    Summary: Bug for young ages (<-30) will be in r17
 {  
   free((FREE_ARG)(m[nrl]+ncl-NR_END));  
   free((FREE_ARG)(m+nrl-NR_END));  
 }  
   
 /******************* ma3x *******************************/    Revision 1.288  2018/05/02 20:58:27  brouard
 double ***ma3x(long nrl, long nrh, long ncl, long nch, long nll, long nlh)    Summary: Some bugs fixed
 {  
   long i, j, nrow=nrh-nrl+1, ncol=nch-ncl+1, nlay=nlh-nll+1;  
   double ***m;  
   
   m=(double ***) malloc((size_t)((nrow+NR_END)*sizeof(double*)));    Revision 1.287  2018/05/01 17:57:25  brouard
   if (!m) nrerror("allocation failure 1 in matrix()");    Summary: Bug fixed by providing frequencies only for non missing covariates
   m += NR_END;  
   m -= nrl;  
   
   m[nrl]=(double **) malloc((size_t)((nrow*ncol+NR_END)*sizeof(double)));    Revision 1.286  2018/04/27 14:27:04  brouard
   if (!m[nrl]) nrerror("allocation failure 2 in matrix()");    Summary: some minor bugs
   m[nrl] += NR_END;  
   m[nrl] -= ncl;  
   
   for (i=nrl+1; i<=nrh; i++) m[i]=m[i-1]+ncol;    Revision 1.285  2018/04/21 21:02:16  brouard
     Summary: Some bugs fixed, valgrind tested
   
   m[nrl][ncl]=(double *) malloc((size_t)((nrow*ncol*nlay+NR_END)*sizeof(double)));    Revision 1.284  2018/04/20 05:22:13  brouard
   if (!m[nrl][ncl]) nrerror("allocation failure 3 in matrix()");    Summary: Computing mean and stdeviation of fixed quantitative variables
   m[nrl][ncl] += NR_END;  
   m[nrl][ncl] -= nll;  
   for (j=ncl+1; j<=nch; j++)   
     m[nrl][j]=m[nrl][j-1]+nlay;  
     
   for (i=nrl+1; i<=nrh; i++) {  
     m[i][ncl]=m[i-1l][ncl]+ncol*nlay;  
     for (j=ncl+1; j<=nch; j++)   
       m[i][j]=m[i][j-1]+nlay;  
   }  
   return m;   
   /*  gdb: p *(m+1) <=> p m[1] and p (m+1) <=> p (m+1) <=> p &(m[1])  
            &(m[i][j][k]) <=> *((*(m+i) + j)+k)  
   */  
 }  
   
 /*************************free ma3x ************************/    Revision 1.283  2018/04/19 14:49:16  brouard
 void free_ma3x(double ***m, long nrl, long nrh, long ncl, long nch,long nll, long nlh)    Summary: Some minor bugs fixed
 {  
   free((FREE_ARG)(m[nrl][ncl]+ nll-NR_END));  
   free((FREE_ARG)(m[nrl]+ncl-NR_END));  
   free((FREE_ARG)(m+nrl-NR_END));  
 }  
   
 /***************** f1dim *************************/    Revision 1.282  2018/02/27 22:50:02  brouard
 extern int ncom;     *** empty log message ***
 extern double *pcom,*xicom;  
 extern double (*nrfunc)(double []);   
    
 double f1dim(double x)   
 {   
   int j;   
   double f;  
   double *xt;   
    
   xt=vector(1,ncom);   
   for (j=1;j<=ncom;j++) xt[j]=pcom[j]+x*xicom[j];   
   f=(*nrfunc)(xt);   
   free_vector(xt,1,ncom);   
   return f;   
 }   
   
 /*****************brent *************************/    Revision 1.281  2018/02/27 19:25:23  brouard
 double brent(double ax, double bx, double cx, double (*f)(double), double tol,  double *xmin)     Summary: Adding second argument for quitting
 {   
   int iter;   
   double a,b,d,etemp;  
   double fu,fv,fw,fx;  
   double ftemp;  
   double p,q,r,tol1,tol2,u,v,w,x,xm;   
   double e=0.0;   
    
   a=(ax < cx ? ax : cx);   
   b=(ax > cx ? ax : cx);   
   x=w=v=bx;   
   fw=fv=fx=(*f)(x);   
   for (iter=1;iter<=ITMAX;iter++) {   
     xm=0.5*(a+b);   
     tol2=2.0*(tol1=tol*fabs(x)+ZEPS);   
     /*          if (2.0*fabs(fp-(*fret)) <= ftol*(fabs(fp)+fabs(*fret)))*/  
     printf(".");fflush(stdout);  
     fprintf(ficlog,".");fflush(ficlog);  
 #ifdef DEBUG  
     printf("br %d,x=%.10e xm=%.10e b=%.10e a=%.10e tol=%.10e tol1=%.10e tol2=%.10e x-xm=%.10e fx=%.12e fu=%.12e,fw=%.12e,ftemp=%.12e,ftol=%.12e\n",iter,x,xm,b,a,tol,tol1,tol2,(x-xm),fx,fu,fw,ftemp,ftol);  
     fprintf(ficlog,"br %d,x=%.10e xm=%.10e b=%.10e a=%.10e tol=%.10e tol1=%.10e tol2=%.10e x-xm=%.10e fx=%.12e fu=%.12e,fw=%.12e,ftemp=%.12e,ftol=%.12e\n",iter,x,xm,b,a,tol,tol1,tol2,(x-xm),fx,fu,fw,ftemp,ftol);  
     /*          if ((fabs(x-xm) <= (tol2-0.5*(b-a)))||(2.0*fabs(fu-ftemp) <= ftol*1.e-2*(fabs(fu)+fabs(ftemp)))) { */  
 #endif  
     if (fabs(x-xm) <= (tol2-0.5*(b-a))){   
       *xmin=x;   
       return fx;   
     }   
     ftemp=fu;  
     if (fabs(e) > tol1) {   
       r=(x-w)*(fx-fv);   
       q=(x-v)*(fx-fw);   
       p=(x-v)*q-(x-w)*r;   
       q=2.0*(q-r);   
       if (q > 0.0) p = -p;   
       q=fabs(q);   
       etemp=e;   
       e=d;   
       if (fabs(p) >= fabs(0.5*q*etemp) || p <= q*(a-x) || p >= q*(b-x))   
         d=CGOLD*(e=(x >= xm ? a-x : b-x));   
       else {   
         d=p/q;   
         u=x+d;   
         if (u-a < tol2 || b-u < tol2)   
           d=SIGN(tol1,xm-x);   
       }   
     } else {   
       d=CGOLD*(e=(x >= xm ? a-x : b-x));   
     }   
     u=(fabs(d) >= tol1 ? x+d : x+SIGN(tol1,d));   
     fu=(*f)(u);   
     if (fu <= fx) {   
       if (u >= x) a=x; else b=x;   
       SHFT(v,w,x,u)   
         SHFT(fv,fw,fx,fu)   
         } else {   
           if (u < x) a=u; else b=u;   
           if (fu <= fw || w == x) {   
             v=w;   
             w=u;   
             fv=fw;   
             fw=fu;   
           } else if (fu <= fv || v == x || v == w) {   
             v=u;   
             fv=fu;   
           }   
         }   
   }   
   nrerror("Too many iterations in brent");   
   *xmin=x;   
   return fx;   
 }   
   
 /****************** mnbrak ***********************/    Revision 1.280  2018/02/21 07:58:13  brouard
     Summary: 0.99r15
   
 void mnbrak(double *ax, double *bx, double *cx, double *fa, double *fb, double *fc,     New Makefile with recent VirtualBox 5.26. Bug in sqrt negatve in imach.c
             double (*func)(double))   
 {   
   double ulim,u,r,q, dum;  
   double fu;   
    
   *fa=(*func)(*ax);   
   *fb=(*func)(*bx);   
   if (*fb > *fa) {   
     SHFT(dum,*ax,*bx,dum)   
       SHFT(dum,*fb,*fa,dum)   
       }   
   *cx=(*bx)+GOLD*(*bx-*ax);   
   *fc=(*func)(*cx);   
   while (*fb > *fc) {   
     r=(*bx-*ax)*(*fb-*fc);   
     q=(*bx-*cx)*(*fb-*fa);   
     u=(*bx)-((*bx-*cx)*q-(*bx-*ax)*r)/   
       (2.0*SIGN(FMAX(fabs(q-r),TINY),q-r));   
     ulim=(*bx)+GLIMIT*(*cx-*bx);   
     if ((*bx-u)*(u-*cx) > 0.0) {   
       fu=(*func)(u);   
     } else if ((*cx-u)*(u-ulim) > 0.0) {   
       fu=(*func)(u);   
       if (fu < *fc) {   
         SHFT(*bx,*cx,u,*cx+GOLD*(*cx-*bx))   
           SHFT(*fb,*fc,fu,(*func)(u))   
           }   
     } else if ((u-ulim)*(ulim-*cx) >= 0.0) {   
       u=ulim;   
       fu=(*func)(u);   
     } else {   
       u=(*cx)+GOLD*(*cx-*bx);   
       fu=(*func)(u);   
     }   
     SHFT(*ax,*bx,*cx,u)   
       SHFT(*fa,*fb,*fc,fu)   
       }   
 }   
   
 /*************** linmin ************************/    Revision 1.279  2017/07/20 13:35:01  brouard
     Summary: temporary working
   
 int ncom;     Revision 1.278  2017/07/19 14:09:02  brouard
 double *pcom,*xicom;    Summary: Bug for mobil_average=0 and prevforecast fixed(?)
 double (*nrfunc)(double []);   
    
 void linmin(double p[], double xi[], int n, double *fret,double (*func)(double []))   
 {   
   double brent(double ax, double bx, double cx,   
                double (*f)(double), double tol, double *xmin);   
   double f1dim(double x);   
   void mnbrak(double *ax, double *bx, double *cx, double *fa, double *fb,   
               double *fc, double (*func)(double));   
   int j;   
   double xx,xmin,bx,ax;   
   double fx,fb,fa;  
    
   ncom=n;   
   pcom=vector(1,n);   
   xicom=vector(1,n);   
   nrfunc=func;   
   for (j=1;j<=n;j++) {   
     pcom[j]=p[j];   
     xicom[j]=xi[j];   
   }   
   ax=0.0;   
   xx=1.0;   
   mnbrak(&ax,&xx,&bx,&fa,&fx,&fb,f1dim);   
   *fret=brent(ax,xx,bx,f1dim,TOL,&xmin);   
 #ifdef DEBUG  
   printf("retour brent fret=%.12e xmin=%.12e\n",*fret,xmin);  
   fprintf(ficlog,"retour brent fret=%.12e xmin=%.12e\n",*fret,xmin);  
 #endif  
   for (j=1;j<=n;j++) {   
     xi[j] *= xmin;   
     p[j] += xi[j];   
   }   
   free_vector(xicom,1,n);   
   free_vector(pcom,1,n);   
 }   
   
 /*************** powell ************************/    Revision 1.277  2017/07/17 08:53:49  brouard
 void powell(double p[], double **xi, int n, double ftol, int *iter, double *fret,     Summary: BOM files can be read now
             double (*func)(double []))   
 {   
   void linmin(double p[], double xi[], int n, double *fret,   
               double (*func)(double []));   
   int i,ibig,j;   
   double del,t,*pt,*ptt,*xit;  
   double fp,fptt;  
   double *xits;  
   pt=vector(1,n);   
   ptt=vector(1,n);   
   xit=vector(1,n);   
   xits=vector(1,n);   
   *fret=(*func)(p);   
   for (j=1;j<=n;j++) pt[j]=p[j];   
   for (*iter=1;;++(*iter)) {   
     fp=(*fret);   
     ibig=0;   
     del=0.0;   
     printf("\nPowell iter=%d -2*LL=%.12f",*iter,*fret);  
     fprintf(ficlog,"\nPowell iter=%d -2*LL=%.12f",*iter,*fret);  
     fprintf(ficrespow,"%d %.12f",*iter,*fret);  
     for (i=1;i<=n;i++) {  
       printf(" %d %.12f",i, p[i]);  
       fprintf(ficlog," %d %.12lf",i, p[i]);  
       fprintf(ficrespow," %.12lf", p[i]);  
     }  
     printf("\n");  
     fprintf(ficlog,"\n");  
     fprintf(ficrespow,"\n");  
     for (i=1;i<=n;i++) {   
       for (j=1;j<=n;j++) xit[j]=xi[j][i];   
       fptt=(*fret);   
 #ifdef DEBUG  
       printf("fret=%lf \n",*fret);  
       fprintf(ficlog,"fret=%lf \n",*fret);  
 #endif  
       printf("%d",i);fflush(stdout);  
       fprintf(ficlog,"%d",i);fflush(ficlog);  
       linmin(p,xit,n,fret,func);   
       if (fabs(fptt-(*fret)) > del) {   
         del=fabs(fptt-(*fret));   
         ibig=i;   
       }   
 #ifdef DEBUG  
       printf("%d %.12e",i,(*fret));  
       fprintf(ficlog,"%d %.12e",i,(*fret));  
       for (j=1;j<=n;j++) {  
         xits[j]=FMAX(fabs(p[j]-pt[j]),1.e-5);  
         printf(" x(%d)=%.12e",j,xit[j]);  
         fprintf(ficlog," x(%d)=%.12e",j,xit[j]);  
       }  
       for(j=1;j<=n;j++) {  
         printf(" p=%.12e",p[j]);  
         fprintf(ficlog," p=%.12e",p[j]);  
       }  
       printf("\n");  
       fprintf(ficlog,"\n");  
 #endif  
     }   
     if (2.0*fabs(fp-(*fret)) <= ftol*(fabs(fp)+fabs(*fret))) {  
 #ifdef DEBUG  
       int k[2],l;  
       k[0]=1;  
       k[1]=-1;  
       printf("Max: %.12e",(*func)(p));  
       fprintf(ficlog,"Max: %.12e",(*func)(p));  
       for (j=1;j<=n;j++) {  
         printf(" %.12e",p[j]);  
         fprintf(ficlog," %.12e",p[j]);  
       }  
       printf("\n");  
       fprintf(ficlog,"\n");  
       for(l=0;l<=1;l++) {  
         for (j=1;j<=n;j++) {  
           ptt[j]=p[j]+(p[j]-pt[j])*k[l];  
           printf("l=%d j=%d ptt=%.12e, xits=%.12e, p=%.12e, xit=%.12e", l,j,ptt[j],xits[j],p[j],xit[j]);  
           fprintf(ficlog,"l=%d j=%d ptt=%.12e, xits=%.12e, p=%.12e, xit=%.12e", l,j,ptt[j],xits[j],p[j],xit[j]);  
         }  
         printf("func(ptt)=%.12e, deriv=%.12e\n",(*func)(ptt),(ptt[j]-p[j])/((*func)(ptt)-(*func)(p)));  
         fprintf(ficlog,"func(ptt)=%.12e, deriv=%.12e\n",(*func)(ptt),(ptt[j]-p[j])/((*func)(ptt)-(*func)(p)));  
       }  
 #endif  
   
     Revision 1.276  2017/06/30 15:48:31  brouard
     Summary: Graphs improvements
   
       free_vector(xit,1,n);     Revision 1.275  2017/06/30 13:39:33  brouard
       free_vector(xits,1,n);     Summary: Saito's color
       free_vector(ptt,1,n);   
       free_vector(pt,1,n);   
       return;   
     }   
     if (*iter == ITMAX) nrerror("powell exceeding maximum iterations.");   
     for (j=1;j<=n;j++) {   
       ptt[j]=2.0*p[j]-pt[j];   
       xit[j]=p[j]-pt[j];   
       pt[j]=p[j];   
     }   
     fptt=(*func)(ptt);   
     if (fptt < fp) {   
       t=2.0*(fp-2.0*(*fret)+fptt)*SQR(fp-(*fret)-del)-del*SQR(fp-fptt);   
       if (t < 0.0) {   
         linmin(p,xit,n,fret,func);   
         for (j=1;j<=n;j++) {   
           xi[j][ibig]=xi[j][n];   
           xi[j][n]=xit[j];   
         }  
 #ifdef DEBUG  
         printf("Direction changed  last moved %d in place of ibig=%d, new last is the average:\n",n,ibig);  
         fprintf(ficlog,"Direction changed  last moved %d in place of ibig=%d, new last is the average:\n",n,ibig);  
         for(j=1;j<=n;j++){  
           printf(" %.12e",xit[j]);  
           fprintf(ficlog," %.12e",xit[j]);  
         }  
         printf("\n");  
         fprintf(ficlog,"\n");  
 #endif  
       }  
     }   
   }   
 }   
   
 /**** Prevalence limit (stable prevalence)  ****************/    Revision 1.274  2017/06/29 09:47:08  brouard
     Summary: Version 0.99r14
   
 double **prevalim(double **prlim, int nlstate, double x[], double age, double **oldm, double **savm, double ftolpl, int ij)    Revision 1.273  2017/06/27 11:06:02  brouard
 {    Summary: More documentation on projections
   /* Computes the prevalence limit in each live state at age x by left multiplying the unit  
      matrix by transitions matrix until convergence is reached */  
   
   int i, ii,j,k;    Revision 1.272  2017/06/27 10:22:40  brouard
   double min, max, maxmin, maxmax,sumnew=0.;    Summary: Color of backprojection changed from 6 to 5(yellow)
   double **matprod2();  
   double **out, cov[NCOVMAX], **pmij();  
   double **newm;  
   double agefin, delaymax=50 ; /* Max number of years to converge */  
   
   for (ii=1;ii<=nlstate+ndeath;ii++)    Revision 1.271  2017/06/27 10:17:50  brouard
     for (j=1;j<=nlstate+ndeath;j++){    Summary: Some bug with rint
       oldm[ii][j]=(ii==j ? 1.0 : 0.0);  
     }  
   
    cov[1]=1.;    Revision 1.270  2017/05/24 05:45:29  brouard
      *** empty log message ***
  /* Even if hstepm = 1, at least one multiplication by the unit matrix */  
   for(agefin=age-stepm/YEARM; agefin>=age-delaymax; agefin=agefin-stepm/YEARM){  
     newm=savm;  
     /* Covariates have to be included here again */  
      cov[2]=agefin;  
     
       for (k=1; k<=cptcovn;k++) {  
         cov[2+k]=nbcode[Tvar[k]][codtab[ij][Tvar[k]]];  
         /*      printf("ij=%d k=%d Tvar[k]=%d nbcode=%d cov=%lf codtab[ij][Tvar[k]]=%d \n",ij,k, Tvar[k],nbcode[Tvar[k]][codtab[ij][Tvar[k]]],cov[2+k], codtab[ij][Tvar[k]]);*/  
       }  
       for (k=1; k<=cptcovage;k++) cov[2+Tage[k]]=cov[2+Tage[k]]*cov[2];  
       for (k=1; k<=cptcovprod;k++)  
         cov[2+Tprod[k]]=nbcode[Tvard[k][1]][codtab[ij][Tvard[k][1]]]*nbcode[Tvard[k][2]][codtab[ij][Tvard[k][2]]];  
   
       /*printf("ij=%d cptcovprod=%d tvar=%d ", ij, cptcovprod, Tvar[1]);*/  
       /*printf("ij=%d cov[3]=%lf cov[4]=%lf \n",ij, cov[3],cov[4]);*/  
       /*printf("ij=%d cov[3]=%lf \n",ij, cov[3]);*/  
     out=matprod2(newm, pmij(pmmij,cov,ncovmodel,x,nlstate),1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, oldm);  
   
     savm=oldm;    Revision 1.269  2017/05/23 08:39:25  brouard
     oldm=newm;    Summary: Code into subroutine, cleanings
     maxmax=0.;  
     for(j=1;j<=nlstate;j++){  
       min=1.;  
       max=0.;  
       for(i=1; i<=nlstate; i++) {  
         sumnew=0;  
         for(k=1; k<=ndeath; k++) sumnew+=newm[i][nlstate+k];  
         prlim[i][j]= newm[i][j]/(1-sumnew);  
         max=FMAX(max,prlim[i][j]);  
         min=FMIN(min,prlim[i][j]);  
       }  
       maxmin=max-min;  
       maxmax=FMAX(maxmax,maxmin);  
     }  
     if(maxmax < ftolpl){  
       return prlim;  
     }  
   }  
 }  
   
 /*************** transition probabilities ***************/     Revision 1.268  2017/05/18 20:09:32  brouard
     Summary: backprojection and confidence intervals of backprevalence
   
 double **pmij(double **ps, double *cov, int ncovmodel, double *x, int nlstate )    Revision 1.267  2017/05/13 10:25:05  brouard
 {    Summary: temporary save for backprojection
   double s1, s2;  
   /*double t34;*/  
   int i,j,j1, nc, ii, jj;  
   
     for(i=1; i<= nlstate; i++){    Revision 1.266  2017/05/13 07:26:12  brouard
     for(j=1; j<i;j++){    Summary: Version 0.99r13 (improvements and bugs fixed)
       for (nc=1, s2=0.;nc <=ncovmodel; nc++){  
         /*s2 += param[i][j][nc]*cov[nc];*/  
         s2 += x[(i-1)*nlstate*ncovmodel+(j-1)*ncovmodel+nc+(i-1)*(ndeath-1)*ncovmodel]*cov[nc];  
         /*printf("Int j<i s1=%.17e, s2=%.17e\n",s1,s2);*/  
       }  
       ps[i][j]=s2;  
       /*printf("s1=%.17e, s2=%.17e\n",s1,s2);*/  
     }  
     for(j=i+1; j<=nlstate+ndeath;j++){  
       for (nc=1, s2=0.;nc <=ncovmodel; nc++){  
         s2 += x[(i-1)*nlstate*ncovmodel+(j-2)*ncovmodel+nc+(i-1)*(ndeath-1)*ncovmodel]*cov[nc];  
         /*printf("Int j>i s1=%.17e, s2=%.17e %lx %lx\n",s1,s2,s1,s2);*/  
       }  
       ps[i][j]=s2;  
     }  
   }  
     /*ps[3][2]=1;*/  
   
   for(i=1; i<= nlstate; i++){    Revision 1.265  2017/04/26 16:22:11  brouard
      s1=0;    Summary: imach 0.99r13 Some bugs fixed
     for(j=1; j<i; j++)  
       s1+=exp(ps[i][j]);  
     for(j=i+1; j<=nlstate+ndeath; j++)  
       s1+=exp(ps[i][j]);  
     ps[i][i]=1./(s1+1.);  
     for(j=1; j<i; j++)  
       ps[i][j]= exp(ps[i][j])*ps[i][i];  
     for(j=i+1; j<=nlstate+ndeath; j++)  
       ps[i][j]= exp(ps[i][j])*ps[i][i];  
     /* ps[i][nlstate+1]=1.-s1- ps[i][i];*/ /* Sum should be 1 */  
   } /* end i */  
   
   for(ii=nlstate+1; ii<= nlstate+ndeath; ii++){    Revision 1.264  2017/04/26 06:01:29  brouard
     for(jj=1; jj<= nlstate+ndeath; jj++){    Summary: Labels in graphs
       ps[ii][jj]=0;  
       ps[ii][ii]=1;  
     }  
   }  
   
     Revision 1.263  2017/04/24 15:23:15  brouard
     Summary: to save
   
   /*   for(ii=1; ii<= nlstate+ndeath; ii++){    Revision 1.262  2017/04/18 16:48:12  brouard
     for(jj=1; jj<= nlstate+ndeath; jj++){    *** empty log message ***
      printf("%lf ",ps[ii][jj]);  
    }  
     printf("\n ");  
     }  
     printf("\n ");printf("%lf ",cov[2]);*/  
 /*  
   for(i=1; i<= npar; i++) printf("%f ",x[i]);  
   goto end;*/  
     return ps;  
 }  
   
 /**************** Product of 2 matrices ******************/    Revision 1.261  2017/04/05 10:14:09  brouard
     Summary: Bug in E_ as well as in T_ fixed nres-1 vs k1-1
   
 double **matprod2(double **out, double **in,long nrl, long nrh, long ncl, long nch, long ncolol, long ncoloh, double **b)    Revision 1.260  2017/04/04 17:46:59  brouard
 {    Summary: Gnuplot indexations fixed (humm)
   /* Computes the matrix product of in(1,nrh-nrl+1)(1,nch-ncl+1) times  
      b(1,nch-ncl+1)(1,ncoloh-ncolol+1) into out(...) */  
   /* in, b, out are matrice of pointers which should have been initialized   
      before: only the contents of out is modified. The function returns  
      a pointer to pointers identical to out */  
   long i, j, k;  
   for(i=nrl; i<= nrh; i++)  
     for(k=ncolol; k<=ncoloh; k++)  
       for(j=ncl,out[i][k]=0.; j<=nch; j++)  
         out[i][k] +=in[i][j]*b[j][k];  
   
   return out;    Revision 1.259  2017/04/04 13:01:16  brouard
 }    Summary: Some errors to warnings only if date of death is unknown but status is death we could set to pi3
   
     Revision 1.258  2017/04/03 10:17:47  brouard
     Summary: Version 0.99r12
   
 /************* Higher Matrix Product ***************/    Some cleanings, conformed with updated documentation.
   
 double ***hpxij(double ***po, int nhstepm, double age, int hstepm, double *x, int nlstate, int stepm, double **oldm, double **savm, int ij )    Revision 1.257  2017/03/29 16:53:30  brouard
 {    Summary: Temp
   /* Computes the transition matrix starting at age 'age' over   
      'nhstepm*hstepm*stepm' months (i.e. until  
      age (in years)  age+nhstepm*hstepm*stepm/12) by multiplying   
      nhstepm*hstepm matrices.   
      Output is stored in matrix po[i][j][h] for h every 'hstepm' step   
      (typically every 2 years instead of every month which is too big   
      for the memory).  
      Model is determined by parameters x and covariates have to be   
      included manually here.   
   
      */    Revision 1.256  2017/03/27 05:50:23  brouard
     Summary: Temporary
   
   int i, j, d, h, k;    Revision 1.255  2017/03/08 16:02:28  brouard
   double **out, cov[NCOVMAX];    Summary: IMaCh version 0.99r10 bugs in gnuplot fixed
   double **newm;  
   
   /* Hstepm could be zero and should return the unit matrix */    Revision 1.254  2017/03/08 07:13:00  brouard
   for (i=1;i<=nlstate+ndeath;i++)    Summary: Fixing data parameter line
     for (j=1;j<=nlstate+ndeath;j++){  
       oldm[i][j]=(i==j ? 1.0 : 0.0);  
       po[i][j][0]=(i==j ? 1.0 : 0.0);  
     }  
   /* Even if hstepm = 1, at least one multiplication by the unit matrix */  
   for(h=1; h <=nhstepm; h++){  
     for(d=1; d <=hstepm; d++){  
       newm=savm;  
       /* Covariates have to be included here again */  
       cov[1]=1.;  
       cov[2]=age+((h-1)*hstepm + (d-1))*stepm/YEARM;  
       for (k=1; k<=cptcovn;k++) cov[2+k]=nbcode[Tvar[k]][codtab[ij][Tvar[k]]];  
       for (k=1; k<=cptcovage;k++)  
         cov[2+Tage[k]]=cov[2+Tage[k]]*cov[2];  
       for (k=1; k<=cptcovprod;k++)  
         cov[2+Tprod[k]]=nbcode[Tvard[k][1]][codtab[ij][Tvard[k][1]]]*nbcode[Tvard[k][2]][codtab[ij][Tvard[k][2]]];  
   
     Revision 1.253  2016/12/15 11:59:41  brouard
     Summary: 0.99 in progress
   
       /*printf("hxi cptcov=%d cptcode=%d\n",cptcov,cptcode);*/    Revision 1.252  2016/09/15 21:15:37  brouard
       /*printf("h=%d d=%d age=%f cov=%f\n",h,d,age,cov[2]);*/    *** empty log message ***
       out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath,   
                    pmij(pmmij,cov,ncovmodel,x,nlstate));  
       savm=oldm;  
       oldm=newm;  
     }  
     for(i=1; i<=nlstate+ndeath; i++)  
       for(j=1;j<=nlstate+ndeath;j++) {  
         po[i][j][h]=newm[i][j];  
         /*printf("i=%d j=%d h=%d po[i][j][h]=%f ",i,j,h,po[i][j][h]);  
          */  
       }  
   } /* end h */  
   return po;  
 }  
   
     Revision 1.251  2016/09/15 15:01:13  brouard
     Summary: not working
   
 /*************** log-likelihood *************/    Revision 1.250  2016/09/08 16:07:27  brouard
 double func( double *x)    Summary: continue
 {  
   int i, ii, j, k, mi, d, kk;  
   double l, ll[NLSTATEMAX], cov[NCOVMAX];  
   double **out;  
   double sw; /* Sum of weights */  
   double lli; /* Individual log likelihood */  
   int s1, s2;  
   double bbh, survp;  
   long ipmx;  
   /*extern weight */  
   /* We are differentiating ll according to initial status */  
   /*  for (i=1;i<=npar;i++) printf("%f ", x[i]);*/  
   /*for(i=1;i<imx;i++)   
     printf(" %d\n",s[4][i]);  
   */  
   cov[1]=1.;  
   
   for(k=1; k<=nlstate; k++) ll[k]=0.;    Revision 1.249  2016/09/07 17:14:18  brouard
     Summary: Starting values from frequencies
   
   if(mle==1){    Revision 1.248  2016/09/07 14:10:18  brouard
     for (i=1,ipmx=0, sw=0.; i<=imx; i++){    *** empty log message ***
       for (k=1; k<=cptcovn;k++) cov[2+k]=covar[Tvar[k]][i];  
       for(mi=1; mi<= wav[i]-1; mi++){  
         for (ii=1;ii<=nlstate+ndeath;ii++)  
           for (j=1;j<=nlstate+ndeath;j++){  
             oldm[ii][j]=(ii==j ? 1.0 : 0.0);  
             savm[ii][j]=(ii==j ? 1.0 : 0.0);  
           }  
         for(d=0; d<dh[mi][i]; d++){  
           newm=savm;  
           cov[2]=agev[mw[mi][i]][i]+d*stepm/YEARM;  
           for (kk=1; kk<=cptcovage;kk++) {  
             cov[Tage[kk]+2]=covar[Tvar[Tage[kk]]][i]*cov[2];  
           }  
           out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,  
                        1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));  
           savm=oldm;  
           oldm=newm;  
         } /* end mult */  
         
         /*lli=log(out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]]);*/ /* Original formula */  
         /* But now since version 0.9 we anticipate for bias and large stepm.  
          * If stepm is larger than one month (smallest stepm) and if the exact delay   
          * (in months) between two waves is not a multiple of stepm, we rounded to   
          * the nearest (and in case of equal distance, to the lowest) interval but now  
          * we keep into memory the bias bh[mi][i] and also the previous matrix product  
          * (i.e to dh[mi][i]-1) saved in 'savm'. The we inter(extra)polate the  
          * probability in order to take into account the bias as a fraction of the way  
          * from savm to out if bh is neagtive or even beyond if bh is positive. bh varies  
          * -stepm/2 to stepm/2 .  
          * For stepm=1 the results are the same as for previous versions of Imach.  
          * For stepm > 1 the results are less biased than in previous versions.   
          */  
         s1=s[mw[mi][i]][i];  
         s2=s[mw[mi+1][i]][i];  
         bbh=(double)bh[mi][i]/(double)stepm;   
         /* bias is positive if real duration  
          * is higher than the multiple of stepm and negative otherwise.  
          */  
         /* lli= (savm[s1][s2]>1.e-8 ?(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]):log((1.+bbh)*out[s1][s2]));*/  
         if( s2 > nlstate){   
           /* i.e. if s2 is a death state and if the date of death is known then the contribution  
              to the likelihood is the probability to die between last step unit time and current   
              step unit time, which is also the differences between probability to die before dh   
              and probability to die before dh-stepm .   
              In version up to 0.92 likelihood was computed  
         as if date of death was unknown. Death was treated as any other  
         health state: the date of the interview describes the actual state  
         and not the date of a change in health state. The former idea was  
         to consider that at each interview the state was recorded  
         (healthy, disable or death) and IMaCh was corrected; but when we  
         introduced the exact date of death then we should have modified  
         the contribution of an exact death to the likelihood. This new  
         contribution is smaller and very dependent of the step unit  
         stepm. It is no more the probability to die between last interview  
         and month of death but the probability to survive from last  
         interview up to one month before death multiplied by the  
         probability to die within a month. Thanks to Chris  
         Jackson for correcting this bug.  Former versions increased  
         mortality artificially. The bad side is that we add another loop  
         which slows down the processing. The difference can be up to 10%  
         lower mortality.  
           */  
           lli=log(out[s1][s2] - savm[s1][s2]);  
         }else{  
           lli= log((1.+bbh)*out[s1][s2]- bbh*savm[s1][s2]); /* linear interpolation */  
           /*  lli= (savm[s1][s2]>(double)1.e-8 ?log((1.+bbh)*out[s1][s2]- bbh*(savm[s1][s2])):log((1.+bbh)*out[s1][s2]));*/ /* linear interpolation */  
         }   
         /*lli=(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]);*/  
         /*if(lli ==000.0)*/  
         /*printf("bbh= %f lli=%f savm=%f out=%f %d\n",bbh,lli,savm[s1][s2], out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]],i); */  
         ipmx +=1;  
         sw += weight[i];  
         ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;  
       } /* end of wave */  
     } /* end of individual */  
   }  else if(mle==2){  
     for (i=1,ipmx=0, sw=0.; i<=imx; i++){  
       for (k=1; k<=cptcovn;k++) cov[2+k]=covar[Tvar[k]][i];  
       for(mi=1; mi<= wav[i]-1; mi++){  
         for (ii=1;ii<=nlstate+ndeath;ii++)  
           for (j=1;j<=nlstate+ndeath;j++){  
             oldm[ii][j]=(ii==j ? 1.0 : 0.0);  
             savm[ii][j]=(ii==j ? 1.0 : 0.0);  
           }  
         for(d=0; d<=dh[mi][i]; d++){  
           newm=savm;  
           cov[2]=agev[mw[mi][i]][i]+d*stepm/YEARM;  
           for (kk=1; kk<=cptcovage;kk++) {  
             cov[Tage[kk]+2]=covar[Tvar[Tage[kk]]][i]*cov[2];  
           }  
           out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,  
                        1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));  
           savm=oldm;  
           oldm=newm;  
         } /* end mult */  
         
         /*lli=log(out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]]);*/ /* Original formula */  
         /* But now since version 0.9 we anticipate for bias and large stepm.  
          * If stepm is larger than one month (smallest stepm) and if the exact delay   
          * (in months) between two waves is not a multiple of stepm, we rounded to   
          * the nearest (and in case of equal distance, to the lowest) interval but now  
          * we keep into memory the bias bh[mi][i] and also the previous matrix product  
          * (i.e to dh[mi][i]-1) saved in 'savm'. The we inter(extra)polate the  
          * probability in order to take into account the bias as a fraction of the way  
          * from savm to out if bh is neagtive or even beyond if bh is positive. bh varies  
          * -stepm/2 to stepm/2 .  
          * For stepm=1 the results are the same as for previous versions of Imach.  
          * For stepm > 1 the results are less biased than in previous versions.   
          */  
         s1=s[mw[mi][i]][i];  
         s2=s[mw[mi+1][i]][i];  
         bbh=(double)bh[mi][i]/(double)stepm;   
         /* bias is positive if real duration  
          * is higher than the multiple of stepm and negative otherwise.  
          */  
         lli= (savm[s1][s2]>(double)1.e-8 ?log((1.+bbh)*out[s1][s2]- bbh*(savm[s1][s2])):log((1.+bbh)*out[s1][s2])); /* linear interpolation */  
         /* lli= (savm[s1][s2]>1.e-8 ?(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]):log((1.+bbh)*out[s1][s2]));*/  
         /*lli= (savm[s1][s2]>1.e-8 ?(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]):log((1.-+bh)*out[s1][s2])); */ /* exponential interpolation */  
         /*lli=(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]);*/  
         /*if(lli ==000.0)*/  
         /*printf("bbh= %f lli=%f savm=%f out=%f %d\n",bbh,lli,savm[s1][s2], out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]],i); */  
         ipmx +=1;  
         sw += weight[i];  
         ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;  
       } /* end of wave */  
     } /* end of individual */  
   }  else if(mle==3){  /* exponential inter-extrapolation */  
     for (i=1,ipmx=0, sw=0.; i<=imx; i++){  
       for (k=1; k<=cptcovn;k++) cov[2+k]=covar[Tvar[k]][i];  
       for(mi=1; mi<= wav[i]-1; mi++){  
         for (ii=1;ii<=nlstate+ndeath;ii++)  
           for (j=1;j<=nlstate+ndeath;j++){  
             oldm[ii][j]=(ii==j ? 1.0 : 0.0);  
             savm[ii][j]=(ii==j ? 1.0 : 0.0);  
           }  
         for(d=0; d<dh[mi][i]; d++){  
           newm=savm;  
           cov[2]=agev[mw[mi][i]][i]+d*stepm/YEARM;  
           for (kk=1; kk<=cptcovage;kk++) {  
             cov[Tage[kk]+2]=covar[Tvar[Tage[kk]]][i]*cov[2];  
           }  
           out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,  
                        1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));  
           savm=oldm;  
           oldm=newm;  
         } /* end mult */  
         
         /*lli=log(out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]]);*/ /* Original formula */  
         /* But now since version 0.9 we anticipate for bias and large stepm.  
          * If stepm is larger than one month (smallest stepm) and if the exact delay   
          * (in months) between two waves is not a multiple of stepm, we rounded to   
          * the nearest (and in case of equal distance, to the lowest) interval but now  
          * we keep into memory the bias bh[mi][i] and also the previous matrix product  
          * (i.e to dh[mi][i]-1) saved in 'savm'. The we inter(extra)polate the  
          * probability in order to take into account the bias as a fraction of the way  
          * from savm to out if bh is neagtive or even beyond if bh is positive. bh varies  
          * -stepm/2 to stepm/2 .  
          * For stepm=1 the results are the same as for previous versions of Imach.  
          * For stepm > 1 the results are less biased than in previous versions.   
          */  
         s1=s[mw[mi][i]][i];  
         s2=s[mw[mi+1][i]][i];  
         bbh=(double)bh[mi][i]/(double)stepm;   
         /* bias is positive if real duration  
          * is higher than the multiple of stepm and negative otherwise.  
          */  
         /* lli= (savm[s1][s2]>(double)1.e-8 ?log((1.+bbh)*out[s1][s2]- bbh*(savm[s1][s2])):log((1.+bbh)*out[s1][s2])); */ /* linear interpolation */  
         lli= (savm[s1][s2]>1.e-8 ?(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]):log((1.+bbh)*out[s1][s2])); /* exponential inter-extrapolation */  
         /*lli=(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]);*/  
         /*if(lli ==000.0)*/  
         /*printf("bbh= %f lli=%f savm=%f out=%f %d\n",bbh,lli,savm[s1][s2], out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]],i); */  
         ipmx +=1;  
         sw += weight[i];  
         ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;  
       } /* end of wave */  
     } /* end of individual */  
   }else if (mle==4){  /* ml=4 no inter-extrapolation */  
     for (i=1,ipmx=0, sw=0.; i<=imx; i++){  
       for (k=1; k<=cptcovn;k++) cov[2+k]=covar[Tvar[k]][i];  
       for(mi=1; mi<= wav[i]-1; mi++){  
         for (ii=1;ii<=nlstate+ndeath;ii++)  
           for (j=1;j<=nlstate+ndeath;j++){  
             oldm[ii][j]=(ii==j ? 1.0 : 0.0);  
             savm[ii][j]=(ii==j ? 1.0 : 0.0);  
           }  
         for(d=0; d<dh[mi][i]; d++){  
           newm=savm;  
           cov[2]=agev[mw[mi][i]][i]+d*stepm/YEARM;  
           for (kk=1; kk<=cptcovage;kk++) {  
             cov[Tage[kk]+2]=covar[Tvar[Tage[kk]]][i]*cov[2];  
           }  
           
           out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,  
                        1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));  
           savm=oldm;  
           oldm=newm;  
         } /* end mult */  
         
         s1=s[mw[mi][i]][i];  
         s2=s[mw[mi+1][i]][i];  
         if( s2 > nlstate){   
           lli=log(out[s1][s2] - savm[s1][s2]);  
         }else{  
           lli=log(out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]]); /* Original formula */  
         }  
         ipmx +=1;  
         sw += weight[i];  
         ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;  
         /*printf("i=%6d s1=%1d s2=%1d mi=%1d mw=%1d dh=%3d prob=%10.6f w=%6.4f out=%10.6f sav=%10.6f\n",i,s1,s2,mi,mw[mi][i],dh[mi][i],exp(lli),weight[i],out[s1][s2],savm[s1][s2]);*/  
       } /* end of wave */  
     } /* end of individual */  
   }else{  /* ml=5 no inter-extrapolation no jackson =0.8a */  
     for (i=1,ipmx=0, sw=0.; i<=imx; i++){  
       for (k=1; k<=cptcovn;k++) cov[2+k]=covar[Tvar[k]][i];  
       for(mi=1; mi<= wav[i]-1; mi++){  
         for (ii=1;ii<=nlstate+ndeath;ii++)  
           for (j=1;j<=nlstate+ndeath;j++){  
             oldm[ii][j]=(ii==j ? 1.0 : 0.0);  
             savm[ii][j]=(ii==j ? 1.0 : 0.0);  
           }  
         for(d=0; d<dh[mi][i]; d++){  
           newm=savm;  
           cov[2]=agev[mw[mi][i]][i]+d*stepm/YEARM;  
           for (kk=1; kk<=cptcovage;kk++) {  
             cov[Tage[kk]+2]=covar[Tvar[Tage[kk]]][i]*cov[2];  
           }  
           
           out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,  
                        1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));  
           savm=oldm;  
           oldm=newm;  
         } /* end mult */  
         
         s1=s[mw[mi][i]][i];  
         s2=s[mw[mi+1][i]][i];  
         lli=log(out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]]); /* Original formula */  
         ipmx +=1;  
         sw += weight[i];  
         ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;  
         /*printf("i=%6d s1=%1d s2=%1d mi=%1d mw=%1d dh=%3d prob=%10.6f w=%6.4f out=%10.6f sav=%10.6f\n",i,s1,s2,mi,mw[mi][i],dh[mi][i],exp(lli),weight[i],out[s1][s2],savm[s1][s2]);*/  
       } /* end of wave */  
     } /* end of individual */  
   } /* End of if */  
   for(k=1,l=0.; k<=nlstate; k++) l += ll[k];  
   /* printf("l1=%f l2=%f ",ll[1],ll[2]); */  
   l= l*ipmx/sw; /* To get the same order of magnitude as if weight=1 for every body */  
   /*exit(0); */  
   return -l;  
 }  
   
     Revision 1.247  2016/09/02 11:11:21  brouard
     *** empty log message ***
   
 /*********** Maximum Likelihood Estimation ***************/    Revision 1.246  2016/09/02 08:49:22  brouard
     *** empty log message ***
   
 void mlikeli(FILE *ficres,double p[], int npar, int ncovmodel, int nlstate, double ftol, double (*func)(double []))    Revision 1.245  2016/09/02 07:25:01  brouard
 {    *** empty log message ***
   int i,j, iter;  
   double **xi;  
   double fret;  
   char filerespow[FILENAMELENGTH];  
   xi=matrix(1,npar,1,npar);  
   for (i=1;i<=npar;i++)  
     for (j=1;j<=npar;j++)  
       xi[i][j]=(i==j ? 1.0 : 0.0);  
   printf("Powell\n");  fprintf(ficlog,"Powell\n");  
   strcpy(filerespow,"pow");   
   strcat(filerespow,fileres);  
   if((ficrespow=fopen(filerespow,"w"))==NULL) {  
     printf("Problem with resultfile: %s\n", filerespow);  
     fprintf(ficlog,"Problem with resultfile: %s\n", filerespow);  
   }  
   fprintf(ficrespow,"# Powell\n# iter -2*LL");  
   for (i=1;i<=nlstate;i++)  
     for(j=1;j<=nlstate+ndeath;j++)  
       if(j!=i)fprintf(ficrespow," p%1d%1d",i,j);  
   fprintf(ficrespow,"\n");  
   powell(p,xi,npar,ftol,&iter,&fret,func);  
   
   fclose(ficrespow);    Revision 1.244  2016/09/02 07:17:34  brouard
   printf("\n#Number of iterations = %d, -2 Log likelihood = %.12f\n",iter,func(p));    *** empty log message ***
   fprintf(ficlog,"\n#Number of iterations = %d, -2 Log likelihood = %.12f \n",iter,func(p));  
   fprintf(ficres,"#Number of iterations = %d, -2 Log likelihood = %.12f \n",iter,func(p));  
   
 }    Revision 1.243  2016/09/02 06:45:35  brouard
     *** empty log message ***
   
 /**** Computes Hessian and covariance matrix ***/    Revision 1.242  2016/08/30 15:01:20  brouard
 void hesscov(double **matcov, double p[], int npar, double delti[], double ftolhess, double (*func)(double []))    Summary: Fixing a lots
 {  
   double  **a,**y,*x,pd;  
   double **hess;  
   int i, j,jk;  
   int *indx;  
   
   double hessii(double p[], double delta, int theta, double delti[]);    Revision 1.241  2016/08/29 17:17:25  brouard
   double hessij(double p[], double delti[], int i, int j);    Summary: gnuplot problem in Back projection to fix
   void lubksb(double **a, int npar, int *indx, double b[]) ;  
   void ludcmp(double **a, int npar, int *indx, double *d) ;  
   
   hess=matrix(1,npar,1,npar);    Revision 1.240  2016/08/29 07:53:18  brouard
     Summary: Better
   
   printf("\nCalculation of the hessian matrix. Wait...\n");    Revision 1.239  2016/08/26 15:51:03  brouard
   fprintf(ficlog,"\nCalculation of the hessian matrix. Wait...\n");    Summary: Improvement in Powell output in order to copy and paste
   for (i=1;i<=npar;i++){  
     printf("%d",i);fflush(stdout);  
     fprintf(ficlog,"%d",i);fflush(ficlog);  
     hess[i][i]=hessii(p,ftolhess,i,delti);  
     /*printf(" %f ",p[i]);*/  
     /*printf(" %lf ",hess[i][i]);*/  
   }  
     
   for (i=1;i<=npar;i++) {  
     for (j=1;j<=npar;j++)  {  
       if (j>i) {   
         printf(".%d%d",i,j);fflush(stdout);  
         fprintf(ficlog,".%d%d",i,j);fflush(ficlog);  
         hess[i][j]=hessij(p,delti,i,j);  
         hess[j][i]=hess[i][j];      
         /*printf(" %lf ",hess[i][j]);*/  
       }  
     }  
   }  
   printf("\n");  
   fprintf(ficlog,"\n");  
   
   printf("\nInverting the hessian to get the covariance matrix. Wait...\n");    Author:
   fprintf(ficlog,"\nInverting the hessian to get the covariance matrix. Wait...\n");  
     
   a=matrix(1,npar,1,npar);  
   y=matrix(1,npar,1,npar);  
   x=vector(1,npar);  
   indx=ivector(1,npar);  
   for (i=1;i<=npar;i++)  
     for (j=1;j<=npar;j++) a[i][j]=hess[i][j];  
   ludcmp(a,npar,indx,&pd);  
   
   for (j=1;j<=npar;j++) {    Revision 1.238  2016/08/26 14:23:35  brouard
     for (i=1;i<=npar;i++) x[i]=0;    Summary: Starting tests of 0.99
     x[j]=1;  
     lubksb(a,npar,indx,x);  
     for (i=1;i<=npar;i++){   
       matcov[i][j]=x[i];  
     }  
   }  
   
   printf("\n#Hessian matrix#\n");    Revision 1.237  2016/08/26 09:20:19  brouard
   fprintf(ficlog,"\n#Hessian matrix#\n");    Summary: to valgrind
   for (i=1;i<=npar;i++) {   
     for (j=1;j<=npar;j++) {   
       printf("%.3e ",hess[i][j]);  
       fprintf(ficlog,"%.3e ",hess[i][j]);  
     }  
     printf("\n");  
     fprintf(ficlog,"\n");  
   }  
   
   /* Recompute Inverse */    Revision 1.236  2016/08/25 10:50:18  brouard
   for (i=1;i<=npar;i++)    *** empty log message ***
     for (j=1;j<=npar;j++) a[i][j]=matcov[i][j];  
   ludcmp(a,npar,indx,&pd);  
   
   /*  printf("\n#Hessian matrix recomputed#\n");    Revision 1.235  2016/08/25 06:59:23  brouard
     *** empty log message ***
   
   for (j=1;j<=npar;j++) {    Revision 1.234  2016/08/23 16:51:20  brouard
     for (i=1;i<=npar;i++) x[i]=0;    *** empty log message ***
     x[j]=1;  
     lubksb(a,npar,indx,x);  
     for (i=1;i<=npar;i++){   
       y[i][j]=x[i];  
       printf("%.3e ",y[i][j]);  
       fprintf(ficlog,"%.3e ",y[i][j]);  
     }  
     printf("\n");  
     fprintf(ficlog,"\n");  
   }  
   */  
   
   free_matrix(a,1,npar,1,npar);    Revision 1.233  2016/08/23 07:40:50  brouard
   free_matrix(y,1,npar,1,npar);    Summary: not working
   free_vector(x,1,npar);  
   free_ivector(indx,1,npar);  
   free_matrix(hess,1,npar,1,npar);  
   
     Revision 1.232  2016/08/22 14:20:21  brouard
     Summary: not working
   
 }    Revision 1.231  2016/08/22 07:17:15  brouard
     Summary: not working
   
 /*************** hessian matrix ****************/    Revision 1.230  2016/08/22 06:55:53  brouard
 double hessii( double x[], double delta, int theta, double delti[])    Summary: Not working
 {  
   int i;  
   int l=1, lmax=20;  
   double k1,k2;  
   double p2[NPARMAX+1];  
   double res;  
   double delt, delts, nkhi=10.,nkhif=1., khi=1.e-4;  
   double fx;  
   int k=0,kmax=10;  
   double l1;  
   
   fx=func(x);    Revision 1.229  2016/07/23 09:45:53  brouard
   for (i=1;i<=npar;i++) p2[i]=x[i];    Summary: Completing for func too
   for(l=0 ; l <=lmax; l++){  
     l1=pow(10,l);  
     delts=delt;  
     for(k=1 ; k <kmax; k=k+1){  
       delt = delta*(l1*k);  
       p2[theta]=x[theta] +delt;  
       k1=func(p2)-fx;  
       p2[theta]=x[theta]-delt;  
       k2=func(p2)-fx;  
       /*res= (k1-2.0*fx+k2)/delt/delt; */  
       res= (k1+k2)/delt/delt/2.; /* Divided by because L and not 2*L */  
         
 #ifdef DEBUG  
       printf("%d %d k1=%.12e k2=%.12e xk1=%.12e xk2=%.12e delt=%.12e res=%.12e l=%d k=%d,fx=%.12e\n",theta,theta,k1,k2,x[theta]+delt,x[theta]-delt,delt,res, l, k,fx);  
       fprintf(ficlog,"%d %d k1=%.12e k2=%.12e xk1=%.12e xk2=%.12e delt=%.12e res=%.12e l=%d k=%d,fx=%.12e\n",theta,theta,k1,k2,x[theta]+delt,x[theta]-delt,delt,res, l, k,fx);  
 #endif  
       /*if(fabs(k1-2.0*fx+k2) <1.e-13){ */  
       if((k1 <khi/nkhi/2.) || (k2 <khi/nkhi/2.)){  
         k=kmax;  
       }  
       else if((k1 >khi/nkhif) || (k2 >khi/nkhif)){ /* Keeps lastvalue before 3.84/2 KHI2 5% 1d.f. */  
         k=kmax; l=lmax*10.;  
       }  
       else if((k1 >khi/nkhi) || (k2 >khi/nkhi)){   
         delts=delt;  
       }  
     }  
   }  
   delti[theta]=delts;  
   return res;   
     
 }  
   
 double hessij( double x[], double delti[], int thetai,int thetaj)    Revision 1.228  2016/07/22 17:45:30  brouard
 {    Summary: Fixing some arrays, still debugging
   int i;  
   int l=1, l1, lmax=20;  
   double k1,k2,k3,k4,res,fx;  
   double p2[NPARMAX+1];  
   int k;  
   
   fx=func(x);    Revision 1.226  2016/07/12 18:42:34  brouard
   for (k=1; k<=2; k++) {    Summary: temp
     for (i=1;i<=npar;i++) p2[i]=x[i];  
     p2[thetai]=x[thetai]+delti[thetai]/k;  
     p2[thetaj]=x[thetaj]+delti[thetaj]/k;  
     k1=func(p2)-fx;  
     
     p2[thetai]=x[thetai]+delti[thetai]/k;  
     p2[thetaj]=x[thetaj]-delti[thetaj]/k;  
     k2=func(p2)-fx;  
     
     p2[thetai]=x[thetai]-delti[thetai]/k;  
     p2[thetaj]=x[thetaj]+delti[thetaj]/k;  
     k3=func(p2)-fx;  
     
     p2[thetai]=x[thetai]-delti[thetai]/k;  
     p2[thetaj]=x[thetaj]-delti[thetaj]/k;  
     k4=func(p2)-fx;  
     res=(k1-k2-k3+k4)/4.0/delti[thetai]*k/delti[thetaj]*k/2.; /* Because of L not 2*L */  
 #ifdef DEBUG  
     printf("%d %d k=%d, k1=%.12e k2=%.12e k3=%.12e k4=%.12e delti/k=%.12e deltj/k=%.12e, xi-de/k=%.12e xj-de/k=%.12e  res=%.12e k1234=%.12e,k1-2=%.12e,k3-4=%.12e\n",thetai,thetaj,k,k1,k2,k3,k4,delti[thetai]/k,delti[thetaj]/k,x[thetai]-delti[thetai]/k,x[thetaj]-delti[thetaj]/k, res,k1-k2-k3+k4,k1-k2,k3-k4);  
     fprintf(ficlog,"%d %d k=%d, k1=%.12e k2=%.12e k3=%.12e k4=%.12e delti/k=%.12e deltj/k=%.12e, xi-de/k=%.12e xj-de/k=%.12e  res=%.12e k1234=%.12e,k1-2=%.12e,k3-4=%.12e\n",thetai,thetaj,k,k1,k2,k3,k4,delti[thetai]/k,delti[thetaj]/k,x[thetai]-delti[thetai]/k,x[thetaj]-delti[thetaj]/k, res,k1-k2-k3+k4,k1-k2,k3-k4);  
 #endif  
   }  
   return res;  
 }  
   
 /************** Inverse of matrix **************/    Revision 1.225  2016/07/12 08:40:03  brouard
 void ludcmp(double **a, int n, int *indx, double *d)     Summary: saving but not running
 {   
   int i,imax,j,k;     Revision 1.224  2016/07/01 13:16:01  brouard
   double big,dum,sum,temp;     Summary: Fixes
   double *vv;   
      Revision 1.223  2016/02/19 09:23:35  brouard
   vv=vector(1,n);     Summary: temporary
   *d=1.0;   
   for (i=1;i<=n;i++) {     Revision 1.222  2016/02/17 08:14:50  brouard
     big=0.0;     Summary: Probably last 0.98 stable version 0.98r6
     for (j=1;j<=n;j++)   
       if ((temp=fabs(a[i][j])) > big) big=temp;     Revision 1.221  2016/02/15 23:35:36  brouard
     if (big == 0.0) nrerror("Singular matrix in routine ludcmp");     Summary: minor bug
   
     Revision 1.219  2016/02/15 00:48:12  brouard
     *** empty log message ***
   
     Revision 1.218  2016/02/12 11:29:23  brouard
     Summary: 0.99 Back projections
   
     Revision 1.217  2015/12/23 17:18:31  brouard
     Summary: Experimental backcast
   
     Revision 1.216  2015/12/18 17:32:11  brouard
     Summary: 0.98r4 Warning and status=-2
   
     Version 0.98r4 is now:
      - displaying an error when status is -1, date of interview unknown and date of death known;
      - permitting a status -2 when the vital status is unknown at a known date of right truncation.
     Older changes concerning s=-2, dating from 2005 have been supersed.
   
     Revision 1.215  2015/12/16 08:52:24  brouard
     Summary: 0.98r4 working
   
     Revision 1.214  2015/12/16 06:57:54  brouard
     Summary: temporary not working
   
     Revision 1.213  2015/12/11 18:22:17  brouard
     Summary: 0.98r4
   
     Revision 1.212  2015/11/21 12:47:24  brouard
     Summary: minor typo
   
     Revision 1.211  2015/11/21 12:41:11  brouard
     Summary: 0.98r3 with some graph of projected cross-sectional
   
     Author: Nicolas Brouard
   
     Revision 1.210  2015/11/18 17:41:20  brouard
     Summary: Start working on projected prevalences  Revision 1.209  2015/11/17 22:12:03  brouard
     Summary: Adding ftolpl parameter
     Author: N Brouard
   
     We had difficulties to get smoothed confidence intervals. It was due
     to the period prevalence which wasn't computed accurately. The inner
     parameter ftolpl is now an outer parameter of the .imach parameter
     file after estepm. If ftolpl is small 1.e-4 and estepm too,
     computation are long.
   
     Revision 1.208  2015/11/17 14:31:57  brouard
     Summary: temporary
   
     Revision 1.207  2015/10/27 17:36:57  brouard
     *** empty log message ***
   
     Revision 1.206  2015/10/24 07:14:11  brouard
     *** empty log message ***
   
     Revision 1.205  2015/10/23 15:50:53  brouard
     Summary: 0.98r3 some clarification for graphs on likelihood contributions
   
     Revision 1.204  2015/10/01 16:20:26  brouard
     Summary: Some new graphs of contribution to likelihood
   
     Revision 1.203  2015/09/30 17:45:14  brouard
     Summary: looking at better estimation of the hessian
   
     Also a better criteria for convergence to the period prevalence And
     therefore adding the number of years needed to converge. (The
     prevalence in any alive state shold sum to one
   
     Revision 1.202  2015/09/22 19:45:16  brouard
     Summary: Adding some overall graph on contribution to likelihood. Might change
   
     Revision 1.201  2015/09/15 17:34:58  brouard
     Summary: 0.98r0
   
     - Some new graphs like suvival functions
     - Some bugs fixed like model=1+age+V2.
   
     Revision 1.200  2015/09/09 16:53:55  brouard
     Summary: Big bug thanks to Flavia
   
     Even model=1+age+V2. did not work anymore
   
     Revision 1.199  2015/09/07 14:09:23  brouard
     Summary: 0.98q6 changing default small png format for graph to vectorized svg.
   
     Revision 1.198  2015/09/03 07:14:39  brouard
     Summary: 0.98q5 Flavia
   
     Revision 1.197  2015/09/01 18:24:39  brouard
     *** empty log message ***
   
     Revision 1.196  2015/08/18 23:17:52  brouard
     Summary: 0.98q5
   
     Revision 1.195  2015/08/18 16:28:39  brouard
     Summary: Adding a hack for testing purpose
   
     After reading the title, ftol and model lines, if the comment line has
     a q, starting with #q, the answer at the end of the run is quit. It
     permits to run test files in batch with ctest. The former workaround was
     $ echo q | imach foo.imach
   
     Revision 1.194  2015/08/18 13:32:00  brouard
     Summary:  Adding error when the covariance matrix doesn't contain the exact number of lines required by the model line.
   
     Revision 1.193  2015/08/04 07:17:42  brouard
     Summary: 0.98q4
   
     Revision 1.192  2015/07/16 16:49:02  brouard
     Summary: Fixing some outputs
   
     Revision 1.191  2015/07/14 10:00:33  brouard
     Summary: Some fixes
   
     Revision 1.190  2015/05/05 08:51:13  brouard
     Summary: Adding digits in output parameters (7 digits instead of 6)
   
     Fix 1+age+.
   
     Revision 1.189  2015/04/30 14:45:16  brouard
     Summary: 0.98q2
   
     Revision 1.188  2015/04/30 08:27:53  brouard
     *** empty log message ***
   
     Revision 1.187  2015/04/29 09:11:15  brouard
     *** empty log message ***
   
     Revision 1.186  2015/04/23 12:01:52  brouard
     Summary: V1*age is working now, version 0.98q1
   
     Some codes had been disabled in order to simplify and Vn*age was
     working in the optimization phase, ie, giving correct MLE parameters,
     but, as usual, outputs were not correct and program core dumped.
   
     Revision 1.185  2015/03/11 13:26:42  brouard
     Summary: Inclusion of compile and links command line for Intel Compiler
   
     Revision 1.184  2015/03/11 11:52:39  brouard
     Summary: Back from Windows 8. Intel Compiler
   
     Revision 1.183  2015/03/10 20:34:32  brouard
     Summary: 0.98q0, trying with directest, mnbrak fixed
   
     We use directest instead of original Powell test; probably no
     incidence on the results, but better justifications;
     We fixed Numerical Recipes mnbrak routine which was wrong and gave
     wrong results.
   
     Revision 1.182  2015/02/12 08:19:57  brouard
     Summary: Trying to keep directest which seems simpler and more general
     Author: Nicolas Brouard
   
     Revision 1.181  2015/02/11 23:22:24  brouard
     Summary: Comments on Powell added
   
     Author:
   
     Revision 1.180  2015/02/11 17:33:45  brouard
     Summary: Finishing move from main to function (hpijx and prevalence_limit)
   
     Revision 1.179  2015/01/04 09:57:06  brouard
     Summary: back to OS/X
   
     Revision 1.178  2015/01/04 09:35:48  brouard
     *** empty log message ***
   
     Revision 1.177  2015/01/03 18:40:56  brouard
     Summary: Still testing ilc32 on OSX
   
     Revision 1.176  2015/01/03 16:45:04  brouard
     *** empty log message ***
   
     Revision 1.175  2015/01/03 16:33:42  brouard
     *** empty log message ***
   
     Revision 1.174  2015/01/03 16:15:49  brouard
     Summary: Still in cross-compilation
   
     Revision 1.173  2015/01/03 12:06:26  brouard
     Summary: trying to detect cross-compilation
   
     Revision 1.172  2014/12/27 12:07:47  brouard
     Summary: Back from Visual Studio and Intel, options for compiling for Windows XP
   
     Revision 1.171  2014/12/23 13:26:59  brouard
     Summary: Back from Visual C
   
     Still problem with utsname.h on Windows
   
     Revision 1.170  2014/12/23 11:17:12  brouard
     Summary: Cleaning some \%% back to %%
   
     The escape was mandatory for a specific compiler (which one?), but too many warnings.
   
     Revision 1.169  2014/12/22 23:08:31  brouard
     Summary: 0.98p
   
     Outputs some informations on compiler used, OS etc. Testing on different platforms.
   
     Revision 1.168  2014/12/22 15:17:42  brouard
     Summary: update
   
     Revision 1.167  2014/12/22 13:50:56  brouard
     Summary: Testing uname and compiler version and if compiled 32 or 64
   
     Testing on Linux 64
   
     Revision 1.166  2014/12/22 11:40:47  brouard
     *** empty log message ***
   
     Revision 1.165  2014/12/16 11:20:36  brouard
     Summary: After compiling on Visual C
   
     * imach.c (Module): Merging 1.61 to 1.162
   
     Revision 1.164  2014/12/16 10:52:11  brouard
     Summary: Merging with Visual C after suppressing some warnings for unused variables. Also fixing Saito's bug 0.98Xn
   
     * imach.c (Module): Merging 1.61 to 1.162
   
     Revision 1.163  2014/12/16 10:30:11  brouard
     * imach.c (Module): Merging 1.61 to 1.162
   
     Revision 1.162  2014/09/25 11:43:39  brouard
     Summary: temporary backup 0.99!
   
     Revision 1.1  2014/09/16 11:06:58  brouard
     Summary: With some code (wrong) for nlopt
   
     Author:
   
     Revision 1.161  2014/09/15 20:41:41  brouard
     Summary: Problem with macro SQR on Intel compiler
   
     Revision 1.160  2014/09/02 09:24:05  brouard
     *** empty log message ***
   
     Revision 1.159  2014/09/01 10:34:10  brouard
     Summary: WIN32
     Author: Brouard
   
     Revision 1.158  2014/08/27 17:11:51  brouard
     *** empty log message ***
   
     Revision 1.157  2014/08/27 16:26:55  brouard
     Summary: Preparing windows Visual studio version
     Author: Brouard
   
     In order to compile on Visual studio, time.h is now correct and time_t
     and tm struct should be used. difftime should be used but sometimes I
     just make the differences in raw time format (time(&now).
     Trying to suppress #ifdef LINUX
     Add xdg-open for __linux in order to open default browser.
   
     Revision 1.156  2014/08/25 20:10:10  brouard
     *** empty log message ***
   
     Revision 1.155  2014/08/25 18:32:34  brouard
     Summary: New compile, minor changes
     Author: Brouard
   
     Revision 1.154  2014/06/20 17:32:08  brouard
     Summary: Outputs now all graphs of convergence to period prevalence
   
     Revision 1.153  2014/06/20 16:45:46  brouard
     Summary: If 3 live state, convergence to period prevalence on same graph
     Author: Brouard
   
     Revision 1.152  2014/06/18 17:54:09  brouard
     Summary: open browser, use gnuplot on same dir than imach if not found in the path
   
     Revision 1.151  2014/06/18 16:43:30  brouard
     *** empty log message ***
   
     Revision 1.150  2014/06/18 16:42:35  brouard
     Summary: If gnuplot is not in the path try on same directory than imach binary (OSX)
     Author: brouard
   
     Revision 1.149  2014/06/18 15:51:14  brouard
     Summary: Some fixes in parameter files errors
     Author: Nicolas Brouard
   
     Revision 1.148  2014/06/17 17:38:48  brouard
     Summary: Nothing new
     Author: Brouard
   
     Just a new packaging for OS/X version 0.98nS
   
     Revision 1.147  2014/06/16 10:33:11  brouard
     *** empty log message ***
   
     Revision 1.146  2014/06/16 10:20:28  brouard
     Summary: Merge
     Author: Brouard
   
     Merge, before building revised version.
   
     Revision 1.145  2014/06/10 21:23:15  brouard
     Summary: Debugging with valgrind
     Author: Nicolas Brouard
   
     Lot of changes in order to output the results with some covariates
     After the Edimburgh REVES conference 2014, it seems mandatory to
     improve the code.
     No more memory valgrind error but a lot has to be done in order to
     continue the work of splitting the code into subroutines.
     Also, decodemodel has been improved. Tricode is still not
     optimal. nbcode should be improved. Documentation has been added in
     the source code.
   
     Revision 1.143  2014/01/26 09:45:38  brouard
     Summary: Version 0.98nR (to be improved, but gives same optimization results as 0.98k. Nice, promising
   
     * imach.c (Module): Trying to merge old staffs together while being at Tokyo. Not tested...
     (Module): Version 0.98nR Running ok, but output format still only works for three covariates.
   
     Revision 1.142  2014/01/26 03:57:36  brouard
     Summary: gnuplot changed plot w l 1 has to be changed to plot w l lt 2
   
     * imach.c (Module): Trying to merge old staffs together while being at Tokyo. Not tested...
   
     Revision 1.141  2014/01/26 02:42:01  brouard
     * imach.c (Module): Trying to merge old staffs together while being at Tokyo. Not tested...
   
     Revision 1.140  2011/09/02 10:37:54  brouard
     Summary: times.h is ok with mingw32 now.
   
     Revision 1.139  2010/06/14 07:50:17  brouard
     After the theft of my laptop, I probably lost some lines of codes which were not uploaded to the CVS tree.
     I remember having already fixed agemin agemax which are pointers now but not cvs saved.
   
     Revision 1.138  2010/04/30 18:19:40  brouard
     *** empty log message ***
   
     Revision 1.137  2010/04/29 18:11:38  brouard
     (Module): Checking covariates for more complex models
     than V1+V2. A lot of change to be done. Unstable.
   
     Revision 1.136  2010/04/26 20:30:53  brouard
     (Module): merging some libgsl code. Fixing computation
     of likelione (using inter/intrapolation if mle = 0) in order to
     get same likelihood as if mle=1.
     Some cleaning of code and comments added.
   
     Revision 1.135  2009/10/29 15:33:14  brouard
     (Module): Now imach stops if date of birth, at least year of birth, is not given. Some cleaning of the code.
   
     Revision 1.134  2009/10/29 13:18:53  brouard
     (Module): Now imach stops if date of birth, at least year of birth, is not given. Some cleaning of the code.
   
     Revision 1.133  2009/07/06 10:21:25  brouard
     just nforces
   
     Revision 1.132  2009/07/06 08:22:05  brouard
     Many tings
   
     Revision 1.131  2009/06/20 16:22:47  brouard
     Some dimensions resccaled
   
     Revision 1.130  2009/05/26 06:44:34  brouard
     (Module): Max Covariate is now set to 20 instead of 8. A
     lot of cleaning with variables initialized to 0. Trying to make
     V2+V3*age+V1+V4 strb=V3*age+V1+V4 working better.
   
     Revision 1.129  2007/08/31 13:49:27  lievre
     Modification of the way of exiting when the covariate is not binary in order to see on the window the error message before exiting
   
     Revision 1.128  2006/06/30 13:02:05  brouard
     (Module): Clarifications on computing e.j
   
     Revision 1.127  2006/04/28 18:11:50  brouard
     (Module): Yes the sum of survivors was wrong since
     imach-114 because nhstepm was no more computed in the age
     loop. Now we define nhstepma in the age loop.
     (Module): In order to speed up (in case of numerous covariates) we
     compute health expectancies (without variances) in a first step
     and then all the health expectancies with variances or standard
     deviation (needs data from the Hessian matrices) which slows the
     computation.
     In the future we should be able to stop the program is only health
     expectancies and graph are needed without standard deviations.
   
     Revision 1.126  2006/04/28 17:23:28  brouard
     (Module): Yes the sum of survivors was wrong since
     imach-114 because nhstepm was no more computed in the age
     loop. Now we define nhstepma in the age loop.
     Version 0.98h
   
     Revision 1.125  2006/04/04 15:20:31  lievre
     Errors in calculation of health expectancies. Age was not initialized.
     Forecasting file added.
   
     Revision 1.124  2006/03/22 17:13:53  lievre
     Parameters are printed with %lf instead of %f (more numbers after the comma).
     The log-likelihood is printed in the log file
   
     Revision 1.123  2006/03/20 10:52:43  brouard
     * imach.c (Module): <title> changed, corresponds to .htm file
     name. <head> headers where missing.
   
     * imach.c (Module): Weights can have a decimal point as for
     English (a comma might work with a correct LC_NUMERIC environment,
     otherwise the weight is truncated).
     Modification of warning when the covariates values are not 0 or
     1.
     Version 0.98g
   
     Revision 1.122  2006/03/20 09:45:41  brouard
     (Module): Weights can have a decimal point as for
     English (a comma might work with a correct LC_NUMERIC environment,
     otherwise the weight is truncated).
     Modification of warning when the covariates values are not 0 or
     1.
     Version 0.98g
   
     Revision 1.121  2006/03/16 17:45:01  lievre
     * imach.c (Module): Comments concerning covariates added
   
     * imach.c (Module): refinements in the computation of lli if
     status=-2 in order to have more reliable computation if stepm is
     not 1 month. Version 0.98f
   
     Revision 1.120  2006/03/16 15:10:38  lievre
     (Module): refinements in the computation of lli if
     status=-2 in order to have more reliable computation if stepm is
     not 1 month. Version 0.98f
   
     Revision 1.119  2006/03/15 17:42:26  brouard
     (Module): Bug if status = -2, the loglikelihood was
     computed as likelihood omitting the logarithm. Version O.98e
   
     Revision 1.118  2006/03/14 18:20:07  brouard
     (Module): varevsij Comments added explaining the second
     table of variances if popbased=1 .
     (Module): Covariances of eij, ekl added, graphs fixed, new html link.
     (Module): Function pstamp added
     (Module): Version 0.98d
   
     Revision 1.117  2006/03/14 17:16:22  brouard
     (Module): varevsij Comments added explaining the second
     table of variances if popbased=1 .
     (Module): Covariances of eij, ekl added, graphs fixed, new html link.
     (Module): Function pstamp added
     (Module): Version 0.98d
   
     Revision 1.116  2006/03/06 10:29:27  brouard
     (Module): Variance-covariance wrong links and
     varian-covariance of ej. is needed (Saito).
   
     Revision 1.115  2006/02/27 12:17:45  brouard
     (Module): One freematrix added in mlikeli! 0.98c
   
     Revision 1.114  2006/02/26 12:57:58  brouard
     (Module): Some improvements in processing parameter
     filename with strsep.
   
     Revision 1.113  2006/02/24 14:20:24  brouard
     (Module): Memory leaks checks with valgrind and:
     datafile was not closed, some imatrix were not freed and on matrix
     allocation too.
   
     Revision 1.112  2006/01/30 09:55:26  brouard
     (Module): Back to gnuplot.exe instead of wgnuplot.exe
   
     Revision 1.111  2006/01/25 20:38:18  brouard
     (Module): Lots of cleaning and bugs added (Gompertz)
     (Module): Comments can be added in data file. Missing date values
     can be a simple dot '.'.
   
     Revision 1.110  2006/01/25 00:51:50  brouard
     (Module): Lots of cleaning and bugs added (Gompertz)
   
     Revision 1.109  2006/01/24 19:37:15  brouard
     (Module): Comments (lines starting with a #) are allowed in data.
   
     Revision 1.108  2006/01/19 18:05:42  lievre
     Gnuplot problem appeared...
     To be fixed
   
     Revision 1.107  2006/01/19 16:20:37  brouard
     Test existence of gnuplot in imach path
   
     Revision 1.106  2006/01/19 13:24:36  brouard
     Some cleaning and links added in html output
   
     Revision 1.105  2006/01/05 20:23:19  lievre
     *** empty log message ***
   
     Revision 1.104  2005/09/30 16:11:43  lievre
     (Module): sump fixed, loop imx fixed, and simplifications.
     (Module): If the status is missing at the last wave but we know
     that the person is alive, then we can code his/her status as -2
     (instead of missing=-1 in earlier versions) and his/her
     contributions to the likelihood is 1 - Prob of dying from last
     health status (= 1-p13= p11+p12 in the easiest case of somebody in
     the healthy state at last known wave). Version is 0.98
   
     Revision 1.103  2005/09/30 15:54:49  lievre
     (Module): sump fixed, loop imx fixed, and simplifications.
   
     Revision 1.102  2004/09/15 17:31:30  brouard
     Add the possibility to read data file including tab characters.
   
     Revision 1.101  2004/09/15 10:38:38  brouard
     Fix on curr_time
   
     Revision 1.100  2004/07/12 18:29:06  brouard
     Add version for Mac OS X. Just define UNIX in Makefile
   
     Revision 1.99  2004/06/05 08:57:40  brouard
     *** empty log message ***
   
     Revision 1.98  2004/05/16 15:05:56  brouard
     New version 0.97 . First attempt to estimate force of mortality
     directly from the data i.e. without the need of knowing the health
     state at each age, but using a Gompertz model: log u =a + b*age .
     This is the basic analysis of mortality and should be done before any
     other analysis, in order to test if the mortality estimated from the
     cross-longitudinal survey is different from the mortality estimated
     from other sources like vital statistic data.
   
     The same imach parameter file can be used but the option for mle should be -3.
   
     AgneÌ€s, who wrote this part of the code, tried to keep most of the
     former routines in order to include the new code within the former code.
   
     The output is very simple: only an estimate of the intercept and of
     the slope with 95% confident intervals.
   
     Current limitations:
     A) Even if you enter covariates, i.e. with the
     model= V1+V2 equation for example, the programm does only estimate a unique global model without covariates.
     B) There is no computation of Life Expectancy nor Life Table.
   
     Revision 1.97  2004/02/20 13:25:42  lievre
     Version 0.96d. Population forecasting command line is (temporarily)
     suppressed.
   
     Revision 1.96  2003/07/15 15:38:55  brouard
     * imach.c (Repository): Errors in subdirf, 2, 3 while printing tmpout is
     rewritten within the same printf. Workaround: many printfs.
   
     Revision 1.95  2003/07/08 07:54:34  brouard
     * imach.c (Repository):
     (Repository): Using imachwizard code to output a more meaningful covariance
     matrix (cov(a12,c31) instead of numbers.
   
     Revision 1.94  2003/06/27 13:00:02  brouard
     Just cleaning
   
     Revision 1.93  2003/06/25 16:33:55  brouard
     (Module): On windows (cygwin) function asctime_r doesn't
     exist so I changed back to asctime which exists.
     (Module): Version 0.96b
   
     Revision 1.92  2003/06/25 16:30:45  brouard
     (Module): On windows (cygwin) function asctime_r doesn't
     exist so I changed back to asctime which exists.
   
     Revision 1.91  2003/06/25 15:30:29  brouard
     * imach.c (Repository): Duplicated warning errors corrected.
     (Repository): Elapsed time after each iteration is now output. It
     helps to forecast when convergence will be reached. Elapsed time
     is stamped in powell.  We created a new html file for the graphs
     concerning matrix of covariance. It has extension -cov.htm.
   
     Revision 1.90  2003/06/24 12:34:15  brouard
     (Module): Some bugs corrected for windows. Also, when
     mle=-1 a template is output in file "or"mypar.txt with the design
     of the covariance matrix to be input.
   
     Revision 1.89  2003/06/24 12:30:52  brouard
     (Module): Some bugs corrected for windows. Also, when
     mle=-1 a template is output in file "or"mypar.txt with the design
     of the covariance matrix to be input.
   
     Revision 1.88  2003/06/23 17:54:56  brouard
     * imach.c (Repository): Create a sub-directory where all the secondary files are. Only imach, htm, gp and r(imach) are on the main directory. Correct time and other things.
   
     Revision 1.87  2003/06/18 12:26:01  brouard
     Version 0.96
   
     Revision 1.86  2003/06/17 20:04:08  brouard
     (Module): Change position of html and gnuplot routines and added
     routine fileappend.
   
     Revision 1.85  2003/06/17 13:12:43  brouard
     * imach.c (Repository): Check when date of death was earlier that
     current date of interview. It may happen when the death was just
     prior to the death. In this case, dh was negative and likelihood
     was wrong (infinity). We still send an "Error" but patch by
     assuming that the date of death was just one stepm after the
     interview.
     (Repository): Because some people have very long ID (first column)
     we changed int to long in num[] and we added a new lvector for
     memory allocation. But we also truncated to 8 characters (left
     truncation)
     (Repository): No more line truncation errors.
   
     Revision 1.84  2003/06/13 21:44:43  brouard
     * imach.c (Repository): Replace "freqsummary" at a correct
     place. It differs from routine "prevalence" which may be called
     many times. Probs is memory consuming and must be used with
     parcimony.
     Version 0.95a3 (should output exactly the same maximization than 0.8a2)
   
     Revision 1.83  2003/06/10 13:39:11  lievre
     *** empty log message ***
   
     Revision 1.82  2003/06/05 15:57:20  brouard
     Add log in  imach.c and  fullversion number is now printed.
   
   */
   /*
      Interpolated Markov Chain
   
     Short summary of the programme:
     
     This program computes Healthy Life Expectancies or State-specific
     (if states aren't health statuses) Expectancies from
     cross-longitudinal data. Cross-longitudinal data consist in: 
   
     -1- a first survey ("cross") where individuals from different ages
     are interviewed on their health status or degree of disability (in
     the case of a health survey which is our main interest)
   
     -2- at least a second wave of interviews ("longitudinal") which
     measure each change (if any) in individual health status.  Health
     expectancies are computed from the time spent in each health state
     according to a model. More health states you consider, more time is
     necessary to reach the Maximum Likelihood of the parameters involved
     in the model.  The simplest model is the multinomial logistic model
     where pij is the probability to be observed in state j at the second
     wave conditional to be observed in state i at the first
     wave. Therefore the model is: log(pij/pii)= aij + bij*age+ cij*sex +
     etc , where 'age' is age and 'sex' is a covariate. If you want to
     have a more complex model than "constant and age", you should modify
     the program where the markup *Covariates have to be included here
     again* invites you to do it.  More covariates you add, slower the
     convergence.
   
     The advantage of this computer programme, compared to a simple
     multinomial logistic model, is clear when the delay between waves is not
     identical for each individual. Also, if a individual missed an
     intermediate interview, the information is lost, but taken into
     account using an interpolation or extrapolation.  
   
     hPijx is the probability to be observed in state i at age x+h
     conditional to the observed state i at age x. The delay 'h' can be
     split into an exact number (nh*stepm) of unobserved intermediate
     states. This elementary transition (by month, quarter,
     semester or year) is modelled as a multinomial logistic.  The hPx
     matrix is simply the matrix product of nh*stepm elementary matrices
     and the contribution of each individual to the likelihood is simply
     hPijx.
   
     Also this programme outputs the covariance matrix of the parameters but also
     of the life expectancies. It also computes the period (stable) prevalence.
   
   Back prevalence and projections:
   
    - back_prevalence_limit(double *p, double **bprlim, double ageminpar,
      double agemaxpar, double ftolpl, int *ncvyearp, double
      dateprev1,double dateprev2, int firstpass, int lastpass, int
      mobilavproj)
   
       Computes the back prevalence limit for any combination of
       covariate values k at any age between ageminpar and agemaxpar and
       returns it in **bprlim. In the loops,
   
      - **bprevalim(**bprlim, ***mobaverage, nlstate, *p, age, **oldm,
          **savm, **dnewm, **doldm, **dsavm, ftolpl, ncvyearp, k);
   
      - hBijx Back Probability to be in state i at age x-h being in j at x
      Computes for any combination of covariates k and any age between bage and fage 
      p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
                           oldm=oldms;savm=savms;
   
      - hbxij(p3mat,nhstepm,agedeb,hstepm,p,nlstate,stepm,oldm,savm, k, nres);
        Computes the transition matrix starting at age 'age' over
        'nhstepm*hstepm*stepm' months (i.e. until
        age (in years)  age+nhstepm*hstepm*stepm/12) by multiplying
        nhstepm*hstepm matrices. 
   
        Returns p3mat[i][j][h] after calling
        p3mat[i][j][h]=matprod2(newm,
        bmij(pmmij,cov,ncovmodel,x,nlstate,prevacurrent, dnewm, doldm,
        dsavm,ij),\ 1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath,
        oldm);
   
   Important routines
   
   - func (or funcone), computes logit (pij) distinguishing
     o fixed variables (single or product dummies or quantitative);
     o varying variables by:
      (1) wave (single, product dummies, quantitative), 
      (2) by age (can be month) age (done), age*age (done), age*Vn where Vn can be:
          % fixed dummy (treated) or quantitative (not done because time-consuming);
          % varying dummy (not done) or quantitative (not done);
   - Tricode which tests the modality of dummy variables (in order to warn with wrong or empty modalities)
     and returns the number of efficient covariates cptcoveff and modalities nbcode[Tvar[k]][1]= 0 and nbcode[Tvar[k]][2]= 1 usually.
   - printinghtml which outputs results like life expectancy in and from a state for a combination of modalities of dummy variables
     o There are 2**cptcoveff combinations of (0,1) for cptcoveff variables. Outputting only combinations with people, eÌliminating 1 1 if
       race White (0 0), Black vs White (1 0), Hispanic (0 1) and 1 1 being meaningless.
   
   
     
     Authors: Nicolas Brouard (brouard@ined.fr) and AgneÌ€s LieÌ€vre (lievre@ined.fr).
              Institut national d'eÌtudes deÌmographiques, Paris.
     This software have been partly granted by Euro-REVES, a concerted action
     from the European Union.
     It is copyrighted identically to a GNU software product, ie programme and
     software can be distributed freely for non commercial use. Latest version
     can be accessed at http://euroreves.ined.fr/imach .
   
     Help to debug: LD_PRELOAD=/usr/local/lib/libnjamd.so ./imach foo.imach
     or better on gdb : set env LD_PRELOAD=/usr/local/lib/libnjamd.so
     
     **********************************************************************/
   /*
     main
     read parameterfile
     read datafile
     concatwav
     freqsummary
     if (mle >= 1)
       mlikeli
     print results files
     if mle==1 
        computes hessian
     read end of parameter file: agemin, agemax, bage, fage, estepm
         begin-prev-date,...
     open gnuplot file
     open html file
     period (stable) prevalence      | pl_nom    1-1 2-2 etc by covariate
      for age prevalim()             | #****** V1=0  V2=1  V3=1  V4=0 ******
                                     | 65 1 0 2 1 3 1 4 0  0.96326 0.03674
       freexexit2 possible for memory heap.
   
     h Pij x                         | pij_nom  ficrestpij
      # Cov Agex agex+h hpijx with i,j= 1-1 1-2     1-3     2-1     2-2     2-3
          1  85   85    1.00000             0.00000 0.00000 0.00000 1.00000 0.00000
          1  85   86    0.68299             0.22291 0.09410 0.71093 0.00000 0.28907
   
          1  65   99    0.00364             0.00322 0.99314 0.00350 0.00310 0.99340
          1  65  100    0.00214             0.00204 0.99581 0.00206 0.00196 0.99597
     variance of p one-step probabilities varprob  | prob_nom   ficresprob #One-step probabilities and stand. devi in ()
      Standard deviation of one-step probabilities | probcor_nom   ficresprobcor #One-step probabilities and correlation matrix
      Matrix of variance covariance of one-step probabilities |  probcov_nom ficresprobcov #One-step probabilities and covariance matrix
   
     forecasting if prevfcast==1 prevforecast call prevalence()
     health expectancies
     Variance-covariance of DFLE
     prevalence()
      movingaverage()
     varevsij() 
     if popbased==1 varevsij(,popbased)
     total life expectancies
     Variance of period (stable) prevalence
    end
   */
   
   /* #define DEBUG */
   /* #define DEBUGBRENT */
   /* #define DEBUGLINMIN */
   /* #define DEBUGHESS */
   #define DEBUGHESSIJ
   /* #define LINMINORIGINAL  /\* Don't use loop on scale in linmin (accepting nan) *\/ */
   #define POWELL /* Instead of NLOPT */
   #define POWELLNOF3INFF1TEST /* Skip test */
   /* #define POWELLORIGINAL /\* Don't use Directest to decide new direction but original Powell test *\/ */
   /* #define MNBRAKORIGINAL /\* Don't use mnbrak fix *\/ */
   /* #define FLATSUP  *//* Suppresses directions where likelihood is flat */
   
   #include <math.h>
   #include <stdio.h>
   #include <stdlib.h>
   #include <string.h>
   #include <ctype.h>
   
   #ifdef _WIN32
   #include <io.h>
   #include <windows.h>
   #include <tchar.h>
   #else
   #include <unistd.h>
   #endif
   
   #include <limits.h>
   #include <sys/types.h>
   
   #if defined(__GNUC__)
   #include <sys/utsname.h> /* Doesn't work on Windows */
   #endif
   
   #include <sys/stat.h>
   #include <errno.h>
   /* extern int errno; */
   
   /* #ifdef LINUX */
   /* #include <time.h> */
   /* #include "timeval.h" */
   /* #else */
   /* #include <sys/time.h> */
   /* #endif */
   
   #include <time.h>
   
   #ifdef GSL
   #include <gsl/gsl_errno.h>
   #include <gsl/gsl_multimin.h>
   #endif
   
   
   #ifdef NLOPT
   #include <nlopt.h>
   typedef struct {
     double (* function)(double [] );
   } myfunc_data ;
   #endif
   
   /* #include <libintl.h> */
   /* #define _(String) gettext (String) */
   
   #define MAXLINE 2048 /* Was 256 and 1024. Overflow with 312 with 2 states and 4 covariates. Should be ok */
   
   #define GNUPLOTPROGRAM "gnuplot"
   /*#define GNUPLOTPROGRAM "..\\gp37mgw\\wgnuplot"*/
   #define FILENAMELENGTH 256
   
   #define GLOCK_ERROR_NOPATH              -1      /* empty path */
   #define GLOCK_ERROR_GETCWD              -2      /* cannot get cwd */
   
   #define MAXPARM 128 /**< Maximum number of parameters for the optimization */
   #define NPARMAX 64 /**< (nlstate+ndeath-1)*nlstate*ncovmodel */
   
   #define NINTERVMAX 8
   #define NLSTATEMAX 8 /**< Maximum number of live states (for func) */
   #define NDEATHMAX 8 /**< Maximum number of dead states (for func) */
   #define NCOVMAX 30  /**< Maximum number of covariates used in the model, including generated covariates V1*V2 or V1*age */
   #define codtabm(h,k)  (1 & (h-1) >> (k-1))+1
   /*#define decodtabm(h,k,cptcoveff)= (h <= (1<<cptcoveff)?(((h-1) >> (k-1)) & 1) +1 : -1)*/
   #define decodtabm(h,k,cptcoveff) (((h-1) >> (k-1)) & 1) +1 
   /*#define MAXN 20000 */ /* Should by replaced by nobs, real number of observations and unlimited */
   #define YEARM 12. /**< Number of months per year */
   /* #define AGESUP 130 */
   /* #define AGESUP 150 */
   #define AGESUP 200
   #define AGEINF 0
   #define AGEMARGE 25 /* Marge for agemin and agemax for(iage=agemin-AGEMARGE; iage <= agemax+3+AGEMARGE; iage++) */
   #define AGEBASE 40
   #define AGEOVERFLOW 1.e20
   #define AGEGOMP 10 /**< Minimal age for Gompertz adjustment */
   #ifdef _WIN32
   #define DIRSEPARATOR '\\'
   #define CHARSEPARATOR "\\"
   #define ODIRSEPARATOR '/'
   #else
   #define DIRSEPARATOR '/'
   #define CHARSEPARATOR "/"
   #define ODIRSEPARATOR '\\'
   #endif
   
   /* $Id$ */
   /* $State$ */
   #include "version.h"
   char version[]=__IMACH_VERSION__;
   char copyright[]="August 2022,INED-EUROREVES-Institut de longevite-Japan Society for the Promotion of Science (Grant-in-Aid for Scientific Research 25293121), Intel Software 2015-2020, Nihon University 2021-202, INED 2000-2022";
   char fullversion[]="$Revision$ $Date$"; 
   char strstart[80];
   char optionfilext[10], optionfilefiname[FILENAMELENGTH];
   int erreur=0, nberr=0, nbwarn=0; /* Error number, number of errors number of warnings  */
   int nagesqr=0, nforce=0; /* nagesqr=1 if model is including age*age, number of forces */
   /* Number of covariates model (1)=V2+V1+ V3*age+V2*V4 */
   /* Model(2)  V1 + V2 + V3 + V8 + V7*V8 + V5*V6 + V8*age + V3*age + age*age */
   int cptcovn=0; /**< cptcovn decodemodel: number of covariates k of the models excluding age*products =6 and age*age but including products */
   int cptcovt=0; /**< cptcovt: total number of covariates of the model (2) nbocc(+)+1 = 8 excepting constant and age and age*age */
   int cptcovs=0; /**< cptcovs number of SIMPLE covariates in the model V2+V1 =2 (dummy or quantit or time varying) */
   int cptcovsnq=0; /**< cptcovsnq number of SIMPLE covariates in the model but non quantitative V2+V1 =2 */
   int cptcovage=0; /**< Number of covariates with age: V3*age only =1 */
   int cptcovprodnoage=0; /**< Number of covariate products without age */   
   int cptcoveff=0; /* Total number of single dummy covariates (fixed or time varying) to vary for printing results (2**cptcoveff combinations of dummies)(computed in tricode as cptcov) */
   int ncovf=0; /* Total number of effective fixed covariates (dummy or quantitative) in the model */
   int ncovv=0; /* Total number of effective (wave) varying covariates (dummy or quantitative) in the model */
   int ncova=0; /* Total number of effective (wave and stepm) varying with age covariates (dummy of quantitative) in the model */
   int nsd=0; /**< Total number of single dummy variables (output) */
   int nsq=0; /**< Total number of single quantitative variables (output) */
   int ncoveff=0; /* Total number of effective fixed dummy covariates in the model */
   int nqfveff=0; /**< nqfveff Number of Quantitative Fixed Variables Effective */
   int ntveff=0; /**< ntveff number of effective time varying variables */
   int nqtveff=0; /**< ntqveff number of effective time varying quantitative variables */
   int cptcov=0; /* Working variable */
   int firstobs=1, lastobs=10; /* nobs = lastobs-firstobs+1 declared globally ;*/
   int nobs=10;  /* Number of observations in the data lastobs-firstobs */
   int ncovcombmax=NCOVMAX; /* Maximum calculated number of covariate combination = pow(2, cptcoveff) */
   int npar=NPARMAX; /* Number of parameters (nlstate+ndeath-1)*nlstate*ncovmodel; */
   int nlstate=2; /* Number of live states */
   int ndeath=1; /* Number of dead states */
   int ncovmodel=0, ncovcol=0;     /* Total number of covariables including constant a12*1 +b12*x ncovmodel=2 */
   int  nqv=0, ntv=0, nqtv=0;    /* Total number of quantitative variables, time variable (dummy), quantitative and time variable */ 
   int popbased=0;
   
   int *wav; /* Number of waves for this individuual 0 is possible */
   int maxwav=0; /* Maxim number of waves */
   int jmin=0, jmax=0; /* min, max spacing between 2 waves */
   int ijmin=0, ijmax=0; /* Individuals having jmin and jmax */ 
   int gipmx=0, gsw=0; /* Global variables on the number of contributions 
                      to the likelihood and the sum of weights (done by funcone)*/
   int mle=1, weightopt=0;
   int **mw; /* mw[mi][i] is number of the mi wave for this individual */
   int **dh; /* dh[mi][i] is number of steps between mi,mi+1 for this individual */
   int **bh; /* bh[mi][i] is the bias (+ or -) for this individual if the delay between
              * wave mi and wave mi+1 is not an exact multiple of stepm. */
   int countcallfunc=0;  /* Count the number of calls to func */
   int selected(int kvar); /* Is covariate kvar selected for printing results */
   
   double jmean=1; /* Mean space between 2 waves */
   double **matprod2(); /* test */
   double **oldm, **newm, **savm; /* Working pointers to matrices */
   double **oldms, **newms, **savms; /* Fixed working pointers to matrices */
   double   **ddnewms, **ddoldms, **ddsavms; /* for freeing later */
   
   /*FILE *fic ; */ /* Used in readdata only */
   FILE *ficpar, *ficparo,*ficres, *ficresp, *ficresphtm, *ficresphtmfr, *ficrespl, *ficresplb,*ficrespij, *ficrespijb, *ficrest,*ficresf, *ficresfb,*ficrespop;
   FILE *ficlog, *ficrespow;
   int globpr=0; /* Global variable for printing or not */
   double fretone; /* Only one call to likelihood */
   long ipmx=0; /* Number of contributions */
   double sw; /* Sum of weights */
   char filerespow[FILENAMELENGTH];
   char fileresilk[FILENAMELENGTH]; /* File of individual contributions to the likelihood */
   FILE *ficresilk;
   FILE *ficgp,*ficresprob,*ficpop, *ficresprobcov, *ficresprobcor;
   FILE *ficresprobmorprev;
   FILE *fichtm, *fichtmcov; /* Html File */
   FILE *ficreseij;
   char filerese[FILENAMELENGTH];
   FILE *ficresstdeij;
   char fileresstde[FILENAMELENGTH];
   FILE *ficrescveij;
   char filerescve[FILENAMELENGTH];
   FILE  *ficresvij;
   char fileresv[FILENAMELENGTH];
   
   char title[MAXLINE];
   char model[MAXLINE]; /**< The model line */
   char optionfile[FILENAMELENGTH], datafile[FILENAMELENGTH],  filerespl[FILENAMELENGTH],  fileresplb[FILENAMELENGTH];
   char plotcmd[FILENAMELENGTH], pplotcmd[FILENAMELENGTH];
   char tmpout[FILENAMELENGTH],  tmpout2[FILENAMELENGTH]; 
   char command[FILENAMELENGTH];
   int  outcmd=0;
   
   char fileres[FILENAMELENGTH], filerespij[FILENAMELENGTH], filerespijb[FILENAMELENGTH], filereso[FILENAMELENGTH], rfileres[FILENAMELENGTH];
   char fileresu[FILENAMELENGTH]; /* fileres without r in front */
   char filelog[FILENAMELENGTH]; /* Log file */
   char filerest[FILENAMELENGTH];
   char fileregp[FILENAMELENGTH];
   char popfile[FILENAMELENGTH];
   
   char optionfilegnuplot[FILENAMELENGTH], optionfilehtm[FILENAMELENGTH], optionfilehtmcov[FILENAMELENGTH] ;
   
   /* struct timeval start_time, end_time, curr_time, last_time, forecast_time; */
   /* struct timezone tzp; */
   /* extern int gettimeofday(); */
   struct tm tml, *gmtime(), *localtime();
   
   extern time_t time();
   
   struct tm start_time, end_time, curr_time, last_time, forecast_time;
   time_t  rstart_time, rend_time, rcurr_time, rlast_time, rforecast_time; /* raw time */
   struct tm tm;
   
   char strcurr[80], strfor[80];
   
   char *endptr;
   long lval;
   double dval;
   
   #define NR_END 1
   #define FREE_ARG char*
   #define FTOL 1.0e-10
   
   #define NRANSI 
   #define ITMAX 200
   #define ITPOWMAX 20 /* This is now multiplied by the number of parameters */ 
   
   #define TOL 2.0e-4 
   
   #define CGOLD 0.3819660 
   #define ZEPS 1.0e-10 
   #define SHFT(a,b,c,d) (a)=(b);(b)=(c);(c)=(d); 
   
   #define GOLD 1.618034 
   #define GLIMIT 100.0 
   #define TINY 1.0e-20 
   
   static double maxarg1,maxarg2;
   #define FMAX(a,b) (maxarg1=(a),maxarg2=(b),(maxarg1)>(maxarg2)? (maxarg1):(maxarg2))
   #define FMIN(a,b) (maxarg1=(a),maxarg2=(b),(maxarg1)<(maxarg2)? (maxarg1):(maxarg2))
     
   #define SIGN(a,b) ((b)>0.0 ? fabs(a) : -fabs(a))
   #define rint(a) floor(a+0.5)
   /* http://www.thphys.uni-heidelberg.de/~robbers/cmbeasy/doc/html/myutils_8h-source.html */
   #define mytinydouble 1.0e-16
   /* #define DEQUAL(a,b) (fabs((a)-(b))<mytinydouble) */
   /* http://www.thphys.uni-heidelberg.de/~robbers/cmbeasy/doc/html/mynrutils_8h-source.html */
   /* static double dsqrarg; */
   /* #define DSQR(a) (DEQUAL((dsqrarg=(a)),0.0) ? 0.0 : dsqrarg*dsqrarg) */
   static double sqrarg;
   #define SQR(a) ((sqrarg=(a)) == 0.0 ? 0.0 :sqrarg*sqrarg)
   #define SWAP(a,b) {temp=(a);(a)=(b);(b)=temp;} 
   int agegomp= AGEGOMP;
   
   int imx; 
   int stepm=1;
   /* Stepm, step in month: minimum step interpolation*/
   
   int estepm;
   /* Estepm, step in month to interpolate survival function in order to approximate Life Expectancy*/
   
   int m,nb;
   long *num;
   int firstpass=0, lastpass=4,*cod, *cens;
   int *ncodemax;  /* ncodemax[j]= Number of modalities of the j th
                      covariate for which somebody answered excluding 
                      undefined. Usually 2: 0 and 1. */
   int *ncodemaxwundef;  /* ncodemax[j]= Number of modalities of the j th
                                covariate for which somebody answered including 
                                undefined. Usually 3: -1, 0 and 1. */
   double **agev,*moisnais, *annais, *moisdc, *andc,**mint, **anint;
   double **pmmij, ***probs; /* Global pointer */
   double ***mobaverage, ***mobaverages; /* New global variable */
   double **precov; /* New global variable to store for each resultline, values of model covariates given by the resultlines (in order to speed up)  */
   double *ageexmed,*agecens;
   double dateintmean=0;
     double anprojd, mprojd, jprojd; /* For eventual projections */
     double anprojf, mprojf, jprojf;
   
     double anbackd, mbackd, jbackd; /* For eventual backprojections */
     double anbackf, mbackf, jbackf;
     double jintmean,mintmean,aintmean;  
   double *weight;
   int **s; /* Status */
   double *agedc;
   double  **covar; /**< covar[j,i], value of jth covariate for individual i,
                     * covar=matrix(0,NCOVMAX,1,n); 
                     * cov[Tage[kk]+2]=covar[Tvar[Tage[kk]]][i]*age; */
   double **coqvar; /* Fixed quantitative covariate nqv */
   double ***cotvar; /* Time varying covariate ntv */
   double ***cotqvar; /* Time varying quantitative covariate itqv */
   double  idx; 
   int **nbcode, *Tvar; /**< model=V2 => Tvar[1]= 2 */
   /* Some documentation */
         /*   Design original data
          *  V1   V2   V3   V4  V5  V6  V7  V8  Weight ddb ddth d1st s1 V9 V10 V11 V12 s2 V9 V10 V11 V12 
          *  <          ncovcol=6   >   nqv=2 (V7 V8)                   dv dv  dv  qtv    dv dv  dvv qtv
          *                                                             ntv=3     nqtv=1
          *  cptcovn number of covariates (not including constant and age or age*age) = number of plus sign + 1 = 10+1=11
          * For time varying covariate, quanti or dummies
          *       cotqvar[wav][iv(1 to nqtv)][i]= [1][12][i]=(V12) quanti
          *       cotvar[wav][ntv+iv][i]= [3+(1 to nqtv)][i]=(V12) quanti
          *       cotvar[wav][iv(1 to ntv)][i]= [1][1][i]=(V9) dummies at wav 1
          *       cotvar[wav][iv(1 to ntv)][i]= [1][2][i]=(V10) dummies at wav 1
          *       covar[Vk,i], value of the Vkth fixed covariate dummy or quanti for individual i:
          *       covar[1][i]= (V1), covar[4][i]=(V4), covar[8][i]=(V8)
          * Model V2 + V1 + V3*age + V3 + V5*V6 + V7*V8 + V8*age + V8 + V9 + V9*age + V10
          *   k=  1    2      3       4     5       6      7        8   9     10       11 
          */
   /* According to the model, more columns can be added to covar by the product of covariates */
   /* ncovcol=1(Males=0 Females=1) nqv=1(raedyrs) ntv=2(withoutiadl=0 withiadl=1, witoutadl=0 withoutadl=1) nqtv=1(bmi) nlstate=3 ndeath=1
     # States 1=Coresidence, 2 Living alone, 3 Institution
     # V1=sex, V2=raedyrs Quant Fixed, State=livarnb4..livarnb11, V3=iadl4..iald11, V4=adlw4..adlw11, V5=r4bmi..r11bmi
   */
   /*             V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   /*    k        1  2   3   4     5    6    7     8    9 */
   /*Typevar[k]=  0  0   0   2     1    0    2     1    0 *//*0 for simple covariate (dummy, quantitative,*/
                                                            /* fixed or varying), 1 for age product, 2 for*/
                                                            /* product */
   /*Dummy[k]=    1  0   0   1     3    1    1     2    0 *//*Dummy[k] 0=dummy (0 1), 1 quantitative */
                                                            /*(single or product without age), 2 dummy*/
                                                            /* with age product, 3 quant with age product*/
   /*Tvar[k]=     5  4   3   6     5    2    7     1    1 */
   /*    nsd         1   2                              3 */ /* Counting single dummies covar fixed or tv */
   /*TnsdVar[Tvar]   1   2                              3 */ 
   /*TvarsD[nsd]     4   3                              1 */ /* ID of single dummy cova fixed or timevary*/
   /*TvarsDind[k]    2   3                              9 */ /* position K of single dummy cova */
   /*    nsq      1                     2                 */ /* Counting single quantit tv */
   /* TvarsQ[k]   5                     2                 */ /* Number of single quantitative cova */
   /* TvarsQind   1                     6                 */ /* position K of single quantitative cova */
   /* Tprod[i]=k             1               2            */ /* Position in model of the ith prod without age */
   /* cptcovage                    1               2      */ /* Counting cov*age in the model equation */
   /* Tage[cptcovage]=k            5               8      */ /* Position in the model of ith cov*age */
   /* Tvard[1][1]@4={4,3,1,2}    V4*V3 V1*V2              */ /* Position in model of the ith prod without age */
   /* Tvardk[4][1]=4;Tvardk[4][2]=3;Tvardk[7][1]=1;Tvardk[7][2]=2 */ /* Variables of a prod at position in the model equation*/
   /* TvarF TvarF[1]=Tvar[6]=2,  TvarF[2]=Tvar[7]=7, TvarF[3]=Tvar[9]=1  ID of fixed covariates or product V2, V1*V2, V1 */
   /* TvarFind;  TvarFind[1]=6,  TvarFind[2]=7, TvarFind[3]=9 *//* Inverse V2(6) is first fixed (single or prod)  */
   /* Type                    */
   /* V         1  2  3  4  5 */
   /*           F  F  V  V  V */
   /*           D  Q  D  D  Q */
   /*                         */
   int *TvarsD;
   int *TnsdVar;
   int *TvarsDind;
   int *TvarsQ;
   int *TvarsQind;
   
   #define MAXRESULTLINESPONE 10+1
   int nresult=0;
   int parameterline=0; /* # of the parameter (type) line */
   int TKresult[MAXRESULTLINESPONE]; /* TKresult[nres]=k for each resultline nres give the corresponding combination of dummies */
   int resultmodel[MAXRESULTLINESPONE][NCOVMAX];/* resultmodel[k1]=k3: k1th position in the model corresponds to the k3 position in the resultline */
   int modelresult[MAXRESULTLINESPONE][NCOVMAX];/* modelresult[k3]=k1: k1th position in the model corresponds to the k3 position in the resultline */
   int Tresult[MAXRESULTLINESPONE][NCOVMAX];/* Tresult[nres][result_position]= value of the dummy variable at the result_position in the nres resultline */
   int Tinvresult[MAXRESULTLINESPONE][NCOVMAX];/* Tinvresult[nres][Name of a dummy variable]= value of the variable in the result line  */
   double TinvDoQresult[MAXRESULTLINESPONE][NCOVMAX];/* TinvDoQresult[nres][Name of a Dummy or Q variable]= value of the variable in the result line */
   int Tvresult[MAXRESULTLINESPONE][NCOVMAX]; /* Tvresult[nres][result_position]= name of the dummy variable at the result_position in the nres resultline */
   double Tqresult[MAXRESULTLINESPONE][NCOVMAX]; /* Tqresult[nres][result_position]= value of the variable at the result_position in the nres resultline */
   double Tqinvresult[MAXRESULTLINESPONE][NCOVMAX]; /* For quantitative variable , value (output) */
   int Tvqresult[MAXRESULTLINESPONE][NCOVMAX]; /* Tvqresult[nres][result_position]= id of the variable at the result_position in the nres resultline */
   
   /* ncovcol=1(Males=0 Females=1) nqv=1(raedyrs) ntv=2(withoutiadl=0 withiadl=1, witoutadl=0 withoutadl=1) nqtv=1(bmi) nlstate=3 ndeath=1
     # States 1=Coresidence, 2 Living alone, 3 Institution
     # V1=sex, V2=raedyrs Quant Fixed, State=livarnb4..livarnb11, V3=iadl4..iald11, V4=adlw4..adlw11, V5=r4bmi..r11bmi
   */
   /* int *TDvar; /\**< TDvar[1]=4,  TDvarF[2]=3, TDvar[3]=6  in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 *\/ */
   int *TvarF; /**< TvarF[1]=Tvar[6]=2,  TvarF[2]=Tvar[7]=7, TvarF[3]=Tvar[9]=1  in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   int *TvarFind; /**< TvarFind[1]=6,  TvarFind[2]=7, Tvarind[3]=9  in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   int *TvarV; /**< TvarV[1]=Tvar[1]=5, TvarV[2]=Tvar[2]=4  in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   int *TvarVind; /**< TvarVind[1]=1, TvarVind[2]=2  in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   int *TvarA; /**< TvarA[1]=Tvar[5]=5, TvarA[2]=Tvar[8]=1  in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   int *TvarAind; /**< TvarindA[1]=5, TvarAind[2]=8  in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   int *TvarFD; /**< TvarFD[1]=V1 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   int *TvarFDind; /* TvarFDind[1]=9 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
   int *TvarFQ; /* TvarFQ[1]=V2 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple fixed quantitative variable */
   int *TvarFQind; /* TvarFQind[1]=6 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple fixed quantitative variable */
   int *TvarVD; /* TvarVD[1]=V5 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple fixed quantitative variable */
   int *TvarVDind; /* TvarVDind[1]=1 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple fixed quantitative variable */
   int *TvarVQ; /* TvarVQ[1]=V5 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple time varying quantitative variable */
   int *TvarVQind; /* TvarVQind[1]=1 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple time varying quantitative variable */
   
   int *Tvarsel; /**< Selected covariates for output */
   double *Tvalsel; /**< Selected modality value of covariate for output */
   int *Typevar; /**< 0 for simple covariate (dummy, quantitative, fixed or varying), 1 for age product, 2 for  product */
   int *Fixed; /** Fixed[k] 0=fixed, 1 varying, 2 fixed with age product, 3 varying with age product */ 
   int *Dummy; /** Dummy[k] 0=dummy (0 1), 1 quantitative (single or product without age), 2 dummy with age product, 3 quant with age product */ 
   int *DummyV; /** Dummy[v] 0=dummy (0 1), 1 quantitative */
   int *FixedV; /** FixedV[v] 0 fixed, 1 varying */
   int *Tage;
   int anyvaryingduminmodel=0; /**< Any varying dummy in Model=1 yes, 0 no, to avoid a loop on waves in freq */ 
   int *Tmodelind; /** Tmodelind[Tvaraff[3]]=9 for V1 position,Tvaraff[1]@9={4, 3, 1, 0, 0, 0, 0, 0, 0}, model=V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1*/
   int *TmodelInvind; /** Tmodelind[Tvaraff[3]]=9 for V1 position,Tvaraff[1]@9={4, 3, 1, 0, 0, 0, 0, 0, 0}, model=V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1*/ 
   int *TmodelInvQind; /** Tmodelqind[1]=1 for V5(quantitative varying) position,Tvaraff[1]@9={4, 3, 1, 0, 0, 0, 0, 0, 0}, model=V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1  */
   int *Ndum; /** Freq of modality (tricode */
   /* int **codtab;*/ /**< codtab=imatrix(1,100,1,10); */
   int **Tvard;
   int **Tvardk;
   int *Tprod;/**< Gives the k position of the k1 product */
   /* Tprod[k1=1]=3(=V1*V4) for V2+V1+V1*V4+age*V3  */
   int *Tposprod; /**< Gives the k1 product from the k position */
      /* if  V2+V1+V1*V4+age*V3+V3*V2   TProd[k1=2]=5 (V3*V2) */
      /* Tposprod[k]=k1 , Tposprod[3]=1, Tposprod[5(V3*V2)]=2 (2nd product without age) */
   int cptcovprod, *Tvaraff, *invalidvarcomb;
   double *lsurv, *lpop, *tpop;
   
   #define FD 1; /* Fixed dummy covariate */
   #define FQ 2; /* Fixed quantitative covariate */
   #define FP 3; /* Fixed product covariate */
   #define FPDD 7; /* Fixed product dummy*dummy covariate */
   #define FPDQ 8; /* Fixed product dummy*quantitative covariate */
   #define FPQQ 9; /* Fixed product quantitative*quantitative covariate */
   #define VD 10; /* Varying dummy covariate */
   #define VQ 11; /* Varying quantitative covariate */
   #define VP 12; /* Varying product covariate */
   #define VPDD 13; /* Varying product dummy*dummy covariate */
   #define VPDQ 14; /* Varying product dummy*quantitative covariate */
   #define VPQQ 15; /* Varying product quantitative*quantitative covariate */
   #define APFD 16; /* Age product * fixed dummy covariate */
   #define APFQ 17; /* Age product * fixed quantitative covariate */
   #define APVD 18; /* Age product * varying dummy covariate */
   #define APVQ 19; /* Age product * varying quantitative covariate */
   
   #define FTYPE 1; /* Fixed covariate */
   #define VTYPE 2; /* Varying covariate (loop in wave) */
   #define ATYPE 2; /* Age product covariate (loop in dh within wave)*/
   
   struct kmodel{
           int maintype; /* main type */
           int subtype; /* subtype */
   };
   struct kmodel modell[NCOVMAX];
   
   double ftol=FTOL; /**< Tolerance for computing Max Likelihood */
   double ftolhess; /**< Tolerance for computing hessian */
   
   /**************** split *************************/
   static  int split( char *path, char *dirc, char *name, char *ext, char *finame )
   {
     /* From a file name with (full) path (either Unix or Windows) we extract the directory (dirc)
        the name of the file (name), its extension only (ext) and its first part of the name (finame)
     */ 
     char  *ss;                            /* pointer */
     int   l1=0, l2=0;                             /* length counters */
   
     l1 = strlen(path );                   /* length of path */
     if ( l1 == 0 ) return( GLOCK_ERROR_NOPATH );
     ss= strrchr( path, DIRSEPARATOR );            /* find last / */
     if ( ss == NULL ) {                   /* no directory, so determine current directory */
       strcpy( name, path );               /* we got the fullname name because no directory */
       /*if(strrchr(path, ODIRSEPARATOR )==NULL)
         printf("Warning you should use %s as a separator\n",DIRSEPARATOR);*/
       /* get current working directory */
       /*    extern  char* getcwd ( char *buf , int len);*/
   #ifdef WIN32
       if (_getcwd( dirc, FILENAME_MAX ) == NULL ) {
   #else
           if (getcwd(dirc, FILENAME_MAX) == NULL) {
   #endif
         return( GLOCK_ERROR_GETCWD );
       }
       /* got dirc from getcwd*/
       printf(" DIRC = %s \n",dirc);
     } else {                              /* strip directory from path */
       ss++;                               /* after this, the filename */
       l2 = strlen( ss );                  /* length of filename */
       if ( l2 == 0 ) return( GLOCK_ERROR_NOPATH );
       strcpy( name, ss );         /* save file name */
       strncpy( dirc, path, l1 - l2 );     /* now the directory */
       dirc[l1-l2] = '\0';                 /* add zero */
       printf(" DIRC2 = %s \n",dirc);
     }
     /* We add a separator at the end of dirc if not exists */
     l1 = strlen( dirc );                  /* length of directory */
     if( dirc[l1-1] != DIRSEPARATOR ){
       dirc[l1] =  DIRSEPARATOR;
       dirc[l1+1] = 0; 
       printf(" DIRC3 = %s \n",dirc);
     }
     ss = strrchr( name, '.' );            /* find last / */
     if (ss >0){
       ss++;
       strcpy(ext,ss);                     /* save extension */
       l1= strlen( name);
       l2= strlen(ss)+1;
       strncpy( finame, name, l1-l2);
       finame[l1-l2]= 0;
     }
   
     return( 0 );                          /* we're done */
   }
   
   
   /******************************************/
   
   void replace_back_to_slash(char *s, char*t)
   {
     int i;
     int lg=0;
     i=0;
     lg=strlen(t);
     for(i=0; i<= lg; i++) {
       (s[i] = t[i]);
       if (t[i]== '\\') s[i]='/';
     }
   }
   
   char *trimbb(char *out, char *in)
   { /* Trim multiple blanks in line but keeps first blanks if line starts with blanks */
     char *s;
     s=out;
     while (*in != '\0'){
       while( *in == ' ' && *(in+1) == ' '){ /* && *(in+1) != '\0'){*/
         in++;
       }
       *out++ = *in++;
     }
     *out='\0';
     return s;
   }
   
   /* char *substrchaine(char *out, char *in, char *chain) */
   /* { */
   /*   /\* Substract chain 'chain' from 'in', return and output 'out' *\/ */
   /*   char *s, *t; */
   /*   t=in;s=out; */
   /*   while ((*in != *chain) && (*in != '\0')){ */
   /*     *out++ = *in++; */
   /*   } */
   
   /*   /\* *in matches *chain *\/ */
   /*   while ((*in++ == *chain++) && (*in != '\0')){ */
   /*     printf("*in = %c, *out= %c *chain= %c \n", *in, *out, *chain);  */
   /*   } */
   /*   in--; chain--; */
   /*   while ( (*in != '\0')){ */
   /*     printf("Bef *in = %c, *out= %c *chain= %c \n", *in, *out, *chain);  */
   /*     *out++ = *in++; */
   /*     printf("Aft *in = %c, *out= %c *chain= %c \n", *in, *out, *chain);  */
   /*   } */
   /*   *out='\0'; */
   /*   out=s; */
   /*   return out; */
   /* } */
   char *substrchaine(char *out, char *in, char *chain)
   {
     /* Substract chain 'chain' from 'in', return and output 'out' */
     /* in="V1+V1*age+age*age+V2", chain="age*age" */
   
     char *strloc;
   
     strcpy (out, in); 
     strloc = strstr(out, chain); /* strloc points to out at age*age+V2 */
     printf("Bef strloc=%s chain=%s out=%s \n", strloc, chain, out);
     if(strloc != NULL){ 
       /* will affect out */ /* strloc+strlenc(chain)=+V2 */ /* Will also work in Unicode */
       memmove(strloc,strloc+strlen(chain), strlen(strloc+strlen(chain))+1);
       /* strcpy (strloc, strloc +strlen(chain));*/
     }
     printf("Aft strloc=%s chain=%s in=%s out=%s \n", strloc, chain, in, out);
     return out;
   }
   
   
   char *cutl(char *blocc, char *alocc, char *in, char occ)
   {
     /* cuts string in into blocc and alocc where blocc ends before FIRST occurence of char 'occ' 
        and alocc starts after first occurence of char 'occ' : ex cutv(blocc,alocc,"abcdef2ghi2j",'2')
        gives alocc="abcdef" and blocc="ghi2j".
        If occ is not found blocc is null and alocc is equal to in. Returns blocc
     */
     char *s, *t;
     t=in;s=in;
     while ((*in != occ) && (*in != '\0')){
       *alocc++ = *in++;
     }
     if( *in == occ){
       *(alocc)='\0';
       s=++in;
     }
    
     if (s == t) {/* occ not found */
       *(alocc-(in-s))='\0';
       in=s;
     }
     while ( *in != '\0'){
       *blocc++ = *in++;
     }
   
     *blocc='\0';
     return t;
   }
   char *cutv(char *blocc, char *alocc, char *in, char occ)
   {
     /* cuts string in into blocc and alocc where blocc ends before LAST occurence of char 'occ' 
        and alocc starts after last occurence of char 'occ' : ex cutv(blocc,alocc,"abcdef2ghi2j",'2')
        gives blocc="abcdef2ghi" and alocc="j".
        If occ is not found blocc is null and alocc is equal to in. Returns alocc
     */
     char *s, *t;
     t=in;s=in;
     while (*in != '\0'){
       while( *in == occ){
         *blocc++ = *in++;
         s=in;
       }
       *blocc++ = *in++;
     }
     if (s == t) /* occ not found */
       *(blocc-(in-s))='\0';
     else
       *(blocc-(in-s)-1)='\0';
     in=s;
     while ( *in != '\0'){
       *alocc++ = *in++;
     }
   
     *alocc='\0';
     return s;
   }
   
   int nbocc(char *s, char occ)
   {
     int i,j=0;
     int lg=20;
     i=0;
     lg=strlen(s);
     for(i=0; i<= lg; i++) {
       if  (s[i] == occ ) j++;
     }
     return j;
   }
   
   /* void cutv(char *u,char *v, char*t, char occ) */
   /* { */
   /*   /\* cuts string t into u and v where u ends before last occurence of char 'occ'  */
   /*      and v starts after last occurence of char 'occ' : ex cutv(u,v,"abcdef2ghi2j",'2') */
   /*      gives u="abcdef2ghi" and v="j" *\/ */
   /*   int i,lg,j,p=0; */
   /*   i=0; */
   /*   lg=strlen(t); */
   /*   for(j=0; j<=lg-1; j++) { */
   /*     if((t[j]!= occ) && (t[j+1]== occ)) p=j+1; */
   /*   } */
   
   /*   for(j=0; j<p; j++) { */
   /*     (u[j] = t[j]); */
   /*   } */
   /*      u[p]='\0'; */
   
   /*    for(j=0; j<= lg; j++) { */
   /*     if (j>=(p+1))(v[j-p-1] = t[j]); */
   /*   } */
   /* } */
   
   #ifdef _WIN32
   char * strsep(char **pp, const char *delim)
   {
     char *p, *q;
            
     if ((p = *pp) == NULL)
       return 0;
     if ((q = strpbrk (p, delim)) != NULL)
     {
       *pp = q + 1;
       *q = '\0';
     }
     else
       *pp = 0;
     return p;
   }
   #endif
   
   /********************** nrerror ********************/
   
   void nrerror(char error_text[])
   {
     fprintf(stderr,"ERREUR ...\n");
     fprintf(stderr,"%s\n",error_text);
     exit(EXIT_FAILURE);
   }
   /*********************** vector *******************/
   double *vector(int nl, int nh)
   {
     double *v;
     v=(double *) malloc((size_t)((nh-nl+1+NR_END)*sizeof(double)));
     if (!v) nrerror("allocation failure in vector");
     return v-nl+NR_END;
   }
   
   /************************ free vector ******************/
   void free_vector(double*v, int nl, int nh)
   {
     free((FREE_ARG)(v+nl-NR_END));
   }
   
   /************************ivector *******************************/
   int *ivector(long nl,long nh)
   {
     int *v;
     v=(int *) malloc((size_t)((nh-nl+1+NR_END)*sizeof(int)));
     if (!v) nrerror("allocation failure in ivector");
     return v-nl+NR_END;
   }
   
   /******************free ivector **************************/
   void free_ivector(int *v, long nl, long nh)
   {
     free((FREE_ARG)(v+nl-NR_END));
   }
   
   /************************lvector *******************************/
   long *lvector(long nl,long nh)
   {
     long *v;
     v=(long *) malloc((size_t)((nh-nl+1+NR_END)*sizeof(long)));
     if (!v) nrerror("allocation failure in ivector");
     return v-nl+NR_END;
   }
   
   /******************free lvector **************************/
   void free_lvector(long *v, long nl, long nh)
   {
     free((FREE_ARG)(v+nl-NR_END));
   }
   
   /******************* imatrix *******************************/
   int **imatrix(long nrl, long nrh, long ncl, long nch) 
        /* allocate a int matrix with subscript range m[nrl..nrh][ncl..nch] */ 
   { 
     long i, nrow=nrh-nrl+1,ncol=nch-ncl+1; 
     int **m; 
     
     /* allocate pointers to rows */ 
     m=(int **) malloc((size_t)((nrow+NR_END)*sizeof(int*))); 
     if (!m) nrerror("allocation failure 1 in matrix()"); 
     m += NR_END; 
     m -= nrl; 
     
     
     /* allocate rows and set pointers to them */ 
     m[nrl]=(int *) malloc((size_t)((nrow*ncol+NR_END)*sizeof(int))); 
     if (!m[nrl]) nrerror("allocation failure 2 in matrix()"); 
     m[nrl] += NR_END; 
     m[nrl] -= ncl; 
     
     for(i=nrl+1;i<=nrh;i++) m[i]=m[i-1]+ncol; 
     
     /* return pointer to array of pointers to rows */ 
     return m; 
   } 
   
   /****************** free_imatrix *************************/
   void free_imatrix(m,nrl,nrh,ncl,nch)
         int **m;
         long nch,ncl,nrh,nrl; 
        /* free an int matrix allocated by imatrix() */ 
   { 
     free((FREE_ARG) (m[nrl]+ncl-NR_END)); 
     free((FREE_ARG) (m+nrl-NR_END)); 
   } 
   
   /******************* matrix *******************************/
   double **matrix(long nrl, long nrh, long ncl, long nch)
   {
     long i, nrow=nrh-nrl+1, ncol=nch-ncl+1;
     double **m;
   
     m=(double **) malloc((size_t)((nrow+NR_END)*sizeof(double*)));
     if (!m) nrerror("allocation failure 1 in matrix()");
     m += NR_END;
     m -= nrl;
   
     m[nrl]=(double *) malloc((size_t)((nrow*ncol+NR_END)*sizeof(double)));
     if (!m[nrl]) nrerror("allocation failure 2 in matrix()");
     m[nrl] += NR_END;
     m[nrl] -= ncl;
   
     for (i=nrl+1; i<=nrh; i++) m[i]=m[i-1]+ncol;
     return m;
     /* print *(*(m+1)+70) or print m[1][70]; print m+1 or print &(m[1]) or &(m[1][0])
   m[i] = address of ith row of the table. &(m[i]) is its value which is another adress
   that of m[i][0]. In order to get the value p m[i][0] but it is unitialized.
      */
   }
   
   /*************************free matrix ************************/
   void free_matrix(double **m, long nrl, long nrh, long ncl, long nch)
   {
     free((FREE_ARG)(m[nrl]+ncl-NR_END));
     free((FREE_ARG)(m+nrl-NR_END));
   }
   
   /******************* ma3x *******************************/
   double ***ma3x(long nrl, long nrh, long ncl, long nch, long nll, long nlh)
   {
     long i, j, nrow=nrh-nrl+1, ncol=nch-ncl+1, nlay=nlh-nll+1;
     double ***m;
   
     m=(double ***) malloc((size_t)((nrow+NR_END)*sizeof(double*)));
     if (!m) nrerror("allocation failure 1 in matrix()");
     m += NR_END;
     m -= nrl;
   
     m[nrl]=(double **) malloc((size_t)((nrow*ncol+NR_END)*sizeof(double)));
     if (!m[nrl]) nrerror("allocation failure 2 in matrix()");
     m[nrl] += NR_END;
     m[nrl] -= ncl;
   
     for (i=nrl+1; i<=nrh; i++) m[i]=m[i-1]+ncol;
   
     m[nrl][ncl]=(double *) malloc((size_t)((nrow*ncol*nlay+NR_END)*sizeof(double)));
     if (!m[nrl][ncl]) nrerror("allocation failure 3 in matrix()");
     m[nrl][ncl] += NR_END;
     m[nrl][ncl] -= nll;
     for (j=ncl+1; j<=nch; j++) 
       m[nrl][j]=m[nrl][j-1]+nlay;
     
     for (i=nrl+1; i<=nrh; i++) {
       m[i][ncl]=m[i-1l][ncl]+ncol*nlay;
       for (j=ncl+1; j<=nch; j++) 
         m[i][j]=m[i][j-1]+nlay;
     }
     return m; 
     /*  gdb: p *(m+1) <=> p m[1] and p (m+1) <=> p (m+1) <=> p &(m[1])
              &(m[i][j][k]) <=> *((*(m+i) + j)+k)
     */
   }
   
   /*************************free ma3x ************************/
   void free_ma3x(double ***m, long nrl, long nrh, long ncl, long nch,long nll, long nlh)
   {
     free((FREE_ARG)(m[nrl][ncl]+ nll-NR_END));
     free((FREE_ARG)(m[nrl]+ncl-NR_END));
     free((FREE_ARG)(m+nrl-NR_END));
   }
   
   /*************** function subdirf ***********/
   char *subdirf(char fileres[])
   {
     /* Caution optionfilefiname is hidden */
     strcpy(tmpout,optionfilefiname);
     strcat(tmpout,"/"); /* Add to the right */
     strcat(tmpout,fileres);
     return tmpout;
   }
   
   /*************** function subdirf2 ***********/
   char *subdirf2(char fileres[], char *preop)
   {
     /* Example subdirf2(optionfilefiname,"FB_") with optionfilefiname="texte", result="texte/FB_texte"
    Errors in subdirf, 2, 3 while printing tmpout is
    rewritten within the same printf. Workaround: many printfs */
     /* Caution optionfilefiname is hidden */
     strcpy(tmpout,optionfilefiname);
     strcat(tmpout,"/");
     strcat(tmpout,preop);
     strcat(tmpout,fileres);
     return tmpout;
   }
   
   /*************** function subdirf3 ***********/
   char *subdirf3(char fileres[], char *preop, char *preop2)
   {
     
     /* Caution optionfilefiname is hidden */
     strcpy(tmpout,optionfilefiname);
     strcat(tmpout,"/");
     strcat(tmpout,preop);
     strcat(tmpout,preop2);
     strcat(tmpout,fileres);
     return tmpout;
   }
    
   /*************** function subdirfext ***********/
   char *subdirfext(char fileres[], char *preop, char *postop)
   {
     
     strcpy(tmpout,preop);
     strcat(tmpout,fileres);
     strcat(tmpout,postop);
     return tmpout;
   }
   
   /*************** function subdirfext3 ***********/
   char *subdirfext3(char fileres[], char *preop, char *postop)
   {
     
     /* Caution optionfilefiname is hidden */
     strcpy(tmpout,optionfilefiname);
     strcat(tmpout,"/");
     strcat(tmpout,preop);
     strcat(tmpout,fileres);
     strcat(tmpout,postop);
     return tmpout;
   }
    
   char *asc_diff_time(long time_sec, char ascdiff[])
   {
     long sec_left, days, hours, minutes;
     days = (time_sec) / (60*60*24);
     sec_left = (time_sec) % (60*60*24);
     hours = (sec_left) / (60*60) ;
     sec_left = (sec_left) %(60*60);
     minutes = (sec_left) /60;
     sec_left = (sec_left) % (60);
     sprintf(ascdiff,"%ld day(s) %ld hour(s) %ld minute(s) %ld second(s)",days, hours, minutes, sec_left);  
     return ascdiff;
   }
   
   /***************** f1dim *************************/
   extern int ncom; 
   extern double *pcom,*xicom;
   extern double (*nrfunc)(double []); 
    
   double f1dim(double x) 
   { 
     int j; 
     double f;
     double *xt; 
    
     xt=vector(1,ncom); 
     for (j=1;j<=ncom;j++) xt[j]=pcom[j]+x*xicom[j]; 
     f=(*nrfunc)(xt); 
     free_vector(xt,1,ncom); 
     return f; 
   } 
   
   /*****************brent *************************/
   double brent(double ax, double bx, double cx, double (*f)(double), double tol,  double *xmin) 
   {
     /* Given a function f, and given a bracketing triplet of abscissas ax, bx, cx (such that bx is
      * between ax and cx, and f(bx) is less than both f(ax) and f(cx) ), this routine isolates
      * the minimum to a fractional precision of about tol using Brent’s method. The abscissa of
      * the minimum is returned as xmin, and the minimum function value is returned as brent , the
      * returned function value. 
     */
     int iter; 
     double a,b,d,etemp;
     double fu=0,fv,fw,fx;
     double ftemp=0.;
     double p,q,r,tol1,tol2,u,v,w,x,xm; 
     double e=0.0; 
    
     a=(ax < cx ? ax : cx); 
     b=(ax > cx ? ax : cx); 
     x=w=v=bx; 
     fw=fv=fx=(*f)(x); 
     for (iter=1;iter<=ITMAX;iter++) { 
       xm=0.5*(a+b); 
       tol2=2.0*(tol1=tol*fabs(x)+ZEPS); 
       /*          if (2.0*fabs(fp-(*fret)) <= ftol*(fabs(fp)+fabs(*fret)))*/
       printf(".");fflush(stdout);
       fprintf(ficlog,".");fflush(ficlog);
   #ifdef DEBUGBRENT
       printf("br %d,x=%.10e xm=%.10e b=%.10e a=%.10e tol=%.10e tol1=%.10e tol2=%.10e x-xm=%.10e fx=%.12e fu=%.12e,fw=%.12e,ftemp=%.12e,ftol=%.12e\n",iter,x,xm,b,a,tol,tol1,tol2,(x-xm),fx,fu,fw,ftemp,ftol);
       fprintf(ficlog,"br %d,x=%.10e xm=%.10e b=%.10e a=%.10e tol=%.10e tol1=%.10e tol2=%.10e x-xm=%.10e fx=%.12e fu=%.12e,fw=%.12e,ftemp=%.12e,ftol=%.12e\n",iter,x,xm,b,a,tol,tol1,tol2,(x-xm),fx,fu,fw,ftemp,ftol);
       /*          if ((fabs(x-xm) <= (tol2-0.5*(b-a)))||(2.0*fabs(fu-ftemp) <= ftol*1.e-2*(fabs(fu)+fabs(ftemp)))) { */
   #endif
       if (fabs(x-xm) <= (tol2-0.5*(b-a))){ 
         *xmin=x; 
         return fx; 
       } 
       ftemp=fu;
       if (fabs(e) > tol1) { 
         r=(x-w)*(fx-fv); 
         q=(x-v)*(fx-fw); 
         p=(x-v)*q-(x-w)*r; 
         q=2.0*(q-r); 
         if (q > 0.0) p = -p; 
         q=fabs(q); 
         etemp=e; 
         e=d; 
         if (fabs(p) >= fabs(0.5*q*etemp) || p <= q*(a-x) || p >= q*(b-x)) 
                                   d=CGOLD*(e=(x >= xm ? a-x : b-x)); 
         else { 
                                   d=p/q; 
                                   u=x+d; 
                                   if (u-a < tol2 || b-u < tol2) 
                                           d=SIGN(tol1,xm-x); 
         } 
       } else { 
         d=CGOLD*(e=(x >= xm ? a-x : b-x)); 
       } 
       u=(fabs(d) >= tol1 ? x+d : x+SIGN(tol1,d)); 
       fu=(*f)(u); 
       if (fu <= fx) { 
         if (u >= x) a=x; else b=x; 
         SHFT(v,w,x,u) 
         SHFT(fv,fw,fx,fu) 
       } else { 
         if (u < x) a=u; else b=u; 
         if (fu <= fw || w == x) { 
                                   v=w; 
                                   w=u; 
                                   fv=fw; 
                                   fw=fu; 
         } else if (fu <= fv || v == x || v == w) { 
                                   v=u; 
                                   fv=fu; 
         } 
       } 
     } 
     nrerror("Too many iterations in brent"); 
     *xmin=x; 
     return fx; 
   } 
   
   /****************** mnbrak ***********************/
   
   void mnbrak(double *ax, double *bx, double *cx, double *fa, double *fb, double *fc, 
               double (*func)(double)) 
   { /* Given a function func , and given distinct initial points ax and bx , this routine searches in
   the downhill direction (defined by the function as evaluated at the initial points) and returns
   new points ax , bx , cx that bracket a minimum of the function. Also returned are the function
   values at the three points, fa, fb , and fc such that fa > fb and fb < fc.
      */
     double ulim,u,r,q, dum;
     double fu; 
   
     double scale=10.;
     int iterscale=0;
   
     *fa=(*func)(*ax); /*  xta[j]=pcom[j]+(*ax)*xicom[j]; fa=f(xta[j])*/
     *fb=(*func)(*bx); /*  xtb[j]=pcom[j]+(*bx)*xicom[j]; fb=f(xtb[j]) */
   
   
     /* while(*fb != *fb){ /\* *ax should be ok, reducing distance to *ax *\/ */
     /*   printf("Warning mnbrak *fb = %lf, *bx=%lf *ax=%lf *fa==%lf iter=%d\n",*fb, *bx, *ax, *fa, iterscale++); */
     /*   *bx = *ax - (*ax - *bx)/scale; */
     /*   *fb=(*func)(*bx);  /\*  xtb[j]=pcom[j]+(*bx)*xicom[j]; fb=f(xtb[j]) *\/ */
     /* } */
   
     if (*fb > *fa) { 
       SHFT(dum,*ax,*bx,dum) 
       SHFT(dum,*fb,*fa,dum) 
     } 
     *cx=(*bx)+GOLD*(*bx-*ax); 
     *fc=(*func)(*cx); 
   #ifdef DEBUG
     printf("mnbrak0 a=%lf *fa=%lf, b=%lf *fb=%lf, c=%lf *fc=%lf\n",*ax,*fa,*bx,*fb,*cx, *fc);
     fprintf(ficlog,"mnbrak0 a=%lf *fa=%lf, b=%lf *fb=%lf, c=%lf *fc=%lf\n",*ax,*fa,*bx,*fb,*cx, *fc);
   #endif
     while (*fb > *fc) { /* Declining a,b,c with fa> fb > fc. If fc=inf it exits and if flat fb=fc it exits too.*/
       r=(*bx-*ax)*(*fb-*fc); 
       q=(*bx-*cx)*(*fb-*fa); /* What if fa=inf */
       u=(*bx)-((*bx-*cx)*q-(*bx-*ax)*r)/ 
         (2.0*SIGN(FMAX(fabs(q-r),TINY),q-r)); /* Minimum abscissa of a parabolic estimated from (a,fa), (b,fb) and (c,fc). */
       ulim=(*bx)+GLIMIT*(*cx-*bx); /* Maximum abscissa where function should be evaluated */
       if ((*bx-u)*(u-*cx) > 0.0) { /* if u_p is between b and c */
         fu=(*func)(u); 
   #ifdef DEBUG
         /* f(x)=A(x-u)**2+f(u) */
         double A, fparabu; 
         A= (*fb - *fa)/(*bx-*ax)/(*bx+*ax-2*u);
         fparabu= *fa - A*(*ax-u)*(*ax-u);
         printf("\nmnbrak (*ax=%.12f, *fa=%.12lf), (*bx=%.12f, *fb=%.12lf), (*cx=%.12f, *fc=%.12lf),  (*u=%.12f, fu=%.12lf, fparabu=%.12f, q=%lf < %lf=r)\n",*ax,*fa,*bx,*fb,*cx,*fc,u,fu, fparabu,q,r);
         fprintf(ficlog,"\nmnbrak (*ax=%.12f, *fa=%.12lf), (*bx=%.12f, *fb=%.12lf), (*cx=%.12f, *fc=%.12lf),  (*u=%.12f, fu=%.12lf, fparabu=%.12f, q=%lf < %lf=r)\n",*ax,*fa,*bx,*fb,*cx,*fc,u,fu, fparabu,q,r);
         /* And thus,it can be that fu > *fc even if fparabu < *fc */
         /* mnbrak (*ax=7.666299858533, *fa=299039.693133272231), (*bx=8.595447774979, *fb=298976.598289369489),
           (*cx=10.098840694817, *fc=298946.631474258087),  (*u=9.852501168332, fu=298948.773013752128, fparabu=298945.434711494134) */
         /* In that case, there is no bracket in the output! Routine is wrong with many consequences.*/
   #endif 
   #ifdef MNBRAKORIGINAL
   #else
   /*       if (fu > *fc) { */
   /* #ifdef DEBUG */
   /*       printf("mnbrak4  fu > fc \n"); */
   /*       fprintf(ficlog, "mnbrak4 fu > fc\n"); */
   /* #endif */
   /*      /\* SHFT(u,*cx,*cx,u) /\\* ie a=c, c=u and u=c; in that case, next SHFT(a,b,c,u) will give a=b=b, b=c=u, c=u=c and *\\/  *\/ */
   /*      /\* SHFT(*fa,*fc,fu,*fc) /\\* (b, u, c) is a bracket while test fb > fc will be fu > fc  will exit *\\/ *\/ */
   /*      dum=u; /\* Shifting c and u *\/ */
   /*      u = *cx; */
   /*      *cx = dum; */
   /*      dum = fu; */
   /*      fu = *fc; */
   /*      *fc =dum; */
   /*       } else { /\* end *\/ */
   /* #ifdef DEBUG */
   /*       printf("mnbrak3  fu < fc \n"); */
   /*       fprintf(ficlog, "mnbrak3 fu < fc\n"); */
   /* #endif */
   /*      dum=u; /\* Shifting c and u *\/ */
   /*      u = *cx; */
   /*      *cx = dum; */
   /*      dum = fu; */
   /*      fu = *fc; */
   /*      *fc =dum; */
   /*       } */
   #ifdef DEBUGMNBRAK
                    double A, fparabu; 
        A= (*fb - *fa)/(*bx-*ax)/(*bx+*ax-2*u);
        fparabu= *fa - A*(*ax-u)*(*ax-u);
        printf("\nmnbrak35 ax=%lf fa=%lf bx=%lf fb=%lf, u=%lf fp=%lf fu=%lf < or >= fc=%lf cx=%lf, q=%lf < %lf=r \n",*ax, *fa, *bx,*fb,u,fparabu,fu,*fc,*cx,q,r);
        fprintf(ficlog,"\nmnbrak35 ax=%lf fa=%lf bx=%lf fb=%lf, u=%lf fp=%lf fu=%lf < or >= fc=%lf cx=%lf, q=%lf < %lf=r \n",*ax, *fa, *bx,*fb,u,fparabu,fu,*fc,*cx,q,r);
   #endif
         dum=u; /* Shifting c and u */
         u = *cx;
         *cx = dum;
         dum = fu;
         fu = *fc;
         *fc =dum;
   #endif
       } else if ((*cx-u)*(u-ulim) > 0.0) { /* u is after c but before ulim */
   #ifdef DEBUG
         printf("\nmnbrak2  u=%lf after c=%lf but before ulim\n",u,*cx);
         fprintf(ficlog,"\nmnbrak2  u=%lf after c=%lf but before ulim\n",u,*cx);
   #endif
         fu=(*func)(u); 
         if (fu < *fc) { 
   #ifdef DEBUG
                                   printf("\nmnbrak2  u=%lf after c=%lf but before ulim=%lf AND fu=%lf < %lf=fc\n",u,*cx,ulim,fu, *fc);
                             fprintf(ficlog,"\nmnbrak2  u=%lf after c=%lf but before ulim=%lf AND fu=%lf < %lf=fc\n",u,*cx,ulim,fu, *fc);
   #endif
                             SHFT(*bx,*cx,u,*cx+GOLD*(*cx-*bx)) 
                                   SHFT(*fb,*fc,fu,(*func)(u)) 
   #ifdef DEBUG
                                           printf("\nmnbrak2 shift GOLD c=%lf",*cx+GOLD*(*cx-*bx));
   #endif
         } 
       } else if ((u-ulim)*(ulim-*cx) >= 0.0) { /* u outside ulim (verifying that ulim is beyond c) */
   #ifdef DEBUG
         printf("\nmnbrak2  u=%lf outside ulim=%lf (verifying that ulim is beyond c=%lf)\n",u,ulim,*cx);
         fprintf(ficlog,"\nmnbrak2  u=%lf outside ulim=%lf (verifying that ulim is beyond c=%lf)\n",u,ulim,*cx);
   #endif
         u=ulim; 
         fu=(*func)(u); 
       } else { /* u could be left to b (if r > q parabola has a maximum) */
   #ifdef DEBUG
         printf("\nmnbrak2  u=%lf could be left to b=%lf (if r=%lf > q=%lf parabola has a maximum)\n",u,*bx,r,q);
         fprintf(ficlog,"\nmnbrak2  u=%lf could be left to b=%lf (if r=%lf > q=%lf parabola has a maximum)\n",u,*bx,r,q);
   #endif
         u=(*cx)+GOLD*(*cx-*bx); 
         fu=(*func)(u); 
   #ifdef DEBUG
         printf("\nmnbrak2 new u=%lf fu=%lf shifted gold left from c=%lf and b=%lf \n",u,fu,*cx,*bx);
         fprintf(ficlog,"\nmnbrak2 new u=%lf fu=%lf shifted gold left from c=%lf and b=%lf \n",u,fu,*cx,*bx);
   #endif
       } /* end tests */
       SHFT(*ax,*bx,*cx,u) 
       SHFT(*fa,*fb,*fc,fu) 
   #ifdef DEBUG
         printf("\nmnbrak2 shift (*ax=%.12f, *fa=%.12lf), (*bx=%.12f, *fb=%.12lf), (*cx=%.12f, *fc=%.12lf)\n",*ax,*fa,*bx,*fb,*cx,*fc);
         fprintf(ficlog, "\nmnbrak2 shift (*ax=%.12f, *fa=%.12lf), (*bx=%.12f, *fb=%.12lf), (*cx=%.12f, *fc=%.12lf)\n",*ax,*fa,*bx,*fb,*cx,*fc);
   #endif
     } /* end while; ie return (a, b, c, fa, fb, fc) such that a < b < c with f(a) > f(b) and fb < f(c) */
   } 
   
   /*************** linmin ************************/
   /* Given an n -dimensional point p[1..n] and an n -dimensional direction xi[1..n] , moves and
   resets p to where the function func(p) takes on a minimum along the direction xi from p ,
   and replaces xi by the actual vector displacement that p was moved. Also returns as fret
   the value of func at the returned location p . This is actually all accomplished by calling the
   routines mnbrak and brent .*/
   int ncom; 
   double *pcom,*xicom;
   double (*nrfunc)(double []); 
    
   #ifdef LINMINORIGINAL
   void linmin(double p[], double xi[], int n, double *fret,double (*func)(double [])) 
   #else
   void linmin(double p[], double xi[], int n, double *fret,double (*func)(double []), int *flat) 
   #endif
   { 
     double brent(double ax, double bx, double cx, 
                  double (*f)(double), double tol, double *xmin); 
     double f1dim(double x); 
     void mnbrak(double *ax, double *bx, double *cx, double *fa, double *fb, 
                 double *fc, double (*func)(double)); 
     int j; 
     double xx,xmin,bx,ax; 
     double fx,fb,fa;
   
   #ifdef LINMINORIGINAL
   #else
     double scale=10., axs, xxs; /* Scale added for infinity */
   #endif
     
     ncom=n; 
     pcom=vector(1,n); 
     xicom=vector(1,n); 
     nrfunc=func; 
     for (j=1;j<=n;j++) { 
       pcom[j]=p[j]; 
       xicom[j]=xi[j]; /* Former scale xi[j] of currrent direction i */
     } 
   
   #ifdef LINMINORIGINAL
     xx=1.;
   #else
     axs=0.0;
     xxs=1.;
     do{
       xx= xxs;
   #endif
       ax=0.;
       mnbrak(&ax,&xx,&bx,&fa,&fx,&fb,f1dim);  /* Outputs: xtx[j]=pcom[j]+(*xx)*xicom[j]; fx=f(xtx[j]) */
       /* brackets with inputs ax=0 and xx=1, but points, pcom=p, and directions values, xicom=xi, are sent via f1dim(x) */
       /* xt[x,j]=pcom[j]+x*xicom[j]  f(ax) = f(xt(a,j=1,n)) = f(p(j) + 0 * xi(j)) and  f(xx) = f(xt(x, j=1,n)) = f(p(j) + 1 * xi(j))   */
       /* Outputs: fa=f(p(j)) and fx=f(p(j) + xxs * xi(j) ) and f(bx)= f(p(j)+ bx* xi(j)) */
       /* Given input ax=axs and xx=xxs, xx might be too far from ax to get a finite f(xx) */
       /* Searches on line, outputs (ax, xx, bx) such that fx < min(fa and fb) */
       /* Find a bracket a,x,b in direction n=xi ie xicom, order may change. Scale is [0:xxs*xi[j]] et non plus  [0:xi[j]]*/
   #ifdef LINMINORIGINAL
   #else
       if (fx != fx){
                           xxs=xxs/scale; /* Trying a smaller xx, closer to initial ax=0 */
                           printf("|");
                           fprintf(ficlog,"|");
   #ifdef DEBUGLINMIN
                           printf("\nLinmin NAN : input [axs=%lf:xxs=%lf], mnbrak outputs fx=%lf <(fb=%lf and fa=%lf) with xx=%lf in [ax=%lf:bx=%lf] \n",  axs, xxs, fx,fb, fa, xx, ax, bx);
   #endif
       }
     }while(fx != fx && xxs > 1.e-5);
   #endif
     
   #ifdef DEBUGLINMIN
     printf("\nLinmin after mnbrak: ax=%12.7f xx=%12.7f bx=%12.7f fa=%12.2f fx=%12.2f fb=%12.2f\n",  ax,xx,bx,fa,fx,fb);
     fprintf(ficlog,"\nLinmin after mnbrak: ax=%12.7f xx=%12.7f bx=%12.7f fa=%12.2f fx=%12.2f fb=%12.2f\n",  ax,xx,bx,fa,fx,fb);
   #endif
   #ifdef LINMINORIGINAL
   #else
     if(fb == fx){ /* Flat function in the direction */
       xmin=xx;
       *flat=1;
     }else{
       *flat=0;
   #endif
                   /*Flat mnbrak2 shift (*ax=0.000000000000, *fa=51626.272983130431), (*bx=-1.618034000000, *fb=51590.149499362531), (*cx=-4.236068025156, *fc=51590.149499362531) */
     *fret=brent(ax,xx,bx,f1dim,TOL,&xmin); /* Giving a bracketting triplet (ax, xx, bx), find a minimum, xmin, according to f1dim, *fret(xmin),*/
     /* fa = f(p[j] + ax * xi[j]), fx = f(p[j] + xx * xi[j]), fb = f(p[j] + bx * xi[j]) */
     /* fmin = f(p[j] + xmin * xi[j]) */
     /* P+lambda n in that direction (lambdamin), with TOL between abscisses */
     /* f1dim(xmin): for (j=1;j<=ncom;j++) xt[j]=pcom[j]+xmin*xicom[j]; */
   #ifdef DEBUG
     printf("retour brent from bracket (a=%lf fa=%lf, xx=%lf fx=%lf, b=%lf fb=%lf): fret=%lf xmin=%lf\n",ax,fa,xx,fx,bx,fb,*fret,xmin);
     fprintf(ficlog,"retour brent from bracket (a=%lf fa=%lf, xx=%lf fx=%lf, b=%lf fb=%lf): fret=%lf xmin=%lf\n",ax,fa,xx,fx,bx,fb,*fret,xmin);
   #endif
   #ifdef LINMINORIGINAL
   #else
                           }
   #endif
   #ifdef DEBUGLINMIN
     printf("linmin end ");
     fprintf(ficlog,"linmin end ");
   #endif
     for (j=1;j<=n;j++) { 
   #ifdef LINMINORIGINAL
       xi[j] *= xmin; 
   #else
   #ifdef DEBUGLINMIN
       if(xxs <1.0)
         printf(" before xi[%d]=%12.8f", j,xi[j]);
   #endif
       xi[j] *= xmin*xxs; /* xi rescaled by xmin and number of loops: if xmin=-1.237 and xi=(1,0,...,0) xi=(-1.237,0,...,0) */
   #ifdef DEBUGLINMIN
       if(xxs <1.0)
         printf(" after xi[%d]=%12.8f, xmin=%12.8f, ax=%12.8f, xx=%12.8f, bx=%12.8f, xxs=%12.8f", j,xi[j], xmin, ax, xx, bx,xxs );
   #endif
   #endif
       p[j] += xi[j]; /* Parameters values are updated accordingly */
     } 
   #ifdef DEBUGLINMIN
     printf("\n");
     printf("Comparing last *frec(xmin=%12.8f)=%12.8f from Brent and frec(0.)=%12.8f \n", xmin, *fret, (*func)(p));
     fprintf(ficlog,"Comparing last *frec(xmin=%12.8f)=%12.8f from Brent and frec(0.)=%12.8f \n", xmin, *fret, (*func)(p));
     for (j=1;j<=n;j++) { 
       printf(" xi[%d]= %14.10f p[%d]= %12.7f",j,xi[j],j,p[j]);
       fprintf(ficlog," xi[%d]= %14.10f p[%d]= %12.7f",j,xi[j],j,p[j]);
       if(j % ncovmodel == 0){
         printf("\n");
         fprintf(ficlog,"\n");
       }
     }
   #else
   #endif
     free_vector(xicom,1,n); 
     free_vector(pcom,1,n); 
   } 
   
   
   /*************** powell ************************/
   /*
   Minimization of a function func of n variables. Input consists in an initial starting point
   p[1..n] ; an initial matrix xi[1..n][1..n]  whose columns contain the initial set of di-
   rections (usually the n unit vectors); and ftol, the fractional tolerance in the function value
   such that failure to decrease by more than this amount in one iteration signals doneness. On
   output, p is set to the best point found, xi is the then-current direction set, fret is the returned
   function value at p , and iter is the number of iterations taken. The routine linmin is used.
    */
   #ifdef LINMINORIGINAL
   #else
           int *flatdir; /* Function is vanishing in that direction */
           int flat=0, flatd=0; /* Function is vanishing in that direction */
   #endif
   void powell(double p[], double **xi, int n, double ftol, int *iter, double *fret, 
               double (*func)(double [])) 
   { 
   #ifdef LINMINORIGINAL
    void linmin(double p[], double xi[], int n, double *fret, 
                 double (*func)(double [])); 
   #else 
    void linmin(double p[], double xi[], int n, double *fret,
                double (*func)(double []),int *flat); 
   #endif
    int i,ibig,j,jk,k; 
     double del,t,*pt,*ptt,*xit;
     double directest;
     double fp,fptt;
     double *xits;
     int niterf, itmp;
   
     pt=vector(1,n); 
     ptt=vector(1,n); 
     xit=vector(1,n); 
     xits=vector(1,n); 
     *fret=(*func)(p); 
     for (j=1;j<=n;j++) pt[j]=p[j]; 
     rcurr_time = time(NULL);  
     for (*iter=1;;++(*iter)) { 
       ibig=0; 
       del=0.0; 
       rlast_time=rcurr_time;
       /* (void) gettimeofday(&curr_time,&tzp); */
       rcurr_time = time(NULL);  
       curr_time = *localtime(&rcurr_time);
       printf("\nPowell iter=%d -2*LL=%.12f gain=%.12f=%.3g %ld sec. %ld sec.",*iter,*fret, fp-*fret,fp-*fret, rcurr_time-rlast_time, rcurr_time-rstart_time);fflush(stdout);
       fprintf(ficlog,"\nPowell iter=%d -2*LL=%.12f gain=%.12f=%.3g %ld sec. %ld sec.",*iter,*fret, fp-*fret,fp-*fret,rcurr_time-rlast_time, rcurr_time-rstart_time); fflush(ficlog);
   /*     fprintf(ficrespow,"%d %.12f %ld",*iter,*fret,curr_time.tm_sec-start_time.tm_sec); */
       fp=(*fret); /* From former iteration or initial value */
       for (i=1;i<=n;i++) {
         fprintf(ficrespow," %.12lf", p[i]);
       }
       fprintf(ficrespow,"\n");fflush(ficrespow);
       printf("\n#model=  1      +     age ");
       fprintf(ficlog,"\n#model=  1      +     age ");
       if(nagesqr==1){
           printf("  + age*age  ");
           fprintf(ficlog,"  + age*age  ");
       }
       for(j=1;j <=ncovmodel-2;j++){
         if(Typevar[j]==0) {
           printf("  +      V%d  ",Tvar[j]);
           fprintf(ficlog,"  +      V%d  ",Tvar[j]);
         }else if(Typevar[j]==1) {
           printf("  +    V%d*age ",Tvar[j]);
           fprintf(ficlog,"  +    V%d*age ",Tvar[j]);
         }else if(Typevar[j]==2) {
           printf("  +    V%d*V%d ",Tvard[Tposprod[j]][1],Tvard[Tposprod[j]][2]);
           fprintf(ficlog,"  +    V%d*V%d ",Tvard[Tposprod[j]][1],Tvard[Tposprod[j]][2]);
         }
       }
       printf("\n");
   /*     printf("12   47.0114589    0.0154322   33.2424412    0.3279905    2.3731903  */
   /* 13  -21.5392400    0.1118147    1.2680506    1.2973408   -1.0663662  */
       fprintf(ficlog,"\n");
       for(i=1,jk=1; i <=nlstate; i++){
         for(k=1; k <=(nlstate+ndeath); k++){
           if (k != i) {
             printf("%d%d ",i,k);
             fprintf(ficlog,"%d%d ",i,k);
             for(j=1; j <=ncovmodel; j++){
               printf("%12.7f ",p[jk]);
               fprintf(ficlog,"%12.7f ",p[jk]);
               jk++; 
             }
             printf("\n");
             fprintf(ficlog,"\n");
           }
         }
       }
       if(*iter <=3 && *iter >1){
         tml = *localtime(&rcurr_time);
         strcpy(strcurr,asctime(&tml));
         rforecast_time=rcurr_time; 
         itmp = strlen(strcurr);
         if(strcurr[itmp-1]=='\n')  /* Windows outputs with a new line */
           strcurr[itmp-1]='\0';
         printf("\nConsidering the time needed for the last iteration #%d: %ld seconds,\n",*iter,rcurr_time-rlast_time);
         fprintf(ficlog,"\nConsidering the time needed for this last iteration #%d: %ld seconds,\n",*iter,rcurr_time-rlast_time);
         for(niterf=10;niterf<=30;niterf+=10){
           rforecast_time=rcurr_time+(niterf-*iter)*(rcurr_time-rlast_time);
           forecast_time = *localtime(&rforecast_time);
           strcpy(strfor,asctime(&forecast_time));
           itmp = strlen(strfor);
           if(strfor[itmp-1]=='\n')
             strfor[itmp-1]='\0';
           printf("   - if your program needs %d iterations to converge, convergence will be \n   reached in %s i.e.\n   on %s (current time is %s);\n",niterf, asc_diff_time(rforecast_time-rcurr_time,tmpout),strfor,strcurr);
           fprintf(ficlog,"   - if your program needs %d iterations to converge, convergence will be \n   reached in %s i.e.\n   on %s (current time is %s);\n",niterf, asc_diff_time(rforecast_time-rcurr_time,tmpout),strfor,strcurr);
         }
       }
       for (i=1;i<=n;i++) { /* For each direction i */
         for (j=1;j<=n;j++) xit[j]=xi[j][i]; /* Directions stored from previous iteration with previous scales */
         fptt=(*fret); 
   #ifdef DEBUG
         printf("fret=%lf, %lf, %lf \n", *fret, *fret, *fret);
         fprintf(ficlog, "fret=%lf, %lf, %lf \n", *fret, *fret, *fret);
   #endif
         printf("%d",i);fflush(stdout); /* print direction (parameter) i */
         fprintf(ficlog,"%d",i);fflush(ficlog);
   #ifdef LINMINORIGINAL
         linmin(p,xit,n,fret,func); /* Point p[n]. xit[n] has been loaded for direction i as input.*/
   #else
         linmin(p,xit,n,fret,func,&flat); /* Point p[n]. xit[n] has been loaded for direction i as input.*/
                           flatdir[i]=flat; /* Function is vanishing in that direction i */
   #endif
                           /* Outputs are fret(new point p) p is updated and xit rescaled */
         if (fabs(fptt-(*fret)) > del) { /* We are keeping the max gain on each of the n directions */
                                   /* because that direction will be replaced unless the gain del is small */
                                   /* in comparison with the 'probable' gain, mu^2, with the last average direction. */
                                   /* Unless the n directions are conjugate some gain in the determinant may be obtained */
                                   /* with the new direction. */
                                   del=fabs(fptt-(*fret)); 
                                   ibig=i; 
         } 
   #ifdef DEBUG
         printf("%d %.12e",i,(*fret));
         fprintf(ficlog,"%d %.12e",i,(*fret));
         for (j=1;j<=n;j++) {
                                   xits[j]=FMAX(fabs(p[j]-pt[j]),1.e-5);
                                   printf(" x(%d)=%.12e",j,xit[j]);
                                   fprintf(ficlog," x(%d)=%.12e",j,xit[j]);
         }
         for(j=1;j<=n;j++) {
                                   printf(" p(%d)=%.12e",j,p[j]);
                                   fprintf(ficlog," p(%d)=%.12e",j,p[j]);
         }
         printf("\n");
         fprintf(ficlog,"\n");
   #endif
       } /* end loop on each direction i */
       /* Convergence test will use last linmin estimation (fret) and compare former iteration (fp) */ 
       /* But p and xit have been updated at the end of linmin, *fret corresponds to new p, xit  */
       /* New value of last point Pn is not computed, P(n-1) */
       for(j=1;j<=n;j++) {
         if(flatdir[j] >0){
           printf(" p(%d)=%lf flat=%d ",j,p[j],flatdir[j]);
           fprintf(ficlog," p(%d)=%lf flat=%d ",j,p[j],flatdir[j]);
         }
         /* printf("\n"); */
         /* fprintf(ficlog,"\n"); */
       }
       /* if (2.0*fabs(fp-(*fret)) <= ftol*(fabs(fp)+fabs(*fret))) { /\* Did we reach enough precision? *\/ */
       if (2.0*fabs(fp-(*fret)) <= ftol) { /* Did we reach enough precision? */
         /* We could compare with a chi^2. chisquare(0.95,ddl=1)=3.84 */
         /* By adding age*age in a model, the new -2LL should be lower and the difference follows a */
         /* a chisquare statistics with 1 degree. To be significant at the 95% level, it should have */
         /* decreased of more than 3.84  */
         /* By adding age*age and V1*age the gain (-2LL) should be more than 5.99 (ddl=2) */
         /* By using V1+V2+V3, the gain should be  7.82, compared with basic 1+age. */
         /* By adding 10 parameters more the gain should be 18.31 */
                           
         /* Starting the program with initial values given by a former maximization will simply change */
         /* the scales of the directions and the directions, because the are reset to canonical directions */
         /* Thus the first calls to linmin will give new points and better maximizations until fp-(*fret) is */
         /* under the tolerance value. If the tolerance is very small 1.e-9, it could last long.  */
   #ifdef DEBUG
         int k[2],l;
         k[0]=1;
         k[1]=-1;
         printf("Max: %.12e",(*func)(p));
         fprintf(ficlog,"Max: %.12e",(*func)(p));
         for (j=1;j<=n;j++) {
           printf(" %.12e",p[j]);
           fprintf(ficlog," %.12e",p[j]);
         }
         printf("\n");
         fprintf(ficlog,"\n");
         for(l=0;l<=1;l++) {
           for (j=1;j<=n;j++) {
             ptt[j]=p[j]+(p[j]-pt[j])*k[l];
             printf("l=%d j=%d ptt=%.12e, xits=%.12e, p=%.12e, xit=%.12e", l,j,ptt[j],xits[j],p[j],xit[j]);
             fprintf(ficlog,"l=%d j=%d ptt=%.12e, xits=%.12e, p=%.12e, xit=%.12e", l,j,ptt[j],xits[j],p[j],xit[j]);
           }
           printf("func(ptt)=%.12e, deriv=%.12e\n",(*func)(ptt),(ptt[j]-p[j])/((*func)(ptt)-(*func)(p)));
           fprintf(ficlog,"func(ptt)=%.12e, deriv=%.12e\n",(*func)(ptt),(ptt[j]-p[j])/((*func)(ptt)-(*func)(p)));
         }
   #endif
   
         free_vector(xit,1,n); 
         free_vector(xits,1,n); 
         free_vector(ptt,1,n); 
         free_vector(pt,1,n); 
         return; 
       } /* enough precision */ 
       if (*iter == ITMAX*n) nrerror("powell exceeding maximum iterations."); 
       for (j=1;j<=n;j++) { /* Computes the extrapolated point P_0 + 2 (P_n-P_0) */
         ptt[j]=2.0*p[j]-pt[j]; 
         xit[j]=p[j]-pt[j]; 
         pt[j]=p[j]; 
       } 
       fptt=(*func)(ptt); /* f_3 */
   #ifdef NODIRECTIONCHANGEDUNTILNITER  /* No change in drections until some iterations are done */
                   if (*iter <=4) {
   #else
   #endif
   #ifdef POWELLNOF3INFF1TEST    /* skips test F3 <F1 */
   #else
       if (fptt < fp) { /* If extrapolated point is better, decide if we keep that new direction or not */
   #endif
         /* (x1 f1=fp), (x2 f2=*fret), (x3 f3=fptt), (xm fm) */
         /* From x1 (P0) distance of x2 is at h and x3 is 2h */
         /* Let f"(x2) be the 2nd derivative equal everywhere.  */
         /* Then the parabolic through (x1,f1), (x2,f2) and (x3,f3) */
         /* will reach at f3 = fm + h^2/2 f"m  ; f" = (f1 -2f2 +f3 ) / h**2 */
         /* Conditional for using this new direction is that mu^2 = (f1-2f2+f3)^2 /2 < del or directest <0 */
         /* also  lamda^2=(f1-f2)^2/mu² is a parasite solution of powell */
         /* For powell, inclusion of this average direction is only if t(del)<0 or del inbetween mu^2 and lambda^2 */
         /* t=2.0*(fp-2.0*(*fret)+fptt)*SQR(fp-(*fret)-del)-del*SQR(fp-fptt); */
         /*  Even if f3 <f1, directest can be negative and t >0 */
         /* mu² and del² are equal when f3=f1 */
                           /* f3 < f1 : mu² < del <= lambda^2 both test are equivalent */
                           /* f3 < f1 : mu² < lambda^2 < del then directtest is negative and powell t is positive */
                           /* f3 > f1 : lambda² < mu^2 < del then t is negative and directest >0  */
                           /* f3 > f1 : lambda² < del < mu^2 then t is positive and directest >0  */
   #ifdef NRCORIGINAL
         t=2.0*(fp-2.0*(*fret)+fptt)*SQR(fp-(*fret)-del)- del*SQR(fp-fptt); /* Original Numerical Recipes in C*/
   #else
         t=2.0*(fp-2.0*(*fret)+fptt)*SQR(fp-(*fret)-del); /* Intel compiler doesn't work on one line; bug reported */
         t= t- del*SQR(fp-fptt);
   #endif
         directest = fp-2.0*(*fret)+fptt - 2.0 * del; /* If delta was big enough we change it for a new direction */
   #ifdef DEBUG
         printf("t1= %.12lf, t2= %.12lf, t=%.12lf  directest=%.12lf\n", 2.0*(fp-2.0*(*fret)+fptt)*SQR(fp-(*fret)-del),del*SQR(fp-fptt),t,directest);
         fprintf(ficlog,"t1= %.12lf, t2= %.12lf, t=%.12lf directest=%.12lf\n", 2.0*(fp-2.0*(*fret)+fptt)*SQR(fp-(*fret)-del),del*SQR(fp-fptt),t,directest);
         printf("t3= %.12lf, t4= %.12lf, t3*= %.12lf, t4*= %.12lf\n",SQR(fp-(*fret)-del),SQR(fp-fptt),
                (fp-(*fret)-del)*(fp-(*fret)-del),(fp-fptt)*(fp-fptt));
         fprintf(ficlog,"t3= %.12lf, t4= %.12lf, t3*= %.12lf, t4*= %.12lf\n",SQR(fp-(*fret)-del),SQR(fp-fptt),
                (fp-(*fret)-del)*(fp-(*fret)-del),(fp-fptt)*(fp-fptt));
         printf("tt= %.12lf, t=%.12lf\n",2.0*(fp-2.0*(*fret)+fptt)*(fp-(*fret)-del)*(fp-(*fret)-del)-del*(fp-fptt)*(fp-fptt),t);
         fprintf(ficlog, "tt= %.12lf, t=%.12lf\n",2.0*(fp-2.0*(*fret)+fptt)*(fp-(*fret)-del)*(fp-(*fret)-del)-del*(fp-fptt)*(fp-fptt),t);
   #endif
   #ifdef POWELLORIGINAL
         if (t < 0.0) { /* Then we use it for new direction */
   #else
         if (directest*t < 0.0) { /* Contradiction between both tests */
                                   printf("directest= %.12lf (if <0 we include P0 Pn as new direction), t= %.12lf, f1= %.12lf,f2= %.12lf,f3= %.12lf, del= %.12lf\n",directest, t, fp,(*fret),fptt,del);
           printf("f1-2f2+f3= %.12lf, f1-f2-del= %.12lf, f1-f3= %.12lf\n",fp-2.0*(*fret)+fptt, fp -(*fret) -del, fp-fptt);
           fprintf(ficlog,"directest= %.12lf (if directest<0 or t<0 we include P0 Pn as new direction), t= %.12lf, f1= %.12lf,f2= %.12lf,f3= %.12lf, del= %.12lf\n",directest, t, fp,(*fret),fptt, del);
           fprintf(ficlog,"f1-2f2+f3= %.12lf, f1-f2-del= %.12lf, f1-f3= %.12lf\n",fp-2.0*(*fret)+fptt, fp -(*fret) -del, fp-fptt);
         } 
         if (directest < 0.0) { /* Then we use it for new direction */
   #endif
   #ifdef DEBUGLINMIN
           printf("Before linmin in direction P%d-P0\n",n);
           for (j=1;j<=n;j++) {
             printf(" Before xit[%d]= %12.7f p[%d]= %12.7f",j,xit[j],j,p[j]);
             fprintf(ficlog," Before xit[%d]= %12.7f p[%d]= %12.7f",j,xit[j],j,p[j]);
             if(j % ncovmodel == 0){
               printf("\n");
               fprintf(ficlog,"\n");
             }
           }
   #endif
   #ifdef LINMINORIGINAL
           linmin(p,xit,n,fret,func); /* computes minimum on the extrapolated direction: changes p and rescales xit.*/
   #else
           linmin(p,xit,n,fret,func,&flat); /* computes minimum on the extrapolated direction: changes p and rescales xit.*/
           flatdir[i]=flat; /* Function is vanishing in that direction i */
   #endif
           
   #ifdef DEBUGLINMIN
           for (j=1;j<=n;j++) { 
             printf("After xit[%d]= %12.7f p[%d]= %12.7f",j,xit[j],j,p[j]);
             fprintf(ficlog,"After xit[%d]= %12.7f p[%d]= %12.7f",j,xit[j],j,p[j]);
             if(j % ncovmodel == 0){
               printf("\n");
               fprintf(ficlog,"\n");
             }
           }
   #endif
           for (j=1;j<=n;j++) { 
             xi[j][ibig]=xi[j][n]; /* Replace direction with biggest decrease by last direction n */
             xi[j][n]=xit[j];      /* and this nth direction by the by the average p_0 p_n */
           }
   #ifdef LINMINORIGINAL
   #else
           for (j=1, flatd=0;j<=n;j++) {
             if(flatdir[j]>0)
               flatd++;
           }
           if(flatd >0){
             printf("%d flat directions: ",flatd);
             fprintf(ficlog,"%d flat directions :",flatd);
             for (j=1;j<=n;j++) { 
               if(flatdir[j]>0){
                 printf("%d ",j);
                 fprintf(ficlog,"%d ",j);
               }
             }
             printf("\n");
             fprintf(ficlog,"\n");
   #ifdef FLATSUP
             free_vector(xit,1,n); 
             free_vector(xits,1,n); 
             free_vector(ptt,1,n); 
             free_vector(pt,1,n); 
             return;
   #endif
           }
   #endif
           printf("Gaining to use new average direction of P0 P%d instead of biggest increase direction %d :\n",n,ibig);
           fprintf(ficlog,"Gaining to use new average direction of P0 P%d instead of biggest increase direction %d :\n",n,ibig);
           
   #ifdef DEBUG
           printf("Direction changed  last moved %d in place of ibig=%d, new last is the average:\n",n,ibig);
           fprintf(ficlog,"Direction changed  last moved %d in place of ibig=%d, new last is the average:\n",n,ibig);
           for(j=1;j<=n;j++){
             printf(" %lf",xit[j]);
             fprintf(ficlog," %lf",xit[j]);
           }
           printf("\n");
           fprintf(ficlog,"\n");
   #endif
         } /* end of t or directest negative */
   #ifdef POWELLNOF3INFF1TEST
   #else
         } /* end if (fptt < fp)  */
   #endif
   #ifdef NODIRECTIONCHANGEDUNTILNITER  /* No change in drections until some iterations are done */
       } /*NODIRECTIONCHANGEDUNTILNITER  No change in drections until some iterations are done */
   #else
   #endif
                   } /* loop iteration */ 
   } 
     
   /**** Prevalence limit (stable or period prevalence)  ****************/
     
     double **prevalim(double **prlim, int nlstate, double x[], double age, double **oldm, double **savm, double ftolpl, int *ncvyear, int ij, int nres)
     {
       /**< Computes the prevalence limit in each live state at age x and for covariate combination ij 
        *   (and selected quantitative values in nres)
        *  by left multiplying the unit
        *  matrix by transitions matrix until convergence is reached with precision ftolpl 
        * Wx= Wx-1 Px-1= Wx-2 Px-2 Px-1  = Wx-n Px-n ... Px-2 Px-1 I
        * Wx is row vector: population in state 1, population in state 2, population dead
        * or prevalence in state 1, prevalence in state 2, 0
        * newm is the matrix after multiplications, its rows are identical at a factor.
        * Inputs are the parameter, age, a tolerance for the prevalence limit ftolpl.
        * Output is prlim.
        * Initial matrix pimij 
        */
     /* {0.85204250825084937, 0.13044499163996345, 0.017512500109187184, */
     /* 0.090851990222114765, 0.88271245433047185, 0.026435555447413338, */
     /*  0,                   0                  , 1} */
     /*
      * and after some iteration: */
     /* {0.45504275246439968, 0.42731458730878791, 0.11764266022681241, */
     /*  0.45201005341706885, 0.42865420071559901, 0.11933574586733192, */
     /*  0,                   0                  , 1} */
     /* And prevalence by suppressing the deaths are close to identical rows in prlim: */
     /* {0.51571254859325999, 0.4842874514067399, */
     /*  0.51326036147820708, 0.48673963852179264} */
     /* If we start from prlim again, prlim tends to a constant matrix */
       
       int i, ii,j,k, k1;
     double *min, *max, *meandiff, maxmax,sumnew=0.;
     /* double **matprod2(); */ /* test */
     double **out, cov[NCOVMAX+1], **pmij(); /* **pmmij is a global variable feeded with oldms etc */
     double **newm;
     double agefin, delaymax=200. ; /* 100 Max number of years to converge */
     int ncvloop=0;
     int first=0;
     
     min=vector(1,nlstate);
     max=vector(1,nlstate);
     meandiff=vector(1,nlstate);
   
           /* Starting with matrix unity */
     for (ii=1;ii<=nlstate+ndeath;ii++)
       for (j=1;j<=nlstate+ndeath;j++){
         oldm[ii][j]=(ii==j ? 1.0 : 0.0);
       }
     
     cov[1]=1.;
     
     /* Even if hstepm = 1, at least one multiplication by the unit matrix */
     /* Start at agefin= age, computes the matrix of passage and loops decreasing agefin until convergence is reached */
     for(agefin=age-stepm/YEARM; agefin>=age-delaymax; agefin=agefin-stepm/YEARM){
       ncvloop++;
       newm=savm;
       /* Covariates have to be included here again */
       cov[2]=agefin;
        if(nagesqr==1){
         cov[3]= agefin*agefin;
        }
        /* Model(2)  V1 + V2 + V3 + V8 + V7*V8 + V5*V6 + V8*age + V3*age + age*age */
        /* total number of covariates of the model nbocc(+)+1 = 8 excepting constant and age and age*age */
        for(k1=1;k1<=cptcovt;k1++){ /* loop on model equation (including products) */ 
          if(Typevar[k1]==1){ /* A product with age */
            cov[2+nagesqr+k1]=precov[nres][k1]*cov[2];
          }else{
            cov[2+nagesqr+k1]=precov[nres][k1];
          }
        }/* End of loop on model equation */
        
   /* Start of old code (replaced by a loop on position in the model equation */
       /* for (k=1; k<=nsd;k++) { /\* For single dummy covariates only of the model *\/ */
       /*                  /\* Here comes the value of the covariate 'ij' after renumbering k with single dummy covariates *\/ */
       /*   /\* cov[2+nagesqr+TvarsDind[k]]=nbcode[TvarsD[k]][codtabm(ij,TvarsD[k])]; *\/ */
       /*   cov[2+nagesqr+TvarsDind[k]]=nbcode[TvarsD[k]][codtabm(ij,TnsdVar[TvarsD[k]])]; */
       /*   /\* model = 1 +age + V1*V3 + age*V1 + V2 + V1 + age*V2 + V3 + V3*age + V1*V2  */
       /*    * k                  1        2      3    4      5      6     7        8 */
       /*    *cov[]   1    2      3        4      5    6      7      8     9       10 */
       /*    *TypeVar[k]          2        1      0    0      1      0     1        2 */
       /*    *Dummy[k]            0        2      0    0      2      0     2        0 */
       /*    *Tvar[k]             4        1      2    1      2      3     3        5 */
       /*    *nsd=3                              (1)  (2)           (3) */
       /*    *TvarsD[nsd]                      [1]=2    1             3 */
       /*    *TnsdVar                          [2]=2 [1]=1         [3]=3 */
       /*    *TvarsDind[nsd](=k)               [1]=3 [2]=4         [3]=6 */
       /*    *Tage[]                  [1]=1                  [2]=2      [3]=3 */
       /*    *Tvard[]       [1][1]=1                                           [2][1]=1 */
       /*    *                   [1][2]=3                                           [2][2]=2 */
       /*    *Tprod[](=k)     [1]=1                                              [2]=8 */
       /*    *TvarsDp(=Tvar)   [1]=1            [2]=2             [3]=3          [4]=5 */
       /*    *TvarD (=k)       [1]=1            [2]=3 [3]=4       [3]=6          [4]=6 */
       /*    *TvarsDpType */
       /*    *si model= 1 + age + V3 + V2*age + V2 + V3*age */
       /*    * nsd=1              (1)           (2) */
       /*    *TvarsD[nsd]          3             2 */
       /*    *TnsdVar           (3)=1          (2)=2 */
       /*    *TvarsDind[nsd](=k)  [1]=1        [2]=3 */
       /*    *Tage[]                  [1]=2           [2]= 3    */
       /*    *\/ */
       /*   /\* cov[++k1]=nbcode[TvarsD[k]][codtabm(ij,k)]; *\/ */
       /*   /\* printf("prevalim Dummy combi=%d k=%d TvarsD[%d]=V%d TvarsDind[%d]=%d nbcode=%d cov=%lf codtabm(%d,Tvar[%d])=%d \n",ij,k, k, TvarsD[k],k,TvarsDind[k],nbcode[TvarsD[k]][codtabm(ij,k)],cov[2+nagesqr+TvarsDind[k]], ij, k, codtabm(ij,k)); *\/ */
       /* } */
       /* for (k=1; k<=nsq;k++) { /\* For single quantitative varying covariates only of the model *\/ */
       /*                  /\* Here comes the value of quantitative after renumbering k with single quantitative covariates *\/ */
       /*   /\* Tqresult[nres][result_position]= value of the variable at the result_position in the nres resultline                                 *\/ */
       /*   /\* cov[2+nagesqr+TvarsQind[k]]=Tqresult[nres][k]; *\/ */
       /*   cov[2+nagesqr+TvarsQind[k]]=Tqresult[nres][resultmodel[nres][k1]] */
       /*   /\* cov[++k1]=Tqresult[nres][k];  *\/ */
       /*   /\* printf("prevalim Quantitative k=%d  TvarsQind[%d]=%d, TvarsQ[%d]=V%d,Tqresult[%d][%d]=%f\n",k,k,TvarsQind[k],k,TvarsQ[k],nres,k,Tqresult[nres][k]); *\/ */
       /* } */
       /* for (k=1; k<=cptcovage;k++){  /\* For product with age *\/ */
       /*   if(Dummy[Tage[k]]==2){ /\* dummy with age *\/ */
       /*  cov[2+nagesqr+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(ij,Tvar[Tage[k]])]*cov[2]; */
       /*  /\* cov[++k1]=nbcode[Tvar[Tage[k]]][codtabm(ij,k)]*cov[2]; *\/ */
       /*   } else if(Dummy[Tage[k]]==3){ /\* quantitative with age *\/ */
       /*  cov[2+nagesqr+Tage[k]]=Tqresult[nres][k]; */
       /*  /\* cov[++k1]=Tqresult[nres][k];  *\/ */
       /*   } */
       /*   /\* printf("prevalim Age combi=%d k=%d  Tage[%d]=V%d Tqresult[%d][%d]=%f\n",ij,k,k,Tage[k],nres,k,Tqresult[nres][k]); *\/ */
       /* } */
       /* for (k=1; k<=cptcovprod;k++){ /\* For product without age *\/ */
       /*   /\* printf("prevalim Prod ij=%d k=%d  Tprod[%d]=%d Tvard[%d][1]=V%d, Tvard[%d][2]=V%d\n",ij,k,k,Tprod[k], k,Tvard[k][1], k,Tvard[k][2]); *\/ */
       /*   if(Dummy[Tvard[k][1]]==0){ */
       /*  if(Dummy[Tvard[k][2]]==0){ */
       /*    cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])] * nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][2])]; */
       /*    /\* cov[++k1]=nbcode[Tvard[k][1]][codtabm(ij,k)] * nbcode[Tvard[k][2]][codtabm(ij,k)]; *\/ */
       /*  }else{ */
       /*    cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])] * Tqresult[nres][k]; */
       /*    /\* cov[++k1]=nbcode[Tvard[k][1]][codtabm(ij,k)] * Tqresult[nres][k]; *\/ */
       /*  } */
       /*   }else{ */
       /*  if(Dummy[Tvard[k][2]]==0){ */
       /*    cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][2])] * Tqinvresult[nres][Tvard[k][1]]; */
       /*    /\* cov[++k1]=nbcode[Tvard[k][2]][codtabm(ij,k)] * Tqinvresult[nres][Tvard[k][1]]; *\/ */
       /*  }else{ */
       /*    cov[2+nagesqr+Tprod[k]]=Tqinvresult[nres][Tvard[k][1]]*  Tqinvresult[nres][Tvard[k][2]]; */
       /*    /\* cov[++k1]=Tqinvresult[nres][Tvard[k][1]]*  Tqinvresult[nres][Tvard[k][2]]; *\/ */
       /*  } */
       /*   } */
       /* } /\* End product without age *\/ */
   /* ENd of old code */
       /*printf("ij=%d cptcovprod=%d tvar=%d ", ij, cptcovprod, Tvar[1]);*/
       /*printf("ij=%d cov[3]=%lf cov[4]=%lf \n",ij, cov[3],cov[4]);*/
       /*printf("ij=%d cov[3]=%lf \n",ij, cov[3]);*/
       /* savm=pmij(pmmij,cov,ncovmodel,x,nlstate); */
       /* out=matprod2(newm, pmij(pmmij,cov,ncovmodel,x,nlstate),1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, oldm); /\* Bug Valgrind *\/ */
       /* age and covariate values of ij are in 'cov' */
       out=matprod2(newm, pmij(pmmij,cov,ncovmodel,x,nlstate),1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, oldm); /* Bug Valgrind */
       
       savm=oldm;
       oldm=newm;
   
       for(j=1; j<=nlstate; j++){
         max[j]=0.;
         min[j]=1.;
       }
       for(i=1;i<=nlstate;i++){
         sumnew=0;
         for(k=1; k<=ndeath; k++) sumnew+=newm[i][nlstate+k];
         for(j=1; j<=nlstate; j++){ 
           prlim[i][j]= newm[i][j]/(1-sumnew);
           max[j]=FMAX(max[j],prlim[i][j]);
           min[j]=FMIN(min[j],prlim[i][j]);
         }
       }
   
       maxmax=0.;
       for(j=1; j<=nlstate; j++){
         meandiff[j]=(max[j]-min[j])/(max[j]+min[j])*2.; /* mean difference for each column */
         maxmax=FMAX(maxmax,meandiff[j]);
         /* printf(" age= %d meandiff[%d]=%f, agefin=%d max[%d]=%f min[%d]=%f maxmax=%f\n", (int)age, j, meandiff[j],(int)agefin, j, max[j], j, min[j],maxmax); */
       } /* j loop */
       *ncvyear= (int)age- (int)agefin;
       /* printf("maxmax=%lf maxmin=%lf ncvloop=%d, age=%d, agefin=%d ncvyear=%d \n", maxmax, maxmin, ncvloop, (int)age, (int)agefin, *ncvyear); */
       if(maxmax < ftolpl){
         /* printf("maxmax=%lf ncvloop=%ld, age=%d, agefin=%d ncvyear=%d \n", maxmax, ncvloop, (int)age, (int)agefin, *ncvyear); */
         free_vector(min,1,nlstate);
         free_vector(max,1,nlstate);
         free_vector(meandiff,1,nlstate);
         return prlim;
       }
     } /* agefin loop */
       /* After some age loop it doesn't converge */
     if(!first){
       first=1;
       printf("Warning: the stable prevalence at age %d did not converge with the required precision (%g > ftolpl=%g) within %.d years and %d loops. Try to lower 'ftolpl'. Youngest age to start was %d=(%d-%d). Others in log file only...\n", (int)age, maxmax, ftolpl, *ncvyear, ncvloop, (int)(agefin+stepm/YEARM),  (int)(age-stepm/YEARM), (int)delaymax);
       fprintf(ficlog, "Warning: the stable prevalence at age %d did not converge with the required precision (%g > ftolpl=%g) within %.d years and %d loops. Try to lower 'ftolpl'. Youngest age to start was %d=(%d-%d).\n", (int)age, maxmax, ftolpl, *ncvyear, ncvloop, (int)(agefin+stepm/YEARM),  (int)(age-stepm/YEARM), (int)delaymax);
     }else if (first >=1 && first <10){
       fprintf(ficlog, "Warning: the stable prevalence at age %d did not converge with the required precision (%g > ftolpl=%g) within %.d years and %d loops. Try to lower 'ftolpl'. Youngest age to start was %d=(%d-%d).\n", (int)age, maxmax, ftolpl, *ncvyear, ncvloop, (int)(agefin+stepm/YEARM),  (int)(age-stepm/YEARM), (int)delaymax);
       first++;
     }else if (first ==10){
       fprintf(ficlog, "Warning: the stable prevalence at age %d did not converge with the required precision (%g > ftolpl=%g) within %.d years and %d loops. Try to lower 'ftolpl'. Youngest age to start was %d=(%d-%d).\n", (int)age, maxmax, ftolpl, *ncvyear, ncvloop, (int)(agefin+stepm/YEARM),  (int)(age-stepm/YEARM), (int)delaymax);
       printf("Warning: the stable prevalence dit not converge. This warning came too often, IMaCh will stop notifying, even in its log file. Look at the graphs to appreciate the non convergence.\n");
       fprintf(ficlog,"Warning: the stable prevalence no convergence; too many cases, giving up noticing, even in log file\n");
       first++;
     }
   
     /* Try to lower 'ftol', for example from 1.e-8 to 6.e-9.\n", ftolpl, (int)age, (int)delaymax, (int)agefin, ncvloop, (int)age-(int)agefin); */
     free_vector(min,1,nlstate);
     free_vector(max,1,nlstate);
     free_vector(meandiff,1,nlstate);
     
     return prlim; /* should not reach here */
   }
   
   
    /**** Back Prevalence limit (stable or period prevalence)  ****************/
   
    /* double **bprevalim(double **bprlim, double ***prevacurrent, int nlstate, double x[], double age, double ageminpar, double agemaxpar, double **oldm, double **savm, double **dnewm, double **doldm, double **dsavm, double ftolpl, int *ncvyear, int ij) */
    /* double **bprevalim(double **bprlim, double ***prevacurrent, int nlstate, double x[], double age, double **oldm, double **savm, double **dnewm, double **doldm, double **dsavm, double ftolpl, int *ncvyear, int ij) */
     double **bprevalim(double **bprlim, double ***prevacurrent, int nlstate, double x[], double age, double ftolpl, int *ncvyear, int ij, int nres)
   {
     /* Computes the prevalence limit in each live state at age x and for covariate combination ij (<=2**cptcoveff) by left multiplying the unit
        matrix by transitions matrix until convergence is reached with precision ftolpl */
     /* Wx= Wx-1 Px-1= Wx-2 Px-2 Px-1  = Wx-n Px-n ... Px-2 Px-1 I */
     /* Wx is row vector: population in state 1, population in state 2, population dead */
     /* or prevalence in state 1, prevalence in state 2, 0 */
     /* newm is the matrix after multiplications, its rows are identical at a factor */
     /* Initial matrix pimij */
     /* {0.85204250825084937, 0.13044499163996345, 0.017512500109187184, */
     /* 0.090851990222114765, 0.88271245433047185, 0.026435555447413338, */
     /*  0,                   0                  , 1} */
     /*
      * and after some iteration: */
     /* {0.45504275246439968, 0.42731458730878791, 0.11764266022681241, */
     /*  0.45201005341706885, 0.42865420071559901, 0.11933574586733192, */
     /*  0,                   0                  , 1} */
     /* And prevalence by suppressing the deaths are close to identical rows in prlim: */
     /* {0.51571254859325999, 0.4842874514067399, */
     /*  0.51326036147820708, 0.48673963852179264} */
     /* If we start from prlim again, prlim tends to a constant matrix */
   
     int i, ii,j,k, k1;
     int first=0;
     double *min, *max, *meandiff, maxmax,sumnew=0.;
     /* double **matprod2(); */ /* test */
     double **out, cov[NCOVMAX+1], **bmij();
     double **newm;
     double         **dnewm, **doldm, **dsavm;  /* for use */
     double         **oldm, **savm;  /* for use */
   
     double agefin, delaymax=200. ; /* 100 Max number of years to converge */
     int ncvloop=0;
     
     min=vector(1,nlstate);
     max=vector(1,nlstate);
     meandiff=vector(1,nlstate);
   
     dnewm=ddnewms; doldm=ddoldms; dsavm=ddsavms;
     oldm=oldms; savm=savms;
     
     /* Starting with matrix unity */
     for (ii=1;ii<=nlstate+ndeath;ii++)
       for (j=1;j<=nlstate+ndeath;j++){
         oldm[ii][j]=(ii==j ? 1.0 : 0.0);
       }
     
     cov[1]=1.;
     
     /* Even if hstepm = 1, at least one multiplication by the unit matrix */
     /* Start at agefin= age, computes the matrix of passage and loops decreasing agefin until convergence is reached */
     /* for(agefin=age+stepm/YEARM; agefin<=age+delaymax; agefin=agefin+stepm/YEARM){ /\* A changer en age *\/ */
     /* for(agefin=age; agefin<AGESUP; agefin=agefin+stepm/YEARM){ /\* A changer en age *\/ */
     for(agefin=age; agefin<FMIN(AGESUP,age+delaymax); agefin=agefin+stepm/YEARM){ /* A changer en age */
       ncvloop++;
       newm=savm; /* oldm should be kept from previous iteration or unity at start */
                   /* newm points to the allocated table savm passed by the function it can be written, savm could be reallocated */
       /* Covariates have to be included here again */
       cov[2]=agefin;
       if(nagesqr==1){
         cov[3]= agefin*agefin;;
       }
       for(k1=1;k1<=cptcovt;k1++){ /* loop on model equation (including products) */ 
         if(Typevar[k1]==1){ /* A product with age */
           cov[2+nagesqr+k1]=precov[nres][k1]*cov[2];
         }else{
           cov[2+nagesqr+k1]=precov[nres][k1];
         }
       }/* End of loop on model equation */
   
   /* Old code */ 
   
       /* for (k=1; k<=nsd;k++) { /\* For single dummy covariates only *\/ */
       /*                  /\* Here comes the value of the covariate 'ij' after renumbering k with single dummy covariates *\/ */
       /*   cov[2+nagesqr+TvarsDind[k]]=nbcode[TvarsD[k]][codtabm(ij,TvarsD[k])]; */
       /*   /\* printf("bprevalim Dummy agefin=%.0f combi=%d k=%d TvarsD[%d]=V%d TvarsDind[%d]=%d nbcode=%d cov[%d]=%lf codtabm(%d,Tvar[%d])=%d \n",agefin,ij,k, k, TvarsD[k],k,TvarsDind[k],nbcode[TvarsD[k]][codtabm(ij,k)],2+nagesqr+TvarsDind[k],cov[2+nagesqr+TvarsDind[k]], ij, k, codtabm(ij,k)); *\/ */
       /* } */
       /* /\* for (k=1; k<=cptcovn;k++) { *\/ */
       /* /\*   /\\* cov[2+nagesqr+k]=nbcode[Tvar[k]][codtabm(ij,Tvar[k])]; *\\/ *\/ */
       /* /\*   cov[2+nagesqr+k]=nbcode[Tvar[k]][codtabm(ij,k)]; *\/ */
       /* /\*   /\\* printf("prevalim ij=%d k=%d Tvar[%d]=%d nbcode=%d cov=%lf codtabm(%d,Tvar[%d])=%d \n",ij,k, k, Tvar[k],nbcode[Tvar[k]][codtabm(ij,Tvar[k])],cov[2+k], ij, k, codtabm(ij,Tvar[k])]); *\\/ *\/ */
       /* /\* } *\/ */
       /* for (k=1; k<=nsq;k++) { /\* For single varying covariates only *\/ */
       /*                  /\* Here comes the value of quantitative after renumbering k with single quantitative covariates *\/ */
       /*   cov[2+nagesqr+TvarsQind[k]]=Tqresult[nres][k];  */
       /*   /\* printf("prevalim Quantitative k=%d  TvarsQind[%d]=%d, TvarsQ[%d]=V%d,Tqresult[%d][%d]=%f\n",k,k,TvarsQind[k],k,TvarsQ[k],nres,k,Tqresult[nres][k]); *\/ */
       /* } */
       /* /\* for (k=1; k<=cptcovage;k++) cov[2+nagesqr+Tage[k]]=nbcode[Tvar[k]][codtabm(ij,k)]*cov[2]; *\/ */
       /* /\* for (k=1; k<=cptcovprod;k++) /\\* Useless *\\/ *\/ */
       /* /\*   /\\* cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])] * nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][2])]; *\\/ *\/ */
       /* /\*   cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,k)] * nbcode[Tvard[k][2]][codtabm(ij,k)]; *\/ */
       /* for (k=1; k<=cptcovage;k++){  /\* For product with age *\/ */
       /*   /\* if(Dummy[Tvar[Tage[k]]]== 2){ /\\* dummy with age *\\/ ERROR ???*\/ */
       /*   if(Dummy[Tage[k]]== 2){ /\* dummy with age *\/ */
       /*  cov[2+nagesqr+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(ij,Tvar[Tage[k]])]*cov[2]; */
       /*   } else if(Dummy[Tage[k]]== 3){ /\* quantitative with age *\/ */
       /*  cov[2+nagesqr+Tage[k]]=Tqresult[nres][k]; */
       /*   } */
       /*   /\* printf("prevalim Age combi=%d k=%d  Tage[%d]=V%d Tqresult[%d][%d]=%f\n",ij,k,k,Tage[k],nres,k,Tqresult[nres][k]); *\/ */
       /* } */
       /* for (k=1; k<=cptcovprod;k++){ /\* For product without age *\/ */
       /*   /\* printf("prevalim Prod ij=%d k=%d  Tprod[%d]=%d Tvard[%d][1]=V%d, Tvard[%d][2]=V%d\n",ij,k,k,Tprod[k], k,Tvard[k][1], k,Tvard[k][2]); *\/ */
       /*   if(Dummy[Tvard[k][1]]==0){ */
       /*  if(Dummy[Tvard[k][2]]==0){ */
       /*    cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])] * nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][2])]; */
       /*  }else{ */
       /*    cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])] * Tqresult[nres][k]; */
       /*  } */
       /*   }else{ */
       /*  if(Dummy[Tvard[k][2]]==0){ */
       /*    cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][2])] * Tqinvresult[nres][Tvard[k][1]]; */
       /*  }else{ */
       /*    cov[2+nagesqr+Tprod[k]]=Tqinvresult[nres][Tvard[k][1]]*  Tqinvresult[nres][Tvard[k][2]]; */
       /*  } */
       /*   } */
       /* } */
       
       /*printf("ij=%d cptcovprod=%d tvar=%d ", ij, cptcovprod, Tvar[1]);*/
       /*printf("ij=%d cov[3]=%lf cov[4]=%lf \n",ij, cov[3],cov[4]);*/
       /*printf("ij=%d cov[3]=%lf \n",ij, cov[3]);*/
       /* savm=pmij(pmmij,cov,ncovmodel,x,nlstate); */
       /* out=matprod2(newm, pmij(pmmij,cov,ncovmodel,x,nlstate),1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, oldm); /\* Bug Valgrind *\/ */
                   /* ij should be linked to the correct index of cov */
                   /* age and covariate values ij are in 'cov', but we need to pass
                    * ij for the observed prevalence at age and status and covariate
                    * number:  prevacurrent[(int)agefin][ii][ij]
                    */
       /* out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, bmij(pmmij,cov,ncovmodel,x,nlstate,prevacurrent, ageminpar, agemaxpar, dnewm, doldm, dsavm,ij)); /\* Bug Valgrind *\/ */
       /* out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, bmij(pmmij,cov,ncovmodel,x,nlstate,prevacurrent, dnewm, doldm, dsavm,ij)); /\* Bug Valgrind *\/ */
       out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, bmij(pmmij,cov,ncovmodel,x,nlstate,prevacurrent,ij)); /* Bug Valgrind */
       /* if((int)age == 86 || (int)age == 87){ */
       /*   printf(" Backward prevalim age=%d agefin=%d \n", (int) age, (int) agefin); */
       /*   for(i=1; i<=nlstate+ndeath; i++) { */
       /*  printf("%d newm= ",i); */
       /*  for(j=1;j<=nlstate+ndeath;j++) { */
       /*    printf("%f ",newm[i][j]); */
       /*  } */
       /*  printf("oldm * "); */
       /*  for(j=1;j<=nlstate+ndeath;j++) { */
       /*    printf("%f ",oldm[i][j]); */
       /*  } */
       /*  printf(" bmmij "); */
       /*  for(j=1;j<=nlstate+ndeath;j++) { */
       /*    printf("%f ",pmmij[i][j]); */
       /*  } */
       /*  printf("\n"); */
       /*   } */
       /* } */
       savm=oldm;
       oldm=newm;
   
       for(j=1; j<=nlstate; j++){
         max[j]=0.;
         min[j]=1.;
       }
       for(j=1; j<=nlstate; j++){ 
         for(i=1;i<=nlstate;i++){
           /* bprlim[i][j]= newm[i][j]/(1-sumnew); */
           bprlim[i][j]= newm[i][j];
           max[i]=FMAX(max[i],bprlim[i][j]); /* Max in line */
           min[i]=FMIN(min[i],bprlim[i][j]);
         }
       }
                   
       maxmax=0.;
       for(i=1; i<=nlstate; i++){
         meandiff[i]=(max[i]-min[i])/(max[i]+min[i])*2.; /* mean difference for each column, could be nan! */
         maxmax=FMAX(maxmax,meandiff[i]);
         /* printf("Back age= %d meandiff[%d]=%f, agefin=%d max[%d]=%f min[%d]=%f maxmax=%f\n", (int)age, i, meandiff[i],(int)agefin, i, max[i], i, min[i],maxmax); */
       } /* i loop */
       *ncvyear= -( (int)age- (int)agefin);
       /* printf("Back maxmax=%lf ncvloop=%d, age=%d, agefin=%d ncvyear=%d \n", maxmax, ncvloop, (int)age, (int)agefin, *ncvyear); */
       if(maxmax < ftolpl){
         /* printf("OK Back maxmax=%lf ncvloop=%d, age=%d, agefin=%d ncvyear=%d \n", maxmax, ncvloop, (int)age, (int)agefin, *ncvyear); */
         free_vector(min,1,nlstate);
         free_vector(max,1,nlstate);
         free_vector(meandiff,1,nlstate);
         return bprlim;
       }
     } /* agefin loop */
       /* After some age loop it doesn't converge */
     if(!first){
       first=1;
       printf("Warning: the back stable prevalence at age %d did not converge with the required precision (%g > ftolpl=%g) within %.0f years. Try to lower 'ftolpl'. Others in log file only...\n\
   Oldest age to start was %d-%d=%d, ncvloop=%d, ncvyear=%d\n", (int)age, maxmax, ftolpl, delaymax, (int)age, (int)delaymax, (int)agefin, ncvloop, *ncvyear);
     }
     fprintf(ficlog,"Warning: the back stable prevalence at age %d did not converge with the required precision (%g > ftolpl=%g) within %.0f years. Try to lower 'ftolpl'. \n\
   Oldest age to start was %d-%d=%d, ncvloop=%d, ncvyear=%d\n", (int)age, maxmax, ftolpl, delaymax, (int)age, (int)delaymax, (int)agefin, ncvloop, *ncvyear);
     /* Try to lower 'ftol', for example from 1.e-8 to 6.e-9.\n", ftolpl, (int)age, (int)delaymax, (int)agefin, ncvloop, (int)age-(int)agefin); */
     free_vector(min,1,nlstate);
     free_vector(max,1,nlstate);
     free_vector(meandiff,1,nlstate);
     
     return bprlim; /* should not reach here */
   }
   
   /*************** transition probabilities ***************/ 
   
   double **pmij(double **ps, double *cov, int ncovmodel, double *x, int nlstate )
   {
     /* According to parameters values stored in x and the covariate's values stored in cov,
        computes the probability to be observed in state j (after stepm years) being in state i by appying the
        model to the ncovmodel covariates (including constant and age).
        lnpijopii=ln(pij/pii)= aij+bij*age+cij*v1+dij*v2+... = sum_nc=1^ncovmodel xij(nc)*cov[nc]
        and, according on how parameters are entered, the position of the coefficient xij(nc) of the
        ncth covariate in the global vector x is given by the formula:
        j<i nc+((i-1)*(nlstate+ndeath-1)+j-1)*ncovmodel
        j>=i nc + ((i-1)*(nlstate+ndeath-1)+(j-2))*ncovmodel
        Computes ln(pij/pii) (lnpijopii), deduces pij/pii by exponentiation,
        sums on j different of i to get 1-pii/pii, deduces pii, and then all pij.
        Outputs ps[i][j] or probability to be observed in j being in i according to
        the values of the covariates cov[nc] and corresponding parameter values x[nc+shiftij]
        Sum on j ps[i][j] should equal to 1.
     */
     double s1, lnpijopii;
     /*double t34;*/
     int i,j, nc, ii, jj;
   
     for(i=1; i<= nlstate; i++){
       for(j=1; j<i;j++){
         for (nc=1, lnpijopii=0.;nc <=ncovmodel; nc++){
           /*lnpijopii += param[i][j][nc]*cov[nc];*/
           lnpijopii += x[nc+((i-1)*(nlstate+ndeath-1)+j-1)*ncovmodel]*cov[nc];
           /*       printf("Int j<i s1=%.17e, lnpijopii=%.17e\n",s1,lnpijopii); */
         }
         ps[i][j]=lnpijopii; /* In fact ln(pij/pii) */
         /* printf("Debug pmij() i=%d j=%d nc=%d s1=%.17f, lnpijopii=%.17f\n",i,j,nc, s1,lnpijopii); */
       }
       for(j=i+1; j<=nlstate+ndeath;j++){
         for (nc=1, lnpijopii=0.;nc <=ncovmodel; nc++){
           /*lnpijopii += x[(i-1)*nlstate*ncovmodel+(j-2)*ncovmodel+nc+(i-1)*(ndeath-1)*ncovmodel]*cov[nc];*/
           lnpijopii += x[nc + ((i-1)*(nlstate+ndeath-1)+(j-2))*ncovmodel]*cov[nc];
           /*        printf("Int j>i s1=%.17e, lnpijopii=%.17e %lx %lx\n",s1,lnpijopii,s1,lnpijopii); */
         }
         ps[i][j]=lnpijopii; /* In fact ln(pij/pii) */
         /* printf("Debug pmij() i=%d j=%d nc=%d s1=%.17f, lnpijopii=%.17f\n",i,j,nc, s1,lnpijopii); */
       }
     }
     
     for(i=1; i<= nlstate; i++){
       s1=0;
       for(j=1; j<i; j++){
         s1+=exp(ps[i][j]); /* In fact sums pij/pii */
         /* printf("debug1 %d %d ps=%lf exp(ps)=%lf s1+=%lf\n",i,j,ps[i][j],exp(ps[i][j]),s1); */
       }
       for(j=i+1; j<=nlstate+ndeath; j++){
         s1+=exp(ps[i][j]); /* In fact sums pij/pii */
         /* printf("debug2 %d %d ps=%lf exp(ps)=%lf s1+=%lf\n",i,j,ps[i][j],exp(ps[i][j]),s1); */
       }
       /* s1= sum_{j<>i} pij/pii=(1-pii)/pii and thus pii is known from s1 */
       ps[i][i]=1./(s1+1.);
       /* Computing other pijs */
       for(j=1; j<i; j++)
         ps[i][j]= exp(ps[i][j])*ps[i][i];/* Bug valgrind */
       for(j=i+1; j<=nlstate+ndeath; j++)
         ps[i][j]= exp(ps[i][j])*ps[i][i];
       /* ps[i][nlstate+1]=1.-s1- ps[i][i];*/ /* Sum should be 1 */
     } /* end i */
     
     for(ii=nlstate+1; ii<= nlstate+ndeath; ii++){
       for(jj=1; jj<= nlstate+ndeath; jj++){
         ps[ii][jj]=0;
         ps[ii][ii]=1;
       }
     }
   
   
     /* for(ii=1; ii<= nlstate+ndeath; ii++){ */
     /*   for(jj=1; jj<= nlstate+ndeath; jj++){ */
     /*    printf(" pmij  ps[%d][%d]=%lf ",ii,jj,ps[ii][jj]); */
     /*   } */
     /*   printf("\n "); */
     /* } */
     /* printf("\n ");printf("%lf ",cov[2]);*/
     /*
       for(i=1; i<= npar; i++) printf("%f ",x[i]);
                   goto end;*/
     return ps; /* Pointer is unchanged since its call */
   }
   
   /*************** backward transition probabilities ***************/ 
   
    /* double **bmij(double **ps, double *cov, int ncovmodel, double *x, int nlstate,  double ***prevacurrent, double ageminpar, double agemaxpar, double ***dnewm, double **doldm, double **dsavm, int ij ) */
   /* double **bmij(double **ps, double *cov, int ncovmodel, double *x, int nlstate,  double ***prevacurrent, double ***dnewm, double **doldm, double **dsavm, int ij ) */
    double **bmij(double **ps, double *cov, int ncovmodel, double *x, int nlstate,  double ***prevacurrent, int ij )
   {
     /* Computes the backward probability at age agefin, cov[2], and covariate combination 'ij'. In fact cov is already filled and x too.
      * Call to pmij(cov and x), call to cross prevalence, sums and inverses, left multiply, and returns in **ps as well as **bmij.
      */
     int i, ii, j,k;
     
     double **out, **pmij();
     double sumnew=0.;
     double agefin;
     double k3=0.; /* constant of the w_x diagonal matrix (in order for B to sum to 1 even for death state) */
     double **dnewm, **dsavm, **doldm;
     double **bbmij;
     
     doldm=ddoldms; /* global pointers */
     dnewm=ddnewms;
     dsavm=ddsavms;
   
     /* Debug */
     /* printf("Bmij ij=%d, cov[2}=%f\n", ij, cov[2]); */
     agefin=cov[2];
     /* Bx = Diag(w_x) P_x Diag(Sum_i w^i_x p^ij_x */
     /* bmij *//* age is cov[2], ij is included in cov, but we need for
        the observed prevalence (with this covariate ij) at beginning of transition */
     /* dsavm=pmij(pmmij,cov,ncovmodel,x,nlstate); */
   
     /* P_x */
     pmmij=pmij(pmmij,cov,ncovmodel,x,nlstate); /*This is forward probability from agefin to agefin + stepm *//* Bug valgrind */
     /* outputs pmmij which is a stochastic matrix in row */
   
     /* Diag(w_x) */
     /* Rescaling the cross-sectional prevalence: Problem with prevacurrent which can be zero */
     sumnew=0.;
     /*for (ii=1;ii<=nlstate+ndeath;ii++){*/
     for (ii=1;ii<=nlstate;ii++){ /* Only on live states */
       /* printf(" agefin=%d, ii=%d, ij=%d, prev=%f\n",(int)agefin,ii, ij, prevacurrent[(int)agefin][ii][ij]); */
       sumnew+=prevacurrent[(int)agefin][ii][ij];
     }
     if(sumnew >0.01){  /* At least some value in the prevalence */
       for (ii=1;ii<=nlstate+ndeath;ii++){
         for (j=1;j<=nlstate+ndeath;j++)
           doldm[ii][j]=(ii==j ? prevacurrent[(int)agefin][ii][ij]/sumnew : 0.0);
       }
     }else{
       for (ii=1;ii<=nlstate+ndeath;ii++){
         for (j=1;j<=nlstate+ndeath;j++)
         doldm[ii][j]=(ii==j ? 1./nlstate : 0.0);
       }
       /* if(sumnew <0.9){ */
       /*   printf("Problem internal bmij B: sum on i wi <0.9: j=%d, sum_i wi=%lf,agefin=%d\n",j,sumnew, (int)agefin); */
       /* } */
     }
     k3=0.0;  /* We put the last diagonal to 0 */
     for (ii=nlstate+1;ii<=nlstate+ndeath;ii++){
         doldm[ii][ii]= k3;
     }
     /* End doldm, At the end doldm is diag[(w_i)] */
     
     /* Left product of this diag matrix by pmmij=Px (dnewm=dsavm*doldm): diag[(w_i)*Px */
     bbmij=matprod2(dnewm, doldm,1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, pmmij); /* was a Bug Valgrind */
   
     /* Diag(Sum_i w^i_x p^ij_x, should be the prevalence at age x+stepm */
     /* w1 p11 + w2 p21 only on live states N1./N..*N11/N1. + N2./N..*N21/N2.=(N11+N21)/N..=N.1/N.. */
     for (j=1;j<=nlstate+ndeath;j++){
       sumnew=0.;
       for (ii=1;ii<=nlstate;ii++){
         /* sumnew+=dsavm[ii][j]*prevacurrent[(int)agefin][ii][ij]; */
         sumnew+=pmmij[ii][j]*doldm[ii][ii]; /* Yes prevalence at beginning of transition */
       } /* sumnew is (N11+N21)/N..= N.1/N.. = sum on i of w_i pij */
       for (ii=1;ii<=nlstate+ndeath;ii++){
           /* if(agefin >= agemaxpar && agefin <= agemaxpar+stepm/YEARM){ */
           /*      dsavm[ii][j]=(ii==j ? 1./sumnew : 0.0); */
           /* }else if(agefin >= agemaxpar+stepm/YEARM){ */
           /*      dsavm[ii][j]=(ii==j ? 1./sumnew : 0.0); */
           /* }else */
         dsavm[ii][j]=(ii==j ? 1./sumnew : 0.0);
       } /*End ii */
     } /* End j, At the end dsavm is diag[1/(w_1p1i+w_2 p2i)] for ALL states even if the sum is only for live states */
   
     ps=matprod2(ps, dnewm,1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, dsavm); /* was a Bug Valgrind */
     /* ps is now diag[w_i] * Px * diag [1/(w_1p1i+w_2 p2i)] */
     /* end bmij */
     return ps; /*pointer is unchanged */
   }
   /*************** transition probabilities ***************/ 
   
   double **bpmij(double **ps, double *cov, int ncovmodel, double *x, int nlstate )
   {
     /* According to parameters values stored in x and the covariate's values stored in cov,
        computes the probability to be observed in state j being in state i by appying the
        model to the ncovmodel covariates (including constant and age).
        lnpijopii=ln(pij/pii)= aij+bij*age+cij*v1+dij*v2+... = sum_nc=1^ncovmodel xij(nc)*cov[nc]
        and, according on how parameters are entered, the position of the coefficient xij(nc) of the
        ncth covariate in the global vector x is given by the formula:
        j<i nc+((i-1)*(nlstate+ndeath-1)+j-1)*ncovmodel
        j>=i nc + ((i-1)*(nlstate+ndeath-1)+(j-2))*ncovmodel
        Computes ln(pij/pii) (lnpijopii), deduces pij/pii by exponentiation,
        sums on j different of i to get 1-pii/pii, deduces pii, and then all pij.
        Outputs ps[i][j] the probability to be observed in j being in j according to
        the values of the covariates cov[nc] and corresponding parameter values x[nc+shiftij]
     */
     double s1, lnpijopii;
     /*double t34;*/
     int i,j, nc, ii, jj;
   
     for(i=1; i<= nlstate; i++){
       for(j=1; j<i;j++){
         for (nc=1, lnpijopii=0.;nc <=ncovmodel; nc++){
           /*lnpijopii += param[i][j][nc]*cov[nc];*/
           lnpijopii += x[nc+((i-1)*(nlstate+ndeath-1)+j-1)*ncovmodel]*cov[nc];
           /*       printf("Int j<i s1=%.17e, lnpijopii=%.17e\n",s1,lnpijopii); */
         }
         ps[i][j]=lnpijopii; /* In fact ln(pij/pii) */
         /*        printf("s1=%.17e, lnpijopii=%.17e\n",s1,lnpijopii); */
       }
       for(j=i+1; j<=nlstate+ndeath;j++){
         for (nc=1, lnpijopii=0.;nc <=ncovmodel; nc++){
           /*lnpijopii += x[(i-1)*nlstate*ncovmodel+(j-2)*ncovmodel+nc+(i-1)*(ndeath-1)*ncovmodel]*cov[nc];*/
           lnpijopii += x[nc + ((i-1)*(nlstate+ndeath-1)+(j-2))*ncovmodel]*cov[nc];
           /*        printf("Int j>i s1=%.17e, lnpijopii=%.17e %lx %lx\n",s1,lnpijopii,s1,lnpijopii); */
         }
         ps[i][j]=lnpijopii; /* In fact ln(pij/pii) */
       }
     }
     
     for(i=1; i<= nlstate; i++){
       s1=0;
       for(j=1; j<i; j++){
         s1+=exp(ps[i][j]); /* In fact sums pij/pii */
         /*printf("debug1 %d %d ps=%lf exp(ps)=%lf s1+=%lf\n",i,j,ps[i][j],exp(ps[i][j]),s1); */
       }
       for(j=i+1; j<=nlstate+ndeath; j++){
         s1+=exp(ps[i][j]); /* In fact sums pij/pii */
         /*printf("debug2 %d %d ps=%lf exp(ps)=%lf s1+=%lf\n",i,j,ps[i][j],exp(ps[i][j]),s1); */
       }
       /* s1= sum_{j<>i} pij/pii=(1-pii)/pii and thus pii is known from s1 */
       ps[i][i]=1./(s1+1.);
       /* Computing other pijs */
       for(j=1; j<i; j++)
         ps[i][j]= exp(ps[i][j])*ps[i][i];
       for(j=i+1; j<=nlstate+ndeath; j++)
         ps[i][j]= exp(ps[i][j])*ps[i][i];
       /* ps[i][nlstate+1]=1.-s1- ps[i][i];*/ /* Sum should be 1 */
     } /* end i */
     
     for(ii=nlstate+1; ii<= nlstate+ndeath; ii++){
       for(jj=1; jj<= nlstate+ndeath; jj++){
         ps[ii][jj]=0;
         ps[ii][ii]=1;
       }
     }
     /* Added for prevbcast */ /* Transposed matrix too */
     for(jj=1; jj<= nlstate+ndeath; jj++){
       s1=0.;
       for(ii=1; ii<= nlstate+ndeath; ii++){
         s1+=ps[ii][jj];
       }
       for(ii=1; ii<= nlstate; ii++){
         ps[ii][jj]=ps[ii][jj]/s1;
       }
     }
     /* Transposition */
     for(jj=1; jj<= nlstate+ndeath; jj++){
       for(ii=jj; ii<= nlstate+ndeath; ii++){
         s1=ps[ii][jj];
         ps[ii][jj]=ps[jj][ii];
         ps[jj][ii]=s1;
       }
     }
     /* for(ii=1; ii<= nlstate+ndeath; ii++){ */
     /*   for(jj=1; jj<= nlstate+ndeath; jj++){ */
     /*    printf(" pmij  ps[%d][%d]=%lf ",ii,jj,ps[ii][jj]); */
     /*   } */
     /*   printf("\n "); */
     /* } */
     /* printf("\n ");printf("%lf ",cov[2]);*/
     /*
       for(i=1; i<= npar; i++) printf("%f ",x[i]);
       goto end;*/
     return ps;
   }
   
   
   /**************** Product of 2 matrices ******************/
   
   double **matprod2(double **out, double **in,int nrl, int nrh, int ncl, int nch, int ncolol, int ncoloh, double **b)
   {
     /* Computes the matrix product of in(1,nrh-nrl+1)(1,nch-ncl+1) times
        b(1,nch-ncl+1)(1,ncoloh-ncolol+1) into out(...) */
     /* in, b, out are matrice of pointers which should have been initialized 
        before: only the contents of out is modified. The function returns
        a pointer to pointers identical to out */
     int i, j, k;
     for(i=nrl; i<= nrh; i++)
       for(k=ncolol; k<=ncoloh; k++){
         out[i][k]=0.;
         for(j=ncl; j<=nch; j++)
           out[i][k] +=in[i][j]*b[j][k];
       }
     return out;
   }
   
   
   /************* Higher Matrix Product ***************/
   
   double ***hpxij(double ***po, int nhstepm, double age, int hstepm, double *x, int nlstate, int stepm, double **oldm, double **savm, int ij, int nres )
   {
     /* Computes the transition matrix starting at age 'age' and dummies values in each resultline (loop on ij to find the corresponding combination) to over 
        'nhstepm*hstepm*stepm' months (i.e. until
        age (in years)  age+nhstepm*hstepm*stepm/12) by multiplying 
        nhstepm*hstepm matrices. 
        Output is stored in matrix po[i][j][h] for h every 'hstepm' step 
        (typically every 2 years instead of every month which is too big 
        for the memory).
        Model is determined by parameters x and covariates have to be 
        included manually here. 
   
        */
   
     int i, j, d, h, k, k1;
     double **out, cov[NCOVMAX+1];
     double **newm;
     double agexact;
     double agebegin, ageend;
   
     /* Hstepm could be zero and should return the unit matrix */
     for (i=1;i<=nlstate+ndeath;i++)
       for (j=1;j<=nlstate+ndeath;j++){
         oldm[i][j]=(i==j ? 1.0 : 0.0);
         po[i][j][0]=(i==j ? 1.0 : 0.0);
       }
     /* Even if hstepm = 1, at least one multiplication by the unit matrix */
     for(h=1; h <=nhstepm; h++){
       for(d=1; d <=hstepm; d++){
         newm=savm;
         /* Covariates have to be included here again */
         cov[1]=1.;
         agexact=age+((h-1)*hstepm + (d-1))*stepm/YEARM; /* age just before transition */
         cov[2]=agexact;
         if(nagesqr==1){
           cov[3]= agexact*agexact;
         }
         /* Model(2)  V1 + V2 + V3 + V8 + V7*V8 + V5*V6 + V8*age + V3*age + age*age */
         /* total number of covariates of the model nbocc(+)+1 = 8 excepting constant and age and age*age */
         for(k1=1;k1<=cptcovt;k1++){ /* loop on model equation (including products) */ 
           if(Typevar[k1]==1){ /* A product with age */
             cov[2+nagesqr+k1]=precov[nres][k1]*cov[2];
           }else{
             cov[2+nagesqr+k1]=precov[nres][k1];
           }
         }/* End of loop on model equation */
           /* Old code */ 
   /*      if( Dummy[k1]==0 && Typevar[k1]==0 ){ /\* Single dummy  *\/ */
   /* /\*     V(Tvarsel)=Tvalsel=Tresult[nres][pos](value); V(Tvresult[nres][pos] (variable): V(variable)=value) *\/ */
   /* /\*       for (k=1; k<=nsd;k++) { /\\* For single dummy covariates only *\\/ *\/ */
   /* /\* /\\* Here comes the value of the covariate 'ij' after renumbering k with single dummy covariates *\\/ *\/ */
   /*      /\* codtabm(ij,k)  (1 & (ij-1) >> (k-1))+1 *\/ */
   /* /\*             V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 *\/ */
   /* /\*    k        1  2   3   4     5    6    7     8    9 *\/ */
   /* /\*Tvar[k]=     5  4   3   6     5    2    7     1    1 *\/ */
   /* /\*    nsd         1   2                              3 *\/ /\* Counting single dummies covar fixed or tv *\/ */
   /* /\*TvarsD[nsd]     4   3                              1 *\/ /\* ID of single dummy cova fixed or timevary*\/ */
   /* /\*TvarsDind[k]    2   3                              9 *\/ /\* position K of single dummy cova *\/ */
   /*        /\* cov[2+nagesqr+TvarsDind[k]]=nbcode[TvarsD[k]][codtabm(ij,k)];or [codtabm(ij,TnsdVar[TvarsD[k]] *\/ */
   /*        cov[2+nagesqr+k1]=Tresult[nres][resultmodel[nres][k1]]; */
   /*        /\* printf("hpxij Dummy combi=%d k=%d TvarsD[%d]=V%d TvarsDind[%d]=%d nbcode=%d cov=%lf codtabm(%d,TnsdVar[TvarsD[%d])=%d \n",ij,k, k, TvarsD[k],k,TvarsDind[k],nbcode[TvarsD[k]][codtabm(ij,TnsdVar[TvarsD[k]])],cov[2+nagesqr+TvarsDind[k]], ij, k, codtabm(ij,TnsdVar[TvarsD[k]])); *\/ */
   /*        printf("hpxij Dummy combi=%d k1=%d Tvar[%d]=V%d cov[2+%d+%d]=%lf resultmodel[nres][%d]=%d nres/nresult=%d/%d \n",ij,k1,k1, Tvar[k1],nagesqr,k1,cov[2+nagesqr+k1],k1,resultmodel[nres][k1],nres,nresult); */
   /*        printf("hpxij new Dummy precov[nres=%d][k1=%d]=%.4f\n", nres, k1, precov[nres][k1]); */
   /*      }else if( Dummy[k1]==1 && Typevar[k1]==0 ){ /\* Single quantitative variables  *\/ */
   /*        /\* resultmodel[nres][k1]=k3: k1th position in the model correspond to the k3 position in the resultline *\/ */
   /*        cov[2+nagesqr+k1]=Tqresult[nres][resultmodel[nres][k1]];  */
   /*        /\* for (k=1; k<=nsq;k++) { /\\* For single varying covariates only *\\/ *\/ */
   /*        /\*   /\\* Here comes the value of quantitative after renumbering k with single quantitative covariates *\\/ *\/ */
   /*        /\*   cov[2+nagesqr+TvarsQind[k]]=Tqresult[nres][k]; *\/ */
   /*        printf("hPxij Quantitative k1=%d resultmodel[nres][%d]=%d,Tqresult[%d][%d]=%f\n",k1,k1,resultmodel[nres][k1],nres,resultmodel[nres][k1],Tqresult[nres][resultmodel[nres][k1]]); */
   /*        printf("hpxij new Quanti precov[nres=%d][k1=%d]=%.4f\n", nres, k1, precov[nres][k1]); */
   /*      }else if( Dummy[k1]==2 ){ /\* For dummy with age product *\/ */
   /*        /\* Tvar[k1] Variable in the age product age*V1 is 1 *\/ */
   /*        /\* [Tinvresult[nres][V1] is its value in the resultline nres *\/ */
   /*        cov[2+nagesqr+k1]=TinvDoQresult[nres][Tvar[k1]]*cov[2]; */
   /*        printf("DhPxij Dummy with age k1=%d Tvar[%d]=%d TinvDoQresult[nres=%d][%d]=%.f age=%.2f,cov[2+%d+%d]=%.3f\n",k1,k1,Tvar[k1],nres,TinvDoQresult[nres][Tvar[k1]],cov[2],nagesqr,k1,cov[2+nagesqr+k1]); */
   /*        printf("hpxij new Dummy with age product precov[nres=%d][k1=%d]=%.4f * age=%.2f\n", nres, k1, precov[nres][k1], cov[2]); */
   
   /*        /\* cov[2+nagesqr+k1]=Tresult[nres][resultmodel[nres][k1]];    *\/ */
   /*        /\* for (k=1; k<=cptcovage;k++){ /\\* For product with age V1+V1*age +V4 +age*V3 *\\/ *\/ */
   /*        /\* 1+2 Tage[1]=2 TVar[2]=1 Dummy[2]=2, Tage[2]=4 TVar[4]=3 Dummy[4]=3 quant*\/ */
   /*        /\* *\/ */
   /* /\*             V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 *\/ */
   /* /\*    k        1  2   3   4     5    6    7     8    9 *\/ */
   /* /\*Tvar[k]=     5  4   3   6     5    2    7     1    1 *\/ */
   /* /\*cptcovage=2                   1               2      *\/ */
   /* /\*Tage[k]=                      5               8      *\/   */
   /*      }else if( Dummy[k1]==3 ){ /\* For quant with age product *\/ */
   /*        cov[2+nagesqr+k1]=Tresult[nres][resultmodel[nres][k1]];        */
   /*        printf("QhPxij Quant with age k1=%d resultmodel[nres][%d]=%d,Tqresult[%d][%d]=%f\n",k1,k1,resultmodel[nres][k1],nres,resultmodel[nres][k1],Tqresult[nres][resultmodel[nres][k1]]); */
   /*        printf("hpxij new Quanti with age product precov[nres=%d][k1=%d] * age=%.2f\n", nres, k1, precov[nres][k1], cov[2]); */
   /*        /\* if(Dummy[Tage[k]]== 2){ /\\* dummy with age *\\/ *\/ */
   /*        /\* /\\* if(Dummy[Tvar[Tage[k]]]== 2){ /\\\* dummy with age *\\\/ *\\/ *\/ */
   /*        /\*   /\\* cov[2+nagesqr+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(ij,k)]*cov[2]; *\\/ *\/ */
   /*        /\*   /\\* cov[2+nagesqr+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(ij,TnsdVar[TvarsD[Tvar[Tage[k]]]])]*cov[2]; *\\/ *\/ */
   /*        /\*   cov[2+nagesqr+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(ij,TnsdVar[TvarsD[Tvar[Tage[k]]]])]*cov[2]; *\/ */
   /*        /\*   printf("hPxij Age combi=%d k=%d cptcovage=%d Tage[%d]=%d Tvar[Tage[%d]]=V%d nbcode[Tvar[Tage[k]]][codtabm(ij,TnsdVar[Tvar[Tage[k]]]])]=%d nres=%d\n",ij,k,cptcovage,k,Tage[k],k,Tvar[Tage[k]], nbcode[Tvar[Tage[k]]][codtabm(ij,TnsdVar[Tvar[Tage[k]]])],nres); *\/ */
   /*        /\* } else if(Dummy[Tage[k]]== 3){ /\\* quantitative with age *\\/ *\/ */
   /*        /\*   cov[2+nagesqr+Tage[k]]=Tqresult[nres][k]; *\/ */
   /*        /\* } *\/ */
   /*        /\* printf("hPxij Age combi=%d k=%d  Tage[%d]=V%d Tqresult[%d][%d]=%f\n",ij,k,k,Tage[k],nres,k,Tqresult[nres][k]); *\/ */
   /*      }else if(Typevar[k1]==2 ){ /\* For product (not with age) *\/ */
   /* /\*       for (k=1; k<=cptcovprod;k++){ /\\*  For product without age *\\/ *\/ */
   /* /\* /\\*             V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 *\\/ *\/ */
   /* /\* /\\*    k        1  2   3   4     5    6    7     8    9 *\\/ *\/ */
   /* /\* /\\*Tvar[k]=     5  4   3   6     5    2    7     1    1 *\\/ *\/ */
   /* /\* /\\*cptcovprod=1            1               2            *\\/ *\/ */
   /* /\* /\\*Tprod[]=                4               7            *\\/ *\/ */
   /* /\* /\\*Tvard[][1]             4               1             *\\/ *\/ */
   /* /\* /\\*Tvard[][2]               3               2           *\\/ *\/ */
             
   /*        /\* printf("hPxij Prod ij=%d k=%d  Tprod[%d]=%d Tvard[%d][1]=V%d, Tvard[%d][2]=V%d nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])]=%d nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][1])]=%d\n",ij,k,k,Tprod[k], k,Tvard[k][1], k,Tvard[k][2],nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])],nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][2])]); *\/ */
   /*        /\* cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,k)] * nbcode[Tvard[k][2]][codtabm(ij,k)]; *\/ */
   /*        cov[2+nagesqr+k1]=TinvDoQresult[nres][Tvardk[k1][1]] * TinvDoQresult[nres][Tvardk[k1][2]];     */
   /*        printf("hPxij Prod ij=%d k1=%d  cov[2+%d+%d]=%.5f Tvard[%d][1]=V%d * Tvard[%d][2]=V%d ; TinvDoQresult[nres][Tvardk[k1][1]]=%.4f * TinvDoQresult[nres][Tvardk[k1][1]]=%.4f\n",ij,k1,nagesqr,k1,cov[2+nagesqr+k1],k1,Tvardk[k1][1], k1,Tvardk[k1][2], TinvDoQresult[nres][Tvardk[k1][1]], TinvDoQresult[nres][Tvardk[k1][2]]); */
   /*        printf("hpxij new Product no age product precov[nres=%d][k1=%d]=%.4f\n", nres, k1, precov[nres][k1]); */
   
   /*        /\* if(Dummy[Tvardk[k1][1]]==0){ *\/ */
   /*        /\*   if(Dummy[Tvardk[k1][2]]==0){ /\\* Product of dummies *\\/ *\/ */
   /*            /\* cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,k)] * nbcode[Tvard[k][2]][codtabm(ij,k)]; *\/ */
   /*            /\* cov[2+nagesqr+k1]=Tinvresult[nres][Tvardk[k1][1]] * Tinvresult[nres][Tvardk[k1][2]];   *\/ */
   /*            /\* cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,TnsdVar[Tvard[k][1]])] * nbcode[Tvard[k][2]][codtabm(ij,TnsdVar[Tvard[k][2]])]; *\/ */
   /*          /\* }else{ /\\* Product of dummy by quantitative *\\/ *\/ */
   /*            /\* cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,TnsdVar[Tvard[k][1]])] * Tqresult[nres][k]; *\/ */
   /*            /\* cov[2+nagesqr+k1]=Tresult[nres][Tinvresult[nres][Tvardk[k1][1]]] * Tqresult[nres][Tinvresult[nres][Tvardk[k1][2]]]; *\/ */
   /*        /\*   } *\/ */
   /*        /\* }else{ /\\* Product of quantitative by...*\\/ *\/ */
   /*        /\*   if(Dummy[Tvard[k][2]]==0){  /\\* quant by dummy *\\/ *\/ */
   /*        /\*     /\\* cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][2]][codtabm(ij,TnsdVar[Tvard[k][2]])] * Tqinvresult[nres][Tvard[k][1]]; *\\/ *\/ */
   /*        /\*     cov[2+nagesqr+k1]=Tqresult[nres][Tinvresult[nres][Tvardk[k1][1]]] * Tresult[nres][Tinvresult[nres][Tvardk[k1][2]]]  ; *\/ */
   /*        /\*   }else{ /\\* Product of two quant *\\/ *\/ */
   /*        /\*     /\\* cov[2+nagesqr+Tprod[k]]=Tqinvresult[nres][Tvard[k][1]]*  Tqinvresult[nres][Tvard[k][2]]; *\\/ *\/ */
   /*        /\*     cov[2+nagesqr+k1]=Tqresult[nres][Tinvresult[nres][Tvardk[k1][1]]] * Tqresult[nres][Tinvresult[nres][Tvardk[k1][2]]]  ; *\/ */
   /*        /\*   } *\/ */
   /*        /\* }/\\*end of products quantitative *\\/ *\/ */
   /*      }/\*end of products *\/ */
         /* } /\* End of loop on model equation *\/ */
         /* for (k=1; k<=cptcovn;k++)  */
         /*        cov[2+nagesqr+k]=nbcode[Tvar[k]][codtabm(ij,k)]; */
         /* for (k=1; k<=cptcovage;k++) /\* Should start at cptcovn+1 *\/ */
         /*        cov[2+nagesqr+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(ij,k)]*cov[2]; */
         /* for (k=1; k<=cptcovprod;k++) /\* Useless because included in cptcovn *\/ */
         /*        cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,k)]*nbcode[Tvard[k][2]][codtabm(ij,k)]; */
         
         
         /*printf("hxi cptcov=%d cptcode=%d\n",cptcov,cptcode);*/
         /*printf("h=%d d=%d age=%f cov=%f\n",h,d,age,cov[2]);*/
         /* right multiplication of oldm by the current matrix */
         out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, 
                      pmij(pmmij,cov,ncovmodel,x,nlstate));
         /* if((int)age == 70){ */
         /*        printf(" Forward hpxij age=%d agexact=%f d=%d nhstepm=%d hstepm=%d\n", (int) age, agexact, d, nhstepm, hstepm); */
         /*        for(i=1; i<=nlstate+ndeath; i++) { */
         /*          printf("%d pmmij ",i); */
         /*          for(j=1;j<=nlstate+ndeath;j++) { */
         /*            printf("%f ",pmmij[i][j]); */
         /*          } */
         /*          printf(" oldm "); */
         /*          for(j=1;j<=nlstate+ndeath;j++) { */
         /*            printf("%f ",oldm[i][j]); */
         /*          } */
         /*          printf("\n"); */
         /*        } */
         /* } */
         savm=oldm;
         oldm=newm;
       }
       for(i=1; i<=nlstate+ndeath; i++)
         for(j=1;j<=nlstate+ndeath;j++) {
           po[i][j][h]=newm[i][j];
           /*if(h==nhstepm) printf("po[%d][%d][%d]=%f ",i,j,h,po[i][j][h]);*/
         }
       /*printf("h=%d ",h);*/
     } /* end h */
     /*     printf("\n H=%d \n",h); */
     return po;
   }
   
   /************* Higher Back Matrix Product ***************/
   /* double ***hbxij(double ***po, int nhstepm, double age, int hstepm, double *x, double ***prevacurrent, int nlstate, int stepm, double **oldm, double **savm, double **dnewm, double **doldm, double **dsavm, int ij ) */
   double ***hbxij(double ***po, int nhstepm, double age, int hstepm, double *x, double ***prevacurrent, int nlstate, int stepm, int ij, int nres )
   {
     /* For dummy covariates given in each resultline (for historical, computes the corresponding combination ij),
        computes the transition matrix starting at age 'age' over
        'nhstepm*hstepm*stepm' months (i.e. until
        age (in years)  age+nhstepm*hstepm*stepm/12) by multiplying
        nhstepm*hstepm matrices.
        Output is stored in matrix po[i][j][h] for h every 'hstepm' step
        (typically every 2 years instead of every month which is too big
        for the memory).
        Model is determined by parameters x and covariates have to be
        included manually here. Then we use a call to bmij(x and cov)
        The addresss of po (p3mat allocated to the dimension of nhstepm) should be stored for output
     */
   
     int i, j, d, h, k, k1;
     double **out, cov[NCOVMAX+1], **bmij();
     double **newm, ***newmm;
     double agexact;
     double agebegin, ageend;
     double **oldm, **savm;
   
     newmm=po; /* To be saved */
     oldm=oldms;savm=savms; /* Global pointers */
     /* Hstepm could be zero and should return the unit matrix */
     for (i=1;i<=nlstate+ndeath;i++)
       for (j=1;j<=nlstate+ndeath;j++){
         oldm[i][j]=(i==j ? 1.0 : 0.0);
         po[i][j][0]=(i==j ? 1.0 : 0.0);
       }
     /* Even if hstepm = 1, at least one multiplication by the unit matrix */
     for(h=1; h <=nhstepm; h++){
       for(d=1; d <=hstepm; d++){
         newm=savm;
         /* Covariates have to be included here again */
         cov[1]=1.;
         agexact=age-( (h-1)*hstepm + (d)  )*stepm/YEARM; /* age just before transition, d or d-1? */
         /* agexact=age+((h-1)*hstepm + (d-1))*stepm/YEARM; /\* age just before transition *\/ */
           /* Debug */
         /* printf("hBxij age=%lf, agexact=%lf\n", age, agexact); */
         cov[2]=agexact;
         if(nagesqr==1){
           cov[3]= agexact*agexact;
         }
         /** New code */
         for(k1=1;k1<=cptcovt;k1++){ /* loop on model equation (including products) */ 
           if(Typevar[k1]==1){ /* A product with age */
             cov[2+nagesqr+k1]=precov[nres][k1]*cov[2];
           }else{
             cov[2+nagesqr+k1]=precov[nres][k1];
           }
         }/* End of loop on model equation */
         /** End of new code */
     /** This was old code */
         /* for (k=1; k<=nsd;k++){ /\* For single dummy covariates only *\//\* cptcovn error *\/ */
         /* /\*    cov[2+nagesqr+k]=nbcode[Tvar[k]][codtabm(ij,k)]; *\/ */
         /* /\* /\\* cov[2+nagesqr+k]=nbcode[Tvar[k]][codtabm(ij,Tvar[k])]; *\\/ *\/ */
         /*        cov[2+nagesqr+TvarsDind[k]]=nbcode[TvarsD[k]][codtabm(ij,TvarsD[k])];/\* Bug valgrind *\/ */
         /*   /\* printf("hbxij Dummy agexact=%.0f combi=%d k=%d TvarsD[%d]=V%d TvarsDind[%d]=%d nbcode=%d cov[%d]=%lf codtabm(%d,Tvar[%d])=%d \n",agexact,ij,k, k, TvarsD[k],k,TvarsDind[k],nbcode[TvarsD[k]][codtabm(ij,k)],2+nagesqr+TvarsDind[k],cov[2+nagesqr+TvarsDind[k]], ij, k, codtabm(ij,k)); *\/ */
         /* } */
         /* for (k=1; k<=nsq;k++) { /\* For single varying covariates only *\/ */
         /*        /\* Here comes the value of quantitative after renumbering k with single quantitative covariates *\/ */
         /*        cov[2+nagesqr+TvarsQind[k]]=Tqresult[nres][k];  */
         /*        /\* printf("hPxij Quantitative k=%d  TvarsQind[%d]=%d, TvarsQ[%d]=V%d,Tqresult[%d][%d]=%f\n",k,k,TvarsQind[k],k,TvarsQ[k],nres,k,Tqresult[nres][k]); *\/ */
         /* } */
         /* for (k=1; k<=cptcovage;k++){ /\* Should start at cptcovn+1 *\//\* For product with age *\/ */
         /*        /\* if(Dummy[Tvar[Tage[k]]]== 2){ /\\* dummy with age error!!!*\\/ *\/ */
         /*        if(Dummy[Tage[k]]== 2){ /\* dummy with age *\/ */
         /*          cov[2+nagesqr+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(ij,Tvar[Tage[k]])]*cov[2]; */
         /*        } else if(Dummy[Tage[k]]== 3){ /\* quantitative with age *\/ */
         /*          cov[2+nagesqr+Tage[k]]=Tqresult[nres][k];  */
         /*        } */
         /*        /\* printf("hBxij Age combi=%d k=%d  Tage[%d]=V%d Tqresult[%d][%d]=%f\n",ij,k,k,Tage[k],nres,k,Tqresult[nres][k]); *\/ */
         /* } */
         /* for (k=1; k<=cptcovprod;k++){ /\* Useless because included in cptcovn *\/ */
         /*        cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])]*nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][2])]; */
         /*        if(Dummy[Tvard[k][1]]==0){ */
         /*          if(Dummy[Tvard[k][2]]==0){ */
         /*            cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])] * nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][1])]; */
         /*          }else{ */
         /*            cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,Tvard[k][1])] * Tqresult[nres][k]; */
         /*          } */
         /*        }else{ */
         /*          if(Dummy[Tvard[k][2]]==0){ */
         /*            cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][2]][codtabm(ij,Tvard[k][2])] * Tqinvresult[nres][Tvard[k][1]]; */
         /*          }else{ */
         /*            cov[2+nagesqr+Tprod[k]]=Tqinvresult[nres][Tvard[k][1]]*  Tqinvresult[nres][Tvard[k][2]]; */
         /*          } */
         /*        } */
         /* }                       */
         /* /\*printf("hxi cptcov=%d cptcode=%d\n",cptcov,cptcode);*\/ */
         /* /\*printf("h=%d d=%d age=%f cov=%f\n",h,d,age,cov[2]);*\/ */
   /** End of old code */
         
         /* Careful transposed matrix */
         /* age is in cov[2], prevacurrent at beginning of transition. */
         /* out=matprod2(newm, bmij(pmmij,cov,ncovmodel,x,nlstate,prevacurrent, dnewm, doldm, dsavm,ij),\ */
         /*                                                 1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, oldm); */
         out=matprod2(newm, bmij(pmmij,cov,ncovmodel,x,nlstate,prevacurrent,ij),\
                      1,nlstate+ndeath,1,nlstate+ndeath,1,nlstate+ndeath, oldm);/* Bug valgrind */
         /* if((int)age == 70){ */
         /*        printf(" Backward hbxij age=%d agexact=%f d=%d nhstepm=%d hstepm=%d\n", (int) age, agexact, d, nhstepm, hstepm); */
         /*        for(i=1; i<=nlstate+ndeath; i++) { */
         /*          printf("%d pmmij ",i); */
         /*          for(j=1;j<=nlstate+ndeath;j++) { */
         /*            printf("%f ",pmmij[i][j]); */
         /*          } */
         /*          printf(" oldm "); */
         /*          for(j=1;j<=nlstate+ndeath;j++) { */
         /*            printf("%f ",oldm[i][j]); */
         /*          } */
         /*          printf("\n"); */
         /*        } */
         /* } */
         savm=oldm;
         oldm=newm;
       }
       for(i=1; i<=nlstate+ndeath; i++)
         for(j=1;j<=nlstate+ndeath;j++) {
           po[i][j][h]=newm[i][j];
           /* if(h==nhstepm) */
           /*   printf("po[%d][%d][%d]=%f ",i,j,h,po[i][j][h]); */
         }
       /* printf("h=%d %.1f ",h, agexact); */
     } /* end h */
     /* printf("\n H=%d nhs=%d \n",h, nhstepm); */
     return po;
   }
   
   
   #ifdef NLOPT
     double  myfunc(unsigned n, const double *p1, double *grad, void *pd){
     double fret;
     double *xt;
     int j;
     myfunc_data *d2 = (myfunc_data *) pd;
   /* xt = (p1-1); */
     xt=vector(1,n); 
     for (j=1;j<=n;j++)   xt[j]=p1[j-1]; /* xt[1]=p1[0] */
   
     fret=(d2->function)(xt); /*  p xt[1]@8 is fine */
     /* fret=(*func)(xt); /\*  p xt[1]@8 is fine *\/ */
     printf("Function = %.12lf ",fret);
     for (j=1;j<=n;j++) printf(" %d %.8lf", j, xt[j]); 
     printf("\n");
    free_vector(xt,1,n);
     return fret;
   }
   #endif
   
   /*************** log-likelihood *************/
   double func( double *x)
   {
     int i, ii, j, k, mi, d, kk;
     int ioffset=0;
     double l, ll[NLSTATEMAX+1], cov[NCOVMAX+1];
     double **out;
     double lli; /* Individual log likelihood */
     int s1, s2;
     int iv=0, iqv=0, itv=0, iqtv=0 ; /* Index of varying covariate, fixed quantitative cov, time varying covariate, quantitative time varying covariate */
     double bbh, survp;
     long ipmx;
     double agexact;
     /*extern weight */
     /* We are differentiating ll according to initial status */
     /*  for (i=1;i<=npar;i++) printf("%f ", x[i]);*/
     /*for(i=1;i<imx;i++) 
       printf(" %d\n",s[4][i]);
     */
   
     ++countcallfunc;
   
     cov[1]=1.;
   
     for(k=1; k<=nlstate; k++) ll[k]=0.;
     ioffset=0;
     if(mle==1){
       for (i=1,ipmx=0, sw=0.; i<=imx; i++){
         /* Computes the values of the ncovmodel covariates of the model
            depending if the covariates are fixed or varying (age dependent) and stores them in cov[]
            Then computes with function pmij which return a matrix p[i][j] giving the elementary probability
            to be observed in j being in i according to the model.
         */
         ioffset=2+nagesqr ;
      /* Fixed */
         for (k=1; k<=ncovf;k++){ /* For each fixed covariate dummu or quant or prod */
           /* # V1=sex, V2=raedyrs Quant Fixed, State=livarnb4..livarnb11, V3=iadl4..iald11, V4=adlw4..adlw11, V5=r4bmi..r11bmi */
           /*             V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
           /*  TvarF[1]=Tvar[6]=2,  TvarF[2]=Tvar[7]=7, TvarF[3]=Tvar[9]=1  ID of fixed covariates or product V2, V1*V2, V1 */
           /* TvarFind;  TvarFind[1]=6,  TvarFind[2]=7, TvarFind[3]=9 *//* Inverse V2(6) is first fixed (single or prod)  */
           cov[ioffset+TvarFind[k]]=covar[Tvar[TvarFind[k]]][i];/* V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1, only V1 is fixed (TvarFind[1]=6)*/
           /* V1*V2 (7)  TvarFind[2]=7, TvarFind[3]=9 */
         }
         /* In model V2+V1*V4+age*V3+V3*V2 Tvar[1] is V2, Tvar[2=V1*V4] 
            is 5, Tvar[3=age*V3] should not be computed because of age Tvar[4=V3*V2]=6 
            has been calculated etc */
         /* For an individual i, wav[i] gives the number of effective waves */
         /* We compute the contribution to Likelihood of each effective transition
            mw[mi][i] is real wave of the mi th effectve wave */
         /* Then statuses are computed at each begin and end of an effective wave s1=s[ mw[mi][i] ][i];
            s2=s[mw[mi+1][i]][i];
            And the iv th varying covariate is the cotvar[mw[mi+1][i]][iv][i]
            But if the variable is not in the model TTvar[iv] is the real variable effective in the model:
            meaning that decodemodel should be used cotvar[mw[mi+1][i]][TTvar[iv]][i]
         */
         for(mi=1; mi<= wav[i]-1; mi++){
           for(k=1; k <= ncovv ; k++){ /* Varying  covariates in the model (single and product but no age )"V5+V4+V3+V4*V3+V5*age+V1*age+V1" +TvarVind 1,2,3,4(V4*V3)  Tvar[1]@7{5, 4, 3, 6, 5, 1, 1 ; 6 because the created covar is after V5 and is 6, minus 1+1, 3,2,1,4 positions in cotvar*/
             /* cov[ioffset+TvarVind[k]]=cotvar[mw[mi][i]][Tvar[TvarVind[k]]][i]; but where is the crossproduct? */
             cov[ioffset+TvarVind[k]]=cotvar[mw[mi][i]][Tvar[TvarVind[k]]-ncovcol-nqv][i];
           }
           for (ii=1;ii<=nlstate+ndeath;ii++)
             for (j=1;j<=nlstate+ndeath;j++){
               oldm[ii][j]=(ii==j ? 1.0 : 0.0);
               savm[ii][j]=(ii==j ? 1.0 : 0.0);
             }
           for(d=0; d<dh[mi][i]; d++){
             newm=savm;
             agexact=agev[mw[mi][i]][i]+d*stepm/YEARM;
             cov[2]=agexact;
             if(nagesqr==1)
               cov[3]= agexact*agexact;  /* Should be changed here */
             for (kk=1; kk<=cptcovage;kk++) {
               if(!FixedV[Tvar[Tage[kk]]])
                 cov[Tage[kk]+2+nagesqr]=covar[Tvar[Tage[kk]]][i]*agexact; /* Tage[kk] gives the data-covariate associated with age */
               else
                 cov[Tage[kk]+2+nagesqr]=cotvar[mw[mi][i]][Tvar[Tage[kk]]-ncovcol-nqv][i]*agexact;
             }
             out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,
                          1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));
             savm=oldm;
             oldm=newm;
           } /* end mult */
           
           /*lli=log(out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]]);*/ /* Original formula */
           /* But now since version 0.9 we anticipate for bias at large stepm.
            * If stepm is larger than one month (smallest stepm) and if the exact delay 
            * (in months) between two waves is not a multiple of stepm, we rounded to 
            * the nearest (and in case of equal distance, to the lowest) interval but now
            * we keep into memory the bias bh[mi][i] and also the previous matrix product
            * (i.e to dh[mi][i]-1) saved in 'savm'. Then we inter(extra)polate the
            * probability in order to take into account the bias as a fraction of the way
                                    * from savm to out if bh is negative or even beyond if bh is positive. bh varies
                                    * -stepm/2 to stepm/2 .
                                    * For stepm=1 the results are the same as for previous versions of Imach.
                                    * For stepm > 1 the results are less biased than in previous versions. 
                                    */
           s1=s[mw[mi][i]][i];
           s2=s[mw[mi+1][i]][i];
           bbh=(double)bh[mi][i]/(double)stepm; 
           /* bias bh is positive if real duration
            * is higher than the multiple of stepm and negative otherwise.
            */
           /* lli= (savm[s1][s2]>1.e-8 ?(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]):log((1.+bbh)*out[s1][s2]));*/
           if( s2 > nlstate){ 
             /* i.e. if s2 is a death state and if the date of death is known 
                then the contribution to the likelihood is the probability to 
                die between last step unit time and current  step unit time, 
                which is also equal to probability to die before dh 
                minus probability to die before dh-stepm . 
                In version up to 0.92 likelihood was computed
                as if date of death was unknown. Death was treated as any other
                health state: the date of the interview describes the actual state
                and not the date of a change in health state. The former idea was
                to consider that at each interview the state was recorded
                (healthy, disable or death) and IMaCh was corrected; but when we
                introduced the exact date of death then we should have modified
                the contribution of an exact death to the likelihood. This new
                contribution is smaller and very dependent of the step unit
                stepm. It is no more the probability to die between last interview
                and month of death but the probability to survive from last
                interview up to one month before death multiplied by the
                probability to die within a month. Thanks to Chris
                Jackson for correcting this bug.  Former versions increased
                mortality artificially. The bad side is that we add another loop
                which slows down the processing. The difference can be up to 10%
                lower mortality.
             */
             /* If, at the beginning of the maximization mostly, the
                cumulative probability or probability to be dead is
                constant (ie = 1) over time d, the difference is equal to
                0.  out[s1][3] = savm[s1][3]: probability, being at state
                s1 at precedent wave, to be dead a month before current
                wave is equal to probability, being at state s1 at
                precedent wave, to be dead at mont of the current
                wave. Then the observed probability (that this person died)
                is null according to current estimated parameter. In fact,
                it should be very low but not zero otherwise the log go to
                infinity.
             */
   /* #ifdef INFINITYORIGINAL */
   /*          lli=log(out[s1][s2] - savm[s1][s2]); */
   /* #else */
   /*        if ((out[s1][s2] - savm[s1][s2]) < mytinydouble)  */
   /*          lli=log(mytinydouble); */
   /*        else */
   /*          lli=log(out[s1][s2] - savm[s1][s2]); */
   /* #endif */
             lli=log(out[s1][s2] - savm[s1][s2]);
             
           } else if  ( s2==-1 ) { /* alive */
             for (j=1,survp=0. ; j<=nlstate; j++) 
               survp += (1.+bbh)*out[s1][j]- bbh*savm[s1][j];
             /*survp += out[s1][j]; */
             lli= log(survp);
           }
           else if  (s2==-4) { 
             for (j=3,survp=0. ; j<=nlstate; j++)  
               survp += (1.+bbh)*out[s1][j]- bbh*savm[s1][j];
             lli= log(survp); 
           } 
           else if  (s2==-5) { 
             for (j=1,survp=0. ; j<=2; j++)  
               survp += (1.+bbh)*out[s1][j]- bbh*savm[s1][j];
             lli= log(survp); 
           } 
           else{
             lli= log((1.+bbh)*out[s1][s2]- bbh*savm[s1][s2]); /* linear interpolation */
             /*  lli= (savm[s1][s2]>(double)1.e-8 ?log((1.+bbh)*out[s1][s2]- bbh*(savm[s1][s2])):log((1.+bbh)*out[s1][s2]));*/ /* linear interpolation */
           } 
           /*lli=(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]);*/
           /*if(lli ==000.0)*/
           /*printf("bbh= %f lli=%f savm=%f out=%f %d\n",bbh,lli,savm[s1][s2], out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]],i); */
           ipmx +=1;
           sw += weight[i];
           ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;
           /* if (lli < log(mytinydouble)){ */
           /*   printf("Close to inf lli = %.10lf <  %.10lf i= %d mi= %d, s[%d][i]=%d s1=%d s2=%d\n", lli,log(mytinydouble), i, mi,mw[mi][i], s[mw[mi][i]][i], s1,s2); */
           /*   fprintf(ficlog,"Close to inf lli = %.10lf i= %d mi= %d, s[mw[mi][i]][i]=%d\n", lli, i, mi,s[mw[mi][i]][i]); */
           /* } */
         } /* end of wave */
       } /* end of individual */
     }  else if(mle==2){
       for (i=1,ipmx=0, sw=0.; i<=imx; i++){
         ioffset=2+nagesqr ;
         for (k=1; k<=ncovf;k++)
           cov[ioffset+TvarFind[k]]=covar[Tvar[TvarFind[k]]][i];
         for(mi=1; mi<= wav[i]-1; mi++){
           for(k=1; k <= ncovv ; k++){
             cov[ioffset+TvarVind[k]]=cotvar[mw[mi][i]][Tvar[TvarVind[k]]-ncovcol-nqv][i];
           }
           for (ii=1;ii<=nlstate+ndeath;ii++)
             for (j=1;j<=nlstate+ndeath;j++){
               oldm[ii][j]=(ii==j ? 1.0 : 0.0);
               savm[ii][j]=(ii==j ? 1.0 : 0.0);
             }
           for(d=0; d<=dh[mi][i]; d++){
             newm=savm;
             agexact=agev[mw[mi][i]][i]+d*stepm/YEARM;
             cov[2]=agexact;
             if(nagesqr==1)
               cov[3]= agexact*agexact;
             for (kk=1; kk<=cptcovage;kk++) {
               cov[Tage[kk]+2+nagesqr]=covar[Tvar[Tage[kk]]][i]*agexact;
             }
             out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,
                          1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));
             savm=oldm;
             oldm=newm;
           } /* end mult */
         
           s1=s[mw[mi][i]][i];
           s2=s[mw[mi+1][i]][i];
           bbh=(double)bh[mi][i]/(double)stepm; 
           lli= (savm[s1][s2]>(double)1.e-8 ?log((1.+bbh)*out[s1][s2]- bbh*(savm[s1][s2])):log((1.+bbh)*out[s1][s2])); /* linear interpolation */
           ipmx +=1;
           sw += weight[i];
           ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;
         } /* end of wave */
       } /* end of individual */
     }  else if(mle==3){  /* exponential inter-extrapolation */
       for (i=1,ipmx=0, sw=0.; i<=imx; i++){
         for (k=1; k<=cptcovn;k++) cov[2+nagesqr+k]=covar[Tvar[k]][i];
         for(mi=1; mi<= wav[i]-1; mi++){
           for (ii=1;ii<=nlstate+ndeath;ii++)
             for (j=1;j<=nlstate+ndeath;j++){
               oldm[ii][j]=(ii==j ? 1.0 : 0.0);
               savm[ii][j]=(ii==j ? 1.0 : 0.0);
             }
           for(d=0; d<dh[mi][i]; d++){
             newm=savm;
             agexact=agev[mw[mi][i]][i]+d*stepm/YEARM;
             cov[2]=agexact;
             if(nagesqr==1)
               cov[3]= agexact*agexact;
             for (kk=1; kk<=cptcovage;kk++) {
               cov[Tage[kk]+2+nagesqr]=covar[Tvar[Tage[kk]]][i]*agexact;
             }
             out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,
                          1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));
             savm=oldm;
             oldm=newm;
           } /* end mult */
         
           s1=s[mw[mi][i]][i];
           s2=s[mw[mi+1][i]][i];
           bbh=(double)bh[mi][i]/(double)stepm; 
           lli= (savm[s1][s2]>1.e-8 ?(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]):log((1.+bbh)*out[s1][s2])); /* exponential inter-extrapolation */
           ipmx +=1;
           sw += weight[i];
           ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;
         } /* end of wave */
       } /* end of individual */
     }else if (mle==4){  /* ml=4 no inter-extrapolation */
       for (i=1,ipmx=0, sw=0.; i<=imx; i++){
         for (k=1; k<=cptcovn;k++) cov[2+nagesqr+k]=covar[Tvar[k]][i];
         for(mi=1; mi<= wav[i]-1; mi++){
           for (ii=1;ii<=nlstate+ndeath;ii++)
             for (j=1;j<=nlstate+ndeath;j++){
               oldm[ii][j]=(ii==j ? 1.0 : 0.0);
               savm[ii][j]=(ii==j ? 1.0 : 0.0);
             }
           for(d=0; d<dh[mi][i]; d++){
             newm=savm;
             agexact=agev[mw[mi][i]][i]+d*stepm/YEARM;
             cov[2]=agexact;
             if(nagesqr==1)
               cov[3]= agexact*agexact;
             for (kk=1; kk<=cptcovage;kk++) {
               cov[Tage[kk]+2+nagesqr]=covar[Tvar[Tage[kk]]][i]*agexact;
             }
           
             out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,
                          1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));
             savm=oldm;
             oldm=newm;
           } /* end mult */
         
           s1=s[mw[mi][i]][i];
           s2=s[mw[mi+1][i]][i];
           if( s2 > nlstate){ 
             lli=log(out[s1][s2] - savm[s1][s2]);
           } else if  ( s2==-1 ) { /* alive */
             for (j=1,survp=0. ; j<=nlstate; j++) 
               survp += out[s1][j];
             lli= log(survp);
           }else{
             lli=log(out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]]); /* Original formula */
           }
           ipmx +=1;
           sw += weight[i];
           ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;
   /*      printf("i=%6d s1=%1d s2=%1d mi=%1d mw=%1d dh=%3d prob=%10.6f w=%6.4f out=%10.6f sav=%10.6f\n",i,s1,s2,mi,mw[mi][i],dh[mi][i],exp(lli),weight[i],out[s1][s2],savm[s1][s2]); */
         } /* end of wave */
       } /* end of individual */
     }else{  /* ml=5 no inter-extrapolation no jackson =0.8a */
       for (i=1,ipmx=0, sw=0.; i<=imx; i++){
         for (k=1; k<=cptcovn;k++) cov[2+nagesqr+k]=covar[Tvar[k]][i];
         for(mi=1; mi<= wav[i]-1; mi++){
           for (ii=1;ii<=nlstate+ndeath;ii++)
             for (j=1;j<=nlstate+ndeath;j++){
               oldm[ii][j]=(ii==j ? 1.0 : 0.0);
               savm[ii][j]=(ii==j ? 1.0 : 0.0);
             }
           for(d=0; d<dh[mi][i]; d++){
             newm=savm;
             agexact=agev[mw[mi][i]][i]+d*stepm/YEARM;
             cov[2]=agexact;
             if(nagesqr==1)
               cov[3]= agexact*agexact;
             for (kk=1; kk<=cptcovage;kk++) {
               cov[Tage[kk]+2+nagesqr]=covar[Tvar[Tage[kk]]][i]*agexact;
             }
           
             out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,
                          1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));
             savm=oldm;
             oldm=newm;
           } /* end mult */
         
           s1=s[mw[mi][i]][i];
           s2=s[mw[mi+1][i]][i];
           lli=log(out[s[mw[mi][i]][i]][s[mw[mi+1][i]][i]]); /* Original formula */
           ipmx +=1;
           sw += weight[i];
           ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;
           /*printf("i=%6d s1=%1d s2=%1d mi=%1d mw=%1d dh=%3d prob=%10.6f w=%6.4f out=%10.6f sav=%10.6f\n",i,s1,s2,mi,mw[mi][i],dh[mi][i],exp(lli),weight[i],out[s1][s2],savm[s1][s2]);*/
         } /* end of wave */
       } /* end of individual */
     } /* End of if */
     for(k=1,l=0.; k<=nlstate; k++) l += ll[k];
     /* printf("l1=%f l2=%f ",ll[1],ll[2]); */
     l= l*ipmx/sw; /* To get the same order of magnitude as if weight=1 for every body */
     return -l;
   }
   
   /*************** log-likelihood *************/
   double funcone( double *x)
   {
     /* Same as func but slower because of a lot of printf and if */
     int i, ii, j, k, mi, d, kk, kf=0;
     int ioffset=0;
     double l, ll[NLSTATEMAX+1], cov[NCOVMAX+1];
     double **out;
     double lli; /* Individual log likelihood */
     double llt;
     int s1, s2;
     int iv=0, iqv=0, itv=0, iqtv=0 ; /* Index of varying covariate, fixed quantitative cov, time varying covariate, quantitative time varying covariate */
   
     double bbh, survp;
     double agexact;
     double agebegin, ageend;
     /*extern weight */
     /* We are differentiating ll according to initial status */
     /*  for (i=1;i<=npar;i++) printf("%f ", x[i]);*/
     /*for(i=1;i<imx;i++) 
       printf(" %d\n",s[4][i]);
     */
     cov[1]=1.;
   
     for(k=1; k<=nlstate; k++) ll[k]=0.;
     ioffset=0;
     for (i=1,ipmx=0, sw=0.; i<=imx; i++){
       /* ioffset=2+nagesqr+cptcovage; */
       ioffset=2+nagesqr;
       /* Fixed */
       /* for (k=1; k<=cptcovn;k++) cov[2+nagesqr+k]=covar[Tvar[k]][i]; */
       /* for (k=1; k<=ncoveff;k++){ /\* Simple and product fixed Dummy covariates without age* products *\/ */
       for (kf=1; kf<=ncovf;kf++){ /* Simple and product fixed covariates without age* products *//* Missing values are set to -1 but should be dropped */
         cov[ioffset+TvarFind[kf]]=covar[Tvar[TvarFind[kf]]][i];/* V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1, only V1 is fixed (k=6)*/
   /*    cov[ioffset+TvarFind[1]]=covar[Tvar[TvarFind[1]]][i];  */
   /*    cov[2+6]=covar[Tvar[6]][i];  */
   /*    cov[2+6]=covar[2][i]; V2  */
   /*    cov[TvarFind[2]]=covar[Tvar[TvarFind[2]]][i];  */
   /*    cov[2+7]=covar[Tvar[7]][i];  */
   /*    cov[2+7]=covar[7][i]; V7=V1*V2  */
   /*    cov[TvarFind[3]]=covar[Tvar[TvarFind[3]]][i];  */
   /*    cov[2+9]=covar[Tvar[9]][i];  */
   /*    cov[2+9]=covar[1][i]; V1  */
       }
       /* for (k=1; k<=nqfveff;k++){ /\* Simple and product fixed Quantitative covariates without age* products *\/ */
       /*   cov[++ioffset]=coqvar[TvarFQ[k]][i];/\* V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1, only V2 and V1*V2 is fixed (k=6 and 7?)*\/ */
       /* } */
       /* for(iqv=1; iqv <= nqfveff; iqv++){ /\* Quantitative fixed covariates *\/ */
       /*   cov[++ioffset]=coqvar[Tvar[iqv]][i]; /\* Only V2 k=6 and V1*V2 7 *\/ */
       /* } */
       
   
       for(mi=1; mi<= wav[i]-1; mi++){  /* Varying with waves */
       /* Wave varying (but not age varying) */
         for(k=1; k <= ncovv ; k++){ /* Varying  covariates (single and product but no age )*/
           /* cov[ioffset+TvarVind[k]]=cotvar[mw[mi][i]][Tvar[TvarVind[k]]][i]; */
           cov[ioffset+TvarVind[k]]=cotvar[mw[mi][i]][Tvar[TvarVind[k]]-ncovcol-nqv][i];
         }
         /* for(itv=1; itv <= ntveff; itv++){ /\* Varying dummy covariates (single??)*\/ */
         /* iv= Tvar[Tmodelind[ioffset-2-nagesqr-cptcovage+itv]]-ncovcol-nqv; /\* Counting the # varying covariate from 1 to ntveff *\/ */
         /* cov[ioffset+iv]=cotvar[mw[mi][i]][iv][i]; */
         /* k=ioffset-2-nagesqr-cptcovage+itv; /\* position in simple model *\/ */
         /* cov[ioffset+itv]=cotvar[mw[mi][i]][TmodelInvind[itv]][i]; */
         /* printf(" i=%d,mi=%d,itv=%d,TmodelInvind[itv]=%d,cotvar[mw[mi][i]][TmodelInvind[itv]][i]=%f\n", i, mi, itv, TmodelInvind[itv],cotvar[mw[mi][i]][TmodelInvind[itv]][i]); */
         /* for(iqtv=1; iqtv <= nqtveff; iqtv++){ /\* Varying quantitatives covariates *\/ */
         /*        iv=TmodelInvQind[iqtv]; /\* Counting the # varying covariate from 1 to ntveff *\/ */
         /*        /\* printf(" i=%d,mi=%d,iqtv=%d,TmodelInvQind[iqtv]=%d,cotqvar[mw[mi][i]][TmodelInvQind[iqtv]][i]=%f\n", i, mi, iqtv, TmodelInvQind[iqtv],cotqvar[mw[mi][i]][TmodelInvQind[iqtv]][i]); *\/ */
         /*        cov[ioffset+ntveff+iqtv]=cotqvar[mw[mi][i]][TmodelInvQind[iqtv]][i]; */
         /* } */
         for (ii=1;ii<=nlstate+ndeath;ii++)
           for (j=1;j<=nlstate+ndeath;j++){
             oldm[ii][j]=(ii==j ? 1.0 : 0.0);
             savm[ii][j]=(ii==j ? 1.0 : 0.0);
           }
         
         agebegin=agev[mw[mi][i]][i]; /* Age at beginning of effective wave */
         ageend=agev[mw[mi][i]][i] + (dh[mi][i])*stepm/YEARM; /* Age at end of effective wave and at the end of transition */
         for(d=0; d<dh[mi][i]; d++){  /* Delay between two effective waves */
         /* for(d=0; d<=0; d++){  /\* Delay between two effective waves Only one matrix to speed up*\/ */
           /*dh[m][i] or dh[mw[mi][i]][i] is the delay between two effective waves m=mw[mi][i]
             and mw[mi+1][i]. dh depends on stepm.*/
           newm=savm;
           agexact=agev[mw[mi][i]][i]+d*stepm/YEARM;  /* Here d is needed */
           cov[2]=agexact;
           if(nagesqr==1)
             cov[3]= agexact*agexact;
           for (kk=1; kk<=cptcovage;kk++) {
             if(!FixedV[Tvar[Tage[kk]]])
               cov[Tage[kk]+2+nagesqr]=covar[Tvar[Tage[kk]]][i]*agexact;
             else
               cov[Tage[kk]+2+nagesqr]=cotvar[mw[mi][i]][Tvar[Tage[kk]]-ncovcol-nqv][i]*agexact;
           }
           /* printf("i=%d,mi=%d,d=%d,mw[mi][i]=%d\n",i, mi,d,mw[mi][i]); */
           /* savm=pmij(pmmij,cov,ncovmodel,x,nlstate); */
           out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath,
                        1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate));
           /* out=matprod2(newm,oldm,1,nlstate+ndeath,1,nlstate+ndeath, */
           /*           1,nlstate+ndeath,pmij(pmmij,cov,ncovmodel,x,nlstate)); */
           savm=oldm;
           oldm=newm;
         } /* end mult */
         
         s1=s[mw[mi][i]][i];
         s2=s[mw[mi+1][i]][i];
         /* if(s2==-1){ */
         /*        printf(" ERROR s1=%d, s2=%d i=%d \n", s1, s2, i); */
         /*        /\* exit(1); *\/ */
         /* } */
         bbh=(double)bh[mi][i]/(double)stepm; 
         /* bias is positive if real duration
          * is higher than the multiple of stepm and negative otherwise.
          */
         if( s2 > nlstate && (mle <5) ){  /* Jackson */
           lli=log(out[s1][s2] - savm[s1][s2]);
         } else if  ( s2==-1 ) { /* alive */
           for (j=1,survp=0. ; j<=nlstate; j++) 
             survp += (1.+bbh)*out[s1][j]- bbh*savm[s1][j];
           lli= log(survp);
         }else if (mle==1){
           lli= log((1.+bbh)*out[s1][s2]- bbh*savm[s1][s2]); /* linear interpolation */
         } else if(mle==2){
           lli= (savm[s1][s2]>(double)1.e-8 ?log((1.+bbh)*out[s1][s2]- bbh*savm[s1][s2]):log((1.+bbh)*out[s1][s2])); /* linear interpolation */
         } else if(mle==3){  /* exponential inter-extrapolation */
           lli= (savm[s1][s2]>(double)1.e-8 ?(1.+bbh)*log(out[s1][s2])- bbh*log(savm[s1][s2]):log((1.+bbh)*out[s1][s2])); /* exponential inter-extrapolation */
         } else if (mle==4){  /* mle=4 no inter-extrapolation */
           lli=log(out[s1][s2]); /* Original formula */
         } else{  /* mle=0 back to 1 */
           lli= log((1.+bbh)*out[s1][s2]- bbh*savm[s1][s2]); /* linear interpolation */
           /*lli=log(out[s1][s2]); */ /* Original formula */
         } /* End of if */
         ipmx +=1;
         sw += weight[i];
         ll[s[mw[mi][i]][i]] += 2*weight[i]*lli;
         /* printf("Funcone i=%6d s1=%1d s2=%1d mi=%1d mw=%1d dh=%3d prob=%10.6f w=%6.4f out=%10.6f sav=%10.6f\n",i,s1,s2,mi,mw[mi][i],dh[mi][i],exp(lli),weight[i],(s2==-1? -1: out[s1][s2]),(s2==-1? -1: savm[s1][s2])); */
         if(globpr){
           fprintf(ficresilk,"%09ld %6.1f %6.1f %6d %2d %2d %2d %2d %3d %15.6f %8.4f %8.3f\
    %11.6f %11.6f %11.6f ", \
                   num[i], agebegin, ageend, i,s1,s2,mi,mw[mi][i],dh[mi][i],exp(lli),weight[i],weight[i]*gipmx/gsw,
                   2*weight[i]*lli,(s2==-1? -1: out[s1][s2]),(s2==-1? -1: savm[s1][s2]));
    /*     printf("%09ld %6.1f %6.1f %6d %2d %2d %2d %2d %3d %15.6f %8.4f %8.3f\ */
    /* %11.6f %11.6f %11.6f ", \ */
    /*             num[i], agebegin, ageend, i,s1,s2,mi,mw[mi][i],dh[mi][i],exp(lli),weight[i],weight[i]*gipmx/gsw, */
    /*             2*weight[i]*lli,(s2==-1? -1: out[s1][s2]),(s2==-1? -1: savm[s1][s2])); */
           for(k=1,llt=0.,l=0.; k<=nlstate; k++){
             llt +=ll[k]*gipmx/gsw;
             fprintf(ficresilk," %10.6f",-ll[k]*gipmx/gsw);
             /* printf(" %10.6f",-ll[k]*gipmx/gsw); */
           }
           fprintf(ficresilk," %10.6f\n", -llt);
           /* printf(" %10.6f\n", -llt); */
         }
       } /* end of wave */
     } /* end of individual */
     for(k=1,l=0.; k<=nlstate; k++) l += ll[k];
   /* printf("l1=%f l2=%f ",ll[1],ll[2]); */
     l= l*ipmx/sw; /* To get the same order of magnitude as if weight=1 for every body */
     if(globpr==0){ /* First time we count the contributions and weights */
       gipmx=ipmx;
       gsw=sw;
     }
   return -l;
   }
   
   
   /*************** function likelione ***********/
   void likelione(FILE *ficres,double p[], int npar, int nlstate, int *globpri, long *ipmx, double *sw, double *fretone, double (*func)(double []))
   {
     /* This routine should help understanding what is done with 
        the selection of individuals/waves and
        to check the exact contribution to the likelihood.
        Plotting could be done.
      */
     int k;
   
     if(*globpri !=0){ /* Just counts and sums, no printings */
       strcpy(fileresilk,"ILK_"); 
       strcat(fileresilk,fileresu);
       if((ficresilk=fopen(fileresilk,"w"))==NULL) {
         printf("Problem with resultfile: %s\n", fileresilk);
         fprintf(ficlog,"Problem with resultfile: %s\n", fileresilk);
       }
       fprintf(ficresilk, "#individual(line's_record) count ageb ageend s1 s2 wave# effective_wave# number_of_matrices_product pij weight weight/gpw -2ln(pij)*weight 0pij_x 0pij_(x-stepm) cumulating_loglikeli_by_health_state(reweighted=-2ll*weightXnumber_of_contribs/sum_of_weights) and_total\n");
       fprintf(ficresilk, "#num_i ageb agend i s1 s2 mi mw dh likeli weight %%weight 2wlli out sav ");
       /*  i,s1,s2,mi,mw[mi][i],dh[mi][i],exp(lli),weight[i],2*weight[i]*lli,out[s1][s2],savm[s1][s2]); */
       for(k=1; k<=nlstate; k++) 
         fprintf(ficresilk," -2*gipw/gsw*weight*ll[%d]++",k);
       fprintf(ficresilk," -2*gipw/gsw*weight*ll(total)\n");
     }
   
     *fretone=(*func)(p);
     if(*globpri !=0){
       fclose(ficresilk);
       if (mle ==0)
         fprintf(fichtm,"\n<br>File of contributions to the likelihood computed with initial parameters and mle = %d.",mle);
       else if(mle >=1)
         fprintf(fichtm,"\n<br>File of contributions to the likelihood computed with optimized parameters mle = %d.",mle);
       fprintf(fichtm," You should at least run with mle >= 1 to get starting values corresponding to the optimized parameters in order to visualize the real contribution of each individual/wave: <a href=\"%s\">%s</a><br>\n",subdirf(fileresilk),subdirf(fileresilk));
       fprintf(fichtm,"\n<br>Equation of the model: <b>model=1+age+%s</b><br>\n",model); 
         
       for (k=1; k<= nlstate ; k++) {
         fprintf(fichtm,"<br>- Probability p<sub>%dj</sub> by origin %d and destination j. Dot's sizes are related to corresponding weight: <a href=\"%s-p%dj.png\">%s-p%dj.png</a><br> \
   <img src=\"%s-p%dj.png\">",k,k,subdirf2(optionfilefiname,"ILK_"),k,subdirf2(optionfilefiname,"ILK_"),k,subdirf2(optionfilefiname,"ILK_"),k);
       }
       fprintf(fichtm,"<br>- The function drawn is -2Log(L) in Log scale: by state of origin <a href=\"%s-ori.png\">%s-ori.png</a><br> \
   <img src=\"%s-ori.png\">",subdirf2(optionfilefiname,"ILK_"),subdirf2(optionfilefiname,"ILK_"),subdirf2(optionfilefiname,"ILK_"));
       fprintf(fichtm,"<br>- and by state of destination <a href=\"%s-dest.png\">%s-dest.png</a><br> \
   <img src=\"%s-dest.png\">",subdirf2(optionfilefiname,"ILK_"),subdirf2(optionfilefiname,"ILK_"),subdirf2(optionfilefiname,"ILK_"));
       fflush(fichtm);
     }
     return;
   }
   
   
   /*********** Maximum Likelihood Estimation ***************/
   
   void mlikeli(FILE *ficres,double p[], int npar, int ncovmodel, int nlstate, double ftol, double (*func)(double []))
   {
     int i,j,k, jk, jkk=0, iter=0;
     double **xi;
     double fret;
     double fretone; /* Only one call to likelihood */
     /*  char filerespow[FILENAMELENGTH];*/
   
   #ifdef NLOPT
     int creturn;
     nlopt_opt opt;
     /* double lb[9] = { -HUGE_VAL, -HUGE_VAL, -HUGE_VAL, -HUGE_VAL, -HUGE_VAL, -HUGE_VAL, -HUGE_VAL, -HUGE_VAL, -HUGE_VAL }; /\* lower bounds *\/ */
     double *lb;
     double minf; /* the minimum objective value, upon return */
     double * p1; /* Shifted parameters from 0 instead of 1 */
     myfunc_data dinst, *d = &dinst;
   #endif
   
   
     xi=matrix(1,npar,1,npar);
     for (i=1;i<=npar;i++)
       for (j=1;j<=npar;j++)
         xi[i][j]=(i==j ? 1.0 : 0.0);
     printf("Powell\n");  fprintf(ficlog,"Powell\n");
     strcpy(filerespow,"POW_"); 
     strcat(filerespow,fileres);
     if((ficrespow=fopen(filerespow,"w"))==NULL) {
       printf("Problem with resultfile: %s\n", filerespow);
       fprintf(ficlog,"Problem with resultfile: %s\n", filerespow);
     }
     fprintf(ficrespow,"# Powell\n# iter -2*LL");
     for (i=1;i<=nlstate;i++)
       for(j=1;j<=nlstate+ndeath;j++)
         if(j!=i)fprintf(ficrespow," p%1d%1d",i,j);
     fprintf(ficrespow,"\n");
   #ifdef POWELL
   #ifdef LINMINORIGINAL
   #else /* LINMINORIGINAL */
     
     flatdir=ivector(1,npar); 
     for (j=1;j<=npar;j++) flatdir[j]=0; 
   #endif /*LINMINORIGINAL */
   
   #ifdef FLATSUP
     powell(p,xi,npar,ftol,&iter,&fret,flatdir,func);
     /* reorganizing p by suppressing flat directions */
     for(i=1, jk=1; i <=nlstate; i++){
       for(k=1; k <=(nlstate+ndeath); k++){
         if (k != i) {
           printf("%d%d flatdir[%d]=%d",i,k,jk, flatdir[jk]);
           if(flatdir[jk]==1){
             printf(" To be skipped %d%d flatdir[%d]=%d ",i,k,jk, flatdir[jk]);
           }
           for(j=1; j <=ncovmodel; j++){
             printf("%12.7f ",p[jk]);
             jk++; 
           }
           printf("\n");
         }
       }
     }
   /* skipping */
     /* for(i=1, jk=1, jkk=1;(flatdir[jk]==0)&& (i <=nlstate); i++){ */
     for(i=1, jk=1, jkk=1;i <=nlstate; i++){
       for(k=1; k <=(nlstate+ndeath); k++){
         if (k != i) {
           printf("%d%d flatdir[%d]=%d",i,k,jk, flatdir[jk]);
           if(flatdir[jk]==1){
             printf(" To be skipped %d%d flatdir[%d]=%d jk=%d p[%d] ",i,k,jk, flatdir[jk],jk, jk);
             for(j=1; j <=ncovmodel;  jk++,j++){
               printf(" p[%d]=%12.7f",jk, p[jk]);
               /*q[jjk]=p[jk];*/
             }
           }else{
             printf(" To be kept %d%d flatdir[%d]=%d jk=%d q[%d]=p[%d] ",i,k,jk, flatdir[jk],jk, jkk, jk);
             for(j=1; j <=ncovmodel;  jk++,jkk++,j++){
               printf(" p[%d]=%12.7f=q[%d]",jk, p[jk],jkk);
               /*q[jjk]=p[jk];*/
             }
           }
           printf("\n");
         }
         fflush(stdout);
       }
     }
     powell(p,xi,npar,ftol,&iter,&fret,flatdir,func);
   #else  /* FLATSUP */
     powell(p,xi,npar,ftol,&iter,&fret,func);
   #endif  /* FLATSUP */
   
   #ifdef LINMINORIGINAL
   #else
         free_ivector(flatdir,1,npar); 
   #endif  /* LINMINORIGINAL*/
   #endif /* POWELL */
   
   #ifdef NLOPT
   #ifdef NEWUOA
     opt = nlopt_create(NLOPT_LN_NEWUOA,npar);
   #else
     opt = nlopt_create(NLOPT_LN_BOBYQA,npar);
   #endif
     lb=vector(0,npar-1);
     for (i=0;i<npar;i++) lb[i]= -HUGE_VAL;
     nlopt_set_lower_bounds(opt, lb);
     nlopt_set_initial_step1(opt, 0.1);
     
     p1= (p+1); /*  p *(p+1)@8 and p *(p1)@8 are equal p1[0]=p[1] */
     d->function = func;
     printf(" Func %.12lf \n",myfunc(npar,p1,NULL,d));
     nlopt_set_min_objective(opt, myfunc, d);
     nlopt_set_xtol_rel(opt, ftol);
     if ((creturn=nlopt_optimize(opt, p1, &minf)) < 0) {
       printf("nlopt failed! %d\n",creturn); 
     }
     else {
       printf("found minimum after %d evaluations (NLOPT=%d)\n", countcallfunc ,NLOPT);
       printf("found minimum at f(%g,%g) = %0.10g\n", p[0], p[1], minf);
       iter=1; /* not equal */
     }
     nlopt_destroy(opt);
   #endif
   #ifdef FLATSUP
     /* npared = npar -flatd/ncovmodel; */
     /* xired= matrix(1,npared,1,npared); */
     /* paramred= ma3x(1,nlstate,1,nlstate+ndeath-1,1,ncovmodel); */
     /* powell(pred,xired,npared,ftol,&iter,&fret,flatdir,func); */
     /* free_matrix(xire,1,npared,1,npared); */
   #else  /* FLATSUP */
   #endif /* FLATSUP */
     free_matrix(xi,1,npar,1,npar);
     fclose(ficrespow);
     printf("\n#Number of iterations & function calls = %d & %d, -2 Log likelihood = %.12f\n",iter, countcallfunc,func(p));
     fprintf(ficlog,"\n#Number of iterations & function calls = %d & %d, -2 Log likelihood = %.12f\n",iter, countcallfunc,func(p));
     fprintf(ficres,"#Number of iterations & function calls = %d & %d, -2 Log likelihood = %.12f\n",iter, countcallfunc,func(p));
   
   }
   
   /**** Computes Hessian and covariance matrix ***/
   void hesscov(double **matcov, double **hess, double p[], int npar, double delti[], double ftolhess, double (*func)(double []))
   {
     double  **a,**y,*x,pd;
     /* double **hess; */
     int i, j;
     int *indx;
   
     double hessii(double p[], double delta, int theta, double delti[],double (*func)(double []),int npar);
     double hessij(double p[], double **hess, double delti[], int i, int j,double (*func)(double []),int npar);
     void lubksb(double **a, int npar, int *indx, double b[]) ;
     void ludcmp(double **a, int npar, int *indx, double *d) ;
     double gompertz(double p[]);
     /* hess=matrix(1,npar,1,npar); */
   
     printf("\nCalculation of the hessian matrix. Wait...\n");
     fprintf(ficlog,"\nCalculation of the hessian matrix. Wait...\n");
     for (i=1;i<=npar;i++){
       printf("%d-",i);fflush(stdout);
       fprintf(ficlog,"%d-",i);fflush(ficlog);
      
        hess[i][i]=hessii(p,ftolhess,i,delti,func,npar);
       
       /*  printf(" %f ",p[i]);
           printf(" %lf %lf %lf",hess[i][i],ftolhess,delti[i]);*/
     }
     
     for (i=1;i<=npar;i++) {
       for (j=1;j<=npar;j++)  {
         if (j>i) { 
           printf(".%d-%d",i,j);fflush(stdout);
           fprintf(ficlog,".%d-%d",i,j);fflush(ficlog);
           hess[i][j]=hessij(p,hess, delti,i,j,func,npar);
           
           hess[j][i]=hess[i][j];    
           /*printf(" %lf ",hess[i][j]);*/
         }
       }
     }
     printf("\n");
     fprintf(ficlog,"\n");
   
     printf("\nInverting the hessian to get the covariance matrix. Wait...\n");
     fprintf(ficlog,"\nInverting the hessian to get the covariance matrix. Wait...\n");
     
     a=matrix(1,npar,1,npar);
     y=matrix(1,npar,1,npar);
     x=vector(1,npar);
     indx=ivector(1,npar);
     for (i=1;i<=npar;i++)
       for (j=1;j<=npar;j++) a[i][j]=hess[i][j];
     ludcmp(a,npar,indx,&pd);
   
     for (j=1;j<=npar;j++) {
       for (i=1;i<=npar;i++) x[i]=0;
       x[j]=1;
       lubksb(a,npar,indx,x);
       for (i=1;i<=npar;i++){ 
         matcov[i][j]=x[i];
       }
     }
   
     printf("\n#Hessian matrix#\n");
     fprintf(ficlog,"\n#Hessian matrix#\n");
     for (i=1;i<=npar;i++) { 
       for (j=1;j<=npar;j++) { 
         printf("%.6e ",hess[i][j]);
         fprintf(ficlog,"%.6e ",hess[i][j]);
       }
       printf("\n");
       fprintf(ficlog,"\n");
     }
   
     /* printf("\n#Covariance matrix#\n"); */
     /* fprintf(ficlog,"\n#Covariance matrix#\n"); */
     /* for (i=1;i<=npar;i++) {  */
     /*   for (j=1;j<=npar;j++) {  */
     /*     printf("%.6e ",matcov[i][j]); */
     /*     fprintf(ficlog,"%.6e ",matcov[i][j]); */
     /*   } */
     /*   printf("\n"); */
     /*   fprintf(ficlog,"\n"); */
     /* } */
   
     /* Recompute Inverse */
     /* for (i=1;i<=npar;i++) */
     /*   for (j=1;j<=npar;j++) a[i][j]=matcov[i][j]; */
     /* ludcmp(a,npar,indx,&pd); */
   
     /*  printf("\n#Hessian matrix recomputed#\n"); */
   
     /* for (j=1;j<=npar;j++) { */
     /*   for (i=1;i<=npar;i++) x[i]=0; */
     /*   x[j]=1; */
     /*   lubksb(a,npar,indx,x); */
     /*   for (i=1;i<=npar;i++){  */
     /*     y[i][j]=x[i]; */
     /*     printf("%.3e ",y[i][j]); */
     /*     fprintf(ficlog,"%.3e ",y[i][j]); */
     /*   } */
     /*   printf("\n"); */
     /*   fprintf(ficlog,"\n"); */
     /* } */
   
     /* Verifying the inverse matrix */
   #ifdef DEBUGHESS
     y=matprod2(y,hess,1,npar,1,npar,1,npar,matcov);
   
      printf("\n#Verification: multiplying the matrix of covariance by the Hessian matrix, should be unity:#\n");
      fprintf(ficlog,"\n#Verification: multiplying the matrix of covariance by the Hessian matrix. Should be unity:#\n");
   
     for (j=1;j<=npar;j++) {
       for (i=1;i<=npar;i++){ 
         printf("%.2f ",y[i][j]);
         fprintf(ficlog,"%.2f ",y[i][j]);
       }
       printf("\n");
       fprintf(ficlog,"\n");
     }
   #endif
   
     free_matrix(a,1,npar,1,npar);
     free_matrix(y,1,npar,1,npar);
     free_vector(x,1,npar);
     free_ivector(indx,1,npar);
     /* free_matrix(hess,1,npar,1,npar); */
   
   
   }
   
   /*************** hessian matrix ****************/
   double hessii(double x[], double delta, int theta, double delti[], double (*func)(double []), int npar)
   { /* Around values of x, computes the function func and returns the scales delti and hessian */
     int i;
     int l=1, lmax=20;
     double k1,k2, res, fx;
     double p2[MAXPARM+1]; /* identical to x */
     double delt=0.0001, delts, nkhi=10.,nkhif=1., khi=1.e-4;
     int k=0,kmax=10;
     double l1;
   
     fx=func(x);
     for (i=1;i<=npar;i++) p2[i]=x[i];
     for(l=0 ; l <=lmax; l++){  /* Enlarging the zone around the Maximum */
       l1=pow(10,l);
       delts=delt;
       for(k=1 ; k <kmax; k=k+1){
         delt = delta*(l1*k);
         p2[theta]=x[theta] +delt;
         k1=func(p2)-fx;   /* Might be negative if too close to the theoretical maximum */
         p2[theta]=x[theta]-delt;
         k2=func(p2)-fx;
         /*res= (k1-2.0*fx+k2)/delt/delt; */
         res= (k1+k2)/delt/delt/2.; /* Divided by 2 because L and not 2*L */
         
   #ifdef DEBUGHESSII
         printf("%d %d k1=%.12e k2=%.12e xk1=%.12e xk2=%.12e delt=%.12e res=%.12e l=%d k=%d,fx=%.12e\n",theta,theta,k1,k2,x[theta]+delt,x[theta]-delt,delt,res, l, k,fx);
         fprintf(ficlog,"%d %d k1=%.12e k2=%.12e xk1=%.12e xk2=%.12e delt=%.12e res=%.12e l=%d k=%d,fx=%.12e\n",theta,theta,k1,k2,x[theta]+delt,x[theta]-delt,delt,res, l, k,fx);
   #endif
         /*if(fabs(k1-2.0*fx+k2) <1.e-13){ */
         if((k1 <khi/nkhi/2.) || (k2 <khi/nkhi/2.)){
           k=kmax;
         }
         else if((k1 >khi/nkhif) || (k2 >khi/nkhif)){ /* Keeps lastvalue before 3.84/2 KHI2 5% 1d.f. */
           k=kmax; l=lmax*10;
         }
         else if((k1 >khi/nkhi) || (k2 >khi/nkhi)){ 
           delts=delt;
         }
       } /* End loop k */
     }
     delti[theta]=delts;
     return res; 
     
   }
   
   double hessij( double x[], double **hess, double delti[], int thetai,int thetaj,double (*func)(double []),int npar)
   {
     int i;
     int l=1, lmax=20;
     double k1,k2,k3,k4,res,fx;
     double p2[MAXPARM+1];
     int k, kmax=1;
     double v1, v2, cv12, lc1, lc2;
   
     int firstime=0;
     
     fx=func(x);
     for (k=1; k<=kmax; k=k+10) {
       for (i=1;i<=npar;i++) p2[i]=x[i];
       p2[thetai]=x[thetai]+delti[thetai]*k;
       p2[thetaj]=x[thetaj]+delti[thetaj]*k;
       k1=func(p2)-fx;
     
       p2[thetai]=x[thetai]+delti[thetai]*k;
       p2[thetaj]=x[thetaj]-delti[thetaj]*k;
       k2=func(p2)-fx;
     
       p2[thetai]=x[thetai]-delti[thetai]*k;
       p2[thetaj]=x[thetaj]+delti[thetaj]*k;
       k3=func(p2)-fx;
     
       p2[thetai]=x[thetai]-delti[thetai]*k;
       p2[thetaj]=x[thetaj]-delti[thetaj]*k;
       k4=func(p2)-fx;
       res=(k1-k2-k3+k4)/4.0/delti[thetai]/k/delti[thetaj]/k/2.; /* Because of L not 2*L */
       if(k1*k2*k3*k4 <0.){
         firstime=1;
         kmax=kmax+10;
       }
       if(kmax >=10 || firstime ==1){
         printf("Warning: directions %d-%d, you are not estimating the Hessian at the exact maximum likelihood; you could increase ftol=%.2e\n",thetai,thetaj, ftol);
         fprintf(ficlog,"Warning: directions %d-%d, you are not estimating the Hessian at the exact maximum likelihood; you could increase ftol=%.2e\n",thetai,thetaj, ftol);
         printf("%d %d k=%d, k1=%.12e k2=%.12e k3=%.12e k4=%.12e delti*k=%.12e deltj*k=%.12e, xi-de*k=%.12e xj-de*k=%.12e  res=%.12e k1234=%.12e,k1-2=%.12e,k3-4=%.12e\n",thetai,thetaj,k,k1,k2,k3,k4,delti[thetai]/k,delti[thetaj]/k,x[thetai]-delti[thetai]/k,x[thetaj]-delti[thetaj]/k, res,k1-k2-k3+k4,k1-k2,k3-k4);
         fprintf(ficlog,"%d %d k=%d, k1=%.12e k2=%.12e k3=%.12e k4=%.12e delti*k=%.12e deltj*k=%.12e, xi-de*k=%.12e xj-de*k=%.12e  res=%.12e k1234=%.12e,k1-2=%.12e,k3-4=%.12e\n",thetai,thetaj,k,k1,k2,k3,k4,delti[thetai]/k,delti[thetaj]/k,x[thetai]-delti[thetai]/k,x[thetaj]-delti[thetaj]/k, res,k1-k2-k3+k4,k1-k2,k3-k4);
       }
   #ifdef DEBUGHESSIJ
       v1=hess[thetai][thetai];
       v2=hess[thetaj][thetaj];
       cv12=res;
       /* Computing eigen value of Hessian matrix */
       lc1=((v1+v2)+sqrt((v1+v2)*(v1+v2) - 4*(v1*v2-cv12*cv12)))/2.;
       lc2=((v1+v2)-sqrt((v1+v2)*(v1+v2) - 4*(v1*v2-cv12*cv12)))/2.;
       if ((lc2 <0) || (lc1 <0) ){
         printf("Warning: sub Hessian matrix '%d%d' does not have positive eigen values \n",thetai,thetaj);
         fprintf(ficlog, "Warning: sub Hessian matrix '%d%d' does not have positive eigen values \n",thetai,thetaj);
         printf("%d %d k=%d, k1=%.12e k2=%.12e k3=%.12e k4=%.12e delti/k=%.12e deltj/k=%.12e, xi-de/k=%.12e xj-de/k=%.12e  res=%.12e k1234=%.12e,k1-2=%.12e,k3-4=%.12e\n",thetai,thetaj,k,k1,k2,k3,k4,delti[thetai]/k,delti[thetaj]/k,x[thetai]-delti[thetai]/k,x[thetaj]-delti[thetaj]/k, res,k1-k2-k3+k4,k1-k2,k3-k4);
         fprintf(ficlog,"%d %d k=%d, k1=%.12e k2=%.12e k3=%.12e k4=%.12e delti/k=%.12e deltj/k=%.12e, xi-de/k=%.12e xj-de/k=%.12e  res=%.12e k1234=%.12e,k1-2=%.12e,k3-4=%.12e\n",thetai,thetaj,k,k1,k2,k3,k4,delti[thetai]/k,delti[thetaj]/k,x[thetai]-delti[thetai]/k,x[thetaj]-delti[thetaj]/k, res,k1-k2-k3+k4,k1-k2,k3-k4);
       }
   #endif
     }
     return res;
   }
   
       /* Not done yet: Was supposed to fix if not exactly at the maximum */
   /* double hessij( double x[], double delti[], int thetai,int thetaj,double (*func)(double []),int npar) */
   /* { */
   /*   int i; */
   /*   int l=1, lmax=20; */
   /*   double k1,k2,k3,k4,res,fx; */
   /*   double p2[MAXPARM+1]; */
   /*   double delt=0.0001, delts, nkhi=10.,nkhif=1., khi=1.e-4; */
   /*   int k=0,kmax=10; */
   /*   double l1; */
     
   /*   fx=func(x); */
   /*   for(l=0 ; l <=lmax; l++){  /\* Enlarging the zone around the Maximum *\/ */
   /*     l1=pow(10,l); */
   /*     delts=delt; */
   /*     for(k=1 ; k <kmax; k=k+1){ */
   /*       delt = delti*(l1*k); */
   /*       for (i=1;i<=npar;i++) p2[i]=x[i]; */
   /*       p2[thetai]=x[thetai]+delti[thetai]/k; */
   /*       p2[thetaj]=x[thetaj]+delti[thetaj]/k; */
   /*       k1=func(p2)-fx; */
         
   /*       p2[thetai]=x[thetai]+delti[thetai]/k; */
   /*       p2[thetaj]=x[thetaj]-delti[thetaj]/k; */
   /*       k2=func(p2)-fx; */
         
   /*       p2[thetai]=x[thetai]-delti[thetai]/k; */
   /*       p2[thetaj]=x[thetaj]+delti[thetaj]/k; */
   /*       k3=func(p2)-fx; */
         
   /*       p2[thetai]=x[thetai]-delti[thetai]/k; */
   /*       p2[thetaj]=x[thetaj]-delti[thetaj]/k; */
   /*       k4=func(p2)-fx; */
   /*       res=(k1-k2-k3+k4)/4.0/delti[thetai]*k/delti[thetaj]*k/2.; /\* Because of L not 2*L *\/ */
   /* #ifdef DEBUGHESSIJ */
   /*       printf("%d %d k=%d, k1=%.12e k2=%.12e k3=%.12e k4=%.12e delti/k=%.12e deltj/k=%.12e, xi-de/k=%.12e xj-de/k=%.12e  res=%.12e k1234=%.12e,k1-2=%.12e,k3-4=%.12e\n",thetai,thetaj,k,k1,k2,k3,k4,delti[thetai]/k,delti[thetaj]/k,x[thetai]-delti[thetai]/k,x[thetaj]-delti[thetaj]/k, res,k1-k2-k3+k4,k1-k2,k3-k4); */
   /*       fprintf(ficlog,"%d %d k=%d, k1=%.12e k2=%.12e k3=%.12e k4=%.12e delti/k=%.12e deltj/k=%.12e, xi-de/k=%.12e xj-de/k=%.12e  res=%.12e k1234=%.12e,k1-2=%.12e,k3-4=%.12e\n",thetai,thetaj,k,k1,k2,k3,k4,delti[thetai]/k,delti[thetaj]/k,x[thetai]-delti[thetai]/k,x[thetaj]-delti[thetaj]/k, res,k1-k2-k3+k4,k1-k2,k3-k4); */
   /* #endif */
   /*       if((k1 <khi/nkhi/2.) || (k2 <khi/nkhi/2.)|| (k4 <khi/nkhi/2.)|| (k4 <khi/nkhi/2.)){ */
   /*      k=kmax; */
   /*       } */
   /*       else if((k1 >khi/nkhif) || (k2 >khi/nkhif) || (k4 >khi/nkhif) || (k4 >khi/nkhif)){ /\* Keeps lastvalue before 3.84/2 KHI2 5% 1d.f. *\/ */
   /*      k=kmax; l=lmax*10; */
   /*       } */
   /*       else if((k1 >khi/nkhi) || (k2 >khi/nkhi)){  */
   /*      delts=delt; */
   /*       } */
   /*     } /\* End loop k *\/ */
   /*   } */
   /*   delti[theta]=delts; */
   /*   return res;  */
   /* } */
   
   
   /************** Inverse of matrix **************/
   void ludcmp(double **a, int n, int *indx, double *d) 
   { 
     int i,imax,j,k; 
     double big,dum,sum,temp; 
     double *vv; 
    
     vv=vector(1,n); 
     *d=1.0; 
     for (i=1;i<=n;i++) { 
       big=0.0; 
       for (j=1;j<=n;j++) 
         if ((temp=fabs(a[i][j])) > big) big=temp; 
       if (big == 0.0){
         printf(" Singular Hessian matrix at row %d:\n",i);
         for (j=1;j<=n;j++) {
           printf(" a[%d][%d]=%f,",i,j,a[i][j]);
           fprintf(ficlog," a[%d][%d]=%f,",i,j,a[i][j]);
         }
         fflush(ficlog);
         fclose(ficlog);
         nrerror("Singular matrix in routine ludcmp"); 
       }
     vv[i]=1.0/big;       vv[i]=1.0/big; 
   }     } 
   for (j=1;j<=n;j++) {     for (j=1;j<=n;j++) { 
     for (i=1;i<j;i++) {       for (i=1;i<j;i++) { 
       sum=a[i][j];         sum=a[i][j]; 
       for (k=1;k<i;k++) sum -= a[i][k]*a[k][j];         for (k=1;k<i;k++) sum -= a[i][k]*a[k][j]; 
       a[i][j]=sum;         a[i][j]=sum; 
     }       } 
     big=0.0;       big=0.0; 
     for (i=j;i<=n;i++) {       for (i=j;i<=n;i++) { 
       sum=a[i][j];         sum=a[i][j]; 
       for (k=1;k<j;k++)         for (k=1;k<j;k++) 
         sum -= a[i][k]*a[k][j];           sum -= a[i][k]*a[k][j]; 
       a[i][j]=sum;         a[i][j]=sum; 
       if ( (dum=vv[i]*fabs(sum)) >= big) {         if ( (dum=vv[i]*fabs(sum)) >= big) { 
         big=dum;           big=dum; 
         imax=i;           imax=i; 
       }         } 
     }       } 
     if (j != imax) {       if (j != imax) { 
       for (k=1;k<=n;k++) {         for (k=1;k<=n;k++) { 
         dum=a[imax][k];           dum=a[imax][k]; 
         a[imax][k]=a[j][k];           a[imax][k]=a[j][k]; 
         a[j][k]=dum;           a[j][k]=dum; 
       }         } 
       *d = -(*d);         *d = -(*d); 
       vv[imax]=vv[j];         vv[imax]=vv[j]; 
     }       } 
     indx[j]=imax;       indx[j]=imax; 
     if (a[j][j] == 0.0) a[j][j]=TINY;       if (a[j][j] == 0.0) a[j][j]=TINY; 
     if (j != n) {       if (j != n) { 
       dum=1.0/(a[j][j]);         dum=1.0/(a[j][j]); 
       for (i=j+1;i<=n;i++) a[i][j] *= dum;         for (i=j+1;i<=n;i++) a[i][j] *= dum; 
     }       } 
     } 
     free_vector(vv,1,n);  /* Doesn't work */
   ;
   } 
   
   void lubksb(double **a, int n, int *indx, double b[]) 
   { 
     int i,ii=0,ip,j; 
     double sum; 
    
     for (i=1;i<=n;i++) { 
       ip=indx[i]; 
       sum=b[ip]; 
       b[ip]=b[i]; 
       if (ii) 
         for (j=ii;j<=i-1;j++) sum -= a[i][j]*b[j]; 
       else if (sum) ii=i; 
       b[i]=sum; 
     } 
     for (i=n;i>=1;i--) { 
       sum=b[i]; 
       for (j=i+1;j<=n;j++) sum -= a[i][j]*b[j]; 
       b[i]=sum/a[i][i]; 
     } 
   } 
   
   void pstamp(FILE *fichier)
   {
     fprintf(fichier,"# %s.%s\n#IMaCh version %s, %s\n#%s\n# %s", optionfilefiname,optionfilext,version,copyright, fullversion, strstart);
   }
   
   void date2dmy(double date,double *day, double *month, double *year){
     double yp=0., yp1=0., yp2=0.;
     
     yp1=modf(date,&yp);/* extracts integral of date in yp  and
                           fractional in yp1 */
     *year=yp;
     yp2=modf((yp1*12),&yp);
     *month=yp;
     yp1=modf((yp2*30.5),&yp);
     *day=yp;
     if(*day==0) *day=1;
     if(*month==0) *month=1;
   }
   
   
   
   /************ Frequencies ********************/
   void  freqsummary(char fileres[], double p[], double pstart[], int iagemin, int iagemax, int **s, double **agev, int nlstate, int imx, \
                     int *Tvaraff, int *invalidvarcomb, int **nbcode, int *ncodemax,double **mint,double **anint, char strstart[], \
                     int firstpass,  int lastpass, int stepm, int weightopt, char model[])
   {  /* Some frequencies as well as proposing some starting values */
     /* Frequencies of any combination of dummy covariate used in the model equation */ 
     int i, m, jk, j1, bool, z1,j, nj, nl, k, iv, jj=0, s1=1, s2=1;
     int iind=0, iage=0;
     int mi; /* Effective wave */
     int first;
     double ***freq; /* Frequencies */
     double *x, *y, a=0.,b=0.,r=1., sa=0., sb=0.; /* for regression, y=b+m*x and r is the correlation coefficient */
     int no=0, linreg(int ifi, int ila, int *no, const double x[], const double y[], double* a, double* b, double* r, double* sa, double * sb);
     double *meanq, *stdq, *idq;
     double **meanqt;
     double *pp, **prop, *posprop, *pospropt;
     double pos=0., posproptt=0., pospropta=0., k2, dateintsum=0,k2cpt=0;
     char fileresp[FILENAMELENGTH], fileresphtm[FILENAMELENGTH], fileresphtmfr[FILENAMELENGTH];
     double agebegin, ageend;
       
     pp=vector(1,nlstate);
     prop=matrix(1,nlstate,iagemin-AGEMARGE,iagemax+4+AGEMARGE); 
     posprop=vector(1,nlstate); /* Counting the number of transition starting from a live state per age */ 
     pospropt=vector(1,nlstate); /* Counting the number of transition starting from a live state */ 
     /* prop=matrix(1,nlstate,iagemin,iagemax+3); */
     meanq=vector(1,nqfveff); /* Number of Quantitative Fixed Variables Effective */
     stdq=vector(1,nqfveff); /* Number of Quantitative Fixed Variables Effective */
     idq=vector(1,nqfveff); /* Number of Quantitative Fixed Variables Effective */
     meanqt=matrix(1,lastpass,1,nqtveff);
     strcpy(fileresp,"P_");
     strcat(fileresp,fileresu);
     /*strcat(fileresphtm,fileresu);*/
     if((ficresp=fopen(fileresp,"w"))==NULL) {
       printf("Problem with prevalence resultfile: %s\n", fileresp);
       fprintf(ficlog,"Problem with prevalence resultfile: %s\n", fileresp);
       exit(0);
     }
     
     strcpy(fileresphtm,subdirfext(optionfilefiname,"PHTM_",".htm"));
     if((ficresphtm=fopen(fileresphtm,"w"))==NULL) {
       printf("Problem with prevalence HTM resultfile '%s' with errno='%s'\n",fileresphtm,strerror(errno));
       fprintf(ficlog,"Problem with prevalence HTM resultfile '%s' with errno='%s'\n",fileresphtm,strerror(errno));
       fflush(ficlog);
       exit(70); 
     }
     else{
       fprintf(ficresphtm,"<html><head>\n<title>IMaCh PHTM_ %s</title></head>\n <body><font size=\"2\">%s <br> %s</font> \
   <hr size=\"2\" color=\"#EC5E5E\"> \n                                    \
   Title=%s <br>Datafile=%s Firstpass=%d Lastpass=%d Stepm=%d Weight=%d Model=1+age+%s<br>\n",\
               fileresphtm,version,fullversion,title,datafile,firstpass,lastpass,stepm, weightopt, model);
     }
     fprintf(ficresphtm,"Current page is file <a href=\"%s\">%s</a><br>\n\n<h4>Frequencies (weight=%d) and prevalence by age at begin of transition and dummy covariate value at beginning of transition</h4>\n",fileresphtm, fileresphtm, weightopt);
     
     strcpy(fileresphtmfr,subdirfext(optionfilefiname,"PHTMFR_",".htm"));
     if((ficresphtmfr=fopen(fileresphtmfr,"w"))==NULL) {
       printf("Problem with frequency table HTM resultfile '%s' with errno='%s'\n",fileresphtmfr,strerror(errno));
       fprintf(ficlog,"Problem with frequency table HTM resultfile '%s' with errno='%s'\n",fileresphtmfr,strerror(errno));
       fflush(ficlog);
       exit(70); 
     } else{
       fprintf(ficresphtmfr,"<html><head>\n<title>IMaCh PHTM_Frequency table %s</title></head>\n <body><font size=\"2\">%s <br> %s</font> \
   ,<hr size=\"2\" color=\"#EC5E5E\"> \n                                   \
   Title=%s <br>Datafile=%s Firstpass=%d Lastpass=%d Stepm=%d Weight=%d Model=1+age+%s<br>\n",\
               fileresphtmfr,version,fullversion,title,datafile,firstpass,lastpass,stepm, weightopt, model);
     }
     fprintf(ficresphtmfr,"Current page is file <a href=\"%s\">%s</a><br>\n\n<h4>(weight=%d) frequencies of all effective transitions of the model, by age at begin of transition, and covariate value at the begin of transition (if the covariate is a varying covariate) </h4>Unknown status is -1<br/>\n",fileresphtmfr, fileresphtmfr,weightopt);
     
     y= vector(iagemin-AGEMARGE,iagemax+4+AGEMARGE);
     x= vector(iagemin-AGEMARGE,iagemax+4+AGEMARGE);
     freq= ma3x(-5,nlstate+ndeath,-5,nlstate+ndeath,iagemin-AGEMARGE,iagemax+4+AGEMARGE);
     j1=0;
     
     /* j=ncoveff;  /\* Only fixed dummy covariates *\/ */
     j=cptcoveff;  /* Only simple dummy covariates used in the model */
     /* j=cptcovn;  /\* Only dummy covariates of the model *\/ */
     if (cptcovn<1) {j=1;ncodemax[1]=1;}
     
     
     /* Detects if a combination j1 is empty: for a multinomial variable like 3 education levels:
        reference=low_education V1=0,V2=0
        med_educ                V1=1 V2=0, 
        high_educ               V1=0 V2=1
        Then V1=1 and V2=1 is a noisy combination that we want to exclude for the list 2**cptcovn 
     */
     dateintsum=0;
     k2cpt=0;
   
     if(cptcoveff == 0 )
       nl=1;  /* Constant and age model only */
     else
       nl=2;
   
     /* if a constant only model, one pass to compute frequency tables and to write it on ficresp */
     /* Loop on nj=1 or 2 if dummy covariates j!=0
      *   Loop on j1(1 to 2**cptcoveff) covariate combination
      *     freq[s1][s2][iage] =0.
      *     Loop on iind
      *       ++freq[s1][s2][iage] weighted
      *     end iind
      *     if covariate and j!0
      *       headers Variable on one line
      *     endif cov j!=0
      *     header of frequency table by age
      *     Loop on age
      *       pp[s1]+=freq[s1][s2][iage] weighted
      *       pos+=freq[s1][s2][iage] weighted
      *       Loop on s1 initial state
      *         fprintf(ficresp
      *       end s1
      *     end age
      *     if j!=0 computes starting values
      *     end compute starting values
      *   end j1
      * end nl 
      */
     for (nj = 1; nj <= nl; nj++){   /* nj= 1 constant model, nl number of loops. */
       if(nj==1)
         j=0;  /* First pass for the constant */
       else{
         j=cptcoveff; /* Other passes for the covariate values number of simple covariates in the model V2+V1 =2 (simple dummy fixed or time varying) */
       }
       first=1;
       for (j1 = 1; j1 <= (int) pow(2,j); j1++){ /* Loop on all dummy covariates combination of the model, ie excluding quantitatives, V4=0, V3=0 for example, fixed or varying covariates */
         posproptt=0.;
         /*printf("cptcovn=%d Tvaraff=%d", cptcovn,Tvaraff[1]);
           scanf("%d", i);*/
         for (i=-5; i<=nlstate+ndeath; i++)  
           for (s2=-5; s2<=nlstate+ndeath; s2++)  
             for(m=iagemin; m <= iagemax+3; m++)
               freq[i][s2][m]=0;
         
         for (i=1; i<=nlstate; i++)  {
           for(m=iagemin; m <= iagemax+3; m++)
             prop[i][m]=0;
           posprop[i]=0;
           pospropt[i]=0;
         }
         for (z1=1; z1<= nqfveff; z1++) { /* zeroing for each combination j1 as well as for the total */
           idq[z1]=0.;
           meanq[z1]=0.;
           stdq[z1]=0.;
         }
         /* for (z1=1; z1<= nqtveff; z1++) { */
         /*   for(m=1;m<=lastpass;m++){ */
         /*          meanqt[m][z1]=0.; */
         /*        } */
         /* }       */
         /* dateintsum=0; */
         /* k2cpt=0; */
         
         /* For that combination of covariates j1 (V4=1 V3=0 for example), we count and print the frequencies in one pass */
         for (iind=1; iind<=imx; iind++) { /* For each individual iind */
           bool=1;
           if(j !=0){
             if(anyvaryingduminmodel==0){ /* If All fixed covariates */
               if (cptcoveff >0) { /* Filter is here: Must be looked at for model=V1+V2+V3+V4 */
                 for (z1=1; z1<=cptcoveff; z1++) { /* loops on covariates in the model */
                   /* if(Tvaraff[z1] ==-20){ */
                   /*       /\* sumnew+=cotvar[mw[mi][iind]][z1][iind]; *\/ */
                   /* }else  if(Tvaraff[z1] ==-10){ */
                   /*       /\* sumnew+=coqvar[z1][iind]; *\/ */
                   /* }else  */ /* TODO TODO codtabm(j1,z1) or codtabm(j1,Tvaraff[z1]]z1)*/
                   /* if( iind >=imx-3) printf("Searching error iind=%d Tvaraff[z1]=%d covar[Tvaraff[z1]][iind]=%.f TnsdVar[Tvaraff[z1]]=%d, cptcoveff=%d, cptcovs=%d \n",iind, Tvaraff[z1], covar[Tvaraff[z1]][iind],TnsdVar[Tvaraff[z1]],cptcoveff, cptcovs); */
                   if(Tvaraff[z1]<1 || Tvaraff[z1]>=NCOVMAX)
                     printf("Error Tvaraff[z1]=%d<1 or >=%d, cptcoveff=%d model=%s\n",Tvaraff[z1],NCOVMAX, cptcoveff, model);
                   if (covar[Tvaraff[z1]][iind]!= nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]){ /* for combination j1 of covariates */
                     /* Tests if the value of the covariate z1 for this individual iind responded to combination j1 (V4=1 V3=0) */
                     bool=0; /* bool should be equal to 1 to be selected, one covariate value failed */
                     /* printf("bool=%d i=%d, z1=%d, Tvaraff[%d]=%d, covar[Tvarff][%d]=%2f, codtabm(%d,%d)=%d, nbcode[Tvaraff][codtabm(%d,%d)=%d, j1=%d\n", */
                     /*   bool,i,z1, z1, Tvaraff[z1],i,covar[Tvaraff[z1]][i],j1,z1,codtabm(j1,z1),*/
                     /*   j1,z1,nbcode[Tvaraff[z1]][codtabm(j1,z1)],j1);*/
                     /* For j1=7 in V1+V2+V3+V4 = 0 1 1 0 and codtabm(7,3)=1 and nbcde[3][?]=1*/
                   } /* Onlyf fixed */
                 } /* end z1 */
               } /* cptcoveff > 0 */
             } /* end any */
           }/* end j==0 */
           if (bool==1){ /* We selected an individual iind satisfying combination j1 (V4=1 V3=0) or all fixed covariates */
             /* for(m=firstpass; m<=lastpass; m++){ */
             for(mi=1; mi<wav[iind];mi++){ /* For each wave */
               m=mw[mi][iind];
               if(j!=0){
                 if(anyvaryingduminmodel==1){ /* Some are varying covariates */
                   for (z1=1; z1<=cptcoveff; z1++) {
                     if( Fixed[Tmodelind[z1]]==1){
                       iv= Tvar[Tmodelind[z1]]-ncovcol-nqv;
                       if (cotvar[m][iv][iind]!= nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]) /* iv=1 to ntv, right modality. If covariate's 
                                                                                         value is -1, we don't select. It differs from the 
                                                                                         constant and age model which counts them. */
                         bool=0; /* not selected */
                     }else if( Fixed[Tmodelind[z1]]== 0) { /* fixed */
                       /* i1=Tvaraff[z1]; */
                       /* i2=TnsdVar[i1]; */
                       /* i3=nbcode[i1][i2]; */
                       /* i4=covar[i1][iind]; */
                       /* if(i4 != i3){ */
                       if (covar[Tvaraff[z1]][iind]!= nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]) { /* Bug valgrind */
                         bool=0;
                       }
                     }
                   }
                 }/* Some are varying covariates, we tried to speed up if all fixed covariates in the model, avoiding waves loop  */
               } /* end j==0 */
               /* bool =0 we keep that guy which corresponds to the combination of dummy values */
               if(bool==1){ /*Selected */
                 /* dh[m][iind] or dh[mw[mi][iind]][iind] is the delay between two effective (mi) waves m=mw[mi][iind]
                    and mw[mi+1][iind]. dh depends on stepm. */
                 agebegin=agev[m][iind]; /* Age at beginning of wave before transition*/
                 ageend=agev[m][iind]+(dh[m][iind])*stepm/YEARM; /* Age at end of wave and transition */
                 if(m >=firstpass && m <=lastpass){
                   k2=anint[m][iind]+(mint[m][iind]/12.);
                   /*if ((k2>=dateprev1) && (k2<=dateprev2)) {*/
                   if(agev[m][iind]==0) agev[m][iind]=iagemax+1;  /* All ages equal to 0 are in iagemax+1 */
                   if(agev[m][iind]==1) agev[m][iind]=iagemax+2;  /* All ages equal to 1 are in iagemax+2 */
                   if (s[m][iind]>0 && s[m][iind]<=nlstate)  /* If status at wave m is known and a live state */
                     prop[s[m][iind]][(int)agev[m][iind]] += weight[iind];  /* At age of beginning of transition, where status is known */
                   if (m<lastpass) {
                     /* if(s[m][iind]==4 && s[m+1][iind]==4) */
                     /*   printf(" num=%ld m=%d, iind=%d s1=%d s2=%d agev at m=%d\n", num[iind], m, iind,s[m][iind],s[m+1][iind], (int)agev[m][iind]); */
                     if(s[m][iind]==-1)
                       printf(" num=%ld m=%d, iind=%d s1=%d s2=%d agev at m=%d agebegin=%.2f ageend=%.2f, agemed=%d\n", num[iind], m, iind,s[m][iind],s[m+1][iind], (int)agev[m][iind],agebegin, ageend, (int)((agebegin+ageend)/2.));
                     freq[s[m][iind]][s[m+1][iind]][(int)agev[m][iind]] += weight[iind]; /* At age of beginning of transition, where status is known */
                     for (z1=1; z1<= nqfveff; z1++) { /* Quantitative variables, calculating mean on known values only */
                       if(!isnan(covar[ncovcol+z1][iind])){
                         idq[z1]=idq[z1]+weight[iind];
                         meanq[z1]+=covar[ncovcol+z1][iind]*weight[iind];  /* Computes mean of quantitative with selected filter */
                         /* stdq[z1]+=covar[ncovcol+z1][iind]*covar[ncovcol+z1][iind]*weight[iind]*weight[iind]; *//*error*/
                         stdq[z1]+=covar[ncovcol+z1][iind]*covar[ncovcol+z1][iind]*weight[iind]; /* *weight[iind];*/  /* Computes mean of quantitative with selected filter */
                       }
                     }
                     /* if((int)agev[m][iind] == 55) */
                     /*   printf("j=%d, j1=%d Age %d, iind=%d, num=%09ld m=%d\n",j,j1,(int)agev[m][iind],iind, num[iind],m); */
                     /* freq[s[m][iind]][s[m+1][iind]][(int)((agebegin+ageend)/2.)] += weight[iind]; */
                     freq[s[m][iind]][s[m+1][iind]][iagemax+3] += weight[iind]; /* Total is in iagemax+3 *//* At age of beginning of transition, where status is known */
                   }
                 } /* end if between passes */  
                 if ((agev[m][iind]>1) && (agev[m][iind]< (iagemax+3)) && (anint[m][iind]!=9999) && (mint[m][iind]!=99) && (j==0)) {
                   dateintsum=dateintsum+k2; /* on all covariates ?*/
                   k2cpt++;
                   /* printf("iind=%ld dateintmean = %lf dateintsum=%lf k2cpt=%lf k2=%lf\n",iind, dateintsum/k2cpt, dateintsum,k2cpt, k2); */
                 }
               }else{
                 bool=1;
               }/* end bool 2 */
             } /* end m */
             /* for (z1=1; z1<= nqfveff; z1++) { /\* Quantitative variables, calculating mean *\/ */
             /*   idq[z1]=idq[z1]+weight[iind]; */
             /*   meanq[z1]+=covar[ncovcol+z1][iind]*weight[iind];  /\* Computes mean of quantitative with selected filter *\/ */
             /*   stdq[z1]+=covar[ncovcol+z1][iind]*covar[ncovcol+z1][iind]*weight[iind]*weight[iind]; /\* *weight[iind];*\/  /\* Computes mean of quantitative with selected filter *\/ */
             /* } */
           } /* end bool */
         } /* end iind = 1 to imx */
         /* prop[s][age] is fed for any initial and valid live state as well as
            freq[s1][s2][age] at single age of beginning the transition, for a combination j1 */
         
         
         /*      fprintf(ficresp, "#Count between %.lf/%.lf/%.lf and %.lf/%.lf/%.lf\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2);*/
         if(cptcoveff==0 && nj==1) /* no covariate and first pass */
           pstamp(ficresp);
         if  (cptcoveff>0 && j!=0){
           pstamp(ficresp);
           printf( "\n#********** Variable "); 
           fprintf(ficresp, "\n#********** Variable "); 
           fprintf(ficresphtm, "\n<br/><br/><h3>********** Variable "); 
           fprintf(ficresphtmfr, "\n<br/><br/><h3>********** Variable "); 
           fprintf(ficlog, "\n#********** Variable "); 
           for (z1=1; z1<=cptcovs; z1++){
             if(!FixedV[Tvaraff[z1]]){
               printf( "V%d(fixed)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
               fprintf(ficresp, "V%d(fixed)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
               fprintf(ficresphtm, "V%d(fixed)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
               fprintf(ficresphtmfr, "V%d(fixed)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
               fprintf(ficlog, "V%d(fixed)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
             }else{
               printf( "V%d(varying)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
               fprintf(ficresp, "V%d(varying)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
               fprintf(ficresphtm, "V%d(varying)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
               fprintf(ficresphtmfr, "V%d(varying)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
               fprintf(ficlog, "V%d(varying)=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
             }
           }
           printf( "**********\n#");
           fprintf(ficresp, "**********\n#");
           fprintf(ficresphtm, "**********</h3>\n");
           fprintf(ficresphtmfr, "**********</h3>\n");
           fprintf(ficlog, "**********\n");
         }
         /*
           Printing means of quantitative variables if any
         */
         for (z1=1; z1<= nqfveff; z1++) {
           fprintf(ficlog,"Mean of fixed quantitative variable V%d on %.3g (weighted) individuals sum=%f", ncovcol+z1, idq[z1], meanq[z1]);
           fprintf(ficlog,", mean=%.3g\n",meanq[z1]/idq[z1]);
           if(weightopt==1){
             printf(" Weighted mean and standard deviation of");
             fprintf(ficlog," Weighted mean and standard deviation of");
             fprintf(ficresphtmfr," Weighted mean and standard deviation of");
           }
           /* mu = \frac{w x}{\sum w}
              var = \frac{\sum w (x-mu)^2}{\sum w} = \frac{w x^2}{\sum w} - mu^2 
           */
           printf(" fixed quantitative variable V%d on  %.3g (weighted) representatives of the population : %8.5g (%8.5g)\n", ncovcol+z1, idq[z1],meanq[z1]/idq[z1], sqrt(stdq[z1]/idq[z1]-meanq[z1]*meanq[z1]/idq[z1]/idq[z1]));
           fprintf(ficlog," fixed quantitative variable V%d on  %.3g (weighted) representatives of the population : %8.5g (%8.5g)\n", ncovcol+z1, idq[z1],meanq[z1]/idq[z1], sqrt(stdq[z1]/idq[z1]-meanq[z1]*meanq[z1]/idq[z1]/idq[z1]));
           fprintf(ficresphtmfr," fixed quantitative variable V%d on %.3g (weighted) representatives of the population : %8.5g (%8.5g)<p>\n", ncovcol+z1, idq[z1],meanq[z1]/idq[z1], sqrt(stdq[z1]/idq[z1]-meanq[z1]*meanq[z1]/idq[z1]/idq[z1]));
         }
         /* for (z1=1; z1<= nqtveff; z1++) { */
         /*        for(m=1;m<=lastpass;m++){ */
         /*          fprintf(ficresphtmfr,"V quantitative id %d, pass id=%d, mean=%f<p>\n", z1, m, meanqt[m][z1]); */
         /*   } */
         /* } */
   
         fprintf(ficresphtm,"<table style=\"text-align:center; border: 1px solid\">");
         if((cptcoveff==0 && nj==1)|| nj==2 ) /* no covariate and first pass */
           fprintf(ficresp, " Age");
         if(nj==2) for (z1=1; z1<=cptcoveff; z1++) {
             printf(" V%d=%d, z1=%d, Tvaraff[z1]=%d, j1=%d, TnsdVar[Tvaraff[%d]]=%d |",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])], z1, Tvaraff[z1], j1,z1,TnsdVar[Tvaraff[z1]]);
             fprintf(ficresp, " V%d=%d",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
           }
         for(i=1; i<=nlstate;i++) {
           if((cptcoveff==0 && nj==1)|| nj==2 ) fprintf(ficresp," Prev(%d)  N(%d)  N  ",i,i);
           fprintf(ficresphtm, "<th>Age</th><th>Prev(%d)</th><th>N(%d)</th><th>N</th>",i,i);
         }
         if((cptcoveff==0 && nj==1)|| nj==2 ) fprintf(ficresp, "\n");
         fprintf(ficresphtm, "\n");
         
         /* Header of frequency table by age */
         fprintf(ficresphtmfr,"<table style=\"text-align:center; border: 1px solid\">");
         fprintf(ficresphtmfr,"<th>Age</th> ");
         for(s2=-1; s2 <=nlstate+ndeath; s2++){
           for(m=-1; m <=nlstate+ndeath; m++){
             if(s2!=0 && m!=0)
               fprintf(ficresphtmfr,"<th>%d%d</th> ",s2,m);
           }
         }
         fprintf(ficresphtmfr, "\n");
       
         /* For each age */
         for(iage=iagemin; iage <= iagemax+3; iage++){
           fprintf(ficresphtm,"<tr>");
           if(iage==iagemax+1){
             fprintf(ficlog,"1");
             fprintf(ficresphtmfr,"<tr><th>0</th> ");
           }else if(iage==iagemax+2){
             fprintf(ficlog,"0");
             fprintf(ficresphtmfr,"<tr><th>Unknown</th> ");
           }else if(iage==iagemax+3){
             fprintf(ficlog,"Total");
             fprintf(ficresphtmfr,"<tr><th>Total</th> ");
           }else{
             if(first==1){
               first=0;
               printf("See log file for details...\n");
             }
             fprintf(ficresphtmfr,"<tr><th>%d</th> ",iage);
             fprintf(ficlog,"Age %d", iage);
           }
           for(s1=1; s1 <=nlstate ; s1++){
             for(m=-1, pp[s1]=0; m <=nlstate+ndeath ; m++)
               pp[s1] += freq[s1][m][iage]; 
           }
           for(s1=1; s1 <=nlstate ; s1++){
             for(m=-1, pos=0; m <=0 ; m++)
               pos += freq[s1][m][iage];
             if(pp[s1]>=1.e-10){
               if(first==1){
                 printf(" %d.=%.0f loss[%d]=%.1f%%",s1,pp[s1],s1,100*pos/pp[s1]);
               }
               fprintf(ficlog," %d.=%.0f loss[%d]=%.1f%%",s1,pp[s1],s1,100*pos/pp[s1]);
             }else{
               if(first==1)
                 printf(" %d.=%.0f loss[%d]=NaNQ%%",s1,pp[s1],s1);
               fprintf(ficlog," %d.=%.0f loss[%d]=NaNQ%%",s1,pp[s1],s1);
             }
           }
         
           for(s1=1; s1 <=nlstate ; s1++){ 
             /* posprop[s1]=0; */
             for(m=0, pp[s1]=0; m <=nlstate+ndeath; m++)/* Summing on all ages */
               pp[s1] += freq[s1][m][iage];
           }       /* pp[s1] is the total number of transitions starting from state s1 and any ending status until this age */
         
           for(s1=1,pos=0, pospropta=0.; s1 <=nlstate ; s1++){
             pos += pp[s1]; /* pos is the total number of transitions until this age */
             posprop[s1] += prop[s1][iage]; /* prop is the number of transitions from a live state
                                               from s1 at age iage prop[s[m][iind]][(int)agev[m][iind]] += weight[iind] */
             pospropta += prop[s1][iage]; /* prop is the number of transitions from a live state
                                             from s1 at age iage prop[s[m][iind]][(int)agev[m][iind]] += weight[iind] */
           }
           
           /* Writing ficresp */
           if(cptcoveff==0 && nj==1){ /* no covariate and first pass */
             if( iage <= iagemax){
               fprintf(ficresp," %d",iage);
             }
           }else if( nj==2){
             if( iage <= iagemax){
               fprintf(ficresp," %d",iage);
               for (z1=1; z1<=cptcoveff; z1++) fprintf(ficresp, " %d %d",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]);
             }
           }
           for(s1=1; s1 <=nlstate ; s1++){
             if(pos>=1.e-5){
               if(first==1)
                 printf(" %d.=%.0f prev[%d]=%.1f%%",s1,pp[s1],s1,100*pp[s1]/pos);
               fprintf(ficlog," %d.=%.0f prev[%d]=%.1f%%",s1,pp[s1],s1,100*pp[s1]/pos);
             }else{
               if(first==1)
                 printf(" %d.=%.0f prev[%d]=NaNQ%%",s1,pp[s1],s1);
               fprintf(ficlog," %d.=%.0f prev[%d]=NaNQ%%",s1,pp[s1],s1);
             }
             if( iage <= iagemax){
               if(pos>=1.e-5){
                 if(cptcoveff==0 && nj==1){ /* no covariate and first pass */
                   fprintf(ficresp," %.5f %.0f %.0f",prop[s1][iage]/pospropta, prop[s1][iage],pospropta);
                 }else if( nj==2){
                   fprintf(ficresp," %.5f %.0f %.0f",prop[s1][iage]/pospropta, prop[s1][iage],pospropta);
                 }
                 fprintf(ficresphtm,"<th>%d</th><td>%.5f</td><td>%.0f</td><td>%.0f</td>",iage,prop[s1][iage]/pospropta, prop[s1][iage],pospropta);
                 /*probs[iage][s1][j1]= pp[s1]/pos;*/
                 /*printf("\niage=%d s1=%d j1=%d %.5f %.0f %.0f %f",iage,s1,j1,pp[s1]/pos, pp[s1],pos,probs[iage][s1][j1]);*/
               } else{
                 if((cptcoveff==0 && nj==1)|| nj==2 ) fprintf(ficresp," NaNq %.0f %.0f",prop[s1][iage],pospropta);
                 fprintf(ficresphtm,"<th>%d</th><td>NaNq</td><td>%.0f</td><td>%.0f</td>",iage, prop[s1][iage],pospropta);
               }
             }
             pospropt[s1] +=posprop[s1];
           } /* end loop s1 */
           /* pospropt=0.; */
           for(s1=-1; s1 <=nlstate+ndeath; s1++){
             for(m=-1; m <=nlstate+ndeath; m++){
               if(freq[s1][m][iage] !=0 ) { /* minimizing output */
                 if(first==1){
                   printf(" %d%d=%.0f",s1,m,freq[s1][m][iage]);
                 }
                 /* printf(" %d%d=%.0f",s1,m,freq[s1][m][iage]); */
                 fprintf(ficlog," %d%d=%.0f",s1,m,freq[s1][m][iage]);
               }
               if(s1!=0 && m!=0)
                 fprintf(ficresphtmfr,"<td>%.0f</td> ",freq[s1][m][iage]);
             }
           } /* end loop s1 */
           posproptt=0.; 
           for(s1=1; s1 <=nlstate; s1++){
             posproptt += pospropt[s1];
           }
           fprintf(ficresphtmfr,"</tr>\n ");
           fprintf(ficresphtm,"</tr>\n");
           if((cptcoveff==0 && nj==1)|| nj==2 ) {
             if(iage <= iagemax)
               fprintf(ficresp,"\n");
           }
           if(first==1)
             printf("Others in log...\n");
           fprintf(ficlog,"\n");
         } /* end loop age iage */
         
         fprintf(ficresphtm,"<tr><th>Tot</th>");
         for(s1=1; s1 <=nlstate ; s1++){
           if(posproptt < 1.e-5){
             fprintf(ficresphtm,"<td>Nanq</td><td>%.0f</td><td>%.0f</td>",pospropt[s1],posproptt); 
           }else{
             fprintf(ficresphtm,"<td>%.5f</td><td>%.0f</td><td>%.0f</td>",pospropt[s1]/posproptt,pospropt[s1],posproptt);  
           }
         }
         fprintf(ficresphtm,"</tr>\n");
         fprintf(ficresphtm,"</table>\n");
         fprintf(ficresphtmfr,"</table>\n");
         if(posproptt < 1.e-5){
           fprintf(ficresphtm,"\n <p><b> This combination (%d) is not valid and no result will be produced</b></p>",j1);
           fprintf(ficresphtmfr,"\n <p><b> This combination (%d) is not valid and no result will be produced</b></p>",j1);
           fprintf(ficlog,"#  This combination (%d) is not valid and no result will be produced\n",j1);
           printf("#  This combination (%d) is not valid and no result will be produced\n",j1);
           invalidvarcomb[j1]=1;
         }else{
           fprintf(ficresphtm,"\n <p> This combination (%d) is valid and result will be produced.</p>",j1);
           invalidvarcomb[j1]=0;
         }
         fprintf(ficresphtmfr,"</table>\n");
         fprintf(ficlog,"\n");
         if(j!=0){
           printf("#Freqsummary: Starting values for combination j1=%d:\n", j1);
           for(i=1,s1=1; i <=nlstate; i++){
             for(k=1; k <=(nlstate+ndeath); k++){
               if (k != i) {
                 for(jj=1; jj <=ncovmodel; jj++){ /* For counting s1 */
                   if(jj==1){  /* Constant case (in fact cste + age) */
                     if(j1==1){ /* All dummy covariates to zero */
                       freq[i][k][iagemax+4]=freq[i][k][iagemax+3]; /* Stores case 0 0 0 */
                       freq[i][i][iagemax+4]=freq[i][i][iagemax+3]; /* Stores case 0 0 0 */
                       printf("%d%d ",i,k);
                       fprintf(ficlog,"%d%d ",i,k);
                       printf("%12.7f ln(%.0f/%.0f)= %f, OR=%f sd=%f \n",p[s1],freq[i][k][iagemax+3],freq[i][i][iagemax+3], log(freq[i][k][iagemax+3]/freq[i][i][iagemax+3]),freq[i][k][iagemax+3]/freq[i][i][iagemax+3], sqrt(1/freq[i][k][iagemax+3]+1/freq[i][i][iagemax+3]));
                       fprintf(ficlog,"%12.7f ln(%.0f/%.0f)= %12.7f \n",p[s1],freq[i][k][iagemax+3],freq[i][i][iagemax+3], log(freq[i][k][iagemax+3]/freq[i][i][iagemax+3]));
                       pstart[s1]= log(freq[i][k][iagemax+3]/freq[i][i][iagemax+3]);
                     }
                   }else if((j1==1) && (jj==2 || nagesqr==1)){ /* age or age*age parameter without covariate V4*age (to be done later) */
                     for(iage=iagemin; iage <= iagemax+3; iage++){
                       x[iage]= (double)iage;
                       y[iage]= log(freq[i][k][iage]/freq[i][i][iage]);
                       /* printf("i=%d, k=%d, s1=%d, j1=%d, jj=%d, y[%d]=%f\n",i,k,s1,j1,jj, iage, y[iage]); */
                     }
                     /* Some are not finite, but linreg will ignore these ages */
                     no=0;
                     linreg(iagemin,iagemax,&no,x,y,&a,&b,&r, &sa, &sb ); /* y= a+b*x with standard errors */
                     pstart[s1]=b;
                     pstart[s1-1]=a;
                   }else if( j1!=1 && (j1==2 || (log(j1-1.)/log(2.)-(int)(log(j1-1.)/log(2.))) <0.010) && ( TvarsDind[(int)(log(j1-1.)/log(2.))+1]+2+nagesqr == jj)  && Dummy[jj-2-nagesqr]==0){ /* We want only if the position, jj, in model corresponds to unique covariate equal to 1 in j1 combination */ 
                     printf("j1=%d, jj=%d, (int)(log(j1-1.)/log(2.))+1=%d, TvarsDind[(int)(log(j1-1.)/log(2.))+1]=%d\n",j1, jj,(int)(log(j1-1.)/log(2.))+1,TvarsDind[(int)(log(j1-1.)/log(2.))+1]);
                     printf("j1=%d, jj=%d, (log(j1-1.)/log(2.))+1=%f, TvarsDind[(int)(log(j1-1.)/log(2.))+1]=%d\n",j1, jj,(log(j1-1.)/log(2.))+1,TvarsDind[(int)(log(j1-1.)/log(2.))+1]);
                     pstart[s1]= log((freq[i][k][iagemax+3]/freq[i][i][iagemax+3])/(freq[i][k][iagemax+4]/freq[i][i][iagemax+4]));
                     printf("%d%d ",i,k);
                     fprintf(ficlog,"%d%d ",i,k);
                     printf("s1=%d,i=%d,k=%d,p[%d]=%12.7f ln((%.0f/%.0f)/(%.0f/%.0f))= %f, OR=%f sd=%f \n",s1,i,k,s1,p[s1],freq[i][k][iagemax+3],freq[i][i][iagemax+3],freq[i][k][iagemax+4],freq[i][i][iagemax+4], log((freq[i][k][iagemax+3]/freq[i][i][iagemax+3])/(freq[i][k][iagemax+4]/freq[i][i][iagemax+4])),(freq[i][k][iagemax+3]/freq[i][i][iagemax+3])/(freq[i][k][iagemax+4]/freq[i][i][iagemax+4]), sqrt(1/freq[i][k][iagemax+3]+1/freq[i][i][iagemax+3]+1/freq[i][k][iagemax+4]+1/freq[i][i][iagemax+4]));
                   }else{ /* Other cases, like quantitative fixed or varying covariates */
                     ;
                   }
                   /* printf("%12.7f )", param[i][jj][k]); */
                   /* fprintf(ficlog,"%12.7f )", param[i][jj][k]); */
                   s1++; 
                 } /* end jj */
               } /* end k!= i */
             } /* end k */
           } /* end i, s1 */
         } /* end j !=0 */
       } /* end selected combination of covariate j1 */
       if(j==0){ /* We can estimate starting values from the occurences in each case */
         printf("#Freqsummary: Starting values for the constants:\n");
         fprintf(ficlog,"\n");
         for(i=1,s1=1; i <=nlstate; i++){
           for(k=1; k <=(nlstate+ndeath); k++){
             if (k != i) {
               printf("%d%d ",i,k);
               fprintf(ficlog,"%d%d ",i,k);
               for(jj=1; jj <=ncovmodel; jj++){
                 pstart[s1]=p[s1]; /* Setting pstart to p values by default */
                 if(jj==1){ /* Age has to be done */
                   pstart[s1]= log(freq[i][k][iagemax+3]/freq[i][i][iagemax+3]);
                   printf("%12.7f ln(%.0f/%.0f)= %12.7f ",p[s1],freq[i][k][iagemax+3],freq[i][i][iagemax+3], log(freq[i][k][iagemax+3]/freq[i][i][iagemax+3]));
                   fprintf(ficlog,"%12.7f ln(%.0f/%.0f)= %12.7f ",p[s1],freq[i][k][iagemax+3],freq[i][i][iagemax+3], log(freq[i][k][iagemax+3]/freq[i][i][iagemax+3]));
                 }
                 /* printf("%12.7f )", param[i][jj][k]); */
                 /* fprintf(ficlog,"%12.7f )", param[i][jj][k]); */
                 s1++; 
               }
               printf("\n");
               fprintf(ficlog,"\n");
             }
           }
         } /* end of state i */
         printf("#Freqsummary\n");
         fprintf(ficlog,"\n");
         for(s1=-1; s1 <=nlstate+ndeath; s1++){
           for(s2=-1; s2 <=nlstate+ndeath; s2++){
             /* param[i]|j][k]= freq[s1][s2][iagemax+3] */
             printf(" %d%d=%.0f",s1,s2,freq[s1][s2][iagemax+3]);
             fprintf(ficlog," %d%d=%.0f",s1,s2,freq[s1][s2][iagemax+3]);
             /* if(freq[s1][s2][iage] !=0 ) { /\* minimizing output *\/ */
             /*   printf(" %d%d=%.0f",s1,s2,freq[s1][s2][iagemax+3]); */
             /*   fprintf(ficlog," %d%d=%.0f",s1,s2,freq[s1][s2][iagemax+3]); */
             /* } */
           }
         } /* end loop s1 */
         
         printf("\n");
         fprintf(ficlog,"\n");
       } /* end j=0 */
     } /* end j */
   
     if(mle == -2){  /* We want to use these values as starting values */
       for(i=1, jk=1; i <=nlstate; i++){
         for(j=1; j <=nlstate+ndeath; j++){
           if(j!=i){
             /*ca[0]= k+'a'-1;ca[1]='\0';*/
             printf("%1d%1d",i,j);
             fprintf(ficparo,"%1d%1d",i,j);
             for(k=1; k<=ncovmodel;k++){
               /*    printf(" %lf",param[i][j][k]); */
               /*    fprintf(ficparo," %lf",param[i][j][k]); */
               p[jk]=pstart[jk];
               printf(" %f ",pstart[jk]);
               fprintf(ficparo," %f ",pstart[jk]);
               jk++;
             }
             printf("\n");
             fprintf(ficparo,"\n");
           }
         }
       }
     } /* end mle=-2 */
     dateintmean=dateintsum/k2cpt; 
     date2dmy(dateintmean,&jintmean,&mintmean,&aintmean);
     
     fclose(ficresp);
     fclose(ficresphtm);
     fclose(ficresphtmfr);
     free_vector(idq,1,nqfveff);
     free_vector(meanq,1,nqfveff);
     free_vector(stdq,1,nqfveff);
     free_matrix(meanqt,1,lastpass,1,nqtveff);
     free_vector(x, iagemin-AGEMARGE, iagemax+4+AGEMARGE);
     free_vector(y, iagemin-AGEMARGE, iagemax+4+AGEMARGE);
     free_ma3x(freq,-5,nlstate+ndeath,-5,nlstate+ndeath, iagemin-AGEMARGE, iagemax+4+AGEMARGE);
     free_vector(pospropt,1,nlstate);
     free_vector(posprop,1,nlstate);
     free_matrix(prop,1,nlstate,iagemin-AGEMARGE, iagemax+4+AGEMARGE);
     free_vector(pp,1,nlstate);
     /* End of freqsummary */
   }
   
   /* Simple linear regression */
   int linreg(int ifi, int ila, int *no, const double x[], const double y[], double* a, double* b, double* r, double* sa, double * sb) {
   
     /* y=a+bx regression */
     double   sumx = 0.0;                        /* sum of x                      */
     double   sumx2 = 0.0;                       /* sum of x**2                   */
     double   sumxy = 0.0;                       /* sum of x * y                  */
     double   sumy = 0.0;                        /* sum of y                      */
     double   sumy2 = 0.0;                       /* sum of y**2                   */
     double   sume2 = 0.0;                       /* sum of square or residuals */
     double yhat;
     
     double denom=0;
     int i;
     int ne=*no;
     
     for ( i=ifi, ne=0;i<=ila;i++) {
       if(!isfinite(x[i]) || !isfinite(y[i])){
         /* printf(" x[%d]=%f, y[%d]=%f\n",i,x[i],i,y[i]); */
         continue;
       }
       ne=ne+1;
       sumx  += x[i];       
       sumx2 += x[i]*x[i];  
       sumxy += x[i] * y[i];
       sumy  += y[i];      
       sumy2 += y[i]*y[i]; 
       denom = (ne * sumx2 - sumx*sumx);
       /* printf("ne=%d, i=%d,x[%d]=%f, y[%d]=%f sumx=%f, sumx2=%f, sumxy=%f, sumy=%f, sumy2=%f, denom=%f\n",ne,i,i,x[i],i,y[i], sumx, sumx2,sumxy, sumy, sumy2,denom); */
     } 
     
     denom = (ne * sumx2 - sumx*sumx);
     if (denom == 0) {
       // vertical, slope m is infinity
       *b = INFINITY;
       *a = 0;
       if (r) *r = 0;
       return 1;
     }
     
     *b = (ne * sumxy  -  sumx * sumy) / denom;
     *a = (sumy * sumx2  -  sumx * sumxy) / denom;
     if (r!=NULL) {
       *r = (sumxy - sumx * sumy / ne) /          /* compute correlation coeff     */
         sqrt((sumx2 - sumx*sumx/ne) *
              (sumy2 - sumy*sumy/ne));
     }
     *no=ne;
     for ( i=ifi, ne=0;i<=ila;i++) {
       if(!isfinite(x[i]) || !isfinite(y[i])){
         /* printf(" x[%d]=%f, y[%d]=%f\n",i,x[i],i,y[i]); */
         continue;
       }
       ne=ne+1;
       yhat = y[i] - *a -*b* x[i];
       sume2  += yhat * yhat ;       
       
       denom = (ne * sumx2 - sumx*sumx);
       /* printf("ne=%d, i=%d,x[%d]=%f, y[%d]=%f sumx=%f, sumx2=%f, sumxy=%f, sumy=%f, sumy2=%f, denom=%f\n",ne,i,i,x[i],i,y[i], sumx, sumx2,sumxy, sumy, sumy2,denom); */
   }     } 
   free_vector(vv,1,n);  /* Doesn't work */    *sb = sqrt(sume2/(double)(ne-2)/(sumx2 - sumx * sumx /(double)ne));
 ;    *sa= *sb * sqrt(sumx2/ne);
 }     
     return 0; 
   }
   
   /************ Prevalence ********************/
   void prevalence(double ***probs, double agemin, double agemax, int **s, double **agev, int nlstate, int imx, int *Tvar, int **nbcode, int *ncodemax,double **mint,double **anint, double dateprev1,double dateprev2, int firstpass, int lastpass)
   {  
     /* Compute observed prevalence between dateprev1 and dateprev2 by counting the number of people
        in each health status at the date of interview (if between dateprev1 and dateprev2).
        We still use firstpass and lastpass as another selection.
     */
    
     int i, m, jk, j1, bool, z1,j, iv;
     int mi; /* Effective wave */
     int iage;
     double agebegin, ageend;
   
     double **prop;
     double posprop; 
     double  y2; /* in fractional years */
     int iagemin, iagemax;
     int first; /** to stop verbosity which is redirected to log file */
   
     iagemin= (int) agemin;
     iagemax= (int) agemax;
     /*pp=vector(1,nlstate);*/
     prop=matrix(1,nlstate,iagemin-AGEMARGE,iagemax+4+AGEMARGE); 
     /*  freq=ma3x(-1,nlstate+ndeath,-1,nlstate+ndeath,iagemin,iagemax+3);*/
     j1=0;
     
     /*j=cptcoveff;*/
     if (cptcovn<1) {j=1;ncodemax[1]=1;}
     
     first=0;
     for(j1=1; j1<= (int) pow(2,cptcoveff);j1++){ /* For each combination of simple dummy covariates */
       for (i=1; i<=nlstate; i++)  
         for(iage=iagemin-AGEMARGE; iage <= iagemax+4+AGEMARGE; iage++)
           prop[i][iage]=0.0;
       printf("Prevalence combination of varying and fixed dummies %d\n",j1);
       /* fprintf(ficlog," V%d=%d ",Tvaraff[j1],nbcode[Tvaraff[j1]][codtabm(k,j1)]); */
       fprintf(ficlog,"Prevalence combination of varying and fixed dummies %d\n",j1);
       
       for (i=1; i<=imx; i++) { /* Each individual */
         bool=1;
         /* for(m=firstpass; m<=lastpass; m++){/\* Other selection (we can limit to certain interviews*\/ */
         for(mi=1; mi<wav[i];mi++){ /* For this wave too look where individual can be counted V4=0 V3=0 */
           m=mw[mi][i];
           /* Tmodelind[z1]=k is the position of the varying covariate in the model, but which # within 1 to ntv? */
           /* Tvar[Tmodelind[z1]] is the n of Vn; n-ncovcol-nqv is the first time varying covariate or iv */
           for (z1=1; z1<=cptcoveff; z1++){
             if( Fixed[Tmodelind[z1]]==1){
               iv= Tvar[Tmodelind[z1]]-ncovcol-nqv;
               if (cotvar[m][iv][i]!= nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]) /* iv=1 to ntv, right modality */
                 bool=0;
             }else if( Fixed[Tmodelind[z1]]== 0)  /* fixed */
               if (covar[Tvaraff[z1]][i]!= nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]) {
                 bool=0;
               }
           }
           if(bool==1){ /* Otherwise we skip that wave/person */
             agebegin=agev[m][i]; /* Age at beginning of wave before transition*/
             /* ageend=agev[m][i]+(dh[m][i])*stepm/YEARM; /\* Age at end of wave and transition *\/ */
             if(m >=firstpass && m <=lastpass){
               y2=anint[m][i]+(mint[m][i]/12.); /* Fractional date in year */
               if ((y2>=dateprev1) && (y2<=dateprev2)) { /* Here is the main selection (fractional years) */
                 if(agev[m][i]==0) agev[m][i]=iagemax+1;
                 if(agev[m][i]==1) agev[m][i]=iagemax+2;
                 if((int)agev[m][i] <iagemin-AGEMARGE || (int)agev[m][i] >iagemax+4+AGEMARGE){
                   printf("Error on individual # %d agev[m][i]=%f <%d-%d or > %d+3+%d  m=%d; either change agemin or agemax or fix data\n",i, agev[m][i],iagemin,AGEMARGE, iagemax,AGEMARGE,m); 
                   exit(1);
                 }
                 if (s[m][i]>0 && s[m][i]<=nlstate) { 
                   /*if(i>4620) printf(" i=%d m=%d s[m][i]=%d (int)agev[m][i]=%d weight[i]=%f prop=%f\n",i,m,s[m][i],(int)agev[m][m],weight[i],prop[s[m][i]][(int)agev[m][i]]);*/
                   prop[s[m][i]][(int)agev[m][i]] += weight[i];/* At age of beginning of transition, where status is known */
                   prop[s[m][i]][iagemax+3] += weight[i]; 
                 } /* end valid statuses */ 
               } /* end selection of dates */
             } /* end selection of waves */
           } /* end bool */
         } /* end wave */
       } /* end individual */
       for(i=iagemin; i <= iagemax+3; i++){  
         for(jk=1,posprop=0; jk <=nlstate ; jk++) { 
           posprop += prop[jk][i]; 
         } 
         
         for(jk=1; jk <=nlstate ; jk++){       
           if( i <=  iagemax){ 
             if(posprop>=1.e-5){ 
               probs[i][jk][j1]= prop[jk][i]/posprop;
             } else{
               if(!first){
                 first=1;
                 printf("Warning Observed prevalence doesn't sum to 1 for state %d: probs[%d][%d][%d]=%lf because of lack of cases\nSee others in log file...\n",jk,i,jk, j1,probs[i][jk][j1]);
               }else{
                 fprintf(ficlog,"Warning Observed prevalence doesn't sum to 1 for state %d: probs[%d][%d][%d]=%lf because of lack of cases.\n",jk,i,jk, j1,probs[i][jk][j1]);
               }
             }
           } 
         }/* end jk */ 
       }/* end i */ 
        /*} *//* end i1 */
     } /* end j1 */
     
     /*  free_ma3x(freq,-1,nlstate+ndeath,-1,nlstate+ndeath, iagemin, iagemax+3);*/
     /*free_vector(pp,1,nlstate);*/
     free_matrix(prop,1,nlstate, iagemin-AGEMARGE,iagemax+4+AGEMARGE);
   }  /* End of prevalence */
   
   /************* Waves Concatenation ***************/
   
   void  concatwav(int wav[], int **dh, int **bh,  int **mw, int **s, double *agedc, double **agev, int  firstpass, int lastpass, int imx, int nlstate, int stepm)
   {
     /* Concatenates waves: wav[i] is the number of effective (useful waves in the sense that a non interview is useless) of individual i.
        Death is a valid wave (if date is known).
        mw[mi][i] is the mi (mi=1 to wav[i])  effective wave of individual i
        dh[m][i] or dh[mw[mi][i]][i] is the delay between two effective waves m=mw[mi][i]
        and mw[mi+1][i]. dh depends on stepm. s[m][i] exists for any wave from firstpass to lastpass
     */
   
     int i=0, mi=0, m=0, mli=0;
     /* int j, k=0,jk, ju, jl,jmin=1e+5, jmax=-1;
        double sum=0., jmean=0.;*/
     int first=0, firstwo=0, firsthree=0, firstfour=0, firstfiv=0;
     int j, k=0,jk, ju, jl;
     double sum=0.;
     first=0;
     firstwo=0;
     firsthree=0;
     firstfour=0;
     jmin=100000;
     jmax=-1;
     jmean=0.;
   
   /* Treating live states */
     for(i=1; i<=imx; i++){  /* For simple cases and if state is death */
       mi=0;  /* First valid wave */
       mli=0; /* Last valid wave */
       m=firstpass;  /* Loop on waves */
       while(s[m][i] <= nlstate){  /* a live state or unknown state  */
         if(m >firstpass && s[m][i]==s[m-1][i] && mint[m][i]==mint[m-1][i] && anint[m][i]==anint[m-1][i]){/* Two succesive identical information on wave m */
           mli=m-1;/* mw[++mi][i]=m-1; */
         }else if(s[m][i]>=1 || s[m][i]==-4 || s[m][i]==-5){ /* Since 0.98r4 if status=-2 vital status is really unknown, wave should be skipped */
           mw[++mi][i]=m; /* Valid wave: incrementing mi and updating mi; mw[mi] is the wave number of mi_th valid transition   */
           mli=m;
         } /* else might be a useless wave  -1 and mi is not incremented and mw[mi] not updated */
         if(m < lastpass){ /* m < lastpass, standard case */
           m++; /* mi gives the "effective" current wave, m the current wave, go to next wave by incrementing m */
         }
         else{ /* m = lastpass, eventual special issue with warning */
   #ifdef UNKNOWNSTATUSNOTCONTRIBUTING
           break;
   #else
           if(s[m][i]==-1 && (int) andc[i] == 9999 && (int)anint[m][i] != 9999){ /* no death date and known date of interview, case -2 (vital status unknown is warned later */
             if(firsthree == 0){
               printf("Information! Unknown status for individual %ld line=%d occurred at last wave %d at known date %d/%d. Please, check if your unknown date of death %d/%d means a live state %d at wave %d. This case(%d)/wave(%d) contributes to the likelihood as 1-p_{%d%d} .\nOthers in log file only\n",num[i],i,lastpass,(int)mint[m][i],(int)anint[m][i], (int) moisdc[i], (int) andc[i], s[m][i], m, i, m, s[m][i], nlstate+ndeath);
               firsthree=1;
             }else if(firsthree >=1 && firsthree < 10){
               fprintf(ficlog,"Information! Unknown status for individual %ld line=%d occurred at last wave %d at known date %d/%d. Please, check if your unknown date of death %d/%d means a live state %d at wave %d. This case(%d)/wave(%d) contributes to the likelihood as 1-p_{%d%d} .\n",num[i],i,lastpass,(int)mint[m][i],(int)anint[m][i], (int) moisdc[i], (int) andc[i], s[m][i], m, i, m, s[m][i], nlstate+ndeath);
               firsthree++;
             }else if(firsthree == 10){
               printf("Information, too many Information flags: no more reported to log either\n");
               fprintf(ficlog,"Information, too many Information flags: no more reported to log either\n");
               firsthree++;
             }else{
               firsthree++;
             }
             mw[++mi][i]=m; /* Valid transition with unknown status */
             mli=m;
           }
           if(s[m][i]==-2){ /* Vital status is really unknown */
             nbwarn++;
             if((int)anint[m][i] == 9999){  /*  Has the vital status really been verified?not a transition */
               printf("Warning! Vital status for individual %ld (line=%d) at last wave %d interviewed at date %d/%d is unknown %d. Please, check if the vital status and the date of death %d/%d are really unknown. This case (%d)/wave (%d) is skipped, no contribution to likelihood.\nOthers in log file only\n",num[i],i,lastpass,(int)mint[m][i],(int)anint[m][i], s[m][i], (int) moisdc[i], (int) andc[i], i, m);
               fprintf(ficlog,"Warning! Vital status for individual %ld (line=%d) at last wave %d interviewed at date %d/%d is unknown %d. Please, check if the vital status and the date of death %d/%d are really unknown. This case (%d)/wave (%d) is skipped, no contribution to likelihood.\n",num[i],i,lastpass,(int)mint[m][i],(int)anint[m][i], s[m][i], (int) moisdc[i], (int) andc[i], i, m);
             }
             break;
           }
           break;
   #endif
         }/* End m >= lastpass */
       }/* end while */
   
       /* mi is the last effective wave, m is lastpass, mw[j][i] gives the # of j-th effective wave for individual i */
       /* After last pass */
   /* Treating death states */
       if (s[m][i] > nlstate){  /* In a death state */
         /* if( mint[m][i]==mdc[m][i] && anint[m][i]==andc[m][i]){ /\* same date of death and date of interview *\/ */
         /* } */
         mi++;     /* Death is another wave */
         /* if(mi==0)  never been interviewed correctly before death */
         /* Only death is a correct wave */
         mw[mi][i]=m;
       } /* else not in a death state */
   #ifndef DISPATCHINGKNOWNDEATHAFTERLASTWAVE
       else if ((int) andc[i] != 9999) {  /* Date of death is known */
         if ((int)anint[m][i]!= 9999) { /* date of last interview is known */
           if((andc[i]+moisdc[i]/12.) <=(anint[m][i]+mint[m][i]/12.)){ /* month of death occured before last wave month and status should have been death instead of -1 */
             nbwarn++;
             if(firstfiv==0){
               printf("Warning! Death for individual %ld line=%d occurred at %d/%d before last wave %d, interviewed on %d/%d and should have been coded as death instead of '%d'. This case (%d)/wave (%d) is contributing to likelihood.\nOthers in log file only\n",num[i],i,(int) moisdc[i], (int) andc[i], lastpass,(int)mint[m][i],(int)anint[m][i], s[m][i], i,m );
               firstfiv=1;
             }else{
               fprintf(ficlog,"Warning! Death for individual %ld line=%d occurred at %d/%d before last wave %d, interviewed on %d/%d and should have been coded as death instead of '%d'. This case (%d)/wave (%d) is contributing to likelihood.\n",num[i],i,(int) moisdc[i], (int) andc[i], lastpass,(int)mint[m][i],(int)anint[m][i], s[m][i], i,m );
             }
               s[m][i]=nlstate+1; /* Fixing the status as death. Be careful if multiple death states */
           }else{ /* Month of Death occured afer last wave month, potential bias */
             nberr++;
             if(firstwo==0){
               printf("Error! Death for individual %ld line=%d occurred at %d/%d after last wave %d interviewed at %d/%d with status %d. Potential bias if other individuals are still alive on this date but ignored. This case (%d)/wave (%d) is skipped, no contribution to likelihood. Please add a new fictitious wave at the date of last vital status scan, with a dead status. See documentation\nOthers in log file only\n",num[i],i,(int) moisdc[i], (int) andc[i], lastpass,(int)mint[m][i],(int)anint[m][i], s[m][i], i,m );
               firstwo=1;
             }
             fprintf(ficlog,"Error! Death for individual %ld line=%d occurred at %d/%d after last wave %d interviewed at %d/%d with status %d. Potential bias if other individuals are still alive on this date but ignored. This case (%d)/wave (%d) is skipped, no contribution to likelihood. Please add a new fictitious wave at the date of last vital status scan, with a dead status. See documentation\n\n",num[i],i,(int) moisdc[i], (int) andc[i], lastpass,(int)mint[m][i],(int)anint[m][i], s[m][i], i,m );
           }
         }else{ /* if date of interview is unknown */
           /* death is known but not confirmed by death status at any wave */
           if(firstfour==0){
             printf("Error! Death for individual %ld line=%d  occurred %d/%d but not confirmed by any death status for any wave, including last wave %d at unknown date %d/%d with status %d. Potential bias if other individuals are still alive at this date but ignored. This case (%d)/wave (%d) is skipped, no contribution to likelihood.\nOthers in log file only\n",num[i],i,(int) moisdc[i], (int) andc[i], lastpass,(int)mint[m][i],(int)anint[m][i], s[m][i], i,m );
             firstfour=1;
           }
           fprintf(ficlog,"Error! Death for individual %ld line=%d  occurred %d/%d but not confirmed by any death status for any wave, including last wave %d at unknown date %d/%d  with status %d. Potential bias if other individuals are still alive at this date but ignored. This case (%d)/wave (%d) is skipped, no contribution to likelihood.\n",num[i],i,(int) moisdc[i], (int) andc[i], lastpass,(int)mint[m][i],(int)anint[m][i], s[m][i], i,m );
         }
       } /* end if date of death is known */
   #endif
       wav[i]=mi; /* mi should be the last effective wave (or mli),  */
       /* wav[i]=mw[mi][i];   */
       if(mi==0){
         nbwarn++;
         if(first==0){
           printf("Warning! No valid information for individual %ld line=%d (skipped) and may be others, see log file\n",num[i],i);
           first=1;
         }
         if(first==1){
           fprintf(ficlog,"Warning! No valid information for individual %ld line=%d (skipped)\n",num[i],i);
         }
       } /* end mi==0 */
     } /* End individuals */
     /* wav and mw are no more changed */
           
     printf("Information, you have to check %d informations which haven't been logged!\n",firsthree);
     fprintf(ficlog,"Information, you have to check %d informations which haven't been logged!\n",firsthree);
   
   
     for(i=1; i<=imx; i++){
       for(mi=1; mi<wav[i];mi++){
         if (stepm <=0)
           dh[mi][i]=1;
         else{
           if (s[mw[mi+1][i]][i] > nlstate) { /* A death, but what if date is unknown? */
             if (agedc[i] < 2*AGESUP) {
               j= rint(agedc[i]*12-agev[mw[mi][i]][i]*12); 
               if(j==0) j=1;  /* Survives at least one month after exam */
               else if(j<0){
                 nberr++;
                 printf("Error! Negative delay (%d to death) between waves %d and %d of individual %ld at line %d who is aged %.1f with statuses from %d to %d\n ",j,mw[mi][i],mw[mi+1][i],num[i], i,agev[mw[mi][i]][i],s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]);
                 j=1; /* Temporary Dangerous patch */
                 printf("   We assumed that the date of interview was correct (and not the date of death) and postponed the death %d month(s) (one stepm) after the interview. You MUST fix the contradiction between dates.\n",stepm);
                 fprintf(ficlog,"Error! Negative delay (%d to death) between waves %d and %d of individual %ld at line %d who is aged %.1f with statuses from %d to %d\n ",j,mw[mi][i],mw[mi+1][i],num[i], i,agev[mw[mi][i]][i],s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]);
                 fprintf(ficlog,"   We assumed that the date of interview was correct (and not the date of death) and postponed the death %d month(s) (one stepm) after the interview. You MUST fix the contradiction between dates.\n",stepm);
               }
               k=k+1;
               if (j >= jmax){
                 jmax=j;
                 ijmax=i;
               }
               if (j <= jmin){
                 jmin=j;
                 ijmin=i;
               }
               sum=sum+j;
               /*if (j<0) printf("j=%d num=%d \n",j,i);*/
               /*    printf("%d %d %d %d\n", s[mw[mi][i]][i] ,s[mw[mi+1][i]][i],j,i);*/
             }
           }
           else{
             j= rint( (agev[mw[mi+1][i]][i]*12 - agev[mw[mi][i]][i]*12));
   /*        if (j<0) printf("%d %lf %lf %d %d %d\n", i,agev[mw[mi+1][i]][i], agev[mw[mi][i]][i],j,s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]); */
                                           
             k=k+1;
             if (j >= jmax) {
               jmax=j;
               ijmax=i;
             }
             else if (j <= jmin){
               jmin=j;
               ijmin=i;
             }
             /*        if (j<10) printf("j=%d jmin=%d num=%d ",j,jmin,i); */
             /*printf("%d %lf %d %d %d\n", i,agev[mw[mi][i]][i],j,s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]);*/
             if(j<0){
               nberr++;
               printf("Error! Negative delay (%d) between waves %d and %d of individual %ld at line %d who is aged %.1f with statuses from %d to %d\n ",j,mw[mi][i],mw[mi+1][i],num[i], i,agev[mw[mi][i]][i],s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]);
               fprintf(ficlog,"Error! Negative delay (%d) between waves %d and %d of individual %ld at line %d who is aged %.1f with statuses from %d to %d\n ",j,mw[mi][i],mw[mi+1][i],num[i], i,agev[mw[mi][i]][i],s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]);
             }
             sum=sum+j;
           }
           jk= j/stepm;
           jl= j -jk*stepm;
           ju= j -(jk+1)*stepm;
           if(mle <=1){ /* only if we use a the linear-interpoloation pseudo-likelihood */
             if(jl==0){
               dh[mi][i]=jk;
               bh[mi][i]=0;
             }else{ /* We want a negative bias in order to only have interpolation ie
                     * to avoid the price of an extra matrix product in likelihood */
               dh[mi][i]=jk+1;
               bh[mi][i]=ju;
             }
           }else{
             if(jl <= -ju){
               dh[mi][i]=jk;
               bh[mi][i]=jl;       /* bias is positive if real duration
                                    * is higher than the multiple of stepm and negative otherwise.
                                    */
             }
             else{
               dh[mi][i]=jk+1;
               bh[mi][i]=ju;
             }
             if(dh[mi][i]==0){
               dh[mi][i]=1; /* At least one step */
               bh[mi][i]=ju; /* At least one step */
               /*  printf(" bh=%d ju=%d jl=%d dh=%d jk=%d stepm=%d %d\n",bh[mi][i],ju,jl,dh[mi][i],jk,stepm,i);*/
             }
           } /* end if mle */
         }
       } /* end wave */
     }
     jmean=sum/k;
     printf("Delay (in months) between two waves Min=%d (for indiviudal %ld) Max=%d (%ld) Mean=%f\n\n ",jmin, num[ijmin], jmax, num[ijmax], jmean);
     fprintf(ficlog,"Delay (in months) between two waves Min=%d (for indiviudal %d) Max=%d (%d) Mean=%f\n\n ",jmin, ijmin, jmax, ijmax, jmean);
   }
   
   /*********** Tricode ****************************/
    void tricode(int *cptcov, int *Tvar, int **nbcode, int imx, int *Ndum)
    {
      /**< Uses cptcovn+2*cptcovprod as the number of covariates */
      /*     Tvar[i]=atoi(stre);  find 'n' in Vn and stores in Tvar. If model=V2+V1 Tvar[1]=2 and Tvar[2]=1 
       * Boring subroutine which should only output nbcode[Tvar[j]][k]
       * Tvar[5] in V2+V1+V3*age+V2*V4 is 4 (V4) even it is a time varying or quantitative variable
       * nbcode[Tvar[5]][1]= nbcode[4][1]=0, nbcode[4][2]=1 (usually);
       */
   
      int ij=1, k=0, j=0, i=0, maxncov=NCOVMAX;
      int modmaxcovj=0; /* Modality max of covariates j */
      int cptcode=0; /* Modality max of covariates j */
      int modmincovj=0; /* Modality min of covariates j */
   
   
      /* cptcoveff=0;  */
      /* *cptcov=0; */
    
      for (k=1; k <= maxncov; k++) ncodemax[k]=0; /* Horrible constant again replaced by NCOVMAX */
      for (k=1; k <= maxncov; k++)
        for(j=1; j<=2; j++)
          nbcode[k][j]=0; /* Valgrind */
   
      /* Loop on covariates without age and products and no quantitative variable */
      for (k=1; k<=cptcovt; k++) { /* cptcovt: total number of covariates of the model (2) nbocc(+)+1 = 8 excepting constant and age and age*age */
        for (j=-1; (j < maxncov); j++) Ndum[j]=0;
        if(Dummy[k]==0 && Typevar[k] !=1){ /* Dummy covariate and not age product */ 
          switch(Fixed[k]) {
          case 0: /* Testing on fixed dummy covariate, simple or product of fixed */
            modmaxcovj=0;
            modmincovj=0;
            for (i=1; i<=imx; i++) { /* Loop on individuals: reads the data file to get the maximum value of the  modality of this covariate Vj*/
              ij=(int)(covar[Tvar[k]][i]);
              /* ij=0 or 1 or -1. Value of the covariate Tvar[j] for individual i
               * If product of Vn*Vm, still boolean *:
               * If it was coded 1, 2, 3, 4 should be splitted into 3 boolean variables
               * 1 => 0 0 0, 2 => 0 0 1, 3 => 0 1 1, 4=1 0 0   */
              /* Finds for covariate j, n=Tvar[j] of Vn . ij is the
                 modality of the nth covariate of individual i. */
              if (ij > modmaxcovj)
                modmaxcovj=ij; 
              else if (ij < modmincovj) 
                modmincovj=ij; 
              if (ij <0 || ij >1 ){
                printf("ERROR, IMaCh doesn't treat covariate with missing values V%d=-1, individual %d will be skipped.\n",Tvar[k],i);
                fprintf(ficlog,"ERROR, currently IMaCh doesn't treat covariate with missing values V%d=-1, individual %d will be skipped.\n",Tvar[k],i);
                fflush(ficlog);
                exit(1);
              }
              if ((ij < -1) || (ij > NCOVMAX)){
                printf( "Error: minimal is less than -1 or maximal is bigger than %d. Exiting. \n", NCOVMAX );
                exit(1);
              }else
                Ndum[ij]++; /*counts and stores the occurence of this modality 0, 1, -1*/
              /*  If coded 1, 2, 3 , counts the number of 1 Ndum[1], number of 2, Ndum[2], etc */
              /*printf("i=%d ij=%d Ndum[ij]=%d imx=%d",i,ij,Ndum[ij],imx);*/
              /* getting the maximum value of the modality of the covariate
                 (should be 0 or 1 now) Tvar[j]. If V=sex and male is coded 0 and
                 female ies 1, then modmaxcovj=1.
              */
            } /* end for loop on individuals i */
            printf(" Minimal and maximal values of %d th (fixed) covariate V%d: min=%d max=%d \n", k, Tvar[k], modmincovj, modmaxcovj);
            fprintf(ficlog," Minimal and maximal values of %d th (fixed) covariate V%d: min=%d max=%d \n", k, Tvar[k], modmincovj, modmaxcovj);
            cptcode=modmaxcovj;
            /* Ndum[0] = frequency of 0 for model-covariate j, Ndum[1] frequency of 1 etc. */
            /*for (i=0; i<=cptcode; i++) {*/
            for (j=modmincovj;  j<=modmaxcovj; j++) { /* j=-1 ? 0 and 1*//* For each value j of the modality of model-cov k */
              printf("Frequencies of (fixed) covariate %d ie V%d with value %d: %d\n", k, Tvar[k], j, Ndum[j]);
              fprintf(ficlog, "Frequencies of (fixed) covariate %d ie V%d with value %d: %d\n", k, Tvar[k], j, Ndum[j]);
              if( Ndum[j] != 0 ){ /* Counts if nobody answered modality j ie empty modality, we skip it and reorder */
                if( j != -1){
                  ncodemax[k]++;  /* ncodemax[k]= Number of modalities of the k th
                                     covariate for which somebody answered excluding 
                                     undefined. Usually 2: 0 and 1. */
                }
                ncodemaxwundef[k]++; /* ncodemax[j]= Number of modalities of the k th
                                        covariate for which somebody answered including 
                                        undefined. Usually 3: -1, 0 and 1. */
              }    /* In fact  ncodemax[k]=2 (dichotom. variables only) but it could be more for
                    * historical reasons: 3 if coded 1, 2, 3 and 4 and Ndum[2]=0 */
            } /* Ndum[-1] number of undefined modalities */
                           
            /* j is a covariate, n=Tvar[j] of Vn; Fills nbcode */
            /* For covariate j, modalities could be 1, 2, 3, 4, 5, 6, 7. */
            /* If Ndum[1]=0, Ndum[2]=0, Ndum[3]= 635, Ndum[4]=0, Ndum[5]=0, Ndum[6]=27, Ndum[7]=125; */
            /* modmincovj=3; modmaxcovj = 7; */
            /* There are only 3 modalities non empty 3, 6, 7 (or 2 if 27 is too few) : ncodemax[j]=3; */
            /* which will be coded 0, 1, 2 which in binary on 2=3-1 digits are 0=00 1=01, 2=10; */
            /*              defining two dummy variables: variables V1_1 and V1_2.*/
            /* nbcode[Tvar[j]][ij]=k; */
            /* nbcode[Tvar[j]][1]=0; */
            /* nbcode[Tvar[j]][2]=1; */
            /* nbcode[Tvar[j]][3]=2; */
            /* To be continued (not working yet). */
            ij=0; /* ij is similar to i but can jump over null modalities */
   
            /* for (i=modmincovj; i<=modmaxcovj; i++) { */ /* i= 1 to 2 for dichotomous, or from 1 to 3 or from -1 or 0 to 1 currently*/
            /* Skipping the case of missing values by reducing nbcode to 0 and 1 and not -1, 0, 1 */
            /* model=V1+V2+V3, if V2=-1, 0 or 1, then nbcode[2][1]=0 and nbcode[2][2]=1 instead of
             * nbcode[2][1]=-1, nbcode[2][2]=0 and nbcode[2][3]=1 */
            /*, could be restored in the future */
            for (i=0; i<=1; i++) { /* i= 1 to 2 for dichotomous, or from 1 to 3 or from -1 or 0 to 1 currently*/
              if (Ndum[i] == 0) { /* If nobody responded to this modality k */
                break;
              }
              ij++;
              nbcode[Tvar[k]][ij]=i;  /* stores the original value of modality i in an array nbcode, ij modality from 1 to last non-nul modality. nbcode[1][1]=0 nbcode[1][2]=1 . Could be -1*/
              cptcode = ij; /* New max modality for covar j */
            } /* end of loop on modality i=-1 to 1 or more */
            break;
          case 1: /* Testing on varying covariate, could be simple and
                   * should look at waves or product of fixed *
                   * varying. No time to test -1, assuming 0 and 1 only */
            ij=0;
            for(i=0; i<=1;i++){
              nbcode[Tvar[k]][++ij]=i;
            }
            break;
          default:
            break;
          } /* end switch */
        } /* end dummy test */
        if(Dummy[k]==1 && Typevar[k] !=1){ /* Quantitative covariate and not age product */ 
          for (i=1; i<=imx; i++) { /* Loop on individuals: reads the data file to get the maximum value of the  modality of this covariate Vj*/
            if(Tvar[k]<=0 || Tvar[k]>=NCOVMAX){
              printf("Error k=%d \n",k);
              exit(1);
            }
            if(isnan(covar[Tvar[k]][i])){
              printf("ERROR, IMaCh doesn't treat fixed quantitative covariate with missing values V%d=., individual %d will be skipped.\n",Tvar[k],i);
              fprintf(ficlog,"ERROR, currently IMaCh doesn't treat covariate with missing values V%d=., individual %d will be skipped.\n",Tvar[k],i);
              fflush(ficlog);
              exit(1);
            }
          }
        } /* end Quanti */
      } /* end of loop on model-covariate k. nbcode[Tvark][1]=-1, nbcode[Tvark][1]=0 and nbcode[Tvark][2]=1 sets the value of covariate k*/  
     
      for (k=-1; k< maxncov; k++) Ndum[k]=0; 
      /* Look at fixed dummy (single or product) covariates to check empty modalities */
      for (i=1; i<=ncovmodel-2-nagesqr; i++) { /* -2, cste and age and eventually age*age */ 
        /* Listing of all covariables in statement model to see if some covariates appear twice. For example, V1 appears twice in V1+V1*V2.*/ 
        ij=Tvar[i]; /* Tvar 5,4,3,6,5,7,1,4 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V4*age */ 
        Ndum[ij]++; /* Count the # of 1, 2 etc: {1,1,1,2,2,1,1} because V1 once, V2 once, two V4 and V5 in above */
        /* V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1,  {2, 1, 1, 1, 2, 1, 1, 0, 0} */
      } /* V4+V3+V5, Ndum[1]@5={0, 0, 1, 1, 1} */
     
      ij=0;
      /* for (i=0; i<=  maxncov-1; i++) { /\* modmaxcovj is unknown here. Only Ndum[2(V2),3(age*V3), 5(V3*V2) 6(V1*V4) *\/ */
      for (k=1; k<=  cptcovt; k++) { /* cptcovt: total number of covariates of the model (2) nbocc(+)+1 = 8 excepting constant and age and age*age */
        /* modmaxcovj is unknown here. Only Ndum[2(V2),3(age*V3), 5(V3*V2) 6(V1*V4) */
        /*printf("Ndum[%d]=%d\n",i, Ndum[i]);*/
        /* if((Ndum[i]!=0) && (i<=ncovcol)){  /\* Tvar[i] <= ncovmodel ? *\/ */
        if(Ndum[Tvar[k]]!=0 && Dummy[k] == 0 && Typevar[k]==0){  /* Only Dummy simple and non empty in the model */
          /* Typevar[k] =0 for simple covariate (dummy, quantitative, fixed or varying), 1 for age product, 2 for product */
          /* Dummy[k] 0=dummy (0 1), 1 quantitative (single or product without age), 2 dummy with age product, 3 quant with age product*/
          /* If product not in single variable we don't print results */
          /*printf("diff Ndum[%d]=%d\n",i, Ndum[i]);*/
          ++ij;/*    V5 + V4 + V3 + V4*V3 + V5*age + V2 +  V1*V2 + V1*age + V1, *//* V5 quanti, V2 quanti, V4, V3, V1 dummies */
          /* k=       1    2   3     4       5       6      7       8        9  */
          /* Tvar[k]= 5    4    3    6       5       2      7       1        1  */
          /* ij            1    2                                            3  */  
          /* Tvaraff[ij]=  4    3                                            1  */
          /* Tmodelind[ij]=2    3                                            9  */
          /* TmodelInvind[ij]=2 1                                            1  */
          Tvaraff[ij]=Tvar[k]; /* For printing combination *//* V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1, Tvar {5, 4, 3, 6, 5, 2, 7, 1, 1} Tvaraff={4, 3, 1} V4, V3, V1*/
          Tmodelind[ij]=k; /* Tmodelind: index in model of dummies Tmodelind[1]=2 V4: pos=2; V3: pos=3, V1=9 {2, 3, 9, ?, ?,} */
          TmodelInvind[ij]=Tvar[k]- ncovcol-nqv; /* Inverse TmodelInvind[2=V4]=2 second dummy varying cov (V4)4-1-1 {0, 2, 1, } TmodelInvind[3]=1 */
          if(Fixed[k]!=0)
            anyvaryingduminmodel=1;
          /* }else if((Ndum[i]!=0) && (i<=ncovcol+nqv)){ */
          /*   Tvaraff[++ij]=-10; /\* Dont'n know how to treat quantitative variables yet *\/ */
          /* }else if((Ndum[i]!=0) && (i<=ncovcol+nqv+ntv)){ */
          /*   Tvaraff[++ij]=i; /\*For printing (unclear) *\/ */
          /* }else if((Ndum[i]!=0) && (i<=ncovcol+nqv+ntv+nqtv)){ */
          /*   Tvaraff[++ij]=-20; /\* Dont'n know how to treat quantitative variables yet *\/ */
        } 
      } /* Tvaraff[1]@5 {3, 4, -20, 0, 0} Very strange */
      /* ij--; */
      /* cptcoveff=ij; /\*Number of total covariates*\/ */
      *cptcov=ij; /* cptcov= Number of total real effective simple dummies (fixed or time  arying) effective (used as cptcoveff in other functions)
                   * because they can be excluded from the model and real
                   * if in the model but excluded because missing values, but how to get k from ij?*/
      for(j=ij+1; j<= cptcovt; j++){
        Tvaraff[j]=0;
        Tmodelind[j]=0;
      }
      for(j=ntveff+1; j<= cptcovt; j++){
        TmodelInvind[j]=0;
      }
      /* To be sorted */
      ;
    }
   
   
   /*********** Health Expectancies ****************/
   
    void evsij(double ***eij, double x[], int nlstate, int stepm, int bage, int fage, double **oldm, double **savm, int cij, int estepm,char strstart[], int nres )
   
   {
     /* Health expectancies, no variances */
     /* cij is the combination in the list of combination of dummy covariates */
     /* strstart is a string of time at start of computing */
     int i, j, nhstepm, hstepm, h, nstepm;
     int nhstepma, nstepma; /* Decreasing with age */
     double age, agelim, hf;
     double ***p3mat;
     double eip;
   
     /* pstamp(ficreseij); */
     fprintf(ficreseij,"# (a) Life expectancies by health status at initial age and (b) health expectancies by health status at initial age\n");
     fprintf(ficreseij,"# Age");
     for(i=1; i<=nlstate;i++){
       for(j=1; j<=nlstate;j++){
         fprintf(ficreseij," e%1d%1d ",i,j);
       }
       fprintf(ficreseij," e%1d. ",i);
     }
     fprintf(ficreseij,"\n");
   
 void lubksb(double **a, int n, int *indx, double b[])     
 {     if(estepm < stepm){
   int i,ii=0,ip,j;       printf ("Problem %d lower than %d\n",estepm, stepm);
   double sum;     }
      else  hstepm=estepm;   
   for (i=1;i<=n;i++) {     /* We compute the life expectancy from trapezoids spaced every estepm months
     ip=indx[i];      * This is mainly to measure the difference between two models: for example
     sum=b[ip];      * if stepm=24 months pijx are given only every 2 years and by summing them
     b[ip]=b[i];      * we are calculating an estimate of the Life Expectancy assuming a linear 
     if (ii)      * progression in between and thus overestimating or underestimating according
       for (j=ii;j<=i-1;j++) sum -= a[i][j]*b[j];      * to the curvature of the survival function. If, for the same date, we 
     else if (sum) ii=i;      * estimate the model with stepm=1 month, we can keep estepm to 24 months
     b[i]=sum;      * to compare the new estimate of Life expectancy with the same linear 
   }      * hypothesis. A more precise result, taking into account a more precise
   for (i=n;i>=1;i--) {      * curvature will be obtained if estepm is as small as stepm. */
     sum=b[i];   
     for (j=i+1;j<=n;j++) sum -= a[i][j]*b[j];   
     b[i]=sum/a[i][i];   
   }   
 }   
   
 /************ Frequencies ********************/    /* For example we decided to compute the life expectancy with the smallest unit */
 void  freqsummary(char fileres[], int iagemin, int iagemax, int **s, double **agev, int nlstate, int imx, int *Tvaraff, int **nbcode, int *ncodemax,double **mint,double **anint)    /* hstepm beeing the number of stepms, if hstepm=1 the length of hstepm is stepm. 
 {  /* Some frequencies */       nhstepm is the number of hstepm from age to agelim 
        nstepm is the number of stepm from age to agelin. 
        Look at hpijx to understand the reason which relies in memory size consideration
        and note for a fixed period like estepm months */
     /* We decided (b) to get a life expectancy respecting the most precise curvature of the
        survival function given by stepm (the optimization length). Unfortunately it
        means that if the survival funtion is printed only each two years of age and if
        you sum them up and add 1 year (area under the trapezoids) you won't get the same 
        results. So we changed our mind and took the option of the best precision.
     */
     hstepm=hstepm/stepm; /* Typically in stepm units, if stepm=6 & estepm=24 , = 24/6 months = 4 */ 
   
     agelim=AGESUP;
     /* If stepm=6 months */
       /* Computed by stepm unit matrices, product of hstepm matrices, stored
          in an array of nhstepm length: nhstepm=10, hstepm=4, stepm=6 months */
       
   /* nhstepm age range expressed in number of stepm */
     nstepm=(int) rint((agelim-bage)*YEARM/stepm); /* Biggest nstepm */
     /* Typically if 20 years nstepm = 20*12/6=40 stepm */ 
     /* if (stepm >= YEARM) hstepm=1;*/
     nhstepm = nstepm/hstepm;/* Expressed in hstepm, typically nhstepm=40/4=10 */
     p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
   
     for (age=bage; age<=fage; age ++){ 
       nstepma=(int) rint((agelim-bage)*YEARM/stepm); /* Biggest nstepm */
       /* Typically if 20 years nstepm = 20*12/6=40 stepm */ 
       /* if (stepm >= YEARM) hstepm=1;*/
       nhstepma = nstepma/hstepm;/* Expressed in hstepm, typically nhstepma=40/4=10 */
   
       /* If stepm=6 months */
       /* Computed by stepm unit matrices, product of hstepma matrices, stored
          in an array of nhstepma length: nhstepma=10, hstepm=4, stepm=6 months */
       /* printf("HELLO evsij Entering hpxij age=%d cij=%d hstepm=%d x[1]=%f nres=%d\n",(int) age, cij, hstepm, x[1], nres); */
       hpxij(p3mat,nhstepma,age,hstepm,x,nlstate,stepm,oldm, savm, cij, nres);  
       
       hf=hstepm*stepm/YEARM;  /* Duration of hstepm expressed in year unit. */
       
       printf("%d|",(int)age);fflush(stdout);
       fprintf(ficlog,"%d|",(int)age);fflush(ficlog);
       
       /* Computing expectancies */
       for(i=1; i<=nlstate;i++)
         for(j=1; j<=nlstate;j++)
           for (h=0, eij[i][j][(int)age]=0; h<=nhstepm-1; h++){
             eij[i][j][(int)age] += (p3mat[i][j][h]+p3mat[i][j][h+1])/2.0*hf;
             
             /* if((int)age==70)printf("i=%2d,j=%2d,h=%2d,age=%3d,%9.4f,%9.4f,%9.4f\n",i,j,h,(int)age,p3mat[i][j][h],hf,eij[i][j][(int)age]);*/
   
           }
   
       fprintf(ficreseij,"%3.0f",age );
       for(i=1; i<=nlstate;i++){
         eip=0;
         for(j=1; j<=nlstate;j++){
           eip +=eij[i][j][(int)age];
           fprintf(ficreseij,"%9.4f", eij[i][j][(int)age] );
         }
         fprintf(ficreseij,"%9.4f", eip );
       }
       fprintf(ficreseij,"\n");
       
     }
     free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
     printf("\n");
     fprintf(ficlog,"\n");
       
   int i, m, jk, k1,i1, j1, bool, z1,z2,j;  }
   int first;  
   double ***freq; /* Frequencies */   void cvevsij(double ***eij, double x[], int nlstate, int stepm, int bage, int fage, double **oldm, double **savm, int cij, int estepm,double delti[],double **matcov,char strstart[], int nres )
   double *pp, **prop;  
   double pos,posprop, k2, dateintsum=0,k2cpt=0;  {
   FILE *ficresp;    /* Covariances of health expectancies eij and of total life expectancies according
   char fileresp[FILENAMELENGTH];       to initial status i, ei. .
     */
     int i, j, nhstepm, hstepm, h, nstepm, k, cptj, cptj2, i2, j2, ij, ji;
     int nhstepma, nstepma; /* Decreasing with age */
     double age, agelim, hf;
     double ***p3matp, ***p3matm, ***varhe;
     double **dnewm,**doldm;
     double *xp, *xm;
     double **gp, **gm;
     double ***gradg, ***trgradg;
     int theta;
   
     double eip, vip;
   
     varhe=ma3x(1,nlstate*nlstate,1,nlstate*nlstate,(int) bage, (int) fage);
     xp=vector(1,npar);
     xm=vector(1,npar);
     dnewm=matrix(1,nlstate*nlstate,1,npar);
     doldm=matrix(1,nlstate*nlstate,1,nlstate*nlstate);
       
   pp=vector(1,nlstate);    pstamp(ficresstdeij);
   prop=matrix(1,nlstate,iagemin,iagemax+3);    fprintf(ficresstdeij,"# Health expectancies with standard errors\n");
   strcpy(fileresp,"p");    fprintf(ficresstdeij,"# Age");
   strcat(fileresp,fileres);    for(i=1; i<=nlstate;i++){
   if((ficresp=fopen(fileresp,"w"))==NULL) {      for(j=1; j<=nlstate;j++)
     printf("Problem with prevalence resultfile: %s\n", fileresp);        fprintf(ficresstdeij," e%1d%1d (SE)",i,j);
     fprintf(ficlog,"Problem with prevalence resultfile: %s\n", fileresp);      fprintf(ficresstdeij," e%1d. ",i);
     exit(0);  
   }    }
   freq= ma3x(-1,nlstate+ndeath,-1,nlstate+ndeath,iagemin,iagemax+3);    fprintf(ficresstdeij,"\n");
   j1=0;  
     pstamp(ficrescveij);
     fprintf(ficrescveij,"# Subdiagonal matrix of covariances of health expectancies by age: cov(eij,ekl)\n");
     fprintf(ficrescveij,"# Age");
     for(i=1; i<=nlstate;i++)
       for(j=1; j<=nlstate;j++){
         cptj= (j-1)*nlstate+i;
         for(i2=1; i2<=nlstate;i2++)
           for(j2=1; j2<=nlstate;j2++){
             cptj2= (j2-1)*nlstate+i2;
             if(cptj2 <= cptj)
               fprintf(ficrescveij,"  %1d%1d,%1d%1d",i,j,i2,j2);
           }
       }
     fprintf(ficrescveij,"\n");
       
   j=cptcoveff;    if(estepm < stepm){
   if (cptcovn<1) {j=1;ncodemax[1]=1;}      printf ("Problem %d lower than %d\n",estepm, stepm);
     }
     else  hstepm=estepm;   
     /* We compute the life expectancy from trapezoids spaced every estepm months
      * This is mainly to measure the difference between two models: for example
      * if stepm=24 months pijx are given only every 2 years and by summing them
      * we are calculating an estimate of the Life Expectancy assuming a linear 
      * progression in between and thus overestimating or underestimating according
      * to the curvature of the survival function. If, for the same date, we 
      * estimate the model with stepm=1 month, we can keep estepm to 24 months
      * to compare the new estimate of Life expectancy with the same linear 
      * hypothesis. A more precise result, taking into account a more precise
      * curvature will be obtained if estepm is as small as stepm. */
   
     /* For example we decided to compute the life expectancy with the smallest unit */
     /* hstepm beeing the number of stepms, if hstepm=1 the length of hstepm is stepm. 
        nhstepm is the number of hstepm from age to agelim 
        nstepm is the number of stepm from age to agelin. 
        Look at hpijx to understand the reason of that which relies in memory size
        and note for a fixed period like estepm months */
     /* We decided (b) to get a life expectancy respecting the most precise curvature of the
        survival function given by stepm (the optimization length). Unfortunately it
        means that if the survival funtion is printed only each two years of age and if
        you sum them up and add 1 year (area under the trapezoids) you won't get the same 
        results. So we changed our mind and took the option of the best precision.
     */
     hstepm=hstepm/stepm; /* Typically in stepm units, if stepm=6 & estepm=24 , = 24/6 months = 4 */ 
   
     /* If stepm=6 months */
     /* nhstepm age range expressed in number of stepm */
     agelim=AGESUP;
     nstepm=(int) rint((agelim-bage)*YEARM/stepm); 
     /* Typically if 20 years nstepm = 20*12/6=40 stepm */ 
     /* if (stepm >= YEARM) hstepm=1;*/
     nhstepm = nstepm/hstepm;/* Expressed in hstepm, typically nhstepm=40/4=10 */
     
     p3matp=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
     p3matm=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
     gradg=ma3x(0,nhstepm,1,npar,1,nlstate*nlstate);
     trgradg =ma3x(0,nhstepm,1,nlstate*nlstate,1,npar);
     gp=matrix(0,nhstepm,1,nlstate*nlstate);
     gm=matrix(0,nhstepm,1,nlstate*nlstate);
   
     for (age=bage; age<=fage; age ++){ 
       nstepma=(int) rint((agelim-bage)*YEARM/stepm); /* Biggest nstepm */
       /* Typically if 20 years nstepm = 20*12/6=40 stepm */ 
       /* if (stepm >= YEARM) hstepm=1;*/
       nhstepma = nstepma/hstepm;/* Expressed in hstepm, typically nhstepma=40/4=10 */
                   
       /* If stepm=6 months */
       /* Computed by stepm unit matrices, product of hstepma matrices, stored
          in an array of nhstepma length: nhstepma=10, hstepm=4, stepm=6 months */
       
       hf=hstepm*stepm/YEARM;  /* Duration of hstepm expressed in year unit. */
                   
       /* Computing  Variances of health expectancies */
       /* Gradient is computed with plus gp and minus gm. Code is duplicated in order to
          decrease memory allocation */
       for(theta=1; theta <=npar; theta++){
         for(i=1; i<=npar; i++){ 
           xp[i] = x[i] + (i==theta ?delti[theta]:0);
           xm[i] = x[i] - (i==theta ?delti[theta]:0);
         }
         hpxij(p3matp,nhstepm,age,hstepm,xp,nlstate,stepm,oldm,savm, cij, nres);  
         hpxij(p3matm,nhstepm,age,hstepm,xm,nlstate,stepm,oldm,savm, cij, nres);  
                           
         for(j=1; j<= nlstate; j++){
           for(i=1; i<=nlstate; i++){
             for(h=0; h<=nhstepm-1; h++){
               gp[h][(j-1)*nlstate + i] = (p3matp[i][j][h]+p3matp[i][j][h+1])/2.;
               gm[h][(j-1)*nlstate + i] = (p3matm[i][j][h]+p3matm[i][j][h+1])/2.;
             }
           }
         }
                           
         for(ij=1; ij<= nlstate*nlstate; ij++)
           for(h=0; h<=nhstepm-1; h++){
             gradg[h][theta][ij]= (gp[h][ij]-gm[h][ij])/2./delti[theta];
           }
       }/* End theta */
       
       
       for(h=0; h<=nhstepm-1; h++)
         for(j=1; j<=nlstate*nlstate;j++)
           for(theta=1; theta <=npar; theta++)
             trgradg[h][j][theta]=gradg[h][theta][j];
       
                   
       for(ij=1;ij<=nlstate*nlstate;ij++)
         for(ji=1;ji<=nlstate*nlstate;ji++)
           varhe[ij][ji][(int)age] =0.;
                   
       printf("%d|",(int)age);fflush(stdout);
       fprintf(ficlog,"%d|",(int)age);fflush(ficlog);
       for(h=0;h<=nhstepm-1;h++){
         for(k=0;k<=nhstepm-1;k++){
           matprod2(dnewm,trgradg[h],1,nlstate*nlstate,1,npar,1,npar,matcov);
           matprod2(doldm,dnewm,1,nlstate*nlstate,1,npar,1,nlstate*nlstate,gradg[k]);
           for(ij=1;ij<=nlstate*nlstate;ij++)
             for(ji=1;ji<=nlstate*nlstate;ji++)
               varhe[ij][ji][(int)age] += doldm[ij][ji]*hf*hf;
         }
       }
       /* if((int)age ==50){ */
       /*   printf(" age=%d cij=%d nres=%d varhe[%d][%d]=%f ",(int)age, cij, nres, 1,2,varhe[1][2]); */
       /* } */
       /* Computing expectancies */
       hpxij(p3matm,nhstepm,age,hstepm,x,nlstate,stepm,oldm, savm, cij,nres);  
       for(i=1; i<=nlstate;i++)
         for(j=1; j<=nlstate;j++)
           for (h=0, eij[i][j][(int)age]=0; h<=nhstepm-1; h++){
             eij[i][j][(int)age] += (p3matm[i][j][h]+p3matm[i][j][h+1])/2.0*hf;
                                           
             /* if((int)age==70)printf("i=%2d,j=%2d,h=%2d,age=%3d,%9.4f,%9.4f,%9.4f\n",i,j,h,(int)age,p3mat[i][j][h],hf,eij[i][j][(int)age]);*/
                                           
           }
   
       /* Standard deviation of expectancies ij */         
       fprintf(ficresstdeij,"%3.0f",age );
       for(i=1; i<=nlstate;i++){
         eip=0.;
         vip=0.;
         for(j=1; j<=nlstate;j++){
           eip += eij[i][j][(int)age];
           for(k=1; k<=nlstate;k++) /* Sum on j and k of cov(eij,eik) */
             vip += varhe[(j-1)*nlstate+i][(k-1)*nlstate+i][(int)age];
           fprintf(ficresstdeij," %9.4f (%.4f)", eij[i][j][(int)age], sqrt(varhe[(j-1)*nlstate+i][(j-1)*nlstate+i][(int)age]) );
         }
         fprintf(ficresstdeij," %9.4f (%.4f)", eip, sqrt(vip));
       }
       fprintf(ficresstdeij,"\n");
                   
       /* Variance of expectancies ij */           
       fprintf(ficrescveij,"%3.0f",age );
       for(i=1; i<=nlstate;i++)
         for(j=1; j<=nlstate;j++){
           cptj= (j-1)*nlstate+i;
           for(i2=1; i2<=nlstate;i2++)
             for(j2=1; j2<=nlstate;j2++){
               cptj2= (j2-1)*nlstate+i2;
               if(cptj2 <= cptj)
                 fprintf(ficrescveij," %.4f", varhe[cptj][cptj2][(int)age]);
             }
         }
       fprintf(ficrescveij,"\n");
                   
     }
     free_matrix(gm,0,nhstepm,1,nlstate*nlstate);
     free_matrix(gp,0,nhstepm,1,nlstate*nlstate);
     free_ma3x(gradg,0,nhstepm,1,npar,1,nlstate*nlstate);
     free_ma3x(trgradg,0,nhstepm,1,nlstate*nlstate,1,npar);
     free_ma3x(p3matm,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
     free_ma3x(p3matp,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
     printf("\n");
     fprintf(ficlog,"\n");
           
     free_vector(xm,1,npar);
     free_vector(xp,1,npar);
     free_matrix(dnewm,1,nlstate*nlstate,1,npar);
     free_matrix(doldm,1,nlstate*nlstate,1,nlstate*nlstate);
     free_ma3x(varhe,1,nlstate*nlstate,1,nlstate*nlstate,(int) bage, (int)fage);
   }
    
   /************ Variance ******************/
    void varevsij(char optionfilefiname[], double ***vareij, double **matcov, double x[], double delti[], int nlstate, int stepm, double bage, double fage, double **oldm, double **savm, double **prlim, double ftolpl, int *ncvyearp, int ij, int estepm, int cptcov, int cptcod, int popbased, int mobilav, char strstart[], int nres)
    {
      /** Variance of health expectancies 
       *  double **prevalim(double **prlim, int nlstate, double *xp, double age, double **oldm, double ** savm,double ftolpl);
       * double **newm;
       * int movingaverage(double ***probs, double bage,double fage, double ***mobaverage, int mobilav) 
       */
     
      /* int movingaverage(); */
      double **dnewm,**doldm;
      double **dnewmp,**doldmp;
      int i, j, nhstepm, hstepm, h, nstepm ;
      int first=0;
      int k;
      double *xp;
      double **gp, **gm;  /**< for var eij */
      double ***gradg, ***trgradg; /**< for var eij */
      double **gradgp, **trgradgp; /**< for var p point j */
      double *gpp, *gmp; /**< for var p point j */
      double **varppt; /**< for var p point j nlstate to nlstate+ndeath */
      double ***p3mat;
      double age,agelim, hf;
      /* double ***mobaverage; */
      int theta;
      char digit[4];
      char digitp[25];
   
      char fileresprobmorprev[FILENAMELENGTH];
   
      if(popbased==1){
        if(mobilav!=0)
          strcpy(digitp,"-POPULBASED-MOBILAV_");
        else strcpy(digitp,"-POPULBASED-NOMOBIL_");
      }
      else 
        strcpy(digitp,"-STABLBASED_");
   
      /* if (mobilav!=0) { */
      /*   mobaverage= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX); */
      /*   if (movingaverage(probs, bage, fage, mobaverage,mobilav)!=0){ */
      /*     fprintf(ficlog," Error in movingaverage mobilav=%d\n",mobilav); */
      /*     printf(" Error in movingaverage mobilav=%d\n",mobilav); */
      /*   } */
      /* } */
   
      strcpy(fileresprobmorprev,"PRMORPREV-"); 
      sprintf(digit,"%-d",ij);
      /*printf("DIGIT=%s, ij=%d ijr=%-d|\n",digit, ij,ij);*/
      strcat(fileresprobmorprev,digit); /* Tvar to be done */
      strcat(fileresprobmorprev,digitp); /* Popbased or not, mobilav or not */
      strcat(fileresprobmorprev,fileresu);
      if((ficresprobmorprev=fopen(fileresprobmorprev,"w"))==NULL) {
        printf("Problem with resultfile: %s\n", fileresprobmorprev);
        fprintf(ficlog,"Problem with resultfile: %s\n", fileresprobmorprev);
      }
      printf("Computing total mortality p.j=w1*p1j+w2*p2j+..: result on file '%s' \n",fileresprobmorprev);
      fprintf(ficlog,"Computing total mortality p.j=w1*p1j+w2*p2j+..: result on file '%s' \n",fileresprobmorprev);
      pstamp(ficresprobmorprev);
      fprintf(ficresprobmorprev,"# probabilities of dying before estepm=%d months for people of exact age and weighted probabilities w1*p1j+w2*p2j+... stand dev in()\n",estepm);
      fprintf(ficresprobmorprev,"# Selected quantitative variables and dummies");
      for (j=1; j<= nsq; j++){ /* For each selected (single) quantitative value */ /* To be done*/
        fprintf(ficresprobmorprev," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
      }
      for(j=1;j<=cptcoveff;j++) 
        fprintf(ficresprobmorprev," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(ij,TnsdVar[Tvaraff[j]])]);
      fprintf(ficresprobmorprev,"\n");
   
      fprintf(ficresprobmorprev,"# Age cov=%-d",ij);
      for(j=nlstate+1; j<=(nlstate+ndeath);j++){
        fprintf(ficresprobmorprev," p.%-d SE",j);
        for(i=1; i<=nlstate;i++)
          fprintf(ficresprobmorprev," w%1d p%-d%-d",i,i,j);
      }  
      fprintf(ficresprobmorprev,"\n");
     
      fprintf(ficgp,"\n# Routine varevsij");
      fprintf(ficgp,"\nunset title \n");
      /* fprintf(fichtm, "#Local time at start: %s", strstart);*/
      fprintf(fichtm,"\n<li><h4> Computing probabilities of dying over estepm months as a weighted average (i.e global mortality independent of initial healh state)</h4></li>\n");
      fprintf(fichtm,"\n<br>%s  <br>\n",digitp);
   
      varppt = matrix(nlstate+1,nlstate+ndeath,nlstate+1,nlstate+ndeath);
      pstamp(ficresvij);
      fprintf(ficresvij,"# Variance and covariance of health expectancies e.j \n#  (weighted average of eij where weights are ");
      if(popbased==1)
        fprintf(ficresvij,"the age specific prevalence observed (cross-sectionally) in the population i.e cross-sectionally\n in each health state (popbased=1) (mobilav=%d\n",mobilav);
      else
        fprintf(ficresvij,"the age specific period (stable) prevalences in each health state \n");
      fprintf(ficresvij,"# Age");
      for(i=1; i<=nlstate;i++)
        for(j=1; j<=nlstate;j++)
          fprintf(ficresvij," Cov(e.%1d, e.%1d)",i,j);
      fprintf(ficresvij,"\n");
   
      xp=vector(1,npar);
      dnewm=matrix(1,nlstate,1,npar);
      doldm=matrix(1,nlstate,1,nlstate);
      dnewmp= matrix(nlstate+1,nlstate+ndeath,1,npar);
      doldmp= matrix(nlstate+1,nlstate+ndeath,nlstate+1,nlstate+ndeath);
   
      gradgp=matrix(1,npar,nlstate+1,nlstate+ndeath);
      gpp=vector(nlstate+1,nlstate+ndeath);
      gmp=vector(nlstate+1,nlstate+ndeath);
      trgradgp =matrix(nlstate+1,nlstate+ndeath,1,npar); /* mu or p point j*/
     
      if(estepm < stepm){
        printf ("Problem %d lower than %d\n",estepm, stepm);
      }
      else  hstepm=estepm;   
      /* For example we decided to compute the life expectancy with the smallest unit */
      /* hstepm beeing the number of stepms, if hstepm=1 the length of hstepm is stepm. 
         nhstepm is the number of hstepm from age to agelim 
         nstepm is the number of stepm from age to agelim. 
         Look at function hpijx to understand why because of memory size limitations, 
         we decided (b) to get a life expectancy respecting the most precise curvature of the
         survival function given by stepm (the optimization length). Unfortunately it
         means that if the survival funtion is printed every two years of age and if
         you sum them up and add 1 year (area under the trapezoids) you won't get the same 
         results. So we changed our mind and took the option of the best precision.
      */
      hstepm=hstepm/stepm; /* Typically in stepm units, if stepm=6 & estepm=24 , = 24/6 months = 4 */ 
      agelim = AGESUP;
      for (age=bage; age<=fage; age ++){ /* If stepm=6 months */
        nstepm=(int) rint((agelim-age)*YEARM/stepm); /* Typically 20 years = 20*12/6=40 */ 
        nhstepm = nstepm/hstepm;/* Expressed in hstepm, typically nhstepm=40/4=10 */
        p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
        gradg=ma3x(0,nhstepm,1,npar,1,nlstate);
        gp=matrix(0,nhstepm,1,nlstate);
        gm=matrix(0,nhstepm,1,nlstate);
                   
                   
        for(theta=1; theta <=npar; theta++){
          for(i=1; i<=npar; i++){ /* Computes gradient x + delta*/
            xp[i] = x[i] + (i==theta ?delti[theta]:0);
          }
          /**< Computes the prevalence limit with parameter theta shifted of delta up to ftolpl precision and 
           * returns into prlim .
           */
          prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ncvyearp,ij, nres);
   
          /* If popbased = 1 we use crossection prevalences. Previous step is useless but prlim is created */
          if (popbased==1) {
            if(mobilav ==0){
              for(i=1; i<=nlstate;i++)
                prlim[i][i]=probs[(int)age][i][ij];
            }else{ /* mobilav */ 
              for(i=1; i<=nlstate;i++)
                prlim[i][i]=mobaverage[(int)age][i][ij];
            }
          }
          /**< Computes the shifted transition matrix \f$ {}{h}_p^{ij}x\f$ at horizon h.
           */                      
          hpxij(p3mat,nhstepm,age,hstepm,xp,nlstate,stepm,oldm,savm, ij,nres);  /* Returns p3mat[i][j][h] for h=0 to nhstepm */
          /**< And for each alive state j, sums over i \f$ w^i_x {}{h}_p^{ij}x\f$, which are the probability
           * at horizon h in state j including mortality.
           */
          for(j=1; j<= nlstate; j++){
            for(h=0; h<=nhstepm; h++){
              for(i=1, gp[h][j]=0.;i<=nlstate;i++)
                gp[h][j] += prlim[i][i]*p3mat[i][j][h];
            }
          }
          /* Next for computing shifted+ probability of death (h=1 means
             computed over hstepm matrices product = hstepm*stepm months) 
             as a weighted average of prlim(i) * p(i,j) p.3=w1*p13 + w2*p23 .
          */
          for(j=nlstate+1;j<=nlstate+ndeath;j++){
            for(i=1,gpp[j]=0.; i<= nlstate; i++)
              gpp[j] += prlim[i][i]*p3mat[i][j][1];
          }
          
          /* Again with minus shift */
                           
          for(i=1; i<=npar; i++) /* Computes gradient x - delta */
            xp[i] = x[i] - (i==theta ?delti[theta]:0);
   
          prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ncvyearp, ij, nres);
                           
          if (popbased==1) {
            if(mobilav ==0){
              for(i=1; i<=nlstate;i++)
                prlim[i][i]=probs[(int)age][i][ij];
            }else{ /* mobilav */ 
              for(i=1; i<=nlstate;i++)
                prlim[i][i]=mobaverage[(int)age][i][ij];
            }
          }
                           
          hpxij(p3mat,nhstepm,age,hstepm,xp,nlstate,stepm,oldm,savm, ij,nres);  
                           
          for(j=1; j<= nlstate; j++){  /* Sum of wi * eij = e.j */
            for(h=0; h<=nhstepm; h++){
              for(i=1, gm[h][j]=0.;i<=nlstate;i++)
                gm[h][j] += prlim[i][i]*p3mat[i][j][h];
            }
          }
          /* This for computing probability of death (h=1 means
             computed over hstepm matrices product = hstepm*stepm months) 
             as a weighted average of prlim.
          */
          for(j=nlstate+1;j<=nlstate+ndeath;j++){
            for(i=1,gmp[j]=0.; i<= nlstate; i++)
              gmp[j] += prlim[i][i]*p3mat[i][j][1];
          }    
          /* end shifting computations */
   
          /**< Computing gradient matrix at horizon h 
           */
          for(j=1; j<= nlstate; j++) /* vareij */
            for(h=0; h<=nhstepm; h++){
              gradg[h][theta][j]= (gp[h][j]-gm[h][j])/2./delti[theta];
            }
          /**< Gradient of overall mortality p.3 (or p.j) 
           */
          for(j=nlstate+1; j<= nlstate+ndeath; j++){ /* var mu mortality from j */
            gradgp[theta][j]= (gpp[j]-gmp[j])/2./delti[theta];
          }
                           
        } /* End theta */
        
        /* We got the gradient matrix for each theta and state j */                
        trgradg =ma3x(0,nhstepm,1,nlstate,1,npar); /* veij */
                   
        for(h=0; h<=nhstepm; h++) /* veij */
          for(j=1; j<=nlstate;j++)
            for(theta=1; theta <=npar; theta++)
              trgradg[h][j][theta]=gradg[h][theta][j];
                   
        for(j=nlstate+1; j<=nlstate+ndeath;j++) /* mu */
          for(theta=1; theta <=npar; theta++)
            trgradgp[j][theta]=gradgp[theta][j];
        /**< as well as its transposed matrix 
         */                
                   
        hf=hstepm*stepm/YEARM;  /* Duration of hstepm expressed in year unit. */
        for(i=1;i<=nlstate;i++)
          for(j=1;j<=nlstate;j++)
            vareij[i][j][(int)age] =0.;
   
        /* Computing trgradg by matcov by gradg at age and summing over h
         * and k (nhstepm) formula 15 of article
         * Lievre-Brouard-Heathcote
         */
        
        for(h=0;h<=nhstepm;h++){
          for(k=0;k<=nhstepm;k++){
            matprod2(dnewm,trgradg[h],1,nlstate,1,npar,1,npar,matcov);
            matprod2(doldm,dnewm,1,nlstate,1,npar,1,nlstate,gradg[k]);
            for(i=1;i<=nlstate;i++)
              for(j=1;j<=nlstate;j++)
                vareij[i][j][(int)age] += doldm[i][j]*hf*hf;
          }
        }
                   
        /* pptj is p.3 or p.j = trgradgp by cov by gradgp, variance of
         * p.j overall mortality formula 49 but computed directly because
         * we compute the grad (wix pijx) instead of grad (pijx),even if
         * wix is independent of theta.
         */
        matprod2(dnewmp,trgradgp,nlstate+1,nlstate+ndeath,1,npar,1,npar,matcov);
        matprod2(doldmp,dnewmp,nlstate+1,nlstate+ndeath,1,npar,nlstate+1,nlstate+ndeath,gradgp);
        for(j=nlstate+1;j<=nlstate+ndeath;j++)
          for(i=nlstate+1;i<=nlstate+ndeath;i++)
            varppt[j][i]=doldmp[j][i];
        /* end ppptj */
        /*  x centered again */
                   
        prevalim(prlim,nlstate,x,age,oldm,savm,ftolpl,ncvyearp,ij, nres);
                   
        if (popbased==1) {
          if(mobilav ==0){
            for(i=1; i<=nlstate;i++)
              prlim[i][i]=probs[(int)age][i][ij];
          }else{ /* mobilav */ 
            for(i=1; i<=nlstate;i++)
              prlim[i][i]=mobaverage[(int)age][i][ij];
          }
        }
                   
        /* This for computing probability of death (h=1 means
           computed over hstepm (estepm) matrices product = hstepm*stepm months) 
           as a weighted average of prlim.
        */
        hpxij(p3mat,nhstepm,age,hstepm,x,nlstate,stepm,oldm,savm, ij, nres);  
        for(j=nlstate+1;j<=nlstate+ndeath;j++){
          for(i=1,gmp[j]=0.;i<= nlstate; i++) 
            gmp[j] += prlim[i][i]*p3mat[i][j][1]; 
        }    
        /* end probability of death */
                   
        fprintf(ficresprobmorprev,"%3d %d ",(int) age, ij);
        for(j=nlstate+1; j<=(nlstate+ndeath);j++){
          fprintf(ficresprobmorprev," %11.3e %11.3e",gmp[j], sqrt(varppt[j][j]));
          for(i=1; i<=nlstate;i++){
            fprintf(ficresprobmorprev," %11.3e %11.3e ",prlim[i][i],p3mat[i][j][1]);
          }
        } 
        fprintf(ficresprobmorprev,"\n");
                   
        fprintf(ficresvij,"%.0f ",age );
        for(i=1; i<=nlstate;i++)
          for(j=1; j<=nlstate;j++){
            fprintf(ficresvij," %.4f", vareij[i][j][(int)age]);
          }
        fprintf(ficresvij,"\n");
        free_matrix(gp,0,nhstepm,1,nlstate);
        free_matrix(gm,0,nhstepm,1,nlstate);
        free_ma3x(gradg,0,nhstepm,1,npar,1,nlstate);
        free_ma3x(trgradg,0,nhstepm,1,nlstate,1,npar);
        free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
      } /* End age */
      free_vector(gpp,nlstate+1,nlstate+ndeath);
      free_vector(gmp,nlstate+1,nlstate+ndeath);
      free_matrix(gradgp,1,npar,nlstate+1,nlstate+ndeath);
      free_matrix(trgradgp,nlstate+1,nlstate+ndeath,1,npar); /* mu or p point j*/
      /* fprintf(ficgp,"\nunset parametric;unset label; set ter png small size 320, 240"); */
      fprintf(ficgp,"\nunset parametric;unset label; set ter svg size 640, 480");
      /* for(j=nlstate+1; j<= nlstate+ndeath; j++){ *//* Only the first actually */
      fprintf(ficgp,"\n set log y; unset log x;set xlabel \"Age\"; set ylabel \"Force of mortality (year-1)\";");
      fprintf(ficgp,"\nset out \"%s%s.svg\";",subdirf3(optionfilefiname,"VARMUPTJGR-",digitp),digit);
      /*   fprintf(ficgp,"\n plot \"%s\"  u 1:($3*%6.3f) not w l 1 ",fileresprobmorprev,YEARM/estepm); */
      /*   fprintf(ficgp,"\n replot \"%s\"  u 1:(($3+1.96*$4)*%6.3f) t \"95\%% interval\" w l 2 ",fileresprobmorprev,YEARM/estepm); */
      /*   fprintf(ficgp,"\n replot \"%s\"  u 1:(($3-1.96*$4)*%6.3f) not w l 2 ",fileresprobmorprev,YEARM/estepm); */
      fprintf(ficgp,"\n plot \"%s\"  u 1:($3) not w l lt 1 ",subdirf(fileresprobmorprev));
      fprintf(ficgp,"\n replot \"%s\"  u 1:(($3+1.96*$4)) t \"95%% interval\" w l lt 2 ",subdirf(fileresprobmorprev));
      fprintf(ficgp,"\n replot \"%s\"  u 1:(($3-1.96*$4)) not w l lt 2 ",subdirf(fileresprobmorprev));
      fprintf(fichtm,"\n<br> File (multiple files are possible if covariates are present): <A href=\"%s\">%s</a>\n",subdirf(fileresprobmorprev),subdirf(fileresprobmorprev));
      fprintf(fichtm,"\n<br> Probability is computed over estepm=%d months. <br> <img src=\"%s%s.svg\"> <br>\n", estepm,subdirf3(optionfilefiname,"VARMUPTJGR-",digitp),digit);
      /*  fprintf(fichtm,"\n<br> Probability is computed over estepm=%d months and then divided by estepm and multiplied by %.0f in order to have the probability to die over a year <br> <img src=\"varmuptjgr%s%s.svg\"> <br>\n", stepm,YEARM,digitp,digit);
       */
      /*   fprintf(ficgp,"\nset out \"varmuptjgr%s%s%s.svg\";replot;",digitp,optionfilefiname,digit); */
      fprintf(ficgp,"\nset out;\nset out \"%s%s.svg\";replot;set out;\n",subdirf3(optionfilefiname,"VARMUPTJGR-",digitp),digit);
   
      free_vector(xp,1,npar);
      free_matrix(doldm,1,nlstate,1,nlstate);
      free_matrix(dnewm,1,nlstate,1,npar);
      free_matrix(doldmp,nlstate+1,nlstate+ndeath,nlstate+1,nlstate+ndeath);
      free_matrix(dnewmp,nlstate+1,nlstate+ndeath,1,npar);
      free_matrix(varppt,nlstate+1,nlstate+ndeath,nlstate+1,nlstate+ndeath);
      /* if (mobilav!=0) free_ma3x(mobaverage,1, AGESUP,1,NCOVMAX, 1,NCOVMAX); */
      fclose(ficresprobmorprev);
      fflush(ficgp);
      fflush(fichtm); 
    }  /* end varevsij */
   
   /************ Variance of prevlim ******************/
    void varprevlim(char fileresvpl[], FILE *ficresvpl, double **varpl, double **matcov, double x[], double delti[], int nlstate, int stepm, double bage, double fage, double **oldm, double **savm, double **prlim, double ftolpl, int *ncvyearp, int ij, char strstart[], int nres)
   {
     /* Variance of prevalence limit  for each state ij using current parameters x[] and estimates of neighbourhood give by delti*/
     /*  double **prevalim(double **prlim, int nlstate, double *xp, double age, double **oldm, double **savm,double ftolpl);*/
   
     double **dnewmpar,**doldm;
     int i, j, nhstepm, hstepm;
     double *xp;
     double *gp, *gm;
     double **gradg, **trgradg;
     double **mgm, **mgp;
     double age,agelim;
     int theta;
     
     pstamp(ficresvpl);
     fprintf(ficresvpl,"# Standard deviation of period (forward stable) prevalences \n");
     fprintf(ficresvpl,"# Age ");
     if(nresult >=1)
       fprintf(ficresvpl," Result# ");
     for(i=1; i<=nlstate;i++)
         fprintf(ficresvpl," %1d-%1d",i,i);
     fprintf(ficresvpl,"\n");
   
     xp=vector(1,npar);
     dnewmpar=matrix(1,nlstate,1,npar);
     doldm=matrix(1,nlstate,1,nlstate);
     
     hstepm=1*YEARM; /* Every year of age */
     hstepm=hstepm/stepm; /* Typically in stepm units, if j= 2 years, = 2/6 months = 4 */ 
     agelim = AGESUP;
     for (age=bage; age<=fage; age ++){ /* If stepm=6 months */
       nhstepm=(int) rint((agelim-age)*YEARM/stepm); /* Typically 20 years = 20*12/6=40 */ 
       if (stepm >= YEARM) hstepm=1;
       nhstepm = nhstepm/hstepm; /* Typically 40/4=10 */
       gradg=matrix(1,npar,1,nlstate);
       mgp=matrix(1,npar,1,nlstate);
       mgm=matrix(1,npar,1,nlstate);
       gp=vector(1,nlstate);
       gm=vector(1,nlstate);
   
   first=1;      for(theta=1; theta <=npar; theta++){
         for(i=1; i<=npar; i++){ /* Computes gradient */
           xp[i] = x[i] + (i==theta ?delti[theta]:0);
         }
         /* if((int)age==79 ||(int)age== 80 ||(int)age== 81 ) */
         /*        prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ncvyearp,ij,nres); */
         /* else */
         prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ncvyearp,ij,nres);
         for(i=1;i<=nlstate;i++){
           gp[i] = prlim[i][i];
           mgp[theta][i] = prlim[i][i];
         }
         for(i=1; i<=npar; i++) /* Computes gradient */
           xp[i] = x[i] - (i==theta ?delti[theta]:0);
         /* if((int)age==79 ||(int)age== 80 ||(int)age== 81 ) */
         /*        prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ncvyearp,ij,nres); */
         /* else */
         prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ncvyearp,ij,nres);
         for(i=1;i<=nlstate;i++){
           gm[i] = prlim[i][i];
           mgm[theta][i] = prlim[i][i];
         }
         for(i=1;i<=nlstate;i++)
           gradg[theta][i]= (gp[i]-gm[i])/2./delti[theta];
         /* gradg[theta][2]= -gradg[theta][1]; */ /* For testing if nlstate=2 */
       } /* End theta */
   
   for(k1=1; k1<=j;k1++){      trgradg =matrix(1,nlstate,1,npar);
     for(i1=1; i1<=ncodemax[k1];i1++){  
       j1++;  
       /*printf("cptcoveff=%d Tvaraff=%d", cptcoveff,Tvaraff[1]);  
         scanf("%d", i);*/  
       for (i=-1; i<=nlstate+ndeath; i++)    
         for (jk=-1; jk<=nlstate+ndeath; jk++)    
           for(m=iagemin; m <= iagemax+3; m++)  
             freq[i][jk][m]=0;  
   
     for (i=1; i<=nlstate; i++)        for(j=1; j<=nlstate;j++)
       for(m=iagemin; m <= iagemax+3; m++)        for(theta=1; theta <=npar; theta++)
         prop[i][m]=0;          trgradg[j][theta]=gradg[theta][j];
             /* if((int)age==79 ||(int)age== 80 ||(int)age== 81 ){ */
       dateintsum=0;      /*   printf("\nmgm mgp %d ",(int)age); */
       k2cpt=0;      /*   for(j=1; j<=nlstate;j++){ */
       for (i=1; i<=imx; i++) {      /*  printf(" %d ",j); */
         bool=1;      /*  for(theta=1; theta <=npar; theta++) */
         if  (cptcovn>0) {      /*    printf(" %d %lf %lf",theta,mgm[theta][j],mgp[theta][j]); */
           for (z1=1; z1<=cptcoveff; z1++)       /*  printf("\n "); */
             if (covar[Tvaraff[z1]][i]!= nbcode[Tvaraff[z1]][codtab[j1][z1]])       /*   } */
               bool=0;      /* } */
         }      /* if((int)age==79 ||(int)age== 80 ||(int)age== 81 ){ */
         if (bool==1){      /*   printf("\n gradg %d ",(int)age); */
           for(m=firstpass; m<=lastpass; m++){      /*   for(j=1; j<=nlstate;j++){ */
             k2=anint[m][i]+(mint[m][i]/12.);      /*  printf("%d ",j); */
             /*if ((k2>=dateprev1) && (k2<=dateprev2)) {*/      /*  for(theta=1; theta <=npar; theta++) */
               if(agev[m][i]==0) agev[m][i]=iagemax+1;      /*    printf("%d %lf ",theta,gradg[theta][j]); */
               if(agev[m][i]==1) agev[m][i]=iagemax+2;      /*  printf("\n "); */
               if (s[m][i]>0 && s[m][i]<=nlstate) prop[s[m][i]][(int)agev[m][i]] += weight[i];      /*   } */
               if (m<lastpass) {      /* } */
                 freq[s[m][i]][s[m+1][i]][(int)agev[m][i]] += weight[i];  
                 freq[s[m][i]][s[m+1][i]][iagemax+3] += weight[i];  
               }  
                 
               if ((agev[m][i]>1) && (agev[m][i]< (iagemax+3))) {  
                 dateintsum=dateintsum+k2;  
                 k2cpt++;  
               }  
               /*}*/  
           }  
         }  
       }  
          
       /*      fprintf(ficresp, "#Count between %.lf/%.lf/%.lf and %.lf/%.lf/%.lf\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2);*/  
   
       if  (cptcovn>0) {      for(i=1;i<=nlstate;i++)
         fprintf(ficresp, "\n#********** Variable ");         varpl[i][(int)age] =0.;
         for (z1=1; z1<=cptcoveff; z1++) fprintf(ficresp, "V%d=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtab[j1][z1]]);      if((int)age==79 ||(int)age== 80  ||(int)age== 81){
         fprintf(ficresp, "**********\n#");      matprod2(dnewmpar,trgradg,1,nlstate,1,npar,1,npar,matcov);
       }      matprod2(doldm,dnewmpar,1,nlstate,1,npar,1,nlstate,gradg);
       for(i=1; i<=nlstate;i++)       }else{
         fprintf(ficresp, " Age Prev(%d) N(%d) N",i,i);      matprod2(dnewmpar,trgradg,1,nlstate,1,npar,1,npar,matcov);
       fprintf(ficresp, "\n");      matprod2(doldm,dnewmpar,1,nlstate,1,npar,1,nlstate,gradg);
             }
       for(i=iagemin; i <= iagemax+3; i++){      for(i=1;i<=nlstate;i++)
         if(i==iagemax+3){        varpl[i][(int)age] = doldm[i][i]; /* Covariances are useless */
           fprintf(ficlog,"Total");  
         }else{  
           if(first==1){  
             first=0;  
             printf("See log file for details...\n");  
           }  
           fprintf(ficlog,"Age %d", i);  
         }  
         for(jk=1; jk <=nlstate ; jk++){  
           for(m=-1, pp[jk]=0; m <=nlstate+ndeath ; m++)  
             pp[jk] += freq[jk][m][i];   
         }  
         for(jk=1; jk <=nlstate ; jk++){  
           for(m=-1, pos=0; m <=0 ; m++)  
             pos += freq[jk][m][i];  
           if(pp[jk]>=1.e-10){  
             if(first==1){  
             printf(" %d.=%.0f loss[%d]=%.1f%%",jk,pp[jk],jk,100*pos/pp[jk]);  
             }  
             fprintf(ficlog," %d.=%.0f loss[%d]=%.1f%%",jk,pp[jk],jk,100*pos/pp[jk]);  
           }else{  
             if(first==1)  
               printf(" %d.=%.0f loss[%d]=NaNQ%%",jk,pp[jk],jk);  
             fprintf(ficlog," %d.=%.0f loss[%d]=NaNQ%%",jk,pp[jk],jk);  
           }  
         }  
   
         for(jk=1; jk <=nlstate ; jk++){      fprintf(ficresvpl,"%.0f ",age );
           for(m=0, pp[jk]=0; m <=nlstate+ndeath; m++)      if(nresult >=1)
             pp[jk] += freq[jk][m][i];        fprintf(ficresvpl,"%d ",nres );
         }             for(i=1; i<=nlstate;i++){
         for(jk=1,pos=0,posprop=0; jk <=nlstate ; jk++){        fprintf(ficresvpl," %.5f (%.5f)",prlim[i][i],sqrt(varpl[i][(int)age]));
           pos += pp[jk];        /* for(j=1;j<=nlstate;j++) */
           posprop += prop[jk][i];        /*        fprintf(ficresvpl," %d %.5f ",j,prlim[j][i]); */
         }  
         for(jk=1; jk <=nlstate ; jk++){  
           if(pos>=1.e-5){  
             if(first==1)  
               printf(" %d.=%.0f prev[%d]=%.1f%%",jk,pp[jk],jk,100*pp[jk]/pos);  
             fprintf(ficlog," %d.=%.0f prev[%d]=%.1f%%",jk,pp[jk],jk,100*pp[jk]/pos);  
           }else{  
             if(first==1)  
               printf(" %d.=%.0f prev[%d]=NaNQ%%",jk,pp[jk],jk);  
             fprintf(ficlog," %d.=%.0f prev[%d]=NaNQ%%",jk,pp[jk],jk);  
           }  
           if( i <= iagemax){  
             if(pos>=1.e-5){  
               fprintf(ficresp," %d %.5f %.0f %.0f",i,prop[jk][i]/posprop, prop[jk][i],posprop);  
               /*probs[i][jk][j1]= pp[jk]/pos;*/  
               /*printf("\ni=%d jk=%d j1=%d %.5f %.0f %.0f %f",i,jk,j1,pp[jk]/pos, pp[jk],pos,probs[i][jk][j1]);*/  
             }  
             else  
               fprintf(ficresp," %d NaNq %.0f %.0f",i,prop[jk][i],posprop);  
           }  
         }  
           
         for(jk=-1; jk <=nlstate+ndeath; jk++)  
           for(m=-1; m <=nlstate+ndeath; m++)  
             if(freq[jk][m][i] !=0 ) {  
             if(first==1)  
               printf(" %d%d=%.0f",jk,m,freq[jk][m][i]);  
               fprintf(ficlog," %d%d=%.0f",jk,m,freq[jk][m][i]);  
             }  
         if(i <= iagemax)  
           fprintf(ficresp,"\n");  
         if(first==1)  
           printf("Others in log...\n");  
         fprintf(ficlog,"\n");  
       }  
     }      }
   }      fprintf(ficresvpl,"\n");
   dateintmean=dateintsum/k2cpt;       free_vector(gp,1,nlstate);
        free_vector(gm,1,nlstate);
   fclose(ficresp);      free_matrix(mgm,1,npar,1,nlstate);
   free_ma3x(freq,-1,nlstate+ndeath,-1,nlstate+ndeath, iagemin, iagemax+3);      free_matrix(mgp,1,npar,1,nlstate);
   free_vector(pp,1,nlstate);      free_matrix(gradg,1,npar,1,nlstate);
   free_matrix(prop,1,nlstate,iagemin, iagemax+3);      free_matrix(trgradg,1,nlstate,1,npar);
   /* End of Freq */    } /* End age */
   
     free_vector(xp,1,npar);
     free_matrix(doldm,1,nlstate,1,npar);
     free_matrix(dnewmpar,1,nlstate,1,nlstate);
   
 }  }
   
 /************ Prevalence ********************/  
 void prevalence(double ***probs, double agemin, double agemax, int **s, double **agev, int nlstate, int imx, int *Tvar, int **nbcode, int *ncodemax,double **mint,double **anint, double dateprev1,double dateprev2, int firstpass, int lastpass)  
 {    
   /* Compute observed prevalence between dateprev1 and dateprev2 by counting the number of people  
      in each health status at the date of interview (if between dateprev1 and dateprev2).  
      We still use firstpass and lastpass as another selection.  
   */  
    
   int i, m, jk, k1, i1, j1, bool, z1,z2,j;  
   double ***freq; /* Frequencies */  
   double *pp, **prop;  
   double pos,posprop;   
   double  y2; /* in fractional years */  
   int iagemin, iagemax;  
   
   iagemin= (int) agemin;  /************ Variance of backprevalence limit ******************/
   iagemax= (int) agemax;   void varbrevlim(char fileresvbl[], FILE  *ficresvbl, double **varbpl, double **matcov, double x[], double delti[], int nlstate, int stepm, double bage, double fage, double **oldm, double **savm, double **bprlim, double ftolpl, int mobilavproj, int *ncvyearp, int ij, char strstart[], int nres)
   /*pp=vector(1,nlstate);*/  {
   prop=matrix(1,nlstate,iagemin,iagemax+3);     /* Variance of backward prevalence limit  for each state ij using current parameters x[] and estimates of neighbourhood give by delti*/
   /*  freq=ma3x(-1,nlstate+ndeath,-1,nlstate+ndeath,iagemin,iagemax+3);*/    /*  double **prevalim(double **prlim, int nlstate, double *xp, double age, double **oldm, double **savm,double ftolpl);*/
   j1=0;  
     double **dnewmpar,**doldm;
     int i, j, nhstepm, hstepm;
     double *xp;
     double *gp, *gm;
     double **gradg, **trgradg;
     double **mgm, **mgp;
     double age,agelim;
     int theta;
       
   j=cptcoveff;    pstamp(ficresvbl);
   if (cptcovn<1) {j=1;ncodemax[1]=1;}    fprintf(ficresvbl,"# Standard deviation of back (stable) prevalences \n");
     fprintf(ficresvbl,"# Age ");
     if(nresult >=1)
       fprintf(ficresvbl," Result# ");
     for(i=1; i<=nlstate;i++)
         fprintf(ficresvbl," %1d-%1d",i,i);
     fprintf(ficresvbl,"\n");
   
     xp=vector(1,npar);
     dnewmpar=matrix(1,nlstate,1,npar);
     doldm=matrix(1,nlstate,1,nlstate);
       
   for(k1=1; k1<=j;k1++){    hstepm=1*YEARM; /* Every year of age */
     for(i1=1; i1<=ncodemax[k1];i1++){    hstepm=hstepm/stepm; /* Typically in stepm units, if j= 2 years, = 2/6 months = 4 */ 
       j1++;    agelim = AGEINF;
           for (age=fage; age>=bage; age --){ /* If stepm=6 months */
       for (i=1; i<=nlstate; i++)        nhstepm=(int) rint((age-agelim)*YEARM/stepm); /* Typically 20 years = 20*12/6=40 */ 
         for(m=iagemin; m <= iagemax+3; m++)      if (stepm >= YEARM) hstepm=1;
           prop[i][m]=0.0;      nhstepm = nhstepm/hstepm; /* Typically 40/4=10 */
            gradg=matrix(1,npar,1,nlstate);
       for (i=1; i<=imx; i++) { /* Each individual */      mgp=matrix(1,npar,1,nlstate);
         bool=1;      mgm=matrix(1,npar,1,nlstate);
         if  (cptcovn>0) {      gp=vector(1,nlstate);
           for (z1=1; z1<=cptcoveff; z1++)       gm=vector(1,nlstate);
             if (covar[Tvaraff[z1]][i]!= nbcode[Tvaraff[z1]][codtab[j1][z1]])   
               bool=0;      for(theta=1; theta <=npar; theta++){
         }         for(i=1; i<=npar; i++){ /* Computes gradient */
         if (bool==1) {           xp[i] = x[i] + (i==theta ?delti[theta]:0);
           for(m=firstpass; m<=lastpass; m++){/* Other selection (we can limit to certain interviews*/  
             y2=anint[m][i]+(mint[m][i]/12.); /* Fractional date in year */  
             if ((y2>=dateprev1) && (y2<=dateprev2)) { /* Here is the main selection (fractional years) */  
               if(agev[m][i]==0) agev[m][i]=iagemax+1;  
               if(agev[m][i]==1) agev[m][i]=iagemax+2;  
               if((int)agev[m][i] <iagemin || (int)agev[m][i] >iagemax+3) printf("Error on individual =%d agev[m][i]=%f m=%d\n",i, agev[m][i],m);   
               if (s[m][i]>0 && s[m][i]<=nlstate) {   
                 /*if(i>4620) printf(" i=%d m=%d s[m][i]=%d (int)agev[m][i]=%d weight[i]=%f prop=%f\n",i,m,s[m][i],(int)agev[m][m],weight[i],prop[s[m][i]][(int)agev[m][i]]);*/  
                 prop[s[m][i]][(int)agev[m][i]] += weight[i];  
                 prop[s[m][i]][iagemax+3] += weight[i];   
               }   
             }  
           } /* end selection of waves */  
         }  
       }        }
       for(i=iagemin; i <= iagemax+3; i++){          if(mobilavproj > 0 )
                   bprevalim(bprlim, mobaverage,nlstate,xp,age,ftolpl,ncvyearp,ij,nres);
         for(jk=1,posprop=0; jk <=nlstate ; jk++) {         else
           posprop += prop[jk][i];           bprevalim(bprlim, mobaverage,nlstate,xp,age,ftolpl,ncvyearp,ij,nres);
         }         for(i=1;i<=nlstate;i++){
           gp[i] = bprlim[i][i];
         for(jk=1; jk <=nlstate ; jk++){               mgp[theta][i] = bprlim[i][i];
           if( i <=  iagemax){         }
             if(posprop>=1.e-5){        for(i=1; i<=npar; i++) /* Computes gradient */
               probs[i][jk][j1]= prop[jk][i]/posprop;          xp[i] = x[i] - (i==theta ?delti[theta]:0);
             }          if(mobilavproj > 0 )
           }           bprevalim(bprlim, mobaverage,nlstate,xp,age,ftolpl,ncvyearp,ij,nres);
         }/* end jk */          else
       }/* end i */           bprevalim(bprlim, mobaverage,nlstate,xp,age,ftolpl,ncvyearp,ij,nres);
     } /* end i1 */        for(i=1;i<=nlstate;i++){
   } /* end k1 */          gm[i] = bprlim[i][i];
             mgm[theta][i] = bprlim[i][i];
   /*  free_ma3x(freq,-1,nlstate+ndeath,-1,nlstate+ndeath, iagemin, iagemax+3);*/        }
   /*free_vector(pp,1,nlstate);*/        for(i=1;i<=nlstate;i++)
   free_matrix(prop,1,nlstate, iagemin,iagemax+3);          gradg[theta][i]= (gp[i]-gm[i])/2./delti[theta];
 }  /* End of prevalence */        /* gradg[theta][2]= -gradg[theta][1]; */ /* For testing if nlstate=2 */
       } /* End theta */
   
 /************* Waves Concatenation ***************/      trgradg =matrix(1,nlstate,1,npar);
   
 void  concatwav(int wav[], int **dh, int **bh,  int **mw, int **s, double *agedc, double **agev, int  firstpass, int lastpass, int imx, int nlstate, int stepm)      for(j=1; j<=nlstate;j++)
 {        for(theta=1; theta <=npar; theta++)
   /* Concatenates waves: wav[i] is the number of effective (useful waves) of individual i.          trgradg[j][theta]=gradg[theta][j];
      Death is a valid wave (if date is known).      /* if((int)age==79 ||(int)age== 80 ||(int)age== 81 ){ */
      mw[mi][i] is the mi (mi=1 to wav[i])  effective wave of individual i      /*   printf("\nmgm mgp %d ",(int)age); */
      dh[m][i] or dh[mw[mi][i]][i] is the delay between two effective waves m=mw[mi][i]      /*   for(j=1; j<=nlstate;j++){ */
      and mw[mi+1][i]. dh depends on stepm.      /*  printf(" %d ",j); */
      */      /*  for(theta=1; theta <=npar; theta++) */
       /*    printf(" %d %lf %lf",theta,mgm[theta][j],mgp[theta][j]); */
       /*  printf("\n "); */
       /*   } */
       /* } */
       /* if((int)age==79 ||(int)age== 80 ||(int)age== 81 ){ */
       /*   printf("\n gradg %d ",(int)age); */
       /*   for(j=1; j<=nlstate;j++){ */
       /*  printf("%d ",j); */
       /*  for(theta=1; theta <=npar; theta++) */
       /*    printf("%d %lf ",theta,gradg[theta][j]); */
       /*  printf("\n "); */
       /*   } */
       /* } */
   
   int i, mi, m;      for(i=1;i<=nlstate;i++)
   /* int j, k=0,jk, ju, jl,jmin=1e+5, jmax=-1;        varbpl[i][(int)age] =0.;
      double sum=0., jmean=0.;*/      if((int)age==79 ||(int)age== 80  ||(int)age== 81){
   int first;      matprod2(dnewmpar,trgradg,1,nlstate,1,npar,1,npar,matcov);
   int j, k=0,jk, ju, jl;      matprod2(doldm,dnewmpar,1,nlstate,1,npar,1,nlstate,gradg);
   double sum=0.;      }else{
   first=0;      matprod2(dnewmpar,trgradg,1,nlstate,1,npar,1,npar,matcov);
   jmin=1e+5;      matprod2(doldm,dnewmpar,1,nlstate,1,npar,1,nlstate,gradg);
   jmax=-1;  
   jmean=0.;  
   for(i=1; i<=imx; i++){  
     mi=0;  
     m=firstpass;  
     while(s[m][i] <= nlstate){  
       if(s[m][i]>=1)  
         mw[++mi][i]=m;  
       if(m >=lastpass)  
         break;  
       else  
         m++;  
     }/* end while */  
     if (s[m][i] > nlstate){  
       mi++;     /* Death is another wave */  
       /* if(mi==0)  never been interviewed correctly before death */  
          /* Only death is a correct wave */  
       mw[mi][i]=m;  
     }      }
       for(i=1;i<=nlstate;i++)
         varbpl[i][(int)age] = doldm[i][i]; /* Covariances are useless */
   
     wav[i]=mi;      fprintf(ficresvbl,"%.0f ",age );
     if(mi==0){      if(nresult >=1)
       if(first==0){        fprintf(ficresvbl,"%d ",nres );
         printf("Warning! None valid information for:%d line=%d (skipped) and may be others, see log file\n",num[i],i);      for(i=1; i<=nlstate;i++)
         first=1;        fprintf(ficresvbl," %.5f (%.5f)",bprlim[i][i],sqrt(varbpl[i][(int)age]));
       }      fprintf(ficresvbl,"\n");
       if(first==1){      free_vector(gp,1,nlstate);
         fprintf(ficlog,"Warning! None valid information for:%d line=%d (skipped)\n",num[i],i);      free_vector(gm,1,nlstate);
       }      free_matrix(mgm,1,npar,1,nlstate);
     } /* end mi==0 */      free_matrix(mgp,1,npar,1,nlstate);
   } /* End individuals */      free_matrix(gradg,1,npar,1,nlstate);
       free_matrix(trgradg,1,nlstate,1,npar);
   for(i=1; i<=imx; i++){    } /* End age */
     for(mi=1; mi<wav[i];mi++){  
       if (stepm <=0)  
         dh[mi][i]=1;  
       else{  
         if (s[mw[mi+1][i]][i] > nlstate) { /* A death */  
           if (agedc[i] < 2*AGESUP) {  
           j= rint(agedc[i]*12-agev[mw[mi][i]][i]*12);   
           if(j==0) j=1;  /* Survives at least one month after exam */  
           k=k+1;  
           if (j >= jmax) jmax=j;  
           if (j <= jmin) jmin=j;  
           sum=sum+j;  
           /*if (j<0) printf("j=%d num=%d \n",j,i);*/  
           /*      printf("%d %d %d %d\n", s[mw[mi][i]][i] ,s[mw[mi+1][i]][i],j,i);*/  
           if(j<0)printf("Error! Negative delay (%d to death) between waves %d and %d of individual %d at line %d who is aged %.1f with statuses from %d to %d\n ",j,mw[mi][i],mw[mi+1][i],num[i], i,agev[mw[mi][i]][i],s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]);  
           }  
         }  
         else{  
           j= rint( (agev[mw[mi+1][i]][i]*12 - agev[mw[mi][i]][i]*12));  
           /*      printf("%d %d %d %d\n", s[mw[mi][i]][i] ,s[mw[mi+1][i]][i],j,i);*/  
           k=k+1;  
           if (j >= jmax) jmax=j;  
           else if (j <= jmin)jmin=j;  
           /*        if (j<10) printf("j=%d jmin=%d num=%d ",j,jmin,i); */  
           /*printf("%d %lf %d %d %d\n", i,agev[mw[mi][i]][i],j,s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]);*/  
           if(j<0)printf("Error! Negative delay (%d) between waves %d and %d of individual %d at line %d who is aged %.1f with statuses from %d to %d\n ",j,mw[mi][i],mw[mi+1][i],num[i], i,agev[mw[mi][i]][i],s[mw[mi][i]][i] ,s[mw[mi+1][i]][i]);  
           sum=sum+j;  
         }  
         jk= j/stepm;  
         jl= j -jk*stepm;  
         ju= j -(jk+1)*stepm;  
         if(mle <=1){   
           if(jl==0){  
             dh[mi][i]=jk;  
             bh[mi][i]=0;  
           }else{ /* We want a negative bias in order to only have interpolation ie  
                   * at the price of an extra matrix product in likelihood */  
             dh[mi][i]=jk+1;  
             bh[mi][i]=ju;  
           }  
         }else{  
           if(jl <= -ju){  
             dh[mi][i]=jk;  
             bh[mi][i]=jl;       /* bias is positive if real duration  
                                  * is higher than the multiple of stepm and negative otherwise.  
                                  */  
           }  
           else{  
             dh[mi][i]=jk+1;  
             bh[mi][i]=ju;  
           }  
           if(dh[mi][i]==0){  
             dh[mi][i]=1; /* At least one step */  
             bh[mi][i]=ju; /* At least one step */  
             /*  printf(" bh=%d ju=%d jl=%d dh=%d jk=%d stepm=%d %d\n",bh[mi][i],ju,jl,dh[mi][i],jk,stepm,i);*/  
           }  
         }  
       } /* end if mle */  
     } /* end wave */  
   }  
   jmean=sum/k;  
   printf("Delay (in months) between two waves Min=%d Max=%d Mean=%f\n\n ",jmin, jmax,jmean);  
   fprintf(ficlog,"Delay (in months) between two waves Min=%d Max=%d Mean=%f\n\n ",jmin, jmax,jmean);  
  }  
   
 /*********** Tricode ****************************/    free_vector(xp,1,npar);
 void tricode(int *Tvar, int **nbcode, int imx)    free_matrix(doldm,1,nlstate,1,npar);
 {    free_matrix(dnewmpar,1,nlstate,1,nlstate);
     
   int Ndum[20],ij=1, k, j, i, maxncov=19;  
   int cptcode=0;  
   cptcoveff=0;   
    
   for (k=0; k<maxncov; k++) Ndum[k]=0;  
   for (k=1; k<=7; k++) ncodemax[k]=0;  
   
   for (j=1; j<=(cptcovn+2*cptcovprod); j++) {  }
     for (i=1; i<=imx; i++) { /*reads the data file to get the maximum   
                                modality*/   
       ij=(int)(covar[Tvar[j]][i]); /* ij is the modality of this individual*/  
       Ndum[ij]++; /*store the modality */  
       /*printf("i=%d ij=%d Ndum[ij]=%d imx=%d",i,ij,Ndum[ij],imx);*/  
       if (ij > cptcode) cptcode=ij; /* getting the maximum of covariable   
                                        Tvar[j]. If V=sex and male is 0 and   
                                        female is 1, then  cptcode=1.*/  
     }  
   
     for (i=0; i<=cptcode; i++) {  
       if(Ndum[i]!=0) ncodemax[j]++; /* Nomber of modalities of the j th covariates. In fact ncodemax[j]=2 (dichotom. variables) but it can be more */  
     }  
   
     ij=1;   
     for (i=1; i<=ncodemax[j]; i++) {  
       for (k=0; k<= maxncov; k++) {  
         if (Ndum[k] != 0) {  
           nbcode[Tvar[j]][ij]=k;   
           /* store the modality in an array. k is a modality. If we have model=V1+V1*sex then: nbcode[1][1]=0 ; nbcode[1][2]=1; nbcode[2][1]=0 ; nbcode[2][2]=1; */  
             
           ij++;  
         }  
         if (ij > ncodemax[j]) break;   
       }    
     }   
   }    
   
  for (k=0; k< maxncov; k++) Ndum[k]=0;  /************ Variance of one-step probabilities  ******************/
   void varprob(char optionfilefiname[], double **matcov, double x[], double delti[], int nlstate, double bage, double fage, int ij, int *Tvar, int **nbcode, int *ncodemax, char strstart[])
    {
      int i, j=0,  k1, l1, tj;
      int k2, l2, j1,  z1;
      int k=0, l;
      int first=1, first1, first2;
      int nres=0; /* New */
      double cv12, mu1, mu2, lc1, lc2, v12, v21, v11, v22,v1,v2, c12, tnalp;
      double **dnewm,**doldm;
      double *xp;
      double *gp, *gm;
      double **gradg, **trgradg;
      double **mu;
      double age, cov[NCOVMAX+1];
      double std=2.0; /* Number of standard deviation wide of confidence ellipsoids */
      int theta;
      char fileresprob[FILENAMELENGTH];
      char fileresprobcov[FILENAMELENGTH];
      char fileresprobcor[FILENAMELENGTH];
      double ***varpij;
   
      strcpy(fileresprob,"PROB_"); 
      strcat(fileresprob,fileres);
      if((ficresprob=fopen(fileresprob,"w"))==NULL) {
        printf("Problem with resultfile: %s\n", fileresprob);
        fprintf(ficlog,"Problem with resultfile: %s\n", fileresprob);
      }
      strcpy(fileresprobcov,"PROBCOV_"); 
      strcat(fileresprobcov,fileresu);
      if((ficresprobcov=fopen(fileresprobcov,"w"))==NULL) {
        printf("Problem with resultfile: %s\n", fileresprobcov);
        fprintf(ficlog,"Problem with resultfile: %s\n", fileresprobcov);
      }
      strcpy(fileresprobcor,"PROBCOR_"); 
      strcat(fileresprobcor,fileresu);
      if((ficresprobcor=fopen(fileresprobcor,"w"))==NULL) {
        printf("Problem with resultfile: %s\n", fileresprobcor);
        fprintf(ficlog,"Problem with resultfile: %s\n", fileresprobcor);
      }
      printf("Computing standard deviation of one-step probabilities: result on file '%s' \n",fileresprob);
      fprintf(ficlog,"Computing standard deviation of one-step probabilities: result on file '%s' \n",fileresprob);
      printf("Computing matrix of variance covariance of one-step probabilities: result on file '%s' \n",fileresprobcov);
      fprintf(ficlog,"Computing matrix of variance covariance of one-step probabilities: result on file '%s' \n",fileresprobcov);
      printf("and correlation matrix of one-step probabilities: result on file '%s' \n",fileresprobcor);
      fprintf(ficlog,"and correlation matrix of one-step probabilities: result on file '%s' \n",fileresprobcor);
      pstamp(ficresprob);
      fprintf(ficresprob,"#One-step probabilities and stand. devi in ()\n");
      fprintf(ficresprob,"# Age");
      pstamp(ficresprobcov);
      fprintf(ficresprobcov,"#One-step probabilities and covariance matrix\n");
      fprintf(ficresprobcov,"# Age");
      pstamp(ficresprobcor);
      fprintf(ficresprobcor,"#One-step probabilities and correlation matrix\n");
      fprintf(ficresprobcor,"# Age");
   
   
      for(i=1; i<=nlstate;i++)
        for(j=1; j<=(nlstate+ndeath);j++){
          fprintf(ficresprob," p%1d-%1d (SE)",i,j);
          fprintf(ficresprobcov," p%1d-%1d ",i,j);
          fprintf(ficresprobcor," p%1d-%1d ",i,j);
        }  
      /* fprintf(ficresprob,"\n");
         fprintf(ficresprobcov,"\n");
         fprintf(ficresprobcor,"\n");
      */
      xp=vector(1,npar);
      dnewm=matrix(1,(nlstate)*(nlstate+ndeath),1,npar);
      doldm=matrix(1,(nlstate)*(nlstate+ndeath),1,(nlstate)*(nlstate+ndeath));
      mu=matrix(1,(nlstate)*(nlstate+ndeath), (int) bage, (int)fage);
      varpij=ma3x(1,nlstate*(nlstate+ndeath),1,nlstate*(nlstate+ndeath),(int) bage, (int) fage);
      first=1;
      fprintf(ficgp,"\n# Routine varprob");
      fprintf(fichtm,"\n<li><h4> Computing and drawing one step probabilities with their confidence intervals</h4></li>\n");
      fprintf(fichtm,"\n");
   
      fprintf(fichtm,"\n<li><h4> <a href=\"%s\">Matrix of variance-covariance of one-step probabilities (drawings)</a></h4> this page is important in order to visualize confidence intervals and especially correlation between disability and recovery, or more generally, way in and way back. File %s</li>\n",optionfilehtmcov,optionfilehtmcov);
      fprintf(fichtmcov,"Current page is file <a href=\"%s\">%s</a><br>\n\n<h4>Matrix of variance-covariance of pairs of step probabilities</h4>\n",optionfilehtmcov, optionfilehtmcov);
      fprintf(fichtmcov,"\nEllipsoids of confidence centered on point (p<inf>ij</inf>, p<inf>kl</inf>) are estimated \
   and drawn. It helps understanding how is the covariance between two incidences.\
    They are expressed in year<sup>-1</sup> in order to be less dependent of stepm.<br>\n");
      fprintf(fichtmcov,"\n<br> Contour plot corresponding to x'cov<sup>-1</sup>x = 4 (where x is the column vector (pij,pkl)) are drawn. \
   It can be understood this way: if pij and pkl where uncorrelated the (2x2) matrix of covariance \
   would have been (1/(var pij), 0 , 0, 1/(var pkl)), and the confidence interval would be 2 \
   standard deviations wide on each axis. <br>\
    Now, if both incidences are correlated (usual case) we diagonalised the inverse of the covariance matrix\
    and made the appropriate rotation to look at the uncorrelated principal directions.<br>\
   To be simple, these graphs help to understand the significativity of each parameter in relation to a second other one.<br> \n");
   
      cov[1]=1;
      /* tj=cptcoveff; */
      tj = (int) pow(2,cptcoveff);
      if (cptcovn<1) {tj=1;ncodemax[1]=1;}
      j1=0;
   
      for(nres=1;nres <=nresult; nres++){ /* For each resultline */
      for(j1=1; j1<=tj;j1++){ /* For any combination of dummy covariates, fixed and varying */
        printf("Varprob  TKresult[nres]=%d j1=%d, nres=%d, cptcovn=%d, cptcoveff=%d tj=%d cptcovs=%d\n",  TKresult[nres], j1, nres, cptcovn, cptcoveff, tj, cptcovs);
        if(tj != 1 && TKresult[nres]!= j1)
          continue;
   
      /* for(j1=1; j1<=tj;j1++){  /\* For each valid combination of covariates or only once*\/ */
        /* for(nres=1;nres <=1; nres++){ /\* For each resultline *\/ */
        /* /\* for(nres=1;nres <=nresult; nres++){ /\\* For each resultline *\\/ *\/ */
        if  (cptcovn>0) {
          fprintf(ficresprob, "\n#********** Variable ");
          fprintf(ficresprobcov, "\n#********** Variable "); 
          fprintf(ficgp, "\n#********** Variable ");
          fprintf(fichtmcov, "\n<hr  size=\"2\" color=\"#EC5E5E\">********** Variable "); 
          fprintf(ficresprobcor, "\n#********** Variable ");    
   
          /* Including quantitative variables of the resultline to be done */
          for (z1=1; z1<=cptcovs; z1++){ /* Loop on each variable of this resultline  */
            printf("Varprob modelresult[%d][%d]=%d model=%s \n",nres, z1, modelresult[nres][z1], model);
            fprintf(ficlog,"Varprob modelresult[%d][%d]=%d model=%s \n",nres, z1, modelresult[nres][z1], model);
            /* fprintf(ficlog,"Varprob modelresult[%d][%d]=%d model=%s resultline[%d]=%s \n",nres, z1, modelresult[nres][z1], model, nres, resultline[nres]); */
            if(Dummy[modelresult[nres][z1]]==0){/* Dummy variable of the variable in position modelresult in the model corresponding to z1 in resultline  */
              if(Fixed[modelresult[nres][z1]]==0){ /* Fixed referenced to model equation */
                fprintf(ficresprob,"V%d=%d ",Tvresult[nres][z1],Tresult[nres][z1]); /* Output of each value for the combination TKresult[nres], ordere by the covariate values in the resultline  */
                fprintf(ficresprobcov,"V%d=%d ",Tvresult[nres][z1],Tresult[nres][z1]); /* Output of each value for the combination TKresult[nres], ordere by the covariate values in the resultline  */
                fprintf(ficgp,"V%d=%d ",Tvresult[nres][z1],Tresult[nres][z1]); /* Output of each value for the combination TKresult[nres], ordere by the covariate values in the resultline  */
                fprintf(fichtmcov,"V%d=%d ",Tvresult[nres][z1],Tresult[nres][z1]); /* Output of each value for the combination TKresult[nres], ordere by the covariate values in the resultline  */
                fprintf(ficresprobcor,"V%d=%d ",Tvresult[nres][z1],Tresult[nres][z1]); /* Output of each value for the combination TKresult[nres], ordere by the covariate values in the resultline  */
                fprintf(ficresprob,"fixed ");
                fprintf(ficresprobcov,"fixed ");
                fprintf(ficgp,"fixed ");
                fprintf(fichtmcov,"fixed ");
                fprintf(ficresprobcor,"fixed ");
              }else{
                fprintf(ficresprob,"varyi ");
                fprintf(ficresprobcov,"varyi ");
                fprintf(ficgp,"varyi ");
                fprintf(fichtmcov,"varyi ");
                fprintf(ficresprobcor,"varyi ");
              }
            }else if(Dummy[modelresult[nres][z1]]==1){ /* Quanti variable */
              /* For each selected (single) quantitative value */
              fprintf(ficresprob," V%d=%f ",Tvqresult[nres][z1],Tqresult[nres][z1]);
              if(Fixed[modelresult[nres][z1]]==0){ /* Fixed */
                fprintf(ficresprob,"fixed ");
                fprintf(ficresprobcov,"fixed ");
                fprintf(ficgp,"fixed ");
                fprintf(fichtmcov,"fixed ");
                fprintf(ficresprobcor,"fixed ");
              }else{
                fprintf(ficresprob,"varyi ");
                fprintf(ficresprobcov,"varyi ");
                fprintf(ficgp,"varyi ");
                fprintf(fichtmcov,"varyi ");
                fprintf(ficresprobcor,"varyi ");
              }
            }else{
              printf("Error in varprob() Dummy[modelresult[%d][%d]]=%d, modelresult[%d][%d]=V%d cptcovs=%d, cptcoveff=%d \n", nres, z1, Dummy[modelresult[nres][z1]],nres,z1,modelresult[nres][z1],cptcovs, cptcoveff);  /* end if dummy  or quanti */
              fprintf(ficlog,"Error in varprob() Dummy[modelresult[%d][%d]]=%d, modelresult[%d][%d]=V%d cptcovs=%d, cptcoveff=%d \n", nres, z1, Dummy[modelresult[nres][z1]],nres,z1,modelresult[nres][z1],cptcovs, cptcoveff);  /* end if dummy  or quanti */
              exit(1);
            }
          } /* End loop on variable of this resultline */
          /* for (z1=1; z1<=cptcoveff; z1++) fprintf(ficresprob, "V%d=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtabm(j1,TnsdVar[Tvaraff[z1]])]); */
          fprintf(ficresprob, "**********\n#\n");
          fprintf(ficresprobcov, "**********\n#\n");
          fprintf(ficgp, "**********\n#\n");
          fprintf(fichtmcov, "**********\n<hr size=\"2\" color=\"#EC5E5E\">");
          fprintf(ficresprobcor, "**********\n#");    
          if(invalidvarcomb[j1]){
            fprintf(ficgp,"\n#Combination (%d) ignored because no cases \n",j1); 
            fprintf(fichtmcov,"\n<h3>Combination (%d) ignored because no cases </h3>\n",j1); 
            continue;
          }
        }
        gradg=matrix(1,npar,1,(nlstate)*(nlstate+ndeath));
        trgradg=matrix(1,(nlstate)*(nlstate+ndeath),1,npar);
        gp=vector(1,(nlstate)*(nlstate+ndeath));
        gm=vector(1,(nlstate)*(nlstate+ndeath));
        for (age=bage; age<=fage; age ++){ /* Fo each age we feed the model equation with covariates, using precov as in hpxij() ? */
          cov[2]=age;
          if(nagesqr==1)
            cov[3]= age*age;
          /* New code end of combination but for each resultline */
          for(k1=1;k1<=cptcovt;k1++){ /* loop on model equation (including products) */ 
            if(Typevar[k1]==1){ /* A product with age */
              cov[2+nagesqr+k1]=precov[nres][k1]*cov[2];
            }else{
              cov[2+nagesqr+k1]=precov[nres][k1];
            }
          }/* End of loop on model equation */
   /* Old code */
          /* /\* for (k=1; k<=cptcovn;k++) { *\/ */
          /* /\*    cov[2+nagesqr+k]=nbcode[Tvar[k]][codtabm(j1,k)]; *\/ */
          /* for (k=1; k<=nsd;k++) { /\* For single dummy covariates only *\/ */
          /*        /\* Here comes the value of the covariate 'j1' after renumbering k with single dummy covariates *\/ */
          /*        cov[2+nagesqr+TvarsDind[k]]=nbcode[TvarsD[k]][codtabm(j1,TnsdVar[TvarsD[k]])]; */
          /*        /\*cov[2+nagesqr+k]=nbcode[Tvar[k]][codtabm(j1,Tvar[k])];*\//\* j1 1 2 3 4 */
          /*                                                                   * 1  1 1 1 1 */
          /*                                                                   * 2  2 1 1 1 */
          /*                                                                   * 3  1 2 1 1 */
          /*                                                                   *\/ */
          /*        /\* nbcode[1][1]=0 nbcode[1][2]=1;*\/ */
          /* } */
          /* /\* V2+V1+V4+V3*age Tvar[4]=3 ; V1+V2*age Tvar[2]=2; V1+V1*age Tvar[2]=1, Tage[1]=2 *\/ */
          /* /\* ) p nbcode[Tvar[Tage[k]]][(1 & (ij-1) >> (k-1))+1] *\/ */
          /* /\*for (k=1; k<=cptcovage;k++) cov[2+Tage[k]]=cov[2+Tage[k]]*cov[2]; *\/ */
          /* for (k=1; k<=cptcovage;k++){  /\* For product with age *\/ */
          /*        if(Dummy[Tage[k]]==2){ /\* dummy with age *\/ */
          /*          cov[2+nagesqr+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(j1,TnsdVar[Tvar[Tage[k]]])]*cov[2]; */
          /*          /\* cov[++k1]=nbcode[Tvar[Tage[k]]][codtabm(ij,k)]*cov[2]; *\/ */
          /*        } else if(Dummy[Tage[k]]==3){ /\* quantitative with age *\/ */
          /*          printf("Internal IMaCh error, don't know which value for quantitative covariate with age, Tage[k]%d, k=%d, Tvar[Tage[k]]=V%d, age=%d\n",Tage[k],k ,Tvar[Tage[k]], (int)cov[2]); */
          /*          /\* cov[2+nagesqr+Tage[k]]=meanq[k]/idq[k]*cov[2];/\\* Using the mean of quantitative variable Tvar[Tage[k]] /\\* Tqresult[nres][k]; *\\/ *\/ */
          /*          /\* exit(1); *\/ */
          /*          /\* cov[++k1]=Tqresult[nres][k];  *\/ */
          /*        } */
          /*        /\* cov[2+Tage[k]+nagesqr]=nbcode[Tvar[Tage[k]]][codtabm(ij,k)]*cov[2]; *\/ */
          /* } */
          /* for (k=1; k<=cptcovprod;k++){/\* For product without age *\/ */
          /*        if(Dummy[Tvard[k][1]]==0){ */
          /*          if(Dummy[Tvard[k][2]]==0){ */
          /*            cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(j1,TnsdVar[Tvard[k][1]])] * nbcode[Tvard[k][2]][codtabm(j1,TnsdVar[Tvard[k][2]])]; */
          /*            /\* cov[++k1]=nbcode[Tvard[k][1]][codtabm(ij,k)] * nbcode[Tvard[k][2]][codtabm(ij,k)]; *\/ */
          /*          }else{ /\* Should we use the mean of the quantitative variables? *\/ */
          /*            cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(j1,TnsdVar[Tvard[k][1]])] * Tqresult[nres][resultmodel[nres][k]]; */
          /*            /\* cov[++k1]=nbcode[Tvard[k][1]][codtabm(ij,k)] * Tqresult[nres][k]; *\/ */
          /*          } */
          /*        }else{ */
          /*          if(Dummy[Tvard[k][2]]==0){ */
          /*            cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][2]][codtabm(j1,TnsdVar[Tvard[k][2]])] * Tqinvresult[nres][TnsdVar[Tvard[k][1]]]; */
          /*            /\* cov[++k1]=nbcode[Tvard[k][2]][codtabm(ij,k)] * Tqinvresult[nres][Tvard[k][1]]; *\/ */
          /*          }else{ */
          /*            cov[2+nagesqr+Tprod[k]]=Tqinvresult[nres][TnsdVar[Tvard[k][1]]]*  Tqinvresult[nres][TnsdVar[Tvard[k][2]]]; */
          /*            /\* cov[++k1]=Tqinvresult[nres][Tvard[k][1]]*  Tqinvresult[nres][Tvard[k][2]]; *\/ */
          /*          } */
          /*        } */
          /*        /\* cov[2+nagesqr+Tprod[k]]=nbcode[Tvard[k][1]][codtabm(ij,k)]*nbcode[Tvard[k][2]][codtabm(ij,k)]; *\/ */
          /* } */                  
   /* For each age and combination of dummy covariates we slightly move the parameters of delti in order to get the gradient*/                     
          for(theta=1; theta <=npar; theta++){
            for(i=1; i<=npar; i++)
              xp[i] = x[i] + (i==theta ?delti[theta]:(double)0);
                                   
            pmij(pmmij,cov,ncovmodel,xp,nlstate);
                                   
            k=0;
            for(i=1; i<= (nlstate); i++){
              for(j=1; j<=(nlstate+ndeath);j++){
                k=k+1;
                gp[k]=pmmij[i][j];
              }
            }
                                   
            for(i=1; i<=npar; i++)
              xp[i] = x[i] - (i==theta ?delti[theta]:(double)0);
                                   
            pmij(pmmij,cov,ncovmodel,xp,nlstate);
            k=0;
            for(i=1; i<=(nlstate); i++){
              for(j=1; j<=(nlstate+ndeath);j++){
                k=k+1;
                gm[k]=pmmij[i][j];
              }
            }
                                   
            for(i=1; i<= (nlstate)*(nlstate+ndeath); i++) 
              gradg[theta][i]=(gp[i]-gm[i])/(double)2./delti[theta];  
          }
   
  for (i=1; i<=ncovmodel-2; i++) {          for(j=1; j<=(nlstate)*(nlstate+ndeath);j++)
    /* Listing of all covariables in staement model to see if some covariates appear twice. For example, V1 appears twice in V1+V1*V2.*/           for(theta=1; theta <=npar; theta++)
    ij=Tvar[i];             trgradg[j][theta]=gradg[theta][j];
    Ndum[ij]++;                          
          matprod2(dnewm,trgradg,1,(nlstate)*(nlstate+ndeath),1,npar,1,npar,matcov); 
          matprod2(doldm,dnewm,1,(nlstate)*(nlstate+ndeath),1,npar,1,(nlstate)*(nlstate+ndeath),gradg);
                           
          pmij(pmmij,cov,ncovmodel,x,nlstate);
                           
          k=0;
          for(i=1; i<=(nlstate); i++){
            for(j=1; j<=(nlstate+ndeath);j++){
              k=k+1;
              mu[k][(int) age]=pmmij[i][j];
            }
          }
          for(i=1;i<=(nlstate)*(nlstate+ndeath);i++)
            for(j=1;j<=(nlstate)*(nlstate+ndeath);j++)
              varpij[i][j][(int)age] = doldm[i][j];
                           
          /*printf("\n%d ",(int)age);
            for (i=1; i<=(nlstate)*(nlstate+ndeath);i++){
            printf("%e [%e ;%e] ",gm[i],gm[i]-2*sqrt(doldm[i][i]),gm[i]+2*sqrt(doldm[i][i]));
            fprintf(ficlog,"%e [%e ;%e] ",gm[i],gm[i]-2*sqrt(doldm[i][i]),gm[i]+2*sqrt(doldm[i][i]));
            }*/
                           
          fprintf(ficresprob,"\n%d ",(int)age);
          fprintf(ficresprobcov,"\n%d ",(int)age);
          fprintf(ficresprobcor,"\n%d ",(int)age);
                           
          for (i=1; i<=(nlstate)*(nlstate+ndeath);i++)
            fprintf(ficresprob,"%11.3e (%11.3e) ",mu[i][(int) age],sqrt(varpij[i][i][(int)age]));
          for (i=1; i<=(nlstate)*(nlstate+ndeath);i++){
            fprintf(ficresprobcov,"%11.3e ",mu[i][(int) age]);
            fprintf(ficresprobcor,"%11.3e ",mu[i][(int) age]);
          }
          i=0;
          for (k=1; k<=(nlstate);k++){
            for (l=1; l<=(nlstate+ndeath);l++){ 
              i++;
              fprintf(ficresprobcov,"\n%d %d-%d",(int)age,k,l);
              fprintf(ficresprobcor,"\n%d %d-%d",(int)age,k,l);
              for (j=1; j<=i;j++){
                /* printf(" k=%d l=%d i=%d j=%d\n",k,l,i,j);fflush(stdout); */
                fprintf(ficresprobcov," %11.3e",varpij[i][j][(int)age]);
                fprintf(ficresprobcor," %11.3e",varpij[i][j][(int) age]/sqrt(varpij[i][i][(int) age])/sqrt(varpij[j][j][(int)age]));
              }
            }
          }/* end of loop for state */
        } /* end of loop for age */
        free_vector(gp,1,(nlstate+ndeath)*(nlstate+ndeath));
        free_vector(gm,1,(nlstate+ndeath)*(nlstate+ndeath));
        free_matrix(trgradg,1,(nlstate+ndeath)*(nlstate+ndeath),1,npar);
        free_matrix(gradg,1,(nlstate+ndeath)*(nlstate+ndeath),1,npar);
       
        /* Confidence intervalle of pij  */
        /*
          fprintf(ficgp,"\nunset parametric;unset label");
          fprintf(ficgp,"\nset log y;unset log x; set xlabel \"Age\";set ylabel \"probability (year-1)\"");
          fprintf(ficgp,"\nset ter png small\nset size 0.65,0.65");
          fprintf(fichtm,"\n<br>Probability with  confidence intervals expressed in year<sup>-1</sup> :<a href=\"pijgr%s.png\">pijgr%s.png</A>, ",optionfilefiname,optionfilefiname);
          fprintf(fichtm,"\n<br><img src=\"pijgr%s.png\"> ",optionfilefiname);
          fprintf(ficgp,"\nset out \"pijgr%s.png\"",optionfilefiname);
          fprintf(ficgp,"\nplot \"%s\" every :::%d::%d u 1:2 \"\%%lf",k1,k2,xfilevarprob);
        */
                   
        /* Drawing ellipsoids of confidence of two variables p(k1-l1,k2-l2)*/
        first1=1;first2=2;
        for (k2=1; k2<=(nlstate);k2++){
          for (l2=1; l2<=(nlstate+ndeath);l2++){ 
            if(l2==k2) continue;
            j=(k2-1)*(nlstate+ndeath)+l2;
            for (k1=1; k1<=(nlstate);k1++){
              for (l1=1; l1<=(nlstate+ndeath);l1++){ 
                if(l1==k1) continue;
                i=(k1-1)*(nlstate+ndeath)+l1;
                if(i<=j) continue;
                for (age=bage; age<=fage; age ++){ 
                  if ((int)age %5==0){
                    v1=varpij[i][i][(int)age]/stepm*YEARM/stepm*YEARM;
                    v2=varpij[j][j][(int)age]/stepm*YEARM/stepm*YEARM;
                    cv12=varpij[i][j][(int)age]/stepm*YEARM/stepm*YEARM;
                    mu1=mu[i][(int) age]/stepm*YEARM ;
                    mu2=mu[j][(int) age]/stepm*YEARM;
                    c12=cv12/sqrt(v1*v2);
                    /* Computing eigen value of matrix of covariance */
                    lc1=((v1+v2)+sqrt((v1+v2)*(v1+v2) - 4*(v1*v2-cv12*cv12)))/2.;
                    lc2=((v1+v2)-sqrt((v1+v2)*(v1+v2) - 4*(v1*v2-cv12*cv12)))/2.;
                    if ((lc2 <0) || (lc1 <0) ){
                      if(first2==1){
                        first1=0;
                        printf("Strange: j1=%d One eigen value of 2x2 matrix of covariance is negative, lc1=%11.3e, lc2=%11.3e, v1=%11.3e, v2=%11.3e, cv12=%11.3e.\n It means that the matrix was not well estimated (varpij), for i=%2d, j=%2d, age=%4d .\n See files %s and %s. Probably WRONG RESULTS. See log file for details...\n", j1, lc1, lc2, v1, v2, cv12, i, j, (int)age,fileresprobcov, fileresprobcor);
                      }
                      fprintf(ficlog,"Strange: j1=%d One eigen value of 2x2 matrix of covariance is negative, lc1=%11.3e, lc2=%11.3e, v1=%11.3e, v2=%11.3e, cv12=%11.3e.\n It means that the matrix was not well estimated (varpij), for i=%2d, j=%2d, age=%4d .\n See files %s and %s. Probably WRONG RESULTS.\n", j1, lc1, lc2, v1, v2, cv12, i, j, (int)age,fileresprobcov, fileresprobcor);fflush(ficlog);
                      /* lc1=fabs(lc1); */ /* If we want to have them positive */
                      /* lc2=fabs(lc2); */
                    }
                                                                   
                    /* Eigen vectors */
                    if(1+(v1-lc1)*(v1-lc1)/cv12/cv12 <1.e-5){
                      printf(" Error sqrt of a negative number: %lf\n",1+(v1-lc1)*(v1-lc1)/cv12/cv12);
                      fprintf(ficlog," Error sqrt of a negative number: %lf\n",1+(v1-lc1)*(v1-lc1)/cv12/cv12);
                      v11=(1./sqrt(fabs(1+(v1-lc1)*(v1-lc1)/cv12/cv12)));
                    }else
                      v11=(1./sqrt(1+(v1-lc1)*(v1-lc1)/cv12/cv12));
                    /*v21=sqrt(1.-v11*v11); *//* error */
                    v21=(lc1-v1)/cv12*v11;
                    v12=-v21;
                    v22=v11;
                    tnalp=v21/v11;
                    if(first1==1){
                      first1=0;
                      printf("%d %d%d-%d%d mu %.4e %.4e Var %.4e %.4e cor %.3f cov %.4e Eig %.3e %.3e 1stv %.3f %.3f tang %.3f\nOthers in log...\n",(int) age,k1,l1,k2,l2,mu1,mu2,v1,v2,c12,cv12,lc1,lc2,v11,v21,tnalp);
                    }
                    fprintf(ficlog,"%d %d%d-%d%d mu %.4e %.4e Var %.4e %.4e cor %.3f cov %.4e Eig %.3e %.3e 1stv %.3f %.3f tan %.3f\n",(int) age,k1,l1,k2,l2,mu1,mu2,v1,v2,c12,cv12,lc1,lc2,v11,v21,tnalp);
                    /*printf(fignu*/
                    /* mu1+ v11*lc1*cost + v12*lc2*sin(t) */
                    /* mu2+ v21*lc1*cost + v22*lc2*sin(t) */
                    if(first==1){
                      first=0;
                      fprintf(ficgp,"\n# Ellipsoids of confidence\n#\n");
                      fprintf(ficgp,"\nset parametric;unset label");
                      fprintf(ficgp,"\nset log y;set log x; set xlabel \"p%1d%1d (year-1)\";set ylabel \"p%1d%1d (year-1)\"",k1,l1,k2,l2);
                      fprintf(ficgp,"\nset ter svg size 640, 480");
                      fprintf(fichtmcov,"\n<p><br>Ellipsoids of confidence cov(p%1d%1d,p%1d%1d) expressed in year<sup>-1</sup>\
    :<a href=\"%s_%d%1d%1d-%1d%1d.svg\">                                                                                                                                           \
   %s_%d%1d%1d-%1d%1d.svg</A>, ",k1,l1,k2,l2,\
                              subdirf2(optionfilefiname,"VARPIJGR_"), j1,k1,l1,k2,l2,      \
                              subdirf2(optionfilefiname,"VARPIJGR_"), j1,k1,l1,k2,l2);
                      fprintf(fichtmcov,"\n<br><img src=\"%s_%d%1d%1d-%1d%1d.svg\"> ",subdirf2(optionfilefiname,"VARPIJGR_"), j1,k1,l1,k2,l2);
                      fprintf(fichtmcov,"\n<br> Correlation at age %d (%.3f),",(int) age, c12);
                      fprintf(ficgp,"\nset out \"%s_%d%1d%1d-%1d%1d.svg\"",subdirf2(optionfilefiname,"VARPIJGR_"), j1,k1,l1,k2,l2);
                      fprintf(ficgp,"\nset label \"%d\" at %11.3e,%11.3e center",(int) age, mu1,mu2);
                      fprintf(ficgp,"\n# Age %d, p%1d%1d - p%1d%1d",(int) age, k1,l1,k2,l2);
                      fprintf(ficgp,"\nplot [-pi:pi] %11.3e+ %.3f*(%11.3e*%11.3e*cos(t)+%11.3e*%11.3e*sin(t)), %11.3e +%.3f*(%11.3e*%11.3e*cos(t)+%11.3e*%11.3e*sin(t)) not",      \
                              mu1,std,v11,sqrt(fabs(lc1)),v12,sqrt(fabs(lc2)), \
                              mu2,std,v21,sqrt(fabs(lc1)),v22,sqrt(fabs(lc2))); /* For gnuplot only */
                    }else{
                      first=0;
                      fprintf(fichtmcov," %d (%.3f),",(int) age, c12);
                      fprintf(ficgp,"\n# Age %d, p%1d%1d - p%1d%1d",(int) age, k1,l1,k2,l2);
                      fprintf(ficgp,"\nset label \"%d\" at %11.3e,%11.3e center",(int) age, mu1,mu2);
                      fprintf(ficgp,"\nreplot %11.3e+ %.3f*(%11.3e*%11.3e*cos(t)+%11.3e*%11.3e*sin(t)), %11.3e +%.3f*(%11.3e*%11.3e*cos(t)+%11.3e*%11.3e*sin(t)) not", \
                              mu1,std,v11,sqrt(lc1),v12,sqrt(fabs(lc2)),   \
                              mu2,std,v21,sqrt(lc1),v22,sqrt(fabs(lc2)));
                    }/* if first */
                  } /* age mod 5 */
                } /* end loop age */
                fprintf(ficgp,"\nset out;\nset out \"%s_%d%1d%1d-%1d%1d.svg\";replot;set out;",subdirf2(optionfilefiname,"VARPIJGR_"), j1,k1,l1,k2,l2);
                first=1;
              } /*l12 */
            } /* k12 */
          } /*l1 */
        }/* k1 */
      }  /* loop on combination of covariates j1 */
      } /* loop on nres */
      free_ma3x(varpij,1,nlstate,1,nlstate+ndeath,(int) bage, (int)fage);
      free_matrix(mu,1,(nlstate+ndeath)*(nlstate+ndeath),(int) bage, (int)fage);
      free_matrix(doldm,1,(nlstate)*(nlstate+ndeath),1,(nlstate)*(nlstate+ndeath));
      free_matrix(dnewm,1,(nlstate)*(nlstate+ndeath),1,npar);
      free_vector(xp,1,npar);
      fclose(ficresprob);
      fclose(ficresprobcov);
      fclose(ficresprobcor);
      fflush(ficgp);
      fflush(fichtmcov);
  }   }
   
  ij=1;  
  for (i=1; i<= maxncov; i++) {  /******************* Printing html file ***********/
    if((Ndum[i]!=0) && (i<=ncovcol)){  void printinghtml(char fileresu[], char title[], char datafile[], int firstpass, \
      Tvaraff[ij]=i; /*For printing */                    int lastpass, int stepm, int weightopt, char model[],\
      ij++;                    int imx,int jmin, int jmax, double jmeanint,char rfileres[],\
                     int popforecast, int mobilav, int prevfcast, int mobilavproj, int prevbcast, int estepm , \
                     double jprev1, double mprev1,double anprev1, double dateprev1, double dateprojd, double dateback1, \
                     double jprev2, double mprev2,double anprev2, double dateprev2, double dateprojf, double dateback2){
     int jj1, k1, i1, cpt, k4, nres;
     /* In fact some results are already printed in fichtm which is open */
      fprintf(fichtm,"<ul><li><a href='#firstorder'>Result files (first order: no variance)</a>\n \
      <li><a href='#secondorder'>Result files (second order (variance)</a>\n \
   </ul>");
   /*    fprintf(fichtm,"<ul><li> model=1+age+%s\n \ */
   /* </ul>", model); */
      fprintf(fichtm,"<ul><li><h4><a name='firstorder'>Result files (first order: no variance)</a></h4>\n");
      fprintf(fichtm,"<li>- Observed frequency between two states (during the period defined between %.lf/%.lf/%.lf and %.lf/%.lf/%.lf): <a href=\"%s\">%s</a> (html file)<br/>\n",
              jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,subdirfext3(optionfilefiname,"PHTMFR_",".htm"),subdirfext3(optionfilefiname,"PHTMFR_",".htm"));
      fprintf(fichtm,"<li> - Observed prevalence (cross-sectional prevalence) in each state (during the period defined between %.lf/%.lf/%.lf and %.lf/%.lf/%.lf): <a href=\"%s\">%s</a> (html file) ",
              jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,subdirfext3(optionfilefiname,"PHTM_",".htm"),subdirfext3(optionfilefiname,"PHTM_",".htm"));
      fprintf(fichtm,",  <a href=\"%s\">%s</a> (text file) <br>\n",subdirf2(fileresu,"P_"),subdirf2(fileresu,"P_"));
      fprintf(fichtm,"\
    - Estimated transition probabilities over %d (stepm) months: <a href=\"%s\">%s</a><br>\n ",
              stepm,subdirf2(fileresu,"PIJ_"),subdirf2(fileresu,"PIJ_"));
      fprintf(fichtm,"\
    - Estimated back transition probabilities over %d (stepm) months: <a href=\"%s\">%s</a><br>\n ",
              stepm,subdirf2(fileresu,"PIJB_"),subdirf2(fileresu,"PIJB_"));
      fprintf(fichtm,"\
    - Period (forward) prevalence in each health state: <a href=\"%s\">%s</a> <br>\n",
              subdirf2(fileresu,"PL_"),subdirf2(fileresu,"PL_"));
      fprintf(fichtm,"\
    - Backward prevalence in each health state: <a href=\"%s\">%s</a> <br>\n",
              subdirf2(fileresu,"PLB_"),subdirf2(fileresu,"PLB_"));
      fprintf(fichtm,"\
    - (a) Life expectancies by health status at initial age, e<sub>i.</sub> (b) health expectancies by health status at initial age, e<sub>ij</sub> . If one or more covariates are included, specific tables for each value of the covariate are output in sequences within the same file (estepm=%2d months): \
      <a href=\"%s\">%s</a> <br>\n",
              estepm,subdirf2(fileresu,"E_"),subdirf2(fileresu,"E_"));
      if(prevfcast==1){
        fprintf(fichtm,"\
    - Prevalence projections by age and states:                            \
      <a href=\"%s\">%s</a> <br>\n</li>", subdirf2(fileresu,"F_"),subdirf2(fileresu,"F_"));
    }     }
  }  
    
  cptcoveff=ij-1; /*Number of simple covariates*/  
 }  
   
 /*********** Health Expectancies ****************/  
   
 void evsij(char fileres[], double ***eij, double x[], int nlstate, int stepm, int bage, int fage, double **oldm, double **savm, int ij, int estepm,double delti[],double **matcov )     m=pow(2,cptcoveff);
      if (cptcovn < 1) {m=1;ncodemax[1]=1;}
   
 {     fprintf(fichtm," \n<ul><li><b>Graphs (first order)</b></li><p>");
   /* Health expectancies */  
   int i, j, nhstepm, hstepm, h, nstepm, k, cptj;  
   double age, agelim, hf;  
   double ***p3mat,***varhe;  
   double **dnewm,**doldm;  
   double *xp;  
   double **gp, **gm;  
   double ***gradg, ***trgradg;  
   int theta;  
   
   varhe=ma3x(1,nlstate*nlstate,1,nlstate*nlstate,(int) bage, (int) fage);     jj1=0;
   xp=vector(1,npar);  
   dnewm=matrix(1,nlstate*nlstate,1,npar);  
   doldm=matrix(1,nlstate*nlstate,1,nlstate*nlstate);  
     
   fprintf(ficreseij,"# Health expectancies\n");  
   fprintf(ficreseij,"# Age");  
   for(i=1; i<=nlstate;i++)  
     for(j=1; j<=nlstate;j++)  
       fprintf(ficreseij," %1d-%1d (SE)",i,j);  
   fprintf(ficreseij,"\n");  
   
   if(estepm < stepm){     fprintf(fichtm," \n<ul>");
     printf ("Problem %d lower than %d\n",estepm, stepm);     for(nres=1; nres <= nresult; nres++) /* For each resultline */
   }     for(k1=1; k1<=m;k1++){ /* For each combination of covariate */
   else  hstepm=estepm;          if(m != 1 && TKresult[nres]!= k1)
   /* We compute the life expectancy from trapezoids spaced every estepm months         continue;
    * This is mainly to measure the difference between two models: for example       jj1++;
    * if stepm=24 months pijx are given only every 2 years and by summing them       if (cptcovn > 0) {
    * we are calculating an estimate of the Life Expectancy assuming a linear          fprintf(fichtm,"\n<li><a  size=\"1\" color=\"#EC5E5E\" href=\"#rescov");
    * progression in between and thus overestimating or underestimating according         for (cpt=1; cpt<=cptcoveff;cpt++){ 
    * to the curvature of the survival function. If, for the same date, we            fprintf(fichtm,"_V%d=%d_",Tvresult[nres][cpt],(int)Tresult[nres][cpt]);
    * estimate the model with stepm=1 month, we can keep estepm to 24 months         }
    * to compare the new estimate of Life expectancy with the same linear          for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
    * hypothesis. A more precise result, taking into account a more precise           fprintf(fichtm,"_V%d=%f_",Tvqresult[nres][k4],Tqresult[nres][k4]);
    * curvature will be obtained if estepm is as small as stepm. */         }
          fprintf(fichtm,"\">");
          
          /* if(nqfveff+nqtveff 0) */ /* Test to be done */
          fprintf(fichtm,"************ Results for covariates");
          for (cpt=1; cpt<=cptcoveff;cpt++){ 
            fprintf(fichtm," V%d=%d ",Tvresult[nres][cpt],(int)Tresult[nres][cpt]);
          }
          for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
            fprintf(fichtm," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
          }
          if(invalidvarcomb[k1]){
            fprintf(fichtm," Warning Combination (%d) ignored because no cases ",k1); 
            continue;
          }
          fprintf(fichtm,"</a></li>");
        } /* cptcovn >0 */
      }
      fprintf(fichtm," \n</ul>");
   
   /* For example we decided to compute the life expectancy with the smallest unit */     jj1=0;
   /* hstepm beeing the number of stepms, if hstepm=1 the length of hstepm is stepm.   
      nhstepm is the number of hstepm from age to agelim   
      nstepm is the number of stepm from age to agelin.   
      Look at hpijx to understand the reason of that which relies in memory size  
      and note for a fixed period like estepm months */  
   /* We decided (b) to get a life expectancy respecting the most precise curvature of the  
      survival function given by stepm (the optimization length). Unfortunately it  
      means that if the survival funtion is printed only each two years of age and if  
      you sum them up and add 1 year (area under the trapezoids) you won't get the same   
      results. So we changed our mind and took the option of the best precision.  
   */  
   hstepm=hstepm/stepm; /* Typically in stepm units, if stepm=6 & estepm=24 , = 24/6 months = 4 */   
   
   agelim=AGESUP;     for(nres=1; nres <= nresult; nres++) /* For each resultline */
   for (age=bage; age<=fage; age ++){ /* If stepm=6 months */     for(k1=1; k1<=m;k1++){ /* For each combination of covariate */
     /* nhstepm age range expressed in number of stepm */       if(m != 1 && TKresult[nres]!= k1)
     nstepm=(int) rint((agelim-age)*YEARM/stepm);          continue;
     /* Typically if 20 years nstepm = 20*12/6=40 stepm */   
     /* if (stepm >= YEARM) hstepm=1;*/  
     nhstepm = nstepm/hstepm;/* Expressed in hstepm, typically nhstepm=40/4=10 */  
     p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);  
     gradg=ma3x(0,nhstepm,1,npar,1,nlstate*nlstate);  
     gp=matrix(0,nhstepm,1,nlstate*nlstate);  
     gm=matrix(0,nhstepm,1,nlstate*nlstate);  
   
     /* Computed by stepm unit matrices, product of hstepm matrices, stored       /* for(i1=1; i1<=ncodemax[k1];i1++){ */
        in an array of nhstepm length: nhstepm=10, hstepm=4, stepm=6 months */       jj1++;
     hpxij(p3mat,nhstepm,age,hstepm,x,nlstate,stepm,oldm, savm, ij);         if (cptcovn > 0) {
          fprintf(fichtm,"\n<p><a name=\"rescov");
          for (cpt=1; cpt<=cptcoveff;cpt++){ 
            fprintf(fichtm,"_V%d=%d_",Tvresult[nres][cpt],(int)Tresult[nres][cpt]);
          }
          for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
            fprintf(fichtm,"_V%d=%f_",Tvqresult[nres][k4],Tqresult[nres][k4]);
          }
          fprintf(fichtm,"\"</a>");
     
          fprintf(fichtm,"<hr  size=\"2\" color=\"#EC5E5E\">************ Results for covariates");
          for (cpt=1; cpt<=cptcoveff;cpt++){ 
            fprintf(fichtm," V%d=%d ",Tvresult[nres][cpt],(int)Tresult[nres][cpt]);
            printf(" V%d=%d ",Tvresult[nres][cpt],Tresult[nres][cpt]);fflush(stdout);
            /* fprintf(fichtm," V%d=%d ",Tvaraff[cpt],nbcode[Tvaraff[cpt]][codtabm(jj1,cpt)]); */
            /* printf(" V%d=%d ",Tvaraff[cpt],nbcode[Tvaraff[cpt]][codtabm(jj1,cpt)]);fflush(stdout); */
          }
          for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
           fprintf(fichtm," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
           printf(" V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);fflush(stdout);
         }
          
          /* if(nqfveff+nqtveff 0) */ /* Test to be done */
          fprintf(fichtm," (model=%s) ************\n<hr size=\"2\" color=\"#EC5E5E\">",model);
          if(invalidvarcomb[k1]){
            fprintf(fichtm,"\n<h3>Combination (%d) ignored because no cases </h3>\n",k1); 
            printf("\nCombination (%d) ignored because no cases \n",k1); 
            continue;
          }
        }
        /* aij, bij */
        fprintf(fichtm,"<br>- Logit model (yours is: logit(pij)=log(pij/pii)= aij+ bij age+%s) as a function of age: <a href=\"%s_%d-1-%d.svg\">%s_%d-1-%d.svg</a><br> \
   <img src=\"%s_%d-1-%d.svg\">",model,subdirf2(optionfilefiname,"PE_"),k1,nres,subdirf2(optionfilefiname,"PE_"),k1,nres,subdirf2(optionfilefiname,"PE_"),k1,nres);
        /* Pij */
        fprintf(fichtm,"<br>\n- P<sub>ij</sub> or conditional probabilities to be observed in state j being in state i, %d (stepm) months before: <a href=\"%s_%d-2-%d.svg\">%s_%d-2-%d.svg</a><br> \
   <img src=\"%s_%d-2-%d.svg\">",stepm,subdirf2(optionfilefiname,"PE_"),k1,nres,subdirf2(optionfilefiname,"PE_"),k1,nres,subdirf2(optionfilefiname,"PE_"),k1,nres);     
        /* Quasi-incidences */
        fprintf(fichtm,"<br>\n- I<sub>ij</sub> or Conditional probabilities to be observed in state j being in state i %d (stepm) months\
    before but expressed in per year i.e. quasi incidences if stepm is small and probabilities too, \
    incidence (rates) are the limit when h tends to zero of the ratio of the probability  <sub>h</sub>P<sub>ij</sub> \
   divided by h: <sub>h</sub>P<sub>ij</sub>/h : <a href=\"%s_%d-3-%d.svg\">%s_%d-3-%d.svg</a><br> \
   <img src=\"%s_%d-3-%d.svg\">",stepm,subdirf2(optionfilefiname,"PE_"),k1,nres,subdirf2(optionfilefiname,"PE_"),k1,nres,subdirf2(optionfilefiname,"PE_"),k1,nres); 
        /* Survival functions (period) in state j */
        for(cpt=1; cpt<=nlstate;cpt++){
          fprintf(fichtm,"<br>\n- Survival functions in state %d. And probability to be observed in state %d being in state (1 to %d) at different ages. <a href=\"%s_%d-%d-%d.svg\">%s_%d-%d-%d.svg</a><br>", cpt, cpt, nlstate, subdirf2(optionfilefiname,"LIJ_"),cpt,k1,nres,subdirf2(optionfilefiname,"LIJ_"),cpt,k1,nres);
          fprintf(fichtm," (data from text file  <a href=\"%s.txt\">%s.txt</a>)\n<br>",subdirf2(optionfilefiname,"PIJ_"),subdirf2(optionfilefiname,"PIJ_"));
          fprintf(fichtm,"<img src=\"%s_%d-%d-%d.svg\">",subdirf2(optionfilefiname,"LIJ_"),cpt,k1,nres);
        }
        /* State specific survival functions (period) */
        for(cpt=1; cpt<=nlstate;cpt++){
          fprintf(fichtm,"<br>\n- Survival functions in state %d and in any other live state (total).\
    And probability to be observed in various states (up to %d) being in state %d at different ages.       \
    <a href=\"%s_%d-%d-%d.svg\">%s_%d-%d-%d.svg</a><br> ", cpt, nlstate, cpt, subdirf2(optionfilefiname,"LIJT_"),cpt,k1,nres,subdirf2(optionfilefiname,"LIJT_"),cpt,k1,nres);
          fprintf(fichtm," (data from text file  <a href=\"%s.txt\">%s.txt</a>)\n<br>",subdirf2(optionfilefiname,"PIJ_"),subdirf2(optionfilefiname,"PIJ_"));
          fprintf(fichtm,"<img src=\"%s_%d-%d-%d.svg\">",subdirf2(optionfilefiname,"LIJT_"),cpt,k1,nres);
        }
        /* Period (forward stable) prevalence in each health state */
        for(cpt=1; cpt<=nlstate;cpt++){
          fprintf(fichtm,"<br>\n- Convergence to period (stable) prevalence in state %d. Or probability for a person being in state (1 to %d) at different ages, to be in state %d some years after. <a href=\"%s_%d-%d-%d.svg\">%s_%d-%d-%d.svg</a><br>", cpt, nlstate, cpt, subdirf2(optionfilefiname,"P_"),cpt,k1,nres,subdirf2(optionfilefiname,"P_"),cpt,k1,nres);
          fprintf(fichtm," (data from text file  <a href=\"%s.txt\">%s.txt</a>)\n<br>",subdirf2(optionfilefiname,"P_"),subdirf2(optionfilefiname,"P_"));
         fprintf(fichtm,"<img src=\"%s_%d-%d-%d.svg\">" ,subdirf2(optionfilefiname,"P_"),cpt,k1,nres);
        }
        if(prevbcast==1){
          /* Backward prevalence in each health state */
          for(cpt=1; cpt<=nlstate;cpt++){
            fprintf(fichtm,"<br>\n- Convergence to mixed (stable) back prevalence in state %d. Or probability for a person to be in state %d at a younger age, knowing that she/he was in state (1 to %d) at different older ages. <a href=\"%s_%d-%d-%d.svg\">%s_%d-%d-%d.svg</a><br> \
   <img src=\"%s_%d-%d-%d.svg\">", cpt, cpt, nlstate, subdirf2(optionfilefiname,"PB_"),cpt,k1,nres,subdirf2(optionfilefiname,"PB_"),cpt,k1,nres,subdirf2(optionfilefiname,"PB_"),cpt,k1,nres);
          }
        }
        if(prevfcast==1){
          /* Projection of prevalence up to period (forward stable) prevalence in each health state */
          for(cpt=1; cpt<=nlstate;cpt++){
            fprintf(fichtm,"<br>\n- Projection of cross-sectional prevalence (estimated with cases observed from %.1f to %.1f and mobil_average=%d), from year %.1f up to year %.1f tending to period (stable) forward prevalence in state %d. Or probability to be in state %d being in an observed weighted state (from 1 to %d). <a href=\"%s_%d-%d-%d.svg\">%s_%d-%d-%d.svg</a>", dateprev1, dateprev2, mobilavproj, dateprojd, dateprojf, cpt, cpt, nlstate, subdirf2(optionfilefiname,"PROJ_"),cpt,k1,nres,subdirf2(optionfilefiname,"PROJ_"),cpt,k1,nres);
            fprintf(fichtm," (data from text file  <a href=\"%s.txt\">%s.txt</a>)\n<br>",subdirf2(optionfilefiname,"F_"),subdirf2(optionfilefiname,"F_"));
            fprintf(fichtm,"<img src=\"%s_%d-%d-%d.svg\">",
                    subdirf2(optionfilefiname,"PROJ_"),cpt,k1,nres);
          }
        }
        if(prevbcast==1){
         /* Back projection of prevalence up to stable (mixed) back-prevalence in each health state */
          for(cpt=1; cpt<=nlstate;cpt++){
            fprintf(fichtm,"<br>\n- Back projection of cross-sectional prevalence (estimated with cases observed from %.1f to %.1f and mobil_average=%d), \
    from year %.1f up to year %.1f (probably close to stable [mixed] back prevalence in state %d (randomness in cross-sectional prevalence is not taken into \
    account but can visually be appreciated). Or probability to have been in an state %d, knowing that the person was in either state (1 or %d) \
   with weights corresponding to observed prevalence at different ages. <a href=\"%s_%d-%d-%d.svg\">%s_%d-%d-%d.svg</a>", dateprev1, dateprev2, mobilavproj, dateback1, dateback2, cpt, cpt, nlstate, subdirf2(optionfilefiname,"PROJB_"),cpt,k1,nres,subdirf2(optionfilefiname,"PROJB_"),cpt,k1,nres);
            fprintf(fichtm," (data from text file  <a href=\"%s.txt\">%s.txt</a>)\n<br>",subdirf2(optionfilefiname,"FB_"),subdirf2(optionfilefiname,"FB_"));
            fprintf(fichtm," <img src=\"%s_%d-%d-%d.svg\">", subdirf2(optionfilefiname,"PROJB_"),cpt,k1,nres);
          }
        }
            
        for(cpt=1; cpt<=nlstate;cpt++) {
          fprintf(fichtm,"\n<br>- Life expectancy by health state (%d) at initial age and its decomposition into health expectancies in each alive state (1 to %d) (or area under each survival functions): <a href=\"%s_%d-%d-%d.svg\">%s_%d-%d-%d.svg</a>",cpt,nlstate,subdirf2(optionfilefiname,"EXP_"),cpt,k1,nres,subdirf2(optionfilefiname,"EXP_"),cpt,k1,nres);
          fprintf(fichtm," (data from text file  <a href=\"%s.txt\"> %s.txt</a>)\n<br>",subdirf2(optionfilefiname,"E_"),subdirf2(optionfilefiname,"E_"));
          fprintf(fichtm,"<img src=\"%s_%d-%d-%d.svg\">", subdirf2(optionfilefiname,"EXP_"),cpt,k1,nres );
        }
        /* } /\* end i1 *\/ */
      }/* End k1 */
      fprintf(fichtm,"</ul>");
   
      fprintf(fichtm,"\
   \n<br><li><h4> <a name='secondorder'>Result files (second order: variances)</a></h4>\n\
    - Parameter file with estimated parameters and covariance matrix: <a href=\"%s\">%s</a> <br> \
    - 95%% confidence intervals and Wald tests of the estimated parameters are in the log file if optimization has been done (mle != 0).<br> \
   But because parameters are usually highly correlated (a higher incidence of disability \
   and a higher incidence of recovery can give very close observed transition) it might \
   be very useful to look not only at linear confidence intervals estimated from the \
   variances but at the covariance matrix. And instead of looking at the estimated coefficients \
   (parameters) of the logistic regression, it might be more meaningful to visualize the \
   covariance matrix of the one-step probabilities. \
   See page 'Matrix of variance-covariance of one-step probabilities' below. \n", rfileres,rfileres);
   
      fprintf(fichtm," - Standard deviation of one-step probabilities: <a href=\"%s\">%s</a> <br>\n",
              subdirf2(fileresu,"PROB_"),subdirf2(fileresu,"PROB_"));
      fprintf(fichtm,"\
    - Variance-covariance of one-step probabilities: <a href=\"%s\">%s</a> <br>\n",
              subdirf2(fileresu,"PROBCOV_"),subdirf2(fileresu,"PROBCOV_"));
   
      fprintf(fichtm,"\
    - Correlation matrix of one-step probabilities: <a href=\"%s\">%s</a> <br>\n",
              subdirf2(fileresu,"PROBCOR_"),subdirf2(fileresu,"PROBCOR_"));
      fprintf(fichtm,"\
    - Variances and covariances of health expectancies by age and <b>initial health status</b> (cov(e<sup>ij</sup>,e<sup>kl</sup>)(estepm=%2d months): \
      <a href=\"%s\">%s</a> <br>\n</li>",
              estepm,subdirf2(fileresu,"CVE_"),subdirf2(fileresu,"CVE_"));
      fprintf(fichtm,"\
    - (a) Health expectancies by health status at initial age (e<sup>ij</sup>) and standard errors (in parentheses) (b) life expectancies and standard errors (e<sup>i.</sup>=e<sup>i1</sup>+e<sup>i2</sup>+...)(estepm=%2d months): \
      <a href=\"%s\">%s</a> <br>\n</li>",
              estepm,subdirf2(fileresu,"STDE_"),subdirf2(fileresu,"STDE_"));
      fprintf(fichtm,"\
    - Variances and covariances of health expectancies by age. Status (i) based health expectancies (in state j), e<sup>ij</sup> are weighted by the forward (period) prevalences in each state i (if popbased=1, an additional computation is done using the cross-sectional prevalences, i.e population based) (estepm=%d months): <a href=\"%s\">%s</a><br>\n",
              estepm, subdirf2(fileresu,"V_"),subdirf2(fileresu,"V_"));
      fprintf(fichtm,"\
    - Total life expectancy and total health expectancies to be spent in each health state e<sup>.j</sup> with their standard errors (if popbased=1, an additional computation is done using the cross-sectional prevalences, i.e population based) (estepm=%d months): <a href=\"%s\">%s</a> <br>\n",
              estepm, subdirf2(fileresu,"T_"),subdirf2(fileresu,"T_"));
      fprintf(fichtm,"\
    - Standard deviation of forward (period) prevalences: <a href=\"%s\">%s</a> <br>\n",\
              subdirf2(fileresu,"VPL_"),subdirf2(fileresu,"VPL_"));
   
     hf=hstepm*stepm/YEARM;  /* Duration of hstepm expressed in year unit. */  /*  if(popforecast==1) fprintf(fichtm,"\n */
   /*  - Prevalences forecasting: <a href=\"f%s\">f%s</a> <br>\n */
   /*  - Population forecasting (if popforecast=1): <a href=\"pop%s\">pop%s</a> <br>\n */
   /*      <br>",fileres,fileres,fileres,fileres); */
   /*  else  */
   /*    fprintf(fichtm,"\n No population forecast: popforecast = %d (instead of 1) or stepm = %d (instead of 1) or model=%s (instead of .)<br><br></li>\n",popforecast, stepm, model); */
      fflush(fichtm);
   
     /* Computing Variances of health expectancies */     m=pow(2,cptcoveff);
      if (cptcovn < 1) {m=1;ncodemax[1]=1;}
   
      for(theta=1; theta <=npar; theta++){     fprintf(fichtm," <ul><li><b>Graphs (second order)</b></li><p>");
       for(i=1; i<=npar; i++){   
         xp[i] = x[i] + (i==theta ?delti[theta]:0);    jj1=0;
   
      fprintf(fichtm," \n<ul>");
      for(nres=1; nres <= nresult; nres++) /* For each resultline */
      for(k1=1; k1<=m;k1++){ /* For each combination of covariate */
        if(m != 1 && TKresult[nres]!= k1)
          continue;
        jj1++;
        if (cptcovn > 0) {
          fprintf(fichtm,"\n<li><a  size=\"1\" color=\"#EC5E5E\" href=\"#rescovsecond");
          for (cpt=1; cpt<=cptcoveff;cpt++){ 
            fprintf(fichtm,"_V%d=%d_",Tvresult[nres][cpt],(int)Tresult[nres][cpt]);
          }
          for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
            fprintf(fichtm,"_V%d=%f_",Tvqresult[nres][k4],Tqresult[nres][k4]);
          }
          fprintf(fichtm,"\">");
          
          /* if(nqfveff+nqtveff 0) */ /* Test to be done */
          fprintf(fichtm,"************ Results for covariates");
          for (cpt=1; cpt<=cptcoveff;cpt++){ 
            fprintf(fichtm," V%d=%d ",Tvresult[nres][cpt],(int)Tresult[nres][cpt]);
          }
          for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
            fprintf(fichtm," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
          }
          if(invalidvarcomb[k1]){
            fprintf(fichtm," Warning Combination (%d) ignored because no cases ",k1); 
            continue;
          }
          fprintf(fichtm,"</a></li>");
        } /* cptcovn >0 */
      }
      fprintf(fichtm," \n</ul>");
   
      jj1=0;
   
      for(nres=1; nres <= nresult; nres++){ /* For each resultline */
      for(k1=1; k1<=m;k1++){
        if(m != 1 && TKresult[nres]!= k1)
          continue;
        /* for(i1=1; i1<=ncodemax[k1];i1++){ */
        jj1++;
        if (cptcovn > 0) {
          fprintf(fichtm,"\n<p><a name=\"rescovsecond");
          for (cpt=1; cpt<=cptcoveff;cpt++){ 
            fprintf(fichtm,"_V%d=%d_",Tvresult[nres][cpt],(int)Tresult[nres][cpt]);
          }
          for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
            fprintf(fichtm,"_V%d=%f_",Tvqresult[nres][k4],Tqresult[nres][k4]);
          }
          fprintf(fichtm,"\"</a>");
          
          fprintf(fichtm,"<hr  size=\"2\" color=\"#EC5E5E\">************ Results for covariates");
          for (cpt=1; cpt<=cptcoveff;cpt++){  /**< cptcoveff number of variables */
            fprintf(fichtm," V%d=%d ",Tvresult[nres][cpt],Tresult[nres][cpt]);
            printf(" V%d=%d ",Tvresult[nres][cpt],Tresult[nres][cpt]);fflush(stdout);
            /* fprintf(fichtm," V%d=%d ",Tvaraff[cpt],nbcode[Tvaraff[cpt]][codtabm(jj1,cpt)]); */
          }
          for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
           fprintf(fichtm," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
       }        }
       hpxij(p3mat,nhstepm,age,hstepm,xp,nlstate,stepm,oldm,savm, ij);    
          fprintf(fichtm," (model=%s) ************\n<hr size=\"2\" color=\"#EC5E5E\">",model);
   
          if(invalidvarcomb[k1]){
            fprintf(fichtm,"\n<h4>Combination (%d) ignored because no cases </h4>\n",k1); 
            continue;
          }
        }
        for(cpt=1; cpt<=nlstate;cpt++) {
          fprintf(fichtm,"\n<br>- Observed (cross-sectional with mov_average=%d) and period (incidence based) \
   prevalence (with 95%% confidence interval) in state (%d): <a href=\"%s_%d-%d-%d.svg\"> %s_%d-%d-%d.svg</a>",mobilav,cpt,subdirf2(optionfilefiname,"V_"),cpt,k1,nres,subdirf2(optionfilefiname,"V_"),cpt,k1,nres);
          fprintf(fichtm," (data from text file  <a href=\"%s\">%s</a>)\n <br>",subdirf2(fileresu,"VPL_"),subdirf2(fileresu,"VPL_"));
          fprintf(fichtm,"<img src=\"%s_%d-%d-%d.svg\">",subdirf2(optionfilefiname,"V_"), cpt,k1,nres);
        }
        fprintf(fichtm,"\n<br>- Total life expectancy by age and \
   health expectancies in each live states (1 to %d). If popbased=1 the smooth (due to the model) \
   true period expectancies (those weighted with period prevalences are also\
    drawn in addition to the population based expectancies computed using\
    observed and cahotic prevalences:  <a href=\"%s_%d-%d.svg\">%s_%d-%d.svg</a>",nlstate, subdirf2(optionfilefiname,"E_"),k1,nres,subdirf2(optionfilefiname,"E_"),k1,nres);
        fprintf(fichtm," (data from text file <a href=\"%s.txt\">%s.txt</a>) \n<br>",subdirf2(optionfilefiname,"T_"),subdirf2(optionfilefiname,"T_"));
        fprintf(fichtm,"<img src=\"%s_%d-%d.svg\">",subdirf2(optionfilefiname,"E_"),k1,nres);
        /* } /\* end i1 *\/ */
      }/* End k1 */
     }/* End nres */
      fprintf(fichtm,"</ul>");
      fflush(fichtm);
   }
   
   /******************* Gnuplot file **************/
   void printinggnuplot(char fileresu[], char optionfilefiname[], double ageminpar, double agemaxpar, double bage, double fage , int prevfcast, int prevbcast, char pathc[], double p[], int offyear, int offbyear){
   
     char dirfileres[132],optfileres[132];
     char gplotcondition[132], gplotlabel[132];
     int cpt=0,k1=0,i=0,k=0,j=0,jk=0,k2=0,k3=0,k4=0,ij=0, ijp=0, l=0;
     int lv=0, vlv=0, kl=0;
     int ng=0;
     int vpopbased;
     int ioffset; /* variable offset for columns */
     int iyearc=1; /* variable column for year of projection  */
     int iagec=1; /* variable column for age of projection  */
     int nres=0; /* Index of resultline */
     int istart=1; /* For starting graphs in projections */
   
   /*   if((ficgp=fopen(optionfilegnuplot,"a"))==NULL) { */
   /*     printf("Problem with file %s",optionfilegnuplot); */
   /*     fprintf(ficlog,"Problem with file %s",optionfilegnuplot); */
   /*   } */
   
     /*#ifdef windows */
     fprintf(ficgp,"cd \"%s\" \n",pathc);
     /*#endif */
     m=pow(2,cptcoveff);
   
     /* diagram of the model */
     fprintf(ficgp,"\n#Diagram of the model \n");
     fprintf(ficgp,"\ndelta=0.03;delta2=0.07;unset arrow;\n");
     fprintf(ficgp,"yoff=(%d > 2? 0:1);\n",nlstate);
     fprintf(ficgp,"\n#Peripheral arrows\nset for [i=1:%d] for [j=1:%d] arrow i*10+j from cos(pi*((1-(%d/2)*2./%d)/2+(i-1)*2./%d))-(i!=j?(i-j)/abs(i-j)*delta:0), yoff +sin(pi*((1-(%d/2)*2./%d)/2+(i-1)*2./%d)) + (i!=j?(i-j)/abs(i-j)*delta:0) rto -0.95*(cos(pi*((1-(%d/2)*2./%d)/2+(i-1)*2./%d))+(i!=j?(i-j)/abs(i-j)*delta:0) - cos(pi*((1-(%d/2)*2./%d)/2+(j-1)*2./%d)) + (i!=j?(i-j)/abs(i-j)*delta2:0)), -0.95*(sin(pi*((1-(%d/2)*2./%d)/2+(i-1)*2./%d)) + (i!=j?(i-j)/abs(i-j)*delta:0) - sin(pi*((1-(%d/2)*2./%d)/2+(j-1)*2./%d))+( i!=j?(i-j)/abs(i-j)*delta2:0)) ls (i < j? 1:2)\n",nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate);
   
     fprintf(ficgp,"\n#Centripete arrows (turning in other direction (1-i) instead of (i-1)) \nset for [i=1:%d] arrow (%d+1)*10+i from cos(pi*((1-(%d/2)*2./%d)/2+(1-i)*2./%d))-(i!=j?(i-j)/abs(i-j)*delta:0), yoff +sin(pi*((1-(%d/2)*2./%d)/2+(1-i)*2./%d)) + (i!=j?(i-j)/abs(i-j)*delta:0) rto -0.80*(cos(pi*((1-(%d/2)*2./%d)/2+(1-i)*2./%d))+(i!=j?(i-j)/abs(i-j)*delta:0)  ), -0.80*(sin(pi*((1-(%d/2)*2./%d)/2+(1-i)*2./%d)) + (i!=j?(i-j)/abs(i-j)*delta:0) + yoff ) ls 4\n",nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate);
     fprintf(ficgp,"\n#show arrow\nunset label\n");
     fprintf(ficgp,"\n#States labels, starting from 2 (2-i) instead of (1-i), was (i-1)\nset for [i=1:%d] label i sprintf(\"State %%d\",i) center at cos(pi*((1-(%d/2)*2./%d)/2+(2-i)*2./%d)), yoff+sin(pi*((1-(%d/2)*2./%d)/2+(2-i)*2./%d)) font \"helvetica, 16\" tc rgbcolor \"blue\"\n",nlstate,nlstate,nlstate,nlstate,nlstate,nlstate,nlstate);
     fprintf(ficgp,"\nset label %d+1 sprintf(\"State %%d\",%d+1) center at 0.,0.  font \"helvetica, 16\" tc rgbcolor \"red\"\n",nlstate,nlstate);
     fprintf(ficgp,"\n#show label\nunset border;unset xtics; unset ytics;\n");
     fprintf(ficgp,"\n\nset ter svg size 640, 480;set out \"%s_.svg\" \n",subdirf2(optionfilefiname,"D_"));
     fprintf(ficgp,"unset log y; plot [-1.2:1.2][yoff-1.2:1.2] 1/0 not; set out;reset;\n");
   
     /* Contribution to likelihood */
     /* Plot the probability implied in the likelihood */
     fprintf(ficgp,"\n# Contributions to the Likelihood, mle >=1. For mle=4 no interpolation, pure matrix products.\n#\n");
     fprintf(ficgp,"\n set log y; unset log x;set xlabel \"Age\"; set ylabel \"Likelihood (-2Log(L))\";");
     /* fprintf(ficgp,"\nset ter svg size 640, 480"); */ /* Too big for svg */
     fprintf(ficgp,"\nset ter pngcairo size 640, 480");
   /* nice for mle=4 plot by number of matrix products.
      replot  "rrtest1/toto.txt" u 2:($4 == 1 && $5==2 ? $9 : 1/0):5 t "p12" with point lc 1 */
   /* replot exp(p1+p2*x)/(1+exp(p1+p2*x)+exp(p3+p4*x)+exp(p5+p6*x)) t "p12(x)"  */
     /* fprintf(ficgp,"\nset out \"%s.svg\";",subdirf2(optionfilefiname,"ILK_")); */
     fprintf(ficgp,"\nset out \"%s-dest.png\";",subdirf2(optionfilefiname,"ILK_"));
     fprintf(ficgp,"\nset log y;plot  \"%s\" u 2:(-$13):6 t \"All sample, transitions colored by destination\" with dots lc variable; set out;\n",subdirf(fileresilk));
     fprintf(ficgp,"\nset out \"%s-ori.png\";",subdirf2(optionfilefiname,"ILK_"));
     fprintf(ficgp,"\nset log y;plot  \"%s\" u 2:(-$13):5 t \"All sample, transitions colored by origin\" with dots lc variable; set out;\n\n",subdirf(fileresilk));
     for (i=1; i<= nlstate ; i ++) {
       fprintf(ficgp,"\nset out \"%s-p%dj.png\";set ylabel \"Probability for each individual/wave\";",subdirf2(optionfilefiname,"ILK_"),i);
       fprintf(ficgp,"unset log;\n# plot weighted, mean weight should have point size of 0.5\n plot  \"%s\"",subdirf(fileresilk));
       fprintf(ficgp,"  u  2:($5 == %d && $6==%d ? $10 : 1/0):($12/4.):6 t \"p%d%d\" with points pointtype 7 ps variable lc variable \\\n",i,1,i,1);
       for (j=2; j<= nlstate+ndeath ; j ++) {
         fprintf(ficgp,",\\\n \"\" u  2:($5 == %d && $6==%d ? $10 : 1/0):($12/4.):6 t \"p%d%d\" with points pointtype 7 ps variable lc variable ",i,j,i,j);
       }
       fprintf(ficgp,";\nset out; unset ylabel;\n"); 
     }
     /* unset log; plot  "rrtest1_sorted_4/ILK_rrtest1_sorted_4.txt" u  2:($4 == 1 && $5==2 ? $9 : 1/0):5 t "p12" with points lc variable */                
     /* fprintf(ficgp,"\nset log y;plot  \"%s\" u 2:(-$11):3 t \"All sample, all transitions\" with dots lc variable",subdirf(fileresilk)); */
     /* fprintf(ficgp,"\nreplot  \"%s\" u 2:($3 <= 3 ? -$11 : 1/0):3 t \"First 3 individuals\" with line lc variable", subdirf(fileresilk)); */
     fprintf(ficgp,"\nset out;unset log\n");
     /* fprintf(ficgp,"\nset out \"%s.svg\"; replot; set out; # bug gnuplot",subdirf2(optionfilefiname,"ILK_")); */
   
     strcpy(dirfileres,optionfilefiname);
     strcpy(optfileres,"vpl");
     /* 1eme*/
     for (cpt=1; cpt<= nlstate ; cpt ++){ /* For each live state */
       for (k1=1; k1<= m ; k1 ++){ /* For each valid combination of covariate */
         for(nres=1; nres <= nresult; nres++){ /* For each resultline */
           /* plot [100000000000000000000:-100000000000000000000] "mysbiaspar/vplrmysbiaspar.txt to check */
           if(m != 1 && TKresult[nres]!= k1)
             continue;
           /* We are interested in selected combination by the resultline */
           /* printf("\n# 1st: Period (stable) prevalence with CI: 'VPL_' files and live state =%d ", cpt); */
           fprintf(ficgp,"\n# 1st: Forward (stable period) prevalence with CI: 'VPL_' files  and live state =%d ", cpt);
           strcpy(gplotlabel,"(");
           for (k=1; k<=cptcoveff; k++){    /* For each covariate k get corresponding value lv for combination k1 */
             /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the value of the covariate corresponding to k1 combination *\/ */
             lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
             /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
             /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
             /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
             vlv= nbcode[Tvaraff[k]][lv]; /* vlv is the value of the covariate lv, 0 or 1 */
             /* For each combination of covariate k1 (V1=1, V3=0), we printed the current covariate k and its value vlv */
             /* printf(" V%d=%d ",Tvaraff[k],vlv); */
             fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
           }
           for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
             /* printf(" V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]); */
             fprintf(ficgp," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
           }
           strcpy(gplotlabel+strlen(gplotlabel),")");
           /* printf("\n#\n"); */
           fprintf(ficgp,"\n#\n");
           if(invalidvarcomb[k1]){
             /*k1=k1-1;*/ /* To be checked */
             fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
             continue;
           }
         
           fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" \n",subdirf2(optionfilefiname,"V_"),cpt,k1,nres);
           fprintf(ficgp,"\n#set out \"V_%s_%d-%d-%d.svg\" \n",optionfilefiname,cpt,k1,nres);
           /* fprintf(ficgp,"set label \"Alive state %d %s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",cpt,gplotlabel); */
           fprintf(ficgp,"set title \"Alive state %d %s model=%s\" font \"Helvetica,12\"\n",cpt,gplotlabel,model);
           fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Probability\" \nset ter svg size 640, 480\nplot [%.f:%.f] \"%s\" every :::%d::%d u 1:($2==%d ? $3:1/0) \"%%lf %%lf",ageminpar,fage,subdirf2(fileresu,"VPL_"),nres-1,nres-1,nres);
           /* fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Probability\" \nset ter svg size 640, 480\nplot [%.f:%.f] \"%s\" every :::%d::%d u 1:($2==%d ? $3:1/0) \"%%lf %%lf",ageminpar,fage,subdirf2(fileresu,"VPL_"),k1-1,k1-1,nres); */
         /* k1-1 error should be nres-1*/
           for (i=1; i<= nlstate ; i ++) {
             if (i==cpt) fprintf(ficgp," %%lf (%%lf)");
             else        fprintf(ficgp," %%*lf (%%*lf)");
           }
           fprintf(ficgp,"\" t\"Forward prevalence\" w l lt 0,\"%s\" every :::%d::%d u 1:($2==%d ? $3+1.96*$4 : 1/0) \"%%lf %%lf",subdirf2(fileresu,"VPL_"),nres-1,nres-1,nres);
           for (i=1; i<= nlstate ; i ++) {
             if (i==cpt) fprintf(ficgp," %%lf (%%lf)");
             else fprintf(ficgp," %%*lf (%%*lf)");
           } 
           fprintf(ficgp,"\" t\"95%% CI\" w l lt 1,\"%s\" every :::%d::%d u 1:($2==%d ? $3-1.96*$4 : 1/0) \"%%lf %%lf",subdirf2(fileresu,"VPL_"),nres-1,nres-1,nres); 
           for (i=1; i<= nlstate ; i ++) {
             if (i==cpt) fprintf(ficgp," %%lf (%%lf)");
             else fprintf(ficgp," %%*lf (%%*lf)");
           }  
           /* fprintf(ficgp,"\" t\"\" w l lt 1,\"%s\" every :::%d::%d u 1:($%d) t\"Observed prevalence\" w l lt 2",subdirf2(fileresu,"P_"),k1-1,k1-1,2+4*(cpt-1)); */
           
           fprintf(ficgp,"\" t\"\" w l lt 1,\"%s\" u 1:((",subdirf2(fileresu,"P_"));
           if(cptcoveff ==0){
             fprintf(ficgp,"$%d)) t 'Observed prevalence in state %d' with line lt 3",      2+3*(cpt-1),  cpt );
           }else{
             kl=0;
             for (k=1; k<=cptcoveff; k++){    /* For each combination of covariate  */
               /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate value corresponding to k1 combination and kth covariate *\/ */
               lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
               /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
               /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
               /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
               vlv= nbcode[Tvaraff[k]][lv];
               kl++;
               /* kl=6+(cpt-1)*(nlstate+1)+1+(i-1); /\* 6+(1-1)*(2+1)+1+(1-1)=7, 6+(2-1)(2+1)+1+(1-1)=10 *\/ */
               /*6+(cpt-1)*(nlstate+1)+1+(i-1)+(nlstate+1)*nlstate; 6+(1-1)*(2+1)+1+(1-1) +(2+1)*2=13 */ 
               /*6+1+(i-1)+(nlstate+1)*nlstate; 6+1+(1-1) +(2+1)*2=13 */ 
               /* ''  u 6:(($1==1 && $2==0 && $3==2 && $4==0)? $9/(1.-$15) : 1/0):($5==2000? 3:2) t 'p.1' with line lc variable*/
               if(k==cptcoveff){
                 fprintf(ficgp,"$%d==%d && $%d==%d)? $%d : 1/0) t 'Observed prevalence in state %d' w l lt 2",kl+1, Tvaraff[k],kl+1+1,nbcode[Tvaraff[k]][lv], \
                         2+cptcoveff*2+3*(cpt-1),  cpt );  /* 4 or 6 ?*/
               }else{
                 fprintf(ficgp,"$%d==%d && $%d==%d && ",kl+1, Tvaraff[k],kl+1+1,nbcode[Tvaraff[k]][lv]);
                 kl++;
               }
             } /* end covariate */
           } /* end if no covariate */
   
           if(prevbcast==1){ /* We need to get the corresponding values of the covariates involved in this combination k1 */
             /* fprintf(ficgp,",\"%s\" every :::%d::%d u 1:($%d) t\"Backward stable prevalence\" w l lt 3",subdirf2(fileresu,"PLB_"),k1-1,k1-1,1+cpt); */
             fprintf(ficgp,",\"%s\" u 1:((",subdirf2(fileresu,"PLB_")); /* Age is in 1, nres in 2 to be fixed */
             if(cptcoveff ==0){
               fprintf(ficgp,"$%d)) t 'Backward prevalence in state %d' with line lt 3",    2+(cpt-1),  cpt );
             }else{
               kl=0;
               for (k=1; k<=cptcoveff; k++){    /* For each combination of covariate  */
                 /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate value corresponding to k1 combination and kth covariate *\/ */
                 lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
                 /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
                 /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
                 /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
                 /* vlv= nbcode[Tvaraff[k]][lv]; */
                 vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
                 kl++;
                 /* kl=6+(cpt-1)*(nlstate+1)+1+(i-1); /\* 6+(1-1)*(2+1)+1+(1-1)=7, 6+(2-1)(2+1)+1+(1-1)=10 *\/ */
                 /*6+(cpt-1)*(nlstate+1)+1+(i-1)+(nlstate+1)*nlstate; 6+(1-1)*(2+1)+1+(1-1) +(2+1)*2=13 */ 
                 /*6+1+(i-1)+(nlstate+1)*nlstate; 6+1+(1-1) +(2+1)*2=13 */ 
                 /* ''  u 6:(($1==1 && $2==0 && $3==2 && $4==0)? $9/(1.-$15) : 1/0):($5==2000? 3:2) t 'p.1' with line lc variable*/
                 if(k==cptcoveff){
                   fprintf(ficgp,"$%d==%d && $%d==%d)? $%d : 1/0) t 'Backward prevalence in state %d' w l lt 3",kl+1, Tvaraff[k],kl+1+1,nbcode[Tvaraff[k]][lv], \
                           2+cptcoveff*2+(cpt-1),  cpt );  /* 4 or 6 ?*/
                 }else{
                   fprintf(ficgp,"$%d==%d && $%d==%d && ",kl+1, Tvaraff[k],kl+1+1,nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])]);
                   kl++;
                 }
               } /* end covariate */
             } /* end if no covariate */
             if(prevbcast == 1){
               fprintf(ficgp,", \"%s\" every :::%d::%d u 1:($2==%d ? $3:1/0) \"%%lf %%lf",subdirf2(fileresu,"VBL_"),nres-1,nres-1,nres);
               /* k1-1 error should be nres-1*/
               for (i=1; i<= nlstate ; i ++) {
                 if (i==cpt) fprintf(ficgp," %%lf (%%lf)");
                 else        fprintf(ficgp," %%*lf (%%*lf)");
               }
               fprintf(ficgp,"\" t\"Backward (stable) prevalence\" w l lt 6 dt 3,\"%s\" every :::%d::%d u 1:($2==%d ? $3+1.96*$4 : 1/0) \"%%lf %%lf",subdirf2(fileresu,"VBL_"),nres-1,nres-1,nres);
               for (i=1; i<= nlstate ; i ++) {
                 if (i==cpt) fprintf(ficgp," %%lf (%%lf)");
                 else fprintf(ficgp," %%*lf (%%*lf)");
               } 
               fprintf(ficgp,"\" t\"95%% CI\" w l lt 4,\"%s\" every :::%d::%d u 1:($2==%d ? $3-1.96*$4 : 1/0) \"%%lf %%lf",subdirf2(fileresu,"VBL_"),nres-1,nres-1,nres); 
               for (i=1; i<= nlstate ; i ++) {
                 if (i==cpt) fprintf(ficgp," %%lf (%%lf)");
                 else fprintf(ficgp," %%*lf (%%*lf)");
               } 
               fprintf(ficgp,"\" t\"\" w l lt 4");
             } /* end if backprojcast */
           } /* end if prevbcast */
           /* fprintf(ficgp,"\nset out ;unset label;\n"); */
           fprintf(ficgp,"\nset out ;unset title;\n");
         } /* nres */
       } /* k1 */
     } /* cpt */
   
       
       cptj=0;    /*2 eme*/
       for(j=1; j<= nlstate; j++){    for (k1=1; k1<= m ; k1 ++){  
         for(i=1; i<=nlstate; i++){      for(nres=1; nres <= nresult; nres++){ /* For each resultline */
           cptj=cptj+1;        if(m != 1 && TKresult[nres]!= k1)
           for(h=0, gp[h][cptj]=0.; h<=nhstepm-1; h++){          continue;
             gp[h][cptj] = (p3mat[i][j][h]+p3mat[i][j][h+1])/2.;        fprintf(ficgp,"\n# 2nd: Total life expectancy with CI: 't' files ");
         strcpy(gplotlabel,"(");
         for (k=1; k<=cptcoveff; k++){    /* For each covariate and each value */
           /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate number corresponding to k1 combination *\/ */
           lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
           /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
           /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
           /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
           /* vlv= nbcode[Tvaraff[k]][lv]; */
           vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
           fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
           sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
         }
         /* for(k=1; k <= ncovds; k++){ */
         for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
           printf(" V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
           fprintf(ficgp," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
           sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
         }
         strcpy(gplotlabel+strlen(gplotlabel),")");
         fprintf(ficgp,"\n#\n");
         if(invalidvarcomb[k1]){
           fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
           continue;
         }
                           
         fprintf(ficgp,"\nset out \"%s_%d-%d.svg\" \n",subdirf2(optionfilefiname,"E_"),k1,nres);
         for(vpopbased=0; vpopbased <= popbased; vpopbased++){ /* Done for vpopbased=0 and vpopbased=1 if popbased==1*/
           fprintf(ficgp,"\nset label \"popbased %d %s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",vpopbased,gplotlabel);
           if(vpopbased==0){
             fprintf(ficgp,"set ylabel \"Years\" \nset ter svg size 640, 480\nplot [%.f:%.f] ",ageminpar,fage);
           }else
             fprintf(ficgp,"\nreplot ");
           for (i=1; i<= nlstate+1 ; i ++) {
             k=2*i;
             fprintf(ficgp,"\"%s\" every :::%d::%d u 1:($2==%d && $4!=0 ?$4 : 1/0) \"%%lf %%lf %%lf",subdirf2(fileresu,"T_"),nres-1,nres-1, vpopbased);
             for (j=1; j<= nlstate+1 ; j ++) {
               if (j==i) fprintf(ficgp," %%lf (%%lf)");
               else fprintf(ficgp," %%*lf (%%*lf)");
             }   
             if (i== 1) fprintf(ficgp,"\" t\"TLE\" w l lt %d, \\\n",i);
             else fprintf(ficgp,"\" t\"LE in state (%d)\" w l lt %d, \\\n",i-1,i+1);
             fprintf(ficgp,"\"%s\" every :::%d::%d u 1:($2==%d && $4!=0 ? $4-$5*2 : 1/0) \"%%lf %%lf %%lf",subdirf2(fileresu,"T_"),nres-1,nres-1,vpopbased);
             for (j=1; j<= nlstate+1 ; j ++) {
               if (j==i) fprintf(ficgp," %%lf (%%lf)");
               else fprintf(ficgp," %%*lf (%%*lf)");
             }   
             fprintf(ficgp,"\" t\"\" w l lt 0,");
             fprintf(ficgp,"\"%s\" every :::%d::%d u 1:($2==%d && $4!=0 ? $4+$5*2 : 1/0) \"%%lf %%lf %%lf",subdirf2(fileresu,"T_"),nres-1,nres-1,vpopbased);
             for (j=1; j<= nlstate+1 ; j ++) {
               if (j==i) fprintf(ficgp," %%lf (%%lf)");
               else fprintf(ficgp," %%*lf (%%*lf)");
             }   
             if (i== (nlstate+1)) fprintf(ficgp,"\" t\"\" w l lt 0");
             else fprintf(ficgp,"\" t\"\" w l lt 0,\\\n");
           } /* state */
         } /* vpopbased */
         fprintf(ficgp,"\nset out;set out \"%s_%d-%d.svg\"; replot; set out; unset label;\n",subdirf2(optionfilefiname,"E_"),k1,nres); /* Buggy gnuplot */
       } /* end nres */
     } /* k1 end 2 eme*/
           
           
     /*3eme*/
     for (k1=1; k1<= m ; k1 ++){
       for(nres=1; nres <= nresult; nres++){ /* For each resultline */
         if(m != 1 && TKresult[nres]!= k1)
           continue;
   
         for (cpt=1; cpt<= nlstate ; cpt ++) { /* Fragile no verification of covariate values */
           fprintf(ficgp,"\n\n# 3d: Life expectancy with EXP_ files:  combination=%d state=%d",k1, cpt);
           strcpy(gplotlabel,"(");
           for (k=1; k<=cptcoveff; k++){    /* For each covariate and each value */
             /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate number corresponding to k1 combination *\/ */
             lv= codtabm(k1,TnsdVar[Tvaraff[k]]); /* Should be the covariate value corresponding to combination k1 and covariate k */
             /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
             /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
             /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
             /* vlv= nbcode[Tvaraff[k]][lv]; */
             vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
             fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
           }
           for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
             fprintf(ficgp," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][resultmodel[nres][k4]]);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][resultmodel[nres][k4]]);
           }       
           strcpy(gplotlabel+strlen(gplotlabel),")");
           fprintf(ficgp,"\n#\n");
           if(invalidvarcomb[k1]){
             fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
             continue;
           }
                           
           /*       k=2+nlstate*(2*cpt-2); */
           k=2+(nlstate+1)*(cpt-1);
           fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" \n",subdirf2(optionfilefiname,"EXP_"),cpt,k1,nres);
           fprintf(ficgp,"set label \"%s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",gplotlabel);
           fprintf(ficgp,"set ter svg size 640, 480\n\
   plot [%.f:%.f] \"%s\" every :::%d::%d u 1:%d t \"e%d1\" w l",ageminpar,fage,subdirf2(fileresu,"E_"),nres-1,nres-1,k,cpt);
           /*fprintf(ficgp,",\"e%s\" every :::%d::%d u 1:($%d-2*$%d) \"\%%lf ",fileres,k1-1,k1-1,k,k+1);
             for (i=1; i<= nlstate*2 ; i ++) fprintf(ficgp,"\%%lf (\%%lf) ");
             fprintf(ficgp,"\" t \"e%d1\" w l",cpt);
             fprintf(ficgp,",\"e%s\" every :::%d::%d u 1:($%d+2*$%d) \"\%%lf ",fileres,k1-1,k1-1,k,k+1);
             for (i=1; i<= nlstate*2 ; i ++) fprintf(ficgp,"\%%lf (\%%lf) ");
             fprintf(ficgp,"\" t \"e%d1\" w l",cpt);
                                   
           */
           for (i=1; i< nlstate ; i ++) {
             fprintf(ficgp," ,\"%s\" every :::%d::%d u 1:%d t \"e%d%d\" w l",subdirf2(fileresu,"E_"),nres-1,nres-1,k+i,cpt,i+1);
             /*    fprintf(ficgp," ,\"%s\" every :::%d::%d u 1:%d t \"e%d%d\" w l",subdirf2(fileres,"e"),k1-1,k1-1,k+2*i,cpt,i+1);*/
                                   
           } 
           fprintf(ficgp," ,\"%s\" every :::%d::%d u 1:%d t \"e%d.\" w l",subdirf2(fileresu,"E_"),nres-1,nres-1,k+nlstate,cpt);
         }
         fprintf(ficgp,"\nunset label;\n");
       } /* end nres */
     } /* end kl 3eme */
     
     /* 4eme */
     /* Survival functions (period) from state i in state j by initial state i */
     for (k1=1; k1<=m; k1++){    /* For each covariate and each value */
       for(nres=1; nres <= nresult; nres++){ /* For each resultline */
         if(m != 1 && TKresult[nres]!= k1)
           continue;
         for (cpt=1; cpt<=nlstate ; cpt ++) { /* For each life state cpt*/
           strcpy(gplotlabel,"(");
           fprintf(ficgp,"\n#\n#\n# Survival functions in state j : 'LIJ_' files, cov=%d state=%d",k1, cpt);
           for (k=1; k<=cptcoveff; k++){    /* For each covariate and each value */
             lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
             /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate number corresponding to k1 combination *\/ */
             /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
             /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
             /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
             /* vlv= nbcode[Tvaraff[k]][lv]; */
             vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
             fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
           }
           for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
             fprintf(ficgp," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
           }       
           strcpy(gplotlabel+strlen(gplotlabel),")");
           fprintf(ficgp,"\n#\n");
           if(invalidvarcomb[k1]){
             fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
             continue;
           }
         
           fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" \n",subdirf2(optionfilefiname,"LIJ_"),cpt,k1,nres);
           fprintf(ficgp,"set label \"Alive state %d %s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",cpt,gplotlabel);
           fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Probability to be alive\" \n\
   set ter svg size 640, 480\nunset log y\nplot [%.f:%.f]  ", ageminpar, agemaxpar);
           k=3;
           for (i=1; i<= nlstate ; i ++){
             if(i==1){
               fprintf(ficgp,"\"%s\"",subdirf2(fileresu,"PIJ_"));
             }else{
               fprintf(ficgp,", '' ");
           }            }
             l=(nlstate+ndeath)*(i-1)+1;
             fprintf(ficgp," u ($1==%d ? ($3):1/0):($%d/($%d",k1,k+l+(cpt-1),k+l);
             for (j=2; j<= nlstate+ndeath ; j ++)
               fprintf(ficgp,"+$%d",k+l+j-1);
             fprintf(ficgp,")) t \"l(%d,%d)\" w l",i,cpt);
           } /* nlstate */
           fprintf(ficgp,"\nset out; unset label;\n");
         } /* end cpt state*/ 
       } /* end nres */
     } /* end covariate k1 */  
   
   /* 5eme */
     /* Survival functions (period) from state i in state j by final state j */
     for (k1=1; k1<= m ; k1++){ /* For each covariate combination if any */
       for(nres=1; nres <= nresult; nres++){ /* For each resultline */
         if(m != 1 && TKresult[nres]!= k1)
           continue;
         for (cpt=1; cpt<=nlstate ; cpt ++) { /* For each inital state  */
           strcpy(gplotlabel,"(");
           fprintf(ficgp,"\n#\n#\n# Survival functions in state j and all livestates from state i by final state j: 'lij' files, cov=%d state=%d",k1, cpt);
           for (k=1; k<=cptcoveff; k++){    /* For each covariate and each value */
             lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
             /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate number corresponding to k1 combination *\/ */
             /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
             /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
             /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
             /* vlv= nbcode[Tvaraff[k]][lv]; */
             vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
             fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
           }
           for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
             fprintf(ficgp," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
           }       
           strcpy(gplotlabel+strlen(gplotlabel),")");
           fprintf(ficgp,"\n#\n");
           if(invalidvarcomb[k1]){
             fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
             continue;
         }          }
         
           fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" \n",subdirf2(optionfilefiname,"LIJT_"),cpt,k1,nres);
           fprintf(ficgp,"set label \"Alive state %d %s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",cpt,gplotlabel);
           fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Probability to be alive\" \n\
   set ter svg size 640, 480\nunset log y\nplot [%.f:%.f]  ", ageminpar, agemaxpar);
           k=3;
           for (j=1; j<= nlstate ; j ++){ /* Lived in state j */
             if(j==1)
               fprintf(ficgp,"\"%s\"",subdirf2(fileresu,"PIJ_"));
             else
               fprintf(ficgp,", '' ");
             l=(nlstate+ndeath)*(cpt-1) +j;
             fprintf(ficgp," u (($1==%d && (floor($2)%%5 == 0)) ? ($3):1/0):($%d",k1,k+l);
             /* for (i=2; i<= nlstate+ndeath ; i ++) */
             /*   fprintf(ficgp,"+$%d",k+l+i-1); */
             fprintf(ficgp,") t \"l(%d,%d)\" w l",cpt,j);
           } /* nlstate */
           fprintf(ficgp,", '' ");
           fprintf(ficgp," u (($1==%d && (floor($2)%%5 == 0)) ? ($3):1/0):(",k1);
           for (j=1; j<= nlstate ; j ++){ /* Lived in state j */
             l=(nlstate+ndeath)*(cpt-1) +j;
             if(j < nlstate)
               fprintf(ficgp,"$%d +",k+l);
             else
               fprintf(ficgp,"$%d) t\"l(%d,.)\" w l",k+l,cpt);
           }
           fprintf(ficgp,"\nset out; unset label;\n");
         } /* end cpt state*/ 
       } /* end covariate */  
     } /* end nres */
     
   /* 6eme */
     /* CV preval stable (period) for each covariate */
     for (k1=1; k1<= m ; k1 ++) /* For each covariate combination if any */
     for(nres=1; nres <= nresult; nres++){ /* For each resultline */
       if(m != 1 && TKresult[nres]!= k1)
         continue;
       for (cpt=1; cpt<=nlstate ; cpt ++) { /* For each life state of arrival */
         strcpy(gplotlabel,"(");      
         fprintf(ficgp,"\n#\n#\n#CV preval stable (forward): 'pij' files, covariatecombination#=%d state=%d",k1, cpt);
         for (k=1; k<=cptcoveff; k++){    /* For each covariate and each value */
           /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate number corresponding to k1 combination *\/ */
           lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
           /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
           /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
           /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
           /* vlv= nbcode[Tvaraff[k]][lv]; */
           vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
           fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
           sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
         }
         for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
           fprintf(ficgp," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
           sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
         } 
         strcpy(gplotlabel+strlen(gplotlabel),")");
         fprintf(ficgp,"\n#\n");
         if(invalidvarcomb[k1]){
           fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
           continue;
       }        }
        
        
       for(i=1; i<=npar; i++)   
         xp[i] = x[i] - (i==theta ?delti[theta]:0);  
       hpxij(p3mat,nhstepm,age,hstepm,xp,nlstate,stepm,oldm,savm, ij);    
               
       cptj=0;        fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" \n",subdirf2(optionfilefiname,"P_"),cpt,k1,nres);
       for(j=1; j<= nlstate; j++){        fprintf(ficgp,"set label \"Alive state %d %s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",cpt,gplotlabel);
         for(i=1;i<=nlstate;i++){        fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Probability\" \n\
           cptj=cptj+1;  set ter svg size 640, 480\nunset log y\nplot [%.f:%.f]  ", ageminpar, agemaxpar);
           for(h=0, gm[h][cptj]=0.; h<=nhstepm-1; h++){        k=3; /* Offset */
         for (i=1; i<= nlstate ; i ++){ /* State of origin */
             gm[h][cptj] = (p3mat[i][j][h]+p3mat[i][j][h+1])/2.;          if(i==1)
             fprintf(ficgp,"\"%s\"",subdirf2(fileresu,"PIJ_"));
           else
             fprintf(ficgp,", '' ");
           l=(nlstate+ndeath)*(i-1)+1; /* 1, 1+ nlstate+ndeath, 1+2*(nlstate+ndeath) */
           fprintf(ficgp," u ($1==%d ? ($3):1/0):($%d/($%d",k1,k+l+(cpt-1),k+l);
           for (j=2; j<= nlstate ; j ++)
             fprintf(ficgp,"+$%d",k+l+j-1);
           fprintf(ficgp,")) t \"prev(%d,%d)\" w l",i,cpt);
         } /* nlstate */
         fprintf(ficgp,"\nset out; unset label;\n");
       } /* end cpt state*/ 
     } /* end covariate */  
     
     
   /* 7eme */
     if(prevbcast == 1){
       /* CV backward prevalence  for each covariate */
       for (k1=1; k1<= m ; k1 ++) /* For each covariate combination if any */
       for(nres=1; nres <= nresult; nres++){ /* For each resultline */
         if(m != 1 && TKresult[nres]!= k1)
           continue;
         for (cpt=1; cpt<=nlstate ; cpt ++) { /* For each life origin state */
           strcpy(gplotlabel,"(");      
           fprintf(ficgp,"\n#\n#\n#CV Backward stable prevalence: 'pijb' files, covariatecombination#=%d state=%d",k1, cpt);
           for (k=1; k<=cptcoveff; k++){    /* For each covariate and each value */
             /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate number corresponding to k1 combination *\/ */
             lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
             /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
             /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
             /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
             /* vlv= nbcode[Tvaraff[k]][lv]; */
             vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
             fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
           }
           for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
             fprintf(ficgp," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
           }       
           strcpy(gplotlabel+strlen(gplotlabel),")");
           fprintf(ficgp,"\n#\n");
           if(invalidvarcomb[k1]){
             fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
             continue;
           }
           
           fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" \n",subdirf2(optionfilefiname,"PB_"),cpt,k1,nres);
           fprintf(ficgp,"set label \"Origin alive state %d %s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",cpt,gplotlabel);
           fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Probability\" \n\
   set ter svg size 640, 480\nunset log y\nplot [%.f:%.f]  ", ageminpar, agemaxpar);
           k=3; /* Offset */
           for (i=1; i<= nlstate ; i ++){ /* State of arrival */
             if(i==1)
               fprintf(ficgp,"\"%s\"",subdirf2(fileresu,"PIJB_"));
             else
               fprintf(ficgp,", '' ");
             /* l=(nlstate+ndeath)*(i-1)+1; */
             l=(nlstate+ndeath)*(cpt-1)+1; /* fixed for i; cpt=1 1, cpt=2 1+ nlstate+ndeath, 1+2*(nlstate+ndeath) */
             /* fprintf(ficgp," u ($1==%d ? ($3):1/0):($%d/($%d",k1,k+l+(cpt-1),k+l); /\* a veÌrifier *\/ */
             /* fprintf(ficgp," u ($1==%d ? ($3):1/0):($%d/($%d",k1,k+l+(cpt-1),k+l+(cpt-1)+i-1); /\* a veÌrifier *\/ */
             fprintf(ficgp," u ($1==%d ? ($3):1/0):($%d",k1,k+l+i-1); /* To be verified */
             /* for (j=2; j<= nlstate ; j ++) */
             /*    fprintf(ficgp,"+$%d",k+l+j-1); */
             /*    /\* fprintf(ficgp,"+$%d",k+l+j-1); *\/ */
             fprintf(ficgp,") t \"bprev(%d,%d)\" w l",cpt,i);
           } /* nlstate */
           fprintf(ficgp,"\nset out; unset label;\n");
         } /* end cpt state*/ 
       } /* end covariate */  
     } /* End if prevbcast */
     
     /* 8eme */
     if(prevfcast==1){
       /* Projection from cross-sectional to forward stable (period) prevalence for each covariate */
       
       for (k1=1; k1<= m ; k1 ++) /* For each covariate combination if any */
       for(nres=1; nres <= nresult; nres++){ /* For each resultline */
         if(m != 1 && TKresult[nres]!= k1)
           continue;
         for (cpt=1; cpt<=nlstate ; cpt ++) { /* For each life state */
           strcpy(gplotlabel,"(");      
           fprintf(ficgp,"\n#\n#\n#Projection of prevalence to forward stable prevalence (period): 'PROJ_' files, covariatecombination#=%d state=%d",k1, cpt);
           for (k=1; k<=cptcoveff; k++){    /* For each correspondig covariate value  */
             /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate value corresponding to k1 combination and kth covariate *\/ */
             lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
             /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
             /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
             /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
             /* vlv= nbcode[Tvaraff[k]][lv]; */
             vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
             fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
           }
           for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
             fprintf(ficgp," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
           }       
           strcpy(gplotlabel+strlen(gplotlabel),")");
           fprintf(ficgp,"\n#\n");
           if(invalidvarcomb[k1]){
             fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
             continue;
           }
           
           fprintf(ficgp,"# hpijx=probability over h years, hp.jx is weighted by observed prev\n ");
           fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" \n",subdirf2(optionfilefiname,"PROJ_"),cpt,k1,nres);
           fprintf(ficgp,"set label \"Alive state %d %s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",cpt,gplotlabel);
           fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Prevalence\" \n\
   set ter svg size 640, 480\nunset log y\nplot [%.f:%.f]  ", ageminpar, agemaxpar);
   
           /* for (i=1; i<= nlstate+1 ; i ++){  /\* nlstate +1 p11 p21 p.1 *\/ */
           istart=nlstate+1; /* Could be one if by state, but nlstate+1 is w.i projection only */
           /*istart=1;*/ /* Could be one if by state, but nlstate+1 is w.i projection only */
           for (i=istart; i<= nlstate+1 ; i ++){  /* nlstate +1 p11 p21 p.1 */
             /*#  V1  = 1  V2 =  0 yearproj age p11 p21 p.1 p12 p22 p.2 p13 p23 p.3*/
             /*#   1    2   3    4    5      6  7   8   9   10   11 12  13   14  15 */   
             /*# yearproj age p11 p21 p.1 p12 p22 p.2 p13 p23 p.3*/
             /*#   1       2   3    4    5      6  7   8   9   10   11 12  13   14  15 */   
             if(i==istart){
               fprintf(ficgp,"\"%s\"",subdirf2(fileresu,"F_"));
             }else{
               fprintf(ficgp,",\\\n '' ");
           }            }
             if(cptcoveff ==0){ /* No covariate */
               ioffset=2; /* Age is in 2 */
               /*# yearproj age p11 p21 p31 p.1 p12 p22 p32 p.2 p13 p23 p33 p.3 p14 p24 p34 p.4*/
               /*#   1       2   3   4   5  6    7  8   9   10  11  12  13  14  15  16  17  18 */
               /*# V1  = 1 yearproj age p11 p21 p31 p.1 p12 p22 p32 p.2 p13 p23 p33 p.3 p14 p24 p34 p.4*/
               /*#  1    2        3   4   5  6    7  8   9   10  11  12  13  14  15  16  17  18 */
               fprintf(ficgp," u %d:(", ioffset); 
               if(i==nlstate+1){
                 fprintf(ficgp," $%d/(1.-$%d)):1 t 'pw.%d' with line lc variable ",        \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1),  ioffset+1+(i-1)+(nlstate+1)*nlstate,cpt );
                 fprintf(ficgp,",\\\n '' ");
                 fprintf(ficgp," u %d:(",ioffset); 
                 fprintf(ficgp," (($1-$2) == %d ) ? $%d/(1.-$%d) : 1/0):1 with labels center not ", \
                        offyear,                           \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1),  ioffset+1+(i-1)+(nlstate+1)*nlstate );
               }else
                 fprintf(ficgp," $%d/(1.-$%d)) t 'p%d%d' with line ",      \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1),  ioffset+1+(i-1)+(nlstate+1)*nlstate,i,cpt );
             }else{ /* more than 2 covariates */
               ioffset=2*cptcoveff+2; /* Age is in 4 or 6 or etc.*/
               /*#  V1  = 1  V2 =  0 yearproj age p11 p21 p.1 p12 p22 p.2 p13 p23 p.3*/
               /*#   1    2   3    4    5      6  7   8   9   10   11 12  13   14  15 */
               iyearc=ioffset-1;
               iagec=ioffset;
               fprintf(ficgp," u %d:(",ioffset); 
               kl=0;
               strcpy(gplotcondition,"(");
               for (k=1; k<=cptcoveff; k++){    /* For each covariate writing the chain of conditions */
                 /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate value corresponding to combination k1 and covariate k *\/ */
                 lv=codtabm(k1,TnsdVar[Tvaraff[k]]);
                 /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
                 /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
                 /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
                 /* vlv= nbcode[Tvaraff[k]][lv]; /\* Value of the modality of Tvaraff[k] *\/ */
                 vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
                 kl++;
                 sprintf(gplotcondition+strlen(gplotcondition),"$%d==%d && $%d==%d " ,kl,Tvaraff[k], kl+1, nbcode[Tvaraff[k]][lv]);
                 kl++;
                 if(k <cptcoveff && cptcoveff>1)
                   sprintf(gplotcondition+strlen(gplotcondition)," && ");
               }
               strcpy(gplotcondition+strlen(gplotcondition),")");
               /* kl=6+(cpt-1)*(nlstate+1)+1+(i-1); /\* 6+(1-1)*(2+1)+1+(1-1)=7, 6+(2-1)(2+1)+1+(1-1)=10 *\/ */
               /*6+(cpt-1)*(nlstate+1)+1+(i-1)+(nlstate+1)*nlstate; 6+(1-1)*(2+1)+1+(1-1) +(2+1)*2=13 */ 
               /*6+1+(i-1)+(nlstate+1)*nlstate; 6+1+(1-1) +(2+1)*2=13 */ 
               /* ''  u 6:(($1==1 && $2==0 && $3==2 && $4==0)? $9/(1.-$15) : 1/0):($5==2000? 3:2) t 'p.1' with line lc variable*/
               if(i==nlstate+1){
                 fprintf(ficgp,"%s ? $%d/(1.-$%d) : 1/0):%d t 'p.%d' with line lc variable", gplotcondition, \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1),  ioffset+1+(i-1)+(nlstate+1)*nlstate,iyearc, cpt );
                 fprintf(ficgp,",\\\n '' ");
                 fprintf(ficgp," u %d:(",iagec); 
                 fprintf(ficgp,"%s && (($%d-$%d) == %d ) ? $%d/(1.-$%d) : 1/0):%d with labels center not ", gplotcondition, \
                         iyearc, iagec, offyear,                           \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1),  ioffset+1+(i-1)+(nlstate+1)*nlstate, iyearc );
   /*  '' u 6:(($1==1 && $2==0  && $3==2 && $4==0) && (($5-$6) == 1947) ? $10/(1.-$22) : 1/0):5 with labels center boxed not*/
               }else{
                 fprintf(ficgp,"%s ? $%d/(1.-$%d) : 1/0) t 'p%d%d' with line ", gplotcondition, \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1), ioffset +1+(i-1)+(nlstate+1)*nlstate,i,cpt );
               }
             } /* end if covariate */
           } /* nlstate */
           fprintf(ficgp,"\nset out; unset label;\n");
         } /* end cpt state*/
       } /* end covariate */
     } /* End if prevfcast */
     
     if(prevbcast==1){
       /* Back projection from cross-sectional to stable (mixed) for each covariate */
       
       for (k1=1; k1<= m ; k1 ++) /* For each covariate combination if any */
       for(nres=1; nres <= nresult; nres++){ /* For each resultline */
         if(m != 1 && TKresult[nres]!= k1)
           continue;
         for (cpt=1; cpt<=nlstate ; cpt ++) { /* For each life state */
           strcpy(gplotlabel,"(");      
           fprintf(ficgp,"\n#\n#\n#Back projection of prevalence to stable (mixed) back prevalence: 'BPROJ_' files, covariatecombination#=%d originstate=%d",k1, cpt);
           for (k=1; k<=cptcoveff; k++){    /* For each correspondig covariate value  */
             /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate value corresponding to k1 combination and kth covariate *\/ */
             lv= codtabm(k1,TnsdVar[Tvaraff[k]]); /* Should be the covariate value corresponding to combination k1 and covariate k */
             /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
             /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
             /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
             /* vlv= nbcode[Tvaraff[k]][lv]; */
             vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
             fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
           }
           for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
             fprintf(ficgp," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
             sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
           }       
           strcpy(gplotlabel+strlen(gplotlabel),")");
           fprintf(ficgp,"\n#\n");
           if(invalidvarcomb[k1]){
             fprintf(ficgp,"#Combination (%d) ignored because no cases \n",k1); 
             continue;
         }          }
       }          
       for(j=1; j<= nlstate*nlstate; j++)          fprintf(ficgp,"# hbijx=backprobability over h years, hb.jx is weighted by observed prev at destination state\n ");
         for(h=0; h<=nhstepm-1; h++){          fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" \n",subdirf2(optionfilefiname,"PROJB_"),cpt,k1,nres);
           gradg[h][theta][j]= (gp[h][j]-gm[h][j])/2./delti[theta];          fprintf(ficgp,"set label \"Origin alive state %d %s\" at graph 0.98,0.5 center rotate font \"Helvetica,12\"\n",cpt,gplotlabel);
           fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Prevalence\" \n\
   set ter svg size 640, 480\nunset log y\nplot [%.f:%.f]  ", ageminpar, agemaxpar);
   
           /* for (i=1; i<= nlstate+1 ; i ++){  /\* nlstate +1 p11 p21 p.1 *\/ */
           istart=nlstate+1; /* Could be one if by state, but nlstate+1 is w.i projection only */
           /*istart=1;*/ /* Could be one if by state, but nlstate+1 is w.i projection only */
           for (i=istart; i<= nlstate+1 ; i ++){  /* nlstate +1 p11 p21 p.1 */
             /*#  V1  = 1  V2 =  0 yearproj age p11 p21 p.1 p12 p22 p.2 p13 p23 p.3*/
             /*#   1    2   3    4    5      6  7   8   9   10   11 12  13   14  15 */   
             /*# yearproj age p11 p21 p.1 p12 p22 p.2 p13 p23 p.3*/
             /*#   1       2   3    4    5      6  7   8   9   10   11 12  13   14  15 */   
             if(i==istart){
               fprintf(ficgp,"\"%s\"",subdirf2(fileresu,"FB_"));
             }else{
               fprintf(ficgp,",\\\n '' ");
             }
             if(cptcoveff ==0){ /* No covariate */
               ioffset=2; /* Age is in 2 */
               /*# yearproj age p11 p21 p31 p.1 p12 p22 p32 p.2 p13 p23 p33 p.3 p14 p24 p34 p.4*/
               /*#   1       2   3   4   5  6    7  8   9   10  11  12  13  14  15  16  17  18 */
               /*# V1  = 1 yearproj age p11 p21 p31 p.1 p12 p22 p32 p.2 p13 p23 p33 p.3 p14 p24 p34 p.4*/
               /*#  1    2        3   4   5  6    7  8   9   10  11  12  13  14  15  16  17  18 */
               fprintf(ficgp," u %d:(", ioffset); 
               if(i==nlstate+1){
                 fprintf(ficgp," $%d/(1.-$%d)):1 t 'bw%d' with line lc variable ", \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1),  ioffset+1+(i-1)+(nlstate+1)*nlstate,cpt );
                 fprintf(ficgp,",\\\n '' ");
                 fprintf(ficgp," u %d:(",ioffset); 
                 fprintf(ficgp," (($1-$2) == %d ) ? $%d : 1/0):1 with labels center not ", \
                        offbyear,                          \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1) );
               }else
                 fprintf(ficgp," $%d/(1.-$%d)) t 'b%d%d' with line ",      \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1),  ioffset+1+(i-1)+(nlstate+1)*nlstate,cpt,i );
             }else{ /* more than 2 covariates */
               ioffset=2*cptcoveff+2; /* Age is in 4 or 6 or etc.*/
               /*#  V1  = 1  V2 =  0 yearproj age p11 p21 p.1 p12 p22 p.2 p13 p23 p.3*/
               /*#   1    2   3    4    5      6  7   8   9   10   11 12  13   14  15 */
               iyearc=ioffset-1;
               iagec=ioffset;
               fprintf(ficgp," u %d:(",ioffset); 
               kl=0;
               strcpy(gplotcondition,"(");
               for (k=1; k<=cptcoveff; k++){    /* For each covariate writing the chain of conditions */
                 /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate value corresponding to combination k1 and covariate k *\/ */
                 lv= codtabm(k1,TnsdVar[Tvaraff[k]]); /* Should be the covariate value corresponding to combination k1 and covariate k */
                 /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
                 /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
                 /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
                 /* vlv= nbcode[Tvaraff[k]][lv]; /\* Value of the modality of Tvaraff[k] *\/ */
                 vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
                 kl++;
                 sprintf(gplotcondition+strlen(gplotcondition),"$%d==%d && $%d==%d " ,kl,Tvaraff[k], kl+1, nbcode[Tvaraff[k]][lv]);
                 kl++;
                 if(k <cptcoveff && cptcoveff>1)
                   sprintf(gplotcondition+strlen(gplotcondition)," && ");
               }
               strcpy(gplotcondition+strlen(gplotcondition),")");
               /* kl=6+(cpt-1)*(nlstate+1)+1+(i-1); /\* 6+(1-1)*(2+1)+1+(1-1)=7, 6+(2-1)(2+1)+1+(1-1)=10 *\/ */
               /*6+(cpt-1)*(nlstate+1)+1+(i-1)+(nlstate+1)*nlstate; 6+(1-1)*(2+1)+1+(1-1) +(2+1)*2=13 */ 
               /*6+1+(i-1)+(nlstate+1)*nlstate; 6+1+(1-1) +(2+1)*2=13 */ 
               /* ''  u 6:(($1==1 && $2==0 && $3==2 && $4==0)? $9/(1.-$15) : 1/0):($5==2000? 3:2) t 'p.1' with line lc variable*/
               if(i==nlstate+1){
                 fprintf(ficgp,"%s ? $%d : 1/0):%d t 'bw%d' with line lc variable", gplotcondition, \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1),iyearc,cpt );
                 fprintf(ficgp,",\\\n '' ");
                 fprintf(ficgp," u %d:(",iagec); 
                 /* fprintf(ficgp,"%s && (($5-$6) == %d ) ? $%d/(1.-$%d) : 1/0):5 with labels center not ", gplotcondition, \ */
                 fprintf(ficgp,"%s && (($%d-$%d) == %d ) ? $%d : 1/0):%d with labels center not ", gplotcondition, \
                         iyearc,iagec,offbyear,                            \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1), iyearc );
   /*  '' u 6:(($1==1 && $2==0  && $3==2 && $4==0) && (($5-$6) == 1947) ? $10/(1.-$22) : 1/0):5 with labels center boxed not*/
               }else{
                 /* fprintf(ficgp,"%s ? $%d/(1.-$%d) : 1/0) t 'p%d%d' with line ", gplotcondition, \ */
                 fprintf(ficgp,"%s ? $%d : 1/0) t 'b%d%d' with line ", gplotcondition, \
                         ioffset+(cpt-1)*(nlstate+1)+1+(i-1), cpt,i );
               }
             } /* end if covariate */
           } /* nlstate */
           fprintf(ficgp,"\nset out; unset label;\n");
         } /* end cpt state*/
       } /* end covariate */
     } /* End if prevbcast */
     
     
     /* 9eme writing MLE parameters */
     fprintf(ficgp,"\n##############\n#9eme MLE estimated parameters\n#############\n");
     for(i=1,jk=1; i <=nlstate; i++){
       fprintf(ficgp,"# initial state %d\n",i);
       for(k=1; k <=(nlstate+ndeath); k++){
         if (k != i) {
           fprintf(ficgp,"#   current state %d\n",k);
           for(j=1; j <=ncovmodel; j++){
             fprintf(ficgp,"p%d=%f; ",jk,p[jk]);
             jk++; 
         }          }
      }           fprintf(ficgp,"\n");
      
 /* End theta */  
   
      trgradg =ma3x(0,nhstepm,1,nlstate*nlstate,1,npar);  
   
      for(h=0; h<=nhstepm-1; h++)  
       for(j=1; j<=nlstate*nlstate;j++)  
         for(theta=1; theta <=npar; theta++)  
           trgradg[h][j][theta]=gradg[h][theta][j];  
        
   
      for(i=1;i<=nlstate*nlstate;i++)  
       for(j=1;j<=nlstate*nlstate;j++)  
         varhe[i][j][(int)age] =0.;  
   
      printf("%d|",(int)age);fflush(stdout);  
      fprintf(ficlog,"%d|",(int)age);fflush(ficlog);  
      for(h=0;h<=nhstepm-1;h++){  
       for(k=0;k<=nhstepm-1;k++){  
         matprod2(dnewm,trgradg[h],1,nlstate*nlstate,1,npar,1,npar,matcov);  
         matprod2(doldm,dnewm,1,nlstate*nlstate,1,npar,1,nlstate*nlstate,gradg[k]);  
         for(i=1;i<=nlstate*nlstate;i++)  
           for(j=1;j<=nlstate*nlstate;j++)  
             varhe[i][j][(int)age] += doldm[i][j]*hf*hf;  
       }        }
     }      }
     /* Computing expectancies */    }
     for(i=1; i<=nlstate;i++)    fprintf(ficgp,"##############\n#\n");
       for(j=1; j<=nlstate;j++)    
         for (h=0, eij[i][j][(int)age]=0; h<=nhstepm-1; h++){    /*goto avoid;*/
           eij[i][j][(int)age] += (p3mat[i][j][h]+p3mat[i][j][h+1])/2.0*hf;    /* 10eme Graphics of probabilities or incidences using written MLE parameters */
     fprintf(ficgp,"\n##############\n#10eme Graphics of probabilities or incidences\n#############\n");
     fprintf(ficgp,"# logi(p12/p11)=a12+b12*age+c12age*age+d12*V1+e12*V1*age\n");
     fprintf(ficgp,"# logi(p12/p11)=p1 +p2*age +p3*age*age+ p4*V1+ p5*V1*age\n");
     fprintf(ficgp,"# logi(p13/p11)=a13+b13*age+c13age*age+d13*V1+e13*V1*age\n");
     fprintf(ficgp,"# logi(p13/p11)=p6 +p7*age +p8*age*age+ p9*V1+ p10*V1*age\n");
     fprintf(ficgp,"# p12+p13+p14+p11=1=p11(1+exp(a12+b12*age+c12age*age+d12*V1+e12*V1*age)\n");
     fprintf(ficgp,"#                      +exp(a13+b13*age+c13age*age+d13*V1+e13*V1*age)+...)\n");
     fprintf(ficgp,"# p11=1/(1+exp(a12+b12*age+c12age*age+d12*V1+e12*V1*age)\n");
     fprintf(ficgp,"#                      +exp(a13+b13*age+c13age*age+d13*V1+e13*V1*age)+...)\n");
     fprintf(ficgp,"# p12=exp(a12+b12*age+c12age*age+d12*V1+e12*V1*age)/\n");
     fprintf(ficgp,"#     (1+exp(a12+b12*age+c12age*age+d12*V1+e12*V1*age)\n");
     fprintf(ficgp,"#       +exp(a13+b13*age+c13age*age+d13*V1+e13*V1*age))\n");
     fprintf(ficgp,"#       +exp(a14+b14*age+c14age*age+d14*V1+e14*V1*age)+...)\n");
     fprintf(ficgp,"#\n");
     for(ng=1; ng<=3;ng++){ /* Number of graphics: first is logit, 2nd is probabilities, third is incidences per year*/
       fprintf(ficgp,"#Number of graphics: first is logit, 2nd is probabilities, third is incidences per year\n");
       fprintf(ficgp,"#model=%s \n",model);
       fprintf(ficgp,"# Type of graphic ng=%d\n",ng);
       fprintf(ficgp,"#   k1=1 to 2^%d=%d\n",cptcoveff,m);/* to be checked */
       for(k1=1; k1 <=m; k1++)  /* For each combination of covariate */
       for(nres=1; nres <= nresult; nres++){ /* For each resultline */
         if(m != 1 && TKresult[nres]!= k1)
           continue;
         fprintf(ficgp,"\n\n# Combination of dummy  k1=%d which is ",k1);
         strcpy(gplotlabel,"(");
         /*sprintf(gplotlabel+strlen(gplotlabel)," Dummy combination %d ",k1);*/
         for (k=1; k<=cptcoveff; k++){    /* For each correspondig covariate value  */
           /* lv= decodtabm(k1,k,cptcoveff); /\* Should be the covariate value corresponding to k1 combination and kth covariate *\/ */
           lv= codtabm(k1,TnsdVar[Tvaraff[k]]); /* Should be the covariate value corresponding to combination k1 and covariate k */
           /* decodtabm(1,1,4) = 1 because h=1  k= (1) 1  1  1 */
           /* decodtabm(1,2,4) = 1 because h=1  k=  1 (1) 1  1 */
           /* decodtabm(13,3,4)= 2 because h=13 k=  1  1 (2) 2 */
           /* vlv= nbcode[Tvaraff[k]][lv]; */
           vlv= nbcode[Tvaraff[k]][codtabm(k1,TnsdVar[Tvaraff[k]])];
           fprintf(ficgp," V%d=%d ",Tvaraff[k],vlv);
           sprintf(gplotlabel+strlen(gplotlabel)," V%d=%d ",Tvaraff[k],vlv);
         }
         for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
           fprintf(ficgp," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
           sprintf(gplotlabel+strlen(gplotlabel)," V%d=%f ",Tvqresult[nres][resultmodel[nres][k4]],Tqresult[nres][resultmodel[nres][k4]]);
         } 
         strcpy(gplotlabel+strlen(gplotlabel),")");
         fprintf(ficgp,"\n#\n");
         fprintf(ficgp,"\nset out \"%s_%d-%d-%d.svg\" ",subdirf2(optionfilefiname,"PE_"),k1,ng,nres);
         fprintf(ficgp,"\nset key outside ");
         /* fprintf(ficgp,"\nset label \"%s\" at graph 1.2,0.5 center rotate font \"Helvetica,12\"\n",gplotlabel); */
         fprintf(ficgp,"\nset title \"%s\" font \"Helvetica,12\"\n",gplotlabel);
         fprintf(ficgp,"\nset ter svg size 640, 480 ");
         if (ng==1){
           fprintf(ficgp,"\nset ylabel \"Value of the logit of the model\"\n"); /* exp(a12+b12*x) could be nice */
           fprintf(ficgp,"\nunset log y");
         }else if (ng==2){
           fprintf(ficgp,"\nset ylabel \"Probability\"\n");
           fprintf(ficgp,"\nset log y");
         }else if (ng==3){
           fprintf(ficgp,"\nset ylabel \"Quasi-incidence per year\"\n");
           fprintf(ficgp,"\nset log y");
         }else
           fprintf(ficgp,"\nunset title ");
         fprintf(ficgp,"\nplot  [%.f:%.f] ",ageminpar,agemaxpar);
         i=1;
         for(k2=1; k2<=nlstate; k2++) {
           k3=i;
           for(k=1; k<=(nlstate+ndeath); k++) {
             if (k != k2){
               switch( ng) {
               case 1:
                 if(nagesqr==0)
                   fprintf(ficgp," p%d+p%d*x",i,i+1);
                 else /* nagesqr =1 */
                   fprintf(ficgp," p%d+p%d*x+p%d*x*x",i,i+1,i+1+nagesqr);
                 break;
               case 2: /* ng=2 */
                 if(nagesqr==0)
                   fprintf(ficgp," exp(p%d+p%d*x",i,i+1);
                 else /* nagesqr =1 */
                   fprintf(ficgp," exp(p%d+p%d*x+p%d*x*x",i,i+1,i+1+nagesqr);
                 break;
               case 3:
                 if(nagesqr==0)
                   fprintf(ficgp," %f*exp(p%d+p%d*x",YEARM/stepm,i,i+1);
                 else /* nagesqr =1 */
                   fprintf(ficgp," %f*exp(p%d+p%d*x+p%d*x*x",YEARM/stepm,i,i+1,i+1+nagesqr);
                 break;
               }
               ij=1;/* To be checked else nbcode[0][0] wrong */
               ijp=1; /* product no age */
               /* for(j=3; j <=ncovmodel-nagesqr; j++) { */
               for(j=1; j <=cptcovt; j++) { /* For each covariate of the simplified model */
                 /* printf("Tage[%d]=%d, j=%d\n", ij, Tage[ij], j); */
                 switch(Typevar[j]){
                 case 1:
                   if(cptcovage >0){ /* V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1, 2 V5 and V1 */
                     if(j==Tage[ij]) { /* Product by age  To be looked at!!*//* Bug valgrind */
                       if(ij <=cptcovage) { /* V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1, 2 V5 and V1 */
                         if(DummyV[j]==0){/* Bug valgrind */
                           fprintf(ficgp,"+p%d*%d*x",i+j+2+nagesqr-1,Tinvresult[nres][Tvar[j]]);;
                         }else{ /* quantitative */
                           fprintf(ficgp,"+p%d*%f*x",i+j+2+nagesqr-1,Tqinvresult[nres][Tvar[j]]); /* Tqinvresult in decoderesult */
                           /* fprintf(ficgp,"+p%d*%d*x",i+j+nagesqr-1,nbcode[Tvar[j-2]][codtabm(k1,Tvar[j-2])]); */
                         }
                         ij++;
                       }
                     }
                   }
                   break;
                 case 2:
                   if(cptcovprod >0){
                     if(j==Tprod[ijp]) { /* */ 
                       /* printf("Tprod[%d]=%d, j=%d\n", ij, Tprod[ijp], j); */
                       if(ijp <=cptcovprod) { /* Product */
                         if(DummyV[Tvard[ijp][1]]==0){/* Vn is dummy */
                           if(DummyV[Tvard[ijp][2]]==0){/* Vn and Vm are dummy */
                             /* fprintf(ficgp,"+p%d*%d*%d",i+j+2+nagesqr-1,nbcode[Tvard[ijp][1]][codtabm(k1,j)],nbcode[Tvard[ijp][2]][codtabm(k1,j)]); */
                             fprintf(ficgp,"+p%d*%d*%d",i+j+2+nagesqr-1,Tinvresult[nres][Tvard[ijp][1]],Tinvresult[nres][Tvard[ijp][2]]);
                           }else{ /* Vn is dummy and Vm is quanti */
                             /* fprintf(ficgp,"+p%d*%d*%f",i+j+2+nagesqr-1,nbcode[Tvard[ijp][1]][codtabm(k1,j)],Tqinvresult[nres][Tvard[ijp][2]]); */
                             fprintf(ficgp,"+p%d*%d*%f",i+j+2+nagesqr-1,Tinvresult[nres][Tvard[ijp][1]],Tqinvresult[nres][Tvard[ijp][2]]);
                           }
                         }else{ /* Vn*Vm Vn is quanti */
                           if(DummyV[Tvard[ijp][2]]==0){
                             fprintf(ficgp,"+p%d*%d*%f",i+j+2+nagesqr-1,Tinvresult[nres][Tvard[ijp][2]],Tqinvresult[nres][Tvard[ijp][1]]);
                           }else{ /* Both quanti */
                             fprintf(ficgp,"+p%d*%f*%f",i+j+2+nagesqr-1,Tqinvresult[nres][Tvard[ijp][1]],Tqinvresult[nres][Tvard[ijp][2]]);
                           }
                         }
                         ijp++;
                       }
                     } /* end Tprod */
                   }
                   break;
                 case 0:
                   /* simple covariate */
                   /* fprintf(ficgp,"+p%d*%d",i+j+2+nagesqr-1,nbcode[Tvar[j]][codtabm(k1,j)]); /\* Valgrind bug nbcode *\/ */
                   if(Dummy[j]==0){
                     fprintf(ficgp,"+p%d*%d",i+j+2+nagesqr-1,Tinvresult[nres][Tvar[j]]); /*  */
                   }else{ /* quantitative */
                     fprintf(ficgp,"+p%d*%f",i+j+2+nagesqr-1,Tqinvresult[nres][Tvar[j]]); /* */
                     /* fprintf(ficgp,"+p%d*%d*x",i+j+nagesqr-1,nbcode[Tvar[j-2]][codtabm(k1,Tvar[j-2])]); */
                   }
                  /* end simple */
                   break;
                 default:
                   break;
                 } /* end switch */
               } /* end j */
             }else{ /* k=k2 */
               if(ng !=1 ){ /* For logit formula of log p11 is more difficult to get */
                 fprintf(ficgp," (1.");i=i-ncovmodel;
               }else
                 i=i-ncovmodel;
             }
                       
 /* if((int)age==70)printf("i=%2d,j=%2d,h=%2d,age=%3d,%9.4f,%9.4f,%9.4f\n",i,j,h,(int)age,p3mat[i][j][h],hf,eij[i][j][(int)age]);*/            if(ng != 1){
               fprintf(ficgp,")/(1");
               
               for(cpt=1; cpt <=nlstate; cpt++){ 
                 if(nagesqr==0)
                   fprintf(ficgp,"+exp(p%d+p%d*x",k3+(cpt-1)*ncovmodel,k3+(cpt-1)*ncovmodel+1);
                 else /* nagesqr =1 */
                   fprintf(ficgp,"+exp(p%d+p%d*x+p%d*x*x",k3+(cpt-1)*ncovmodel,k3+(cpt-1)*ncovmodel+1,k3+(cpt-1)*ncovmodel+1+nagesqr);
                  
                 ij=1;
                 ijp=1;
                 /* for(j=3; j <=ncovmodel-nagesqr; j++){ */
                 for(j=1; j <=cptcovt; j++) { /* For each covariate of the simplified model */
                   switch(Typevar[j]){
                   case 1:
                     if(cptcovage >0){ 
                       if(j==Tage[ij]) { /* Bug valgrind */
                         if(ij <=cptcovage) { /* Bug valgrind */
                           if(DummyV[j]==0){/* Bug valgrind */
                             /* fprintf(ficgp,"+p%d*%d*x",k3+(cpt-1)*ncovmodel+1+j-2+nagesqr,nbcode[Tvar[j-2]][codtabm(k1,j-2)]); */
                             /* fprintf(ficgp,"+p%d*%d*x",k3+(cpt-1)*ncovmodel+1+j+nagesqr,nbcode[Tvar[j]][codtabm(k1,j)]); */
                             fprintf(ficgp,"+p%d*%d*x",k3+(cpt-1)*ncovmodel+1+j+nagesqr,Tinvresult[nres][Tvar[j]]);
                             /* fprintf(ficgp,"+p%d*%d*x",i+j+2+nagesqr-1,Tinvresult[nres][Tvar[j]]);; */
                             /* fprintf(ficgp,"+p%d*%d*x",k3+(cpt-1)*ncovmodel+1+j-2+nagesqr,nbcode[Tvar[j-2]][codtabm(k1,Tvar[j-2])]); */
                           }else{ /* quantitative */
                             /* fprintf(ficgp,"+p%d*%f*x",i+j+2+nagesqr-1,Tqinvresult[nres][Tvar[j]]); /\* Tqinvresult in decoderesult *\/ */
                             fprintf(ficgp,"+p%d*%f*x",k3+(cpt-1)*ncovmodel+1+j+nagesqr,Tqinvresult[nres][Tvar[j]]); /* Tqinvresult in decoderesult */
                             /* fprintf(ficgp,"+p%d*%f*x",i+j+2+nagesqr-1,Tqinvresult[nres][Tvar[j]]); /\* Tqinvresult in decoderesult *\/ */
                             /* fprintf(ficgp,"+p%d*%d*x",i+j+nagesqr-1,nbcode[Tvar[j-2]][codtabm(k1,Tvar[j-2])]); */
                           }
                           ij++;
                         }
                       }
                     }
                     break;
                   case 2:
                     if(cptcovprod >0){
                       if(j==Tprod[ijp]) { /* */ 
                         /* printf("Tprod[%d]=%d, j=%d\n", ij, Tprod[ijp], j); */
                         if(ijp <=cptcovprod) { /* Product */
                           if(DummyV[Tvard[ijp][1]]==0){/* Vn is dummy */
                             if(DummyV[Tvard[ijp][2]]==0){/* Vn and Vm are dummy */
                               /* fprintf(ficgp,"+p%d*%d*%d",i+j+2+nagesqr-1,nbcode[Tvard[ijp][1]][codtabm(k1,j)],nbcode[Tvard[ijp][2]][codtabm(k1,j)]); */
                               fprintf(ficgp,"+p%d*%d*%d",k3+(cpt-1)*ncovmodel+1+j+nagesqr,Tinvresult[nres][Tvard[ijp][1]],Tinvresult[nres][Tvard[ijp][2]]);
                               /* fprintf(ficgp,"+p%d*%d*%d",i+j+2+nagesqr-1,Tinvresult[nres][Tvard[ijp][1]],Tinvresult[nres][Tvard[ijp][2]]); */
                             }else{ /* Vn is dummy and Vm is quanti */
                               /* fprintf(ficgp,"+p%d*%d*%f",i+j+2+nagesqr-1,nbcode[Tvard[ijp][1]][codtabm(k1,j)],Tqinvresult[nres][Tvard[ijp][2]]); */
                               fprintf(ficgp,"+p%d*%d*%f",k3+(cpt-1)*ncovmodel+1+j+nagesqr,Tinvresult[nres][Tvard[ijp][1]],Tqinvresult[nres][Tvard[ijp][2]]);
                               /* fprintf(ficgp,"+p%d*%d*%f",i+j+2+nagesqr-1,Tinvresult[nres][Tvard[ijp][1]],Tqinvresult[nres][Tvard[ijp][2]]); */
                             }
                           }else{ /* Vn*Vm Vn is quanti */
                             if(DummyV[Tvard[ijp][2]]==0){
                               fprintf(ficgp,"+p%d*%d*%f",k3+(cpt-1)*ncovmodel+1+j+nagesqr,Tinvresult[nres][Tvard[ijp][2]],Tqinvresult[nres][Tvard[ijp][1]]);
                               /* fprintf(ficgp,"+p%d*%d*%f",i+j+2+nagesqr-1,Tinvresult[nres][Tvard[ijp][2]],Tqinvresult[nres][Tvard[ijp][1]]); */
                             }else{ /* Both quanti */
                               fprintf(ficgp,"+p%d*%f*%f",k3+(cpt-1)*ncovmodel+1+j+nagesqr,Tqinvresult[nres][Tvard[ijp][1]],Tqinvresult[nres][Tvard[ijp][2]]);
                               /* fprintf(ficgp,"+p%d*%f*%f",i+j+2+nagesqr-1,Tqinvresult[nres][Tvard[ijp][1]],Tqinvresult[nres][Tvard[ijp][2]]); */
                             } 
                           }
                           ijp++;
                         }
                       } /* end Tprod */
                     } /* end if */
                     break;
                   case 0: 
                     /* simple covariate */
                     /* fprintf(ficgp,"+p%d*%d",i+j+2+nagesqr-1,nbcode[Tvar[j]][codtabm(k1,j)]); /\* Valgrind bug nbcode *\/ */
                     if(Dummy[j]==0){
                       /* fprintf(ficgp,"+p%d*%d",i+j+2+nagesqr-1,Tinvresult[nres][Tvar[j]]); /\*  *\/ */
                       fprintf(ficgp,"+p%d*%d",k3+(cpt-1)*ncovmodel+1+j+nagesqr,Tinvresult[nres][Tvar[j]]); /*  */
                       /* fprintf(ficgp,"+p%d*%d",i+j+2+nagesqr-1,Tinvresult[nres][Tvar[j]]); /\*  *\/ */
                     }else{ /* quantitative */
                       fprintf(ficgp,"+p%d*%f",k3+(cpt-1)*ncovmodel+1+j+nagesqr,Tqinvresult[nres][Tvar[j]]); /* */
                       /* fprintf(ficgp,"+p%d*%f",i+j+2+nagesqr-1,Tqinvresult[nres][Tvar[j]]); /\* *\/ */
                       /* fprintf(ficgp,"+p%d*%d*x",i+j+nagesqr-1,nbcode[Tvar[j-2]][codtabm(k1,Tvar[j-2])]); */
                     }
                     /* end simple */
                     /* fprintf(ficgp,"+p%d*%d",k3+(cpt-1)*ncovmodel+1+j-2+nagesqr,nbcode[Tvar[j-2]][codtabm(k1,j-2)]);/\* Valgrind bug nbcode *\/ */
                     break;
                   default:
                     break;
                   } /* end switch */
                 }
                 fprintf(ficgp,")");
               }
               fprintf(ficgp,")");
               if(ng ==2)
                 fprintf(ficgp," w l lw 2 lt (%d*%d+%d)%%%d+1 dt %d t \"p%d%d\" ", nlstate+ndeath, k2, k, nlstate+ndeath, k2, k2,k);
               else /* ng= 3 */
                 fprintf(ficgp," w l lw 2 lt (%d*%d+%d)%%%d+1 dt %d t \"i%d%d\" ",  nlstate+ndeath, k2, k, nlstate+ndeath, k2, k2,k);
             }else{ /* end ng <> 1 */
               if( k !=k2) /* logit p11 is hard to draw */
                 fprintf(ficgp," w l lw 2 lt (%d*%d+%d)%%%d+1 dt %d t \"logit(p%d%d)\" ",  nlstate+ndeath, k2, k, nlstate+ndeath, k2, k2,k);
             }
             if ((k+k2)!= (nlstate*2+ndeath) && ng != 1)
               fprintf(ficgp,",");
             if (ng == 1 && k!=k2 && (k+k2)!= (nlstate*2+ndeath))
               fprintf(ficgp,",");
             i=i+ncovmodel;
           } /* end k */
         } /* end k2 */
         /* fprintf(ficgp,"\n set out; unset label;set key default;\n"); */
         fprintf(ficgp,"\n set out; unset title;set key default;\n");
       } /* end k1 */
     } /* end ng */
     /* avoid: */
     fflush(ficgp); 
   }  /* end gnuplot */
   
   
   /*************** Moving average **************/
   /* int movingaverage(double ***probs, double bage, double fage, double ***mobaverage, int mobilav, double bageout, double fageout){ */
    int movingaverage(double ***probs, double bage, double fage, double ***mobaverage, int mobilav){
      
      int i, cpt, cptcod;
      int modcovmax =1;
      int mobilavrange, mob;
      int iage=0;
      int firstA1=0, firstA2=0;
   
      double sum=0., sumr=0.;
      double age;
      double *sumnewp, *sumnewm, *sumnewmr;
      double *agemingood, *agemaxgood; 
      double *agemingoodr, *agemaxgoodr; 
     
     
      /* modcovmax=2*cptcoveff;  Max number of modalities. We suppose  */
      /*              a covariate has 2 modalities, should be equal to ncovcombmax   */
   
      sumnewp = vector(1,ncovcombmax);
      sumnewm = vector(1,ncovcombmax);
      sumnewmr = vector(1,ncovcombmax);
      agemingood = vector(1,ncovcombmax);  
      agemingoodr = vector(1,ncovcombmax); 
      agemaxgood = vector(1,ncovcombmax);
      agemaxgoodr = vector(1,ncovcombmax);
   
      for (cptcod=1;cptcod<=ncovcombmax;cptcod++){
        sumnewm[cptcod]=0.; sumnewmr[cptcod]=0.;
        sumnewp[cptcod]=0.;
        agemingood[cptcod]=0, agemingoodr[cptcod]=0;
        agemaxgood[cptcod]=0, agemaxgoodr[cptcod]=0;
      }
      if (cptcovn<1) ncovcombmax=1; /* At least 1 pass */
     
      if(mobilav==-1 || mobilav==1||mobilav ==3 ||mobilav==5 ||mobilav== 7){
        if(mobilav==1 || mobilav==-1) mobilavrange=5; /* default */
        else mobilavrange=mobilav;
        for (age=bage; age<=fage; age++)
          for (i=1; i<=nlstate;i++)
            for (cptcod=1;cptcod<=ncovcombmax;cptcod++)
              mobaverage[(int)age][i][cptcod]=probs[(int)age][i][cptcod];
        /* We keep the original values on the extreme ages bage, fage and for 
           fage+1 and bage-1 we use a 3 terms moving average; for fage+2 bage+2
           we use a 5 terms etc. until the borders are no more concerned. 
        */ 
        for (mob=3;mob <=mobilavrange;mob=mob+2){
          for (age=bage+(mob-1)/2; age<=fage-(mob-1)/2; age++){
            for (cptcod=1;cptcod<=ncovcombmax;cptcod++){
              sumnewm[cptcod]=0.;
              for (i=1; i<=nlstate;i++){
                mobaverage[(int)age][i][cptcod] =probs[(int)age][i][cptcod];
                for (cpt=1;cpt<=(mob-1)/2;cpt++){
                  mobaverage[(int)age][i][cptcod] +=probs[(int)age-cpt][i][cptcod];
                  mobaverage[(int)age][i][cptcod] +=probs[(int)age+cpt][i][cptcod];
                }
                mobaverage[(int)age][i][cptcod]=mobaverage[(int)age][i][cptcod]/mob;
                sumnewm[cptcod]+=mobaverage[(int)age][i][cptcod];
              } /* end i */
              if(sumnewm[cptcod] >1.e-3) mobaverage[(int)age][i][cptcod]=mobaverage[(int)age][i][cptcod]/sumnewm[cptcod]; /* Rescaling to sum one */
            } /* end cptcod */
          }/* end age */
        }/* end mob */
      }else{
        printf("Error internal in movingaverage, mobilav=%d.\n",mobilav);
        return -1;
      }
   
         }     for (cptcod=1;cptcod<=ncovcombmax;cptcod++){ /* for each combination */
        /* for (age=bage+(mob-1)/2; age<=fage-(mob-1)/2; age++){ */
        if(invalidvarcomb[cptcod]){
          printf("\nCombination (%d) ignored because no cases \n",cptcod); 
          continue;
        }
   
     fprintf(ficreseij,"%3.0f",age );       for (age=fage-(mob-1)/2; age>=bage+(mob-1)/2; age--){ /*looking for the youngest and oldest good age */
     cptj=0;         sumnewm[cptcod]=0.;
     for(i=1; i<=nlstate;i++)         sumnewmr[cptcod]=0.;
       for(j=1; j<=nlstate;j++){         for (i=1; i<=nlstate;i++){
         cptj++;           sumnewm[cptcod]+=mobaverage[(int)age][i][cptcod];
         fprintf(ficreseij," %9.4f (%.4f)", eij[i][j][(int)age], sqrt(varhe[cptj][cptj][(int)age]) );           sumnewmr[cptcod]+=probs[(int)age][i][cptcod];
       }         }
     fprintf(ficreseij,"\n");         if(fabs(sumnewmr[cptcod] - 1.) <= 1.e-3) { /* good without smoothing */
               agemingoodr[cptcod]=age;
     free_matrix(gm,0,nhstepm,1,nlstate*nlstate);         }
     free_matrix(gp,0,nhstepm,1,nlstate*nlstate);         if(fabs(sumnewm[cptcod] - 1.) <= 1.e-3) { /* good */
     free_ma3x(gradg,0,nhstepm,1,npar,1,nlstate*nlstate);             agemingood[cptcod]=age;
     free_ma3x(trgradg,0,nhstepm,1,nlstate*nlstate,1,npar);         }
     free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);       } /* age */
   }       for (age=bage+(mob-1)/2; age<=fage-(mob-1)/2; age++){ /*looking for the youngest and oldest good age */
   printf("\n");         sumnewm[cptcod]=0.;
   fprintf(ficlog,"\n");         sumnewmr[cptcod]=0.;
          for (i=1; i<=nlstate;i++){
            sumnewm[cptcod]+=mobaverage[(int)age][i][cptcod];
            sumnewmr[cptcod]+=probs[(int)age][i][cptcod];
          }
          if(fabs(sumnewmr[cptcod] - 1.) <= 1.e-3) { /* good without smoothing */
            agemaxgoodr[cptcod]=age;
          }
          if(fabs(sumnewm[cptcod] - 1.) <= 1.e-3) { /* good */
            agemaxgood[cptcod]=age;
          }
        } /* age */
        /* Thus we have agemingood and agemaxgood as well as goodr for raw (preobs) */
        /* but they will change */
        firstA1=0;firstA2=0;
        for (age=fage-(mob-1)/2; age>=bage; age--){/* From oldest to youngest, filling up to the youngest */
          sumnewm[cptcod]=0.;
          sumnewmr[cptcod]=0.;
          for (i=1; i<=nlstate;i++){
            sumnewm[cptcod]+=mobaverage[(int)age][i][cptcod];
            sumnewmr[cptcod]+=probs[(int)age][i][cptcod];
          }
          if(mobilav==-1){ /* Forcing raw ages if good else agemingood */
            if(fabs(sumnewmr[cptcod] - 1.) <= 1.e-3) { /* good without smoothing */
              agemaxgoodr[cptcod]=age;  /* age min */
              for (i=1; i<=nlstate;i++)
                mobaverage[(int)age][i][cptcod]=probs[(int)age][i][cptcod];
            }else{ /* bad we change the value with the values of good ages */
              for (i=1; i<=nlstate;i++){
                mobaverage[(int)age][i][cptcod]=mobaverage[(int)agemaxgoodr[cptcod]][i][cptcod];
              } /* i */
            } /* end bad */
          }else{
            if(fabs(sumnewm[cptcod] - 1.) <= 1.e-3) { /* good */
              agemaxgood[cptcod]=age;
            }else{ /* bad we change the value with the values of good ages */
              for (i=1; i<=nlstate;i++){
                mobaverage[(int)age][i][cptcod]=mobaverage[(int)agemaxgood[cptcod]][i][cptcod];
              } /* i */
            } /* end bad */
          }/* end else */
          sum=0.;sumr=0.;
          for (i=1; i<=nlstate;i++){
            sum+=mobaverage[(int)age][i][cptcod];
            sumr+=probs[(int)age][i][cptcod];
          }
          if(fabs(sum - 1.) > 1.e-3) { /* bad */
            if(!firstA1){
              firstA1=1;
              printf("Moving average A1: For this combination of covariate cptcod=%d, we can't get a smoothed prevalence which sums to one (%f) at any descending age! age=%d, could you increase bage=%d. Others in log file...\n",cptcod,sumr, (int)age, (int)bage);
            }
            fprintf(ficlog,"Moving average A1: For this combination of covariate cptcod=%d, we can't get a smoothed prevalence which sums to one (%f) at any descending age! age=%d, could you increase bage=%d\n",cptcod,sumr, (int)age, (int)bage);
          } /* end bad */
          /* else{ /\* We found some ages summing to one, we will smooth the oldest *\/ */
          if(fabs(sumr - 1.) > 1.e-3) { /* bad */
            if(!firstA2){
              firstA2=1;
              printf("Moving average A2: For this combination of covariate cptcod=%d, the raw prevalence doesn't sums to one (%f) even with smoothed values at young ages! age=%d, could you increase bage=%d. Others in log file...\n",cptcod,sumr, (int)age, (int)bage);
            }
            fprintf(ficlog,"Moving average A2: For this combination of covariate cptcod=%d, the raw prevalence doesn't sums to one (%f) even with smoothed values at young ages! age=%d, could you increase bage=%d\n",cptcod,sumr, (int)age, (int)bage);
          } /* end bad */
        }/* age */
   
        for (age=bage+(mob-1)/2; age<=fage; age++){/* From youngest, finding the oldest wrong */
          sumnewm[cptcod]=0.;
          sumnewmr[cptcod]=0.;
          for (i=1; i<=nlstate;i++){
            sumnewm[cptcod]+=mobaverage[(int)age][i][cptcod];
            sumnewmr[cptcod]+=probs[(int)age][i][cptcod];
          } 
          if(mobilav==-1){ /* Forcing raw ages if good else agemingood */
            if(fabs(sumnewmr[cptcod] - 1.) <= 1.e-3) { /* good */
              agemingoodr[cptcod]=age;
              for (i=1; i<=nlstate;i++)
                mobaverage[(int)age][i][cptcod]=probs[(int)age][i][cptcod];
            }else{ /* bad we change the value with the values of good ages */
              for (i=1; i<=nlstate;i++){
                mobaverage[(int)age][i][cptcod]=mobaverage[(int)agemingoodr[cptcod]][i][cptcod];
              } /* i */
            } /* end bad */
          }else{
            if(fabs(sumnewm[cptcod] - 1.) <= 1.e-3) { /* good */
              agemingood[cptcod]=age;
            }else{ /* bad */
              for (i=1; i<=nlstate;i++){
                mobaverage[(int)age][i][cptcod]=mobaverage[(int)agemingood[cptcod]][i][cptcod];
              } /* i */
            } /* end bad */
          }/* end else */
          sum=0.;sumr=0.;
          for (i=1; i<=nlstate;i++){
            sum+=mobaverage[(int)age][i][cptcod];
            sumr+=mobaverage[(int)age][i][cptcod];
          }
          if(fabs(sum - 1.) > 1.e-3) { /* bad */
            printf("Moving average B1: For this combination of covariate cptcod=%d, we can't get a smoothed prevalence which sums to one (%f) at any descending age! age=%d, could you decrease fage=%d?\n",cptcod, sum, (int) age, (int)fage);
          } /* end bad */
          /* else{ /\* We found some ages summing to one, we will smooth the oldest *\/ */
          if(fabs(sumr - 1.) > 1.e-3) { /* bad */
            printf("Moving average B2: For this combination of covariate cptcod=%d, the raw prevalence doesn't sums to one (%f) even with smoothed values at young ages! age=%d, could you increase fage=%d\n",cptcod,sumr, (int)age, (int)fage);
          } /* end bad */
        }/* age */
   
                   
        for (age=bage; age<=fage; age++){
          /* printf("%d %d ", cptcod, (int)age); */
          sumnewp[cptcod]=0.;
          sumnewm[cptcod]=0.;
          for (i=1; i<=nlstate;i++){
            sumnewp[cptcod]+=probs[(int)age][i][cptcod];
            sumnewm[cptcod]+=mobaverage[(int)age][i][cptcod];
            /* printf("%.4f %.4f ",probs[(int)age][i][cptcod], mobaverage[(int)age][i][cptcod]); */
          }
          /* printf("%.4f %.4f \n",sumnewp[cptcod], sumnewm[cptcod]); */
        }
        /* printf("\n"); */
        /* } */
   
   free_vector(xp,1,npar);       /* brutal averaging */
   free_matrix(dnewm,1,nlstate*nlstate,1,npar);       /* for (i=1; i<=nlstate;i++){ */
   free_matrix(doldm,1,nlstate*nlstate,1,nlstate*nlstate);       /*   for (age=1; age<=bage; age++){ */
   free_ma3x(varhe,1,nlstate*nlstate,1,nlstate*nlstate,(int) bage, (int)fage);       /*          mobaverage[(int)age][i][cptcod]=mobaverage[(int)agemingood[cptcod]][i][cptcod]; */
 }       /*          /\* printf("age=%d i=%d cptcod=%d mobaverage=%.4f \n",(int)age,i, cptcod, mobaverage[(int)age][i][cptcod]); *\/ */
        /*   }      */
        /*   for (age=fage; age<=AGESUP; age++){ */
        /*          mobaverage[(int)age][i][cptcod]=mobaverage[(int)agemaxgood[cptcod]][i][cptcod]; */
        /*          /\* printf("age=%d i=%d cptcod=%d mobaverage=%.4f \n",(int)age,i, cptcod, mobaverage[(int)age][i][cptcod]); *\/ */
        /*   } */
        /* } /\* end i status *\/ */
        /* for (i=nlstate+1; i<=nlstate+ndeath;i++){ */
        /*   for (age=1; age<=AGESUP; age++){ */
        /*          /\*printf("i=%d, age=%d, cptcod=%d\n",i, (int)age, cptcod);*\/ */
        /*          mobaverage[(int)age][i][cptcod]=0.; */
        /*   } */
        /* } */
      }/* end cptcod */
      free_vector(agemaxgoodr,1, ncovcombmax);
      free_vector(agemaxgood,1, ncovcombmax);
      free_vector(agemingood,1, ncovcombmax);
      free_vector(agemingoodr,1, ncovcombmax);
      free_vector(sumnewmr,1, ncovcombmax);
      free_vector(sumnewm,1, ncovcombmax);
      free_vector(sumnewp,1, ncovcombmax);
      return 0;
    }/* End movingaverage */
    
   
 /************ Variance ******************/   
 void varevsij(char optionfilefiname[], double ***vareij, double **matcov, double x[], double delti[], int nlstate, int stepm, double bage, double fage, double **oldm, double **savm, double **prlim, double ftolpl, int ij, int estepm, int cptcov, int cptcod, int popbased, int mobilav)  /************** Forecasting ******************/
 {  /* void prevforecast(char fileres[], double dateintmean, double anprojd, double mprojd, double jprojd, double ageminpar, double agemax, double dateprev1, double dateprev2, int mobilav, double ***prev, double bage, double fage, int firstpass, int lastpass, double anprojf, double p[], int cptcoveff)*/
   /* Variance of health expectancies */  void prevforecast(char fileres[], double dateintmean, double dateprojd, double dateprojf, double ageminpar, double agemax, double dateprev1, double dateprev2, int mobilav, double ***prev, double bage, double fage, int firstpass, int lastpass, double p[], int cptcoveff){
   /*  double **prevalim(double **prlim, int nlstate, double *xp, double age, double **oldm, double ** savm,double ftolpl);*/    /* dateintemean, mean date of interviews
   /* double **newm;*/       dateprojd, year, month, day of starting projection 
   double **dnewm,**doldm;       dateprojf date of end of projection;year of end of projection (same day and month as proj1).
   double **dnewmp,**doldmp;       agemin, agemax range of age
   int i, j, nhstepm, hstepm, h, nstepm ;       dateprev1 dateprev2 range of dates during which prevalence is computed
   int k, cptcode;    */
   double *xp;    /* double anprojd, mprojd, jprojd; */
   double **gp, **gm;  /* for var eij */    /* double anprojf, mprojf, jprojf; */
   double ***gradg, ***trgradg; /*for var eij */    int yearp, stepsize, hstepm, nhstepm, j, k, cptcod, i, h, i1, k4, nres=0;
   double **gradgp, **trgradgp; /* for var p point j */    double agec; /* generic age */
   double *gpp, *gmp; /* for var p point j */    double agelim, ppij, yp,yp1,yp2;
   double **varppt; /* for var p point j nlstate to nlstate+ndeath */    double *popeffectif,*popcount;
   double ***p3mat;    double ***p3mat;
   double age,agelim, hf;    /* double ***mobaverage; */
   double ***mobaverage;    char fileresf[FILENAMELENGTH];
   int theta;  
   char digit[4];  
   char digitp[25];  
   
   char fileresprobmorprev[FILENAMELENGTH];  
   
   if(popbased==1){    agelim=AGESUP;
     if(mobilav!=0)    /* Compute observed prevalence between dateprev1 and dateprev2 by counting the number of people
       strcpy(digitp,"-populbased-mobilav-");       in each health status at the date of interview (if between dateprev1 and dateprev2).
     else strcpy(digitp,"-populbased-nomobil-");       We still use firstpass and lastpass as another selection.
   }    */
   else     /* freqsummary(fileres, agemin, agemax, s, agev, nlstate, imx,Tvaraff,nbcode, ncodemax,mint,anint,strstart,\ */
     strcpy(digitp,"-stablbased-");    /*          firstpass, lastpass,  stepm,  weightopt, model); */
    
   if (mobilav!=0) {    strcpy(fileresf,"F_"); 
     mobaverage= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX);    strcat(fileresf,fileresu);
     if (movingaverage(probs, bage, fage, mobaverage,mobilav)!=0){    if((ficresf=fopen(fileresf,"w"))==NULL) {
       fprintf(ficlog," Error in movingaverage mobilav=%d\n",mobilav);      printf("Problem with forecast resultfile: %s\n", fileresf);
       printf(" Error in movingaverage mobilav=%d\n",mobilav);      fprintf(ficlog,"Problem with forecast resultfile: %s\n", fileresf);
     }  
   }    }
     printf("\nComputing forecasting: result on file '%s', please wait... \n", fileresf);
     fprintf(ficlog,"\nComputing forecasting: result on file '%s', please wait... \n", fileresf);
   
   strcpy(fileresprobmorprev,"prmorprev");     if (cptcoveff==0) ncodemax[cptcoveff]=1;
   sprintf(digit,"%-d",ij);  
   /*printf("DIGIT=%s, ij=%d ijr=%-d|\n",digit, ij,ij);*/  
   strcat(fileresprobmorprev,digit); /* Tvar to be done */    stepsize=(int) (stepm+YEARM-1)/YEARM;
   strcat(fileresprobmorprev,digitp); /* Popbased or not, mobilav or not */    if (stepm<=12) stepsize=1;
   strcat(fileresprobmorprev,fileres);    if(estepm < stepm){
   if((ficresprobmorprev=fopen(fileresprobmorprev,"w"))==NULL) {      printf ("Problem %d lower than %d\n",estepm, stepm);
     printf("Problem with resultfile: %s\n", fileresprobmorprev);  
     fprintf(ficlog,"Problem with resultfile: %s\n", fileresprobmorprev);  
   }  
   printf("Computing total mortality p.j=w1*p1j+w2*p2j+..: result on file '%s' \n",fileresprobmorprev);  
   fprintf(ficlog,"Computing total mortality p.j=w1*p1j+w2*p2j+..: result on file '%s' \n",fileresprobmorprev);  
   fprintf(ficresprobmorprev,"# probabilities of dying before estepm=%d months for people of exact age and weighted probabilities w1*p1j+w2*p2j+... stand dev in()\n",estepm);  
   fprintf(ficresprobmorprev,"# Age cov=%-d",ij);  
   for(j=nlstate+1; j<=(nlstate+ndeath);j++){  
     fprintf(ficresprobmorprev," p.%-d SE",j);  
     for(i=1; i<=nlstate;i++)  
       fprintf(ficresprobmorprev," w%1d p%-d%-d",i,i,j);  
   }    
   fprintf(ficresprobmorprev,"\n");  
   if((ficgp=fopen(optionfilegnuplot,"a"))==NULL) {  
     printf("Problem with gnuplot file: %s\n", optionfilegnuplot);  
     fprintf(ficlog,"Problem with gnuplot file: %s\n", optionfilegnuplot);  
     exit(0);  
   }    }
   else{    else{
     fprintf(ficgp,"\n# Routine varevsij");      hstepm=estepm;   
   }  
   if((fichtm=fopen(optionfilehtm,"a"))==NULL) {  
     printf("Problem with html file: %s\n", optionfilehtm);  
     fprintf(ficlog,"Problem with html file: %s\n", optionfilehtm);  
     exit(0);  
   }    }
   else{    if(estepm > stepm){ /* Yes every two year */
     fprintf(fichtm,"\n<li><h4> Computing probabilities of dying over estepm months as a weighted average (i.e global mortality independent of initial healh state)</h4></li>\n");      stepsize=2;
     fprintf(fichtm,"\n<br>%s  <br>\n",digitp);  
   }    }
   varppt = matrix(nlstate+1,nlstate+ndeath,nlstate+1,nlstate+ndeath);    hstepm=hstepm/stepm;
   
   fprintf(ficresvij,"# Variance and covariance of health expectancies e.j \n#  (weighted average of eij where weights are the stable prevalence in health states i\n");    
   fprintf(ficresvij,"# Age");    /* yp1=modf(dateintmean,&yp);/\* extracts integral of datemean in yp  and */
   for(i=1; i<=nlstate;i++)    /*                              fractional in yp1 *\/ */
     for(j=1; j<=nlstate;j++)    /* aintmean=yp; */
       fprintf(ficresvij," Cov(e%1d, e%1d)",i,j);    /* yp2=modf((yp1*12),&yp); */
   fprintf(ficresvij,"\n");    /* mintmean=yp; */
     /* yp1=modf((yp2*30.5),&yp); */
     /* jintmean=yp; */
     /* if(jintmean==0) jintmean=1; */
     /* if(mintmean==0) mintmean=1; */
   
   xp=vector(1,npar);  
   dnewm=matrix(1,nlstate,1,npar);  
   doldm=matrix(1,nlstate,1,nlstate);  
   dnewmp= matrix(nlstate+1,nlstate+ndeath,1,npar);  
   doldmp= matrix(nlstate+1,nlstate+ndeath,nlstate+1,nlstate+ndeath);  
   
   gradgp=matrix(1,npar,nlstate+1,nlstate+ndeath);    /* date2dmy(dateintmean,&jintmean,&mintmean,&aintmean); */
   gpp=vector(nlstate+1,nlstate+ndeath);    /* date2dmy(dateprojd,&jprojd, &mprojd, &anprojd); */
   gmp=vector(nlstate+1,nlstate+ndeath);    /* date2dmy(dateprojf,&jprojf, &mprojf, &anprojf); */
   trgradgp =matrix(nlstate+1,nlstate+ndeath,1,npar); /* mu or p point j*/    i1=pow(2,cptcoveff);
     if (cptcovn < 1){i1=1;}
       
   if(estepm < stepm){    fprintf(ficresf,"# Mean day of interviews %.lf/%.lf/%.lf (%.2f) between %.2f and %.2f \n",jintmean,mintmean,aintmean,dateintmean,dateprev1,dateprev2); 
     printf ("Problem %d lower than %d\n",estepm, stepm);    
   }    fprintf(ficresf,"#****** Routine prevforecast **\n");
   else  hstepm=estepm;       
   /* For example we decided to compute the life expectancy with the smallest unit */  /*            if (h==(int)(YEARM*yearp)){ */
   /* hstepm beeing the number of stepms, if hstepm=1 the length of hstepm is stepm.     for(nres=1; nres <= nresult; nres++) /* For each resultline */
      nhstepm is the number of hstepm from age to agelim       for(k=1; k<=i1;k++){ /* We want to find the combination k corresponding to the values of the dummies given in this resut line (to be cleaned one day) */
      nstepm is the number of stepm from age to agelin.       if(i1 != 1 && TKresult[nres]!= k)
      Look at hpijx to understand the reason of that which relies in memory size        continue;
      and note for a fixed period like k years */      if(invalidvarcomb[k]){
   /* We decided (b) to get a life expectancy respecting the most precise curvature of the        printf("\nCombination (%d) projection ignored because no cases \n",k); 
      survival function given by stepm (the optimization length). Unfortunately it        continue;
      means that if the survival funtion is printed every two years of age and if      }
      you sum them up and add 1 year (area under the trapezoids) you won't get the same       fprintf(ficresf,"\n#****** hpijx=probability over h years, hp.jx is weighted by observed prev \n#");
      results. So we changed our mind and took the option of the best precision.      for(j=1;j<=cptcoveff;j++) {
   */        /* fprintf(ficresf," V%d (=) %d",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,Tvaraff[j])]); */
   hstepm=hstepm/stepm; /* Typically in stepm units, if stepm=6 & estepm=24 , = 24/6 months = 4 */         fprintf(ficresf," V%d (=) %d",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
   agelim = AGESUP;      }
   for (age=bage; age<=fage; age ++){ /* If stepm=6 months */      for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
     nstepm=(int) rint((agelim-age)*YEARM/stepm); /* Typically 20 years = 20*12/6=40 */         fprintf(ficresf," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
     nhstepm = nstepm/hstepm;/* Expressed in hstepm, typically nhstepm=40/4=10 */      }
     p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);      fprintf(ficresf," yearproj age");
     gradg=ma3x(0,nhstepm,1,npar,1,nlstate);      for(j=1; j<=nlstate+ndeath;j++){ 
     gp=matrix(0,nhstepm,1,nlstate);        for(i=1; i<=nlstate;i++)        
     gm=matrix(0,nhstepm,1,nlstate);          fprintf(ficresf," p%d%d",i,j);
         fprintf(ficresf," wp.%d",j);
       }
       for (yearp=0; yearp<=(anprojf-anprojd);yearp +=stepsize) {
         fprintf(ficresf,"\n");
         fprintf(ficresf,"\n# Forecasting at date %.lf/%.lf/%.lf ",jprojd,mprojd,anprojd+yearp);   
         /* for (agec=fage; agec>=(ageminpar-1); agec--){  */
         for (agec=fage; agec>=(bage); agec--){ 
           nhstepm=(int) rint((agelim-agec)*YEARM/stepm); 
           nhstepm = nhstepm/hstepm; 
           p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
           oldm=oldms;savm=savms;
           /* We compute pii at age agec over nhstepm);*/
           hpxij(p3mat,nhstepm,agec,hstepm,p,nlstate,stepm,oldm,savm, k,nres);
           /* Then we print p3mat for h corresponding to the right agec+h*stepms=yearp */
           for (h=0; h<=nhstepm; h++){
             if (h*hstepm/YEARM*stepm ==yearp) {
               break;
             }
           }
           fprintf(ficresf,"\n");
           for(j=1;j<=cptcoveff;j++) 
             /* fprintf(ficresf,"%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,Tvaraff[j])]); /\* Tvaraff not correct *\/ */
             fprintf(ficresf,"%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]); /* TnsdVar[Tvaraff]  correct */
           fprintf(ficresf,"%.f %.f ",anprojd+yearp,agec+h*hstepm/YEARM*stepm);
           
           for(j=1; j<=nlstate+ndeath;j++) {
             ppij=0.;
             for(i=1; i<=nlstate;i++) {
               if (mobilav>=1)
                ppij=ppij+p3mat[i][j][h]*prev[(int)agec][i][k];
               else { /* even if mobilav==-1 we use mobaverage, probs may not sums to 1 */
                   ppij=ppij+p3mat[i][j][h]*probs[(int)(agec)][i][k];
               }
               fprintf(ficresf," %.3f", p3mat[i][j][h]);
             } /* end i */
             fprintf(ficresf," %.3f", ppij);
           }/* end j */
           free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
         } /* end agec */
         /* diffyear=(int) anproj1+yearp-ageminpar-1; */
         /*printf("Prevforecast %d+%d-%d=diffyear=%d\n",(int) anproj1, (int)yearp,(int)ageminpar,(int) anproj1-(int)ageminpar);*/
       } /* end yearp */
     } /* end  k */
           
     fclose(ficresf);
     printf("End of Computing forecasting \n");
     fprintf(ficlog,"End of Computing forecasting\n");
   
   }
   
     for(theta=1; theta <=npar; theta++){  /************** Back Forecasting ******************/
       for(i=1; i<=npar; i++){ /* Computes gradient x + delta*/   /* void prevbackforecast(char fileres[], double ***prevacurrent, double anback1, double mback1, double jback1, double ageminpar, double agemax, double dateprev1, double dateprev2, int mobilav, double bage, double fage, int firstpass, int lastpass, double anback2, double p[], int cptcoveff){ */
         xp[i] = x[i] + (i==theta ?delti[theta]:0);   void prevbackforecast(char fileres[], double ***prevacurrent, double dateintmean, double dateprojd, double dateprojf, double ageminpar, double agemax, double dateprev1, double dateprev2, int mobilav, double bage, double fage, int firstpass, int lastpass, double p[], int cptcoveff){
       }    /* back1, year, month, day of starting backprojection
       hpxij(p3mat,nhstepm,age,hstepm,xp,nlstate,stepm,oldm,savm, ij);         agemin, agemax range of age
       prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ij);       dateprev1 dateprev2 range of dates during which prevalence is computed
        anback2 year of end of backprojection (same day and month as back1).
        prevacurrent and prev are prevalences.
     */
     int yearp, stepsize, hstepm, nhstepm, j, k, cptcod, i, h, i1, k4, nres=0;
     double agec; /* generic age */
     double agelim, ppij, ppi, yp,yp1,yp2; /* ,jintmean,mintmean,aintmean;*/
     double *popeffectif,*popcount;
     double ***p3mat;
     /* double ***mobaverage; */
     char fileresfb[FILENAMELENGTH];
    
     agelim=AGEINF;
     /* Compute observed prevalence between dateprev1 and dateprev2 by counting the number of people
        in each health status at the date of interview (if between dateprev1 and dateprev2).
        We still use firstpass and lastpass as another selection.
     */
     /* freqsummary(fileres, agemin, agemax, s, agev, nlstate, imx,Tvaraff,nbcode, ncodemax,mint,anint,strstart,\ */
     /*          firstpass, lastpass,  stepm,  weightopt, model); */
   
       if (popbased==1) {    /*Do we need to compute prevalence again?*/
         if(mobilav ==0){  
           for(i=1; i<=nlstate;i++)    /* prevalence(probs, ageminpar, agemax, s, agev, nlstate, imx, Tvar, nbcode, ncodemax, mint, anint, dateprev1, dateprev2, firstpass, lastpass); */
             prlim[i][i]=probs[(int)age][i][ij];  
         }else{ /* mobilav */   
           for(i=1; i<=nlstate;i++)  
             prlim[i][i]=mobaverage[(int)age][i][ij];  
         }  
       }  
       
       for(j=1; j<= nlstate; j++){    strcpy(fileresfb,"FB_");
         for(h=0; h<=nhstepm; h++){    strcat(fileresfb,fileresu);
           for(i=1, gp[h][j]=0.;i<=nlstate;i++)    if((ficresfb=fopen(fileresfb,"w"))==NULL) {
             gp[h][j] += prlim[i][i]*p3mat[i][j][h];      printf("Problem with back forecast resultfile: %s\n", fileresfb);
       fprintf(ficlog,"Problem with back forecast resultfile: %s\n", fileresfb);
     }
     printf("\nComputing back forecasting: result on file '%s', please wait... \n", fileresfb);
     fprintf(ficlog,"\nComputing back forecasting: result on file '%s', please wait... \n", fileresfb);
     
     if (cptcoveff==0) ncodemax[cptcoveff]=1;
     
      
     stepsize=(int) (stepm+YEARM-1)/YEARM;
     if (stepm<=12) stepsize=1;
     if(estepm < stepm){
       printf ("Problem %d lower than %d\n",estepm, stepm);
     }
     else{
       hstepm=estepm;   
     }
     if(estepm >= stepm){ /* Yes every two year */
       stepsize=2;
     }
     
     hstepm=hstepm/stepm;
     /* yp1=modf(dateintmean,&yp);/\* extracts integral of datemean in yp  and */
     /*                              fractional in yp1 *\/ */
     /* aintmean=yp; */
     /* yp2=modf((yp1*12),&yp); */
     /* mintmean=yp; */
     /* yp1=modf((yp2*30.5),&yp); */
     /* jintmean=yp; */
     /* if(jintmean==0) jintmean=1; */
     /* if(mintmean==0) jintmean=1; */
     
     i1=pow(2,cptcoveff);
     if (cptcovn < 1){i1=1;}
     
     fprintf(ficresfb,"# Mean day of interviews %.lf/%.lf/%.lf (%.2f) between %.2f and %.2f \n",jintmean,mintmean,aintmean,dateintmean,dateprev1,dateprev2);
     printf("# Mean day of interviews %.lf/%.lf/%.lf (%.2f) between %.2f and %.2f \n",jintmean,mintmean,aintmean,dateintmean,dateprev1,dateprev2);
     
     fprintf(ficresfb,"#****** Routine prevbackforecast **\n");
     
     for(nres=1; nres <= nresult; nres++) /* For each resultline */
     for(k=1; k<=i1;k++){
       if(i1 != 1 && TKresult[nres]!= k)
         continue;
       if(invalidvarcomb[k]){
         printf("\nCombination (%d) projection ignored because no cases \n",k); 
         continue;
       }
       fprintf(ficresfb,"\n#****** hbijx=probability over h years, hb.jx is weighted by observed prev \n#");
       for(j=1;j<=cptcoveff;j++) {
         fprintf(ficresfb," V%d (=) %d",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
       }
       for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
         fprintf(ficresf," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
       }
       fprintf(ficresfb," yearbproj age");
       for(j=1; j<=nlstate+ndeath;j++){
         for(i=1; i<=nlstate;i++)
           fprintf(ficresfb," b%d%d",i,j);
         fprintf(ficresfb," b.%d",j);
       }
       for (yearp=0; yearp>=(anbackf-anbackd);yearp -=stepsize) {
         /* for (yearp=0; yearp<=(anproj2-anproj1);yearp +=stepsize) {  */
         fprintf(ficresfb,"\n");
         fprintf(ficresfb,"\n# Back Forecasting at date %.lf/%.lf/%.lf ",jbackd,mbackd,anbackd+yearp);
         /* printf("\n# Back Forecasting at date %.lf/%.lf/%.lf ",jback1,mback1,anback1+yearp); */
         /* for (agec=bage; agec<=agemax-1; agec++){  /\* testing *\/ */
         for (agec=bage; agec<=fage; agec++){  /* testing */
           /* We compute bij at age agec over nhstepm, nhstepm decreases when agec increases because of agemax;*/
           nhstepm=(int) (agec-agelim) *YEARM/stepm;/*     nhstepm=(int) rint((agec-agelim)*YEARM/stepm);*/
           nhstepm = nhstepm/hstepm;
           p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
           oldm=oldms;savm=savms;
           /* computes hbxij at age agec over 1 to nhstepm */
           /* printf("####prevbackforecast debug  agec=%.2f nhstepm=%d\n",agec, nhstepm);fflush(stdout); */
           hbxij(p3mat,nhstepm,agec,hstepm,p,prevacurrent,nlstate,stepm, k, nres);
           /* hpxij(p3mat,nhstepm,agec,hstepm,p,             nlstate,stepm,oldm,savm, k,nres); */
           /* Then we print p3mat for h corresponding to the right agec+h*stepms=yearp */
           /* printf(" agec=%.2f\n",agec);fflush(stdout); */
           for (h=0; h<=nhstepm; h++){
             if (h*hstepm/YEARM*stepm ==-yearp) {
               break;
             }
         }          }
       }          fprintf(ficresfb,"\n");
       /* This for computing probability of death (h=1 means          for(j=1;j<=cptcoveff;j++)
          computed over hstepm matrices product = hstepm*stepm months)             fprintf(ficresfb,"%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
          as a weighted average of prlim.          fprintf(ficresfb,"%.f %.f ",anbackd+yearp,agec-h*hstepm/YEARM*stepm);
       */          for(i=1; i<=nlstate+ndeath;i++) {
       for(j=nlstate+1;j<=nlstate+ndeath;j++){            ppij=0.;ppi=0.;
         for(i=1,gpp[j]=0.; i<= nlstate; i++)            for(j=1; j<=nlstate;j++) {
           gpp[j] += prlim[i][i]*p3mat[i][j][1];              /* if (mobilav==1) */
       }                  ppij=ppij+p3mat[i][j][h]*prevacurrent[(int)agec][j][k];
       /* end probability of death */              ppi=ppi+prevacurrent[(int)agec][j][k];
               /* ppij=ppij+p3mat[i][j][h]*mobaverage[(int)agec][j][k]; */
               /* ppi=ppi+mobaverage[(int)agec][j][k]; */
                 /* else { */
                 /*        ppij=ppij+p3mat[i][j][h]*probs[(int)(agec)][i][k]; */
                 /* } */
               fprintf(ficresfb," %.3f", p3mat[i][j][h]);
             } /* end j */
             if(ppi <0.99){
               printf("Error in prevbackforecast, prevalence doesn't sum to 1 for state %d: %3f\n",i, ppi);
               fprintf(ficlog,"Error in prevbackforecast, prevalence doesn't sum to 1 for state %d: %3f\n",i, ppi);
             }
             fprintf(ficresfb," %.3f", ppij);
           }/* end j */
           free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
         } /* end agec */
       } /* end yearp */
     } /* end k */
     
     /* if (mobilav!=0) free_ma3x(mobaverage,1, AGESUP,1,NCOVMAX, 1,NCOVMAX); */
     
     fclose(ficresfb);
     printf("End of Computing Back forecasting \n");
     fprintf(ficlog,"End of Computing Back forecasting\n");
           
   }
   
       for(i=1; i<=npar; i++) /* Computes gradient x - delta */  /* Variance of prevalence limit: varprlim */
         xp[i] = x[i] - (i==theta ?delti[theta]:0);   void varprlim(char fileresu[], int nresult, double ***prevacurrent, int mobilavproj, double bage, double fage, double **prlim, int *ncvyearp, double ftolpl, double p[], double **matcov, double *delti, int stepm, int cptcoveff){
       hpxij(p3mat,nhstepm,age,hstepm,xp,nlstate,stepm,oldm,savm, ij);        /*------- Variance of forward period (stable) prevalence------*/   
       prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ij);  
     
       if (popbased==1) {     char fileresvpl[FILENAMELENGTH];  
         if(mobilav ==0){     FILE *ficresvpl;
           for(i=1; i<=nlstate;i++)     double **oldm, **savm;
             prlim[i][i]=probs[(int)age][i][ij];     double **varpl; /* Variances of prevalence limits by age */   
         }else{ /* mobilav */      int i1, k, nres, j ;
           for(i=1; i<=nlstate;i++)     
             prlim[i][i]=mobaverage[(int)age][i][ij];      strcpy(fileresvpl,"VPL_");
         }      strcat(fileresvpl,fileresu);
       }      if((ficresvpl=fopen(fileresvpl,"w"))==NULL) {
         printf("Problem with variance of forward period (stable) prevalence  resultfile: %s\n", fileresvpl);
       for(j=1; j<= nlstate; j++){        exit(0);
         for(h=0; h<=nhstepm; h++){      }
           for(i=1, gm[h][j]=0.;i<=nlstate;i++)      printf("Computing Variance-covariance of forward period (stable) prevalence: file '%s' ...", fileresvpl);fflush(stdout);
             gm[h][j] += prlim[i][i]*p3mat[i][j][h];      fprintf(ficlog, "Computing Variance-covariance of forward period (stable) prevalence: file '%s' ...", fileresvpl);fflush(ficlog);
         }      
       }      /*for(cptcov=1,k=0;cptcov<=i1;cptcov++){
       /* This for computing probability of death (h=1 means        for(cptcod=1;cptcod<=ncodemax[cptcov];cptcod++){*/
          computed over hstepm matrices product = hstepm*stepm months)       
          as a weighted average of prlim.      i1=pow(2,cptcoveff);
       */      if (cptcovn < 1){i1=1;}
       for(j=nlstate+1;j<=nlstate+ndeath;j++){  
         for(i=1,gmp[j]=0.; i<= nlstate; i++)  
          gmp[j] += prlim[i][i]*p3mat[i][j][1];  
       }      
       /* end probability of death */  
   
       for(j=1; j<= nlstate; j++) /* vareij */  
         for(h=0; h<=nhstepm; h++){  
           gradg[h][theta][j]= (gp[h][j]-gm[h][j])/2./delti[theta];  
         }  
   
       for(j=nlstate+1; j<= nlstate+ndeath; j++){ /* var mu */      for(nres=1; nres <= nresult; nres++) /* For each resultline */
         gradgp[theta][j]= (gpp[j]-gmp[j])/2./delti[theta];        for(k=1; k<=i1;k++){ /* We find the combination equivalent to result line values of dummies */
       }        if(i1 != 1 && TKresult[nres]!= k)
           continue;
         fprintf(ficresvpl,"\n#****** ");
         printf("\n#****** ");
         fprintf(ficlog,"\n#****** ");
         for(j=1;j<=cptcoveff;j++) {
           fprintf(ficresvpl,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
           fprintf(ficlog,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
           printf("V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
         }
         for (j=1; j<= nsq; j++){ /* For each selected (single) quantitative value */
           printf(" V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
           fprintf(ficresvpl," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
           fprintf(ficlog," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
         } 
         fprintf(ficresvpl,"******\n");
         printf("******\n");
         fprintf(ficlog,"******\n");
         
         varpl=matrix(1,nlstate,(int) bage, (int) fage);
         oldm=oldms;savm=savms;
         varprevlim(fileresvpl, ficresvpl, varpl, matcov, p, delti, nlstate, stepm, (int) bage, (int) fage, oldm, savm, prlim, ftolpl, ncvyearp, k, strstart, nres);
         free_matrix(varpl,1,nlstate,(int) bage, (int)fage);
         /*}*/
       }
       
       fclose(ficresvpl);
       printf("done variance-covariance of forward period prevalence\n");fflush(stdout);
       fprintf(ficlog,"done variance-covariance of forward period prevalence\n");fflush(ficlog);
   
     } /* End theta */   }
   /* Variance of back prevalence: varbprlim */
    void varbprlim(char fileresu[], int nresult, double ***prevacurrent, int mobilavproj, double bage, double fage, double **bprlim, int *ncvyearp, double ftolpl, double p[], double **matcov, double *delti, int stepm, int cptcoveff){
         /*------- Variance of back (stable) prevalence------*/
   
      char fileresvbl[FILENAMELENGTH];  
      FILE  *ficresvbl;
   
      double **oldm, **savm;
      double **varbpl; /* Variances of back prevalence limits by age */   
      int i1, k, nres, j ;
   
      strcpy(fileresvbl,"VBL_");
      strcat(fileresvbl,fileresu);
      if((ficresvbl=fopen(fileresvbl,"w"))==NULL) {
        printf("Problem with variance of back (stable) prevalence  resultfile: %s\n", fileresvbl);
        exit(0);
      }
      printf("Computing Variance-covariance of back (stable) prevalence: file '%s' ...", fileresvbl);fflush(stdout);
      fprintf(ficlog, "Computing Variance-covariance of back (stable) prevalence: file '%s' ...", fileresvbl);fflush(ficlog);
      
      
      i1=pow(2,cptcoveff);
      if (cptcovn < 1){i1=1;}
      
      for(nres=1; nres <= nresult; nres++) /* For each resultline */
        for(k=1; k<=i1;k++){
          if(i1 != 1 && TKresult[nres]!= k)
            continue;
          fprintf(ficresvbl,"\n#****** ");
          printf("\n#****** ");
          fprintf(ficlog,"\n#****** ");
          for(j=1;j<=cptcoveff;j++) {
            fprintf(ficresvbl,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
            fprintf(ficlog,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
            printf("V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
          }
          for (j=1; j<= nsq; j++){ /* For each selected (single) quantitative value */
            printf(" V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
            fprintf(ficresvbl," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
            fprintf(ficlog," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
          }
          fprintf(ficresvbl,"******\n");
          printf("******\n");
          fprintf(ficlog,"******\n");
          
          varbpl=matrix(1,nlstate,(int) bage, (int) fage);
          oldm=oldms;savm=savms;
          
          varbrevlim(fileresvbl, ficresvbl, varbpl, matcov, p, delti, nlstate, stepm, (int) bage, (int) fage, oldm, savm, bprlim, ftolpl, mobilavproj, ncvyearp, k, strstart, nres);
          free_matrix(varbpl,1,nlstate,(int) bage, (int)fage);
          /*}*/
        }
      
      fclose(ficresvbl);
      printf("done variance-covariance of back prevalence\n");fflush(stdout);
      fprintf(ficlog,"done variance-covariance of back prevalence\n");fflush(ficlog);
   
     trgradg =ma3x(0,nhstepm,1,nlstate,1,npar); /* veij */   } /* End of varbprlim */
   
     for(h=0; h<=nhstepm; h++) /* veij */  /************** Forecasting *****not tested NB*************/
       for(j=1; j<=nlstate;j++)  /* void populforecast(char fileres[], double anpyram,double mpyram,double jpyram,double ageminpar, double agemax,double dateprev1, double dateprev2s, int mobilav, double agedeb, double fage, int popforecast, char popfile[], double anpyram1,double p[], int i2){ */
         for(theta=1; theta <=npar; theta++)    
           trgradg[h][j][theta]=gradg[h][theta][j];  /*   int cpt, stepsize, hstepm, nhstepm, j,k,c, cptcod, i,h; */
   /*   int *popage; */
   /*   double calagedatem, agelim, kk1, kk2; */
   /*   double *popeffectif,*popcount; */
   /*   double ***p3mat,***tabpop,***tabpopprev; */
   /*   /\* double ***mobaverage; *\/ */
   /*   char filerespop[FILENAMELENGTH]; */
   
   /*   tabpop= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX); */
   /*   tabpopprev= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX); */
   /*   agelim=AGESUP; */
   /*   calagedatem=(anpyram+mpyram/12.+jpyram/365.-dateintmean)*YEARM; */
     
   /*   prevalence(probs, ageminpar, agemax, s, agev, nlstate, imx, Tvar, nbcode, ncodemax, mint, anint, dateprev1, dateprev2, firstpass, lastpass); */
     
     
   /*   strcpy(filerespop,"POP_");  */
   /*   strcat(filerespop,fileresu); */
   /*   if((ficrespop=fopen(filerespop,"w"))==NULL) { */
   /*     printf("Problem with forecast resultfile: %s\n", filerespop); */
   /*     fprintf(ficlog,"Problem with forecast resultfile: %s\n", filerespop); */
   /*   } */
   /*   printf("Computing forecasting: result on file '%s' \n", filerespop); */
   /*   fprintf(ficlog,"Computing forecasting: result on file '%s' \n", filerespop); */
   
   /*   if (cptcoveff==0) ncodemax[cptcoveff]=1; */
   
   /*   /\* if (mobilav!=0) { *\/ */
   /*   /\*   mobaverage= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX); *\/ */
   /*   /\*   if (movingaverage(probs, ageminpar, fage, mobaverage,mobilav)!=0){ *\/ */
   /*   /\*     fprintf(ficlog," Error in movingaverage mobilav=%d\n",mobilav); *\/ */
   /*   /\*     printf(" Error in movingaverage mobilav=%d\n",mobilav); *\/ */
   /*   /\*   } *\/ */
   /*   /\* } *\/ */
   
     for(j=nlstate+1; j<=nlstate+ndeath;j++) /* mu */  /*   stepsize=(int) (stepm+YEARM-1)/YEARM; */
       for(theta=1; theta <=npar; theta++)  /*   if (stepm<=12) stepsize=1; */
         trgradgp[j][theta]=gradgp[theta][j];    
   /*   agelim=AGESUP; */
       
   /*   hstepm=1; */
   /*   hstepm=hstepm/stepm;  */
           
   /*   if (popforecast==1) { */
   /*     if((ficpop=fopen(popfile,"r"))==NULL) { */
   /*       printf("Problem with population file : %s\n",popfile);exit(0); */
   /*       fprintf(ficlog,"Problem with population file : %s\n",popfile);exit(0); */
   /*     }  */
   /*     popage=ivector(0,AGESUP); */
   /*     popeffectif=vector(0,AGESUP); */
   /*     popcount=vector(0,AGESUP); */
       
   /*     i=1;    */
   /*     while ((c=fscanf(ficpop,"%d %lf\n",&popage[i],&popcount[i])) != EOF) i=i+1; */
       
   /*     imx=i; */
   /*     for (i=1; i<imx;i++) popeffectif[popage[i]]=popcount[i]; */
   /*   } */
     
   /*   for(cptcov=1,k=0;cptcov<=i2;cptcov++){ */
   /*     for(cptcod=1;cptcod<=ncodemax[cptcoveff];cptcod++){ */
   /*       k=k+1; */
   /*       fprintf(ficrespop,"\n#******"); */
   /*       for(j=1;j<=cptcoveff;j++) { */
   /*      fprintf(ficrespop," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,j)]); */
   /*       } */
   /*       fprintf(ficrespop,"******\n"); */
   /*       fprintf(ficrespop,"# Age"); */
   /*       for(j=1; j<=nlstate+ndeath;j++) fprintf(ficrespop," P.%d",j); */
   /*       if (popforecast==1)  fprintf(ficrespop," [Population]"); */
         
   /*       for (cpt=0; cpt<=0;cpt++) {  */
   /*      fprintf(ficrespop,"\n\n# Forecasting at date %.lf/%.lf/%.lf ",jpyram,mpyram,anpyram+cpt);    */
           
   /*      for (agedeb=(fage-((int)calagedatem %12/12.)); agedeb>=(ageminpar-((int)calagedatem %12)/12.); agedeb--){  */
   /*        nhstepm=(int) rint((agelim-agedeb)*YEARM/stepm);  */
   /*        nhstepm = nhstepm/hstepm;  */
             
   /*        p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm); */
   /*        oldm=oldms;savm=savms; */
   /*        hpxij(p3mat,nhstepm,agedeb,hstepm,p,nlstate,stepm,oldm,savm, k);   */
             
   /*        for (h=0; h<=nhstepm; h++){ */
   /*          if (h==(int) (calagedatem+YEARM*cpt)) { */
   /*            fprintf(ficrespop,"\n %3.f ",agedeb+h*hstepm/YEARM*stepm); */
   /*          }  */
   /*          for(j=1; j<=nlstate+ndeath;j++) { */
   /*            kk1=0.;kk2=0; */
   /*            for(i=1; i<=nlstate;i++) {               */
   /*              if (mobilav==1)  */
   /*                kk1=kk1+p3mat[i][j][h]*mobaverage[(int)agedeb+1][i][cptcod]; */
   /*              else { */
   /*                kk1=kk1+p3mat[i][j][h]*probs[(int)(agedeb+1)][i][cptcod]; */
   /*              } */
   /*            } */
   /*            if (h==(int)(calagedatem+12*cpt)){ */
   /*              tabpop[(int)(agedeb)][j][cptcod]=kk1; */
   /*              /\*fprintf(ficrespop," %.3f", kk1); */
   /*                if (popforecast==1) fprintf(ficrespop," [%.f]", kk1*popeffectif[(int)agedeb+1]);*\/ */
   /*            } */
   /*          } */
   /*          for(i=1; i<=nlstate;i++){ */
   /*            kk1=0.; */
   /*            for(j=1; j<=nlstate;j++){ */
   /*              kk1= kk1+tabpop[(int)(agedeb)][j][cptcod];  */
   /*            } */
   /*            tabpopprev[(int)(agedeb)][i][cptcod]=tabpop[(int)(agedeb)][i][cptcod]/kk1*popeffectif[(int)(agedeb+(calagedatem+12*cpt)*hstepm/YEARM*stepm-1)]; */
   /*          } */
               
   /*          if (h==(int)(calagedatem+12*cpt)) */
   /*            for(j=1; j<=nlstate;j++)  */
   /*              fprintf(ficrespop," %15.2f",tabpopprev[(int)(agedeb+1)][j][cptcod]); */
   /*        } */
   /*        free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm); */
   /*      } */
   /*       } */
         
   /*       /\******\/ */
         
   /*       for (cpt=1; cpt<=(anpyram1-anpyram);cpt++) {  */
   /*      fprintf(ficrespop,"\n\n# Forecasting at date %.lf/%.lf/%.lf ",jpyram,mpyram,anpyram+cpt);    */
   /*      for (agedeb=(fage-((int)calagedatem %12/12.)); agedeb>=(ageminpar-((int)calagedatem %12)/12.); agedeb--){  */
   /*        nhstepm=(int) rint((agelim-agedeb)*YEARM/stepm);  */
   /*        nhstepm = nhstepm/hstepm;  */
             
   /*        p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm); */
   /*        oldm=oldms;savm=savms; */
   /*        hpxij(p3mat,nhstepm,agedeb,hstepm,p,nlstate,stepm,oldm,savm, k);   */
   /*        for (h=0; h<=nhstepm; h++){ */
   /*          if (h==(int) (calagedatem+YEARM*cpt)) { */
   /*            fprintf(ficresf,"\n %3.f ",agedeb+h*hstepm/YEARM*stepm); */
   /*          }  */
   /*          for(j=1; j<=nlstate+ndeath;j++) { */
   /*            kk1=0.;kk2=0; */
   /*            for(i=1; i<=nlstate;i++) {               */
   /*              kk1=kk1+p3mat[i][j][h]*tabpopprev[(int)agedeb+1][i][cptcod];     */
   /*            } */
   /*            if (h==(int)(calagedatem+12*cpt)) fprintf(ficresf," %15.2f", kk1);         */
   /*          } */
   /*        } */
   /*        free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm); */
   /*      } */
   /*       } */
   /*     }  */
   /*   } */
     
   /*   /\* if (mobilav!=0) free_ma3x(mobaverage,1, AGESUP,1,NCOVMAX, 1,NCOVMAX); *\/ */
     
   /*   if (popforecast==1) { */
   /*     free_ivector(popage,0,AGESUP); */
   /*     free_vector(popeffectif,0,AGESUP); */
   /*     free_vector(popcount,0,AGESUP); */
   /*   } */
   /*   free_ma3x(tabpop,1, AGESUP,1,NCOVMAX, 1,NCOVMAX); */
   /*   free_ma3x(tabpopprev,1, AGESUP,1,NCOVMAX, 1,NCOVMAX); */
   /*   fclose(ficrespop); */
   /* } /\* End of popforecast *\/ */
    
   int fileappend(FILE *fichier, char *optionfich)
   {
     if((fichier=fopen(optionfich,"a"))==NULL) {
       printf("Problem with file: %s\n", optionfich);
       fprintf(ficlog,"Problem with file: %s\n", optionfich);
       return (0);
     }
     fflush(fichier);
     return (1);
   }
   
     hf=hstepm*stepm/YEARM;  /* Duration of hstepm expressed in year unit. */  
     for(i=1;i<=nlstate;i++)  
       for(j=1;j<=nlstate;j++)  
         vareij[i][j][(int)age] =0.;  
   
     for(h=0;h<=nhstepm;h++){  /**************** function prwizard **********************/
       for(k=0;k<=nhstepm;k++){  void prwizard(int ncovmodel, int nlstate, int ndeath,  char model[], FILE *ficparo)
         matprod2(dnewm,trgradg[h],1,nlstate,1,npar,1,npar,matcov);  {
         matprod2(doldm,dnewm,1,nlstate,1,npar,1,nlstate,gradg[k]);  
         for(i=1;i<=nlstate;i++)    /* Wizard to print covariance matrix template */
           for(j=1;j<=nlstate;j++)  
             vareij[i][j][(int)age] += doldm[i][j]*hf*hf;    char ca[32], cb[32];
     int i,j, k, li, lj, lk, ll, jj, npar, itimes;
     int numlinepar;
   
     printf("# Parameters nlstate*nlstate*ncov a12*1 + b12 * age + ...\n");
     fprintf(ficparo,"# Parameters nlstate*nlstate*ncov a12*1 + b12 * age + ...\n");
     for(i=1; i <=nlstate; i++){
       jj=0;
       for(j=1; j <=nlstate+ndeath; j++){
         if(j==i) continue;
         jj++;
         /*ca[0]= k+'a'-1;ca[1]='\0';*/
         printf("%1d%1d",i,j);
         fprintf(ficparo,"%1d%1d",i,j);
         for(k=1; k<=ncovmodel;k++){
           /*        printf(" %lf",param[i][j][k]); */
           /*        fprintf(ficparo," %lf",param[i][j][k]); */
           printf(" 0.");
           fprintf(ficparo," 0.");
       }        }
         printf("\n");
         fprintf(ficparo,"\n");
     }      }
       }
     /* pptj */    printf("# Scales (for hessian or gradient estimation)\n");
     matprod2(dnewmp,trgradgp,nlstate+1,nlstate+ndeath,1,npar,1,npar,matcov);    fprintf(ficparo,"# Scales (for hessian or gradient estimation)\n");
     matprod2(doldmp,dnewmp,nlstate+1,nlstate+ndeath,1,npar,nlstate+1,nlstate+ndeath,gradgp);    npar= (nlstate+ndeath-1)*nlstate*ncovmodel; /* Number of parameters*/ 
     for(j=nlstate+1;j<=nlstate+ndeath;j++)    for(i=1; i <=nlstate; i++){
       for(i=nlstate+1;i<=nlstate+ndeath;i++)      jj=0;
         varppt[j][i]=doldmp[j][i];      for(j=1; j <=nlstate+ndeath; j++){
     /* end ppptj */        if(j==i) continue;
     /*  x centered again */        jj++;
     hpxij(p3mat,nhstepm,age,hstepm,x,nlstate,stepm,oldm,savm, ij);          fprintf(ficparo,"%1d%1d",i,j);
     prevalim(prlim,nlstate,x,age,oldm,savm,ftolpl,ij);        printf("%1d%1d",i,j);
         fflush(stdout);
         for(k=1; k<=ncovmodel;k++){
           /*      printf(" %le",delti3[i][j][k]); */
           /*      fprintf(ficparo," %le",delti3[i][j][k]); */
           printf(" 0.");
           fprintf(ficparo," 0.");
         }
         numlinepar++;
         printf("\n");
         fprintf(ficparo,"\n");
       }
     }
     printf("# Covariance matrix\n");
   /* # 121 Var(a12)\n\ */
   /* # 122 Cov(b12,a12) Var(b12)\n\ */
   /* # 131 Cov(a13,a12) Cov(a13,b12, Var(a13)\n\ */
   /* # 132 Cov(b13,a12) Cov(b13,b12, Cov(b13,a13) Var(b13)\n\ */
   /* # 212 Cov(a21,a12) Cov(a21,b12, Cov(a21,a13) Cov(a21,b13) Var(a21)\n\ */
   /* # 212 Cov(b21,a12) Cov(b21,b12, Cov(b21,a13) Cov(b21,b13) Cov(b21,a21) Var(b21)\n\ */
   /* # 232 Cov(a23,a12) Cov(a23,b12, Cov(a23,a13) Cov(a23,b13) Cov(a23,a21) Cov(a23,b21) Var(a23)\n\ */
   /* # 232 Cov(b23,a12) Cov(b23,b12) ... Var (b23)\n" */
     fflush(stdout);
     fprintf(ficparo,"# Covariance matrix\n");
     /* # 121 Var(a12)\n\ */
     /* # 122 Cov(b12,a12) Var(b12)\n\ */
     /* #   ...\n\ */
     /* # 232 Cov(b23,a12)  Cov(b23,b12) ... Var (b23)\n" */
     
     for(itimes=1;itimes<=2;itimes++){
       jj=0;
       for(i=1; i <=nlstate; i++){
         for(j=1; j <=nlstate+ndeath; j++){
           if(j==i) continue;
           for(k=1; k<=ncovmodel;k++){
             jj++;
             ca[0]= k+'a'-1;ca[1]='\0';
             if(itimes==1){
               printf("#%1d%1d%d",i,j,k);
               fprintf(ficparo,"#%1d%1d%d",i,j,k);
             }else{
               printf("%1d%1d%d",i,j,k);
               fprintf(ficparo,"%1d%1d%d",i,j,k);
               /*  printf(" %.5le",matcov[i][j]); */
             }
             ll=0;
             for(li=1;li <=nlstate; li++){
               for(lj=1;lj <=nlstate+ndeath; lj++){
                 if(lj==li) continue;
                 for(lk=1;lk<=ncovmodel;lk++){
                   ll++;
                   if(ll<=jj){
                     cb[0]= lk +'a'-1;cb[1]='\0';
                     if(ll<jj){
                       if(itimes==1){
                         printf(" Cov(%s%1d%1d,%s%1d%1d)",ca,i,j,cb, li,lj);
                         fprintf(ficparo," Cov(%s%1d%1d,%s%1d%1d)",ca,i,j,cb, li,lj);
                       }else{
                         printf(" 0.");
                         fprintf(ficparo," 0.");
                       }
                     }else{
                       if(itimes==1){
                         printf(" Var(%s%1d%1d)",ca,i,j);
                         fprintf(ficparo," Var(%s%1d%1d)",ca,i,j);
                       }else{
                         printf(" 0.");
                         fprintf(ficparo," 0.");
                       }
                     }
                   }
                 } /* end lk */
               } /* end lj */
             } /* end li */
             printf("\n");
             fprintf(ficparo,"\n");
             numlinepar++;
           } /* end k*/
         } /*end j */
       } /* end i */
     } /* end itimes */
   
   } /* end of prwizard */
   /******************* Gompertz Likelihood ******************************/
   double gompertz(double x[])
   { 
     double A=0.0,B=0.,L=0.0,sump=0.,num=0.;
     int i,n=0; /* n is the size of the sample */
   
     for (i=1;i<=imx ; i++) {
       sump=sump+weight[i];
       /*    sump=sump+1;*/
       num=num+1;
     }
     L=0.0;
     /* agegomp=AGEGOMP; */
     /* for (i=0; i<=imx; i++) 
        if (wav[i]>0) printf("i=%d ageex=%lf agecens=%lf agedc=%lf cens=%d %d\n" ,i,ageexmed[i],agecens[i],agedc[i],cens[i],wav[i]);*/
   
     for (i=1;i<=imx ; i++) {
       /* mu(a)=mu(agecomp)*exp(teta*(age-agegomp))
          mu(a)=x[1]*exp(x[2]*(age-agegomp)); x[1] and x[2] are per year.
        * L= Product mu(agedeces)exp(-\int_ageexam^agedc mu(u) du ) for a death between agedc (in month) 
        *   and agedc +1 month, cens[i]=0: log(x[1]/YEARM)
        * +
        * exp(-\int_ageexam^agecens mu(u) du ) when censored, cens[i]=1
        */
        if (wav[i] > 1 || agedc[i] < AGESUP) {
          if (cens[i] == 1){
            A=-x[1]/(x[2])*(exp(x[2]*(agecens[i]-agegomp))-exp(x[2]*(ageexmed[i]-agegomp)));
          } else if (cens[i] == 0){
           A=-x[1]/(x[2])*(exp(x[2]*(agedc[i]-agegomp))-exp(x[2]*(ageexmed[i]-agegomp)))
             +log(x[1]/YEARM) +x[2]*(agedc[i]-agegomp)+log(YEARM);
         } else
            printf("Gompertz cens[%d] neither 1 nor 0\n",i);
         /*if (wav[i] > 1 && agecens[i] > 15) {*/ /* ??? */
          L=L+A*weight[i];
           /*      printf("\ni=%d A=%f L=%lf x[1]=%lf x[2]=%lf ageex=%lf agecens=%lf cens=%d agedc=%lf weight=%lf\n",i,A,L,x[1],x[2],ageexmed[i]*12,agecens[i]*12,cens[i],agedc[i]*12,weight[i]);*/
        }
     }
   
     /*printf("x1=%2.9f x2=%2.9f x3=%2.9f L=%f\n",x[1],x[2],x[3],L);*/
    
     return -2*L*num/sump;
   }
   
   #ifdef GSL
   /******************* Gompertz_f Likelihood ******************************/
   double gompertz_f(const gsl_vector *v, void *params)
   { 
     double A=0.,B=0.,LL=0.0,sump=0.,num=0.;
     double *x= (double *) v->data;
     int i,n=0; /* n is the size of the sample */
   
     for (i=0;i<=imx-1 ; i++) {
       sump=sump+weight[i];
       /*    sump=sump+1;*/
       num=num+1;
     }
     
     if (popbased==1) {   
       if(mobilav ==0){    /* for (i=0; i<=imx; i++) 
         for(i=1; i<=nlstate;i++)       if (wav[i]>0) printf("i=%d ageex=%lf agecens=%lf agedc=%lf cens=%d %d\n" ,i,ageexmed[i],agecens[i],agedc[i],cens[i],wav[i]);*/
           prlim[i][i]=probs[(int)age][i][ij];    printf("x[0]=%lf x[1]=%lf\n",x[0],x[1]);
       }else{ /* mobilav */     for (i=1;i<=imx ; i++)
         for(i=1; i<=nlstate;i++)      {
           prlim[i][i]=mobaverage[(int)age][i][ij];        if (cens[i] == 1 && wav[i]>1)
           A=-x[0]/(x[1])*(exp(x[1]*(agecens[i]-agegomp))-exp(x[1]*(ageexmed[i]-agegomp)));
         
         if (cens[i] == 0 && wav[i]>1)
           A=-x[0]/(x[1])*(exp(x[1]*(agedc[i]-agegomp))-exp(x[1]*(ageexmed[i]-agegomp)))
                +log(x[0]/YEARM)+x[1]*(agedc[i]-agegomp)+log(YEARM);  
         
         /*if (wav[i] > 1 && agecens[i] > 15) {*/ /* ??? */
         if (wav[i] > 1 ) { /* ??? */
           LL=LL+A*weight[i];
           /*      printf("\ni=%d A=%f L=%lf x[1]=%lf x[2]=%lf ageex=%lf agecens=%lf cens=%d agedc=%lf weight=%lf\n",i,A,L,x[1],x[2],ageexmed[i]*12,agecens[i]*12,cens[i],agedc[i]*12,weight[i]);*/
       }        }
     }      }
                
     /* This for computing probability of death (h=1 means  
        computed over hstepm (estepm) matrices product = hstepm*stepm months)   
        as a weighted average of prlim.  
     */  
     for(j=nlstate+1;j<=nlstate+ndeath;j++){  
       for(i=1,gmp[j]=0.;i<= nlstate; i++)   
         gmp[j] += prlim[i][i]*p3mat[i][j][1];   
     }      
     /* end probability of death */  
   
     fprintf(ficresprobmorprev,"%3d %d ",(int) age, ij);  
     for(j=nlstate+1; j<=(nlstate+ndeath);j++){  
       fprintf(ficresprobmorprev," %11.3e %11.3e",gmp[j], sqrt(varppt[j][j]));  
       for(i=1; i<=nlstate;i++){  
         fprintf(ficresprobmorprev," %11.3e %11.3e ",prlim[i][i],p3mat[i][j][1]);  
       }  
     }   
     fprintf(ficresprobmorprev,"\n");  
   
     fprintf(ficresvij,"%.0f ",age );   /*printf("x1=%2.9f x2=%2.9f x3=%2.9f L=%f\n",x[1],x[2],x[3],L);*/
     for(i=1; i<=nlstate;i++)    printf("x[0]=%lf x[1]=%lf -2*LL*num/sump=%lf\n",x[0],x[1],-2*LL*num/sump);
       for(j=1; j<=nlstate;j++){   
         fprintf(ficresvij," %.4f", vareij[i][j][(int)age]);    return -2*LL*num/sump;
       }  }
     fprintf(ficresvij,"\n");  #endif
     free_matrix(gp,0,nhstepm,1,nlstate);  
     free_matrix(gm,0,nhstepm,1,nlstate);  
     free_ma3x(gradg,0,nhstepm,1,npar,1,nlstate);  
     free_ma3x(trgradg,0,nhstepm,1,nlstate,1,npar);  
     free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);  
   } /* End age */  
   free_vector(gpp,nlstate+1,nlstate+ndeath);  
   free_vector(gmp,nlstate+1,nlstate+ndeath);  
   free_matrix(gradgp,1,npar,nlstate+1,nlstate+ndeath);  
   free_matrix(trgradgp,nlstate+1,nlstate+ndeath,1,npar); /* mu or p point j*/  
   fprintf(ficgp,"\nset noparametric;set nolabel; set ter png small;set size 0.65, 0.65");  
   /* for(j=nlstate+1; j<= nlstate+ndeath; j++){ *//* Only the first actually */  
   fprintf(ficgp,"\n set log y; set nolog x;set xlabel \"Age\"; set ylabel \"Force of mortality (year-1)\";");  
 /*   fprintf(ficgp,"\n plot \"%s\"  u 1:($3*%6.3f) not w l 1 ",fileresprobmorprev,YEARM/estepm); */  
 /*   fprintf(ficgp,"\n replot \"%s\"  u 1:(($3+1.96*$4)*%6.3f) t \"95\%% interval\" w l 2 ",fileresprobmorprev,YEARM/estepm); */  
 /*   fprintf(ficgp,"\n replot \"%s\"  u 1:(($3-1.96*$4)*%6.3f) not w l 2 ",fileresprobmorprev,YEARM/estepm); */  
   fprintf(ficgp,"\n plot \"%s\"  u 1:($3) not w l 1 ",fileresprobmorprev);  
   fprintf(ficgp,"\n replot \"%s\"  u 1:(($3+1.96*$4)) t \"95\%% interval\" w l 2 ",fileresprobmorprev);  
   fprintf(ficgp,"\n replot \"%s\"  u 1:(($3-1.96*$4)) not w l 2 ",fileresprobmorprev);  
   fprintf(fichtm,"\n<br> File (multiple files are possible if covariates are present): <A href=\"%s\">%s</a>\n",fileresprobmorprev,fileresprobmorprev);  
   fprintf(fichtm,"\n<br> Probability is computed over estepm=%d months. <br> <img src=\"varmuptjgr%s%s%s.png\"> <br>\n", estepm,digitp,optionfilefiname,digit);  
   /*  fprintf(fichtm,"\n<br> Probability is computed over estepm=%d months and then divided by estepm and multiplied by %.0f in order to have the probability to die over a year <br> <img src=\"varmuptjgr%s%s.png\"> <br>\n", stepm,YEARM,digitp,digit);  
 */  
   fprintf(ficgp,"\nset out \"varmuptjgr%s%s%s.png\";replot;",digitp,optionfilefiname,digit);  
   
   free_vector(xp,1,npar);  /******************* Printing html file ***********/
   free_matrix(doldm,1,nlstate,1,nlstate);  void printinghtmlmort(char fileresu[], char title[], char datafile[], int firstpass, \
   free_matrix(dnewm,1,nlstate,1,npar);                    int lastpass, int stepm, int weightopt, char model[],\
   free_matrix(doldmp,nlstate+1,nlstate+ndeath,nlstate+1,nlstate+ndeath);                    int imx,  double p[],double **matcov,double agemortsup){
   free_matrix(dnewmp,nlstate+1,nlstate+ndeath,1,npar);    int i,k;
   free_matrix(varppt,nlstate+1,nlstate+ndeath,nlstate+1,nlstate+ndeath);  
   if (mobilav!=0) free_ma3x(mobaverage,1, AGESUP,1,NCOVMAX, 1,NCOVMAX);  
   fclose(ficresprobmorprev);  
   fclose(ficgp);  
   fclose(fichtm);  
 }  /* end varevsij */  
   
 /************ Variance of prevlim ******************/    fprintf(fichtm,"<ul><li><h4>Result files </h4>\n Force of mortality. Parameters of the Gompertz fit (with confidence interval in brackets):<br>");
 void varprevlim(char fileres[], double **varpl, double **matcov, double x[], double delti[], int nlstate, int stepm, double bage, double fage, double **oldm, double **savm, double **prlim, double ftolpl, int ij)    fprintf(fichtm,"  mu(age) =%lf*exp(%lf*(age-%d)) per year<br><br>",p[1],p[2],agegomp);
 {    for (i=1;i<=2;i++) 
   /* Variance of prevalence limit */      fprintf(fichtm," p[%d] = %lf [%f ; %f]<br>\n",i,p[i],p[i]-2*sqrt(matcov[i][i]),p[i]+2*sqrt(matcov[i][i]));
   /*  double **prevalim(double **prlim, int nlstate, double *xp, double age, double **oldm, double **savm,double ftolpl);*/    fprintf(fichtm,"<br><br><img src=\"graphmort.svg\">");
   double **newm;    fprintf(fichtm,"</ul>");
   double **dnewm,**doldm;  
   int i, j, nhstepm, hstepm;  
   int k, cptcode;  
   double *xp;  
   double *gp, *gm;  
   double **gradg, **trgradg;  
   double age,agelim;  
   int theta;  
      
   fprintf(ficresvpl,"# Standard deviation of stable prevalences \n");  
   fprintf(ficresvpl,"# Age");  
   for(i=1; i<=nlstate;i++)  
       fprintf(ficresvpl," %1d-%1d",i,i);  
   fprintf(ficresvpl,"\n");  
   
   xp=vector(1,npar);  fprintf(fichtm,"<ul><li><h4>Life table</h4>\n <br>");
   dnewm=matrix(1,nlstate,1,npar);  
   doldm=matrix(1,nlstate,1,nlstate);  
     
   hstepm=1*YEARM; /* Every year of age */  
   hstepm=hstepm/stepm; /* Typically in stepm units, if j= 2 years, = 2/6 months = 4 */   
   agelim = AGESUP;  
   for (age=bage; age<=fage; age ++){ /* If stepm=6 months */  
     nhstepm=(int) rint((agelim-age)*YEARM/stepm); /* Typically 20 years = 20*12/6=40 */   
     if (stepm >= YEARM) hstepm=1;  
     nhstepm = nhstepm/hstepm; /* Typically 40/4=10 */  
     gradg=matrix(1,npar,1,nlstate);  
     gp=vector(1,nlstate);  
     gm=vector(1,nlstate);  
   
     for(theta=1; theta <=npar; theta++){   fprintf(fichtm,"\nAge   l<inf>x</inf>     q<inf>x</inf> d(x,x+1)    L<inf>x</inf>     T<inf>x</inf>     e<infx</inf><br>");
       for(i=1; i<=npar; i++){ /* Computes gradient */  
         xp[i] = x[i] + (i==theta ?delti[theta]:0);  
       }  
       prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ij);  
       for(i=1;i<=nlstate;i++)  
         gp[i] = prlim[i][i];  
       
       for(i=1; i<=npar; i++) /* Computes gradient */  
         xp[i] = x[i] - (i==theta ?delti[theta]:0);  
       prevalim(prlim,nlstate,xp,age,oldm,savm,ftolpl,ij);  
       for(i=1;i<=nlstate;i++)  
         gm[i] = prlim[i][i];  
   
       for(i=1;i<=nlstate;i++)   for (k=agegomp;k<(agemortsup-2);k++) 
         gradg[theta][i]= (gp[i]-gm[i])/2./delti[theta];     fprintf(fichtm,"%d %.0lf %lf %.0lf %.0lf %.0lf %lf<br>\n",k,lsurv[k],p[1]*exp(p[2]*(k-agegomp)),(p[1]*exp(p[2]*(k-agegomp)))*lsurv[k],lpop[k],tpop[k],tpop[k]/lsurv[k]);
     } /* End theta */  
   
     trgradg =matrix(1,nlstate,1,npar);   
     fflush(fichtm);
   }
   
     for(j=1; j<=nlstate;j++)  /******************* Gnuplot file **************/
       for(theta=1; theta <=npar; theta++)  void printinggnuplotmort(char fileresu[], char optionfilefiname[], double ageminpar, double agemaxpar, double fage , char pathc[], double p[]){
         trgradg[j][theta]=gradg[theta][j];  
   
     for(i=1;i<=nlstate;i++)    char dirfileres[132],optfileres[132];
       varpl[i][(int)age] =0.;  
     matprod2(dnewm,trgradg,1,nlstate,1,npar,1,npar,matcov);  
     matprod2(doldm,dnewm,1,nlstate,1,npar,1,nlstate,gradg);  
     for(i=1;i<=nlstate;i++)  
       varpl[i][(int)age] = doldm[i][i]; /* Covariances are useless */  
   
     fprintf(ficresvpl,"%.0f ",age );    int ng;
     for(i=1; i<=nlstate;i++)  
       fprintf(ficresvpl," %.5f (%.5f)",prlim[i][i],sqrt(varpl[i][(int)age]));  
     fprintf(ficresvpl,"\n");  
     free_vector(gp,1,nlstate);  
     free_vector(gm,1,nlstate);  
     free_matrix(gradg,1,npar,1,nlstate);  
     free_matrix(trgradg,1,nlstate,1,npar);  
   } /* End age */  
   
   free_vector(xp,1,npar);  
   free_matrix(doldm,1,nlstate,1,npar);  
   free_matrix(dnewm,1,nlstate,1,nlstate);  
   
 }    /*#ifdef windows */
     fprintf(ficgp,"cd \"%s\" \n",pathc);
       /*#endif */
   
 /************ Variance of one-step probabilities  ******************/  
 void varprob(char optionfilefiname[], double **matcov, double x[], double delti[], int nlstate, double bage, double fage, int ij, int *Tvar, int **nbcode, int *ncodemax)    strcpy(dirfileres,optionfilefiname);
     strcpy(optfileres,"vpl");
     fprintf(ficgp,"set out \"graphmort.svg\"\n "); 
     fprintf(ficgp,"set xlabel \"Age\"\n set ylabel \"Force of mortality (per year)\" \n "); 
     fprintf(ficgp, "set ter svg size 640, 480\n set log y\n"); 
     /* fprintf(ficgp, "set size 0.65,0.65\n"); */
     fprintf(ficgp,"plot [%d:100] %lf*exp(%lf*(x-%d))",agegomp,p[1],p[2],agegomp);
   
   } 
   
   int readdata(char datafile[], int firstobs, int lastobs, int *imax)
 {  {
   int i, j=0,  i1, k1, l1, t, tj;  
   int k2, l2, j1,  z1;  
   int k=0,l, cptcode;  
   int first=1, first1;  
   double cv12, mu1, mu2, lc1, lc2, v12, v21, v11, v22,v1,v2, c12, tnalp;  
   double **dnewm,**doldm;  
   double *xp;  
   double *gp, *gm;  
   double **gradg, **trgradg;  
   double **mu;  
   double age,agelim, cov[NCOVMAX];  
   double std=2.0; /* Number of standard deviation wide of confidence ellipsoids */  
   int theta;  
   char fileresprob[FILENAMELENGTH];  
   char fileresprobcov[FILENAMELENGTH];  
   char fileresprobcor[FILENAMELENGTH];  
   
   double ***varpij;  
   
   strcpy(fileresprob,"prob");   
   strcat(fileresprob,fileres);  
   if((ficresprob=fopen(fileresprob,"w"))==NULL) {  
     printf("Problem with resultfile: %s\n", fileresprob);  
     fprintf(ficlog,"Problem with resultfile: %s\n", fileresprob);  
   }  
   strcpy(fileresprobcov,"probcov");   
   strcat(fileresprobcov,fileres);  
   if((ficresprobcov=fopen(fileresprobcov,"w"))==NULL) {  
     printf("Problem with resultfile: %s\n", fileresprobcov);  
     fprintf(ficlog,"Problem with resultfile: %s\n", fileresprobcov);  
   }  
   strcpy(fileresprobcor,"probcor");   
   strcat(fileresprobcor,fileres);  
   if((ficresprobcor=fopen(fileresprobcor,"w"))==NULL) {  
     printf("Problem with resultfile: %s\n", fileresprobcor);  
     fprintf(ficlog,"Problem with resultfile: %s\n", fileresprobcor);  
   }  
   printf("Computing standard deviation of one-step probabilities: result on file '%s' \n",fileresprob);  
   fprintf(ficlog,"Computing standard deviation of one-step probabilities: result on file '%s' \n",fileresprob);  
   printf("Computing matrix of variance covariance of one-step probabilities: result on file '%s' \n",fileresprobcov);  
   fprintf(ficlog,"Computing matrix of variance covariance of one-step probabilities: result on file '%s' \n",fileresprobcov);  
   printf("and correlation matrix of one-step probabilities: result on file '%s' \n",fileresprobcor);  
   fprintf(ficlog,"and correlation matrix of one-step probabilities: result on file '%s' \n",fileresprobcor);  
     
   fprintf(ficresprob,"#One-step probabilities and stand. devi in ()\n");  
   fprintf(ficresprob,"# Age");  
   fprintf(ficresprobcov,"#One-step probabilities and covariance matrix\n");  
   fprintf(ficresprobcov,"# Age");  
   fprintf(ficresprobcor,"#One-step probabilities and correlation matrix\n");  
   fprintf(ficresprobcov,"# Age");  
   
     /*-------- data file ----------*/
     FILE *fic;
     char dummy[]="                         ";
     int i=0, j=0, n=0, iv=0, v;
     int lstra;
     int linei, month, year,iout;
     int noffset=0; /* This is the offset if BOM data file */
     char line[MAXLINE], linetmp[MAXLINE];
     char stra[MAXLINE], strb[MAXLINE];
     char *stratrunc;
   
     DummyV=ivector(1,NCOVMAX); /* 1 to 3 */
     FixedV=ivector(1,NCOVMAX); /* 1 to 3 */
     for(v=1;v<NCOVMAX;v++){
       DummyV[v]=0;
       FixedV[v]=0;
     }
   
     for(v=1; v <=ncovcol;v++){
       DummyV[v]=0;
       FixedV[v]=0;
     }
     for(v=ncovcol+1; v <=ncovcol+nqv;v++){
       DummyV[v]=1;
       FixedV[v]=0;
     }
     for(v=ncovcol+nqv+1; v <=ncovcol+nqv+ntv;v++){
       DummyV[v]=0;
       FixedV[v]=1;
     }
     for(v=ncovcol+nqv+ntv+1; v <=ncovcol+nqv+ntv+nqtv;v++){
       DummyV[v]=1;
       FixedV[v]=1;
     }
     for(v=1; v <=ncovcol+nqv+ntv+nqtv;v++){
       printf("Covariate type in the data: V%d, DummyV(V%d)=%d, FixedV(V%d)=%d\n",v,v,DummyV[v],v,FixedV[v]);
       fprintf(ficlog,"Covariate type in the data: V%d, DummyV(V%d)=%d, FixedV(V%d)=%d\n",v,v,DummyV[v],v,FixedV[v]);
     }
   
   for(i=1; i<=nlstate;i++)    if((fic=fopen(datafile,"r"))==NULL)    {
     for(j=1; j<=(nlstate+ndeath);j++){      printf("Problem while opening datafile: %s with errno='%s'\n", datafile,strerror(errno));fflush(stdout);
       fprintf(ficresprob," p%1d-%1d (SE)",i,j);      fprintf(ficlog,"Problem while opening datafile: %s with errno='%s'\n", datafile,strerror(errno));fflush(ficlog);return 1;
       fprintf(ficresprobcov," p%1d-%1d ",i,j);  
       fprintf(ficresprobcor," p%1d-%1d ",i,j);  
     }    
  /* fprintf(ficresprob,"\n");  
   fprintf(ficresprobcov,"\n");  
   fprintf(ficresprobcor,"\n");  
  */  
  xp=vector(1,npar);  
   dnewm=matrix(1,(nlstate)*(nlstate+ndeath),1,npar);  
   doldm=matrix(1,(nlstate)*(nlstate+ndeath),1,(nlstate)*(nlstate+ndeath));  
   mu=matrix(1,(nlstate)*(nlstate+ndeath), (int) bage, (int)fage);  
   varpij=ma3x(1,nlstate*(nlstate+ndeath),1,nlstate*(nlstate+ndeath),(int) bage, (int) fage);  
   first=1;  
   if((ficgp=fopen(optionfilegnuplot,"a"))==NULL) {  
     printf("Problem with gnuplot file: %s\n", optionfilegnuplot);  
     fprintf(ficlog,"Problem with gnuplot file: %s\n", optionfilegnuplot);  
     exit(0);  
   }    }
   else{  
     fprintf(ficgp,"\n# Routine varprob");      /* Is it a BOM UTF-8 Windows file? */
     /* First data line */
     linei=0;
     while(fgets(line, MAXLINE, fic)) {
       noffset=0;
       if( line[0] == (char)0xEF && line[1] == (char)0xBB) /* EF BB BF */
       {
         noffset=noffset+3;
         printf("# Data file '%s'  is an UTF8 BOM file, please convert to UTF8 or ascii file and rerun.\n",datafile);fflush(stdout);
         fprintf(ficlog,"# Data file '%s'  is an UTF8 BOM file, please convert to UTF8 or ascii file and rerun.\n",datafile);
         fflush(ficlog); return 1;
       }
       /*    else if( line[0] == (char)0xFE && line[1] == (char)0xFF)*/
       else if( line[0] == (char)0xFF && line[1] == (char)0xFE)
       {
         noffset=noffset+2;
         printf("# Error Data file '%s'  is a huge UTF16BE BOM file, please convert to UTF8 or ascii file (for example with dos2unix) and rerun.\n",datafile);fflush(stdout);
         fprintf(ficlog,"# Error Data file '%s'  is a huge UTF16BE BOM file, please convert to UTF8 or ascii file (for example with dos2unix) and rerun.\n",datafile);
         fflush(ficlog); return 1;
       }
       else if( line[0] == 0 && line[1] == 0)
       {
         if( line[2] == (char)0xFE && line[3] == (char)0xFF){
           noffset=noffset+4;
           printf("# Error Data file '%s'  is a huge UTF16BE BOM file, please convert to UTF8 or ascii file (for example with dos2unix) and rerun.\n",datafile);fflush(stdout);
           fprintf(ficlog,"# Error Data file '%s'  is a huge UTF16BE BOM file, please convert to UTF8 or ascii file (for example with dos2unix) and rerun.\n",datafile);
           fflush(ficlog); return 1;
         }
       } else{
         ;/*printf(" Not a BOM file\n");*/
       }
           /* If line starts with a # it is a comment */
       if (line[noffset] == '#') {
         linei=linei+1;
         break;
       }else{
         break;
       }
   }    }
   if((fichtm=fopen(optionfilehtm,"a"))==NULL) {    fclose(fic);
     printf("Problem with html file: %s\n", optionfilehtm);    if((fic=fopen(datafile,"r"))==NULL)    {
     fprintf(ficlog,"Problem with html file: %s\n", optionfilehtm);      printf("Problem while opening datafile: %s with errno='%s'\n", datafile,strerror(errno));fflush(stdout);
     exit(0);      fprintf(ficlog,"Problem while opening datafile: %s with errno='%s'\n", datafile,strerror(errno));fflush(ficlog);return 1;
   }    }
   else{    /* Not a Bom file */
     fprintf(fichtm,"\n<li><h4> Computing and drawing one step probabilities with their confidence intervals</h4></li>\n");    
     fprintf(fichtm,"\n");    i=1;
     while ((fgets(line, MAXLINE, fic) != NULL) &&((i >= firstobs) && (i <=lastobs))) {
       linei=linei+1;
       for(j=strlen(line); j>=0;j--){  /* Untabifies line */
         if(line[j] == '\t')
           line[j] = ' ';
       }
       for(j=strlen(line)-1; (line[j]==' ')||(line[j]==10)||(line[j]==13);j--){
         ;
       };
       line[j+1]=0;  /* Trims blanks at end of line */
       if(line[0]=='#'){
         fprintf(ficlog,"Comment line\n%s\n",line);
         printf("Comment line\n%s\n",line);
         continue;
       }
       trimbb(linetmp,line); /* Trims multiple blanks in line */
       strcpy(line, linetmp);
       
       /* Loops on waves */
       for (j=maxwav;j>=1;j--){
         for (iv=nqtv;iv>=1;iv--){  /* Loop  on time varying quantitative variables */
           cutv(stra, strb, line, ' '); 
           if(strb[0]=='.') { /* Missing value */
             lval=-1;
             cotqvar[j][iv][i]=-1; /* 0.0/0.0 */
             cotvar[j][ntv+iv][i]=-1; /* For performance reasons */
             if(isalpha(strb[1])) { /* .m or .d Really Missing value */
               printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be the %d th quantitative value out of %d measured at wave %d. If missing, you should remove this individual or impute a value.  Exiting.\n", strb, linei,i,line,iv, nqtv, j);
               fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be the %d th quantitative value out of %d measured at wave %d. If missing, you should remove this individual or impute a value.  Exiting.\n", strb, linei,i,line,iv, nqtv, j);fflush(ficlog);
               return 1;
             }
           }else{
             errno=0;
             /* what_kind_of_number(strb); */
             dval=strtod(strb,&endptr); 
             /* if( strb[0]=='\0' || (*endptr != '\0')){ */
             /* if(strb != endptr && *endptr == '\0') */
             /*    dval=dlval; */
             /* if (errno == ERANGE && (lval == LONG_MAX || lval == LONG_MIN)) */
             if( strb[0]=='\0' || (*endptr != '\0')){
               printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be the %d th quantitative value out of %d measured at wave %d. Setting maxwav=%d might be wrong.  Exiting.\n", strb, linei,i,line,iv, nqtv, j,maxwav);
               fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be the %d th quantitative value out of %d measured at wave %d. Setting maxwav=%d might be wrong.  Exiting.\n", strb, linei,i,line, iv, nqtv, j,maxwav);fflush(ficlog);
               return 1;
             }
             cotqvar[j][iv][i]=dval; 
             cotvar[j][ntv+iv][i]=dval; 
           }
           strcpy(line,stra);
         }/* end loop ntqv */
         
         for (iv=ntv;iv>=1;iv--){  /* Loop  on time varying dummies */
           cutv(stra, strb, line, ' '); 
           if(strb[0]=='.') { /* Missing value */
             lval=-1;
           }else{
             errno=0;
             lval=strtol(strb,&endptr,10); 
             /*    if (errno == ERANGE && (lval == LONG_MAX || lval == LONG_MIN))*/
             if( strb[0]=='\0' || (*endptr != '\0')){
               printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be the %d th dummy covariate out of %d measured at wave %d. Setting maxwav=%d might be wrong.  Exiting.\n", strb, linei,i,line,iv, ntv, j,maxwav);
               fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be the %d dummy covariate out of %d measured wave %d. Setting maxwav=%d might be wrong.  Exiting.\n", strb, linei,i,line,iv, ntv,j,maxwav);fflush(ficlog);
               return 1;
             }
           }
           if(lval <-1 || lval >1){
             printf("Error reading data around '%ld' at line number %d for individual %d, '%s'\n \
    Should be a value of %d(nth) covariate of wave %d (0 should be the value for the reference and 1\n \
    for the alternative. IMaCh does not build design variables automatically, do it yourself.\n \
    For example, for multinomial values like 1, 2 and 3,\n                 \
    build V1=0 V2=0 for the reference value (1),\n                         \
           V1=1 V2=0 for (2) \n                                            \
    and V1=0 V2=1 for (3). V1=1 V2=1 should not exist and the corresponding\n \
    output of IMaCh is often meaningless.\n                                \
    Exiting.\n",lval,linei, i,line,iv,j);
             fprintf(ficlog,"Error reading data around '%ld' at line number %d for individual %d, '%s'\n \
    Should be a value of %d(nth) covariate of wave %d (0 should be the value for the reference and 1\n \
    for the alternative. IMaCh does not build design variables automatically, do it yourself.\n \
    For example, for multinomial values like 1, 2 and 3,\n                 \
    build V1=0 V2=0 for the reference value (1),\n                         \
           V1=1 V2=0 for (2) \n                                            \
    and V1=0 V2=1 for (3). V1=1 V2=1 should not exist and the corresponding\n \
    output of IMaCh is often meaningless.\n                                \
    Exiting.\n",lval,linei, i,line,iv,j);fflush(ficlog);
             return 1;
           }
           cotvar[j][iv][i]=(double)(lval);
           strcpy(line,stra);
         }/* end loop ntv */
         
         /* Statuses  at wave */
         cutv(stra, strb, line, ' '); 
         if(strb[0]=='.') { /* Missing value */
           lval=-1;
         }else{
           errno=0;
           lval=strtol(strb,&endptr,10); 
           /*      if (errno == ERANGE && (lval == LONG_MAX || lval == LONG_MIN))*/
           if( strb[0]=='\0' || (*endptr != '\0')){
             printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a status of wave %d. Setting maxwav=%d might be wrong.  Exiting.\n", strb, linei,i,line,j,maxwav);
             fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a status of wave %d. Setting maxwav=%d might be wrong.  Exiting.\n", strb, linei,i,line,j,maxwav);fflush(ficlog);
             return 1;
           }
         }
         
         s[j][i]=lval;
         
         /* Date of Interview */
         strcpy(line,stra);
         cutv(stra, strb,line,' ');
         if( (iout=sscanf(strb,"%d/%d",&month, &year)) != 0){
         }
         else  if( (iout=sscanf(strb,"%s.",dummy)) != 0){
           month=99;
           year=9999;
         }else{
           printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a date of interview (mm/yyyy or .) at wave %d.  Exiting.\n",strb, linei,i, line,j);
           fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a date of interview (mm/yyyy or .) at wave %d.  Exiting.\n",strb, linei,i, line,j);fflush(ficlog);
           return 1;
         }
         anint[j][i]= (double) year; 
         mint[j][i]= (double)month;
         /* if( (int)anint[j][i]+ (int)(mint[j][i])/12. < (int) (moisnais[i]/12.+annais[i])){ */
         /*        printf("Warning reading data around '%s' at line number %d for individual %d, '%s'\nThe date of interview (%2d/%4d) at wave %d occurred before the date of birth (%2d/%4d).\n",strb, linei,i, line, mint[j][i],anint[j][i], moisnais[i],annais[i]); */
         /*        fprintf(ficlog,"Warning reading data around '%s' at line number %d for individual %d, '%s'\nThe date of interview (%2d/%4d) at wave %d occurred before the date of birth (%2d/%4d).\n",strb, linei,i, line, mint[j][i],anint[j][i], moisnais[i],annais[i]); */
         /* } */
         strcpy(line,stra);
       } /* End loop on waves */
       
       /* Date of death */
       cutv(stra, strb,line,' '); 
       if( (iout=sscanf(strb,"%d/%d",&month, &year)) != 0){
       }
       else  if( (iout=sscanf(strb,"%s.",dummy)) != 0){
         month=99;
         year=9999;
       }else{
         printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a date of death (mm/yyyy or .).  Exiting.\n",strb, linei,i,line);
         fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a date of death (mm/yyyy or .).  Exiting.\n",strb, linei,i,line);fflush(ficlog);
         return 1;
       }
       andc[i]=(double) year; 
       moisdc[i]=(double) month; 
       strcpy(line,stra);
       
       /* Date of birth */
       cutv(stra, strb,line,' '); 
       if( (iout=sscanf(strb,"%d/%d",&month, &year)) != 0){
       }
       else  if( (iout=sscanf(strb,"%s.", dummy)) != 0){
         month=99;
         year=9999;
       }else{
         printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a date of birth (mm/yyyy or .).  Exiting.\n",strb, linei,i,line);
         fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a date of birth (mm/yyyy or .).  Exiting.\n",strb, linei,i,line);fflush(ficlog);
         return 1;
       }
       if (year==9999) {
         printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a date of birth (mm/yyyy) but at least the year of birth should be given.  Exiting.\n",strb, linei,i,line);
         fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be a date of birth (mm/yyyy) but at least the year of birth should be given. Exiting.\n",strb, linei,i,line);fflush(ficlog);
         return 1;
         
       }
       annais[i]=(double)(year);
       moisnais[i]=(double)(month);
       for (j=1;j<=maxwav;j++){
         if( (int)anint[j][i]+ (int)(mint[j][i])/12. < (int) (moisnais[i]/12.+annais[i])){
           printf("Warning reading data around '%s' at line number %d for individual %d, '%s'\nThe date of interview (%2d/%4d) at wave %d occurred before the date of birth (%2d/%4d).\n",strb, linei,i, line, (int)mint[j][i],(int)anint[j][i], j,(int)moisnais[i],(int)annais[i]);
           fprintf(ficlog,"Warning reading data around '%s' at line number %d for individual %d, '%s'\nThe date of interview (%2d/%4d) at wave %d occurred before the date of birth (%2d/%4d).\n",strb, linei,i, line, (int)mint[j][i],(int)anint[j][i], j, (int)moisnais[i],(int)annais[i]);
         }
       }
   
       strcpy(line,stra);
       
       /* Sample weight */
       cutv(stra, strb,line,' '); 
       errno=0;
       dval=strtod(strb,&endptr); 
       if( strb[0]=='\0' || (*endptr != '\0')){
         printf("Error reading data around '%f' at line number %d, \"%s\" for individual %d\nShould be a weight.  Exiting.\n",dval, i,line,linei);
         fprintf(ficlog,"Error reading data around '%f' at line number %d, \"%s\" for individual %d\nShould be a weight.  Exiting.\n",dval, i,line,linei);
         fflush(ficlog);
         return 1;
       }
       weight[i]=dval; 
       strcpy(line,stra);
       
       for (iv=nqv;iv>=1;iv--){  /* Loop  on fixed quantitative variables */
         cutv(stra, strb, line, ' '); 
         if(strb[0]=='.') { /* Missing value */
           lval=-1;
           coqvar[iv][i]=NAN; 
           covar[ncovcol+iv][i]=NAN; /* including qvar in standard covar for performance reasons */ 
         }else{
           errno=0;
           /* what_kind_of_number(strb); */
           dval=strtod(strb,&endptr);
           /* if(strb != endptr && *endptr == '\0') */
           /*   dval=dlval; */
           /* if (errno == ERANGE && (lval == LONG_MAX || lval == LONG_MIN)) */
           if( strb[0]=='\0' || (*endptr != '\0')){
             printf("Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be the %d th quantitative value (out of %d) constant for all waves. Setting maxwav=%d might be wrong.  Exiting.\n", strb, linei,i,line, iv, nqv, maxwav);
             fprintf(ficlog,"Error reading data around '%s' at line number %d for individual %d, '%s'\nShould be the %d th quantitative value (out of %d) constant for all waves. Setting maxwav=%d might be wrong.  Exiting.\n", strb, linei,i,line, iv, nqv, maxwav);fflush(ficlog);
             return 1;
           }
           coqvar[iv][i]=dval; 
           covar[ncovcol+iv][i]=dval; /* including qvar in standard covar for performance reasons */ 
         }
         strcpy(line,stra);
       }/* end loop nqv */
       
       /* Covariate values */
       for (j=ncovcol;j>=1;j--){
         cutv(stra, strb,line,' '); 
         if(strb[0]=='.') { /* Missing covariate value */
           lval=-1;
         }else{
           errno=0;
           lval=strtol(strb,&endptr,10); 
           if( strb[0]=='\0' || (*endptr != '\0')){
             printf("Error reading data around '%ld' at line number %d for individual %d, '%s'\nShould be a covariate value (=0 for the reference or 1 for alternative).  Exiting.\n",lval, linei,i, line);
             fprintf(ficlog,"Error reading data around '%ld' at line number %d for individual %d, '%s'\nShould be a covariate value (=0 for the reference or 1 for alternative).  Exiting.\n",lval, linei,i, line);fflush(ficlog);
             return 1;
           }
         }
         if(lval <-1 || lval >1){
           printf("Error reading data around '%ld' at line number %d for individual %d, '%s'\n \
    Should be a value of %d(nth) covariate (0 should be the value for the reference and 1\n \
    for the alternative. IMaCh does not build design variables automatically, do it yourself.\n \
    For example, for multinomial values like 1, 2 and 3,\n                 \
    build V1=0 V2=0 for the reference value (1),\n                         \
           V1=1 V2=0 for (2) \n                                            \
    and V1=0 V2=1 for (3). V1=1 V2=1 should not exist and the corresponding\n \
    output of IMaCh is often meaningless.\n                                \
    Exiting.\n",lval,linei, i,line,j);
           fprintf(ficlog,"Error reading data around '%ld' at line number %d for individual %d, '%s'\n \
    Should be a value of %d(nth) covariate (0 should be the value for the reference and 1\n \
    for the alternative. IMaCh does not build design variables automatically, do it yourself.\n \
    For example, for multinomial values like 1, 2 and 3,\n                 \
    build V1=0 V2=0 for the reference value (1),\n                         \
           V1=1 V2=0 for (2) \n                                            \
    and V1=0 V2=1 for (3). V1=1 V2=1 should not exist and the corresponding\n \
    output of IMaCh is often meaningless.\n                                \
    Exiting.\n",lval,linei, i,line,j);fflush(ficlog);
           return 1;
         }
         covar[j][i]=(double)(lval);
         strcpy(line,stra);
       }  
       lstra=strlen(stra);
       
       if(lstra > 9){ /* More than 2**32 or max of what printf can write with %ld */
         stratrunc = &(stra[lstra-9]);
         num[i]=atol(stratrunc);
       }
       else
         num[i]=atol(stra);
       /*if((s[2][i]==2) && (s[3][i]==-1)&&(s[4][i]==9)){
         printf("%ld %.lf %.lf %.lf %.lf/%.lf %.lf/%.lf %.lf/%.lf %d %.lf/%.lf %d %.lf/%.lf %d %.lf/%.lf %d\n",num[i],(covar[1][i]), (covar[2][i]),weight[i], (moisnais[i]), (annais[i]), (moisdc[i]), (andc[i]), (mint[1][i]), (anint[1][i]), (s[1][i]),  (mint[2][i]), (anint[2][i]), (s[2][i]),  (mint[3][i]), (anint[3][i]), (s[3][i]),  (mint[4][i]), (anint[4][i]), (s[4][i])); ij=ij+1;}*/
       
       i=i+1;
     } /* End loop reading  data */
     
     *imax=i-1; /* Number of individuals */
     fclose(fic);
     
     return (0);
     /* endread: */
     printf("Exiting readdata: ");
     fclose(fic);
     return (1);
   }
   
     fprintf(fichtm,"\n<li><h4> Computing matrix of variance-covariance of step probabilities</h4></li>\n");  void removefirstspace(char **stri){/*, char stro[]) {*/
     fprintf(fichtm,"\nWe have drawn ellipsoids of confidence around the p<inf>ij</inf>, p<inf>kl</inf> to understand the covariance between two incidences. They are expressed in year<sup>-1</sup> in order to be less dependent of stepm.<br>\n");    char *p1 = *stri, *p2 = *stri;
     fprintf(fichtm,"\n<br> We have drawn x'cov<sup>-1</sup>x = 4 where x is the column vector (pij,pkl). It means that if pij and pkl where uncorrelated the (2X2) matrix would have been (1/(var pij), 0 , 0, 1/(var pkl)), and the confidence interval would be 2 standard deviations wide on each axis. <br> When both incidences are correlated we diagonalised the inverse of the covariance matrix and made the appropriate rotation.<br> \n");    while (*p2 == ' ')
       p2++; 
     /* while ((*p1++ = *p2++) !=0) */
     /*   ; */
     /* do */
     /*   while (*p2 == ' ') */
     /*     p2++; */
     /* while (*p1++ == *p2++); */
     *stri=p2; 
   }
   
   int decoderesult( char resultline[], int nres)
   /**< This routine decode one result line and returns the combination # of dummy covariates only **/
   {
     int j=0, k=0, k1=0, k2=0, k3=0, k4=0, match=0, k2q=0, k3q=0, k4q=0;
     char resultsav[MAXLINE];
     /* int resultmodel[MAXLINE]; */
     /* int modelresult[MAXLINE]; */
     char stra[80], strb[80], strc[80], strd[80],stre[80];
   
     removefirstspace(&resultline);
     printf("decoderesult:%s\n",resultline);
   
     strcpy(resultsav,resultline);
     printf("Decoderesult resultsav=\"%s\" resultline=\"%s\"\n", resultsav, resultline);
     if (strlen(resultsav) >1){
       j=nbocc(resultsav,'='); /**< j=Number of covariate values'=' in this resultline */
     }
     if(j == 0){ /* Resultline but no = */
       TKresult[nres]=0; /* Combination for the nresult and the model */
       return (0);
     }
     if( j != cptcovs ){ /* Be careful if a variable is in a product but not single */
       printf("ERROR: the number of variables in the resultline which is %d, differs from the number %d of single variables used in the model line, %s.\n",j, cptcovs, model);
       fprintf(ficlog,"ERROR: the number of variables in the resultline which is %d, differs from the number %d of single variables used in the model line, %s.\n",j, cptcovs, model);
       /* return 1;*/
     }
     for(k=1; k<=j;k++){ /* Loop on any covariate of the RESULT LINE */
       if(nbocc(resultsav,'=') >1){
         cutl(stra,strb,resultsav,' '); /* keeps in strb after the first ' ' (stra is the rest of the resultline to be analyzed in the next loop *//*     resultsav= "V4=1 V5=25.1 V3=0" stra= "V5=25.1 V3=0" strb= "V4=1" */
         /* If resultsav= "V4= 1 V5=25.1 V3=0" with a blank then strb="V4=" and stra="1 V5=25.1 V3=0" */
         cutl(strc,strd,strb,'=');  /* strb:"V4=1" strc="1" strd="V4" */
         /* If a blank, then strc="V4=" and strd='\0' */
         if(strc[0]=='\0'){
         printf("Error in resultline, probably a blank after the \"%s\", \"result:%s\", stra=\"%s\" resultsav=\"%s\"\n",strb,resultline, stra, resultsav);
           fprintf(ficlog,"Error in resultline, probably a blank after the \"V%s=\", resultline=%s\n",strb,resultline);
           return 1;
         }
       }else
         cutl(strc,strd,resultsav,'=');
       Tvalsel[k]=atof(strc); /* 1 */ /* Tvalsel of k is the float value of the kth covariate appearing in this result line */
       
       cutl(strc,stre,strd,'V'); /* strd='V4' strc=4 stre='V' */;
       Tvarsel[k]=atoi(strc);  /* 4 */ /* Tvarsel is the id of the kth covariate in the result line Tvarsel[1] in "V4=1.." is 4.*/
       /* Typevarsel[k]=1;  /\* 1 for age product *\/ */
       /* cptcovsel++;     */
       if (nbocc(stra,'=') >0)
         strcpy(resultsav,stra); /* and analyzes it */
     }
     /* Checking for missing or useless values in comparison of current model needs */
     /* Feeds resultmodel[nres][k1]=k2 for k1th product covariate with age in the model equation fed by the index k2 of the resutline*/
     for(k1=1; k1<= cptcovt ;k1++){ /* Loop on MODEL LINE V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
       if(Typevar[k1]==0){ /* Single covariate in model */
         /* 0 for simple covariate (dummy, quantitative, fixed or varying), 1 for age product, 2 for  product */
         match=0;
         for(k2=1; k2 <=j;k2++){/* Loop on resultline. In result line V4=1 V5=24.1 V3=1  V2=8 V1=0 */
           if(Tvar[k1]==Tvarsel[k2]) {/* Tvar is coming from the model, Tvarsel from the result. Tvar[1]=5 == Tvarsel[2]=5   */
             modelresult[nres][k2]=k1;/* modelresult[2]=1 modelresult[1]=2  modelresult[3]=3  modelresult[6]=4 modelresult[9]=5 */
             match=1; /* modelresult of k2 variable of resultline is identical to k1 variable of the model good */
             break;
           }
         }
         if(match == 0){
           printf("Error in result line (Dummy single): V%d is missing in result: %s according to model=%s. Tvar[k1=%d]=%d is different from Tvarsel[k2=%d]=%d.\n",Tvar[k1], resultline, model,k1, Tvar[k1], k2, Tvarsel[k2]);
           fprintf(ficlog,"Error in result line (Dummy single): V%d is missing in result: %s according to model=%s\n",Tvar[k1], resultline, model);
           return 1;
         }
       }else if(Typevar[k1]==1){ /* Product with age We want to get the position k2 in the resultline of the product k1 in the model line*/
         /* We feed resultmodel[k1]=k2; */
         match=0;
         for(k2=1; k2 <=j;k2++){/* Loop on resultline.  jth occurence of = signs in the result line. In result line V4=1 V5=24.1 V3=1  V2=8 V1=0 */
           if(Tvar[k1]==Tvarsel[k2]) {/* Tvar is coming from the model, Tvarsel from the result. Tvar[1]=5 == Tvarsel[2]=5   */
             modelresult[nres][k2]=k1;/* we found a Vn=1 corrresponding to Vn*age in the model modelresult[2]=1 modelresult[1]=2  modelresult[3]=3  modelresult[6]=4 modelresult[9]=5 */
             resultmodel[nres][k1]=k2; /* Added here */
             printf("Decoderesult first modelresult[k2=%d]=%d (k1) V%d*AGE\n",k2,k1,Tvar[k1]);
             match=1; /* modelresult of k2 variable of resultline is identical to k1 variable of the model good */
             break;
           }
         }
         if(match == 0){
           printf("Error in result line (Product with age): V%d is missing in result: %s according to model=%s (Tvarsel[k2=%d]=%d)\n",Tvar[k1], resultline, model, k2, Tvarsel[k2]);
           fprintf(ficlog,"Error in result line (Product with age): V%d is missing in result: %s according to model=%s (Tvarsel[k2=%d]=%d)\n",Tvar[k1], resultline, model, k2, Tvarsel[k2]);
         return 1;
         }
       }else if(Typevar[k1]==2){ /* Product No age We want to get the position in the resultline of the product in the model line*/
         /* resultmodel[nres][of such a Vn * Vm product k1] is not unique, so can't exist, we feed Tvard[k1][1] and [2] */ 
         match=0;
         printf("Decoderesult very first Product Tvardk[k1=%d][1]=%d Tvardk[k1=%d][2]=%d V%d * V%d\n",k1,Tvardk[k1][1],k1,Tvardk[k1][2],Tvardk[k1][1],Tvardk[k1][2]);
         for(k2=1; k2 <=j;k2++){/* Loop on resultline. In result line V4=1 V5=24.1 V3=1  V2=8 V1=0 */
           if(Tvardk[k1][1]==Tvarsel[k2]) {/* Tvardk is coming from the model, Tvarsel from the result. Tvar[1]=5 == Tvarsel[2]=5   */
             /* modelresult[k2]=k1; */
             printf("Decoderesult first Product modelresult[k2=%d]=%d (k1) V%d * \n",k2,k1,Tvarsel[k2]);
             match=1; /* modelresult of k2 variable of resultline is identical to k1 variable of the model good */
           }
         }
         if(match == 0){
           printf("Error in result line (Product without age first variable): V%d is missing in result: %s according to model=%s\n",Tvardk[k1][1], resultline, model);
           fprintf(ficlog,"Error in result line (Product without age first variable): V%d is missing in result: %s according to model=%s\n",Tvardk[k1][1], resultline, model);
           return 1;
         }
         match=0;
         for(k2=1; k2 <=j;k2++){/* Loop on resultline. In result line V4=1 V5=24.1 V3=1  V2=8 V1=0 */
           if(Tvardk[k1][2]==Tvarsel[k2]) {/* Tvardk is coming from the model, Tvarsel from the result. Tvar[1]=5 == Tvarsel[2]=5   */
             /* modelresult[k2]=k1;*/
             printf("Decoderesult second Product modelresult[k2=%d]=%d (k1) * V%d \n ",k2,k1,Tvarsel[k2]);
             match=1; /* modelresult of k2 variable of resultline is identical to k1 variable of the model good */
             break;
           }
         }
         if(match == 0){
           printf("Error in result line (Product without age second variable): V%d is missing in result: %s according to model=%s\n",Tvardk[k1][2], resultline, model);
           fprintf(ficlog,"Error in result line (Product without age second variable): V%d is missing in result : %s according to model=%s\n",Tvardk[k1][2], resultline, model);
           return 1;
         }
       }/* End of testing */
     }/* End loop cptcovt */
     /* Checking for missing or useless values in comparison of current model needs */
     /* Feeds resultmodel[nres][k1]=k2 for single covariate (k1) in the model equation */
     for(k2=1; k2 <=j;k2++){ /* j or cptcovs is the number of single covariates used either in the model line as well as in the result line (dummy or quantitative)
                              * Loop on resultline variables: result line V4=1 V5=24.1 V3=1  V2=8 V1=0 */
       match=0;
       for(k1=1; k1<= cptcovt ;k1++){ /* loop on model: model line V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
         if(Typevar[k1]==0){ /* Single only */
           if(Tvar[k1]==Tvarsel[k2]) { /* Tvar[2]=4 == Tvarsel[1]=4   */
             resultmodel[nres][k1]=k2;  /* k1th position in the model equation corresponds to k2th position in the result line. resultmodel[2]=1 resultmodel[1]=2  resultmodel[3]=3  resultmodel[6]=4 resultmodel[9]=5 */
             modelresult[nres][k2]=k1; /* k1th position in the model equation corresponds to k2th position in the result line. modelresult[1]=2 modelresult[2]=1  modelresult[3]=3  remodelresult[4]=6 modelresult[5]=9 */
             ++match;
           }
         }
       }
       if(match == 0){
         printf("Error in result line: variable V%d is missing in model; result: %s, model=%s\n",Tvarsel[k2], resultline, model);
         fprintf(ficlog,"Error in result line: variable V%d is missing in model; result: %s, model=%s\n",Tvarsel[k2], resultline, model);
         return 1;
       }else if(match > 1){
         printf("Error in result line: %d doubled; result: %s, model=%s\n",k2, resultline, model);
         fprintf(ficlog,"Error in result line: %d doubled; result: %s, model=%s\n",k2, resultline, model);
         return 1;
       }
     }
     /* cptcovres=j /\* Number of variables in the resultline is equal to cptcovs and thus useless *\/     */
     /* We need to deduce which combination number is chosen and save quantitative values */
     /* model line V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
     /* nres=1st result line: V4=1 V5=25.1 V3=0  V2=8 V1=1 */
     /* should correspond to the combination 6 of dummy: V4=1, V3=0, V1=1 => V4*2**(0) + V3*2**(1) + V1*2**(2) = 1*1 + 0*2 + 1*4 = 5 + (1offset) = 6*/
     /* nres=2nd result line: V4=1 V5=24.1 V3=1  V2=8 V1=0 */
     /* should give a combination of dummy V4=1, V3=1, V1=0 => V4*2**(0) + V3*2**(1) + V1*2**(2) = 3 + (1offset) = 4*/
     /*    1 0 0 0 */
     /*    2 1 0 0 */
     /*    3 0 1 0 */ 
     /*    4 1 1 0 */ /* V4=1, V3=1, V1=0 (nres=2)*/
     /*    5 0 0 1 */
     /*    6 1 0 1 */ /* V4=1, V3=0, V1=1 (nres=1)*/
     /*    7 0 1 1 */
     /*    8 1 1 1 */
     /* V(Tvresult)=Tresult V4=1 V3=0 V1=1 Tresult[nres=1][2]=0 */
     /* V(Tvqresult)=Tqresult V5=25.1 V2=8 Tqresult[nres=1][1]=25.1 */
     /* V5*age V5 known which value for nres?  */
     /* Tqinvresult[2]=8 Tqinvresult[1]=25.1  */
     for(k1=1, k=0, k4=0, k4q=0; k1 <=cptcovt;k1++){ /* cptcovt number of covariates (excluding 1 and age or age*age) in the MODEL equation.
                                                      * loop on position k1 in the MODEL LINE */
       /* k counting number of combination of single dummies in the equation model */
       /* k4 counting single dummies in the equation model */
       /* k4q counting single quantitatives in the equation model */
       if( Dummy[k1]==0 && Typevar[k1]==0 ){ /* Dummy and Single, k1 is sorting according to MODEL, but k3 to resultline */
          /* k4+1= (not always if quant in model) position in the resultline V(Tvarsel)=Tvalsel=Tresult[nres][pos](value); V(Tvresult[nres][pos] (variable): V(variable)=value) */
         /* modelresult[k3]=k1: k3th position in the result line corresponds to the k1 position in the model line (doesn't work with products)*/
         /* Value in the (current nres) resultline of the variable at the k1th position in the model equation resultmodel[nres][k1]= k3 */
         /* resultmodel[nres][k1]=k3: k1th position in the model correspond to the k3 position in the resultline                        */
         /*      k3 is the position in the nres result line of the k1th variable of the model equation                                  */
         /* Tvarsel[k3]: Name of the variable at the k3th position in the result line.                                                  */
         /* Tvalsel[k3]: Value of the variable at the k3th position in the result line.                                                 */
         /* Tresult[nres][result_position]= value of the dummy variable at the result_position in the nres resultline                   */
         /* Tvresult[nres][result_position]= name of the dummy variable at the result_position in the nres resultline                     */
         /* Tinvresult[nres][Name of a dummy variable]= value of the variable in the result line                                        */
         /* TinvDoQresult[nres][Name of a Dummy or Q variable]= value of the variable in the result line                                                      */
         k3= resultmodel[nres][k1]; /* From position k1 in model get position k3 in result line */
         /* nres=1 k1=2 resultmodel[2(V4)] = 1=k3 ; k1=3 resultmodel[3(V3)] = 2=k3*/
         k2=(int)Tvarsel[k3]; /* from position k3 in resultline get name k2: nres=1 k1=2=>k3=1 Tvarsel[resultmodel[2]]= Tvarsel[1] = 4=k2 (V4); k1=3=>k3=2 Tvarsel[2]=3 (V3)*/
         k+=Tvalsel[k3]*pow(2,k4);  /* nres=1 k1=2 Tvalsel[1]=1 (V4=1); k1=3 k3=2 Tvalsel[2]=0 (V3=0) */
         TinvDoQresult[nres][(int)Tvarsel[k3]]=Tvalsel[k3]; /* TinvDoQresult[nres][Name]=Value; stores the value into the name of the variable. */
         /* Tinvresult[nres][4]=1 */
         /* Tresult[nres][k4+1]=Tvalsel[k3];/\* Tresult[nres=2][1]=1(V4=1)  Tresult[nres=2][2]=0(V3=0) *\/ */
         Tresult[nres][k3]=Tvalsel[k3];/* Tresult[nres=2][1]=1(V4=1)  Tresult[nres=2][2]=0(V3=0) */
         /* Tvresult[nres][k4+1]=(int)Tvarsel[k3];/\* Tvresult[nres][1]=4 Tvresult[nres][3]=1 *\/ */
         Tvresult[nres][k3]=(int)Tvarsel[k3];/* Tvresult[nres][1]=4 Tvresult[nres][3]=1 */
         Tinvresult[nres][(int)Tvarsel[k3]]=Tvalsel[k3]; /* Tinvresult[nres][4]=1 */
         precov[nres][k1]=Tvalsel[k3]; /* Value from resultline of the variable at the k1 position in the model */
         printf("Decoderesult Dummy k=%d, k1=%d precov[nres=%d][k1=%d]=%.f V(k2=V%d)= Tvalsel[%d]=%d, 2**(%d)\n",k, k1, nres, k1,precov[nres][k1], k2, k3, (int)Tvalsel[k3], k4);
         k4++;;
       }else if( Dummy[k1]==1 && Typevar[k1]==0 ){ /* Quantitative and single */
         /* Tqresult[nres][result_position]= value of the variable at the result_position in the nres resultline                                 */
         /* Tvqresult[nres][result_position]= id of the variable at the result_position in the nres resultline                                 */
         /* Tqinvresult[nres][Name of a quantitative variable]= value of the variable in the result line                                                      */
         k3q= resultmodel[nres][k1]; /* resultmodel[1(V5)] = 5 =k3q */
         k2q=(int)Tvarsel[k3q]; /*  Name of variable at k3q th position in the resultline */
         /* Tvarsel[resultmodel[1]]= Tvarsel[1] = 4=k2 */
         /* Tqresult[nres][k4q+1]=Tvalsel[k3q]; /\* Tqresult[nres][1]=25.1 *\/ */
         /* Tvresult[nres][k4q+1]=(int)Tvarsel[k3q];/\* Tvresult[nres][1]=4 Tvresult[nres][3]=1 *\/ */
         /* Tvqresult[nres][k4q+1]=(int)Tvarsel[k3q]; /\* Tvqresult[nres][1]=5 *\/ */
         Tqresult[nres][k3q]=Tvalsel[k3q]; /* Tqresult[nres][1]=25.1 */
         Tvresult[nres][k3q]=(int)Tvarsel[k3q];/* Tvresult[nres][1]=4 Tvresult[nres][3]=1 */
         Tvqresult[nres][k3q]=(int)Tvarsel[k3q]; /* Tvqresult[nres][1]=5 */
         Tqinvresult[nres][(int)Tvarsel[k3q]]=Tvalsel[k3q]; /* Tqinvresult[nres][5]=25.1 */
         TinvDoQresult[nres][(int)Tvarsel[k3q]]=Tvalsel[k3q]; /* Tqinvresult[nres][5]=25.1 */
         precov[nres][k1]=Tvalsel[k3q];
         printf("Decoderesult Quantitative nres=%d,precov[nres=%d][k1=%d]=%.f V(k2q=V%d)= Tvalsel[%d]=%d, Tvarsel[%d]=%f\n",nres, nres, k1,precov[nres][k1], k2q, k3q, Tvarsel[k3q], k3q, Tvalsel[k3q]);
         k4q++;;
       }else if( Dummy[k1]==2 ){ /* For dummy with age product */
         /* Tvar[k1]; */ /* Age variable */
         /* Wrong we want the value of variable name Tvar[k1] */
         
         k3= resultmodel[nres][k1]; /* nres=1 k1=2 resultmodel[2(V4)] = 1=k3 ; k1=3 resultmodel[3(V3)] = 2=k3*/
         k2=(int)Tvarsel[k3]; /* nres=1 k1=2=>k3=1 Tvarsel[resultmodel[2]]= Tvarsel[1] = 4=k2 (V4); k1=3=>k3=2 Tvarsel[2]=3 (V3)*/
         TinvDoQresult[nres][(int)Tvarsel[k3]]=Tvalsel[k3]; /* TinvDoQresult[nres][4]=1 */
         precov[nres][k1]=Tvalsel[k3];
         printf("Decoderesult Dummy with age k=%d, k1=%d precov[nres=%d][k1=%d]=%.f Tvar[%d]=V%d k2=Tvarsel[%d]=%d Tvalsel[%d]=%d\n",k, k1, nres, k1,precov[nres][k1], k1, Tvar[k1], k3,(int)Tvarsel[k3], k3, (int)Tvalsel[k3]);
       }else if( Dummy[k1]==3 ){ /* For quant with age product */
         k3q= resultmodel[nres][k1]; /* resultmodel[1(V5)] = 25.1=k3q */
         k2q=(int)Tvarsel[k3q]; /*  Tvarsel[resultmodel[1]]= Tvarsel[1] = 4=k2 */
         TinvDoQresult[nres][(int)Tvarsel[k3q]]=Tvalsel[k3q]; /* TinvDoQresult[nres][5]=25.1 */
         precov[nres][k1]=Tvalsel[k3q];
         printf("Decoderesult Quantitative with age nres=%d, k1=%d, precov[nres=%d][k1=%d]=%f Tvar[%d]=V%d V(k2q=%d)= Tvarsel[%d]=%d, Tvalsel[%d]=%f\n",nres, k1, nres, k1,precov[nres][k1], k1,  Tvar[k1], k2q, k3q, Tvarsel[k3q], k3q, Tvalsel[k3q]);
       }else if(Typevar[k1]==2 ){ /* For product quant or dummy (not with age) */
         precov[nres][k1]=TinvDoQresult[nres][Tvardk[k1][1]] * TinvDoQresult[nres][Tvardk[k1][2]];      
         printf("Decoderesult Quantitative or Dummy (not with age) nres=%d k1=%d precov[nres=%d][k1=%d]=%.f V%d(=%.f) * V%d(=%.f) \n",nres, k1, nres, k1,precov[nres][k1], Tvardk[k1][1], TinvDoQresult[nres][Tvardk[k1][1]], Tvardk[k1][2], TinvDoQresult[nres][Tvardk[k1][2]]);
       }else{
         printf("Error Decoderesult probably a product  Dummy[%d]==%d && Typevar[%d]==%d\n", k1, Dummy[k1], k1, Typevar[k1]);
         fprintf(ficlog,"Error Decoderesult probably a product  Dummy[%d]==%d && Typevar[%d]==%d\n", k1, Dummy[k1], k1, Typevar[k1]);
       }
   }    }
     
     TKresult[nres]=++k; /* Number of combinations of dummies for the nresult and the model =Tvalsel[k3]*pow(2,k4) + 1*/
     return (0);
   }
   
   cov[1]=1;  int decodemodel( char model[], int lastobs)
   tj=cptcoveff;   /**< This routine decodes the model and returns:
   if (cptcovn<1) {tj=1;ncodemax[1]=1;}          * Model  V1+V2+V3+V8+V7*V8+V5*V6+V8*age+V3*age+age*age
   j1=0;          * - nagesqr = 1 if age*age in the model, otherwise 0.
   for(t=1; t<=tj;t++){          * - cptcovt total number of covariates of the model nbocc(+)+1 = 8 excepting constant and age and age*age
     for(i1=1; i1<=ncodemax[t];i1++){           * - cptcovn or number of covariates k of the models excluding age*products =6 and age*age
       j1++;          * - cptcovage number of covariates with age*products =2
       if  (cptcovn>0) {          * - cptcovs number of simple covariates
         fprintf(ficresprob, "\n#********** Variable ");           * - Tvar[k] is the id of the kth covariate Tvar[1]@12 {1, 2, 3, 8, 10, 11, 8, 3, 7, 8, 5, 6}, thus Tvar[5=V7*V8]=10
         for (z1=1; z1<=cptcoveff; z1++) fprintf(ficresprob, "V%d=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtab[j1][z1]]);          *     which is a new column after the 9 (ncovcol) variables. 
         fprintf(ficresprob, "**********\n#\n");          * - if k is a product Vn*Vm, covar[k][i] is filled with correct values for each individual
         fprintf(ficresprobcov, "\n#********** Variable ");           * - Tprod[l] gives the kth covariates of the product Vn*Vm l=1 to cptcovprod-cptcovage
         for (z1=1; z1<=cptcoveff; z1++) fprintf(ficresprobcov, "V%d=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtab[j1][z1]]);          *    Tprod[1]@2 {5, 6}: position of first product V7*V8 is 5, and second V5*V6 is 6.
         fprintf(ficresprobcov, "**********\n#\n");          * - Tvard[k]  p Tvard[1][1]@4 {7, 8, 5, 6} for V7*V8 and V5*V6 .
                   */
         fprintf(ficgp, "\n#********** Variable ");   /* V2+V1+V4+V3*age Tvar[4]=3 ; V1+V2*age Tvar[2]=2; V1+V1*age Tvar[2]=1, Tage[1]=2 */
         for (z1=1; z1<=cptcoveff; z1++) fprintf(ficgp, " V%d=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtab[j1][z1]]);  {
         fprintf(ficgp, "**********\n#\n");    int i, j, k, ks, v;
             int  j1, k1, k2, k3, k4;
             char modelsav[80];
         fprintf(fichtm, "\n<hr  size=\"2\" color=\"#EC5E5E\">********** Variable ");     char stra[80], strb[80], strc[80], strd[80],stre[80];
         for (z1=1; z1<=cptcoveff; z1++) fprintf(fichtm, "V%d=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtab[j1][z1]]);    char *strpt;
         fprintf(fichtm, "**********\n<hr size=\"2\" color=\"#EC5E5E\">");  
             /*removespace(model);*/
         fprintf(ficresprobcor, "\n#********** Variable ");        if (strlen(model) >1){ /* If there is at least 1 covariate */
         for (z1=1; z1<=cptcoveff; z1++) fprintf(ficresprobcor, "V%d=%d ",Tvaraff[z1],nbcode[Tvaraff[z1]][codtab[j1][z1]]);      j=0, j1=0, k1=0, k2=-1, ks=0, cptcovn=0;
         fprintf(ficresprobcor, "**********\n#");          if (strstr(model,"AGE") !=0){
       }        printf("Error. AGE must be in lower case 'age' model=1+age+%s. ",model);
         fprintf(ficlog,"Error. AGE must be in lower case model=1+age+%s. ",model);fflush(ficlog);
         return 1;
       }
       if (strstr(model,"v") !=0){
         printf("Error. 'v' must be in upper case 'V' model=%s ",model);
         fprintf(ficlog,"Error. 'v' must be in upper case model=%s ",model);fflush(ficlog);
         return 1;
       }
       strcpy(modelsav,model); 
       if ((strpt=strstr(model,"age*age")) !=0){
         printf(" strpt=%s, model=%s\n",strpt, model);
         if(strpt != model){
           printf("Error in model: 'model=%s'; 'age*age' should in first place before other covariates\n \
    'model=1+age+age*age+V1.' or 'model=1+age+age*age+V1+V1*age.', please swap as well as \n \
    corresponding column of parameters.\n",model);
           fprintf(ficlog,"Error in model: 'model=%s'; 'age*age' should in first place before other covariates\n \
    'model=1+age+age*age+V1.' or 'model=1+age+age*age+V1+V1*age.', please swap as well as \n \
    corresponding column of parameters.\n",model); fflush(ficlog);
           return 1;
         }
         nagesqr=1;
         if (strstr(model,"+age*age") !=0)
           substrchaine(modelsav, model, "+age*age");
         else if (strstr(model,"age*age+") !=0)
           substrchaine(modelsav, model, "age*age+");
         else 
           substrchaine(modelsav, model, "age*age");
       }else
         nagesqr=0;
       if (strlen(modelsav) >1){
         j=nbocc(modelsav,'+'); /**< j=Number of '+' */
         j1=nbocc(modelsav,'*'); /**< j1=Number of '*' */
         cptcovs=j+1-j1; /**<  Number of simple covariates V1+V1*age+V3 +V3*V4+age*age=> V1 + V3 =5-3=2  */
         cptcovt= j+1; /* Number of total covariates in the model, not including
                        * cst, age and age*age 
                        * V1+V1*age+ V3 + V3*V4+age*age=> 3+1=4*/
         /* including age products which are counted in cptcovage.
          * but the covariates which are products must be treated 
          * separately: ncovn=4- 2=2 (V1+V3). */
         cptcovprod=j1; /**< Number of products  V1*V2 +v3*age = 2 */
         cptcovprodnoage=0; /**< Number of covariate products without age: V3*V4 =1  */
               
       for (age=bage; age<=fage; age ++){         
         cov[2]=age;        /*   Design
         for (k=1; k<=cptcovn;k++) {         *  V1   V2   V3   V4  V5  V6  V7  V8  V9 Weight
           cov[2+k]=nbcode[Tvar[k]][codtab[j1][Tvar[k]]];         *  <          ncovcol=8                >
         }         * Model V2 + V1 + V3*age + V3 + V5*V6 + V7*V8 + V8*age + V8
         for (k=1; k<=cptcovage;k++) cov[2+Tage[k]]=cov[2+Tage[k]]*cov[2];         *   k=  1    2      3       4     5       6      7        8
         for (k=1; k<=cptcovprod;k++)         *  cptcovn number of covariates (not including constant and age ) = # of + plus 1 = 7+1=8
           cov[2+Tprod[k]]=nbcode[Tvard[k][1]][codtab[ij][Tvard[k][1]]]*nbcode[Tvard[k][2]][codtab[ij][Tvard[k][2]]];         *  covar[k,i], value of kth covariate if not including age for individual i:
                  *       covar[1][i]= (V1), covar[4][i]=(V4), covar[8][i]=(V8)
         gradg=matrix(1,npar,1,(nlstate)*(nlstate+ndeath));         *  Tvar[k] # of the kth covariate:  Tvar[1]=2  Tvar[2]=1 Tvar[4]=3 Tvar[8]=8
         trgradg=matrix(1,(nlstate)*(nlstate+ndeath),1,npar);         *       if multiplied by age: V3*age Tvar[3=V3*age]=3 (V3) Tvar[7]=8 and 
         gp=vector(1,(nlstate)*(nlstate+ndeath));         *  Tage[++cptcovage]=k
         gm=vector(1,(nlstate)*(nlstate+ndeath));         *       if products, new covar are created after ncovcol with k1
              *  Tvar[k]=ncovcol+k1; # of the kth covariate product:  Tvar[5]=ncovcol+1=10  Tvar[6]=ncovcol+1=11
         for(theta=1; theta <=npar; theta++){         *  Tprod[k1]=k; Tprod[1]=5 Tprod[2]= 6; gives the position of the k1th product
           for(i=1; i<=npar; i++)         *  Tvard[k1][1]=m Tvard[k1][2]=m; Tvard[1][1]=5 (V5) Tvard[1][2]=6 Tvard[2][1]=7 (V7) Tvard[2][2]=8
             xp[i] = x[i] + (i==theta ?delti[theta]:(double)0);         *  Tvar[cptcovn+k2]=Tvard[k1][1];Tvar[cptcovn+k2+1]=Tvard[k1][2];
                    *  Tvar[8+1]=5;Tvar[8+2]=6;Tvar[8+3]=7;Tvar[8+4]=8 inverted
           pmij(pmmij,cov,ncovmodel,xp,nlstate);         *  V1   V2   V3   V4  V5  V6  V7  V8  V9  V10  V11
                    *  <          ncovcol=8                >
           k=0;         *       Model V2 + V1 + V3*age + V3 + V5*V6 + V7*V8 + V8*age + V8    d1   d1   d2  d2
           for(i=1; i<= (nlstate); i++){         *          k=  1    2      3       4     5       6      7        8    9   10   11  12
             for(j=1; j<=(nlstate+ndeath);j++){         *     Tvar[k]= 2    1      3       3    10      11      8        8    5    6    7   8
               k=k+1;         * p Tvar[1]@12={2,   1,     3,      3,  11,     10,     8,       8,   7,   8,   5,  6}
               gp[k]=pmmij[i][j];         * p Tprod[1]@2={                         6, 5}
          *p Tvard[1][1]@4= {7, 8, 5, 6}
          * covar[k][i]= V2   V1      ?      V3    V5*V6?   V7*V8?  ?       V8   
          *  cov[Tage[kk]+2]=covar[Tvar[Tage[kk]]][i]*cov[2];
          *How to reorganize? Tvars(orted)
          * Model V1 + V2 + V3 + V8 + V5*V6 + V7*V8 + V3*age + V8*age
          * Tvars {2,   1,     3,      3,   11,     10,     8,       8,   7,   8,   5,  6}
          *       {2,   1,     4,      8,    5,      6,     3,       7}
          * Struct []
          */
         
         /* This loop fills the array Tvar from the string 'model'.*/
         /* j is the number of + signs in the model V1+V2+V3 j=2 i=3 to 1 */
         /*   modelsav=V2+V1+V4+age*V3 strb=age*V3 stra=V2+V1+V4  */
         /*        k=4 (age*V3) Tvar[k=4]= 3 (from V3) Tage[cptcovage=1]=4 */
         /*        k=3 V4 Tvar[k=3]= 4 (from V4) */
         /*        k=2 V1 Tvar[k=2]= 1 (from V1) */
         /*        k=1 Tvar[1]=2 (from V2) */
         /*        k=5 Tvar[5] */
         /* for (k=1; k<=cptcovn;k++) { */
         /*        cov[2+k]=nbcode[Tvar[k]][codtabm(ij,Tvar[k])]; */
         /*        } */
         /* for (k=1; k<=cptcovage;k++) cov[2+Tage[k]]=nbcode[Tvar[Tage[k]]][codtabm(ij,Tvar[Tage[k])]]*cov[2]; */
         /*
          * Treating invertedly V2+V1+V3*age+V2*V4 is as if written V2*V4 +V3*age + V1 + V2 */
         for(k=cptcovt; k>=1;k--){ /**< Number of covariates not including constant and age, neither age*age*/
           Tvar[k]=0; Tprod[k]=0; Tposprod[k]=0;
         }
         cptcovage=0;
         for(k=1; k<=cptcovt;k++){ /* Loop on total covariates of the model line */
           cutl(stra,strb,modelsav,'+'); /* keeps in strb after the first '+' cutl from left to right
                                            modelsav==V2+V1+V5*age+V4+V3*age strb=V3*age stra=V2+V1V5*age+V4 */    /* <model> "V5+V4+V3+V4*V3+V5*age+V1*age+V1" strb="V5" stra="V4+V3+V4*V3+V5*age+V1*age+V1" */
           if (nbocc(modelsav,'+')==0)
             strcpy(strb,modelsav); /* and analyzes it */
           /*      printf("i=%d a=%s b=%s sav=%s\n",i, stra,strb,modelsav);*/
           /*scanf("%d",i);*/
           if (strchr(strb,'*')) {  /**< Model includes a product V2+V1+V5*age+ V4+V3*age strb=V3*age */
             cutl(strc,strd,strb,'*'); /**< k=1 strd*strc  Vm*Vn: strb=V3*age(input) strc=age strd=V3 ; V3*V2 strc=V2, strd=V3 */
             if (strcmp(strc,"age")==0) { /**< Model includes age: Vn*age */
               /* covar is not filled and then is empty */
               cptcovprod--;
               cutl(stre,strb,strd,'V'); /* strd=V3(input): stre="3" */
               Tvar[k]=atoi(stre);  /* V2+V1+V5*age+V4+V3*age Tvar[5]=3 ; V1+V2*age Tvar[2]=2; V1+V1*age Tvar[2]=1 */
               Typevar[k]=1;  /* 1 for age product */
               cptcovage++; /* Counts the number of covariates which include age as a product */
               Tage[cptcovage]=k;  /*  V2+V1+V4+V3*age Tvar[4]=3, Tage[1] = 4 or V1+V1*age Tvar[2]=1, Tage[1]=2 */
               /*printf("stre=%s ", stre);*/
             } else if (strcmp(strd,"age")==0) { /* or age*Vn */
               cptcovprod--;
               cutl(stre,strb,strc,'V');
               Tvar[k]=atoi(stre);
               Typevar[k]=1;  /* 1 for age product */
               cptcovage++;
               Tage[cptcovage]=k;
             } else {  /* Age is not in the model product V2+V1+V1*V4+V3*age+V3*V2  strb=V3*V2*/
               /* loops on k1=1 (V3*V2) and k1=2 V4*V3 */
               cptcovn++;
               cptcovprodnoage++;k1++;
               cutl(stre,strb,strc,'V'); /* strc= Vn, stre is n; strb=V3*V2 stre=3 strc=*/
               Tvar[k]=ncovcol+nqv+ntv+nqtv+k1; /* For model-covariate k tells which data-covariate to use but
                                                   because this model-covariate is a construction we invent a new column
                                                   which is after existing variables ncovcol+nqv+ntv+nqtv + k1
                                                   If already ncovcol=4 and model= V2 + V1 + V1*V4 + age*V3 + V3*V2
                                                   thus after V4 we invent V5 and V6 because age*V3 will be computed in 4
                                                   Tvar[3=V1*V4]=4+1=5 Tvar[5=V3*V2]=4 + 2= 6, Tvar[4=age*V3]=4 etc */
               /* Please remark that the new variables are model dependent */
               /* If we have 4 variable but the model uses only 3, like in
                * model= V1 + age*V1 + V2 + V3 + age*V2 + age*V3 + V1*V2 + V1*V3
                *  k=     1     2       3   4     5        6        7       8
                * Tvar[k]=1     1       2   3     2        3       (5       6) (and not 4 5 because of V4 missing)
                * Tage[kk]    [1]= 2           [2]=5      [3]=6                  kk=1 to cptcovage=3
                * Tvar[Tage[kk]][1]=2          [2]=2      [3]=3
                */
               Typevar[k]=2;  /* 2 for double fixed dummy covariates */
               cutl(strc,strb,strd,'V'); /* strd was Vm, strc is m */
               Tprod[k1]=k;  /* Tprod[1]=3(=V1*V4) for V2+V1+V1*V4+age*V3+V3*V2  */
               Tposprod[k]=k1; /* Tposprod[3]=1, Tposprod[2]=5 */
               Tvard[k1][1] =atoi(strc); /* m 1 for V1*/
               Tvardk[k][1] =atoi(strc); /* m 1 for V1*/
               Tvard[k1][2] =atoi(stre); /* n 4 for V4*/
               Tvardk[k][2] =atoi(stre); /* n 4 for V4*/
               k2=k2+2;  /* k2 is initialize to -1, We want to store the n and m in Vn*Vm at the end of Tvar */
               /* Tvar[cptcovt+k2]=Tvard[k1][1]; /\* Tvar[(cptcovt=4+k2=1)=5]= 1 (V1) *\/ */
               /* Tvar[cptcovt+k2+1]=Tvard[k1][2];  /\* Tvar[(cptcovt=4+(k2=1)+1)=6]= 4 (V4) *\/ */
               /*ncovcol=4 and model=V2+V1+V1*V4+age*V3+V3*V2, Tvar[3]=5, Tvar[4]=6, cptcovt=5 */
               /*                     1  2   3      4     5 | Tvar[5+1)=1, Tvar[7]=2   */
               for (i=1; i<=lastobs;i++){
                 /* Computes the new covariate which is a product of
                    covar[n][i]* covar[m][i] and stores it at ncovol+k1 May not be defined */
                 covar[ncovcol+k1][i]=covar[atoi(stre)][i]*covar[atoi(strc)][i];
             }              }
             } /* End age is not in the model */
           } /* End if model includes a product */
           else { /* not a product */
             /*printf("d=%s c=%s b=%s\n", strd,strc,strb);*/
             /*  scanf("%d",i);*/
             cutl(strd,strc,strb,'V');
             ks++; /**< Number of simple covariates dummy or quantitative, fixe or varying */
             cptcovn++; /** V4+V3+V5: V4 and V3 timevarying dummy covariates, V5 timevarying quantitative */
             Tvar[k]=atoi(strd);
             Typevar[k]=0;  /* 0 for simple covariates */
           }
           strcpy(modelsav,stra);  /* modelsav=V2+V1+V4 stra=V2+V1+V4 */ 
                                   /*printf("a=%s b=%s sav=%s\n", stra,strb,modelsav);
                                     scanf("%d",i);*/
         } /* end of loop + on total covariates */
       } /* end if strlen(modelsave == 0) age*age might exist */
     } /* end if strlen(model == 0) */
     
     /*The number n of Vn is stored in Tvar. cptcovage =number of age covariate. Tage gives the position of age. cptcovprod= number of products.
       If model=V1+V1*age then Tvar[1]=1 Tvar[2]=1 cptcovage=1 Tage[1]=2 cptcovprod=0*/
     
     /* printf("tvar1=%d tvar2=%d tvar3=%d cptcovage=%d Tage=%d",Tvar[1],Tvar[2],Tvar[3],cptcovage,Tage[1]);
        printf("cptcovprod=%d ", cptcovprod);
        fprintf(ficlog,"cptcovprod=%d ", cptcovprod);
        scanf("%d ",i);*/
   
   
   /* Until here, decodemodel knows only the grammar (simple, product, age*) of the model but not what kind
      of variable (dummy vs quantitative, fixed vs time varying) is behind. But we know the # of each. */
   /* ncovcol= 1, nqv=1 | ntv=2, nqtv= 1  = 5 possible variables data: 2 fixed 3, varying
      model=        V5 + V4 +V3 + V4*V3 + V5*age + V2 + V1*V2 + V1*age + V5*age, V1 is not used saving its place
      k =           1    2   3     4       5       6      7      8        9
      Tvar[k]=      5    4   3 1+1+2+1+1=6 5       2      7      1        5
      Typevar[k]=   0    0   0     2       1       0      2      1        0
      Fixed[k]      1    1   1     1       3       0    0 or 2   2        3
      Dummy[k]      1    0   0     0       3       1      1      2        3
             Tmodelind[combination of covar]=k;
   */  
   /* Dispatching between quantitative and time varying covariates */
     /* If Tvar[k] >ncovcol it is a product */
     /* Tvar[k] is the value n of Vn with n varying for 1 to nvcol, or p  Vp=Vn*Vm for product */
           /* Computing effective variables, ie used by the model, that is from the cptcovt variables */
     printf("Model=1+age+%s\n\
   Typevar: 0 for simple covariate (dummy, quantitative, fixed or varying), 1 for age product, 2 for  product \n\
   Fixed[k] 0=fixed (product or simple), 1 varying, 2 fixed with age product, 3 varying with age product \n\
   Dummy[k] 0=dummy (0 1), 1 quantitative (single or product without age), 2 dummy with age product, 3 quant with age product\n",model);
     fprintf(ficlog,"Model=1+age+%s\n\
   Typevar: 0 for simple covariate (dummy, quantitative, fixed or varying), 1 for age product, 2 for  product \n\
   Fixed[k] 0=fixed (product or simple), 1 varying, 2 fixed with age product, 3 varying with age product \n\
   Dummy[k] 0=dummy (0 1), 1 quantitative (single or product without age), 2 dummy with age product, 3 quant with age product\n",model);
     for(k=-1;k<=cptcovt; k++){ Fixed[k]=0; Dummy[k]=0;}
     for(k=1, ncovf=0, nsd=0, nsq=0, ncovv=0, ncova=0, ncoveff=0, nqfveff=0, ntveff=0, nqtveff=0;k<=cptcovt; k++){ /* or cptocvt */
       if (Tvar[k] <=ncovcol && Typevar[k]==0 ){ /* Simple fixed dummy (<=ncovcol) covariates */
         Fixed[k]= 0;
         Dummy[k]= 0;
         ncoveff++;
         ncovf++;
         nsd++;
         modell[k].maintype= FTYPE;
         TvarsD[nsd]=Tvar[k];
         TvarsDind[nsd]=k;
         TnsdVar[Tvar[k]]=nsd;
         TvarF[ncovf]=Tvar[k];
         TvarFind[ncovf]=k;
         TvarFD[ncoveff]=Tvar[k]; /* TvarFD[1]=V1 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
         TvarFDind[ncoveff]=k; /* TvarFDind[1]=9 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
       }else if( Tvar[k] <=ncovcol &&  Typevar[k]==2){ /* Product of fixed dummy (<=ncovcol) covariates */
         Fixed[k]= 0;
         Dummy[k]= 0;
         ncoveff++;
         ncovf++;
         modell[k].maintype= FTYPE;
         TvarF[ncovf]=Tvar[k];
         /* TnsdVar[Tvar[k]]=nsd; */ /* To be done */
         TvarFind[ncovf]=k;
         TvarFD[ncoveff]=Tvar[k]; /* TvarFD[1]=V1 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
         TvarFDind[ncoveff]=k; /* TvarFDind[1]=9 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */
       }else if( Tvar[k] <=ncovcol+nqv && Typevar[k]==0){/* Remind that product Vn*Vm are added in k Only simple fixed quantitative variable */
         Fixed[k]= 0;
         Dummy[k]= 1;
         nqfveff++;
         modell[k].maintype= FTYPE;
         modell[k].subtype= FQ;
         nsq++;
         TvarsQ[nsq]=Tvar[k]; /* Gives the variable name (extended to products) of first nsq simple quantitative covariates (fixed or time vary see below */
         TvarsQind[nsq]=k;    /* Gives the position in the model equation of the first nsq simple quantitative covariates (fixed or time vary) */
         ncovf++;
         TvarF[ncovf]=Tvar[k];
         TvarFind[ncovf]=k;
         TvarFQ[nqfveff]=Tvar[k]-ncovcol; /* TvarFQ[1]=V2-1=1st in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple fixed quantitative variable */
         TvarFQind[nqfveff]=k; /* TvarFQind[1]=6 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple fixed quantitative variable */
       }else if( Tvar[k] <=ncovcol+nqv+ntv && Typevar[k]==0){/* Only simple time varying dummy variables */
         Fixed[k]= 1;
         Dummy[k]= 0;
         ntveff++; /* Only simple time varying dummy variable */
         modell[k].maintype= VTYPE;
         modell[k].subtype= VD;
         nsd++;
         TvarsD[nsd]=Tvar[k];
         TvarsDind[nsd]=k;
         TnsdVar[Tvar[k]]=nsd; /* To be verified */
         ncovv++; /* Only simple time varying variables */
         TvarV[ncovv]=Tvar[k];
         TvarVind[ncovv]=k; /* TvarVind[2]=2  TvarVind[3]=3 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Any time varying singele */
         TvarVD[ntveff]=Tvar[k]; /* TvarVD[1]=V4  TvarVD[2]=V3 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple time varying dummy variable */
         TvarVDind[ntveff]=k; /* TvarVDind[1]=2 TvarVDind[2]=3 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple time varying dummy variable */
         printf("Quasi Tmodelind[%d]=%d,Tvar[Tmodelind[%d]]=V%d, ncovcol=%d, nqv=%d,Tvar[k]- ncovcol-nqv=%d\n",ntveff,k,ntveff,Tvar[k], ncovcol, nqv,Tvar[k]- ncovcol-nqv);
         printf("Quasi TmodelInvind[%d]=%d\n",k,Tvar[k]- ncovcol-nqv);
       }else if( Tvar[k] <=ncovcol+nqv+ntv+nqtv  && Typevar[k]==0){ /* Only simple time varying quantitative variable V5*/
         Fixed[k]= 1;
         Dummy[k]= 1;
         nqtveff++;
         modell[k].maintype= VTYPE;
         modell[k].subtype= VQ;
         ncovv++; /* Only simple time varying variables */
         nsq++;
         TvarsQ[nsq]=Tvar[k]; /* k=1 Tvar=5 nsq=1 TvarsQ[1]=5 */ /* Gives the variable name (extended to products) of first nsq simple quantitative covariates (fixed or time vary here) */
         TvarsQind[nsq]=k; /* For single quantitative covariate gives the model position of each single quantitative covariate *//* Gives the position in the model equation of the first nsq simple quantitative covariates (fixed or time vary) */
         TvarV[ncovv]=Tvar[k];
         TvarVind[ncovv]=k; /* TvarVind[1]=1 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Any time varying singele */
         TvarVQ[nqtveff]=Tvar[k]; /* TvarVQ[1]=V5 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple time varying quantitative variable */
         TvarVQind[nqtveff]=k; /* TvarVQind[1]=1 in V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 */ /* Only simple time varying quantitative variable */
         TmodelInvQind[nqtveff]=Tvar[k]- ncovcol-nqv-ntv;/* Only simple time varying quantitative variable */
         /* Tmodeliqind[k]=nqtveff;/\* Only simple time varying quantitative variable *\/ */
         printf("Quasi TmodelQind[%d]=%d,Tvar[TmodelQind[%d]]=V%d, ncovcol=%d, nqv=%d, ntv=%d,Tvar[k]- ncovcol-nqv-ntv=%d\n",nqtveff,k,nqtveff,Tvar[k], ncovcol, nqv, ntv, Tvar[k]- ncovcol-nqv-ntv);
         printf("Quasi TmodelInvQind[%d]=%d\n",k,Tvar[k]- ncovcol-nqv-ntv);
       }else if (Typevar[k] == 1) {  /* product with age */
         ncova++;
         TvarA[ncova]=Tvar[k];
         TvarAind[ncova]=k;
         if (Tvar[k] <=ncovcol ){ /* Product age with fixed dummy covariatee */
           Fixed[k]= 2;
           Dummy[k]= 2;
           modell[k].maintype= ATYPE;
           modell[k].subtype= APFD;
           /* ncoveff++; */
         }else if( Tvar[k] <=ncovcol+nqv) { /* Remind that product Vn*Vm are added in k*/
           Fixed[k]= 2;
           Dummy[k]= 3;
           modell[k].maintype= ATYPE;
           modell[k].subtype= APFQ;                /*      Product age * fixed quantitative */
           /* nqfveff++;  /\* Only simple fixed quantitative variable *\/ */
         }else if( Tvar[k] <=ncovcol+nqv+ntv ){
           Fixed[k]= 3;
           Dummy[k]= 2;
           modell[k].maintype= ATYPE;
           modell[k].subtype= APVD;                /*      Product age * varying dummy */
           /* ntveff++; /\* Only simple time varying dummy variable *\/ */
         }else if( Tvar[k] <=ncovcol+nqv+ntv+nqtv){
           Fixed[k]= 3;
           Dummy[k]= 3;
           modell[k].maintype= ATYPE;
           modell[k].subtype= APVQ;                /*      Product age * varying quantitative */
           /* nqtveff++;/\* Only simple time varying quantitative variable *\/ */
         }
       }else if (Typevar[k] == 2) {  /* product without age */
         k1=Tposprod[k];
         if(Tvard[k1][1] <=ncovcol){
           if(Tvard[k1][2] <=ncovcol){
             Fixed[k]= 1;
             Dummy[k]= 0;
             modell[k].maintype= FTYPE;
             modell[k].subtype= FPDD;              /*      Product fixed dummy * fixed dummy */
             ncovf++; /* Fixed variables without age */
             TvarF[ncovf]=Tvar[k];
             TvarFind[ncovf]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv){
             Fixed[k]= 0;  /* or 2 ?*/
             Dummy[k]= 1;
             modell[k].maintype= FTYPE;
             modell[k].subtype= FPDQ;              /*      Product fixed dummy * fixed quantitative */
             ncovf++; /* Varying variables without age */
             TvarF[ncovf]=Tvar[k];
             TvarFind[ncovf]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv+ntv){
             Fixed[k]= 1;
             Dummy[k]= 0;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDD;              /*      Product fixed dummy * varying dummy */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv+ntv+nqtv){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDQ;              /*      Product fixed dummy * varying quantitative */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }
         }else if(Tvard[k1][1] <=ncovcol+nqv){
           if(Tvard[k1][2] <=ncovcol){
             Fixed[k]= 0;  /* or 2 ?*/
             Dummy[k]= 1;
             modell[k].maintype= FTYPE;
             modell[k].subtype= FPDQ;              /*      Product fixed quantitative * fixed dummy */
             ncovf++; /* Fixed variables without age */
             TvarF[ncovf]=Tvar[k];
             TvarFind[ncovf]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv+ntv){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDQ;              /*      Product fixed quantitative * varying dummy */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv+ntv+nqtv){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPQQ;              /*      Product fixed quantitative * varying quantitative */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }
         }else if(Tvard[k1][1] <=ncovcol+nqv+ntv){
           if(Tvard[k1][2] <=ncovcol){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDD;              /*      Product time varying dummy * fixed dummy */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDQ;              /*      Product time varying dummy * fixed quantitative */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv+ntv){
             Fixed[k]= 1;
             Dummy[k]= 0;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDD;              /*      Product time varying dummy * time varying dummy */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv+ntv+nqtv){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDQ;              /*      Product time varying dummy * time varying quantitative */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }
         }else if(Tvard[k1][1] <=ncovcol+nqv+ntv+nqtv){
           if(Tvard[k1][2] <=ncovcol){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDQ;              /*      Product time varying quantitative * fixed dummy */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPQQ;              /*      Product time varying quantitative * fixed quantitative */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv+ntv){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPDQ;              /*      Product time varying quantitative * time varying dummy */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }else if(Tvard[k1][2] <=ncovcol+nqv+ntv+nqtv){
             Fixed[k]= 1;
             Dummy[k]= 1;
             modell[k].maintype= VTYPE;
             modell[k].subtype= VPQQ;              /*      Product time varying quantitative * time varying quantitative */
             ncovv++; /* Varying variables without age */
             TvarV[ncovv]=Tvar[k];
             TvarVind[ncovv]=k;
           }
         }else{
           printf("Error unknown type of covariate: Tvard[%d][1]=%d,Tvard[%d][2]=%d\n",k1,Tvard[k1][1],k1,Tvard[k1][2]);
           fprintf(ficlog,"Error unknown type of covariate: Tvard[%d][1]=%d,Tvard[%d][2]=%d\n",k1,Tvard[k1][1],k1,Tvard[k1][2]);
         } /*end k1*/
       }else{
         printf("Error, current version can't treat for performance reasons, Tvar[%d]=%d, Typevar[%d]=%d\n", k, Tvar[k], k, Typevar[k]);
         fprintf(ficlog,"Error, current version can't treat for performance reasons, Tvar[%d]=%d, Typevar[%d]=%d\n", k, Tvar[k], k, Typevar[k]);
       }
       printf("Decodemodel, k=%d, Tvar[%d]=V%d,Typevar=%d, Fixed=%d, Dummy=%d\n",k, k,Tvar[k],Typevar[k],Fixed[k],Dummy[k]);
       printf("           modell[%d].maintype=%d, modell[%d].subtype=%d\n",k,modell[k].maintype,k,modell[k].subtype);
       fprintf(ficlog,"Decodemodel, k=%d, Tvar[%d]=V%d,Typevar=%d, Fixed=%d, Dummy=%d\n",k, k,Tvar[k],Typevar[k],Fixed[k],Dummy[k]);
     }
     /* Searching for doublons in the model */
     for(k1=1; k1<= cptcovt;k1++){
       for(k2=1; k2 <k1;k2++){
         /* if((Typevar[k1]==Typevar[k2]) && (Fixed[Tvar[k1]]==Fixed[Tvar[k2]]) && (Dummy[Tvar[k1]]==Dummy[Tvar[k2]] )){ */
         if((Typevar[k1]==Typevar[k2]) && (Fixed[k1]==Fixed[k2]) && (Dummy[k1]==Dummy[k2] )){
           if((Typevar[k1] == 0 || Typevar[k1] == 1)){ /* Simple or age product */
             if(Tvar[k1]==Tvar[k2]){
               printf("Error duplication in the model=%s at positions (+) %d and %d, Tvar[%d]=V%d, Tvar[%d]=V%d, Typevar=%d, Fixed=%d, Dummy=%d\n", model, k1,k2, k1, Tvar[k1], k2, Tvar[k2],Typevar[k1],Fixed[k1],Dummy[k1]);
               fprintf(ficlog,"Error duplication in the model=%s at positions (+) %d and %d, Tvar[%d]=V%d, Tvar[%d]=V%d, Typevar=%d, Fixed=%d, Dummy=%d\n", model, k1,k2, k1, Tvar[k1], k2, Tvar[k2],Typevar[k1],Fixed[k1],Dummy[k1]); fflush(ficlog);
               return(1);
           }            }
                     }else if (Typevar[k1] ==2){
           for(i=1; i<=npar; i++)            k3=Tposprod[k1];
             xp[i] = x[i] - (i==theta ?delti[theta]:(double)0);            k4=Tposprod[k2];
                 if( ((Tvard[k3][1]== Tvard[k4][1])&&(Tvard[k3][2]== Tvard[k4][2])) || ((Tvard[k3][1]== Tvard[k4][2])&&(Tvard[k3][2]== Tvard[k4][1])) ){
           pmij(pmmij,cov,ncovmodel,xp,nlstate);              printf("Error duplication in the model=%s at positions (+) %d and %d, V%d*V%d, Typevar=%d, Fixed=%d, Dummy=%d\n",model, k1,k2, Tvard[k3][1], Tvard[k3][2],Typevar[k1],Fixed[Tvar[k1]],Dummy[Tvar[k1]]);
           k=0;              fprintf(ficlog,"Error duplication in the model=%s at positions (+) %d and %d, V%d*V%d, Typevar=%d, Fixed=%d, Dummy=%d\n",model, k1,k2, Tvard[k3][1], Tvard[k3][2],Typevar[k1],Fixed[Tvar[k1]],Dummy[Tvar[k1]]); fflush(ficlog);
           for(i=1; i<=(nlstate); i++){              return(1);
             for(j=1; j<=(nlstate+ndeath);j++){  
               k=k+1;  
               gm[k]=pmmij[i][j];  
             }  
           }            }
        
           for(i=1; i<= (nlstate)*(nlstate+ndeath); i++)   
             gradg[theta][i]=(gp[i]-gm[i])/(double)2./delti[theta];    
         }          }
         }
       }
     }
     printf("ncoveff=%d, nqfveff=%d, ntveff=%d, nqtveff=%d, cptcovn=%d\n",ncoveff,nqfveff,ntveff,nqtveff,cptcovn);
     fprintf(ficlog,"ncoveff=%d, nqfveff=%d, ntveff=%d, nqtveff=%d, cptcovn=%d\n",ncoveff,nqfveff,ntveff,nqtveff,cptcovn);
     printf("ncovf=%d, ncovv=%d, ncova=%d, nsd=%d, nsq=%d\n",ncovf,ncovv,ncova,nsd,nsq);
     fprintf(ficlog,"ncovf=%d, ncovv=%d, ncova=%d, nsd=%d, nsq=%d\n",ncovf,ncovv,ncova,nsd, nsq);
     return (0); /* with covar[new additional covariate if product] and Tage if age */ 
     /*endread:*/
     printf("Exiting decodemodel: ");
     return (1);
   }
   
         for(j=1; j<=(nlstate)*(nlstate+ndeath);j++)  int calandcheckages(int imx, int maxwav, double *agemin, double *agemax, int *nberr, int *nbwarn )
           for(theta=1; theta <=npar; theta++)  {/* Check ages at death */
             trgradg[j][theta]=gradg[theta][j];    int i, m;
             int firstone=0;
         matprod2(dnewm,trgradg,1,(nlstate)*(nlstate+ndeath),1,npar,1,npar,matcov);     
         matprod2(doldm,dnewm,1,(nlstate)*(nlstate+ndeath),1,npar,1,(nlstate)*(nlstate+ndeath),gradg);    for (i=1; i<=imx; i++) {
         free_vector(gp,1,(nlstate+ndeath)*(nlstate+ndeath));      for(m=2; (m<= maxwav); m++) {
         free_vector(gm,1,(nlstate+ndeath)*(nlstate+ndeath));        if (((int)mint[m][i]== 99) && (s[m][i] <= nlstate)){
         free_matrix(trgradg,1,(nlstate+ndeath)*(nlstate+ndeath),1,npar);          anint[m][i]=9999;
         free_matrix(gradg,1,(nlstate+ndeath)*(nlstate+ndeath),1,npar);          if (s[m][i] != -2) /* Keeping initial status of unknown vital status */
             s[m][i]=-1;
         pmij(pmmij,cov,ncovmodel,x,nlstate);        }
                 if((int)moisdc[i]==99 && (int)andc[i]==9999 && s[m][i]>nlstate){
         k=0;          *nberr = *nberr + 1;
         for(i=1; i<=(nlstate); i++){          if(firstone == 0){
           for(j=1; j<=(nlstate+ndeath);j++){            firstone=1;
             k=k+1;          printf("Warning (#%d)! Date of death (month %2d and year %4d) of individual %ld on line %d was unknown but status is a death state %d at wave %d. If you don't know the vital status, please enter -2. If he/she is still alive but don't know the state, please code with '-1 or '.'. Here, we do not believe in a death, skipped.\nOther similar cases in log file\n", *nberr,(int)moisdc[i],(int)andc[i],num[i],i,s[m][i],m);
             mu[k][(int) age]=pmmij[i][j];  
           }  
         }          }
         for(i=1;i<=(nlstate)*(nlstate+ndeath);i++)          fprintf(ficlog,"Warning (#%d)! Date of death (month %2d and year %4d) of individual %ld on line %d was unknown but status is a death state %d at wave %d. If you don't know the vital status, please enter -2. If he/she is still alive but don't know the state, please code with '-1 or '.'. Here, we do not believe in a death, skipped.\n", *nberr,(int)moisdc[i],(int)andc[i],num[i],i,s[m][i],m);
           for(j=1;j<=(nlstate)*(nlstate+ndeath);j++)          s[m][i]=-1;  /* Droping the death status */
             varpij[i][j][(int)age] = doldm[i][j];        }
         if((int)moisdc[i]==99 && (int)andc[i]!=9999 && s[m][i]>nlstate){
         /*printf("\n%d ",(int)age);          (*nberr)++;
           for (i=1; i<=(nlstate)*(nlstate+ndeath);i++){          printf("Error (#%d)! Month of death of individual %ld on line %d was unknown (%2d) (year of death is %4d) and status is a death state %d at wave %d. Please impute an arbitrary (or not) month and rerun. Currently this transition to death will be skipped (status is set to -2).\nOther similar cases in log file\n", *nberr, num[i],i,(int)moisdc[i],(int)andc[i],s[m][i],m);
           printf("%e [%e ;%e] ",gm[i],gm[i]-2*sqrt(doldm[i][i]),gm[i]+2*sqrt(doldm[i][i]));          fprintf(ficlog,"Error (#%d)! Month of death of individual %ld on line %d was unknown (%2d) (year of death is %4d) and status is a death state %d at wave %d. Please impute an arbitrary (or not) month and rerun. Currently this transition to death will be skipped (status is set to -2).\n", *nberr, num[i],i,(int)moisdc[i],(int)andc[i],s[m][i],m);
           fprintf(ficlog,"%e [%e ;%e] ",gm[i],gm[i]-2*sqrt(doldm[i][i]),gm[i]+2*sqrt(doldm[i][i]));          s[m][i]=-2; /* We prefer to skip it (and to skip it in version 0.8a1 too */
           }*/        }
       }
         fprintf(ficresprob,"\n%d ",(int)age);    }
         fprintf(ficresprobcov,"\n%d ",(int)age);  
         fprintf(ficresprobcor,"\n%d ",(int)age);    for (i=1; i<=imx; i++)  {
       agedc[i]=(moisdc[i]/12.+andc[i])-(moisnais[i]/12.+annais[i]);
         for (i=1; i<=(nlstate)*(nlstate+ndeath);i++)      for(m=firstpass; (m<= lastpass); m++){
           fprintf(ficresprob,"%11.3e (%11.3e) ",mu[i][(int) age],sqrt(varpij[i][i][(int)age]));        if(s[m][i] >0  || s[m][i]==-1 || s[m][i]==-2 || s[m][i]==-4 || s[m][i]==-5){ /* What if s[m][i]=-1 */
         for (i=1; i<=(nlstate)*(nlstate+ndeath);i++){          if (s[m][i] >= nlstate+1) {
           fprintf(ficresprobcov,"%11.3e ",mu[i][(int) age]);            if(agedc[i]>0){
           fprintf(ficresprobcor,"%11.3e ",mu[i][(int) age]);              if((int)moisdc[i]!=99 && (int)andc[i]!=9999){
         }                agev[m][i]=agedc[i];
         i=0;                /*if(moisdc[i]==99 && andc[i]==9999) s[m][i]=-1;*/
         for (k=1; k<=(nlstate);k++){              }else {
           for (l=1; l<=(nlstate+ndeath);l++){                 if ((int)andc[i]!=9999){
             i=i++;                  nbwarn++;
             fprintf(ficresprobcov,"\n%d %d-%d",(int)age,k,l);                  printf("Warning negative age at death: %ld line:%d\n",num[i],i);
             fprintf(ficresprobcor,"\n%d %d-%d",(int)age,k,l);                  fprintf(ficlog,"Warning negative age at death: %ld line:%d\n",num[i],i);
             for (j=1; j<=i;j++){                  agev[m][i]=-1;
               fprintf(ficresprobcov," %11.3e",varpij[i][j][(int)age]);                }
               fprintf(ficresprobcor," %11.3e",varpij[i][j][(int) age]/sqrt(varpij[i][i][(int) age])/sqrt(varpij[j][j][(int)age]));  
             }              }
             } /* agedc > 0 */
           } /* end if */
           else if(s[m][i] !=9){ /* Standard case, age in fractional
                                    years but with the precision of a month */
             agev[m][i]=(mint[m][i]/12.+1./24.+anint[m][i])-(moisnais[i]/12.+1./24.+annais[i]);
             if((int)mint[m][i]==99 || (int)anint[m][i]==9999)
               agev[m][i]=1;
             else if(agev[m][i] < *agemin){ 
               *agemin=agev[m][i];
               printf(" Min anint[%d][%d]=%.2f annais[%d]=%.2f, agemin=%.2f\n",m,i,anint[m][i], i,annais[i], *agemin);
           }            }
         }/* end of loop for state */            else if(agev[m][i] >*agemax){
       } /* end of loop for age */              *agemax=agev[m][i];
               /* printf(" Max anint[%d][%d]=%.0f annais[%d]=%.0f, agemax=%.2f\n",m,i,anint[m][i], i,annais[i], *agemax);*/
       /* Confidence intervalle of pij  */            }
       /*            /*agev[m][i]=anint[m][i]-annais[i];*/
         fprintf(ficgp,"\nset noparametric;unset label");            /*     agev[m][i] = age[i]+2*m;*/
         fprintf(ficgp,"\nset log y;unset log x; set xlabel \"Age\";set ylabel \"probability (year-1)\"");          } /* en if 9*/
         fprintf(ficgp,"\nset ter png small\nset size 0.65,0.65");          else { /* =9 */
         fprintf(fichtm,"\n<br>Probability with  confidence intervals expressed in year<sup>-1</sup> :<a href=\"pijgr%s.png\">pijgr%s.png</A>, ",optionfilefiname,optionfilefiname);            /* printf("Debug num[%d]=%ld s[%d][%d]=%d\n",i,num[i], m,i, s[m][i]); */
         fprintf(fichtm,"\n<br><img src=\"pijgr%s.png\"> ",optionfilefiname);            agev[m][i]=1;
         fprintf(ficgp,"\nset out \"pijgr%s.png\"",optionfilefiname);            s[m][i]=-1;
         fprintf(ficgp,"\nplot \"%s\" every :::%d::%d u 1:2 \"\%%lf",k1,k2,xfilevarprob);          }
       */        }
         else if(s[m][i]==0) /*= 0 Unknown */
       /* Drawing ellipsoids of confidence of two variables p(k1-l1,k2-l2)*/          agev[m][i]=1;
       first1=1;        else{
       for (k2=1; k2<=(nlstate);k2++){          printf("Warning, num[%d]=%ld, s[%d][%d]=%d\n", i, num[i], m, i,s[m][i]); 
         for (l2=1; l2<=(nlstate+ndeath);l2++){           fprintf(ficlog, "Warning, num[%d]=%ld, s[%d][%d]=%d\n", i, num[i], m, i,s[m][i]); 
           if(l2==k2) continue;          agev[m][i]=0;
           j=(k2-1)*(nlstate+ndeath)+l2;        }
           for (k1=1; k1<=(nlstate);k1++){      } /* End for lastpass */
             for (l1=1; l1<=(nlstate+ndeath);l1++){     }
               if(l1==k1) continue;      
               i=(k1-1)*(nlstate+ndeath)+l1;    for (i=1; i<=imx; i++)  {
               if(i<=j) continue;      for(m=firstpass; (m<=lastpass); m++){
               for (age=bage; age<=fage; age ++){         if (s[m][i] > (nlstate+ndeath)) {
                 if ((int)age %5==0){          (*nberr)++;
                   v1=varpij[i][i][(int)age]/stepm*YEARM/stepm*YEARM;          printf("Error: on wave %d of individual %d status %d > (nlstate+ndeath)=(%d+%d)=%d\n",m,i,s[m][i],nlstate, ndeath, nlstate+ndeath);     
                   v2=varpij[j][j][(int)age]/stepm*YEARM/stepm*YEARM;          fprintf(ficlog,"Error: on wave %d of individual %d status %d > (nlstate+ndeath)=(%d+%d)=%d\n",m,i,s[m][i],nlstate, ndeath, nlstate+ndeath);     
                   cv12=varpij[i][j][(int)age]/stepm*YEARM/stepm*YEARM;          return 1;
                   mu1=mu[i][(int) age]/stepm*YEARM ;        }
                   mu2=mu[j][(int) age]/stepm*YEARM;      }
                   c12=cv12/sqrt(v1*v2);  
                   /* Computing eigen value of matrix of covariance */  
                   lc1=((v1+v2)+sqrt((v1+v2)*(v1+v2) - 4*(v1*v2-cv12*cv12)))/2.;  
                   lc2=((v1+v2)-sqrt((v1+v2)*(v1+v2) - 4*(v1*v2-cv12*cv12)))/2.;  
                   /* Eigen vectors */  
                   v11=(1./sqrt(1+(v1-lc1)*(v1-lc1)/cv12/cv12));  
                   /*v21=sqrt(1.-v11*v11); *//* error */  
                   v21=(lc1-v1)/cv12*v11;  
                   v12=-v21;  
                   v22=v11;  
                   tnalp=v21/v11;  
                   if(first1==1){  
                     first1=0;  
                     printf("%d %d%d-%d%d mu %.4e %.4e Var %.4e %.4e cor %.3f cov %.4e Eig %.3e %.3e 1stv %.3f %.3f tang %.3f\nOthers in log...\n",(int) age,k1,l1,k2,l2,mu1,mu2,v1,v2,c12,cv12,lc1,lc2,v11,v21,tnalp);  
                   }  
                   fprintf(ficlog,"%d %d%d-%d%d mu %.4e %.4e Var %.4e %.4e cor %.3f cov %.4e Eig %.3e %.3e 1stv %.3f %.3f tan %.3f\n",(int) age,k1,l1,k2,l2,mu1,mu2,v1,v2,c12,cv12,lc1,lc2,v11,v21,tnalp);  
                   /*printf(fignu*/  
                   /* mu1+ v11*lc1*cost + v12*lc2*sin(t) */  
                   /* mu2+ v21*lc1*cost + v22*lc2*sin(t) */  
                   if(first==1){  
                     first=0;  
                     fprintf(ficgp,"\nset parametric;unset label");  
                     fprintf(ficgp,"\nset log y;set log x; set xlabel \"p%1d%1d (year-1)\";set ylabel \"p%1d%1d (year-1)\"",k1,l1,k2,l2);  
                     fprintf(ficgp,"\nset ter png small\nset size 0.65,0.65");  
                     fprintf(fichtm,"\n<br>Ellipsoids of confidence cov(p%1d%1d,p%1d%1d) expressed in year<sup>-1</sup> :<a href=\"varpijgr%s%d%1d%1d-%1d%1d.png\">varpijgr%s%d%1d%1d-%1d%1d.png</A>, ",k1,l1,k2,l2,optionfilefiname, j1,k1,l1,k2,l2,optionfilefiname, j1,k1,l1,k2,l2);  
                     fprintf(fichtm,"\n<br><img src=\"varpijgr%s%d%1d%1d-%1d%1d.png\"> ",optionfilefiname, j1,k1,l1,k2,l2);  
                     fprintf(fichtm,"\n<br> Correlation at age %d (%.3f),",(int) age, c12);  
                     fprintf(ficgp,"\nset out \"varpijgr%s%d%1d%1d-%1d%1d.png\"",optionfilefiname, j1,k1,l1,k2,l2);  
                     fprintf(ficgp,"\nset label \"%d\" at %11.3e,%11.3e center",(int) age, mu1,mu2);  
                     fprintf(ficgp,"\n# Age %d, p%1d%1d - p%1d%1d",(int) age, k1,l1,k2,l2);  
                     fprintf(ficgp,"\nplot [-pi:pi] %11.3e+ %.3f*(%11.3e*%11.3e*cos(t)+%11.3e*%11.3e*sin(t)), %11.3e +%.3f*(%11.3e*%11.3e*cos(t)+%11.3e*%11.3e*sin(t)) not",\  
                             mu1,std,v11,sqrt(lc1),v12,sqrt(lc2),\  
                             mu2,std,v21,sqrt(lc1),v22,sqrt(lc2));  
                   }else{  
                     first=0;  
                     fprintf(fichtm," %d (%.3f),",(int) age, c12);  
                     fprintf(ficgp,"\n# Age %d, p%1d%1d - p%1d%1d",(int) age, k1,l1,k2,l2);  
                     fprintf(ficgp,"\nset label \"%d\" at %11.3e,%11.3e center",(int) age, mu1,mu2);  
                     fprintf(ficgp,"\nreplot %11.3e+ %.3f*(%11.3e*%11.3e*cos(t)+%11.3e*%11.3e*sin(t)), %11.3e +%.3f*(%11.3e*%11.3e*cos(t)+%11.3e*%11.3e*sin(t)) not",\  
                             mu1,std,v11,sqrt(lc1),v12,sqrt(lc2),\  
                             mu2,std,v21,sqrt(lc1),v22,sqrt(lc2));  
                   }/* if first */  
                 } /* age mod 5 */  
               } /* end loop age */  
               fprintf(ficgp,"\nset out \"varpijgr%s%d%1d%1d-%1d%1d.png\";replot;",optionfilefiname, j1,k1,l1,k2,l2);  
               first=1;  
             } /*l12 */  
           } /* k12 */  
         } /*l1 */  
       }/* k1 */  
     } /* loop covariates */  
   }    }
   free_ma3x(varpij,1,nlstate,1,nlstate+ndeath,(int) bage, (int)fage);  
   free_matrix(mu,1,(nlstate+ndeath)*(nlstate+ndeath),(int) bage, (int)fage);  
   free_vector(xp,1,npar);  
   fclose(ficresprob);  
   fclose(ficresprobcov);  
   fclose(ficresprobcor);  
   fclose(ficgp);  
   fclose(fichtm);  
 }  
   
     /*for (i=1; i<=imx; i++){
     for (m=firstpass; (m<lastpass); m++){
        printf("%ld %d %.lf %d %d\n", num[i],(covar[1][i]),agev[m][i],s[m][i],s[m+1][i]);
   }
   
 /******************* Printing html file ***********/  }*/
 void printinghtml(char fileres[], char title[], char datafile[], int firstpass, \  
                   int lastpass, int stepm, int weightopt, char model[],\  
                   int imx,int jmin, int jmax, double jmeanint,char rfileres[],\  
                   int popforecast, int estepm ,\  
                   double jprev1, double mprev1,double anprev1, \  
                   double jprev2, double mprev2,double anprev2){  
   int jj1, k1, i1, cpt;  
   /*char optionfilehtm[FILENAMELENGTH];*/  
   if((fichtm=fopen(optionfilehtm,"a"))==NULL)    {  
     printf("Problem with %s \n",optionfilehtm), exit(0);  
     fprintf(ficlog,"Problem with %s \n",optionfilehtm), exit(0);  
   }  
   
    fprintf(fichtm,"<ul><li><h4>Result files (first order: no variance)</h4>\n  
  - Observed prevalence in each state (during the period defined between %.lf/%.lf/%.lf and %.lf/%.lf/%.lf): <a href=\"p%s\">p%s</a> <br>\n  
  - Estimated transition probabilities over %d (stepm) months: <a href=\"pij%s\">pij%s</a><br>\n  
  - Stable prevalence in each health state: <a href=\"pl%s\">pl%s</a> <br>\n  
  - Life expectancies by age and initial health status (estepm=%2d months):   
    <a href=\"e%s\">e%s</a> <br>\n</li>", \  
   jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,fileres,fileres,stepm,fileres,fileres,fileres,fileres,estepm,fileres,fileres);  
   
 fprintf(fichtm," \n<ul><li><b>Graphs</b></li><p>");  
   
  m=cptcoveff;  
  if (cptcovn < 1) {m=1;ncodemax[1]=1;}  
   
  jj1=0;  
  for(k1=1; k1<=m;k1++){  
    for(i1=1; i1<=ncodemax[k1];i1++){  
      jj1++;  
      if (cptcovn > 0) {  
        fprintf(fichtm,"<hr  size=\"2\" color=\"#EC5E5E\">************ Results for covariates");  
        for (cpt=1; cpt<=cptcoveff;cpt++)   
          fprintf(fichtm," V%d=%d ",Tvaraff[cpt],nbcode[Tvaraff[cpt]][codtab[jj1][cpt]]);  
        fprintf(fichtm," ************\n<hr size=\"2\" color=\"#EC5E5E\">");  
      }  
      /* Pij */  
      fprintf(fichtm,"<br>- Pij or Conditional probabilities to be observed in state j being in state i, %d (stepm) months before: pe%s%d1.png<br>  
 <img src=\"pe%s%d1.png\">",stepm,strtok(optionfile, "."),jj1,strtok(optionfile, "."),jj1);       
      /* Quasi-incidences */  
      fprintf(fichtm,"<br>- Pij or Conditional probabilities to be observed in state j being in state i %d (stepm) months before but expressed in per year i.e. quasi incidences if stepm is small and probabilities too: pe%s%d2.png<br>  
 <img src=\"pe%s%d2.png\">",stepm,strtok(optionfile, "."),jj1,strtok(optionfile, "."),jj1);   
        /* Stable prevalence in each health state */  
        for(cpt=1; cpt<nlstate;cpt++){  
          fprintf(fichtm,"<br>- Stable prevalence in each health state : p%s%d%d.png<br>  
 <img src=\"p%s%d%d.png\">",strtok(optionfile, "."),cpt,jj1,strtok(optionfile, "."),cpt,jj1);  
        }  
      for(cpt=1; cpt<=nlstate;cpt++) {  
         fprintf(fichtm,"\n<br>- Health life expectancies by age and initial health state (%d): exp%s%d%d.png <br>  
 <img src=\"exp%s%d%d.png\">",cpt,strtok(optionfile, "."),cpt,jj1,strtok(optionfile, "."),cpt,jj1);  
      }  
      fprintf(fichtm,"\n<br>- Total life expectancy by age and  
 health expectancies in states (1) and (2): e%s%d.png<br>  
 <img src=\"e%s%d.png\">",strtok(optionfile, "."),jj1,strtok(optionfile, "."),jj1);  
    } /* end i1 */  
  }/* End k1 */  
  fprintf(fichtm,"</ul>");  
   
   
  fprintf(fichtm,"\n<br><li><h4> Result files (second order: variances)</h4>\n  
  - Parameter file with estimated parameters and covariance matrix: <a href=\"%s\">%s</a> <br>\n  
  - Variance of one-step probabilities: <a href=\"prob%s\">prob%s</a> <br>\n  
  - Variance-covariance of one-step probabilities: <a href=\"probcov%s\">probcov%s</a> <br>\n  
  - Correlation matrix of one-step probabilities: <a href=\"probcor%s\">probcor%s</a> <br>\n  
  - Variances and covariances of life expectancies by age and initial health status (estepm=%d months): <a href=\"v%s\">v%s</a><br>\n   
  - Health expectancies with their variances (no covariance): <a href=\"t%s\">t%s</a> <br>\n  
  - Standard deviation of stable prevalences: <a href=\"vpl%s\">vpl%s</a> <br>\n",rfileres,rfileres,fileres,fileres,fileres,fileres,fileres,fileres, estepm, fileres,fileres,fileres,fileres,fileres,fileres);  
   
 /*  if(popforecast==1) fprintf(fichtm,"\n */  
 /*  - Prevalences forecasting: <a href=\"f%s\">f%s</a> <br>\n */  
 /*  - Population forecasting (if popforecast=1): <a href=\"pop%s\">pop%s</a> <br>\n */  
 /*      <br>",fileres,fileres,fileres,fileres); */  
 /*  else  */  
 /*    fprintf(fichtm,"\n No population forecast: popforecast = %d (instead of 1) or stepm = %d (instead of 1) or model=%s (instead of .)<br><br></li>\n",popforecast, stepm, model); */  
 fprintf(fichtm," <ul><li><b>Graphs</b></li><p>");  
   
  m=cptcoveff;    printf("Total number of individuals= %d, Agemin = %.2f, Agemax= %.2f\n\n", imx, *agemin, *agemax);
  if (cptcovn < 1) {m=1;ncodemax[1]=1;}    fprintf(ficlog,"Total number of individuals= %d, Agemin = %.2f, Agemax= %.2f\n\n", imx, *agemin, *agemax); 
   
  jj1=0;    return (0);
  for(k1=1; k1<=m;k1++){   /* endread:*/
    for(i1=1; i1<=ncodemax[k1];i1++){      printf("Exiting calandcheckages: ");
      jj1++;      return (1);
      if (cptcovn > 0) {  
        fprintf(fichtm,"<hr  size=\"2\" color=\"#EC5E5E\">************ Results for covariates");  
        for (cpt=1; cpt<=cptcoveff;cpt++)   
          fprintf(fichtm," V%d=%d ",Tvaraff[cpt],nbcode[Tvaraff[cpt]][codtab[jj1][cpt]]);  
        fprintf(fichtm," ************\n<hr size=\"2\" color=\"#EC5E5E\">");  
      }  
      for(cpt=1; cpt<=nlstate;cpt++) {  
        fprintf(fichtm,"<br>- Observed and period prevalence (with confident  
 interval) in state (%d): v%s%d%d.png <br>  
 <img src=\"v%s%d%d.png\">",cpt,strtok(optionfile, "."),cpt,jj1,strtok(optionfile, "."),cpt,jj1);    
      }  
    } /* end i1 */  
  }/* End k1 */  
  fprintf(fichtm,"</ul>");  
 fclose(fichtm);  
 }  }
   
 /******************* Gnuplot file **************/  #if defined(_MSC_VER)
 void printinggnuplot(char fileres[], double ageminpar, double agemaxpar, double fage , char pathc[], double p[]){  /*printf("Visual C++ compiler: %s \n;", _MSC_FULL_VER);*/
   /*fprintf(ficlog, "Visual C++ compiler: %s \n;", _MSC_FULL_VER);*/
   //#include "stdafx.h"
   //#include <stdio.h>
   //#include <tchar.h>
   //#include <windows.h>
   //#include <iostream>
   typedef BOOL(WINAPI *LPFN_ISWOW64PROCESS) (HANDLE, PBOOL);
   
   int m,cpt,k1,i,k,j,jk,k2,k3,ij,l;  LPFN_ISWOW64PROCESS fnIsWow64Process;
   int ng;  
   if((ficgp=fopen(optionfilegnuplot,"a"))==NULL) {  
     printf("Problem with file %s",optionfilegnuplot);  
     fprintf(ficlog,"Problem with file %s",optionfilegnuplot);  
   }  
   
   /*#ifdef windows */  BOOL IsWow64()
     fprintf(ficgp,"cd \"%s\" \n",pathc);  {
     /*#endif */          BOOL bIsWow64 = FALSE;
 m=pow(2,cptcoveff);  
     
  /* 1eme*/  
   for (cpt=1; cpt<= nlstate ; cpt ++) {  
    for (k1=1; k1<= m ; k1 ++) {  
      fprintf(ficgp,"\nset out \"v%s%d%d.png\" \n",strtok(optionfile, "."),cpt,k1);  
      fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Probability\" \nset ter png small\nset size 0.65,0.65\nplot [%.f:%.f] \"vpl%s\" every :::%d::%d u 1:2 \"\%%lf",ageminpar,fage,fileres,k1-1,k1-1);  
   
      for (i=1; i<= nlstate ; i ++) {  
        if (i==cpt) fprintf(ficgp," \%%lf (\%%lf)");  
        else fprintf(ficgp," \%%*lf (\%%*lf)");  
      }  
      fprintf(ficgp,"\" t\"Stable prevalence\" w l 0,\"vpl%s\" every :::%d::%d u 1:($2+1.96*$3) \"\%%lf",fileres,k1-1,k1-1);  
      for (i=1; i<= nlstate ; i ++) {  
        if (i==cpt) fprintf(ficgp," \%%lf (\%%lf)");  
        else fprintf(ficgp," \%%*lf (\%%*lf)");  
      }   
      fprintf(ficgp,"\" t\"95\%% CI\" w l 1,\"vpl%s\" every :::%d::%d u 1:($2-1.96*$3) \"\%%lf",fileres,k1-1,k1-1);   
      for (i=1; i<= nlstate ; i ++) {  
        if (i==cpt) fprintf(ficgp," \%%lf (\%%lf)");  
        else fprintf(ficgp," \%%*lf (\%%*lf)");  
      }    
      fprintf(ficgp,"\" t\"\" w l 1,\"p%s\" every :::%d::%d u 1:($%d) t\"Observed prevalence \" w l 2",fileres,k1-1,k1-1,2+4*(cpt-1));  
    }  
   }  
   /*2 eme*/  
     
   for (k1=1; k1<= m ; k1 ++) {   
     fprintf(ficgp,"\nset out \"e%s%d.png\" \n",strtok(optionfile, "."),k1);  
     fprintf(ficgp,"set ylabel \"Years\" \nset ter png small\nset size 0.65,0.65\nplot [%.f:%.f] ",ageminpar,fage);  
       
     for (i=1; i<= nlstate+1 ; i ++) {  
       k=2*i;  
       fprintf(ficgp,"\"t%s\" every :::%d::%d u 1:2 \"\%%lf",fileres,k1-1,k1-1);  
       for (j=1; j<= nlstate+1 ; j ++) {  
         if (j==i) fprintf(ficgp," \%%lf (\%%lf)");  
         else fprintf(ficgp," \%%*lf (\%%*lf)");  
       }     
       if (i== 1) fprintf(ficgp,"\" t\"TLE\" w l ,");  
       else fprintf(ficgp,"\" t\"LE in state (%d)\" w l ,",i-1);  
       fprintf(ficgp,"\"t%s\" every :::%d::%d u 1:($2-$3*2) \"\%%lf",fileres,k1-1,k1-1);  
       for (j=1; j<= nlstate+1 ; j ++) {  
         if (j==i) fprintf(ficgp," \%%lf (\%%lf)");  
         else fprintf(ficgp," \%%*lf (\%%*lf)");  
       }     
       fprintf(ficgp,"\" t\"\" w l 0,");  
       fprintf(ficgp,"\"t%s\" every :::%d::%d u 1:($2+$3*2) \"\%%lf",fileres,k1-1,k1-1);  
       for (j=1; j<= nlstate+1 ; j ++) {  
         if (j==i) fprintf(ficgp," \%%lf (\%%lf)");  
         else fprintf(ficgp," \%%*lf (\%%*lf)");  
       }     
       if (i== (nlstate+1)) fprintf(ficgp,"\" t\"\" w l 0");  
       else fprintf(ficgp,"\" t\"\" w l 0,");  
     }  
   }  
     
   /*3eme*/  
     
   for (k1=1; k1<= m ; k1 ++) {   
     for (cpt=1; cpt<= nlstate ; cpt ++) {  
       k=2+nlstate*(2*cpt-2);  
       fprintf(ficgp,"\nset out \"exp%s%d%d.png\" \n",strtok(optionfile, "."),cpt,k1);  
       fprintf(ficgp,"set ter png small\nset size 0.65,0.65\nplot [%.f:%.f] \"e%s\" every :::%d::%d u 1:%d t \"e%d1\" w l",ageminpar,fage,fileres,k1-1,k1-1,k,cpt);  
       /*fprintf(ficgp,",\"e%s\" every :::%d::%d u 1:($%d-2*$%d) \"\%%lf ",fileres,k1-1,k1-1,k,k+1);  
         for (i=1; i<= nlstate*2 ; i ++) fprintf(ficgp,"\%%lf (\%%lf) ");  
         fprintf(ficgp,"\" t \"e%d1\" w l",cpt);  
         fprintf(ficgp,",\"e%s\" every :::%d::%d u 1:($%d+2*$%d) \"\%%lf ",fileres,k1-1,k1-1,k,k+1);  
         for (i=1; i<= nlstate*2 ; i ++) fprintf(ficgp,"\%%lf (\%%lf) ");  
         fprintf(ficgp,"\" t \"e%d1\" w l",cpt);  
           
       */  
       for (i=1; i< nlstate ; i ++) {  
         fprintf(ficgp," ,\"e%s\" every :::%d::%d u 1:%d t \"e%d%d\" w l",fileres,k1-1,k1-1,k+2*i,cpt,i+1);  
           
       }   
     }  
   }  
     
   /* CV preval stable (period) */  
   for (k1=1; k1<= m ; k1 ++) {   
     for (cpt=1; cpt<=nlstate ; cpt ++) {  
       k=3;  
       fprintf(ficgp,"\nset out \"p%s%d%d.png\" \n",strtok(optionfile, "."),cpt,k1);  
       fprintf(ficgp,"set xlabel \"Age\" \nset ylabel \"Probability\" \nset ter png small\nset size 0.65,0.65\nplot [%.f:%.f] \"pij%s\" u ($1==%d ? ($3):1/0):($%d/($%d",ageminpar,agemaxpar,fileres,k1,k+cpt+1,k+1);  
         
       for (i=1; i< nlstate ; i ++)  
         fprintf(ficgp,"+$%d",k+i+1);  
       fprintf(ficgp,")) t\"prev(%d,%d)\" w l",cpt,cpt+1);  
         
       l=3+(nlstate+ndeath)*cpt;  
       fprintf(ficgp,",\"pij%s\" u ($1==%d ? ($3):1/0):($%d/($%d",fileres,k1,l+cpt+1,l+1);  
       for (i=1; i< nlstate ; i ++) {  
         l=3+(nlstate+ndeath)*cpt;  
         fprintf(ficgp,"+$%d",l+i+1);  
       }  
       fprintf(ficgp,")) t\"prev(%d,%d)\" w l\n",cpt+1,cpt+1);     
     }   
   }    
     
   /* proba elementaires */  
   for(i=1,jk=1; i <=nlstate; i++){  
     for(k=1; k <=(nlstate+ndeath); k++){  
       if (k != i) {  
         for(j=1; j <=ncovmodel; j++){  
           fprintf(ficgp,"p%d=%f ",jk,p[jk]);  
           jk++;   
           fprintf(ficgp,"\n");  
         }  
       }  
     }  
    }  
   
    for(ng=1; ng<=2;ng++){ /* Number of graphics: first is probabilities second is incidence per year*/          //typedef BOOL (APIENTRY *LPFN_ISWOW64PROCESS)
      for(jk=1; jk <=m; jk++) {          //  (HANDLE, PBOOL);
        fprintf(ficgp,"\nset out \"pe%s%d%d.png\" \n",strtok(optionfile, "."),jk,ng);   
        if (ng==2)  
          fprintf(ficgp,"\nset ylabel \"Quasi-incidence per year\"\n");  
        else  
          fprintf(ficgp,"\nset title \"Probability\"\n");  
        fprintf(ficgp,"\nset ter png small\nset size 0.65,0.65\nset log y\nplot  [%.f:%.f] ",ageminpar,agemaxpar);  
        i=1;  
        for(k2=1; k2<=nlstate; k2++) {  
          k3=i;  
          for(k=1; k<=(nlstate+ndeath); k++) {  
            if (k != k2){  
              if(ng==2)  
                fprintf(ficgp," %f*exp(p%d+p%d*x",YEARM/stepm,i,i+1);  
              else  
                fprintf(ficgp," exp(p%d+p%d*x",i,i+1);  
              ij=1;  
              for(j=3; j <=ncovmodel; j++) {  
                if(((j-2)==Tage[ij]) &&(ij <=cptcovage)) {  
                  fprintf(ficgp,"+p%d*%d*x",i+j-1,nbcode[Tvar[j-2]][codtab[jk][Tvar[j-2]]]);  
                  ij++;  
                }  
                else  
                  fprintf(ficgp,"+p%d*%d",i+j-1,nbcode[Tvar[j-2]][codtab[jk][j-2]]);  
              }  
              fprintf(ficgp,")/(1");  
                
              for(k1=1; k1 <=nlstate; k1++){     
                fprintf(ficgp,"+exp(p%d+p%d*x",k3+(k1-1)*ncovmodel,k3+(k1-1)*ncovmodel+1);  
                ij=1;  
                for(j=3; j <=ncovmodel; j++){  
                  if(((j-2)==Tage[ij]) &&(ij <=cptcovage)) {  
                    fprintf(ficgp,"+p%d*%d*x",k3+(k1-1)*ncovmodel+1+j-2,nbcode[Tvar[j-2]][codtab[jk][Tvar[j-2]]]);  
                    ij++;  
                  }  
                  else  
                    fprintf(ficgp,"+p%d*%d",k3+(k1-1)*ncovmodel+1+j-2,nbcode[Tvar[j-2]][codtab[jk][j-2]]);  
                }  
                fprintf(ficgp,")");  
              }  
              fprintf(ficgp,") t \"p%d%d\" ", k2,k);  
              if ((k+k2)!= (nlstate*2+ndeath)) fprintf(ficgp,",");  
              i=i+ncovmodel;  
            }  
          } /* end k */  
        } /* end k2 */  
      } /* end jk */  
    } /* end ng */  
    fclose(ficgp);   
 }  /* end gnuplot */  
   
           //LPFN_ISWOW64PROCESS fnIsWow64Process;
   
 /*************** Moving average **************/          HMODULE module = GetModuleHandle(_T("kernel32"));
 int movingaverage(double ***probs, double bage,double fage, double ***mobaverage, int mobilav){          const char funcName[] = "IsWow64Process";
           fnIsWow64Process = (LPFN_ISWOW64PROCESS)
                   GetProcAddress(module, funcName);
   
   int i, cpt, cptcod;          if (NULL != fnIsWow64Process)
   int modcovmax =1;          {
   int mobilavrange, mob;                  if (!fnIsWow64Process(GetCurrentProcess(),
   double age;                          &bIsWow64))
                           //throw std::exception("Unknown error");
   modcovmax=2*cptcoveff;/* Max number of modalities. We suppose                           printf("Unknown error\n");
                            a covariate has 2 modalities */  
   if (cptcovn<1) modcovmax=1; /* At least 1 pass */  
   
   if(mobilav==1||mobilav ==3 ||mobilav==5 ||mobilav== 7){  
     if(mobilav==1) mobilavrange=5; /* default */  
     else mobilavrange=mobilav;  
     for (age=bage; age<=fage; age++)  
       for (i=1; i<=nlstate;i++)  
         for (cptcod=1;cptcod<=modcovmax;cptcod++)  
           mobaverage[(int)age][i][cptcod]=probs[(int)age][i][cptcod];  
     /* We keep the original values on the extreme ages bage, fage and for   
        fage+1 and bage-1 we use a 3 terms moving average; for fage+2 bage+2  
        we use a 5 terms etc. until the borders are no more concerned.   
     */   
     for (mob=3;mob <=mobilavrange;mob=mob+2){  
       for (age=bage+(mob-1)/2; age<=fage-(mob-1)/2; age++){  
         for (i=1; i<=nlstate;i++){  
           for (cptcod=1;cptcod<=modcovmax;cptcod++){  
             mobaverage[(int)age][i][cptcod] =probs[(int)age][i][cptcod];  
               for (cpt=1;cpt<=(mob-1)/2;cpt++){  
                 mobaverage[(int)age][i][cptcod] +=probs[(int)age-cpt][i][cptcod];  
                 mobaverage[(int)age][i][cptcod] +=probs[(int)age+cpt][i][cptcod];  
               }  
             mobaverage[(int)age][i][cptcod]=mobaverage[(int)age][i][cptcod]/mob;  
           }  
         }          }
       }/* end age */          return bIsWow64 != FALSE;
     }/* end mob */  }
   }else return -1;  #endif
   return 0;  
 }/* End movingaverage */  
   
   void syscompilerinfo(int logged)
   {
   #include <stdint.h>
   
 /************** Forecasting ******************/    /* #include "syscompilerinfo.h"*/
 prevforecast(char fileres[], double anproj1, double mproj1, double jproj1, double ageminpar, double agemax, double dateprev1, double dateprev2, int mobilav, double bage, double fage, int firstpass, int lastpass, double anproj2, double p[], int cptcoveff){     /* command line Intel compiler 32bit windows, XP compatible:*/
   /* proj1, year, month, day of starting projection      /* /GS /W3 /Gy
      agemin, agemax range of age        /Zc:wchar_t /Zi /O2 /Fd"Release\vc120.pdb" /D "WIN32" /D "NDEBUG" /D
      dateprev1 dateprev2 range of dates during which prevalence is computed        "_CONSOLE" /D "_LIB" /D "_USING_V110_SDK71_" /D "_UNICODE" /D
      anproj2 year of en of projection (same day and month as proj1).        "UNICODE" /Qipo /Zc:forScope /Gd /Oi /MT /Fa"Release\" /EHsc /nologo
   */        /Fo"Release\" /Qprof-dir "Release\" /Fp"Release\IMaCh.pch"
   int yearp, stepsize, hstepm, nhstepm, j, k, c, cptcod, i, h, i1;     */ 
   int *popage;     /* 64 bits */
   double agec; /* generic age */     /*
   double agelim, ppij, yp,yp1,yp2,jprojmean,mprojmean,anprojmean;       /GS /W3 /Gy
   double *popeffectif,*popcount;       /Zc:wchar_t /Zi /O2 /Fd"x64\Release\vc120.pdb" /D "WIN32" /D "NDEBUG"
   double ***p3mat;       /D "_CONSOLE" /D "_LIB" /D "_UNICODE" /D "UNICODE" /Qipo /Zc:forScope
   double ***mobaverage;       /Oi /MD /Fa"x64\Release\" /EHsc /nologo /Fo"x64\Release\" /Qprof-dir
   char fileresf[FILENAMELENGTH];       "x64\Release\" /Fp"x64\Release\IMaCh.pch" */
      /* Optimization are useless and O3 is slower than O2 */
      /*
        /GS /W3 /Gy /Zc:wchar_t /Zi /O3 /Fd"x64\Release\vc120.pdb" /D "WIN32" 
        /D "NDEBUG" /D "_CONSOLE" /D "_LIB" /D "_UNICODE" /D "UNICODE" /Qipo 
        /Zc:forScope /Oi /MD /Fa"x64\Release\" /EHsc /nologo /Qparallel 
        /Fo"x64\Release\" /Qprof-dir "x64\Release\" /Fp"x64\Release\IMaCh.pch" 
      */
      /* Link is */ /* /OUT:"visual studio
         2013\Projects\IMaCh\Release\IMaCh.exe" /MANIFEST /NXCOMPAT
         /PDB:"visual studio
         2013\Projects\IMaCh\Release\IMaCh.pdb" /DYNAMICBASE
         "kernel32.lib" "user32.lib" "gdi32.lib" "winspool.lib"
         "comdlg32.lib" "advapi32.lib" "shell32.lib" "ole32.lib"
         "oleaut32.lib" "uuid.lib" "odbc32.lib" "odbccp32.lib"
         /MACHINE:X86 /OPT:REF /SAFESEH /INCREMENTAL:NO
         /SUBSYSTEM:CONSOLE",5.01" /MANIFESTUAC:"level='asInvoker'
         uiAccess='false'"
         /ManifestFile:"Release\IMaCh.exe.intermediate.manifest" /OPT:ICF
         /NOLOGO /TLBID:1
      */
   
   agelim=AGESUP;  
   prevalence(probs, ageminpar, agemax, s, agev, nlstate, imx, Tvar, nbcode, ncodemax, mint, anint, dateprev1, dateprev2, firstpass, lastpass);  
    
   strcpy(fileresf,"f");   
   strcat(fileresf,fileres);  
   if((ficresf=fopen(fileresf,"w"))==NULL) {  
     printf("Problem with forecast resultfile: %s\n", fileresf);  
     fprintf(ficlog,"Problem with forecast resultfile: %s\n", fileresf);  
   }  
   printf("Computing forecasting: result on file '%s' \n", fileresf);  
   fprintf(ficlog,"Computing forecasting: result on file '%s' \n", fileresf);  
   
   if (cptcoveff==0) ncodemax[cptcoveff]=1;  #if defined __INTEL_COMPILER
   #if defined(__GNUC__)
           struct utsname sysInfo;  /* For Intel on Linux and OS/X */
   #endif
   #elif defined(__GNUC__) 
   #ifndef  __APPLE__
   #include <gnu/libc-version.h>  /* Only on gnu */
   #endif
      struct utsname sysInfo;
      int cross = CROSS;
      if (cross){
              printf("Cross-");
              if(logged) fprintf(ficlog, "Cross-");
      }
   #endif
   
   if (mobilav!=0) {     printf("Compiled with:");if(logged)fprintf(ficlog,"Compiled with:");
     mobaverage= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX);  #if defined(__clang__)
     if (movingaverage(probs, ageminpar, fage, mobaverage,mobilav)!=0){     printf(" Clang/LLVM");if(logged)fprintf(ficlog," Clang/LLVM");       /* Clang/LLVM. ---------------------------------------------- */
       fprintf(ficlog," Error in movingaverage mobilav=%d\n",mobilav);  #endif
       printf(" Error in movingaverage mobilav=%d\n",mobilav);  #if defined(__ICC) || defined(__INTEL_COMPILER)
     }     printf(" Intel ICC/ICPC");if(logged)fprintf(ficlog," Intel ICC/ICPC");/* Intel ICC/ICPC. ------------------------------------------ */
   }  #endif
   #if defined(__GNUC__) || defined(__GNUG__)
      printf(" GNU GCC/G++");if(logged)fprintf(ficlog," GNU GCC/G++");/* GNU GCC/G++. --------------------------------------------- */
   #endif
   #if defined(__HP_cc) || defined(__HP_aCC)
      printf(" Hewlett-Packard C/aC++");if(logged)fprintf(fcilog," Hewlett-Packard C/aC++"); /* Hewlett-Packard C/aC++. ---------------------------------- */
   #endif
   #if defined(__IBMC__) || defined(__IBMCPP__)
      printf(" IBM XL C/C++"); if(logged) fprintf(ficlog," IBM XL C/C++");/* IBM XL C/C++. -------------------------------------------- */
   #endif
   #if defined(_MSC_VER)
      printf(" Microsoft Visual Studio");if(logged)fprintf(ficlog," Microsoft Visual Studio");/* Microsoft Visual Studio. --------------------------------- */
   #endif
   #if defined(__PGI)
      printf(" Portland Group PGCC/PGCPP");if(logged) fprintf(ficlog," Portland Group PGCC/PGCPP");/* Portland Group PGCC/PGCPP. ------------------------------- */
   #endif
   #if defined(__SUNPRO_C) || defined(__SUNPRO_CC)
      printf(" Oracle Solaris Studio");if(logged)fprintf(ficlog," Oracle Solaris Studio\n");/* Oracle Solaris Studio. ----------------------------------- */
   #endif
      printf(" for "); if (logged) fprintf(ficlog, " for ");
      
   // http://stackoverflow.com/questions/4605842/how-to-identify-platform-compiler-from-preprocessor-macros
   #ifdef _WIN32 // note the underscore: without it, it's not msdn official!
       // Windows (x64 and x86)
      printf("Windows (x64 and x86) ");if(logged) fprintf(ficlog,"Windows (x64 and x86) ");
   #elif __unix__ // all unices, not all compilers
       // Unix
      printf("Unix ");if(logged) fprintf(ficlog,"Unix ");
   #elif __linux__
       // linux
      printf("linux ");if(logged) fprintf(ficlog,"linux ");
   #elif __APPLE__
       // Mac OS, not sure if this is covered by __posix__ and/or __unix__ though..
      printf("Mac OS ");if(logged) fprintf(ficlog,"Mac OS ");
   #endif
   
   stepsize=(int) (stepm+YEARM-1)/YEARM;  /*  __MINGW32__   */
   if (stepm<=12) stepsize=1;  /*  __CYGWIN__   */
   if(estepm < stepm){  /* __MINGW64__  */
     printf ("Problem %d lower than %d\n",estepm, stepm);  // http://msdn.microsoft.com/en-us/library/b0084kay.aspx
   /* _MSC_VER  //the Visual C++ compiler is 17.00.51106.1, the _MSC_VER macro evaluates to 1700. Type cl /?  */
   /* _MSC_FULL_VER //the Visual C++ compiler is 15.00.20706.01, the _MSC_FULL_VER macro evaluates to 150020706 */
   /* _WIN64  // Defined for applications for Win64. */
   /* _M_X64 // Defined for compilations that target x64 processors. */
   /* _DEBUG // Defined when you compile with /LDd, /MDd, and /MTd. */
   
   #if UINTPTR_MAX == 0xffffffff
      printf(" 32-bit"); if(logged) fprintf(ficlog," 32-bit");/* 32-bit */
   #elif UINTPTR_MAX == 0xffffffffffffffff
      printf(" 64-bit"); if(logged) fprintf(ficlog," 64-bit");/* 64-bit */
   #else
      printf(" wtf-bit"); if(logged) fprintf(ficlog," wtf-bit");/* wtf */
   #endif
   
   #if defined(__GNUC__)
   # if defined(__GNUC_PATCHLEVEL__)
   #  define __GNUC_VERSION__ (__GNUC__ * 10000 \
                               + __GNUC_MINOR__ * 100 \
                               + __GNUC_PATCHLEVEL__)
   # else
   #  define __GNUC_VERSION__ (__GNUC__ * 10000 \
                               + __GNUC_MINOR__ * 100)
   # endif
      printf(" using GNU C version %d.\n", __GNUC_VERSION__);
      if(logged) fprintf(ficlog, " using GNU C version %d.\n", __GNUC_VERSION__);
   
      if (uname(&sysInfo) != -1) {
        printf("Running on: %s %s %s %s %s\n",sysInfo.sysname, sysInfo.nodename, sysInfo.release, sysInfo.version, sysInfo.machine);
            if(logged) fprintf(ficlog,"Running on: %s %s %s %s %s\n ",sysInfo.sysname, sysInfo.nodename, sysInfo.release, sysInfo.version, sysInfo.machine);
      }
      else
         perror("uname() error");
      //#ifndef __INTEL_COMPILER 
   #if !defined (__INTEL_COMPILER) && !defined(__APPLE__)
      printf("GNU libc version: %s\n", gnu_get_libc_version()); 
      if(logged) fprintf(ficlog,"GNU libc version: %s\n", gnu_get_libc_version());
   #endif
   #endif
   
      //   void main ()
      //   {
   #if defined(_MSC_VER)
      if (IsWow64()){
              printf("\nThe program (probably compiled for 32bit) is running under WOW64 (64bit) emulation.\n");
              if (logged) fprintf(ficlog, "\nThe program (probably compiled for 32bit) is running under WOW64 (64bit) emulation.\n");
      }
      else{
              printf("\nThe program is not running under WOW64 (i.e probably on a 64bit Windows).\n");
              if (logged) fprintf(ficlog, "\nThe programm is not running under WOW64 (i.e probably on a 64bit Windows).\n");
      }
      //      printf("\nPress Enter to continue...");
      //      getchar();
      //   }
   
   #endif
      
   
   }
   
   int prevalence_limit(double *p, double **prlim, double ageminpar, double agemaxpar, double ftolpl, int *ncvyearp){
     /*--------------- Prevalence limit  (forward period or forward stable prevalence) --------------*/
     /* Computes the prevalence limit for each combination of the dummy covariates */
     int i, j, k, i1, k4=0, nres=0 ;
     /* double ftolpl = 1.e-10; */
     double age, agebase, agelim;
     double tot;
   
     strcpy(filerespl,"PL_");
     strcat(filerespl,fileresu);
     if((ficrespl=fopen(filerespl,"w"))==NULL) {
       printf("Problem with forward period (stable) prevalence resultfile: %s\n", filerespl);return 1;
       fprintf(ficlog,"Problem with forward period (stable) prevalence resultfile: %s\n", filerespl);return 1;
   }    }
   else  hstepm=estepm;       printf("\nComputing forward period (stable) prevalence: result on file '%s' \n", filerespl);
     fprintf(ficlog,"\nComputing forward period (stable) prevalence: result on file '%s' \n", filerespl);
     pstamp(ficrespl);
     fprintf(ficrespl,"# Forward period (stable) prevalence. Precision given by ftolpl=%g \n", ftolpl);
     fprintf(ficrespl,"#Age ");
     for(i=1; i<=nlstate;i++) fprintf(ficrespl,"%d-%d ",i,i);
     fprintf(ficrespl,"\n");
     
     /* prlim=matrix(1,nlstate,1,nlstate);*/ /* back in main */
   
   hstepm=hstepm/stepm;     agebase=ageminpar;
   yp1=modf(dateintmean,&yp);/* extracts integral of datemean in yp  and    agelim=agemaxpar;
                                fractional in yp1 */  
   anprojmean=yp;  
   yp2=modf((yp1*12),&yp);  
   mprojmean=yp;  
   yp1=modf((yp2*30.5),&yp);  
   jprojmean=yp;  
   if(jprojmean==0) jprojmean=1;  
   if(mprojmean==0) jprojmean=1;  
   
   i1=cptcoveff;    /* i1=pow(2,ncoveff); */
     i1=pow(2,cptcoveff); /* Number of combination of dummy covariates */
   if (cptcovn < 1){i1=1;}    if (cptcovn < 1){i1=1;}
     
   fprintf(ficresf,"# Mean day of interviews %.lf/%.lf/%.lf (%.2f) between %.2f and %.2f \n",jprojmean,mprojmean,anprojmean,dateintmean,dateprev1,dateprev2);   
     
   fprintf(ficresf,"#****** Routine prevforecast **\n");  
   
 /*            if (h==(int)(YEARM*yearp)){ */    for(k=1; k<=i1;k++){ /* For each combination k of dummy covariates in the model */
   for(cptcov=1, k=0;cptcov<=i1;cptcov++){      for(nres=1; nres <= nresult; nres++){ /* For each resultline */
     for(cptcod=1;cptcod<=ncodemax[cptcoveff];cptcod++){        if(i1 != 1 && TKresult[nres]!= k)
       k=k+1;          continue;
       fprintf(ficresf,"\n#******");  
       for(j=1;j<=cptcoveff;j++) {        /* for(cptcov=1,k=0;cptcov<=i1;cptcov++){ */
         fprintf(ficresf," V%d=%d, hpijx=probability over h years, hp.jx is weighted by observed prev ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);        /* for(cptcov=1,k=0;cptcov<=1;cptcov++){ */
         //for(cptcod=1;cptcod<=ncodemax[cptcov];cptcod++){
         /* k=k+1; */
         /* to clean */
         /*printf("cptcov=%d cptcod=%d codtab=%d\n",cptcov, cptcod,codtabm(cptcod,cptcov));*/
         fprintf(ficrespl,"#******");
         printf("#******");
         fprintf(ficlog,"#******");
         for(j=1;j<=cptcoveff ;j++) {/* all covariates */
           /* fprintf(ficrespl," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,Tvaraff[j])]); /\* Here problem for varying dummy*\/ */
           fprintf(ficrespl," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]); /* Here problem for varying dummy*/
           printf(" V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
           fprintf(ficlog," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
         }
         for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
           printf(" V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
           fprintf(ficrespl," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
           fprintf(ficlog," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
       }        }
       fprintf(ficresf,"******\n");        fprintf(ficrespl,"******\n");
       fprintf(ficresf,"# Covariate valuofcovar yearproj age");        printf("******\n");
       for(j=1; j<=nlstate+ndeath;j++){         fprintf(ficlog,"******\n");
         for(i=1; i<=nlstate;i++)                      if(invalidvarcomb[k]){
           fprintf(ficresf," p%d%d",i,j);          printf("\nCombination (%d) ignored because no case \n",k); 
         fprintf(ficresf," p.%d",j);          fprintf(ficrespl,"#Combination (%d) ignored because no case \n",k); 
           fprintf(ficlog,"\nCombination (%d) ignored because no case \n",k); 
           continue;
       }        }
       for (yearp=0; yearp<=(anproj2-anproj1);yearp +=stepsize) {   
         fprintf(ficresf,"\n");  
         fprintf(ficresf,"\n# Forecasting at date %.lf/%.lf/%.lf ",jproj1,mproj1,anproj1+yearp);     
   
         for (agec=fage; agec>=(ageminpar-1); agec--){   
           nhstepm=(int) rint((agelim-agec)*YEARM/stepm);   
           nhstepm = nhstepm/hstepm;   
           p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);  
           oldm=oldms;savm=savms;  
           hpxij(p3mat,nhstepm,agec,hstepm,p,nlstate,stepm,oldm,savm, k);    
           
           for (h=0; h<=nhstepm; h++){  
             if (h*hstepm/YEARM*stepm ==yearp) {  
               fprintf(ficresf,"\n");  
               for(j=1;j<=cptcoveff;j++)   
                 fprintf(ficresf,"%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);  
               fprintf(ficresf,"%.f %.f ",anproj1+yearp,agec+h*hstepm/YEARM*stepm);  
             }   
             for(j=1; j<=nlstate+ndeath;j++) {  
               ppij=0.;  
               for(i=1; i<=nlstate;i++) {  
                 if (mobilav==1)   
                   ppij=ppij+p3mat[i][j][h]*mobaverage[(int)agec][i][cptcod];  
                 else {  
                   ppij=ppij+p3mat[i][j][h]*probs[(int)(agec)][i][cptcod];  
                 }  
                 if (h*hstepm/YEARM*stepm== yearp) {  
                   fprintf(ficresf," %.3f", p3mat[i][j][h]);  
                 }  
               } /* end i */  
               if (h*hstepm/YEARM*stepm==yearp) {  
                 fprintf(ficresf," %.3f", ppij);  
               }  
             }/* end j */  
           } /* end h */  
           free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);  
         } /* end agec */  
       } /* end yearp */  
     } /* end cptcod */  
   } /* end  cptcov */  
          
   if (mobilav!=0) free_ma3x(mobaverage,1, AGESUP,1,NCOVMAX, 1,NCOVMAX);  
   
   fclose(ficresf);        fprintf(ficrespl,"#Age ");
         for(j=1;j<=cptcoveff;j++) {
           fprintf(ficrespl,"V%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
         }
         for(i=1; i<=nlstate;i++) fprintf(ficrespl,"  %d-%d   ",i,i);
         fprintf(ficrespl,"Total Years_to_converge\n");
       
         for (age=agebase; age<=agelim; age++){
           /* for (age=agebase; age<=agebase; age++){ */
           prevalim(prlim, nlstate, p, age, oldm, savm, ftolpl, ncvyearp, k, nres);
           fprintf(ficrespl,"%.0f ",age );
           for(j=1;j<=cptcoveff;j++)
             fprintf(ficrespl,"%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
           tot=0.;
           for(i=1; i<=nlstate;i++){
             tot +=  prlim[i][i];
             fprintf(ficrespl," %.5f", prlim[i][i]);
           }
           fprintf(ficrespl," %.3f %d\n", tot, *ncvyearp);
         } /* Age */
         /* was end of cptcod */
       } /* cptcov */
     } /* nres */
     return 0;
 }  }
   
 /************** Forecasting *****not tested NB*************/  int back_prevalence_limit(double *p, double **bprlim, double ageminpar, double agemaxpar, double ftolpl, int *ncvyearp, double dateprev1,double dateprev2, int firstpass, int lastpass, int mobilavproj){
 populforecast(char fileres[], double anpyram,double mpyram,double jpyram,double ageminpar, double agemax,double dateprev1, double dateprev2, int mobilav, double agedeb, double fage, int popforecast, char popfile[], double anpyram1,double p[], int i2){          /*--------------- Back Prevalence limit  (backward stable prevalence) --------------*/
             
   int cpt, stepsize, hstepm, nhstepm, j,k,c, cptcod, i,h;          /* Computes the back prevalence limit  for any combination      of covariate values 
   int *popage;     * at any age between ageminpar and agemaxpar
   double calagedatem, agelim, kk1, kk2;           */
   double *popeffectif,*popcount;    int i, j, k, i1, nres=0 ;
   double ***p3mat,***tabpop,***tabpopprev;    /* double ftolpl = 1.e-10; */
   double ***mobaverage;    double age, agebase, agelim;
   char filerespop[FILENAMELENGTH];    double tot;
     /* double ***mobaverage; */
   tabpop= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX);    /* double      **dnewm, **doldm, **dsavm;  /\* for use *\/ */
   tabpopprev= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX);  
   agelim=AGESUP;    strcpy(fileresplb,"PLB_");
   calagedatem=(anpyram+mpyram/12.+jpyram/365.-dateintmean)*YEARM;    strcat(fileresplb,fileresu);
     if((ficresplb=fopen(fileresplb,"w"))==NULL) {
       printf("Problem with backward prevalence resultfile: %s\n", fileresplb);return 1;
       fprintf(ficlog,"Problem with backward prevalence resultfile: %s\n", fileresplb);return 1;
     }
     printf("Computing backward prevalence: result on file '%s' \n", fileresplb);
     fprintf(ficlog,"Computing backward prevalence: result on file '%s' \n", fileresplb);
     pstamp(ficresplb);
     fprintf(ficresplb,"# Backward prevalence. Precision given by ftolpl=%g \n", ftolpl);
     fprintf(ficresplb,"#Age ");
     for(i=1; i<=nlstate;i++) fprintf(ficresplb,"%d-%d ",i,i);
     fprintf(ficresplb,"\n");
       
   prevalence(probs, ageminpar, agemax, s, agev, nlstate, imx, Tvar, nbcode, ncodemax, mint, anint, dateprev1, dateprev2, firstpass, lastpass);  
       
     /* prlim=matrix(1,nlstate,1,nlstate);*/ /* back in main */
       
   strcpy(filerespop,"pop");     agebase=ageminpar;
   strcat(filerespop,fileres);    agelim=agemaxpar;
   if((ficrespop=fopen(filerespop,"w"))==NULL) {  
     printf("Problem with forecast resultfile: %s\n", filerespop);  
     fprintf(ficlog,"Problem with forecast resultfile: %s\n", filerespop);  
   }  
   printf("Computing forecasting: result on file '%s' \n", filerespop);  
   fprintf(ficlog,"Computing forecasting: result on file '%s' \n", filerespop);  
   
   if (cptcoveff==0) ncodemax[cptcoveff]=1;  
   
   if (mobilav!=0) {  
     mobaverage= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX);  
     if (movingaverage(probs, ageminpar, fage, mobaverage,mobilav)!=0){  
       fprintf(ficlog," Error in movingaverage mobilav=%d\n",mobilav);  
       printf(" Error in movingaverage mobilav=%d\n",mobilav);  
     }  
   }  
   
   stepsize=(int) (stepm+YEARM-1)/YEARM;  
   if (stepm<=12) stepsize=1;  
       
   agelim=AGESUP;  
       
   hstepm=1;    i1=pow(2,cptcoveff);
   hstepm=hstepm/stepm;     if (cptcovn < 1){i1=1;}
       
   if (popforecast==1) {    for(nres=1; nres <= nresult; nres++){ /* For each resultline */
     if((ficpop=fopen(popfile,"r"))==NULL) {      for(k=1; k<=i1;k++){ /* For any combination of dummy covariates, fixed and varying */
       printf("Problem with population file : %s\n",popfile);exit(0);       if(i1 != 1 && TKresult[nres]!= k)
       fprintf(ficlog,"Problem with population file : %s\n",popfile);exit(0);          continue;
     }        /*printf("cptcov=%d cptcod=%d codtab=%d\n",cptcov, cptcod,codtabm(cptcod,cptcov));*/
     popage=ivector(0,AGESUP);        fprintf(ficresplb,"#******");
     popeffectif=vector(0,AGESUP);        printf("#******");
     popcount=vector(0,AGESUP);        fprintf(ficlog,"#******");
         for(j=1;j<=cptcoveff ;j++) {/* all covariates */
           fprintf(ficresplb," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
           printf(" V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
           fprintf(ficlog," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
         }
         for (j=1; j<= nsq; j++){ /* For each selected (single) quantitative value */
           printf(" V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
           fprintf(ficresplb," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
           fprintf(ficlog," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
         }
         fprintf(ficresplb,"******\n");
         printf("******\n");
         fprintf(ficlog,"******\n");
         if(invalidvarcomb[k]){
           printf("\nCombination (%d) ignored because no cases \n",k); 
           fprintf(ficresplb,"#Combination (%d) ignored because no cases \n",k); 
           fprintf(ficlog,"\nCombination (%d) ignored because no cases \n",k); 
           continue;
         }
           
     i=1;           fprintf(ficresplb,"#Age ");
     while ((c=fscanf(ficpop,"%d %lf\n",&popage[i],&popcount[i])) != EOF) i=i+1;  
      
     imx=i;  
     for (i=1; i<imx;i++) popeffectif[popage[i]]=popcount[i];  
   }  
   
   for(cptcov=1,k=0;cptcov<=i2;cptcov++){  
    for(cptcod=1;cptcod<=ncodemax[cptcoveff];cptcod++){  
       k=k+1;  
       fprintf(ficrespop,"\n#******");  
       for(j=1;j<=cptcoveff;j++) {        for(j=1;j<=cptcoveff;j++) {
         fprintf(ficrespop," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);          fprintf(ficresplb,"V%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
         }
         for(i=1; i<=nlstate;i++) fprintf(ficresplb,"  %d-%d   ",i,i);
         fprintf(ficresplb,"Total Years_to_converge\n");
       
       
         for (age=agebase; age<=agelim; age++){
           /* for (age=agebase; age<=agebase; age++){ */
           if(mobilavproj > 0){
             /* bprevalim(bprlim, mobaverage, nlstate, p, age, ageminpar, agemaxpar, oldm, savm, doldm, dsavm, ftolpl, ncvyearp, k); */
             /* bprevalim(bprlim, mobaverage, nlstate, p, age, oldm, savm, dnewm, doldm, dsavm, ftolpl, ncvyearp, k); */
             bprevalim(bprlim, mobaverage, nlstate, p, age, ftolpl, ncvyearp, k, nres);
           }else if (mobilavproj == 0){
             printf("There is no chance to get back prevalence limit if data aren't non zero and summing to 1, please try a non null mobil_average(=%d) parameter or mobil_average=-1 if you want to try at your own risk.\n",mobilavproj);
             fprintf(ficlog,"There is no chance to get back prevalence limit if data aren't non zero and summing to 1, please try a non null mobil_average(=%d) parameter or mobil_average=-1 if you want to try at your own risk.\n",mobilavproj);
             exit(1);
           }else{
             /* bprevalim(bprlim, probs, nlstate, p, age, oldm, savm, dnewm, doldm, dsavm, ftolpl, ncvyearp, k); */
             bprevalim(bprlim, probs, nlstate, p, age, ftolpl, ncvyearp, k,nres);
             /* printf("TOTOT\n"); */
             /* exit(1); */
           }
           fprintf(ficresplb,"%.0f ",age );
           for(j=1;j<=cptcoveff;j++)
             fprintf(ficresplb,"%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
           tot=0.;
           for(i=1; i<=nlstate;i++){
             tot +=  bprlim[i][i];
             fprintf(ficresplb," %.5f", bprlim[i][i]);
           }
           fprintf(ficresplb," %.3f %d\n", tot, *ncvyearp);
         } /* Age */
         /* was end of cptcod */
         /*fprintf(ficresplb,"\n");*/ /* Seems to be necessary for gnuplot only if two result lines and no covariate. */
       } /* end of any combination */
     } /* end of nres */  
     /* hBijx(p, bage, fage); */
     /* fclose(ficrespijb); */
     
     return 0;
   }
    
   int hPijx(double *p, int bage, int fage){
       /*------------- h Pij x at various ages ------------*/
   
     int stepsize;
     int agelim;
     int hstepm;
     int nhstepm;
     int h, i, i1, j, k, k4, nres=0;
   
     double agedeb;
     double ***p3mat;
   
       strcpy(filerespij,"PIJ_");  strcat(filerespij,fileresu);
       if((ficrespij=fopen(filerespij,"w"))==NULL) {
         printf("Problem with Pij resultfile: %s\n", filerespij); return 1;
         fprintf(ficlog,"Problem with Pij resultfile: %s\n", filerespij); return 1;
       }
       printf("Computing pij: result on file '%s' \n", filerespij);
       fprintf(ficlog,"Computing pij: result on file '%s' \n", filerespij);
     
       stepsize=(int) (stepm+YEARM-1)/YEARM;
       /*if (stepm<=24) stepsize=2;*/
   
       agelim=AGESUP;
       hstepm=stepsize*YEARM; /* Every year of age */
       hstepm=hstepm/stepm; /* Typically 2 years, = 2/6 months = 4 */ 
                   
       /* hstepm=1;   aff par mois*/
       pstamp(ficrespij);
       fprintf(ficrespij,"#****** h Pij x Probability to be in state j at age x+h being in i at x ");
       i1= pow(2,cptcoveff);
                   /* for(cptcov=1,k=0;cptcov<=i1;cptcov++){ */
                   /*    /\*for(cptcod=1;cptcod<=ncodemax[cptcov];cptcod++){*\/ */
                   /*      k=k+1;  */
       for(nres=1; nres <= nresult; nres++) /* For each resultline */
       for(k=1; k<=i1;k++){
         if(i1 != 1 && TKresult[nres]!= k)
           continue;
         fprintf(ficrespij,"\n#****** ");
         for(j=1;j<=cptcoveff;j++) 
           fprintf(ficrespij,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
         for (k4=1; k4<= nsq; k4++){ /* For each selected (single) quantitative value */
           printf(" V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
           fprintf(ficrespij," V%d=%f ",Tvqresult[nres][k4],Tqresult[nres][k4]);
       }        }
       fprintf(ficrespop,"******\n");        fprintf(ficrespij,"******\n");
       fprintf(ficrespop,"# Age");  
       for(j=1; j<=nlstate+ndeath;j++) fprintf(ficrespop," P.%d",j);  
       if (popforecast==1)  fprintf(ficrespop," [Population]");  
               
       for (cpt=0; cpt<=0;cpt++) {         for (agedeb=fage; agedeb>=bage; agedeb--){ /* If stepm=6 months */
         fprintf(ficrespop,"\n\n# Forecasting at date %.lf/%.lf/%.lf ",jpyram,mpyram,anpyram+cpt);             nhstepm=(int) rint((agelim-agedeb)*YEARM/stepm); /* Typically 20 years = 20*12/6=40 */ 
           nhstepm = nhstepm/hstepm; /* Typically 40/4=10 */
                   
         for (agedeb=(fage-((int)calagedatem %12/12.)); agedeb>=(ageminpar-((int)calagedatem %12)/12.); agedeb--){           /*        nhstepm=nhstepm*YEARM; aff par mois*/
           nhstepm=(int) rint((agelim-agedeb)*YEARM/stepm);   
           nhstepm = nhstepm/hstepm;   
             
           p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);  
           oldm=oldms;savm=savms;  
           hpxij(p3mat,nhstepm,agedeb,hstepm,p,nlstate,stepm,oldm,savm, k);    
                   
           for (h=0; h<=nhstepm; h++){          p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
             if (h==(int) (calagedatem+YEARM*cpt)) {          oldm=oldms;savm=savms;
               fprintf(ficrespop,"\n %3.f ",agedeb+h*hstepm/YEARM*stepm);          hpxij(p3mat,nhstepm,agedeb,hstepm,p,nlstate,stepm,oldm,savm, k, nres);  
             }           fprintf(ficrespij,"# Cov Agex agex+h hpijx with i,j=");
             for(j=1; j<=nlstate+ndeath;j++) {          for(i=1; i<=nlstate;i++)
               kk1=0.;kk2=0;            for(j=1; j<=nlstate+ndeath;j++)
               for(i=1; i<=nlstate;i++) {                            fprintf(ficrespij," %1d-%1d",i,j);
                 if (mobilav==1)           fprintf(ficrespij,"\n");
                   kk1=kk1+p3mat[i][j][h]*mobaverage[(int)agedeb+1][i][cptcod];          for (h=0; h<=nhstepm; h++){
                 else {            /*agedebphstep = agedeb + h*hstepm/YEARM*stepm;*/
                   kk1=kk1+p3mat[i][j][h]*probs[(int)(agedeb+1)][i][cptcod];            fprintf(ficrespij,"%d %3.f %3.f",k, agedeb, agedeb + h*hstepm/YEARM*stepm );
                 }            for(i=1; i<=nlstate;i++)
               }              for(j=1; j<=nlstate+ndeath;j++)
               if (h==(int)(calagedatem+12*cpt)){                fprintf(ficrespij," %.5f", p3mat[i][j][h]);
                 tabpop[(int)(agedeb)][j][cptcod]=kk1;            fprintf(ficrespij,"\n");
                   /*fprintf(ficrespop," %.3f", kk1);  
                     if (popforecast==1) fprintf(ficrespop," [%.f]", kk1*popeffectif[(int)agedeb+1]);*/  
               }  
             }  
             for(i=1; i<=nlstate;i++){  
               kk1=0.;  
                 for(j=1; j<=nlstate;j++){  
                   kk1= kk1+tabpop[(int)(agedeb)][j][cptcod];   
                 }  
                   tabpopprev[(int)(agedeb)][i][cptcod]=tabpop[(int)(agedeb)][i][cptcod]/kk1*popeffectif[(int)(agedeb+(calagedatem+12*cpt)*hstepm/YEARM*stepm-1)];  
             }  
   
             if (h==(int)(calagedatem+12*cpt)) for(j=1; j<=nlstate;j++)   
               fprintf(ficrespop," %15.2f",tabpopprev[(int)(agedeb+1)][j][cptcod]);  
           }  
           free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);  
         }          }
           free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
           fprintf(ficrespij,"\n");
         }
         /*}*/
       }
       return 0;
   }
    
    int hBijx(double *p, int bage, int fage, double ***prevacurrent){
       /*------------- h Bij x at various ages ------------*/
   
     int stepsize;
     /* int agelim; */
           int ageminl;
     int hstepm;
     int nhstepm;
     int h, i, i1, j, k, nres;
           
     double agedeb;
     double ***p3mat;
           
     strcpy(filerespijb,"PIJB_");  strcat(filerespijb,fileresu);
     if((ficrespijb=fopen(filerespijb,"w"))==NULL) {
       printf("Problem with Pij back resultfile: %s\n", filerespijb); return 1;
       fprintf(ficlog,"Problem with Pij back resultfile: %s\n", filerespijb); return 1;
     }
     printf("Computing pij back: result on file '%s' \n", filerespijb);
     fprintf(ficlog,"Computing pij back: result on file '%s' \n", filerespijb);
     
     stepsize=(int) (stepm+YEARM-1)/YEARM;
     /*if (stepm<=24) stepsize=2;*/
     
     /* agelim=AGESUP; */
     ageminl=AGEINF; /* was 30 */
     hstepm=stepsize*YEARM; /* Every year of age */
     hstepm=hstepm/stepm; /* Typically 2 years, = 2/6 months = 4 */
     
     /* hstepm=1;   aff par mois*/
     pstamp(ficrespijb);
     fprintf(ficrespijb,"#****** h Bij x Back probability to be in state i at age x-h being in j at x: B1j+B2j+...=1 ");
     i1= pow(2,cptcoveff);
     /* for(cptcov=1,k=0;cptcov<=i1;cptcov++){ */
     /*    /\*for(cptcod=1;cptcod<=ncodemax[cptcov];cptcod++){*\/ */
     /*    k=k+1;  */
     for(nres=1; nres <= nresult; nres++){ /* For each resultline */
       for(k=1; k<=i1;k++){ /* For any combination of dummy covariates, fixed and varying */
         if(i1 != 1 && TKresult[nres]!= k)
           continue;
         fprintf(ficrespijb,"\n#****** ");
         for(j=1;j<=cptcoveff;j++)
           fprintf(ficrespijb,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
         for (j=1; j<= nsq; j++){ /* For each selected (single) quantitative value */
           fprintf(ficrespijb," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]);
         }
         fprintf(ficrespijb,"******\n");
         if(invalidvarcomb[k]){  /* Is it necessary here? */
           fprintf(ficrespijb,"\n#Combination (%d) ignored because no cases \n",k); 
           continue;
       }        }
          
   /******/        /* for (agedeb=fage; agedeb>=bage; agedeb--){ /\* If stepm=6 months *\/ */
         for (agedeb=bage; agedeb<=fage; agedeb++){ /* If stepm=6 months and estepm=24 (2 years) */
           /* nhstepm=(int) rint((agelim-agedeb)*YEARM/stepm); /\* Typically 20 years = 20*12/6=40 *\/ */
           nhstepm=(int) rint((agedeb-ageminl)*YEARM/stepm+0.1)-1; /* Typically 20 years = 20*12/6=40 or 55*12/24=27.5-1.1=>27 */
           nhstepm = nhstepm/hstepm; /* Typically 40/4=10, because estepm=24 stepm=6 => hstepm=24/6=4 or 28*/
           
           /*        nhstepm=nhstepm*YEARM; aff par mois*/
           
           p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm); /* We can't have it at an upper level because of nhstepm */
           /* and memory limitations if stepm is small */
   
       for (cpt=1; cpt<=(anpyram1-anpyram);cpt++) {           /* oldm=oldms;savm=savms; */
         fprintf(ficrespop,"\n\n# Forecasting at date %.lf/%.lf/%.lf ",jpyram,mpyram,anpyram+cpt);             /* hbxij(p3mat,nhstepm,agedeb,hstepm,p,nlstate,stepm,oldm,savm, k);   */
         for (agedeb=(fage-((int)calagedatem %12/12.)); agedeb>=(ageminpar-((int)calagedatem %12)/12.); agedeb--){           hbxij(p3mat,nhstepm,agedeb,hstepm,p,prevacurrent,nlstate,stepm, k, nres);/* Bug valgrind */
           nhstepm=(int) rint((agelim-agedeb)*YEARM/stepm);           /* hbxij(p3mat,nhstepm,agedeb,hstepm,p,prevacurrent,nlstate,stepm,oldm,savm, dnewm, doldm, dsavm, k); */
           nhstepm = nhstepm/hstepm;           fprintf(ficrespijb,"# Cov Agex agex-h hbijx with i,j=");
                     for(i=1; i<=nlstate;i++)
           p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);            for(j=1; j<=nlstate+ndeath;j++)
           oldm=oldms;savm=savms;              fprintf(ficrespijb," %1d-%1d",i,j);
           hpxij(p3mat,nhstepm,agedeb,hstepm,p,nlstate,stepm,oldm,savm, k);            fprintf(ficrespijb,"\n");
           for (h=0; h<=nhstepm; h++){          for (h=0; h<=nhstepm; h++){
             if (h==(int) (calagedatem+YEARM*cpt)) {            /*agedebphstep = agedeb + h*hstepm/YEARM*stepm;*/
               fprintf(ficresf,"\n %3.f ",agedeb+h*hstepm/YEARM*stepm);            fprintf(ficrespijb,"%d %3.f %3.f",k, agedeb, agedeb - h*hstepm/YEARM*stepm );
             }             /* fprintf(ficrespijb,"%d %3.f %3.f",k, agedeb, agedeb + h*hstepm/YEARM*stepm ); */
             for(j=1; j<=nlstate+ndeath;j++) {            for(i=1; i<=nlstate;i++)
               kk1=0.;kk2=0;              for(j=1; j<=nlstate+ndeath;j++)
               for(i=1; i<=nlstate;i++) {                              fprintf(ficrespijb," %.5f", p3mat[i][j][h]);/* Bug valgrind */
                 kk1=kk1+p3mat[i][j][h]*tabpopprev[(int)agedeb+1][i][cptcod];                fprintf(ficrespijb,"\n");
               }  
               if (h==(int)(calagedatem+12*cpt)) fprintf(ficresf," %15.2f", kk1);          
             }  
           }  
           free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);  
         }          }
       }          free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);
    }           fprintf(ficrespijb,"\n");
   }        } /* end age deb */
        } /* end combination */
   if (mobilav!=0) free_ma3x(mobaverage,1, AGESUP,1,NCOVMAX, 1,NCOVMAX);    } /* end nres */
     return 0;
    } /*  hBijx */
   
   if (popforecast==1) {  
     free_ivector(popage,0,AGESUP);  
     free_vector(popeffectif,0,AGESUP);  
     free_vector(popcount,0,AGESUP);  
   }  
   free_ma3x(tabpop,1, AGESUP,1,NCOVMAX, 1,NCOVMAX);  
   free_ma3x(tabpopprev,1, AGESUP,1,NCOVMAX, 1,NCOVMAX);  
   fclose(ficrespop);  
 } /* End of popforecast */  
   
 /***********************************************/  /***********************************************/
 /**************** Main Program *****************/  /**************** Main Program *****************/
Line 3460  populforecast(char fileres[], double anp Line 11779  populforecast(char fileres[], double anp
   
 int main(int argc, char *argv[])  int main(int argc, char *argv[])
 {  {
   #ifdef GSL
     const gsl_multimin_fminimizer_type *T;
     size_t iteri = 0, it;
     int rval = GSL_CONTINUE;
     int status = GSL_SUCCESS;
     double ssval;
   #endif
   int movingaverage(double ***probs, double bage,double fage, double ***mobaverage, int mobilav);    int movingaverage(double ***probs, double bage,double fage, double ***mobaverage, int mobilav);
   int i,j, k, n=MAXN,iter,m,size=100,cptcode, cptcod;    int i,j, k, iter=0,m,size=100, cptcod; /* Suppressing because nobs */
   double agedeb, agefin,hf;    /* int i,j, k, n=MAXN,iter=0,m,size=100, cptcod; */
   double ageminpar=1.e20,agemin=1.e20, agemaxpar=-1.e20, agemax=-1.e20;    int ncvyear=0; /* Number of years needed for the period prevalence to converge */
     int jj, ll, li, lj, lk;
     int numlinepar=0; /* Current linenumber of parameter file */
     int num_filled;
     int itimes;
     int NDIM=2;
     int vpopbased=0;
     int nres=0;
     int endishere=0;
     int noffset=0;
     int ncurrv=0; /* Temporary variable */
     
     char ca[32], cb[32];
     /*  FILE *fichtm; *//* Html File */
     /* FILE *ficgp;*/ /*Gnuplot File */
     struct stat info;
     double agedeb=0.;
   
   double fret;    double ageminpar=AGEOVERFLOW,agemin=AGEOVERFLOW, agemaxpar=-AGEOVERFLOW, agemax=-AGEOVERFLOW;
   double **xi,tmp,delta;    double ageminout=-AGEOVERFLOW,agemaxout=AGEOVERFLOW; /* Smaller Age range redefined after movingaverage */
   
   double dum; /* Dummy variable */    double fret;
     double dum=0.; /* Dummy variable */
   double ***p3mat;    double ***p3mat;
   double ***mobaverage;    /* double ***mobaverage; */
   int *indx;    double wald;
   char line[MAXLINE], linepar[MAXLINE];  
   char path[80],pathc[80],pathcd[80],pathtot[80],model[80];    char line[MAXLINE];
   int firstobs=1, lastobs=10;    char path[MAXLINE],pathc[MAXLINE],pathcd[MAXLINE],pathtot[MAXLINE];
   int sdeb, sfin; /* Status at beginning and end */  
   int c,  h , cpt,l;    char  modeltemp[MAXLINE];
   int ju,jl, mi;    char resultline[MAXLINE], resultlineori[MAXLINE];
   int i1,j1, k1,k2,k3,jk,aa,bb, stepsize, ij;    
   int jnais,jdc,jint4,jint1,jint2,jint3,**outcome,*tab;     char pathr[MAXLINE], pathimach[MAXLINE]; 
     char *tok, *val; /* pathtot */
     /* int firstobs=1, lastobs=10; /\* nobs = lastobs-firstobs declared globally ;*\/ */
     int c,  h , cpt, c2;
     int jl=0;
     int i1, j1, jk, stepsize=0;
     int count=0;
   
     int *tab; 
   int mobilavproj=0 , prevfcast=0 ; /* moving average of prev, If prevfcast=1 prevalence projection */    int mobilavproj=0 , prevfcast=0 ; /* moving average of prev, If prevfcast=1 prevalence projection */
     /* double anprojd, mprojd, jprojd; /\* For eventual projections *\/ */
     /* double anprojf, mprojf, jprojf; */
     /* double jintmean,mintmean,aintmean;   */
     int prvforecast = 0; /* Might be 1 (date of beginning of projection is a choice or 2 is the dateintmean */
     int prvbackcast = 0; /* Might be 1 (date of beginning of projection is a choice or 2 is the dateintmean */
     double yrfproj= 10.0; /* Number of years of forward projections */
     double yrbproj= 10.0; /* Number of years of backward projections */
     int prevbcast=0; /* defined as global for mlikeli and mle, replacing backcast */
   int mobilav=0,popforecast=0;    int mobilav=0,popforecast=0;
   int hstepm, nhstepm;    int hstepm=0, nhstepm=0;
     int agemortsup;
     float  sumlpop=0.;
   double jprev1=1, mprev1=1,anprev1=2000,jprev2=1, mprev2=1,anprev2=2000;    double jprev1=1, mprev1=1,anprev1=2000,jprev2=1, mprev2=1,anprev2=2000;
   double jpyram=1, mpyram=1,anpyram=2000,jpyram1=1, mpyram1=1,anpyram1=2000;    double jpyram=1, mpyram=1,anpyram=2000,jpyram1=1, mpyram1=1,anpyram1=2000;
   
   double bage, fage, age, agelim, agebase;    double bage=0, fage=110., age, agelim=0., agebase=0.;
   double ftolpl=FTOL;    double ftolpl=FTOL;
   double **prlim;    double **prlim;
   double *severity;    double **bprlim;
   double ***param; /* Matrix of parameters */    double ***param; /* Matrix of parameters, param[i][j][k] param=ma3x(1,nlstate,1,nlstate+ndeath-1,1,ncovmodel) 
   double  *p;                      state of origin, state of destination including death, for each covariate: constante, age, and V1 V2 etc. */
     double ***paramstart; /* Matrix of starting parameter values */
     double  *p, *pstart; /* p=param[1][1] pstart is for starting values guessed by freqsummary */
   double **matcov; /* Matrix of covariance */    double **matcov; /* Matrix of covariance */
     double **hess; /* Hessian matrix */
   double ***delti3; /* Scale */    double ***delti3; /* Scale */
   double *delti; /* Scale */    double *delti; /* Scale */
   double ***eij, ***vareij;    double ***eij, ***vareij;
   double **varpl; /* Variances of prevalence limits by age */    double **varpl; /* Variances of prevalence limits by age */
   double *epj, vepp;  
   double kk1, kk2;  
   double dateprev1, dateprev2,jproj1=1,mproj1=1,anproj1=2000,jproj2=1,mproj2=1,anproj2=2000;  
   
   char *alph[]={"a","a","b","c","d","e"}, str[4];    double *epj, vepp;
   
     double dateprev1, dateprev2;
     double jproj1=1,mproj1=1,anproj1=2000,jproj2=1,mproj2=1,anproj2=2000, dateproj1=0, dateproj2=0, dateprojd=0, dateprojf=0;
     double jback1=1,mback1=1,anback1=2000,jback2=1,mback2=1,anback2=2000, dateback1=0, dateback2=0, datebackd=0, datebackf=0;
   
   
     double **ximort;
     char *alph[]={"a","a","b","c","d","e"}, str[4]="1234";
     int *dcwave;
   
     char z[1]="c";
   
     /*char  *strt;*/
     char strtend[80];
   
   
   /*   setlocale (LC_ALL, ""); */
   /*   bindtextdomain (PACKAGE, LOCALEDIR); */
   /*   textdomain (PACKAGE); */
   /*   setlocale (LC_CTYPE, ""); */
   /*   setlocale (LC_MESSAGES, ""); */
   
     /*   gettimeofday(&start_time, (struct timezone*)0); */ /* at first time */
     rstart_time = time(NULL);  
     /*  (void) gettimeofday(&start_time,&tzp);*/
     start_time = *localtime(&rstart_time);
     curr_time=start_time;
     /*tml = *localtime(&start_time.tm_sec);*/
     /* strcpy(strstart,asctime(&tml)); */
     strcpy(strstart,asctime(&start_time));
   
   /*  printf("Localtime (at start)=%s",strstart); */
   /*  tp.tm_sec = tp.tm_sec +86400; */
   /*  tm = *localtime(&start_time.tm_sec); */
   /*   tmg.tm_year=tmg.tm_year +dsign*dyear; */
   /*   tmg.tm_mon=tmg.tm_mon +dsign*dmonth; */
   /*   tmg.tm_hour=tmg.tm_hour + 1; */
   /*   tp.tm_sec = mktime(&tmg); */
   /*   strt=asctime(&tmg); */
   /*   printf("Time(after) =%s",strstart);  */
   /*  (void) time (&time_value);
   *  printf("time=%d,t-=%d\n",time_value,time_value-86400);
   *  tm = *localtime(&time_value);
   *  strstart=asctime(&tm);
   *  printf("tim_value=%d,asctime=%s\n",time_value,strstart); 
   */
   
   char z[1]="c", occ;    nberr=0; /* Number of errors and warnings */
 #include <sys/time.h>    nbwarn=0;
 #include <time.h>  #ifdef WIN32
   char stra[80], strb[80], strc[80], strd[80],stre[80],modelsav[80];    _getcwd(pathcd, size);
    #else
   /* long total_usecs;  
      struct timeval start_time, end_time;  
     
      gettimeofday(&start_time, (struct timezone*)0); */ /* at first time */  
   getcwd(pathcd, size);    getcwd(pathcd, size);
   #endif
   printf("\n%s\n%s",version,fullversion);    syscompilerinfo(0);
     printf("\nIMaCh version %s, %s\n%s",version, copyright, fullversion);
   if(argc <=1){    if(argc <=1){
     printf("\nEnter the parameter file name: ");      printf("\nEnter the parameter file name: ");
     scanf("%s",pathtot);      if(!fgets(pathr,FILENAMELENGTH,stdin)){
         printf("ERROR Empty parameter file name\n");
         goto end;
       }
       i=strlen(pathr);
       if(pathr[i-1]=='\n')
         pathr[i-1]='\0';
       i=strlen(pathr);
       if(i >= 1 && pathr[i-1]==' ') {/* This may happen when dragging on oS/X! */
         pathr[i-1]='\0';
       }
       i=strlen(pathr);
       if( i==0 ){
         printf("ERROR Empty parameter file name\n");
         goto end;
       }
       for (tok = pathr; tok != NULL; ){
         printf("Pathr |%s|\n",pathr);
         while ((val = strsep(&tok, "\"" )) != NULL && *val == '\0');
         printf("val= |%s| pathr=%s\n",val,pathr);
         strcpy (pathtot, val);
         if(pathr[0] == '\0') break; /* Dirty */
       }
     }
     else if (argc<=2){
       strcpy(pathtot,argv[1]);
   }    }
   else{    else{
     strcpy(pathtot,argv[1]);      strcpy(pathtot,argv[1]);
       strcpy(z,argv[2]);
       printf("\nargv[2]=%s z=%c\n",argv[2],z[0]);
   }    }
   /*if(getcwd(pathcd, 80)!= NULL)printf ("Error pathcd\n");*/    /*if(getcwd(pathcd, MAXLINE)!= NULL)printf ("Error pathcd\n");*/
   /*cygwin_split_path(pathtot,path,optionfile);    /*cygwin_split_path(pathtot,path,optionfile);
     printf("pathtot=%s, path=%s, optionfile=%s\n",pathtot,path,optionfile);*/      printf("pathtot=%s, path=%s, optionfile=%s\n",pathtot,path,optionfile);*/
   /* cutv(path,optionfile,pathtot,'\\');*/    /* cutv(path,optionfile,pathtot,'\\');*/
   
     /* Split argv[0], imach program to get pathimach */
     printf("\nargv[0]=%s argv[1]=%s, \n",argv[0],argv[1]);
     split(argv[0],pathimach,optionfile,optionfilext,optionfilefiname);
     printf("\nargv[0]=%s pathimach=%s, \noptionfile=%s \noptionfilext=%s \noptionfilefiname=%s\n",argv[0],pathimach,optionfile,optionfilext,optionfilefiname);
    /*   strcpy(pathimach,argv[0]); */
     /* Split argv[1]=pathtot, parameter file name to get path, optionfile, extension and name */
   split(pathtot,path,optionfile,optionfilext,optionfilefiname);    split(pathtot,path,optionfile,optionfilext,optionfilefiname);
   printf("pathtot=%s, path=%s, optionfile=%s optionfilext=%s optionfilefiname=%s\n",pathtot,path,optionfile,optionfilext,optionfilefiname);    printf("\npathtot=%s,\npath=%s,\noptionfile=%s \noptionfilext=%s \noptionfilefiname=%s\n",pathtot,path,optionfile,optionfilext,optionfilefiname);
   chdir(path);  #ifdef WIN32
   replace(pathc,path);    _chdir(path); /* Can be a relative path */
     if(_getcwd(pathcd,MAXLINE) > 0) /* So pathcd is the full path */
   #else
     chdir(path); /* Can be a relative path */
     if (getcwd(pathcd, MAXLINE) > 0) /* So pathcd is the full path */
   #endif
     printf("Current directory %s!\n",pathcd);
     strcpy(command,"mkdir ");
     strcat(command,optionfilefiname);
     if((outcmd=system(command)) != 0){
       printf("Directory already exists (or can't create it) %s%s, err=%d\n",path,optionfilefiname,outcmd);
       /* fprintf(ficlog,"Problem creating directory %s%s\n",path,optionfilefiname); */
       /* fclose(ficlog); */
   /*     exit(1); */
     }
   /*   if((imk=mkdir(optionfilefiname))<0){ */
   /*     perror("mkdir"); */
   /*   } */
   
   /*-------- arguments in the command line --------*/    /*-------- arguments in the command line --------*/
   
   /* Log file */    /* Main Log file */
   strcat(filelog, optionfilefiname);    strcat(filelog, optionfilefiname);
   strcat(filelog,".log");    /* */    strcat(filelog,".log");    /* */
   if((ficlog=fopen(filelog,"w"))==NULL)    {    if((ficlog=fopen(filelog,"w"))==NULL)    {
Line 3543  int main(int argc, char *argv[]) Line 11996  int main(int argc, char *argv[])
     goto end;      goto end;
   }    }
   fprintf(ficlog,"Log filename:%s\n",filelog);    fprintf(ficlog,"Log filename:%s\n",filelog);
   fprintf(ficlog,"\n%s",version);    fprintf(ficlog,"Version %s %s",version,fullversion);
   fprintf(ficlog,"\nEnter the parameter file name: ");    fprintf(ficlog,"\nEnter the parameter file name: \n");
   fprintf(ficlog,"pathtot=%s, path=%s, optionfile=%s optionfilext=%s optionfilefiname=%s\n",pathtot,path,optionfile,optionfilext,optionfilefiname);    fprintf(ficlog,"pathimach=%s\npathtot=%s\n\
    path=%s \n\
    optionfile=%s\n\
    optionfilext=%s\n\
    optionfilefiname='%s'\n",pathimach,pathtot,path,optionfile,optionfilext,optionfilefiname);
   
     syscompilerinfo(1);
   
     printf("Local time (at start):%s",strstart);
     fprintf(ficlog,"Local time (at start): %s",strstart);
   fflush(ficlog);    fflush(ficlog);
   /*   (void) gettimeofday(&curr_time,&tzp); */
   /*   printf("Elapsed time %d\n", asc_diff_time(curr_time.tm_sec-start_time.tm_sec,tmpout)); */
   
   /* */    /* */
   strcpy(fileres,"r");    strcpy(fileres,"r");
   strcat(fileres, optionfilefiname);    strcat(fileres, optionfilefiname);
     strcat(fileresu, optionfilefiname); /* Without r in front */
   strcat(fileres,".txt");    /* Other files have txt extension */    strcat(fileres,".txt");    /* Other files have txt extension */
     strcat(fileresu,".txt");    /* Other files have txt extension */
   
   /*---------arguments file --------*/    /* Main ---------arguments file --------*/
   
   if((ficpar=fopen(optionfile,"r"))==NULL)    {    if((ficpar=fopen(optionfile,"r"))==NULL)    {
     printf("Problem with optionfile %s\n",optionfile);      printf("Problem with optionfile '%s' with errno='%s'\n",optionfile,strerror(errno));
     fprintf(ficlog,"Problem with optionfile %s\n",optionfile);      fprintf(ficlog,"Problem with optionfile '%s' with errno='%s'\n",optionfile,strerror(errno));
     goto end;      fflush(ficlog);
       /* goto end; */
       exit(70); 
   }    }
   
   strcpy(filereso,"o");    strcpy(filereso,"o");
   strcat(filereso,fileres);    strcat(filereso,fileresu);
   if((ficparo=fopen(filereso,"w"))==NULL) {    if((ficparo=fopen(filereso,"w"))==NULL) { /* opened on subdirectory */
     printf("Problem with Output resultfile: %s\n", filereso);      printf("Problem with Output resultfile: %s\n", filereso);
     fprintf(ficlog,"Problem with Output resultfile: %s\n", filereso);      fprintf(ficlog,"Problem with Output resultfile: %s\n", filereso);
       fflush(ficlog);
     goto end;      goto end;
   }    }
         /*-------- Rewriting parameter file ----------*/
   /* Reads comments: lines beginning with '#' */    strcpy(rfileres,"r");    /* "Rparameterfile */
   while((c=getc(ficpar))=='#' && c!= EOF){    strcat(rfileres,optionfilefiname);    /* Parameter file first name */
     ungetc(c,ficpar);    strcat(rfileres,".");    /* */
     fgets(line, MAXLINE, ficpar);    strcat(rfileres,optionfilext);    /* Other files have txt extension */
     puts(line);    if((ficres =fopen(rfileres,"w"))==NULL) {
     fputs(line,ficparo);      printf("Problem writing new parameter file: %s\n", rfileres);goto end;
   }      fprintf(ficlog,"Problem writing new parameter file: %s\n", rfileres);goto end;
   ungetc(c,ficpar);      fflush(ficlog);
       goto end;
   fscanf(ficpar,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%lf stepm=%d ncovcol=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d model=%s\n",title, datafile, &lastobs, &firstpass,&lastpass,&ftol, &stepm, &ncovcol, &nlstate,&ndeath, &maxwav, &mle, &weightopt,model);    }
   printf("title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%e stepm=%d ncovcol=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\nmodel=%s\n", title, datafile, lastobs, firstpass,lastpass,ftol, stepm, ncovcol, nlstate,ndeath, maxwav, mle, weightopt,model);    fprintf(ficres,"#IMaCh %s\n",version);
   fprintf(ficparo,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%e stepm=%d ncovcol=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\nmodel=%s\n", title, datafile, lastobs, firstpass,lastpass,ftol,stepm,ncovcol,nlstate,ndeath,maxwav, mle, weightopt,model);  
   while((c=getc(ficpar))=='#' && c!= EOF){  
     ungetc(c,ficpar);  
     fgets(line, MAXLINE, ficpar);  
     puts(line);  
     fputs(line,ficparo);  
   }  
   ungetc(c,ficpar);  
     
      
   covar=matrix(0,NCOVMAX,1,n);   
   cptcovn=0; /*Number of covariates, i.e. number of '+' in model statement*/  
   if (strlen(model)>1) cptcovn=nbocc(model,'+')+1;  
   
   ncovmodel=2+cptcovn; /*Number of variables = cptcovn + intercept + age */  
   nvar=ncovmodel-1; /* Suppressing age as a basic covariate */  
     
   /* Read guess parameters */  
   /* Reads comments: lines beginning with '#' */  
   while((c=getc(ficpar))=='#' && c!= EOF){  
     ungetc(c,ficpar);  
     fgets(line, MAXLINE, ficpar);  
     puts(line);  
     fputs(line,ficparo);  
   }  
   ungetc(c,ficpar);  
     
   param= ma3x(1,nlstate,1,nlstate+ndeath-1,1,ncovmodel);  
   for(i=1; i <=nlstate; i++)  
     for(j=1; j <=nlstate+ndeath-1; j++){  
       fscanf(ficpar,"%1d%1d",&i1,&j1);  
       fprintf(ficparo,"%1d%1d",i1,j1);  
       if(mle==1)  
         printf("%1d%1d",i,j);  
       fprintf(ficlog,"%1d%1d",i,j);  
       for(k=1; k<=ncovmodel;k++){  
         fscanf(ficpar," %lf",&param[i][j][k]);  
         if(mle==1){  
           printf(" %lf",param[i][j][k]);  
           fprintf(ficlog," %lf",param[i][j][k]);  
         }  
         else  
           fprintf(ficlog," %lf",param[i][j][k]);  
         fprintf(ficparo," %lf",param[i][j][k]);  
       }  
       fscanf(ficpar,"\n");  
       if(mle==1)  
         printf("\n");  
       fprintf(ficlog,"\n");  
       fprintf(ficparo,"\n");  
     }  
     
   npar= (nlstate+ndeath-1)*nlstate*ncovmodel; /* Number of parameters*/  
   
   p=param[1][1];  
     
   /* Reads comments: lines beginning with '#' */  
   while((c=getc(ficpar))=='#' && c!= EOF){  
     ungetc(c,ficpar);  
     fgets(line, MAXLINE, ficpar);  
     puts(line);  
     fputs(line,ficparo);  
   }  
   ungetc(c,ficpar);  
   
   delti3= ma3x(1,nlstate,1,nlstate+ndeath-1,1,ncovmodel);  
   /* delti=vector(1,npar); *//* Scale of each paramater (output from hesscov) */  
   for(i=1; i <=nlstate; i++){  
     for(j=1; j <=nlstate+ndeath-1; j++){  
       fscanf(ficpar,"%1d%1d",&i1,&j1);  
       printf("%1d%1d",i,j);  
       fprintf(ficparo,"%1d%1d",i1,j1);  
       for(k=1; k<=ncovmodel;k++){  
         fscanf(ficpar,"%le",&delti3[i][j][k]);  
         printf(" %le",delti3[i][j][k]);  
         fprintf(ficparo," %le",delti3[i][j][k]);  
       }  
       fscanf(ficpar,"\n");  
       printf("\n");  
       fprintf(ficparo,"\n");  
     }  
   }  
   delti=delti3[1][1];  
   
   
   /* free_ma3x(delti3,1,nlstate,1,nlstate+ndeath-1,1,ncovmodel); */ /* Hasn't to to freed here otherwise delti is no more allocated */  
     
   /* Reads comments: lines beginning with '#' */  
   while((c=getc(ficpar))=='#' && c!= EOF){  
     ungetc(c,ficpar);  
     fgets(line, MAXLINE, ficpar);  
     puts(line);  
     fputs(line,ficparo);  
   }  
   ungetc(c,ficpar);  
     
   matcov=matrix(1,npar,1,npar);  
   for(i=1; i <=npar; i++){  
     fscanf(ficpar,"%s",&str);  
     if(mle==1)  
       printf("%s",str);  
     fprintf(ficlog,"%s",str);  
     fprintf(ficparo,"%s",str);  
     for(j=1; j <=i; j++){  
       fscanf(ficpar," %le",&matcov[i][j]);  
       if(mle==1){  
         printf(" %.5le",matcov[i][j]);  
         fprintf(ficlog," %.5le",matcov[i][j]);  
       }  
       else  
         fprintf(ficlog," %.5le",matcov[i][j]);  
       fprintf(ficparo," %.5le",matcov[i][j]);  
     }  
     fscanf(ficpar,"\n");  
     if(mle==1)  
       printf("\n");  
     fprintf(ficlog,"\n");  
     fprintf(ficparo,"\n");  
   }  
   for(i=1; i <=npar; i++)  
     for(j=i+1;j<=npar;j++)  
       matcov[i][j]=matcov[j][i];  
      
   if(mle==1)  
     printf("\n");  
   fprintf(ficlog,"\n");  
   
   
   /*-------- Rewriting paramater file ----------*/  
   strcpy(rfileres,"r");    /* "Rparameterfile */  
   strcat(rfileres,optionfilefiname);    /* Parameter file first name*/  
   strcat(rfileres,".");    /* */  
   strcat(rfileres,optionfilext);    /* Other files have txt extension */  
   if((ficres =fopen(rfileres,"w"))==NULL) {  
     printf("Problem writing new parameter file: %s\n", fileres);goto end;  
     fprintf(ficlog,"Problem writing new parameter file: %s\n", fileres);goto end;  
   }  
   fprintf(ficres,"#%s\n",version);  
       
   /*-------- data file ----------*/  
   if((fic=fopen(datafile,"r"))==NULL)    {  
     printf("Problem with datafile: %s\n", datafile);goto end;  
     fprintf(ficlog,"Problem with datafile: %s\n", datafile);goto end;  
   }  
   
   n= lastobs;  
   severity = vector(1,maxwav);  
   outcome=imatrix(1,maxwav+1,1,n);  
   num=ivector(1,n);  
   moisnais=vector(1,n);  
   annais=vector(1,n);  
   moisdc=vector(1,n);  
   andc=vector(1,n);  
   agedc=vector(1,n);  
   cod=ivector(1,n);  
   weight=vector(1,n);  
   for(i=1;i<=n;i++) weight[i]=1.0; /* Equal weights, 1 by default */  
   mint=matrix(1,maxwav,1,n);  
   anint=matrix(1,maxwav,1,n);  
   s=imatrix(1,maxwav+1,1,n);  
   tab=ivector(1,NCOVMAX);  
   ncodemax=ivector(1,8);  
   
   i=1;  
   while (fgets(line, MAXLINE, fic) != NULL)    {  
     if ((i >= firstobs) && (i <=lastobs)) {  
           
       for (j=maxwav;j>=1;j--){  
         cutv(stra, strb,line,' '); s[j][i]=atoi(strb);   
         strcpy(line,stra);  
         cutv(stra, strb,line,'/'); anint[j][i]=(double)(atoi(strb)); strcpy(line,stra);  
         cutv(stra, strb,line,' '); mint[j][i]=(double)(atoi(strb)); strcpy(line,stra);  
       }  
           
       cutv(stra, strb,line,'/'); andc[i]=(double)(atoi(strb)); strcpy(line,stra);  
       cutv(stra, strb,line,' '); moisdc[i]=(double)(atoi(strb)); strcpy(line,stra);  
   
       cutv(stra, strb,line,'/'); annais[i]=(double)(atoi(strb)); strcpy(line,stra);  
       cutv(stra, strb,line,' '); moisnais[i]=(double)(atoi(strb)); strcpy(line,stra);  
   
       cutv(stra, strb,line,' '); weight[i]=(double)(atoi(strb)); strcpy(line,stra);  
       for (j=ncovcol;j>=1;j--){  
         cutv(stra, strb,line,' '); covar[j][i]=(double)(atoi(strb)); strcpy(line,stra);  
       }   
       num[i]=atol(stra);  
           
       /*if((s[2][i]==2) && (s[3][i]==-1)&&(s[4][i]==9)){  
         printf("%d %.lf %.lf %.lf %.lf/%.lf %.lf/%.lf %.lf/%.lf %d %.lf/%.lf %d %.lf/%.lf %d %.lf/%.lf %d\n",num[i],(covar[1][i]), (covar[2][i]),weight[i], (moisnais[i]), (annais[i]), (moisdc[i]), (andc[i]), (mint[1][i]), (anint[1][i]), (s[1][i]),  (mint[2][i]), (anint[2][i]), (s[2][i]),  (mint[3][i]), (anint[3][i]), (s[3][i]),  (mint[4][i]), (anint[4][i]), (s[4][i])); ij=ij+1;}*/  
   
       i=i+1;  
     }  
   }  
   /* printf("ii=%d", ij);  
      scanf("%d",i);*/  
   imx=i-1; /* Number of individuals */  
   
   /* for (i=1; i<=imx; i++){  
     if ((s[1][i]==3) && (s[2][i]==2)) s[2][i]=3;  
     if ((s[2][i]==3) && (s[3][i]==2)) s[3][i]=3;  
     if ((s[3][i]==3) && (s[4][i]==2)) s[4][i]=3;  
     }*/  
    /*  for (i=1; i<=imx; i++){  
      if (s[4][i]==9)  s[4][i]=-1;   
      printf("%d %.lf %.lf %.lf %.lf/%.lf %.lf/%.lf %.lf/%.lf %d %.lf/%.lf %d %.lf/%.lf %d %.lf/%.lf %d\n",num[i],(covar[1][i]), (covar[2][i]), (weight[i]), (moisnais[i]), (annais[i]), (moisdc[i]), (andc[i]), (mint[1][i]), (anint[1][i]), (s[1][i]),  (mint[2][i]), (anint[2][i]), (s[2][i]),  (mint[3][i]), (anint[3][i]), (s[3][i]),  (mint[4][i]), (anint[4][i]), (s[4][i]));}*/  
     
  for (i=1; i<=imx; i++)  
    
    /*if ((s[3][i]==3) ||  (s[4][i]==3)) weight[i]=0.08;  
      else weight[i]=1;*/  
   
   /* Calculation of the number of parameter from char model*/  
   Tvar=ivector(1,15); /* stores the number n of the covariates in Vm+Vn at 1 and m at 2 */  
   Tprod=ivector(1,15);   
   Tvaraff=ivector(1,15);   
   Tvard=imatrix(1,15,1,2);  
   Tage=ivector(1,15);        
      
   if (strlen(model) >1){ /* If there is at least 1 covariate */  
     j=0, j1=0, k1=1, k2=1;  
     j=nbocc(model,'+'); /* j=Number of '+' */  
     j1=nbocc(model,'*'); /* j1=Number of '*' */  
     cptcovn=j+1;   
     cptcovprod=j1; /*Number of products */  
       
     strcpy(modelsav,model);   
     if ((strcmp(model,"age")==0) || (strcmp(model,"age*age")==0)){  
       printf("Error. Non available option model=%s ",model);  
       fprintf(ficlog,"Error. Non available option model=%s ",model);  
       goto end;  
     }  
       
     /* This loop fills the array Tvar from the string 'model'.*/  
   
     for(i=(j+1); i>=1;i--){  
       cutv(stra,strb,modelsav,'+'); /* keeps in strb after the last + */   
       if (nbocc(modelsav,'+')==0) strcpy(strb,modelsav); /* and analyzes it */  
       /*      printf("i=%d a=%s b=%s sav=%s\n",i, stra,strb,modelsav);*/  
       /*scanf("%d",i);*/  
       if (strchr(strb,'*')) {  /* Model includes a product */  
         cutv(strd,strc,strb,'*'); /* strd*strc  Vm*Vn (if not *age)*/  
         if (strcmp(strc,"age")==0) { /* Vn*age */  
           cptcovprod--;  
           cutv(strb,stre,strd,'V');  
           Tvar[i]=atoi(stre); /* computes n in Vn and stores in Tvar*/  
           cptcovage++;  
             Tage[cptcovage]=i;  
             /*printf("stre=%s ", stre);*/  
         }  
         else if (strcmp(strd,"age")==0) { /* or age*Vn */  
           cptcovprod--;  
           cutv(strb,stre,strc,'V');  
           Tvar[i]=atoi(stre);  
           cptcovage++;  
           Tage[cptcovage]=i;  
         }  
         else {  /* Age is not in the model */  
           cutv(strb,stre,strc,'V'); /* strc= Vn, stre is n*/  
           Tvar[i]=ncovcol+k1;  
           cutv(strb,strc,strd,'V'); /* strd was Vm, strc is m */  
           Tprod[k1]=i;  
           Tvard[k1][1]=atoi(strc); /* m*/  
           Tvard[k1][2]=atoi(stre); /* n */  
           Tvar[cptcovn+k2]=Tvard[k1][1];  
           Tvar[cptcovn+k2+1]=Tvard[k1][2];   
           for (k=1; k<=lastobs;k++)   
             covar[ncovcol+k1][k]=covar[atoi(stre)][k]*covar[atoi(strc)][k];  
           k1++;  
           k2=k2+2;  
         }  
       }  
       else { /* no more sum */  
         /*printf("d=%s c=%s b=%s\n", strd,strc,strb);*/  
        /*  scanf("%d",i);*/  
       cutv(strd,strc,strb,'V');  
       Tvar[i]=atoi(strc);  
       }  
       strcpy(modelsav,stra);    
       /*printf("a=%s b=%s sav=%s\n", stra,strb,modelsav);  
         scanf("%d",i);*/  
     } /* end of loop + */  
   } /* end model */  
     
   /*The number n of Vn is stored in Tvar. cptcovage =number of age covariate. Tage gives the position of age. cptcovprod= number of products.  
     If model=V1+V1*age then Tvar[1]=1 Tvar[2]=1 cptcovage=1 Tage[1]=2 cptcovprod=0*/  
   
   /* printf("tvar1=%d tvar2=%d tvar3=%d cptcovage=%d Tage=%d",Tvar[1],Tvar[2],Tvar[3],cptcovage,Tage[1]);  
   printf("cptcovprod=%d ", cptcovprod);  
   fprintf(ficlog,"cptcovprod=%d ", cptcovprod);  
   
   scanf("%d ",i);  
   fclose(fic);*/  
   
     /*  if(mle==1){*/  
   if (weightopt != 1) { /* Maximisation without weights*/  
     for(i=1;i<=n;i++) weight[i]=1.0;  
   }  
     /*-calculation of age at interview from date of interview and age at death -*/  
   agev=matrix(1,maxwav,1,imx);  
   
   for (i=1; i<=imx; i++) {  
     for(m=2; (m<= maxwav); m++) {  
       if (((int)mint[m][i]== 99) && (s[m][i] <= nlstate)){  
         anint[m][i]=9999;  
         s[m][i]=-1;  
       }  
       if((int)moisdc[i]==99 && (int)andc[i]==9999 && s[m][i]>nlstate){  
         printf("Error! Date of death (month %2d and year %4d) of individual %d on line %d was unknown, you must set an arbitrary year of death or he/she is skipped and results are biased\n",(int)moisdc[i],(int)andc[i],num[i],i);  
         fprintf(ficlog,"Error! Date of death (month %2d and year %4d) of individual %d on line %d was unknown, you must set an arbitrary year of death or he/she is skipped and results are biased\n",(int)moisdc[i],(int)andc[i],num[i],i);  
         s[m][i]=-1;  
       }  
       if((int)moisdc[i]==99 && (int)andc[i]!=9999 && s[m][i]>nlstate){  
         printf("Error! Month of death of individual %d on line %d was unknown %2d, you should set it otherwise the information on the death is skipped and results are biased.\n",num[i],i,(int)moisdc[i]);   
         fprintf(ficlog,"Error! Month of death of individual %d on line %d was unknown %f, you should set it otherwise the information on the death is skipped and results are biased.\n",num[i],i,moisdc[i]);   
         s[m][i]=-1; /* We prefer to skip it (and to skip it in version 0.8a1 too */  
       }  
     }  
   }  
   
   for (i=1; i<=imx; i++)  {  
     agedc[i]=(moisdc[i]/12.+andc[i])-(moisnais[i]/12.+annais[i]);  
     for(m=firstpass; (m<= lastpass); m++){  
       if(s[m][i] >0){  
         if (s[m][i] >= nlstate+1) {  
           if(agedc[i]>0)  
             if((int)moisdc[i]!=99 && (int)andc[i]!=9999)  
               agev[m][i]=agedc[i];  
           /*if(moisdc[i]==99 && andc[i]==9999) s[m][i]=-1;*/  
             else {  
               if ((int)andc[i]!=9999){  
                 printf("Warning negative age at death: %d line:%d\n",num[i],i);  
                 fprintf(ficlog,"Warning negative age at death: %d line:%d\n",num[i],i);  
                 agev[m][i]=-1;  
               }  
             }  
         }  
         else if(s[m][i] !=9){ /* Standard case, age in fractional  
                                  years but with the precision of a  
                                  month */  
           agev[m][i]=(mint[m][i]/12.+1./24.+anint[m][i])-(moisnais[i]/12.+1./24.+annais[i]);  
           if((int)mint[m][i]==99 || (int)anint[m][i]==9999)  
             agev[m][i]=1;  
           else if(agev[m][i] <agemin){   
             agemin=agev[m][i];  
             /*printf(" Min anint[%d][%d]=%.2f annais[%d]=%.2f, agemin=%.2f\n",m,i,anint[m][i], i,annais[i], agemin);*/  
           }  
           else if(agev[m][i] >agemax){  
             agemax=agev[m][i];  
             /* printf(" anint[%d][%d]=%.0f annais[%d]=%.0f, agemax=%.0f\n",m,i,anint[m][i], i,annais[i], agemax);*/  
           }  
           /*agev[m][i]=anint[m][i]-annais[i];*/  
           /*     agev[m][i] = age[i]+2*m;*/  
         }  
         else { /* =9 */  
           agev[m][i]=1;  
           s[m][i]=-1;  
         }  
       }  
       else /*= 0 Unknown */  
         agev[m][i]=1;  
     }  
       
   }  
   for (i=1; i<=imx; i++)  {  
     for(m=firstpass; (m<=lastpass); m++){  
       if (s[m][i] > (nlstate+ndeath)) {  
         printf("Error: on wave %d of individual %d status %d > (nlstate+ndeath)=(%d+%d)=%d\n",m,i,s[m][i],nlstate, ndeath, nlstate+ndeath);       
         fprintf(ficlog,"Error: on wave %d of individual %d status %d > (nlstate+ndeath)=(%d+%d)=%d\n",m,i,s[m][i],nlstate, ndeath, nlstate+ndeath);       
         goto end;  
       }  
     }  
   }  
   
   /*for (i=1; i<=imx; i++){  
   for (m=firstpass; (m<lastpass); m++){  
      printf("%d %d %.lf %d %d\n", num[i],(covar[1][i]),agev[m][i],s[m][i],s[m+1][i]);  
 }  
   
 }*/  
   
   printf("Total number of individuals= %d, Agemin = %.2f, Agemax= %.2f\n\n", imx, agemin, agemax);  
   fprintf(ficlog,"Total number of individuals= %d, Agemin = %.2f, Agemax= %.2f\n\n", imx, agemin, agemax);   
   
   free_vector(severity,1,maxwav);  
   free_imatrix(outcome,1,maxwav+1,1,n);  
   free_vector(moisnais,1,n);  
   free_vector(annais,1,n);  
   /* free_matrix(mint,1,maxwav,1,n);  
      free_matrix(anint,1,maxwav,1,n);*/  
   free_vector(moisdc,1,n);  
   free_vector(andc,1,n);  
   
      
   wav=ivector(1,imx);  
   dh=imatrix(1,lastpass-firstpass+1,1,imx);  
   bh=imatrix(1,lastpass-firstpass+1,1,imx);  
   mw=imatrix(1,lastpass-firstpass+1,1,imx);  
      
   /* Concatenates waves */  
   concatwav(wav, dh, bh, mw, s, agedc, agev,  firstpass, lastpass, imx, nlstate, stepm);  
   
   /* Routine tricode is to calculate cptcoveff (real number of unique covariates) and to associate covariable number and modality */  
   
   Tcode=ivector(1,100);                                        
   nbcode=imatrix(0,NCOVMAX,0,NCOVMAX);     /* Reads comments: lines beginning with '#' */
   ncodemax[1]=1;    numlinepar=0;
   if (cptcovn > 0) tricode(Tvar,nbcode,imx);    /* Is it a BOM UTF-8 Windows file? */
           /* First parameter line */
   codtab=imatrix(1,100,1,10); /* Cross tabulation to get the order of     while(fgets(line, MAXLINE, ficpar)) {
                                  the estimations*/      noffset=0;
   h=0;      if( line[0] == (char)0xEF && line[1] == (char)0xBB) /* EF BB BF */
   m=pow(2,cptcoveff);      {
          noffset=noffset+3;
   for(k=1;k<=cptcoveff; k++){        printf("# File is an UTF8 Bom.\n"); // 0xBF
     for(i=1; i <=(m/pow(2,k));i++){      }
       for(j=1; j <= ncodemax[k]; j++){  /*    else if( line[0] == (char)0xFE && line[1] == (char)0xFF)*/
         for(cpt=1; cpt <=(m/pow(2,cptcoveff+1-k)); cpt++){      else if( line[0] == (char)0xFF && line[1] == (char)0xFE)
           h++;      {
           if (h>m) h=1;codtab[h][k]=j;codtab[h][Tvar[k]]=j;        noffset=noffset+2;
           /*  printf("h=%d k=%d j=%d codtab[h][k]=%d tvar[k]=%d \n",h, k,j,codtab[h][k],Tvar[k]);*/        printf("# File is an UTF16BE BOM file\n");
         }       }
       }      else if( line[0] == 0 && line[1] == 0)
       {
         if( line[2] == (char)0xFE && line[3] == (char)0xFF){
           noffset=noffset+4;
           printf("# File is an UTF16BE BOM file\n");
         }
       } else{
         ;/*printf(" Not a BOM file\n");*/
       }
     
       /* If line starts with a # it is a comment */
       if (line[noffset] == '#') {
         numlinepar++;
         fputs(line,stdout);
         fputs(line,ficparo);
         fputs(line,ficres);
         fputs(line,ficlog);
         continue;
       }else
         break;
     }
     if((num_filled=sscanf(line,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\n", \
                           title, datafile, &lastobs, &firstpass,&lastpass)) !=EOF){
       if (num_filled != 5) {
         printf("Should be 5 parameters\n");
         fprintf(ficlog,"Should be 5 parameters\n");
       }
       numlinepar++;
       printf("title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\n", title, datafile, lastobs, firstpass,lastpass);
       fprintf(ficparo,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\n", title, datafile, lastobs, firstpass,lastpass);
       fprintf(ficres,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\n", title, datafile, lastobs, firstpass,lastpass);
       fprintf(ficlog,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\n", title, datafile, lastobs, firstpass,lastpass);
     }
     /* Second parameter line */
     while(fgets(line, MAXLINE, ficpar)) {
       /* while(fscanf(ficpar,"%[^\n]", line)) { */
       /* If line starts with a # it is a comment. Strangely fgets reads the EOL and fputs doesn't */
       if (line[0] == '#') {
         numlinepar++;
         printf("%s",line);
         fprintf(ficres,"%s",line);
         fprintf(ficparo,"%s",line);
         fprintf(ficlog,"%s",line);
         continue;
       }else
         break;
     }
     if((num_filled=sscanf(line,"ftol=%lf stepm=%d ncovcol=%d nqv=%d ntv=%d nqtv=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\n", \
                           &ftol, &stepm, &ncovcol, &nqv, &ntv, &nqtv, &nlstate, &ndeath, &maxwav, &mle, &weightopt)) !=EOF){
       if (num_filled != 11) {
         printf("Not 11 parameters, for example:ftol=1.e-8 stepm=12 ncovcol=2 nqv=1 ntv=2 nqtv=1  nlstate=2 ndeath=1 maxwav=3 mle=1 weight=1\n");
         printf("but line=%s\n",line);
         fprintf(ficlog,"Not 11 parameters, for example:ftol=1.e-8 stepm=12 ncovcol=2 nqv=1 ntv=2 nqtv=1  nlstate=2 ndeath=1 maxwav=3 mle=1 weight=1\n");
         fprintf(ficlog,"but line=%s\n",line);
       }
       if( lastpass > maxwav){
         printf("Error (lastpass = %d) > (maxwav = %d)\n",lastpass, maxwav);
         fprintf(ficlog,"Error (lastpass = %d) > (maxwav = %d)\n",lastpass, maxwav);
         fflush(ficlog);
         goto end;
     }      }
   }         printf("ftol=%e stepm=%d ncovcol=%d nqv=%d ntv=%d nqtv=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\n",ftol, stepm, ncovcol, nqv, ntv, nqtv, nlstate, ndeath, maxwav, mle, weightopt);
   /* printf("codtab[1][2]=%d codtab[2][2]=%d",codtab[1][2],codtab[2][2]);       fprintf(ficparo,"ftol=%e stepm=%d ncovcol=%d nqv=%d ntv=%d nqtv=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\n",ftol, stepm, ncovcol, nqv, ntv, nqtv, nlstate, ndeath, maxwav, mle, weightopt);
      codtab[1][2]=1;codtab[2][2]=2; */      fprintf(ficres,"ftol=%e stepm=%d ncovcol=%d nqv=%d ntv=%d nqtv=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\n",ftol, stepm, ncovcol, nqv, ntv, nqtv, nlstate, ndeath, maxwav, 0, weightopt);
   /* for(i=1; i <=m ;i++){       fprintf(ficlog,"ftol=%e stepm=%d ncovcol=%d nqv=%d ntv=%d nqtv=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\n",ftol, stepm, ncovcol, nqv, ntv, nqtv, nlstate, ndeath, maxwav, mle, weightopt);
      for(k=1; k <=cptcovn; k++){    }
      printf("i=%d k=%d %d %d ",i,k,codtab[i][k], cptcoveff);    /* ftolpl=6*ftol*1.e5; /\* 6.e-3 make convergences in less than 80 loops for the prevalence limit *\/ */
      }    /*ftolpl=6.e-4; *//* 6.e-3 make convergences in less than 80 loops for the prevalence limit */
      printf("\n");    /* Third parameter line */
      }    while(fgets(line, MAXLINE, ficpar)) {
      scanf("%d",i);*/      /* If line starts with a # it is a comment */
           if (line[0] == '#') {
   /* Calculates basic frequencies. Computes observed prevalence at single age        numlinepar++;
      and prints on file fileres'p'. */        printf("%s",line);
   freqsummary(fileres, agemin, agemax, s, agev, nlstate, imx,Tvaraff,nbcode, ncodemax,mint,anint);        fprintf(ficres,"%s",line);
         fprintf(ficparo,"%s",line);
         fprintf(ficlog,"%s",line);
         continue;
       }else
         break;
     }
     if((num_filled=sscanf(line,"model=1+age%[^.\n]", model)) !=EOF){
       if (num_filled != 1){
         printf("ERROR %d: Model should be at minimum 'model=1+age+' instead of '%s'\n",num_filled, line);
         fprintf(ficlog,"ERROR %d: Model should be at minimum 'model=1+age+' instead of '%s'\n",num_filled, line);
         model[0]='\0';
         goto end;
       }
       else{
         if (model[0]=='+'){
           for(i=1; i<=strlen(model);i++)
             modeltemp[i-1]=model[i];
           strcpy(model,modeltemp); 
         }
       }
       /* printf(" model=1+age%s modeltemp= %s, model=%s\n",model, modeltemp, model);fflush(stdout); */
       printf("model=1+age+%s\n",model);fflush(stdout);
       fprintf(ficparo,"model=1+age+%s\n",model);fflush(stdout);
       fprintf(ficres,"model=1+age+%s\n",model);fflush(stdout);
       fprintf(ficlog,"model=1+age+%s\n",model);fflush(stdout);
     }
     /* fscanf(ficpar,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%lf stepm=%d ncovcol=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d model=1+age+%s\n",title, datafile, &lastobs, &firstpass,&lastpass,&ftol, &stepm, &ncovcol, &nlstate,&ndeath, &maxwav, &mle, &weightopt,model); */
     /* numlinepar=numlinepar+3; /\* In general *\/ */
     /* printf("title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%e stepm=%d ncovcol=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\nmodel=1+age+%s\n", title, datafile, lastobs, firstpass,lastpass,ftol, stepm, ncovcol, nlstate,ndeath, maxwav, mle, weightopt,model); */
     /* fprintf(ficparo,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%e stepm=%d ncovcol=%d nqv=%d ntv=%d nqtv=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\nmodel=1+age+%s.\n", title, datafile, lastobs, firstpass,lastpass,ftol,stepm,ncovcol, nqv, ntv, nqtv, nlstate,ndeath,maxwav, mle, weightopt,model); */
     /* fprintf(ficlog,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%e stepm=%d ncovcol=%d nqv=%d ntv=%d nqtv=%d nlstate=%d ndeath=%d maxwav=%d mle=%d weight=%d\nmodel=1+age+%s.\n", title, datafile, lastobs, firstpass,lastpass,ftol,stepm,ncovcol, nqv, ntv, nqtv, nlstate,ndeath,maxwav, mle, weightopt,model); */
     fflush(ficlog);
     /* if(model[0]=='#'|| model[0]== '\0'){ */
     if(model[0]=='#'){
       printf("Error in 'model' line: model should start with 'model=1+age+' and end without space \n \
    'model=1+age+' or 'model=1+age+V1.' or 'model=1+age+age*age+V1+V1*age' or \n \
    'model=1+age+V1+V2' or 'model=1+age+V1+V2+V1*V2' etc. \n");            \
       if(mle != -1){
         printf("Fix the model line and run imach with mle=-1 to get a correct template of the parameter vectors and subdiagonal covariance matrix.\n");
         exit(1);
       }
     }
     while((c=getc(ficpar))=='#' && c!= EOF){
       ungetc(c,ficpar);
       fgets(line, MAXLINE, ficpar);
       numlinepar++;
       if(line[1]=='q'){ /* This #q will quit imach (the answer is q) */
         z[0]=line[1];
       }
       /* printf("****line [1] = %c \n",line[1]); */
       fputs(line, stdout);
       //puts(line);
       fputs(line,ficparo);
       fputs(line,ficlog);
     }
     ungetc(c,ficpar);
   
     pmmij= matrix(1,nlstate+ndeath,1,nlstate+ndeath); /* creation */  
     oldms= matrix(1,nlstate+ndeath,1,nlstate+ndeath); /* creation */  
     newms= matrix(1,nlstate+ndeath,1,nlstate+ndeath); /* creation */  
     savms= matrix(1,nlstate+ndeath,1,nlstate+ndeath); /* creation */  
     oldm=oldms; newm=newms; savm=savms; /* Keeps fixed addresses to free */  
       
         
   /* For Powell, parameters are in a vector p[] starting at p[1]    covar=matrix(0,NCOVMAX,firstobs,lastobs);  /**< used in readdata */
      so we point p on param[1][1] so that p[1] maps on param[1][1][1] */    if(nqv>=1)coqvar=matrix(1,nqv,firstobs,lastobs);  /**< Fixed quantitative covariate */
   p=param[1][1]; /* *(*(*(param +1)+1)+0) */    if(nqtv>=1)cotqvar=ma3x(1,maxwav,1,nqtv,firstobs,lastobs);  /**< Time varying quantitative covariate */
     if(ntv+nqtv>=1)cotvar=ma3x(1,maxwav,1,ntv+nqtv,firstobs,lastobs);  /**< Time varying covariate (dummy and quantitative)*/
   if(mle>=1){ /* Could be 1 or 2 */    cptcovn=0; /*Number of covariates, i.e. number of '+' in model statement plus one, indepently of n in Vn*/
     mlikeli(ficres,p, npar, ncovmodel, nlstate, ftol, func);    /* v1+v2+v3+v2*v4+v5*age makes cptcovn = 5
        v1+v2*age+v2*v3 makes cptcovn = 3
     */
     if (strlen(model)>1) 
       ncovmodel=2+nbocc(model,'+')+1; /*Number of variables including intercept and age = cptcovn + intercept + age : v1+v2+v3+v2*v4+v5*age makes 5+2=7,age*age makes 3*/
     else
       ncovmodel=2; /* Constant and age */
     nforce= (nlstate+ndeath-1)*nlstate; /* Number of forces ij from state i to j */
     npar= nforce*ncovmodel; /* Number of parameters like aij*/
     if(npar >MAXPARM || nlstate >NLSTATEMAX || ndeath >NDEATHMAX || ncovmodel>NCOVMAX){
       printf("Too complex model for current IMaCh: npar=(nlstate+ndeath-1)*nlstate*ncovmodel=%d >= %d(MAXPARM) or nlstate=%d >= %d(NLSTATEMAX) or ndeath=%d >= %d(NDEATHMAX) or ncovmodel=(k+age+#of+signs)=%d(NCOVMAX) >= %d\n",npar, MAXPARM, nlstate, NLSTATEMAX, ndeath, NDEATHMAX, ncovmodel, NCOVMAX);
       fprintf(ficlog,"Too complex model for current IMaCh: %d >=%d(MAXPARM) or %d >=%d(NLSTATEMAX) or %d >=%d(NDEATHMAX) or %d(NCOVMAX) >=%d\n",npar, MAXPARM, nlstate, NLSTATEMAX, ndeath, NDEATHMAX, ncovmodel, NCOVMAX);
       fflush(stdout);
       fclose (ficlog);
       goto end;
   }    }
     delti3= ma3x(1,nlstate,1,nlstate+ndeath-1,1,ncovmodel);
     delti=delti3[1][1];
     /*delti=vector(1,npar); *//* Scale of each paramater (output from hesscov)*/
     if(mle==-1){ /* Print a wizard for help writing covariance matrix */
   /* We could also provide initial parameters values giving by simple logistic regression 
    * only one way, that is without matrix product. We will have nlstate maximizations */
         /* for(i=1;i<nlstate;i++){ */
         /*        /\*reducing xi for 1 to npar to 1 to ncovmodel; *\/ */
         /*    mlikeli(ficres,p, ncovmodel, ncovmodel, nlstate, ftol, funcnoprod); */
         /* } */
       prwizard(ncovmodel, nlstate, ndeath, model, ficparo);
       printf(" You chose mle=-1, look at file %s for a template of covariance matrix \n",filereso);
       fprintf(ficlog," You chose mle=-1, look at file %s for a template of covariance matrix \n",filereso);
       free_ma3x(delti3,1,nlstate,1, nlstate+ndeath-1,1,ncovmodel); 
       fclose (ficparo);
       fclose (ficlog);
       goto end;
       exit(0);
     }  else if(mle==-5) { /* Main Wizard */
       prwizard(ncovmodel, nlstate, ndeath, model, ficparo);
       printf(" You chose mle=-3, look at file %s for a template of covariance matrix \n",filereso);
       fprintf(ficlog," You chose mle=-3, look at file %s for a template of covariance matrix \n",filereso);
       param= ma3x(1,nlstate,1,nlstate+ndeath-1,1,ncovmodel);
       matcov=matrix(1,npar,1,npar);
       hess=matrix(1,npar,1,npar);
     }  else{ /* Begin of mle != -1 or -5 */
       /* Read guessed parameters */
       /* Reads comments: lines beginning with '#' */
       while((c=getc(ficpar))=='#' && c!= EOF){
         ungetc(c,ficpar);
         fgets(line, MAXLINE, ficpar);
         numlinepar++;
         fputs(line,stdout);
         fputs(line,ficparo);
         fputs(line,ficlog);
       }
       ungetc(c,ficpar);
           
   /*--------- results files --------------*/      param= ma3x(1,nlstate,1,nlstate+ndeath-1,1,ncovmodel);
   fprintf(ficres,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%e stepm=%d ncovcol=%d nlstate=%d ndeath=%d maxwav=%d mle= 0 weight=%d\nmodel=%s\n", title, datafile, lastobs, firstpass,lastpass,ftol, stepm, ncovcol, nlstate, ndeath, maxwav, weightopt,model);      paramstart= ma3x(1,nlstate,1,nlstate+ndeath-1,1,ncovmodel);
         for(i=1; i <=nlstate; i++){
         j=0;
   jk=1;        for(jj=1; jj <=nlstate+ndeath; jj++){
   fprintf(ficres,"# Parameters nlstate*nlstate*ncov a12*1 + b12 * age + ...\n");          if(jj==i) continue;
   printf("# Parameters nlstate*nlstate*ncov a12*1 + b12 * age + ...\n");          j++;
   fprintf(ficlog,"# Parameters nlstate*nlstate*ncov a12*1 + b12 * age + ...\n");          while((c=getc(ficpar))=='#' && c!= EOF){
   for(i=1,jk=1; i <=nlstate; i++){            ungetc(c,ficpar);
     for(k=1; k <=(nlstate+ndeath); k++){            fgets(line, MAXLINE, ficpar);
       if (k != i)             numlinepar++;
         {            fputs(line,stdout);
           printf("%d%d ",i,k);            fputs(line,ficparo);
           fprintf(ficlog,"%d%d ",i,k);            fputs(line,ficlog);
           fprintf(ficres,"%1d%1d ",i,k);          }
           for(j=1; j <=ncovmodel; j++){          ungetc(c,ficpar);
             printf("%f ",p[jk]);          fscanf(ficpar,"%1d%1d",&i1,&j1);
             fprintf(ficlog,"%f ",p[jk]);          if ((i1 != i) || (j1 != jj)){
             fprintf(ficres,"%f ",p[jk]);            printf("Error in line parameters number %d, %1d%1d instead of %1d%1d \n \
             jk++;   It might be a problem of design; if ncovcol and the model are correct\n \
   run imach with mle=-1 to get a correct template of the parameter file.\n",numlinepar, i,j, i1, j1);
             exit(1);
           }
           fprintf(ficparo,"%1d%1d",i1,j1);
           if(mle==1)
             printf("%1d%1d",i,jj);
           fprintf(ficlog,"%1d%1d",i,jj);
           for(k=1; k<=ncovmodel;k++){
             fscanf(ficpar," %lf",&param[i][j][k]);
             if(mle==1){
               printf(" %lf",param[i][j][k]);
               fprintf(ficlog," %lf",param[i][j][k]);
           }            }
             else
               fprintf(ficlog," %lf",param[i][j][k]);
             fprintf(ficparo," %lf",param[i][j][k]);
           }
           fscanf(ficpar,"\n");
           numlinepar++;
           if(mle==1)
           printf("\n");            printf("\n");
           fprintf(ficlog,"\n");          fprintf(ficlog,"\n");
           fprintf(ficres,"\n");          fprintf(ficparo,"\n");
         }        }
       }  
       fflush(ficlog);
       
       /* Reads parameters values */
       p=param[1][1];
       pstart=paramstart[1][1];
       
       /* Reads comments: lines beginning with '#' */
       while((c=getc(ficpar))=='#' && c!= EOF){
         ungetc(c,ficpar);
         fgets(line, MAXLINE, ficpar);
         numlinepar++;
         fputs(line,stdout);
         fputs(line,ficparo);
         fputs(line,ficlog);
     }      }
   }      ungetc(c,ficpar);
   if(mle!=0){  
     /* Computing hessian and covariance matrix */      for(i=1; i <=nlstate; i++){
     ftolhess=ftol; /* Usually correct */        for(j=1; j <=nlstate+ndeath-1; j++){
     hesscov(matcov, p, npar, delti, ftolhess, func);          fscanf(ficpar,"%1d%1d",&i1,&j1);
   }          if ( (i1-i) * (j1-j) != 0){
   fprintf(ficres,"# Scales (for hessian or gradient estimation)\n");            printf("Error in line parameters number %d, %1d%1d instead of %1d%1d \n",numlinepar, i,j, i1, j1);
   printf("# Scales (for hessian or gradient estimation)\n");            exit(1);
   fprintf(ficlog,"# Scales (for hessian or gradient estimation)\n");          }
   for(i=1,jk=1; i <=nlstate; i++){  
     for(j=1; j <=nlstate+ndeath; j++){  
       if (j!=i) {  
         fprintf(ficres,"%1d%1d",i,j);  
         printf("%1d%1d",i,j);          printf("%1d%1d",i,j);
         fprintf(ficlog,"%1d%1d",i,j);          fprintf(ficparo,"%1d%1d",i1,j1);
           fprintf(ficlog,"%1d%1d",i1,j1);
         for(k=1; k<=ncovmodel;k++){          for(k=1; k<=ncovmodel;k++){
           printf(" %.5e",delti[jk]);            fscanf(ficpar,"%le",&delti3[i][j][k]);
           fprintf(ficlog," %.5e",delti[jk]);            printf(" %le",delti3[i][j][k]);
           fprintf(ficres," %.5e",delti[jk]);            fprintf(ficparo," %le",delti3[i][j][k]);
           jk++;            fprintf(ficlog," %le",delti3[i][j][k]);
         }          }
           fscanf(ficpar,"\n");
           numlinepar++;
         printf("\n");          printf("\n");
           fprintf(ficparo,"\n");
         fprintf(ficlog,"\n");          fprintf(ficlog,"\n");
         fprintf(ficres,"\n");  
       }        }
     }      }
   }      fflush(ficlog);
          
   fprintf(ficres,"# Covariance matrix \n# 121 Var(a12)\n# 122 Cov(b12,a12) Var(b12)\n#   ...\n# 232 Cov(b23,a12)  Cov(b23,b12) ... Var (b23)\n");      /* Reads covariance matrix */
   if(mle==1)      delti=delti3[1][1];
     printf("# Covariance matrix \n# 121 Var(a12)\n# 122 Cov(b12,a12) Var(b12)\n#   ...\n# 232 Cov(b23,a12)  Cov(b23,b12) ... Var (b23)\n");                  
   fprintf(ficlog,"# Covariance matrix \n# 121 Var(a12)\n# 122 Cov(b12,a12) Var(b12)\n#   ...\n# 232 Cov(b23,a12)  Cov(b23,b12) ... Var (b23)\n");                  
   for(i=1,k=1;i<=npar;i++){      /* free_ma3x(delti3,1,nlstate,1,nlstate+ndeath-1,1,ncovmodel); */ /* Hasn't to to freed here otherwise delti is no more allocated */
     /*  if (k>nlstate) k=1;                  
         i1=(i-1)/(ncovmodel*nlstate)+1;       /* Reads comments: lines beginning with '#' */
         fprintf(ficres,"%s%d%d",alph[k],i1,tab[i]);      while((c=getc(ficpar))=='#' && c!= EOF){
         printf("%s%d%d",alph[k],i1,tab[i]);        ungetc(c,ficpar);
     */        fgets(line, MAXLINE, ficpar);
     fprintf(ficres,"%3d",i);        numlinepar++;
     if(mle==1)        fputs(line,stdout);
       printf("%3d",i);        fputs(line,ficparo);
     fprintf(ficlog,"%3d",i);        fputs(line,ficlog);
     for(j=1; j<=i;j++){      }
       fprintf(ficres," %.5e",matcov[i][j]);      ungetc(c,ficpar);
                   
       matcov=matrix(1,npar,1,npar);
       hess=matrix(1,npar,1,npar);
       for(i=1; i <=npar; i++)
         for(j=1; j <=npar; j++) matcov[i][j]=0.;
                   
       /* Scans npar lines */
       for(i=1; i <=npar; i++){
         count=fscanf(ficpar,"%1d%1d%d",&i1,&j1,&jk);
         if(count != 3){
           printf("Error! Error in parameter file %s at line %d after line starting with %1d%1d%1d\n\
   This is probably because your covariance matrix doesn't \n  contain exactly %d lines corresponding to your model line '1+age+%s'.\n\
   Please run with mle=-1 to get a correct covariance matrix.\n",optionfile,numlinepar, i1,j1,jk, npar, model);
           fprintf(ficlog,"Error! Error in parameter file %s at line %d after line starting with %1d%1d%1d\n\
   This is probably because your covariance matrix doesn't \n  contain exactly %d lines corresponding to your model line '1+age+%s'.\n\
   Please run with mle=-1 to get a correct covariance matrix.\n",optionfile,numlinepar, i1,j1,jk, npar, model);
           exit(1);
         }else{
           if(mle==1)
             printf("%1d%1d%d",i1,j1,jk);
         }
         fprintf(ficlog,"%1d%1d%d",i1,j1,jk);
         fprintf(ficparo,"%1d%1d%d",i1,j1,jk);
         for(j=1; j <=i; j++){
           fscanf(ficpar," %le",&matcov[i][j]);
           if(mle==1){
             printf(" %.5le",matcov[i][j]);
           }
           fprintf(ficlog," %.5le",matcov[i][j]);
           fprintf(ficparo," %.5le",matcov[i][j]);
         }
         fscanf(ficpar,"\n");
         numlinepar++;
       if(mle==1)        if(mle==1)
         printf(" %.5e",matcov[i][j]);                                  printf("\n");
       fprintf(ficlog," %.5e",matcov[i][j]);        fprintf(ficlog,"\n");
         fprintf(ficparo,"\n");
     }      }
     fprintf(ficres,"\n");      /* End of read covariance matrix npar lines */
       for(i=1; i <=npar; i++)
         for(j=i+1;j<=npar;j++)
           matcov[i][j]=matcov[j][i];
       
     if(mle==1)      if(mle==1)
       printf("\n");        printf("\n");
     fprintf(ficlog,"\n");      fprintf(ficlog,"\n");
     k++;      
   }      fflush(ficlog);
          
   while((c=getc(ficpar))=='#' && c!= EOF){    }    /* End of mle != -3 */
     ungetc(c,ficpar);    
     fgets(line, MAXLINE, ficpar);    /*  Main data
     puts(line);     */
     fputs(line,ficparo);    nobs=lastobs-firstobs+1; /* was = lastobs;*/
   }    /* num=lvector(1,n); */
   ungetc(c,ficpar);    /* moisnais=vector(1,n); */
     /* annais=vector(1,n); */
     /* moisdc=vector(1,n); */
     /* andc=vector(1,n); */
     /* weight=vector(1,n); */
     /* agedc=vector(1,n); */
     /* cod=ivector(1,n); */
     /* for(i=1;i<=n;i++){ */
     num=lvector(firstobs,lastobs);
     moisnais=vector(firstobs,lastobs);
     annais=vector(firstobs,lastobs);
     moisdc=vector(firstobs,lastobs);
     andc=vector(firstobs,lastobs);
     weight=vector(firstobs,lastobs);
     agedc=vector(firstobs,lastobs);
     cod=ivector(firstobs,lastobs);
     for(i=firstobs;i<=lastobs;i++){
       num[i]=0;
       moisnais[i]=0;
       annais[i]=0;
       moisdc[i]=0;
       andc[i]=0;
       agedc[i]=0;
       cod[i]=0;
       weight[i]=1.0; /* Equal weights, 1 by default */
     }
     mint=matrix(1,maxwav,firstobs,lastobs);
     anint=matrix(1,maxwav,firstobs,lastobs);
     s=imatrix(1,maxwav+1,firstobs,lastobs); /* s[i][j] health state for wave i and individual j */
     printf("BUG ncovmodel=%d NCOVMAX=%d 2**ncovmodel=%f BUG\n",ncovmodel,NCOVMAX,pow(2,ncovmodel));
     tab=ivector(1,NCOVMAX);
     ncodemax=ivector(1,NCOVMAX); /* Number of code per covariate; if O and 1 only, 2**ncov; V1+V2+V3+V4=>16 */
     ncodemaxwundef=ivector(1,NCOVMAX); /* Number of code per covariate; if - 1 O and 1 only, 2**ncov; V1+V2+V3+V4=>16 */
   
     /* Reads data from file datafile */
     if (readdata(datafile, firstobs, lastobs, &imx)==1)
       goto end;
   
     /* Calculation of the number of parameters from char model */
     /*    modelsav=V2+V1+V4+age*V3 strb=age*V3 stra=V2+V1+V4 
           k=4 (age*V3) Tvar[k=4]= 3 (from V3) Tag[cptcovage=1]=4
           k=3 V4 Tvar[k=3]= 4 (from V4)
           k=2 V1 Tvar[k=2]= 1 (from V1)
           k=1 Tvar[1]=2 (from V2)
     */
     
     Tvar=ivector(1,NCOVMAX); /* Was 15 changed to NCOVMAX. */
     TvarsDind=ivector(1,NCOVMAX); /*  */
     TnsdVar=ivector(1,NCOVMAX); /*  */
       /* for(i=1; i<=NCOVMAX;i++) TnsdVar[i]=3; */
     TvarsD=ivector(1,NCOVMAX); /*  */
     TvarsQind=ivector(1,NCOVMAX); /*  */
     TvarsQ=ivector(1,NCOVMAX); /*  */
     TvarF=ivector(1,NCOVMAX); /*  */
     TvarFind=ivector(1,NCOVMAX); /*  */
     TvarV=ivector(1,NCOVMAX); /*  */
     TvarVind=ivector(1,NCOVMAX); /*  */
     TvarA=ivector(1,NCOVMAX); /*  */
     TvarAind=ivector(1,NCOVMAX); /*  */
     TvarFD=ivector(1,NCOVMAX); /*  */
     TvarFDind=ivector(1,NCOVMAX); /*  */
     TvarFQ=ivector(1,NCOVMAX); /*  */
     TvarFQind=ivector(1,NCOVMAX); /*  */
     TvarVD=ivector(1,NCOVMAX); /*  */
     TvarVDind=ivector(1,NCOVMAX); /*  */
     TvarVQ=ivector(1,NCOVMAX); /*  */
     TvarVQind=ivector(1,NCOVMAX); /*  */
   
     Tvalsel=vector(1,NCOVMAX); /*  */
     Tvarsel=ivector(1,NCOVMAX); /*  */
     Typevar=ivector(-1,NCOVMAX); /* -1 to 2 */
     Fixed=ivector(-1,NCOVMAX); /* -1 to 3 */
     Dummy=ivector(-1,NCOVMAX); /* -1 to 3 */
     /*  V2+V1+V4+age*V3 is a model with 4 covariates (3 plus signs). 
         For each model-covariate stores the data-covariate id. Tvar[1]=2, Tvar[2]=1, Tvar[3]=4, 
         Tvar[4=age*V3] is 3 and 'age' is recorded in Tage.
     */
     /* For model-covariate k tells which data-covariate to use but
       because this model-covariate is a construction we invent a new column
       ncovcol + k1
       If already ncovcol=4 and model=V2+V1+V1*V4+age*V3
       Tvar[3=V1*V4]=4+1 etc */
     Tprod=ivector(1,NCOVMAX); /* Gives the k position of the k1 product */
     Tposprod=ivector(1,NCOVMAX); /* Gives the k1 product from the k position */
     /* Tprod[k1=1]=3(=V1*V4) for V2+V1+V1*V4+age*V3
        if  V2+V1+V1*V4+age*V3+V3*V2   TProd[k1=2]=5 (V3*V2)
        Tposprod[k]=k1 , Tposprod[3]=1, Tposprod[5]=2 
     */
     Tvaraff=ivector(1,NCOVMAX); /* Unclear */
     Tvard=imatrix(1,NCOVMAX,1,2); /* n=Tvard[k1][1]  and m=Tvard[k1][2] gives the couple n,m of the k1 th product Vn*Vm
                               * For V3*V2 (in V2+V1+V1*V4+age*V3+V3*V2), V3*V2 position is 2nd. 
                               * Tvard[k1=2][1]=3 (V3) Tvard[k1=2][2]=2(V2) */
     Tvardk=imatrix(1,NCOVMAX,1,2);
     Tage=ivector(1,NCOVMAX); /* Gives the covariate id of covariates associated with age: V2 + V1 + age*V4 + V3*age
                            4 covariates (3 plus signs)
                            Tage[1=V3*age]= 4; Tage[2=age*V4] = 3
                              */  
     for(i=1;i<NCOVMAX;i++)
       Tage[i]=0;
     Tmodelind=ivector(1,NCOVMAX);/** gives the k model position of an
                                   * individual dummy, fixed or varying:
                                   * Tmodelind[Tvaraff[3]]=9,Tvaraff[1]@9={4,
                                   * 3, 1, 0, 0, 0, 0, 0, 0},
                                   * model=V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1 , 
                                   * V1 df, V2 qf, V3 & V4 dv, V5 qv
                                   * Tmodelind[1]@9={9,0,3,2,}*/
     TmodelInvind=ivector(1,NCOVMAX); /* TmodelInvind=Tvar[k]- ncovcol-nqv={5-2-1=2,*/
     TmodelInvQind=ivector(1,NCOVMAX);/** gives the k model position of an
                                   * individual quantitative, fixed or varying:
                                   * Tmodelqind[1]=1,Tvaraff[1]@9={4,
                                   * 3, 1, 0, 0, 0, 0, 0, 0},
                                   * model=V5+V4+V3+V4*V3+V5*age+V2+V1*V2+V1*age+V1*/
   /* Main decodemodel */
   
   
     if(decodemodel(model, lastobs) == 1) /* In order to get Tvar[k] V4+V3+V5 p Tvar[1]@3  = {4, 3, 5}*/
       goto end;
   
   estepm=0;    if((double)(lastobs-imx)/(double)imx > 1.10){
   fscanf(ficpar,"agemin=%lf agemax=%lf bage=%lf fage=%lf estepm=%d\n",&ageminpar,&agemaxpar, &bage, &fage, &estepm);      nbwarn++;
   if (estepm==0 || estepm < stepm) estepm=stepm;      printf("Warning: The value of parameter lastobs=%d is big compared to the \n  effective number of cases imx=%d, please adjust, \n  otherwise you are allocating more memory than necessary.\n",lastobs, imx); 
   if (fage <= 2) {      fprintf(ficlog,"Warning: The value of parameter lastobs=%d is big compared to the \n  effective number of cases imx=%d, please adjust, \n  otherwise you are allocating more memory than necessary.\n",lastobs, imx); 
     bage = ageminpar;  
     fage = agemaxpar;  
   }    }
          /*  if(mle==1){*/
   fprintf(ficres,"# agemin agemax for life expectancy, bage fage (if mle==0 ie no data nor Max likelihood).\n");    if (weightopt != 1) { /* Maximisation without weights. We can have weights different from 1 but want no weight*/
   fprintf(ficres,"agemin=%.0f agemax=%.0f bage=%.0f fage=%.0f estepm=%d\n",ageminpar,agemaxpar,bage,fage, estepm);      for(i=1;i<=imx;i++) weight[i]=1.0; /* changed to imx */
   fprintf(ficparo,"agemin=%.0f agemax=%.0f bage=%.0f fage=%.0f estepm=%d\n",ageminpar,agemaxpar,bage,fage, estepm);  
      
   while((c=getc(ficpar))=='#' && c!= EOF){  
     ungetc(c,ficpar);  
     fgets(line, MAXLINE, ficpar);  
     puts(line);  
     fputs(line,ficparo);  
   }    }
   ungetc(c,ficpar);  
       /*-calculation of age at interview from date of interview and age at death -*/
     agev=matrix(1,maxwav,1,imx);
   
     if(calandcheckages(imx, maxwav, &agemin, &agemax, &nberr, &nbwarn) == 1)
       goto end;
   
   
     agegomp=(int)agemin;
     free_vector(moisnais,firstobs,lastobs);
     free_vector(annais,firstobs,lastobs);
     /* free_matrix(mint,1,maxwav,1,n);
        free_matrix(anint,1,maxwav,1,n);*/
     /* free_vector(moisdc,1,n); */
     /* free_vector(andc,1,n); */
     /* */
       
   fscanf(ficpar,"begin-prev-date=%lf/%lf/%lf end-prev-date=%lf/%lf/%lf mov_average=%d\n",&jprev1, &mprev1,&anprev1,&jprev2, &mprev2,&anprev2,&mobilav);    wav=ivector(1,imx);
   fprintf(ficparo,"begin-prev-date=%.lf/%.lf/%.lf end-prev-date=%.lf/%.lf/%.lf mov_average=%d\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,mobilav);    /* dh=imatrix(1,lastpass-firstpass+1,1,imx); */
   fprintf(ficres,"begin-prev-date=%.lf/%.lf/%.lf end-prev-date=%.lf/%.lf/%.lf mov_average=%d\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,mobilav);    /* bh=imatrix(1,lastpass-firstpass+1,1,imx); */
   printf("begin-prev-date=%.lf/%.lf/%.lf end-prev-date=%.lf/%.lf/%.lf mov_average=%d\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,mobilav);    /* mw=imatrix(1,lastpass-firstpass+1,1,imx); */
   fprintf(ficlog,"begin-prev-date=%.lf/%.lf/%.lf end-prev-date=%.lf/%.lf/%.lf mov_average=%d\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,mobilav);    dh=imatrix(1,lastpass-firstpass+2,1,imx); /* We are adding a wave if status is unknown at last wave but death occurs after last wave.*/
     bh=imatrix(1,lastpass-firstpass+2,1,imx);
     mw=imatrix(1,lastpass-firstpass+2,1,imx);
         
   while((c=getc(ficpar))=='#' && c!= EOF){    /* Concatenates waves */
     ungetc(c,ficpar);    /* Concatenates waves: wav[i] is the number of effective (useful waves) of individual i.
     fgets(line, MAXLINE, ficpar);       Death is a valid wave (if date is known).
     puts(line);       mw[mi][i] is the number of (mi=1 to wav[i]) effective wave out of mi of individual i
     fputs(line,ficparo);       dh[m][i] or dh[mw[mi][i]][i] is the delay between two effective waves m=mw[mi][i]
   }       and mw[mi+1][i]. dh depends on stepm.
   ungetc(c,ficpar);    */
    
   
   dateprev1=anprev1+(mprev1-1)/12.+(jprev1-1)/365.;    concatwav(wav, dh, bh, mw, s, agedc, agev,  firstpass, lastpass, imx, nlstate, stepm);
   dateprev2=anprev2+(mprev2-1)/12.+(jprev2-1)/365.;    /* Concatenates waves */
    
     free_vector(moisdc,firstobs,lastobs);
     free_vector(andc,firstobs,lastobs);
   
   fscanf(ficpar,"pop_based=%d\n",&popbased);    /* Routine tricode is to calculate cptcoveff (real number of unique covariates) and to associate covariable number and modality */
   fprintf(ficparo,"pop_based=%d\n",popbased);       nbcode=imatrix(0,NCOVMAX,0,NCOVMAX); 
   fprintf(ficres,"pop_based=%d\n",popbased);       ncodemax[1]=1;
       Ndum =ivector(-1,NCOVMAX);  
   while((c=getc(ficpar))=='#' && c!= EOF){    cptcoveff=0;
     ungetc(c,ficpar);    if (ncovmodel-nagesqr > 2 ){ /* That is if covariate other than cst, age and age*age */
     fgets(line, MAXLINE, ficpar);      tricode(&cptcoveff,Tvar,nbcode,imx, Ndum); /**< Fills nbcode[Tvar[j]][l]; as well as calculate cptcoveff or number of total effective dummy covariates*/
     puts(line);    }
     fputs(line,ficparo);    
   }    ncovcombmax=pow(2,cptcoveff);
   ungetc(c,ficpar);    invalidvarcomb=ivector(1, ncovcombmax); 
     for(i=1;i<ncovcombmax;i++)
       invalidvarcomb[i]=0;
     
     /* Nbcode gives the value of the lth modality (currently 1 to 2) of jth covariate, in
        V2+V1*age, there are 3 covariates Tvar[2]=1 (V1).*/
     /* 1 to ncodemax[j] which is the maximum value of this jth covariate */
     
     /*  codtab=imatrix(1,100,1,10);*/ /* codtab[h,k]=( (h-1) - mod(k-1,2**(k-1) )/2**(k-1) */
     /*printf(" codtab[1,1],codtab[100,10]=%d,%d\n", codtab[1][1],codtabm(100,10));*/
     /* codtab gives the value 1 or 2 of the hth combination of k covariates (1 or 2).*/
     /* nbcode[Tvaraff[j]][codtabm(h,j)]) : if there are only 2 modalities for a covariate j, 
      * codtabm(h,j) gives its value classified at position h and nbcode gives how it is coded 
      * (currently 0 or 1) in the data.
      * In a loop on h=1 to 2**k, and a loop on j (=1 to k), we get the value of 
      * corresponding modality (h,j).
      */
   
   fscanf(ficpar,"prevforecast=%d starting-proj-date=%lf/%lf/%lf final-proj-date=%lf/%lf/%lf mobil_average=%d\n",&prevfcast,&jproj1,&mproj1,&anproj1,&jproj2,&mproj2,&anproj2,&mobilavproj);    h=0;
   fprintf(ficparo,"prevforecast=%d starting-proj-date=%.lf/%.lf/%.lf final-proj-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevfcast,jproj1,mproj1,anproj1,jproj2,mproj2,anproj2,mobilavproj);    /*if (cptcovn > 0) */
   printf("prevforecast=%d starting-proj-date=%.lf/%.lf/%.lf final-proj-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevfcast,jproj1,mproj1,anproj1,jproj2,mproj2,anproj2,mobilavproj);    m=pow(2,cptcoveff);
   fprintf(ficlog,"prevforecast=%d starting-proj-date=%.lf/%.lf/%.lf final-proj-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevfcast,jproj1,mproj1,anproj1,jproj2,mproj2,anproj2,mobilavproj);   
   fprintf(ficres,"prevforecast=%d starting-proj-date=%.lf/%.lf/%.lf final-proj-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevfcast,jproj1,mproj1,anproj1,jproj2,mproj2,anproj2,mobilavproj);            /**< codtab(h,k)  k   = codtab[h,k]=( (h-1) - mod(k-1,2**(k-1) )/2**(k-1) + 1
   /* day and month of proj2 are not used but only year anproj2.*/             * For k=4 covariates, h goes from 1 to m=2**k
              * codtabm(h,k)=  (1 & (h-1) >> (k-1)) + 1;
              * #define codtabm(h,k)  (1 & (h-1) >> (k-1))+1
              *     h\k   1     2     3     4   *  h-1\k-1  4  3  2  1          
              *______________________________   *______________________
              *     1 i=1 1 i=1 1 i=1 1 i=1 1   *     0     0  0  0  0 
              *     2     2     1     1     1   *     1     0  0  0  1 
              *     3 i=2 1     2     1     1   *     2     0  0  1  0 
              *     4     2     2     1     1   *     3     0  0  1  1 
              *     5 i=3 1 i=2 1     2     1   *     4     0  1  0  0 
              *     6     2     1     2     1   *     5     0  1  0  1 
              *     7 i=4 1     2     2     1   *     6     0  1  1  0 
              *     8     2     2     2     1   *     7     0  1  1  1 
              *     9 i=5 1 i=3 1 i=2 1     2   *     8     1  0  0  0 
              *    10     2     1     1     2   *     9     1  0  0  1 
              *    11 i=6 1     2     1     2   *    10     1  0  1  0 
              *    12     2     2     1     2   *    11     1  0  1  1 
              *    13 i=7 1 i=4 1     2     2   *    12     1  1  0  0  
              *    14     2     1     2     2   *    13     1  1  0  1 
              *    15 i=8 1     2     2     2   *    14     1  1  1  0 
              *    16     2     2     2     2   *    15     1  1  1  1          
              */                                     
     /* How to do the opposite? From combination h (=1 to 2**k) how to get the value on the covariates? */
        /* from h=5 and m, we get then number of covariates k=log(m)/log(2)=4
        * and the value of each covariate?
        * V1=1, V2=1, V3=2, V4=1 ?
        * h-1=4 and 4 is 0100 or reverse 0010, and +1 is 1121 ok.
        * h=6, 6-1=5, 5 is 0101, 1010, 2121, V1=2nd, V2=1st, V3=2nd, V4=1st.
        * In order to get the real value in the data, we use nbcode
        * nbcode[Tvar[3][2nd]]=1 and nbcode[Tvar[4][1]]=0
        * We are keeping this crazy system in order to be able (in the future?) 
        * to have more than 2 values (0 or 1) for a covariate.
        * #define codtabm(h,k)  (1 & (h-1) >> (k-1))+1
        * h=6, k=2? h-1=5=0101, reverse 1010, +1=2121, k=2nd position: value is 1: codtabm(6,2)=1
        *              bbbbbbbb
        *              76543210     
        *   h-1        00000101 (6-1=5)
        *(h-1)>>(k-1)= 00000010 >> (2-1) = 1 right shift
        *           &
        *     1        00000001 (1)
        *              00000000        = 1 & ((h-1) >> (k-1))
        *          +1= 00000001 =1 
        *
        * h=14, k=3 => h'=h-1=13, k'=k-1=2
        *          h'      1101 =2^3+2^2+0x2^1+2^0
        *    >>k'            11
        *          &   00000001
        *            = 00000001
        *      +1    = 00000010=2    =  codtabm(14,3)   
        * Reverse h=6 and m=16?
        * cptcoveff=log(16)/log(2)=4 covariate: 6-1=5=0101 reversed=1010 +1=2121 =>V1=2, V2=1, V3=2, V4=1.
        * for (j=1 to cptcoveff) Vj=decodtabm(j,h,cptcoveff)
        * decodtabm(h,j,cptcoveff)= (((h-1) >> (j-1)) & 1) +1 
        * decodtabm(h,j,cptcoveff)= (h <= (1<<cptcoveff)?(((h-1) >> (j-1)) & 1) +1 : -1)
        * V3=decodtabm(14,3,2**4)=2
        *          h'=13   1101 =2^3+2^2+0x2^1+2^0
        *(h-1) >> (j-1)    0011 =13 >> 2
        *          &1 000000001
        *           = 000000001
        *         +1= 000000010 =2
        *                  2211
        *                  V1=1+1, V2=0+1, V3=1+1, V4=1+1
        *                  V3=2
                    * codtabm and decodtabm are identical
        */
   
   while((c=getc(ficpar))=='#' && c!= EOF){  
     ungetc(c,ficpar);  
     fgets(line, MAXLINE, ficpar);  
     puts(line);  
     fputs(line,ficparo);  
   }  
   ungetc(c,ficpar);  
   
   fscanf(ficpar,"popforecast=%d popfile=%s popfiledate=%lf/%lf/%lf last-popfiledate=%lf/%lf/%lf\n",&popforecast,popfile,&jpyram,&mpyram,&anpyram,&jpyram1,&mpyram1,&anpyram1);   free_ivector(Ndum,-1,NCOVMAX);
   fprintf(ficparo,"popforecast=%d popfile=%s popfiledate=%.lf/%.lf/%.lf last-popfiledate=%.lf/%.lf/%.lf\n",popforecast,popfile,jpyram,mpyram,anpyram,jpyram1,mpyram1,anpyram1);  
   fprintf(ficres,"popforecast=%d popfile=%s popfiledate=%.lf/%.lf/%.lf last-popfiledate=%.lf/%.lf/%.lf\n",popforecast,popfile,jpyram,mpyram,anpyram,jpyram1,mpyram1,anpyram1);  
   
   freqsummary(fileres, agemin, agemax, s, agev, nlstate, imx,Tvaraff,nbcode, ncodemax,mint,anint);  
   /*,dateprev1,dateprev2,jprev1, mprev1,anprev1,jprev2, mprev2,anprev2);*/  
   
   /*------------ gnuplot -------------*/      
     /* Initialisation of ----------- gnuplot -------------*/
   strcpy(optionfilegnuplot,optionfilefiname);    strcpy(optionfilegnuplot,optionfilefiname);
     if(mle==-3)
       strcat(optionfilegnuplot,"-MORT_");
   strcat(optionfilegnuplot,".gp");    strcat(optionfilegnuplot,".gp");
   
   if((ficgp=fopen(optionfilegnuplot,"w"))==NULL) {    if((ficgp=fopen(optionfilegnuplot,"w"))==NULL) {
     printf("Problem with file %s",optionfilegnuplot);      printf("Problem with file %s",optionfilegnuplot);
   }    }
   else{    else{
     fprintf(ficgp,"\n# %s\n", version);       fprintf(ficgp,"\n# IMaCh-%s\n", version); 
     fprintf(ficgp,"# %s\n", optionfilegnuplot);       fprintf(ficgp,"# %s\n", optionfilegnuplot); 
     fprintf(ficgp,"set missing 'NaNq'\n");      //fprintf(ficgp,"set missing 'NaNq'\n");
       fprintf(ficgp,"set datafile missing 'NaNq'\n");
   }    }
   fclose(ficgp);    /*  fclose(ficgp);*/
   printinggnuplot(fileres, ageminpar,agemaxpar,fage, pathc,p);  
   /*--------- index.htm --------*/  
   
   strcpy(optionfilehtm,optionfile);  
     /* Initialisation of --------- index.htm --------*/
   
     strcpy(optionfilehtm,optionfilefiname); /* Main html file */
     if(mle==-3)
       strcat(optionfilehtm,"-MORT_");
   strcat(optionfilehtm,".htm");    strcat(optionfilehtm,".htm");
   if((fichtm=fopen(optionfilehtm,"w"))==NULL)    {    if((fichtm=fopen(optionfilehtm,"w"))==NULL)    {
     printf("Problem with %s \n",optionfilehtm), exit(0);      printf("Problem with %s \n",optionfilehtm);
       exit(0);
     }
   
     strcpy(optionfilehtmcov,optionfilefiname); /* Only for matrix of covariance */
     strcat(optionfilehtmcov,"-cov.htm");
     if((fichtmcov=fopen(optionfilehtmcov,"w"))==NULL)    {
       printf("Problem with %s \n",optionfilehtmcov), exit(0);
   }    }
     else{
     fprintf(fichtmcov,"<html><head>\n<title>IMaCh Cov %s</title></head>\n <body><font size=\"2\">%s <br> %s</font> \
   <hr size=\"2\" color=\"#EC5E5E\"> \n\
   Title=%s <br>Datafile=%s Firstpass=%d Lastpass=%d Stepm=%d Weight=%d Model=1+age+%s<br>\n",\
             optionfilehtmcov,version,fullversion,title,datafile,firstpass,lastpass,stepm, weightopt, model);
     }
   
     fprintf(fichtm,"<html><head>\n<meta charset=\"utf-8\"/><meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n\
   <title>IMaCh %s</title></head>\n\
    <body><font size=\"7\"><a href=http:/euroreves.ined.fr/imach>IMaCh for Interpolated Markov Chain</a> </font><br>\n\
   <font size=\"3\">Sponsored by Copyright (C)  2002-2015 <a href=http://www.ined.fr>INED</a>\
   -EUROREVES-Institut de longeÌviteÌ-2013-2022-Japan Society for the Promotion of Sciences 日本学術振興会 \
   (<a href=https://www.jsps.go.jp/english/e-grants/>Grant-in-Aid for Scientific Research 25293121</a>) - \
   <a href=https://software.intel.com/en-us>Intel Software 2015-2018</a></font><br> \n", optionfilehtm);
     
     fprintf(fichtm,"<hr size=\"2\" color=\"#EC5E5E\"> \n\
   <font size=\"2\">IMaCh-%s <br> %s</font> \
   <hr size=\"2\" color=\"#EC5E5E\"> \n\
   This file: <a href=\"%s\">%s</a>Title=%s <br>Datafile=<a href=\"%s\">%s</a> Firstpass=%d Lastpass=%d Stepm=%d Weight=%d Model=1+age+%s<br>\n\
   \n\
   <hr  size=\"2\" color=\"#EC5E5E\">\
    <ul><li><h4>Parameter files</h4>\n\
    - Parameter file: <a href=\"%s.%s\">%s.%s</a><br>\n\
    - Copy of the parameter file: <a href=\"o%s\">o%s</a><br>\n\
    - Log file of the run: <a href=\"%s\">%s</a><br>\n\
    - Gnuplot file name: <a href=\"%s\">%s</a><br>\n\
    - Date and time at start: %s</ul>\n",\
             version,fullversion,optionfilehtm,optionfilehtm,title,datafile,datafile,firstpass,lastpass,stepm, weightopt, model, \
             optionfilefiname,optionfilext,optionfilefiname,optionfilext,\
             fileres,fileres,\
             filelog,filelog,optionfilegnuplot,optionfilegnuplot,strstart);
     fflush(fichtm);
   
     strcpy(pathr,path);
     strcat(pathr,optionfilefiname);
   #ifdef WIN32
     _chdir(optionfilefiname); /* Move to directory named optionfile */
   #else
     chdir(optionfilefiname); /* Move to directory named optionfile */
   #endif
             
     
     /* Calculates basic frequencies. Computes observed prevalence at single age 
                    and for any valid combination of covariates
        and prints on file fileres'p'. */
     freqsummary(fileres, p, pstart, agemin, agemax, s, agev, nlstate, imx, Tvaraff, invalidvarcomb, nbcode, ncodemax,mint,anint,strstart, \
                 firstpass, lastpass,  stepm,  weightopt, model);
   
   fprintf(fichtm,"<body> <font size=\"2\">%s </font> <hr size=\"2\" color=\"#EC5E5E\"> \n    fprintf(fichtm,"\n");
 Title=%s <br>Datafile=%s Firstpass=%d Lastpass=%d Stepm=%d Weight=%d Model=%s<br>\n    fprintf(fichtm,"<h4>Parameter line 2</h4><ul><li>Tolerance for the convergence of the likelihood: ftol=%g \n<li>Interval for the elementary matrix (in month): stepm=%d",\
 \n            ftol, stepm);
 Total number of observations=%d <br>\n    fprintf(fichtm,"\n<li>Number of fixed dummy covariates: ncovcol=%d ", ncovcol);
 Youngest age at first (selected) pass %.2f, oldest age %.2f<br>\n    ncurrv=1;
 Interval (in months) between two waves: Min=%d Max=%d Mean=%.2lf<br>\n    for(i=ncurrv; i <=ncovcol; i++) fprintf(fichtm,"V%d ", i);
 <hr  size=\"2\" color=\"#EC5E5E\">    fprintf(fichtm,"\n<li> Number of fixed quantitative variables: nqv=%d ", nqv); 
  <ul><li><h4>Parameter files</h4>\n    ncurrv=i;
  - Copy of the parameter file: <a href=\"o%s\">o%s</a><br>\n    for(i=ncurrv; i <=ncurrv-1+nqv; i++) fprintf(fichtm,"V%d ", i);
  - Log file of the run: <a href=\"%s\">%s</a><br>\n    fprintf(fichtm,"\n<li> Number of time varying (wave varying) dummy covariates: ntv=%d ", ntv);
  - Gnuplot file name: <a href=\"%s\">%s</a></ul>\n",version,title,datafile,firstpass,lastpass,stepm, weightopt,model,imx,agemin,agemax,jmin,jmax,jmean,fileres,fileres,filelog,filelog,optionfilegnuplot,optionfilegnuplot);    ncurrv=i;
    fclose(fichtm);    for(i=ncurrv; i <=ncurrv-1+ntv; i++) fprintf(fichtm,"V%d ", i);
     fprintf(fichtm,"\n<li>Number of time varying  quantitative covariates: nqtv=%d ", nqtv);
     ncurrv=i;
     for(i=ncurrv; i <=ncurrv-1+nqtv; i++) fprintf(fichtm,"V%d ", i);
     fprintf(fichtm,"\n<li>Weights column \n<br>Number of alive states: nlstate=%d <br>Number of death states (not really implemented): ndeath=%d \n<li>Number of waves: maxwav=%d \n<li>Parameter for maximization (1), using parameter values (0), for design of parameters and variance-covariance matrix: mle=%d \n<li>Does the weight column be taken into account (1), or not (0): weight=%d</ul>\n", \
              nlstate, ndeath, maxwav, mle, weightopt);
   
     fprintf(fichtm,"<h4> Diagram of states <a href=\"%s_.svg\">%s_.svg</a></h4> \n\
   <img src=\"%s_.svg\">", subdirf2(optionfilefiname,"D_"),subdirf2(optionfilefiname,"D_"),subdirf2(optionfilefiname,"D_"));
   
     
     fprintf(fichtm,"\n<h4>Some descriptive statistics </h4>\n<br>Number of (used) observations=%d <br>\n\
   Youngest age at first (selected) pass %.2f, oldest age %.2f<br>\n\
   Interval (in months) between two waves: Min=%d Max=%d Mean=%.2lf<br>\n",\
     imx,agemin,agemax,jmin,jmax,jmean);
     pmmij= matrix(1,nlstate+ndeath,1,nlstate+ndeath); /* creation */
     oldms= matrix(1,nlstate+ndeath,1,nlstate+ndeath); /* creation */
     newms= matrix(1,nlstate+ndeath,1,nlstate+ndeath); /* creation */
     savms= matrix(1,nlstate+ndeath,1,nlstate+ndeath); /* creation */
     oldm=oldms; newm=newms; savm=savms; /* Keeps fixed addresses to free */
   
   printinghtml(fileres,title,datafile, firstpass, lastpass, stepm, weightopt,model,imx,jmin,jmax,jmean,rfileres,popforecast,estepm,jprev1,mprev1,anprev1,jprev2,mprev2,anprev2);    /* For Powell, parameters are in a vector p[] starting at p[1]
         so we point p on param[1][1] so that p[1] maps on param[1][1][1] */
   /*------------ free_vector  -------------*/    p=param[1][1]; /* *(*(*(param +1)+1)+0) */
   chdir(path);  
     globpr=0; /* To get the number ipmx of contributions and the sum of weights*/
     /* For mortality only */
     if (mle==-3){
       ximort=matrix(1,NDIM,1,NDIM); 
       for(i=1;i<=NDIM;i++)
         for(j=1;j<=NDIM;j++)
           ximort[i][j]=0.;
       /*     ximort=gsl_matrix_alloc(1,NDIM,1,NDIM); */
       cens=ivector(firstobs,lastobs);
       ageexmed=vector(firstobs,lastobs);
       agecens=vector(firstobs,lastobs);
       dcwave=ivector(firstobs,lastobs);
                   
       for (i=1; i<=imx; i++){
         dcwave[i]=-1;
         for (m=firstpass; m<=lastpass; m++)
           if (s[m][i]>nlstate) {
             dcwave[i]=m;
             /*    printf("i=%d j=%d s=%d dcwave=%d\n",i,j, s[j][i],dcwave[i]);*/
             break;
           }
       }
       
       for (i=1; i<=imx; i++) {
         if (wav[i]>0){
           ageexmed[i]=agev[mw[1][i]][i];
           j=wav[i];
           agecens[i]=1.; 
           
           if (ageexmed[i]> 1 && wav[i] > 0){
             agecens[i]=agev[mw[j][i]][i];
             cens[i]= 1;
           }else if (ageexmed[i]< 1) 
             cens[i]= -1;
           if (agedc[i]< AGESUP && agedc[i]>1 && dcwave[i]>firstpass && dcwave[i]<=lastpass)
             cens[i]=0 ;
         }
         else cens[i]=-1;
       }
       
       for (i=1;i<=NDIM;i++) {
         for (j=1;j<=NDIM;j++)
           ximort[i][j]=(i == j ? 1.0 : 0.0);
       }
       
       p[1]=0.0268; p[NDIM]=0.083;
       /* printf("%lf %lf", p[1], p[2]); */
       
       
   #ifdef GSL
       printf("GSL optimization\n");  fprintf(ficlog,"Powell\n");
   #else
       printf("Powell\n");  fprintf(ficlog,"Powell\n");
   #endif
       strcpy(filerespow,"POW-MORT_"); 
       strcat(filerespow,fileresu);
       if((ficrespow=fopen(filerespow,"w"))==NULL) {
         printf("Problem with resultfile: %s\n", filerespow);
         fprintf(ficlog,"Problem with resultfile: %s\n", filerespow);
       }
   #ifdef GSL
       fprintf(ficrespow,"# GSL optimization\n# iter -2*LL");
   #else
       fprintf(ficrespow,"# Powell\n# iter -2*LL");
   #endif
       /*  for (i=1;i<=nlstate;i++)
           for(j=1;j<=nlstate+ndeath;j++)
           if(j!=i)fprintf(ficrespow," p%1d%1d",i,j);
       */
       fprintf(ficrespow,"\n");
   #ifdef GSL
       /* gsl starts here */ 
       T = gsl_multimin_fminimizer_nmsimplex;
       gsl_multimin_fminimizer *sfm = NULL;
       gsl_vector *ss, *x;
       gsl_multimin_function minex_func;
   
       /* Initial vertex size vector */
       ss = gsl_vector_alloc (NDIM);
       
       if (ss == NULL){
         GSL_ERROR_VAL ("failed to allocate space for ss", GSL_ENOMEM, 0);
       }
       /* Set all step sizes to 1 */
       gsl_vector_set_all (ss, 0.001);
   
       /* Starting point */
       
       x = gsl_vector_alloc (NDIM);
       
       if (x == NULL){
         gsl_vector_free(ss);
         GSL_ERROR_VAL ("failed to allocate space for x", GSL_ENOMEM, 0);
       }
     
       /* Initialize method and iterate */
       /*     p[1]=0.0268; p[NDIM]=0.083; */
       /*     gsl_vector_set(x, 0, 0.0268); */
       /*     gsl_vector_set(x, 1, 0.083); */
       gsl_vector_set(x, 0, p[1]);
       gsl_vector_set(x, 1, p[2]);
   
       minex_func.f = &gompertz_f;
       minex_func.n = NDIM;
       minex_func.params = (void *)&p; /* ??? */
       
       sfm = gsl_multimin_fminimizer_alloc (T, NDIM);
       gsl_multimin_fminimizer_set (sfm, &minex_func, x, ss);
       
       printf("Iterations beginning .....\n\n");
       printf("Iter. #    Intercept       Slope     -Log Likelihood     Simplex size\n");
   
       iteri=0;
       while (rval == GSL_CONTINUE){
         iteri++;
         status = gsl_multimin_fminimizer_iterate(sfm);
         
         if (status) printf("error: %s\n", gsl_strerror (status));
         fflush(0);
         
         if (status) 
           break;
         
         rval = gsl_multimin_test_size (gsl_multimin_fminimizer_size (sfm), 1e-6);
         ssval = gsl_multimin_fminimizer_size (sfm);
         
         if (rval == GSL_SUCCESS)
           printf ("converged to a local maximum at\n");
         
         printf("%5d ", iteri);
         for (it = 0; it < NDIM; it++){
           printf ("%10.5f ", gsl_vector_get (sfm->x, it));
         }
         printf("f() = %-10.5f ssize = %.7f\n", sfm->fval, ssval);
       }
       
       printf("\n\n Please note: Program should be run many times with varying starting points to detemine global maximum\n\n");
       
       gsl_vector_free(x); /* initial values */
       gsl_vector_free(ss); /* inital step size */
       for (it=0; it<NDIM; it++){
         p[it+1]=gsl_vector_get(sfm->x,it);
         fprintf(ficrespow," %.12lf", p[it]);
       }
       gsl_multimin_fminimizer_free (sfm); /* p *(sfm.x.data) et p *(sfm.x.data+1)  */
   #endif
   #ifdef POWELL
        powell(p,ximort,NDIM,ftol,&iter,&fret,gompertz);
   #endif  
       fclose(ficrespow);
       
       hesscov(matcov, hess, p, NDIM, delti, 1e-4, gompertz); 
   
       for(i=1; i <=NDIM; i++)
         for(j=i+1;j<=NDIM;j++)
                                   matcov[i][j]=matcov[j][i];
       
       printf("\nCovariance matrix\n ");
       fprintf(ficlog,"\nCovariance matrix\n ");
       for(i=1; i <=NDIM; i++) {
         for(j=1;j<=NDIM;j++){ 
                                   printf("%f ",matcov[i][j]);
                                   fprintf(ficlog,"%f ",matcov[i][j]);
         }
         printf("\n ");  fprintf(ficlog,"\n ");
       }
       
       printf("iter=%d MLE=%f Eq=%lf*exp(%lf*(age-%d))\n",iter,-gompertz(p),p[1],p[2],agegomp);
       for (i=1;i<=NDIM;i++) {
         printf("%f [%f ; %f]\n",p[i],p[i]-2*sqrt(matcov[i][i]),p[i]+2*sqrt(matcov[i][i]));
         fprintf(ficlog,"%f [%f ; %f]\n",p[i],p[i]-2*sqrt(matcov[i][i]),p[i]+2*sqrt(matcov[i][i]));
       }
       lsurv=vector(agegomp,AGESUP);
       lpop=vector(agegomp,AGESUP);
       tpop=vector(agegomp,AGESUP);
       lsurv[agegomp]=100000;
       
       for (k=agegomp;k<=AGESUP;k++) {
         agemortsup=k;
         if (p[1]*exp(p[2]*(k-agegomp))>1) break;
       }
       
       for (k=agegomp;k<agemortsup;k++)
         lsurv[k+1]=lsurv[k]-lsurv[k]*(p[1]*exp(p[2]*(k-agegomp)));
       
       for (k=agegomp;k<agemortsup;k++){
         lpop[k]=(lsurv[k]+lsurv[k+1])/2.;
         sumlpop=sumlpop+lpop[k];
       }
       
       tpop[agegomp]=sumlpop;
       for (k=agegomp;k<(agemortsup-3);k++){
         /*  tpop[k+1]=2;*/
         tpop[k+1]=tpop[k]-lpop[k];
       }
       
       
       printf("\nAge   lx     qx    dx    Lx     Tx     e(x)\n");
       for (k=agegomp;k<(agemortsup-2);k++) 
         printf("%d %.0lf %lf %.0lf %.0lf %.0lf %lf\n",k,lsurv[k],p[1]*exp(p[2]*(k-agegomp)),(p[1]*exp(p[2]*(k-agegomp)))*lsurv[k],lpop[k],tpop[k],tpop[k]/lsurv[k]);
       
       
       replace_back_to_slash(pathc,pathcd); /* Even gnuplot wants a / */
                   ageminpar=50;
                   agemaxpar=100;
       if(ageminpar == AGEOVERFLOW ||agemaxpar == AGEOVERFLOW){
           printf("Warning! Error in gnuplot file with ageminpar %f or agemaxpar %f overflow\n\
   This is probably because your parameter file doesn't \n  contain the exact number of lines (or columns) corresponding to your model line.\n\
   Please run with mle=-1 to get a correct covariance matrix.\n",ageminpar,agemaxpar);
           fprintf(ficlog,"Warning! Error in gnuplot file with ageminpar %f or agemaxpar %f overflow\n\
   This is probably because your parameter file doesn't \n  contain the exact number of lines (or columns) corresponding to your model line.\n\
   Please run with mle=-1 to get a correct covariance matrix.\n",ageminpar,agemaxpar);
       }else{
                           printf("Warning! ageminpar %f and agemaxpar %f have been fixed because for simplification until it is fixed...\n\n",ageminpar,agemaxpar);
                           fprintf(ficlog,"Warning! ageminpar %f and agemaxpar %f have been fixed because for simplification until it is fixed...\n\n",ageminpar,agemaxpar);
         printinggnuplotmort(fileresu, optionfilefiname,ageminpar,agemaxpar,fage, pathc,p);
                   }
       printinghtmlmort(fileresu,title,datafile, firstpass, lastpass, \
                        stepm, weightopt,\
                        model,imx,p,matcov,agemortsup);
       
       free_vector(lsurv,agegomp,AGESUP);
       free_vector(lpop,agegomp,AGESUP);
       free_vector(tpop,agegomp,AGESUP);
       free_matrix(ximort,1,NDIM,1,NDIM);
       free_ivector(dcwave,firstobs,lastobs);
       free_vector(agecens,firstobs,lastobs);
       free_vector(ageexmed,firstobs,lastobs);
       free_ivector(cens,firstobs,lastobs);
   #ifdef GSL
   #endif
     } /* Endof if mle==-3 mortality only */
     /* Standard  */
     else{ /* For mle !=- 3, could be 0 or 1 or 4 etc. */
       globpr=0;/* Computes sum of likelihood for globpr=1 and funcone */
       /* Computes likelihood for initial parameters, uses funcone to compute gpimx and gsw */
       likelione(ficres, p, npar, nlstate, &globpr, &ipmx, &sw, &fretone, funcone); /* Prints the contributions to the likelihood */
       printf("First Likeli=%12.6f ipmx=%ld sw=%12.6f",fretone,ipmx,sw);
       for (k=1; k<=npar;k++)
         printf(" %d %8.5f",k,p[k]);
       printf("\n");
       if(mle>=1){ /* Could be 1 or 2, Real Maximization */
         /* mlikeli uses func not funcone */
         /* for(i=1;i<nlstate;i++){ */
         /*        /\*reducing xi for 1 to npar to 1 to ncovmodel; *\/ */
         /*    mlikeli(ficres,p, ncovmodel, ncovmodel, nlstate, ftol, funcnoprod); */
         /* } */
         mlikeli(ficres,p, npar, ncovmodel, nlstate, ftol, func);
       }
       if(mle==0) {/* No optimization, will print the likelihoods for the datafile */
         globpr=0;/* Computes sum of likelihood for globpr=1 and funcone */
         /* Computes likelihood for initial parameters, uses funcone to compute gpimx and gsw */
         likelione(ficres, p, npar, nlstate, &globpr, &ipmx, &sw, &fretone, funcone); /* Prints the contributions to the likelihood */
       }
       globpr=1; /* again, to print the individual contributions using computed gpimx and gsw */
       likelione(ficres, p, npar, nlstate, &globpr, &ipmx, &sw, &fretone, funcone); /* Prints the contributions to the likelihood */
       printf("Second Likeli=%12.6f ipmx=%ld sw=%12.6f",fretone,ipmx,sw);
             /* exit(0); */
       for (k=1; k<=npar;k++)
         printf(" %d %8.5f",k,p[k]);
       printf("\n");
       
       /*--------- results files --------------*/
       /* fprintf(ficres,"title=%s datafile=%s lastobs=%d firstpass=%d lastpass=%d\nftol=%e stepm=%d ncovcol=%d nqv=%d ntv=%d nqtv=%d nlstate=%d ndeath=%d maxwav=%d mle= 0 weight=%d\nmodel=1+age+%s.\n", title, datafile, lastobs, firstpass,lastpass,ftol, stepm, ncovcol, nqv, ntv, nqtv, nlstate, ndeath, maxwav, weightopt,model); */
       
       
       fprintf(ficres,"# Parameters nlstate*nlstate*ncov a12*1 + b12 * age + ...\n");
       printf("# Parameters nlstate*nlstate*ncov a12*1 + b12 * age + ...\n"); /* Printing model equation */
       fprintf(ficlog,"# Parameters nlstate*nlstate*ncov a12*1 + b12 * age + ...\n");
   
       printf("#model=  1      +     age ");
       fprintf(ficres,"#model=  1      +     age ");
       fprintf(ficlog,"#model=  1      +     age ");
       fprintf(fichtm,"\n<ul><li> model=1+age+%s\n \
   </ul>", model);
   
       fprintf(fichtm,"\n<table style=\"text-align:center; border: 1px solid\">\n");
       fprintf(fichtm, "<tr><th>Model=</th><th>1</th><th>+ age</th>");
       if(nagesqr==1){
         printf("  + age*age  ");
         fprintf(ficres,"  + age*age  ");
         fprintf(ficlog,"  + age*age  ");
         fprintf(fichtm, "<th>+ age*age</th>");
       }
       for(j=1;j <=ncovmodel-2;j++){
         if(Typevar[j]==0) {
           printf("  +      V%d  ",Tvar[j]);
           fprintf(ficres,"  +      V%d  ",Tvar[j]);
           fprintf(ficlog,"  +      V%d  ",Tvar[j]);
           fprintf(fichtm, "<th>+ V%d</th>",Tvar[j]);
         }else if(Typevar[j]==1) {
           printf("  +    V%d*age ",Tvar[j]);
           fprintf(ficres,"  +    V%d*age ",Tvar[j]);
           fprintf(ficlog,"  +    V%d*age ",Tvar[j]);
           fprintf(fichtm, "<th>+  V%d*age</th>",Tvar[j]);
         }else if(Typevar[j]==2) {
           printf("  +    V%d*V%d ",Tvard[Tposprod[j]][1],Tvard[Tposprod[j]][2]);
           fprintf(ficres,"  +    V%d*V%d ",Tvard[Tposprod[j]][1],Tvard[Tposprod[j]][2]);
           fprintf(ficlog,"  +    V%d*V%d ",Tvard[Tposprod[j]][1],Tvard[Tposprod[j]][2]);
           fprintf(fichtm, "<th>+  V%d*V%d</th>",Tvard[Tposprod[j]][1],Tvard[Tposprod[j]][2]);
         }
       }
       printf("\n");
       fprintf(ficres,"\n");
       fprintf(ficlog,"\n");
       fprintf(fichtm, "</tr>");
       fprintf(fichtm, "\n");
       
       
       for(i=1,jk=1; i <=nlstate; i++){
         for(k=1; k <=(nlstate+ndeath); k++){
           if (k != i) {
             fprintf(fichtm, "<tr>");
             printf("%d%d ",i,k);
             fprintf(ficlog,"%d%d ",i,k);
             fprintf(ficres,"%1d%1d ",i,k);
             fprintf(fichtm, "<td>%1d%1d</td>",i,k);
             for(j=1; j <=ncovmodel; j++){
               printf("%12.7f ",p[jk]);
               fprintf(ficlog,"%12.7f ",p[jk]);
               fprintf(ficres,"%12.7f ",p[jk]);
               fprintf(fichtm, "<td>%12.7f</td>",p[jk]);
               jk++; 
             }
             printf("\n");
             fprintf(ficlog,"\n");
             fprintf(ficres,"\n");
             fprintf(fichtm, "</tr>\n");
           }
         }
       }
       /* fprintf(fichtm,"</tr>\n"); */
       fprintf(fichtm,"</table>\n");
       fprintf(fichtm, "\n");
   
       if(mle != 0){
         /* Computing hessian and covariance matrix only at a peak of the Likelihood, that is after optimization */
         ftolhess=ftol; /* Usually correct */
         hesscov(matcov, hess, p, npar, delti, ftolhess, func);
         printf("Parameters and 95%% confidence intervals\n W is simply the result of the division of the parameter by the square root of covariance of the parameter.\n And Wald-based confidence intervals plus and minus 1.96 * W .\n But be careful that parameters are highly correlated because incidence of disability is highly correlated to incidence of recovery.\n It might be better to visualize the covariance matrix. See the page 'Matrix of variance-covariance of one-step probabilities' and its graphs.\n");
         fprintf(ficlog, "Parameters, Wald tests and Wald-based confidence intervals\n W is simply the result of the division of the parameter by the square root of covariance of the parameter.\n And Wald-based confidence intervals plus and minus 1.96 * W \n  It might be better to visualize the covariance matrix. See the page 'Matrix of variance-covariance of one-step probabilities' and its graphs.\n");
         fprintf(fichtm, "\n<p>The Wald test results are output only if the maximimzation of the Likelihood is performed (mle=1)\n</br>Parameters, Wald tests and Wald-based confidence intervals\n</br> W is simply the result of the division of the parameter by the square root of covariance of the parameter.\n</br> And Wald-based confidence intervals plus and minus 1.96 * W \n </br> It might be better to visualize the covariance matrix. See the page '<a href=\"%s\">Matrix of variance-covariance of one-step probabilities and its graphs</a>'.\n</br>",optionfilehtmcov);
         fprintf(fichtm,"\n<table style=\"text-align:center; border: 1px solid\">");
         fprintf(fichtm, "\n<tr><th>Model=</th><th>1</th><th>+ age</th>");
         if(nagesqr==1){
           printf("  + age*age  ");
           fprintf(ficres,"  + age*age  ");
           fprintf(ficlog,"  + age*age  ");
           fprintf(fichtm, "<th>+ age*age</th>");
         }
         for(j=1;j <=ncovmodel-2;j++){
           if(Typevar[j]==0) {
             printf("  +      V%d  ",Tvar[j]);
             fprintf(fichtm, "<th>+ V%d</th>",Tvar[j]);
           }else if(Typevar[j]==1) {
             printf("  +    V%d*age ",Tvar[j]);
             fprintf(fichtm, "<th>+  V%d*age</th>",Tvar[j]);
           }else if(Typevar[j]==2) {
             fprintf(fichtm, "<th>+  V%d*V%d</th>",Tvard[Tposprod[j]][1],Tvard[Tposprod[j]][2]);
           }
         }
         fprintf(fichtm, "</tr>\n");
     
   free_ivector(wav,1,imx);        for(i=1,jk=1; i <=nlstate; i++){
   free_imatrix(dh,1,lastpass-firstpass+1,1,imx);          for(k=1; k <=(nlstate+ndeath); k++){
   free_imatrix(bh,1,lastpass-firstpass+1,1,imx);            if (k != i) {
   free_imatrix(mw,1,lastpass-firstpass+1,1,imx);                 fprintf(fichtm, "<tr valign=top>");
   free_ivector(num,1,n);              printf("%d%d ",i,k);
   free_vector(agedc,1,n);              fprintf(ficlog,"%d%d ",i,k);
   /*free_matrix(covar,0,NCOVMAX,1,n);*/              fprintf(fichtm, "<td>%1d%1d</td>",i,k);
   /*free_matrix(covar,1,NCOVMAX,1,n);*/              for(j=1; j <=ncovmodel; j++){
   fclose(ficparo);                wald=p[jk]/sqrt(matcov[jk][jk]);
   fclose(ficres);                printf("%12.7f(%12.7f) W=%8.3f CI=[%12.7f ; %12.7f] ",p[jk],sqrt(matcov[jk][jk]), p[jk]/sqrt(matcov[jk][jk]), p[jk]-1.96*sqrt(matcov[jk][jk]),p[jk]+1.96*sqrt(matcov[jk][jk]));
                 fprintf(ficlog,"%12.7f(%12.7f) W=%8.3f CI=[%12.7f ; %12.7f] ",p[jk],sqrt(matcov[jk][jk]), p[jk]/sqrt(matcov[jk][jk]), p[jk]-1.96*sqrt(matcov[jk][jk]),p[jk]+1.96*sqrt(matcov[jk][jk]));
                 if(fabs(wald) > 1.96){
   /*--------------- Prevalence limit  (stable prevalence) --------------*/                  fprintf(fichtm, "<td><b>%12.7f</b></br> (%12.7f)</br>",p[jk],sqrt(matcov[jk][jk]));
                   }else{
   strcpy(filerespl,"pl");                  fprintf(fichtm, "<td>%12.7f (%12.7f)</br>",p[jk],sqrt(matcov[jk][jk]));
   strcat(filerespl,fileres);                }
   if((ficrespl=fopen(filerespl,"w"))==NULL) {                fprintf(fichtm,"W=%8.3f</br>",wald);
     printf("Problem with stable prevalence resultfile: %s\n", filerespl);goto end;                fprintf(fichtm,"[%12.7f;%12.7f]</br></td>", p[jk]-1.96*sqrt(matcov[jk][jk]),p[jk]+1.96*sqrt(matcov[jk][jk]));
     fprintf(ficlog,"Problem with stable prevalence resultfile: %s\n", filerespl);goto end;                jk++; 
   }              }
   printf("Computing stable prevalence: result on file '%s' \n", filerespl);              printf("\n");
   fprintf(ficlog,"Computing stable prevalence: result on file '%s' \n", filerespl);              fprintf(ficlog,"\n");
   fprintf(ficrespl,"#Stable prevalence \n");              fprintf(fichtm, "</tr>\n");
   fprintf(ficrespl,"#Age ");            }
   for(i=1; i<=nlstate;i++) fprintf(ficrespl,"%d-%d ",i,i);          }
   fprintf(ficrespl,"\n");  
     
   prlim=matrix(1,nlstate,1,nlstate);  
   
   agebase=ageminpar;  
   agelim=agemaxpar;  
   ftolpl=1.e-10;  
   i1=cptcoveff;  
   if (cptcovn < 1){i1=1;}  
   
   for(cptcov=1,k=0;cptcov<=i1;cptcov++){  
     for(cptcod=1;cptcod<=ncodemax[cptcov];cptcod++){  
       k=k+1;  
       /*printf("cptcov=%d cptcod=%d codtab=%d nbcode=%d\n",cptcov, cptcod,Tcode[cptcode],codtab[cptcod][cptcov]);*/  
       fprintf(ficrespl,"\n#******");  
       printf("\n#******");  
       fprintf(ficlog,"\n#******");  
       for(j=1;j<=cptcoveff;j++) {  
         fprintf(ficrespl," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);  
         printf(" V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);  
         fprintf(ficlog," V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);  
       }        }
       fprintf(ficrespl,"******\n");      } /* end of hesscov and Wald tests */
       printf("******\n");      fprintf(fichtm,"</table>\n");
       fprintf(ficlog,"******\n");      
               /*  */
       for (age=agebase; age<=agelim; age++){      fprintf(ficres,"# Scales (for hessian or gradient estimation)\n");
         prevalim(prlim, nlstate, p, age, oldm, savm,ftolpl,k);      printf("# Scales (for hessian or gradient estimation)\n");
         fprintf(ficrespl,"%.0f ",age );      fprintf(ficlog,"# Scales (for hessian or gradient estimation)\n");
         for(j=1;j<=cptcoveff;j++)      for(i=1,jk=1; i <=nlstate; i++){
           fprintf(ficrespl,"%d %d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);        for(j=1; j <=nlstate+ndeath; j++){
         for(i=1; i<=nlstate;i++)          if (j!=i) {
           fprintf(ficrespl," %.5f", prlim[i][i]);            fprintf(ficres,"%1d%1d",i,j);
         fprintf(ficrespl,"\n");            printf("%1d%1d",i,j);
             fprintf(ficlog,"%1d%1d",i,j);
             for(k=1; k<=ncovmodel;k++){
               printf(" %.5e",delti[jk]);
               fprintf(ficlog," %.5e",delti[jk]);
               fprintf(ficres," %.5e",delti[jk]);
               jk++;
             }
             printf("\n");
             fprintf(ficlog,"\n");
             fprintf(ficres,"\n");
           }
       }        }
     }      }
   }      
   fclose(ficrespl);      fprintf(ficres,"# Covariance matrix \n# 121 Var(a12)\n# 122 Cov(b12,a12) Var(b12)\n#   ...\n# 232 Cov(b23,a12)  Cov(b23,b12) ... Var (b23)\n");
       if(mle >= 1) /* To big for the screen */
   /*------------- h Pij x at various ages ------------*/        printf("# Covariance matrix \n# 121 Var(a12)\n# 122 Cov(b12,a12) Var(b12)\n#   ...\n# 232 Cov(b23,a12)  Cov(b23,b12) ... Var (b23)\n");
         fprintf(ficlog,"# Covariance matrix \n# 121 Var(a12)\n# 122 Cov(b12,a12) Var(b12)\n#   ...\n# 232 Cov(b23,a12)  Cov(b23,b12) ... Var (b23)\n");
   strcpy(filerespij,"pij");  strcat(filerespij,fileres);      /* # 121 Var(a12)\n\ */
   if((ficrespij=fopen(filerespij,"w"))==NULL) {      /* # 122 Cov(b12,a12) Var(b12)\n\ */
     printf("Problem with Pij resultfile: %s\n", filerespij);goto end;      /* # 131 Cov(a13,a12) Cov(a13,b12, Var(a13)\n\ */
     fprintf(ficlog,"Problem with Pij resultfile: %s\n", filerespij);goto end;      /* # 132 Cov(b13,a12) Cov(b13,b12, Cov(b13,a13) Var(b13)\n\ */
   }      /* # 212 Cov(a21,a12) Cov(a21,b12, Cov(a21,a13) Cov(a21,b13) Var(a21)\n\ */
   printf("Computing pij: result on file '%s' \n", filerespij);      /* # 212 Cov(b21,a12) Cov(b21,b12, Cov(b21,a13) Cov(b21,b13) Cov(b21,a21) Var(b21)\n\ */
   fprintf(ficlog,"Computing pij: result on file '%s' \n", filerespij);      /* # 232 Cov(a23,a12) Cov(a23,b12, Cov(a23,a13) Cov(a23,b13) Cov(a23,a21) Cov(a23,b21) Var(a23)\n\ */
         /* # 232 Cov(b23,a12) Cov(b23,b12) ... Var (b23)\n" */
   stepsize=(int) (stepm+YEARM-1)/YEARM;      
   /*if (stepm<=24) stepsize=2;*/      
       /* Just to have a covariance matrix which will be more understandable
   agelim=AGESUP;         even is we still don't want to manage dictionary of variables
   hstepm=stepsize*YEARM; /* Every year of age */      */
   hstepm=hstepm/stepm; /* Typically 2 years, = 2/6 months = 4 */       for(itimes=1;itimes<=2;itimes++){
         jj=0;
   /* hstepm=1;   aff par mois*/        for(i=1; i <=nlstate; i++){
           for(j=1; j <=nlstate+ndeath; j++){
             if(j==i) continue;
             for(k=1; k<=ncovmodel;k++){
               jj++;
               ca[0]= k+'a'-1;ca[1]='\0';
               if(itimes==1){
                 if(mle>=1)
                   printf("#%1d%1d%d",i,j,k);
                 fprintf(ficlog,"#%1d%1d%d",i,j,k);
                 fprintf(ficres,"#%1d%1d%d",i,j,k);
               }else{
                 if(mle>=1)
                   printf("%1d%1d%d",i,j,k);
                 fprintf(ficlog,"%1d%1d%d",i,j,k);
                 fprintf(ficres,"%1d%1d%d",i,j,k);
               }
               ll=0;
               for(li=1;li <=nlstate; li++){
                 for(lj=1;lj <=nlstate+ndeath; lj++){
                   if(lj==li) continue;
                   for(lk=1;lk<=ncovmodel;lk++){
                     ll++;
                     if(ll<=jj){
                       cb[0]= lk +'a'-1;cb[1]='\0';
                       if(ll<jj){
                         if(itimes==1){
                           if(mle>=1)
                             printf(" Cov(%s%1d%1d,%s%1d%1d)",ca,i,j,cb, li,lj);
                           fprintf(ficlog," Cov(%s%1d%1d,%s%1d%1d)",ca,i,j,cb, li,lj);
                           fprintf(ficres," Cov(%s%1d%1d,%s%1d%1d)",ca,i,j,cb, li,lj);
                         }else{
                           if(mle>=1)
                             printf(" %.5e",matcov[jj][ll]); 
                           fprintf(ficlog," %.5e",matcov[jj][ll]); 
                           fprintf(ficres," %.5e",matcov[jj][ll]); 
                         }
                       }else{
                         if(itimes==1){
                           if(mle>=1)
                             printf(" Var(%s%1d%1d)",ca,i,j);
                           fprintf(ficlog," Var(%s%1d%1d)",ca,i,j);
                           fprintf(ficres," Var(%s%1d%1d)",ca,i,j);
                         }else{
                           if(mle>=1)
                             printf(" %.7e",matcov[jj][ll]); 
                           fprintf(ficlog," %.7e",matcov[jj][ll]); 
                           fprintf(ficres," %.7e",matcov[jj][ll]); 
                         }
                       }
                     }
                   } /* end lk */
                 } /* end lj */
               } /* end li */
               if(mle>=1)
                 printf("\n");
               fprintf(ficlog,"\n");
               fprintf(ficres,"\n");
               numlinepar++;
             } /* end k*/
           } /*end j */
         } /* end i */
       } /* end itimes */
       
       fflush(ficlog);
       fflush(ficres);
       while(fgets(line, MAXLINE, ficpar)) {
         /* If line starts with a # it is a comment */
         if (line[0] == '#') {
           numlinepar++;
           fputs(line,stdout);
           fputs(line,ficparo);
           fputs(line,ficlog);
           fputs(line,ficres);
           continue;
         }else
           break;
       }
       
       /* while((c=getc(ficpar))=='#' && c!= EOF){ */
       /*   ungetc(c,ficpar); */
       /*   fgets(line, MAXLINE, ficpar); */
       /*   fputs(line,stdout); */
       /*   fputs(line,ficparo); */
       /* } */
       /* ungetc(c,ficpar); */
       
       estepm=0;
       if((num_filled=sscanf(line,"agemin=%lf agemax=%lf bage=%lf fage=%lf estepm=%d ftolpl=%lf\n",&ageminpar,&agemaxpar, &bage, &fage, &estepm, &ftolpl)) !=EOF){
         
         if (num_filled != 6) {
           printf("Error: Not 6 parameters in line, for example:agemin=60 agemax=95 bage=55 fage=95 estepm=24 ftolpl=6e-4\n, your line=%s . Probably you are running an older format.\n",line);
           fprintf(ficlog,"Error: Not 6 parameters in line, for example:agemin=60 agemax=95 bage=55 fage=95 estepm=24 ftolpl=6e-4\n, your line=%s . Probably you are running an older format.\n",line);
           goto end;
         }
         printf("agemin=%lf agemax=%lf bage=%lf fage=%lf estepm=%d ftolpl=%lf\n",ageminpar,agemaxpar, bage, fage, estepm, ftolpl);
       }
       /* ftolpl=6*ftol*1.e5; /\* 6.e-3 make convergences in less than 80 loops for the prevalence limit *\/ */
       /*ftolpl=6.e-4;*/ /* 6.e-3 make convergences in less than 80 loops for the prevalence limit */
       
       /* fscanf(ficpar,"agemin=%lf agemax=%lf bage=%lf fage=%lf estepm=%d ftolpl=%\n",&ageminpar,&agemaxpar, &bage, &fage, &estepm); */
       if (estepm==0 || estepm < stepm) estepm=stepm;
       if (fage <= 2) {
         bage = ageminpar;
         fage = agemaxpar;
       }
       
       fprintf(ficres,"# agemin agemax for life expectancy, bage fage (if mle==0 ie no data nor Max likelihood).\n");
       fprintf(ficres,"agemin=%.0f agemax=%.0f bage=%.0f fage=%.0f estepm=%d ftolpl=%e\n",ageminpar,agemaxpar,bage,fage, estepm, ftolpl);
       fprintf(ficparo,"agemin=%.0f agemax=%.0f bage=%.0f fage=%.0f estepm=%d, ftolpl=%e\n",ageminpar,agemaxpar,bage,fage, estepm, ftolpl);
                   
       /* Other stuffs, more or less useful */    
       while(fgets(line, MAXLINE, ficpar)) {
         /* If line starts with a # it is a comment */
         if (line[0] == '#') {
           numlinepar++;
           fputs(line,stdout);
           fputs(line,ficparo);
           fputs(line,ficlog);
           fputs(line,ficres);
           continue;
         }else
           break;
       }
   
   fprintf(ficrespij,"#****** h Pij x Probability to be in state j at age x+h being in i at x ");      if((num_filled=sscanf(line,"begin-prev-date=%lf/%lf/%lf end-prev-date=%lf/%lf/%lf mov_average=%d\n",&jprev1, &mprev1,&anprev1,&jprev2, &mprev2,&anprev2,&mobilav)) !=EOF){
   for(cptcov=1,k=0;cptcov<=i1;cptcov++){        
     for(cptcod=1;cptcod<=ncodemax[cptcov];cptcod++){        if (num_filled != 7) {
       k=k+1;          printf("Error: Not 7 (data)parameters in line but %d, for example:begin-prev-date=1/1/1990 end-prev-date=1/6/2004  mov_average=0\n, your line=%s . Probably you are running an older format.\n",num_filled,line);
       fprintf(ficrespij,"\n#****** ");          fprintf(ficlog,"Error: Not 7 (data)parameters in line but %d, for example:begin-prev-date=1/1/1990 end-prev-date=1/6/2004  mov_average=0\n, your line=%s . Probably you are running an older format.\n",num_filled,line);
       for(j=1;j<=cptcoveff;j++)           goto end;
         fprintf(ficrespij,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);        }
       fprintf(ficrespij,"******\n");        printf("begin-prev-date=%.lf/%.lf/%.lf end-prev-date=%.lf/%.lf/%.lf mov_average=%d\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,mobilav);
         fprintf(ficparo,"begin-prev-date=%.lf/%.lf/%.lf end-prev-date=%.lf/%.lf/%.lf mov_average=%d\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,mobilav);
         fprintf(ficres,"begin-prev-date=%.lf/%.lf/%.lf end-prev-date=%.lf/%.lf/%.lf mov_average=%d\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,mobilav);
         fprintf(ficlog,"begin-prev-date=%.lf/%.lf/%.lf end-prev-date=%.lf/%.lf/%.lf mov_average=%d\n",jprev1, mprev1,anprev1,jprev2, mprev2,anprev2,mobilav);
       }
   
       while(fgets(line, MAXLINE, ficpar)) {
         /* If line starts with a # it is a comment */
         if (line[0] == '#') {
           numlinepar++;
           fputs(line,stdout);
           fputs(line,ficparo);
           fputs(line,ficlog);
           fputs(line,ficres);
           continue;
         }else
           break;
       }
       
       
       dateprev1=anprev1+(mprev1-1)/12.+(jprev1-1)/365.;
       dateprev2=anprev2+(mprev2-1)/12.+(jprev2-1)/365.;
       
       if((num_filled=sscanf(line,"pop_based=%d\n",&popbased)) !=EOF){
         if (num_filled != 1) {
           printf("Error: Not 1 (data)parameters in line but %d, for example:pop_based=0\n, your line=%s . Probably you are running an older format.\n",num_filled,line);
           fprintf(ficlog,"Error: Not 1 (data)parameters in line but %d, for example: pop_based=1\n, your line=%s . Probably you are running an older format.\n",num_filled,line);
           goto end;
         }
         printf("pop_based=%d\n",popbased);
         fprintf(ficlog,"pop_based=%d\n",popbased);
         fprintf(ficparo,"pop_based=%d\n",popbased);   
         fprintf(ficres,"pop_based=%d\n",popbased);   
       }
        
       /* Results */
       /* Value of covariate in each resultine will be compututed (if product) and sorted according to model rank */
       /* It is precov[] because we need the varying age in order to compute the real cov[] of the model equation */  
       precov=matrix(1,MAXRESULTLINESPONE,1,NCOVMAX+1);
       endishere=0;
       nresult=0;
       parameterline=0;
       do{
         if(!fgets(line, MAXLINE, ficpar)){
           endishere=1;
           parameterline=15;
         }else if (line[0] == '#') {
           /* If line starts with a # it is a comment */
           numlinepar++;
           fputs(line,stdout);
           fputs(line,ficparo);
           fputs(line,ficlog);
           fputs(line,ficres);
           continue;
         }else if(sscanf(line,"prevforecast=%[^\n]\n",modeltemp))
           parameterline=11;
         else if(sscanf(line,"prevbackcast=%[^\n]\n",modeltemp))
           parameterline=12;
         else if(sscanf(line,"result:%[^\n]\n",modeltemp)){
           parameterline=13;
         }
         else{
           parameterline=14;
         }
         switch (parameterline){ /* =0 only if only comments */
         case 11:
           if((num_filled=sscanf(line,"prevforecast=%d starting-proj-date=%lf/%lf/%lf final-proj-date=%lf/%lf/%lf mobil_average=%d\n",&prevfcast,&jproj1,&mproj1,&anproj1,&jproj2,&mproj2,&anproj2,&mobilavproj)) !=EOF && (num_filled == 8)){
                     fprintf(ficparo,"prevforecast=%d starting-proj-date=%.lf/%.lf/%.lf final-proj-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevfcast,jproj1,mproj1,anproj1,jproj2,mproj2,anproj2,mobilavproj);
             printf("prevforecast=%d starting-proj-date=%.lf/%.lf/%.lf final-proj-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevfcast,jproj1,mproj1,anproj1,jproj2,mproj2,anproj2,mobilavproj);
             fprintf(ficlog,"prevforecast=%d starting-proj-date=%.lf/%.lf/%.lf final-proj-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevfcast,jproj1,mproj1,anproj1,jproj2,mproj2,anproj2,mobilavproj);
             fprintf(ficres,"prevforecast=%d starting-proj-date=%.lf/%.lf/%.lf final-proj-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevfcast,jproj1,mproj1,anproj1,jproj2,mproj2,anproj2,mobilavproj);
             /* day and month of proj2 are not used but only year anproj2.*/
             dateproj1=anproj1+(mproj1-1)/12.+(jproj1-1)/365.;
             dateproj2=anproj2+(mproj2-1)/12.+(jproj2-1)/365.;
             prvforecast = 1;
           } 
           else if((num_filled=sscanf(line,"prevforecast=%d yearsfproj=%lf mobil_average=%d\n",&prevfcast,&yrfproj,&mobilavproj)) !=EOF){/* && (num_filled == 3))*/
             printf("prevforecast=%d yearsfproj=%.2lf mobil_average=%d\n",prevfcast,yrfproj,mobilavproj);
             fprintf(ficlog,"prevforecast=%d yearsfproj=%.2lf mobil_average=%d\n",prevfcast,yrfproj,mobilavproj);
             fprintf(ficres,"prevforecast=%d yearsfproj=%.2lf mobil_average=%d\n",prevfcast,yrfproj,mobilavproj);
             prvforecast = 2;
           }
           else {
             printf("Error: Not 8 (data)parameters in line but %d, for example:prevforecast=1 starting-proj-date=1/1/1990 final-proj-date=1/1/2000 mobil_average=0\nnor 3 (data)parameters, for example:prevforecast=1 yearsfproj=10 mobil_average=0. Your line=%s . You are running probably an older format.\n, ",num_filled,line);
             fprintf(ficlog,"Error: Not 8 (data)parameters in line but %d, for example:prevforecast=1 starting-proj-date=1/1/1990 final-proj-date=1/1/2000 mobil_average=0\nnor 3 (data)parameters, for example:prevforecast=1 yearproj=10 mobil_average=0. Your line=%s . You are running probably an older format.\n, ",num_filled,line);
             goto end;
           }
           break;
         case 12:
           if((num_filled=sscanf(line,"prevbackcast=%d starting-back-date=%lf/%lf/%lf final-back-date=%lf/%lf/%lf mobil_average=%d\n",&prevbcast,&jback1,&mback1,&anback1,&jback2,&mback2,&anback2,&mobilavproj)) !=EOF && (num_filled == 8)){
             fprintf(ficparo,"prevbackcast=%d starting-back-date=%.lf/%.lf/%.lf final-back-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevbcast,jback1,mback1,anback1,jback2,mback2,anback2,mobilavproj);
             printf("prevbackcast=%d starting-back-date=%.lf/%.lf/%.lf final-back-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevbcast,jback1,mback1,anback1,jback2,mback2,anback2,mobilavproj);
             fprintf(ficlog,"prevbackcast=%d starting-back-date=%.lf/%.lf/%.lf final-back-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevbcast,jback1,mback1,anback1,jback2,mback2,anback2,mobilavproj);
             fprintf(ficres,"prevbackcast=%d starting-back-date=%.lf/%.lf/%.lf final-back-date=%.lf/%.lf/%.lf mobil_average=%d\n",prevbcast,jback1,mback1,anback1,jback2,mback2,anback2,mobilavproj);
             /* day and month of back2 are not used but only year anback2.*/
             dateback1=anback1+(mback1-1)/12.+(jback1-1)/365.;
             dateback2=anback2+(mback2-1)/12.+(jback2-1)/365.;
             prvbackcast = 1;
           } 
           else if((num_filled=sscanf(line,"prevbackcast=%d yearsbproj=%lf mobil_average=%d\n",&prevbcast,&yrbproj,&mobilavproj)) ==3){/* && (num_filled == 3))*/
             printf("prevbackcast=%d yearsbproj=%.2lf mobil_average=%d\n",prevbcast,yrbproj,mobilavproj);
             fprintf(ficlog,"prevbackcast=%d yearsbproj=%.2lf mobil_average=%d\n",prevbcast,yrbproj,mobilavproj);
             fprintf(ficres,"prevbackcast=%d yearsbproj=%.2lf mobil_average=%d\n",prevbcast,yrbproj,mobilavproj);
             prvbackcast = 2;
           }
           else {
             printf("Error: Not 8 (data)parameters in line but %d, for example:prevbackcast=1 starting-back-date=1/1/1990 final-back-date=1/1/2000 mobil_average=0\nnor 3 (data)parameters, for example:prevbackcast=1 yearsbproj=10 mobil_average=0. Your line=%s . You are running probably an older format.\n, ",num_filled,line);
             fprintf(ficlog,"Error: Not 8 (data)parameters in line but %d, for example:prevbackcast=1 starting-back-date=1/1/1990 final-back-date=1/1/2000 mobil_average=0\nnor 3 (data)parameters, for example:prevbackcast=1 yearbproj=10 mobil_average=0. Your line=%s . You are running probably an older format.\n, ",num_filled,line);
             goto end;
           }
           break;
         case 13:
           num_filled=sscanf(line,"result:%[^\n]\n",resultlineori);
           nresult++; /* Sum of resultlines */
           printf("Result %d: result:%s\n",nresult, resultlineori);
           /* removefirstspace(&resultlineori); */
                   
       for (agedeb=fage; agedeb>=bage; agedeb--){ /* If stepm=6 months */          if(strstr(resultlineori,"v") !=0){
         nhstepm=(int) rint((agelim-agedeb)*YEARM/stepm); /* Typically 20 years = 20*12/6=40 */             printf("Error. 'v' must be in upper case 'V' result: %s ",resultlineori);
         nhstepm = nhstepm/hstepm; /* Typically 40/4=10 */            fprintf(ficlog,"Error. 'v' must be in upper case result: %s ",resultlineori);fflush(ficlog);
             return 1;
         /*        nhstepm=nhstepm*YEARM; aff par mois*/          }
           trimbb(resultline, resultlineori); /* Suppressing double blank in the resultline */
         p3mat=ma3x(1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);          printf("Decoderesult resultline=\"%s\" resultlineori=\"%s\"\n", resultline, resultlineori);
         oldm=oldms;savm=savms;          if(nresult > MAXRESULTLINESPONE-1){
         hpxij(p3mat,nhstepm,agedeb,hstepm,p,nlstate,stepm,oldm,savm, k);              printf("ERROR: Current version of IMaCh limits the number of resultlines to %d, you used %d\nYou can use the 'r' parameter file '%s' which uses option mle=0 to get other results. ",MAXRESULTLINESPONE-1,nresult,rfileres);
         fprintf(ficrespij,"# Cov Agex agex+h hpijx with i,j=");            fprintf(ficlog,"ERROR: Current version of IMaCh limits the number of resultlines to %d, you used %d\nYou can use the 'r' parameter file '%s' which uses option mle=0 to get other results. ",MAXRESULTLINESPONE-1,nresult,rfileres);
         for(i=1; i<=nlstate;i++)            goto end;
           for(j=1; j<=nlstate+ndeath;j++)  
             fprintf(ficrespij," %1d-%1d",i,j);  
         fprintf(ficrespij,"\n");  
         for (h=0; h<=nhstepm; h++){  
           fprintf(ficrespij,"%d %3.f %3.f",k,agedeb, agedeb+ h*hstepm/YEARM*stepm );  
           for(i=1; i<=nlstate;i++)  
             for(j=1; j<=nlstate+ndeath;j++)  
               fprintf(ficrespij," %.5f", p3mat[i][j][h]);  
           fprintf(ficrespij,"\n");  
         }          }
         free_ma3x(p3mat,1,nlstate+ndeath,1, nlstate+ndeath, 0,nhstepm);          
         fprintf(ficrespij,"\n");          if(!decoderesult(resultline, nresult)){ /* Fills TKresult[nresult] combination and Tresult[nresult][k4+1] combination values */
             fprintf(ficparo,"result: %s\n",resultline);
             fprintf(ficres,"result: %s\n",resultline);
             fprintf(ficlog,"result: %s\n",resultline);
           } else
             goto end;
           break;
         case 14:
           printf("Error: Unknown command '%s'\n",line);
           fprintf(ficlog,"Error: Unknown command '%s'\n",line);
           if(line[0] == ' ' || line[0] == '\n'){
             printf("It should not be an empty line '%s'\n",line);
             fprintf(ficlog,"It should not be an empty line '%s'\n",line);
           }         
           if(ncovmodel >=2 && nresult==0 ){
             printf("ERROR: no result lines! It should be at minimum 'result: V2=0 V1=1 or result:.' %s\n",line);
             fprintf(ficlog,"ERROR: no result lines! It should be at minimum 'result: V2=0 V1=1 or result:.' %s\n",line);
           }
           /* goto end; */
           break;
         case 15:
           printf("End of resultlines.\n");
           fprintf(ficlog,"End of resultlines.\n");
           break;
         default: /* parameterline =0 */
           nresult=1;
           decoderesult(".",nresult ); /* No covariate */
         } /* End switch parameterline */
       }while(endishere==0); /* End do */
       
       /* freqsummary(fileres, agemin, agemax, s, agev, nlstate, imx,Tvaraff,nbcode, ncodemax,mint,anint); */
       /* ,dateprev1,dateprev2,jprev1, mprev1,anprev1,jprev2, mprev2,anprev2); */
       
       replace_back_to_slash(pathc,pathcd); /* Even gnuplot wants a / */
       if(ageminpar == AGEOVERFLOW ||agemaxpar == -AGEOVERFLOW){
         printf("Warning! Error in gnuplot file with ageminpar %f or agemaxpar %f overflow\n\
   This is probably because your parameter file doesn't \n  contain the exact number of lines (or columns) corresponding to your model line.\n\
   Please run with mle=-1 to get a correct covariance matrix.\n",ageminpar,agemaxpar);
         fprintf(ficlog,"Warning! Error in gnuplot file with ageminpar %f or agemaxpar %f overflow\n\
   This is probably because your parameter file doesn't \n  contain the exact number of lines (or columns) corresponding to your model line.\n\
   Please run with mle=-1 to get a correct covariance matrix.\n",ageminpar,agemaxpar);
       }else{
         /* printinggnuplot(fileresu, optionfilefiname,ageminpar,agemaxpar,fage, prevfcast, backcast, pathc,p, (int)anproj1-(int)agemin, (int)anback1-(int)agemax+1); */
         /* It seems that anprojd which is computed from the mean year at interview which is known yet because of freqsummary */
         /* date2dmy(dateintmean,&jintmean,&mintmean,&aintmean); */ /* Done in freqsummary */
         if(prvforecast==1){
           dateprojd=(jproj1+12*mproj1+365*anproj1)/365;
           jprojd=jproj1;
           mprojd=mproj1;
           anprojd=anproj1;
           dateprojf=(jproj2+12*mproj2+365*anproj2)/365;
           jprojf=jproj2;
           mprojf=mproj2;
           anprojf=anproj2;
         } else if(prvforecast == 2){
           dateprojd=dateintmean;
           date2dmy(dateprojd,&jprojd, &mprojd, &anprojd);
           dateprojf=dateintmean+yrfproj;
           date2dmy(dateprojf,&jprojf, &mprojf, &anprojf);
         }
         if(prvbackcast==1){
           datebackd=(jback1+12*mback1+365*anback1)/365;
           jbackd=jback1;
           mbackd=mback1;
           anbackd=anback1;
           datebackf=(jback2+12*mback2+365*anback2)/365;
           jbackf=jback2;
           mbackf=mback2;
           anbackf=anback2;
         } else if(prvbackcast == 2){
           datebackd=dateintmean;
           date2dmy(datebackd,&jbackd, &mbackd, &anbackd);
           datebackf=dateintmean-yrbproj;
           date2dmy(datebackf,&jbackf, &mbackf, &anbackf);
       }        }
         
         printinggnuplot(fileresu, optionfilefiname,ageminpar,agemaxpar,bage, fage, prevfcast, prevbcast, pathc,p, (int)anprojd-bage, (int)anbackd-fage);
     }      }
   }      printinghtml(fileresu,title,datafile, firstpass, lastpass, stepm, weightopt, \
                    model,imx,jmin,jmax,jmean,rfileres,popforecast,mobilav,prevfcast,mobilavproj,prevbcast, estepm, \
   varprob(optionfilefiname, matcov, p, delti, nlstate, bage, fage,k,Tvar,nbcode, ncodemax);                   jprev1,mprev1,anprev1,dateprev1, dateprojd, datebackd,jprev2,mprev2,anprev2,dateprev2,dateprojf, datebackf);
                   
   fclose(ficrespij);      /*------------ free_vector  -------------*/
       /*  chdir(path); */
   probs= ma3x(1,AGESUP,1,NCOVMAX, 1,NCOVMAX);                  
       /* free_ivector(wav,1,imx); */  /* Moved after last prevalence call */
   /*---------- Forecasting ------------------*/      /* free_imatrix(dh,1,lastpass-firstpass+2,1,imx); */
   /*if((stepm == 1) && (strcmp(model,".")==0)){*/      /* free_imatrix(bh,1,lastpass-firstpass+2,1,imx); */
   if(prevfcast==1){      /* free_imatrix(mw,1,lastpass-firstpass+2,1,imx);    */
     /*    if(stepm ==1){*/      free_lvector(num,firstobs,lastobs);
       prevforecast(fileres, anproj1, mproj1, jproj1, agemin, agemax, dateprev1, dateprev2, mobilavproj, bage, fage, firstpass, lastpass, anproj2, p, cptcoveff);      free_vector(agedc,firstobs,lastobs);
       /* (popforecast==1) populforecast(fileres, anpyram,mpyram,jpyram, agemin,agemax, dateprev1, dateprev2,mobilav, agedeb, fage, popforecast, popfile, anpyram1,p, i1);*/      /*free_matrix(covar,0,NCOVMAX,1,n);*/
 /*      }  */      /*free_matrix(covar,1,NCOVMAX,1,n);*/
 /*      else{ */      fclose(ficparo);
 /*        erreur=108; */      fclose(ficres);
 /*        printf("Warning %d!! You can only forecast the prevalences if the optimization\n  has been performed with stepm = 1 (month) instead of %d or model=. instead of '%s'\n", erreur, stepm, model); */                  
 /*        fprintf(ficlog,"Warning %d!! You can only forecast the prevalences if the optimization\n  has been performed with stepm = 1 (month) instead of %d or model=. instead of '%s'\n", erreur, stepm, model); */                  
 /*      } */      /* Other results (useful)*/
   }                  
                     
       /*--------------- Prevalence limit  (period or stable prevalence) --------------*/
   /*---------- Health expectancies and variances ------------*/      /*#include "prevlim.h"*/  /* Use ficrespl, ficlog */
       prlim=matrix(1,nlstate,1,nlstate);
   strcpy(filerest,"t");      /* Computes the prevalence limit for each combination k of the dummy covariates by calling prevalim(k) */
   strcat(filerest,fileres);      prevalence_limit(p, prlim,  ageminpar, agemaxpar, ftolpl, &ncvyear);
   if((ficrest=fopen(filerest,"w"))==NULL) {      fclose(ficrespl);
     printf("Problem with total LE resultfile: %s\n", filerest);goto end;  
     fprintf(ficlog,"Problem with total LE resultfile: %s\n", filerest);goto end;      /*------------- h Pij x at various ages ------------*/
   }      /*#include "hpijx.h"*/
   printf("Computing Total LEs with variances: file '%s' \n", filerest);       /** h Pij x Probability to be in state j at age x+h being in i at x, for each combination k of dummies in the model line or to nres?*/
   fprintf(ficlog,"Computing Total LEs with variances: file '%s' \n", filerest);       /* calls hpxij with combination k */
       hPijx(p, bage, fage);
       fclose(ficrespij);
   strcpy(filerese,"e");      
   strcat(filerese,fileres);      /* ncovcombmax=  pow(2,cptcoveff); */
   if((ficreseij=fopen(filerese,"w"))==NULL) {      /*-------------- Variance of one-step probabilities for a combination ij or for nres ?---*/
     printf("Problem with Health Exp. resultfile: %s\n", filerese); exit(0);      k=1;
     fprintf(ficlog,"Problem with Health Exp. resultfile: %s\n", filerese); exit(0);      varprob(optionfilefiname, matcov, p, delti, nlstate, bage, fage,k,Tvar,nbcode, ncodemax,strstart);
   }      
   printf("Computing Health Expectancies: result on file '%s' \n", filerese);      /* Prevalence for each covariate combination in probs[age][status][cov] */
   fprintf(ficlog,"Computing Health Expectancies: result on file '%s' \n", filerese);      probs= ma3x(AGEINF,AGESUP,1,nlstate+ndeath, 1,ncovcombmax);
       for(i=AGEINF;i<=AGESUP;i++)
   strcpy(fileresv,"v");        for(j=1;j<=nlstate+ndeath;j++) /* ndeath is useless but a necessity to be compared with mobaverages */
   strcat(fileresv,fileres);          for(k=1;k<=ncovcombmax;k++)
   if((ficresvij=fopen(fileresv,"w"))==NULL) {            probs[i][j][k]=0.;
     printf("Problem with variance resultfile: %s\n", fileresv);exit(0);      prevalence(probs, ageminpar, agemaxpar, s, agev, nlstate, imx, Tvar, nbcode, 
     fprintf(ficlog,"Problem with variance resultfile: %s\n", fileresv);exit(0);                 ncodemax, mint, anint, dateprev1, dateprev2, firstpass, lastpass);
   }      if (mobilav!=0 ||mobilavproj !=0 ) {
   printf("Computing Variance-covariance of DFLEs: file '%s' \n", fileresv);        mobaverages= ma3x(AGEINF, AGESUP,1,nlstate+ndeath, 1,ncovcombmax);
   fprintf(ficlog,"Computing Variance-covariance of DFLEs: file '%s' \n", fileresv);        for(i=AGEINF;i<=AGESUP;i++)
           for(j=1;j<=nlstate+ndeath;j++)
   /* Computes prevalence between agemin (i.e minimal age computed) and no more ageminpar */            for(k=1;k<=ncovcombmax;k++)
   prevalence(probs, agemin, agemax, s, agev, nlstate, imx, Tvar, nbcode, ncodemax, mint, anint, dateprev1, dateprev2, firstpass, lastpass);              mobaverages[i][j][k]=0.;
   /*  printf("ageminpar=%f, agemax=%f, s[lastpass][imx]=%d, agev[lastpass][imx]=%f, nlstate=%d, imx=%d,  mint[lastpass][imx]=%f, anint[lastpass][imx]=%f,dateprev1=%f, dateprev2=%f, firstpass=%d, lastpass=%d\n",\        mobaverage=mobaverages;
 ageminpar, agemax, s[lastpass][imx], agev[lastpass][imx], nlstate, imx, mint[lastpass][imx],anint[lastpass][imx], dateprev1, dateprev2, firstpass, lastpass);        if (mobilav!=0) {
   */          printf("Movingaveraging observed prevalence\n");
           fprintf(ficlog,"Movingaveraging observed prevalence\n");
           if (movingaverage(probs, ageminpar, agemaxpar, mobaverage, mobilav)!=0){
             fprintf(ficlog," Error in movingaverage mobilav=%d\n",mobilav);
             printf(" Error in movingaverage mobilav=%d\n",mobilav);
           }
         } else if (mobilavproj !=0) {
           printf("Movingaveraging projected observed prevalence\n");
           fprintf(ficlog,"Movingaveraging projected observed prevalence\n");
           if (movingaverage(probs, ageminpar, agemaxpar, mobaverage, mobilavproj)!=0){
             fprintf(ficlog," Error in movingaverage mobilavproj=%d\n",mobilavproj);
             printf(" Error in movingaverage mobilavproj=%d\n",mobilavproj);
           }
         }else{
           printf("Internal error moving average\n");
           fflush(stdout);
           exit(1);
         }
       }/* end if moving average */
       
       /*---------- Forecasting ------------------*/
       if(prevfcast==1){ 
         /*   /\*    if(stepm ==1){*\/ */
         /*   /\*  anproj1, mproj1, jproj1 either read explicitly or yrfproj *\/ */
         /*This done previously after freqsummary.*/
         /*   dateprojd=(jproj1+12*mproj1+365*anproj1)/365; */
         /*   dateprojf=(jproj2+12*mproj2+365*anproj2)/365; */
         
         /* } else if (prvforecast==2){ */
         /*   /\*    if(stepm ==1){*\/ */
         /*   /\*  anproj1, mproj1, jproj1 either read explicitly or yrfproj *\/ */
         /* } */
         /*prevforecast(fileresu, dateintmean, anproj1, mproj1, jproj1, agemin, agemax, dateprev1, dateprev2, mobilavproj, mobaverage, bage, fage, firstpass, lastpass, anproj2, p, cptcoveff);*/
         prevforecast(fileresu,dateintmean, dateprojd, dateprojf, agemin, agemax, dateprev1, dateprev2, mobilavproj, mobaverage, bage, fage, firstpass, lastpass, p, cptcoveff);
       }
   
       /* Prevbcasting */
       if(prevbcast==1){
         ddnewms=matrix(1,nlstate+ndeath,1,nlstate+ndeath);        
         ddoldms=matrix(1,nlstate+ndeath,1,nlstate+ndeath);        
         ddsavms=matrix(1,nlstate+ndeath,1,nlstate+ndeath);
   
         /*--------------- Back Prevalence limit  (period or stable prevalence) --------------*/
   
         bprlim=matrix(1,nlstate,1,nlstate);
   
         back_prevalence_limit(p, bprlim,  ageminpar, agemaxpar, ftolpl, &ncvyear, dateprev1, dateprev2, firstpass, lastpass, mobilavproj);
         fclose(ficresplb);
   
         hBijx(p, bage, fage, mobaverage);
         fclose(ficrespijb);
   
         /* /\* prevbackforecast(fileresu, mobaverage, anback1, mback1, jback1, agemin, agemax, dateprev1, dateprev2, *\/ */
         /* /\*                   mobilavproj, bage, fage, firstpass, lastpass, anback2, p, cptcoveff); *\/ */
         /* prevbackforecast(fileresu, mobaverage, anback1, mback1, jback1, agemin, agemax, dateprev1, dateprev2, */
         /*                       mobilavproj, bage, fage, firstpass, lastpass, anback2, p, cptcoveff); */
         prevbackforecast(fileresu, mobaverage, dateintmean, dateprojd, dateprojf, agemin, agemax, dateprev1, dateprev2,
                          mobilavproj, bage, fage, firstpass, lastpass, p, cptcoveff);
   
   if (mobilav!=0) {        
     mobaverage= ma3x(1, AGESUP,1,NCOVMAX, 1,NCOVMAX);        varbprlim(fileresu, nresult, mobaverage, mobilavproj, bage, fage, bprlim, &ncvyear, ftolpl, p, matcov, delti, stepm, cptcoveff);
     if (movingaverage(probs, bage, fage, mobaverage,mobilav)!=0){  
       fprintf(ficlog," Error in movingaverage mobilav=%d\n",mobilav);  
       printf(" Error in movingaverage mobilav=%d\n",mobilav);  
     }  
   }  
   
   for(cptcov=1,k=0;cptcov<=i1;cptcov++){        
     for(cptcod=1;cptcod<=ncodemax[cptcov];cptcod++){        free_matrix(bprlim,1,nlstate,1,nlstate); /*here or after loop ? */
       k=k+1;         free_matrix(ddnewms, 1, nlstate+ndeath, 1, nlstate+ndeath);
       fprintf(ficrest,"\n#****** ");        free_matrix(ddsavms, 1, nlstate+ndeath, 1, nlstate+ndeath);
       for(j=1;j<=cptcoveff;j++)         free_matrix(ddoldms, 1, nlstate+ndeath, 1, nlstate+ndeath);
         fprintf(ficrest,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);      }    /* end  Prevbcasting */
       fprintf(ficrest,"******\n");   
    
       /* ------ Other prevalence ratios------------ */
   
       free_ivector(wav,1,imx);
       free_imatrix(dh,1,lastpass-firstpass+2,1,imx);
       free_imatrix(bh,1,lastpass-firstpass+2,1,imx);
       free_imatrix(mw,1,lastpass-firstpass+2,1,imx);   
                   
                   
       /*---------- Health expectancies, no variances ------------*/
                   
       strcpy(filerese,"E_");
       strcat(filerese,fileresu);
       if((ficreseij=fopen(filerese,"w"))==NULL) {
         printf("Problem with Health Exp. resultfile: %s\n", filerese); exit(0);
         fprintf(ficlog,"Problem with Health Exp. resultfile: %s\n", filerese); exit(0);
       }
       printf("Computing Health Expectancies: result on file '%s' ...", filerese);fflush(stdout);
       fprintf(ficlog,"Computing Health Expectancies: result on file '%s' ...", filerese);fflush(ficlog);
   
       pstamp(ficreseij);
                   
       i1=pow(2,cptcoveff); /* Number of combination of dummy covariates */
       if (cptcovn < 1){i1=1;}
       
       for(nres=1; nres <= nresult; nres++) /* For each resultline */
       for(k=1; k<=i1;k++){ /* For any combination of dummy covariates, fixed and varying */
         if(i1 != 1 && TKresult[nres]!= k)
           continue;
       fprintf(ficreseij,"\n#****** ");        fprintf(ficreseij,"\n#****** ");
       for(j=1;j<=cptcoveff;j++)         printf("\n#****** ");
         fprintf(ficreseij,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);        for(j=1;j<=cptcoveff;j++) {
           fprintf(ficreseij,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
           printf("V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]);
         }
         for (j=1; j<= nsq; j++){ /* For each selected (single) quantitative value */
           printf(" V%d=%f ",TvarsQ[j], TinvDoQresult[nres][TvarsQ[j]]); /* TvarsQ[j] gives the name of the jth quantitative (fixed or time v) */
           fprintf(ficreseij,"V%d=%f ",TvarsQ[j], TinvDoQresult[nres][TvarsQ[j]]);
         }
       fprintf(ficreseij,"******\n");        fprintf(ficreseij,"******\n");
         printf("******\n");
         
         eij=ma3x(1,nlstate,1,nlstate,(int) bage, (int) fage);
         oldm=oldms;savm=savms;
         /* printf("HELLO Entering evsij bage=%d fage=%d k=%d estepm=%d nres=%d\n",(int) bage, (int)fage, k, estepm, nres); */
         evsij(eij, p, nlstate, stepm, (int) bage, (int)fage, oldm, savm, k, estepm, strstart, nres);  
         
         free_ma3x(eij,1,nlstate,1,nlstate,(int) bage, (int)fage);
       }
       fclose(ficreseij);
       printf("done evsij\n");fflush(stdout);
       fprintf(ficlog,"done evsij\n");fflush(ficlog);
   
                   
       /*---------- State-specific expectancies and variances ------------*/
                   
       strcpy(filerest,"T_");
       strcat(filerest,fileresu);
       if((ficrest=fopen(filerest,"w"))==NULL) {
         printf("Problem with total LE resultfile: %s\n", filerest);goto end;
         fprintf(ficlog,"Problem with total LE resultfile: %s\n", filerest);goto end;
       }
       printf("Computing Total Life expectancies with their standard errors: file '%s' ...\n", filerest); fflush(stdout);
       fprintf(ficlog,"Computing Total Life expectancies with their standard errors: file '%s' ...\n", filerest); fflush(ficlog);
       strcpy(fileresstde,"STDE_");
       strcat(fileresstde,fileresu);
       if((ficresstdeij=fopen(fileresstde,"w"))==NULL) {
         printf("Problem with State specific Exp. and std errors resultfile: %s\n", fileresstde); exit(0);
         fprintf(ficlog,"Problem with State specific Exp. and std errors resultfile: %s\n", fileresstde); exit(0);
       }
       printf("  Computing State-specific Expectancies and standard errors: result on file '%s' \n", fileresstde);
       fprintf(ficlog,"  Computing State-specific Expectancies and standard errors: result on file '%s' \n", fileresstde);
   
       strcpy(filerescve,"CVE_");
       strcat(filerescve,fileresu);
       if((ficrescveij=fopen(filerescve,"w"))==NULL) {
         printf("Problem with Covar. State-specific Exp. resultfile: %s\n", filerescve); exit(0);
         fprintf(ficlog,"Problem with Covar. State-specific Exp. resultfile: %s\n", filerescve); exit(0);
       }
       printf("    Computing Covar. of State-specific Expectancies: result on file '%s' \n", filerescve);
       fprintf(ficlog,"    Computing Covar. of State-specific Expectancies: result on file '%s' \n", filerescve);
   
       strcpy(fileresv,"V_");
       strcat(fileresv,fileresu);
       if((ficresvij=fopen(fileresv,"w"))==NULL) {
         printf("Problem with variance resultfile: %s\n", fileresv);exit(0);
         fprintf(ficlog,"Problem with variance resultfile: %s\n", fileresv);exit(0);
       }
       printf("      Computing Variance-covariance of State-specific Expectancies: file '%s' ... ", fileresv);fflush(stdout);
       fprintf(ficlog,"      Computing Variance-covariance of State-specific Expectancies: file '%s' ... ", fileresv);fflush(ficlog);
   
       i1=pow(2,cptcoveff); /* Number of combination of dummy covariates */
       if (cptcovn < 1){i1=1;}
       
       for(nres=1; nres <= nresult; nres++) /* For each resultline, find the combination and output results according to the values of dummies and then quanti.  */
       for(k=1; k<=i1;k++){ /* For any combination of dummy covariates, fixed and varying. For each nres and each value at position k
                             * we know Tresult[nres][result_position]= value of the dummy variable at the result_position in the nres resultline
                             * Tvqresult[nres][result_position]= id of the variable at the result_position in the nres resultline 
                             * and Tqresult[nres][result_position]= value of the variable at the result_position in the nres resultline */
         /* */
         if(i1 != 1 && TKresult[nres]!= k) /* TKresult[nres] is the combination of this nres resultline. All the i1 combinations are not output */
           continue;
         printf("\n# model %s \n#****** Result for:", model);
         fprintf(ficrest,"\n# model %s \n#****** Result for:", model);
         fprintf(ficlog,"\n# model %s \n#****** Result for:", model);
         /* It might not be a good idea to mix dummies and quantitative */
         /* for(j=1;j<=cptcoveff;j++){ /\* j=resultpos. Could be a loop on cptcovs: number of single dummy covariate in the result line as well as in the model *\/ */
         for(j=1;j<=cptcovs;j++){ /* j=resultpos. Could be a loop on cptcovs: number of single covariate (dummy or quantitative) in the result line as well as in the model */
           /* printf("V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]); /\* Output by variables in the resultline *\/ */
           /* Tvaraff[j] is the name of the dummy variable in position j in the equation model:
            * Tvaraff[1]@9={4, 3, 0, 0, 0, 0, 0, 0, 0}, in model=V5+V4+V3+V4*V3+V5*age
            * (V5 is quanti) V4 and V3 are dummies
            * TnsdVar[4] is the position 1 and TnsdVar[3]=2 in codtabm(k,l)(V4  V3)=V4  V3
            *                                                              l=1 l=2
            *                                                           k=1  1   1   0   0
            *                                                           k=2  2   1   1   0
            *                                                           k=3 [1] [2]  0   1
            *                                                           k=4  2   2   1   1
            * If nres=1 result: V3=1 V4=0 then k=3 and outputs
            * If nres=2 result: V4=1 V3=0 then k=2 and outputs
            * nres=1 =>k=3 j=1 V4= nbcode[4][codtabm(3,1)=1)=0; j=2  V3= nbcode[3][codtabm(3,2)=2]=1
            * nres=2 =>k=2 j=1 V4= nbcode[4][codtabm(2,1)=2)=1; j=2  V3= nbcode[3][codtabm(2,2)=1]=0
            */
           /* Tvresult[nres][j] Name of the variable at position j in this resultline */
           /* Tresult[nres][j] Value of this variable at position j could be a float if quantitative  */
   /* We give up with the combinations!! */
           printf("\n j=%d In computing T_ Dummy[modelresult[%d][%d]]=%d, modelresult[%d][%d]=%d cptcovs=%d, cptcoveff=%d Fixed[modelresult[nres][j]]=%d\n", j, nres, j, Dummy[modelresult[nres][j]],nres,j,modelresult[nres][j],cptcovs, cptcoveff,Fixed[modelresult[nres][j]]);  /* end if dummy  or quanti */
   
           if(Dummy[modelresult[nres][j]]==0){/* Dummy variable of the variable in position modelresult in the model corresponding to j in resultline  */
             printf("V%d=%d ",Tvresult[nres][j],Tresult[nres][j]); /* Output of each value for the combination TKresult[nres], ordere by the covariate values in the resultline  */
             fprintf(ficlog,"V%d=%d ",Tvresult[nres][j],Tresult[nres][j]); /* Output of each value for the combination TKresult[nres], ordere by the covariate values in the resultline  */
             fprintf(ficrest,"V%d=%d ",Tvresult[nres][j],Tresult[nres][j]); /* Output of each value for the combination TKresult[nres], ordere by the covariate values in the resultline  */
             if(Fixed[modelresult[nres][j]]==0){ /* Fixed */
               printf("fixed ");fprintf(ficlog,"fixed ");fprintf(ficrest,"fixed ");
             }else{
               printf("varyi ");fprintf(ficlog,"varyi ");fprintf(ficrest,"varyi ");
             }
             /* fprintf(ficrest,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]); */
             /* fprintf(ficlog,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]); */
           }else if(Dummy[modelresult[nres][j]]==1){ /* Quanti variable */
             /* For each selected (single) quantitative value */
             printf(" V%d=%f ",Tvqresult[nres][j],Tqresult[nres][j]);
             if(Fixed[modelresult[nres][j]]==0){ /* Fixed */
               printf("fixed ");fprintf(ficlog,"fixed ");fprintf(ficrest,"fixed ");
             }else{
               printf("varyi ");fprintf(ficlog,"varyi ");fprintf(ficrest,"varyi ");
             }
           }else{
             printf("Error in computing T_ Dummy[modelresult[%d][%d]]=%d, modelresult[%d][%d]=%d cptcovs=%d, cptcoveff=%d \n", nres, j, Dummy[modelresult[nres][j]],nres,j,modelresult[nres][j],cptcovs, cptcoveff);  /* end if dummy  or quanti */
             fprintf(ficlog,"Error in computing T_ Dummy[modelresult[%d][%d]]=%d, modelresult[%d][%d]=%d cptcovs=%d, cptcoveff=%d \n", nres, j, Dummy[modelresult[nres][j]],nres,j,modelresult[nres][j],cptcovs, cptcoveff);  /* end if dummy  or quanti */
             exit(1);
           }
         } /* End loop for each variable in the resultline */
         /* for (j=1; j<= nsq; j++){ /\* For each selected (single) quantitative value *\/ */
         /*        printf(" V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]); /\* Wrong j is not in the equation model *\/ */
         /*        fprintf(ficrest," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]); */
         /*        fprintf(ficlog," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]); */
         /* }       */
         fprintf(ficrest,"******\n");
         fprintf(ficlog,"******\n");
         printf("******\n");
         
         fprintf(ficresstdeij,"\n#****** ");
         fprintf(ficrescveij,"\n#****** ");
         for(j=1;j<=cptcoveff;j++) {
           fprintf(ficresstdeij,"V%d=%d ",Tvresult[nres][j],Tresult[nres][j]);
           /* fprintf(ficresstdeij,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]); */
           /* fprintf(ficrescveij,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[Tvaraff[j]])]); */
         }
         for (j=1; j<= nsq; j++){ /* For each selected (single) quantitative value, TvarsQind gives the position of a quantitative in model equation  */
           fprintf(ficresstdeij," V%d=%f ",Tvar[TvarsQind[j]],Tqresult[nres][resultmodel[nres][TvarsQind[j]]]);
           fprintf(ficrescveij," V%d=%f ",Tvar[TvarsQind[j]],Tqresult[nres][resultmodel[nres][TvarsQind[j]]]);
         } 
         fprintf(ficresstdeij,"******\n");
         fprintf(ficrescveij,"******\n");
         
       fprintf(ficresvij,"\n#****** ");        fprintf(ficresvij,"\n#****** ");
         /* pstamp(ficresvij); */
       for(j=1;j<=cptcoveff;j++)         for(j=1;j<=cptcoveff;j++) 
         fprintf(ficresvij,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);          fprintf(ficresvij,"V%d=%d ",Tvresult[nres][j],Tresult[nres][j]);
           /* fprintf(ficresvij,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,TnsdVar[TnsdVar[Tvaraff[j]]])]); */
         for (j=1; j<= nsq; j++){ /* For each selected (single) quantitative value */
           /* fprintf(ficresvij," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][j]); /\* To solve *\/ */
           fprintf(ficresvij," V%d=%f ",Tvqresult[nres][j],Tqresult[nres][resultmodel[nres][j]]); /* Solved */
         } 
       fprintf(ficresvij,"******\n");        fprintf(ficresvij,"******\n");
         
       eij=ma3x(1,nlstate,1,nlstate,(int) bage, (int) fage);        eij=ma3x(1,nlstate,1,nlstate,(int) bage, (int) fage);
       oldm=oldms;savm=savms;        oldm=oldms;savm=savms;
       evsij(fileres, eij, p, nlstate, stepm, (int) bage, (int)fage, oldm, savm, k, estepm, delti, matcov);          printf(" cvevsij ");
          fprintf(ficlog, " cvevsij ");
         cvevsij(eij, p, nlstate, stepm, (int) bage, (int)fage, oldm, savm, k, estepm, delti, matcov, strstart, nres);
         printf(" end cvevsij \n ");
         fprintf(ficlog, " end cvevsij \n ");
         
         /*
          */
         /* goto endfree; */
         
       vareij=ma3x(1,nlstate,1,nlstate,(int) bage, (int) fage);        vareij=ma3x(1,nlstate,1,nlstate,(int) bage, (int) fage);
       oldm=oldms;savm=savms;        pstamp(ficrest);
       varevsij(optionfilefiname, vareij, matcov, p, delti, nlstate, stepm, (int) bage, (int) fage, oldm, savm, prlim, ftolpl,k, estepm, cptcov,cptcod,0, mobilav);        
       if(popbased==1){  
         varevsij(optionfilefiname, vareij, matcov, p, delti, nlstate, stepm, (int) bage, (int) fage, oldm, savm, prlim, ftolpl,k, estepm, cptcov,cptcod,popbased,mobilav);  
       }  
   
    
       fprintf(ficrest,"#Total LEs with variances: e.. (std) ");  
       for (i=1;i<=nlstate;i++) fprintf(ficrest,"e.%d (std) ",i);  
       fprintf(ficrest,"\n");  
   
       epj=vector(1,nlstate+1);        epj=vector(1,nlstate+1);
       for(age=bage; age <=fage ;age++){        for(vpopbased=0; vpopbased <= popbased; vpopbased++){ /* Done for vpopbased=0 and vpopbased=1 if popbased==1*/
         prevalim(prlim, nlstate, p, age, oldm, savm,ftolpl,k);          oldm=oldms;savm=savms; /* ZZ Segmentation fault */
         if (popbased==1) {          cptcod= 0; /* To be deleted */
           if(mobilav ==0){          printf("varevsij vpopbased=%d \n",vpopbased);
             for(i=1; i<=nlstate;i++)          fprintf(ficlog, "varevsij vpopbased=%d \n",vpopbased);
               prlim[i][i]=probs[(int)age][i][k];          varevsij(optionfilefiname, vareij, matcov, p, delti, nlstate, stepm, (int) bage, (int) fage, oldm, savm, prlim, ftolpl, &ncvyear, k, estepm, cptcov,cptcod,vpopbased,mobilav, strstart, nres); /* cptcod not initialized Intel */
           }else{ /* mobilav */           fprintf(ficrest,"# Total life expectancy with std error and decomposition into time to be expected in each health state\n#  (weighted average of eij where weights are ");
             for(i=1; i<=nlstate;i++)          if(vpopbased==1)
               prlim[i][i]=mobaverage[(int)age][i][k];            fprintf(ficrest,"the age specific prevalence observed (cross-sectionally) in the population i.e cross-sectionally\n in each health state (popbased=1) (mobilav=%d)\n",mobilav);
           else
             fprintf(ficrest,"the age specific forward period (stable) prevalences in each health state \n");
           fprintf(ficrest,"# Age popbased mobilav e.. (std) "); /* Adding covariate values? */
           for (i=1;i<=nlstate;i++) fprintf(ficrest,"e.%d (std) ",i);
           fprintf(ficrest,"\n");
           /* printf("Which p?\n"); for(i=1;i<=npar;i++)printf("p[i=%d]=%lf,",i,p[i]);printf("\n"); */
           printf("Computing age specific forward period (stable) prevalences in each health state \n");
           fprintf(ficlog,"Computing age specific forward period (stable) prevalences in each health state \n");
           for(age=bage; age <=fage ;age++){
             prevalim(prlim, nlstate, p, age, oldm, savm, ftolpl, &ncvyear, k, nres); /*ZZ Is it the correct prevalim */
             if (vpopbased==1) {
               if(mobilav ==0){
                 for(i=1; i<=nlstate;i++)
                   prlim[i][i]=probs[(int)age][i][k];
               }else{ /* mobilav */ 
                 for(i=1; i<=nlstate;i++)
                   prlim[i][i]=mobaverage[(int)age][i][k];
               }
           }            }
             
             fprintf(ficrest," %4.0f %d %d",age, vpopbased, mobilav);
             /* fprintf(ficrest," %4.0f %d %d %d %d",age, vpopbased, mobilav,Tvaraff[j],nbcode[Tvaraff[j]][codtabm(k,j)]); */ /* to be done */
             /* printf(" age %4.0f ",age); */
             for(j=1, epj[nlstate+1]=0.;j <=nlstate;j++){
               for(i=1, epj[j]=0.;i <=nlstate;i++) {
                 epj[j] += prlim[i][i]*eij[i][j][(int)age];
                 /*ZZZ  printf("%lf %lf ", prlim[i][i] ,eij[i][j][(int)age]);*/
                 /* printf("%lf %lf ", prlim[i][i] ,eij[i][j][(int)age]); */
               }
               epj[nlstate+1] +=epj[j];
             }
             /* printf(" age %4.0f \n",age); */
             
             for(i=1, vepp=0.;i <=nlstate;i++)
               for(j=1;j <=nlstate;j++)
                 vepp += vareij[i][j][(int)age];
             fprintf(ficrest," %7.3f (%7.3f)", epj[nlstate+1],sqrt(vepp));
             for(j=1;j <=nlstate;j++){
               fprintf(ficrest," %7.3f (%7.3f)", epj[j],sqrt(vareij[j][j][(int)age]));
             }
             fprintf(ficrest,"\n");
         }          }
                 } /* End vpopbased */
         fprintf(ficrest," %4.0f",age);        free_vector(epj,1,nlstate+1);
         for(j=1, epj[nlstate+1]=0.;j <=nlstate;j++){  
           for(i=1, epj[j]=0.;i <=nlstate;i++) {  
             epj[j] += prlim[i][i]*eij[i][j][(int)age];  
             /*  printf("%lf %lf ", prlim[i][i] ,eij[i][j][(int)age]);*/  
           }  
           epj[nlstate+1] +=epj[j];  
         }  
   
         for(i=1, vepp=0.;i <=nlstate;i++)  
           for(j=1;j <=nlstate;j++)  
             vepp += vareij[i][j][(int)age];  
         fprintf(ficrest," %7.3f (%7.3f)", epj[nlstate+1],sqrt(vepp));  
         for(j=1;j <=nlstate;j++){  
           fprintf(ficrest," %7.3f (%7.3f)", epj[j],sqrt(vareij[j][j][(int)age]));  
         }  
         fprintf(ficrest,"\n");  
       }  
       free_ma3x(eij,1,nlstate,1,nlstate,(int) bage, (int)fage);        free_ma3x(eij,1,nlstate,1,nlstate,(int) bage, (int)fage);
       free_ma3x(vareij,1,nlstate,1,nlstate,(int) bage, (int)fage);        free_ma3x(vareij,1,nlstate,1,nlstate,(int) bage, (int)fage);
       free_vector(epj,1,nlstate+1);        printf("done selection\n");fflush(stdout);
     }        fprintf(ficlog,"done selection\n");fflush(ficlog);
   }  
   free_vector(weight,1,n);  
   free_imatrix(Tvard,1,15,1,2);  
   free_imatrix(s,1,maxwav+1,1,n);  
   free_matrix(anint,1,maxwav,1,n);   
   free_matrix(mint,1,maxwav,1,n);  
   free_ivector(cod,1,n);  
   free_ivector(tab,1,NCOVMAX);  
   fclose(ficreseij);  
   fclose(ficresvij);  
   fclose(ficrest);  
   fclose(ficpar);  
     
   /*------- Variance of stable prevalence------*/     
   
   strcpy(fileresvpl,"vpl");  
   strcat(fileresvpl,fileres);  
   if((ficresvpl=fopen(fileresvpl,"w"))==NULL) {  
     printf("Problem with variance of stable prevalence  resultfile: %s\n", fileresvpl);  
     exit(0);  
   }  
   printf("Computing Variance-covariance of stable prevalence: file '%s' \n", fileresvpl);  
   
   for(cptcov=1,k=0;cptcov<=i1;cptcov++){  
     for(cptcod=1;cptcod<=ncodemax[cptcov];cptcod++){  
       k=k+1;  
       fprintf(ficresvpl,"\n#****** ");  
       for(j=1;j<=cptcoveff;j++)   
         fprintf(ficresvpl,"V%d=%d ",Tvaraff[j],nbcode[Tvaraff[j]][codtab[k][j]]);  
       fprintf(ficresvpl,"******\n");  
               
       varpl=matrix(1,nlstate,(int) bage, (int) fage);      } /* End k selection or end covariate selection for nres */
       oldm=oldms;savm=savms;  
       varprevlim(fileres, varpl, matcov, p, delti, nlstate, stepm, (int) bage, (int) fage, oldm, savm, prlim, ftolpl,k);      printf("done State-specific expectancies\n");fflush(stdout);
       free_matrix(varpl,1,nlstate,(int) bage, (int)fage);      fprintf(ficlog,"done State-specific expectancies\n");fflush(ficlog);
     }  
   }  
   
   fclose(ficresvpl);      /* variance-covariance of forward period prevalence */
       varprlim(fileresu, nresult, mobaverage, mobilavproj, bage, fage, prlim, &ncvyear, ftolpl, p, matcov, delti, stepm, cptcoveff);
   
   /*---------- End : free ----------------*/      
   free_matrix(pmmij,1,nlstate+ndeath,1,nlstate+ndeath);      free_vector(weight,firstobs,lastobs);
       free_imatrix(Tvardk,1,NCOVMAX,1,2);
       free_imatrix(Tvard,1,NCOVMAX,1,2);
       free_imatrix(s,1,maxwav+1,firstobs,lastobs);
       free_matrix(anint,1,maxwav,firstobs,lastobs); 
       free_matrix(mint,1,maxwav,firstobs,lastobs);
       free_ivector(cod,firstobs,lastobs);
       free_ivector(tab,1,NCOVMAX);
       fclose(ficresstdeij);
       fclose(ficrescveij);
       fclose(ficresvij);
       fclose(ficrest);
       fclose(ficpar);
       
       
       /*---------- End : free ----------------*/
       if (mobilav!=0 ||mobilavproj !=0)
         free_ma3x(mobaverages,AGEINF, AGESUP,1,nlstate+ndeath, 1,ncovcombmax); /* We need to have a squared matrix with prevalence of the dead! */
       free_ma3x(probs,AGEINF,AGESUP,1,nlstate+ndeath, 1,ncovcombmax);
       free_matrix(prlim,1,nlstate,1,nlstate); /*here or after loop ? */
       free_matrix(pmmij,1,nlstate+ndeath,1,nlstate+ndeath);
     }  /* mle==-3 arrives here for freeing */
     /* endfree:*/
   free_matrix(oldms, 1,nlstate+ndeath,1,nlstate+ndeath);    free_matrix(oldms, 1,nlstate+ndeath,1,nlstate+ndeath);
   free_matrix(newms, 1,nlstate+ndeath,1,nlstate+ndeath);    free_matrix(newms, 1,nlstate+ndeath,1,nlstate+ndeath);
   free_matrix(savms, 1,nlstate+ndeath,1,nlstate+ndeath);    free_matrix(savms, 1,nlstate+ndeath,1,nlstate+ndeath);
       if(ntv+nqtv>=1)free_ma3x(cotvar,1,maxwav,1,ntv+nqtv,firstobs,lastobs);
   free_matrix(covar,0,NCOVMAX,1,n);    if(nqtv>=1)free_ma3x(cotqvar,1,maxwav,1,nqtv,firstobs,lastobs);
     if(nqv>=1)free_matrix(coqvar,1,nqv,firstobs,lastobs);
     free_matrix(covar,0,NCOVMAX,firstobs,lastobs);
   free_matrix(matcov,1,npar,1,npar);    free_matrix(matcov,1,npar,1,npar);
     free_matrix(hess,1,npar,1,npar);
   /*free_vector(delti,1,npar);*/    /*free_vector(delti,1,npar);*/
   free_ma3x(delti3,1,nlstate,1, nlstate+ndeath-1,1,ncovmodel);     free_ma3x(delti3,1,nlstate,1, nlstate+ndeath-1,1,ncovmodel); 
   free_matrix(agev,1,maxwav,1,imx);    free_matrix(agev,1,maxwav,1,imx);
     free_ma3x(paramstart,1,nlstate,1, nlstate+ndeath-1,1,ncovmodel);
   free_ma3x(param,1,nlstate,1, nlstate+ndeath-1,1,ncovmodel);    free_ma3x(param,1,nlstate,1, nlstate+ndeath-1,1,ncovmodel);
   if (mobilav!=0) free_ma3x(mobaverage,1, AGESUP,1,NCOVMAX, 1,NCOVMAX);    
   free_ma3x(probs,1,AGESUP,1,NCOVMAX, 1,NCOVMAX);    free_ivector(ncodemax,1,NCOVMAX);
     free_ivector(ncodemaxwundef,1,NCOVMAX);
   free_ivector(ncodemax,1,8);    free_ivector(Dummy,-1,NCOVMAX);
   free_ivector(Tvar,1,15);    free_ivector(Fixed,-1,NCOVMAX);
   free_ivector(Tprod,1,15);    free_ivector(DummyV,1,NCOVMAX);
   free_ivector(Tvaraff,1,15);    free_ivector(FixedV,1,NCOVMAX);
   free_ivector(Tage,1,15);    free_ivector(Typevar,-1,NCOVMAX);
   free_ivector(Tcode,1,100);    free_ivector(Tvar,1,NCOVMAX);
     free_ivector(TvarsQ,1,NCOVMAX);
   /*  fclose(fichtm);*/    free_ivector(TvarsQind,1,NCOVMAX);
   /*  fclose(ficgp);*/ /* ALready done */    free_ivector(TvarsD,1,NCOVMAX);
       free_ivector(TnsdVar,1,NCOVMAX);
     free_ivector(TvarsDind,1,NCOVMAX);
   if(erreur >0){    free_ivector(TvarFD,1,NCOVMAX);
     printf("End of Imach with error or warning %d\n",erreur);    free_ivector(TvarFDind,1,NCOVMAX);
     fprintf(ficlog,"End of Imach with error or warning %d\n",erreur);    free_ivector(TvarF,1,NCOVMAX);
     free_ivector(TvarFind,1,NCOVMAX);
     free_ivector(TvarV,1,NCOVMAX);
     free_ivector(TvarVind,1,NCOVMAX);
     free_ivector(TvarA,1,NCOVMAX);
     free_ivector(TvarAind,1,NCOVMAX);
     free_ivector(TvarFQ,1,NCOVMAX);
     free_ivector(TvarFQind,1,NCOVMAX);
     free_ivector(TvarVD,1,NCOVMAX);
     free_ivector(TvarVDind,1,NCOVMAX);
     free_ivector(TvarVQ,1,NCOVMAX);
     free_ivector(TvarVQind,1,NCOVMAX);
     free_ivector(Tvarsel,1,NCOVMAX);
     free_vector(Tvalsel,1,NCOVMAX);
     free_ivector(Tposprod,1,NCOVMAX);
     free_ivector(Tprod,1,NCOVMAX);
     free_ivector(Tvaraff,1,NCOVMAX);
     free_ivector(invalidvarcomb,1,ncovcombmax);
     free_ivector(Tage,1,NCOVMAX);
     free_ivector(Tmodelind,1,NCOVMAX);
     free_ivector(TmodelInvind,1,NCOVMAX);
     free_ivector(TmodelInvQind,1,NCOVMAX);
   
     free_matrix(precov, 1,MAXRESULTLINESPONE,1,NCOVMAX+1); /* Could be elsewhere ?*/
   
     free_imatrix(nbcode,0,NCOVMAX,0,NCOVMAX);
     /* free_imatrix(codtab,1,100,1,10); */
     fflush(fichtm);
     fflush(ficgp);
     
     
     if((nberr >0) || (nbwarn>0)){
       printf("End of Imach with %d errors and/or %d warnings. Please look at the log file for details.\n",nberr,nbwarn);
       fprintf(ficlog,"End of Imach with %d errors and/or warnings %d. Please look at the log file for details.\n",nberr,nbwarn);
   }else{    }else{
    printf("End of Imach\n");      printf("End of Imach\n");
    fprintf(ficlog,"End of Imach\n");      fprintf(ficlog,"End of Imach\n");
   }    }
   printf("See log file on %s\n",filelog);    printf("See log file on %s\n",filelog);
   fclose(ficlog);  
   /*  gettimeofday(&end_time, (struct timezone*)0);*/  /* after time */    /*  gettimeofday(&end_time, (struct timezone*)0);*/  /* after time */
       /*(void) gettimeofday(&end_time,&tzp);*/
   /* printf("Total time was %d Sec. %d uSec.\n", end_time.tv_sec -start_time.tv_sec, end_time.tv_usec -start_time.tv_usec);*/    rend_time = time(NULL);  
   /*printf("Total time was %d uSec.\n", total_usecs);*/    end_time = *localtime(&rend_time);
     /* tml = *localtime(&end_time.tm_sec); */
     strcpy(strtend,asctime(&end_time));
     printf("Local time at start %s\nLocal time at end   %s",strstart, strtend); 
     fprintf(ficlog,"Local time at start %s\nLocal time at end   %s\n",strstart, strtend); 
     printf("Total time used %s\n", asc_diff_time(rend_time -rstart_time,tmpout));
     
     printf("Total time was %.0lf Sec.\n", difftime(rend_time,rstart_time));
     fprintf(ficlog,"Total time used %s\n", asc_diff_time(rend_time -rstart_time,tmpout));
     fprintf(ficlog,"Total time was %.0lf Sec.\n", difftime(rend_time,rstart_time));
     /*  printf("Total time was %d uSec.\n", total_usecs);*/
   /*   if(fileappend(fichtm,optionfilehtm)){ */
     fprintf(fichtm,"<br>Local time at start %s<br>Local time at end   %s<br>\n</body></html>",strstart, strtend);
     fclose(fichtm);
     fprintf(fichtmcov,"<br>Local time at start %s<br>Local time at end   %s<br>\n</body></html>",strstart, strtend);
     fclose(fichtmcov);
     fclose(ficgp);
     fclose(ficlog);
   /*------ End -----------*/    /*------ End -----------*/
     
   
   end:  /* Executes gnuplot */
 #ifdef windows    
   /* chdir(pathcd);*/    printf("Before Current directory %s!\n",pathcd);
   #ifdef WIN32
     if (_chdir(pathcd) != 0)
       printf("Can't move to directory %s!\n",path);
     if(_getcwd(pathcd,MAXLINE) > 0)
   #else
       if(chdir(pathcd) != 0)
         printf("Can't move to directory %s!\n", path);
     if (getcwd(pathcd, MAXLINE) > 0)
 #endif   #endif 
  /*system("wgnuplot graph.plt");*/      printf("Current directory %s!\n",pathcd);
  /*system("../gp37mgw/wgnuplot graph.plt");*/    /*strcat(plotcmd,CHARSEPARATOR);*/
  /*system("cd ../gp37mgw");*/    sprintf(plotcmd,"gnuplot");
  /* system("..\\gp37mgw\\wgnuplot graph.plt");*/  #ifdef _WIN32
   strcpy(plotcmd,GNUPLOTPROGRAM);    sprintf(plotcmd,"\"%sgnuplot.exe\"",pathimach);
   strcat(plotcmd," ");  #endif
   strcat(plotcmd,optionfilegnuplot);    if(!stat(plotcmd,&info)){
   printf("Starting graphs with: %s",plotcmd);fflush(stdout);      printf("Error or gnuplot program not found: '%s'\n",plotcmd);fflush(stdout);
   system(plotcmd);      if(!stat(getenv("GNUPLOTBIN"),&info)){
   printf(" Wait...");        printf("Error or gnuplot program not found: '%s' Environment GNUPLOTBIN not set.\n",plotcmd);fflush(stdout);
       }else
  /*#ifdef windows*/        strcpy(pplotcmd,plotcmd);
   #ifdef __unix
       strcpy(plotcmd,GNUPLOTPROGRAM);
       if(!stat(plotcmd,&info)){
         printf("Error gnuplot program not found: '%s'\n",plotcmd);fflush(stdout);
       }else
         strcpy(pplotcmd,plotcmd);
   #endif
     }else
       strcpy(pplotcmd,plotcmd);
     
     sprintf(plotcmd,"%s %s",pplotcmd, optionfilegnuplot);
     printf("Starting graphs with: '%s'\n",plotcmd);fflush(stdout);
     strcpy(pplotcmd,plotcmd);
     
     if((outcmd=system(plotcmd)) != 0){
       printf("Error in gnuplot, command might not be in your path: '%s', err=%d\n", plotcmd, outcmd);
       printf("\n Trying if gnuplot resides on the same directory that IMaCh\n");
       sprintf(plotcmd,"%sgnuplot %s", pathimach, optionfilegnuplot);
       if((outcmd=system(plotcmd)) != 0){
         printf("\n Still a problem with gnuplot command %s, err=%d\n", plotcmd, outcmd);
         strcpy(plotcmd,pplotcmd);
       }
     }
     printf(" Successful, please wait...");
   while (z[0] != 'q') {    while (z[0] != 'q') {
     /* chdir(path); */      /* chdir(path); */
     printf("\nType e to edit output files, g to graph again, c to start again, and q for exiting: ");      printf("\nType e to edit results with your browser, g to graph again and q for exit: ");
     scanf("%s",z);      scanf("%s",z);
     if (z[0] == 'c') system("./imach");  /*     if (z[0] == 'c') system("./imach"); */
     else if (z[0] == 'e') system(optionfilehtm);      if (z[0] == 'e') {
   #ifdef __APPLE__
         sprintf(pplotcmd, "open %s", optionfilehtm);
   #elif __linux
         sprintf(pplotcmd, "xdg-open %s", optionfilehtm);
   #else
         sprintf(pplotcmd, "%s", optionfilehtm);
   #endif
         printf("Starting browser with: %s",pplotcmd);fflush(stdout);
         system(pplotcmd);
       }
     else if (z[0] == 'g') system(plotcmd);      else if (z[0] == 'g') system(plotcmd);
     else if (z[0] == 'q') exit(0);      else if (z[0] == 'q') exit(0);
   }    }
   /*#endif */  end:
     while (z[0] != 'q') {
       printf("\nType  q for exiting: "); fflush(stdout);
       scanf("%s",z);
     }
     printf("End\n");
     exit(0);
 }  }
   
   
Removed from v.1.84  
changed lines
  Added in v.1.335

FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>