[ADMB Users] Another sdreport mystery

[dave fournier]

Looking at this a bit more one sees that ylog and alog only show up here.
So the alog are like offsets. But one offset is already implicitly 
contained in
the ylogs so you only need one alog as below.  So get rid of alog(2).




  for (int j=1;j<=numcalyrs_d;j++){
                 // age 1-2
                 d_jk(j, 1)= 1 - mfexp(-mfexp( ylog(j) + alog(1) ) );

                 for (int k=2;k<=numages;k++){
                         //age 2-4
                         d_jk(j,k) = 1 - mfexp(-mfexp( ylog(j))+alog(2) );
                 }
Then as I said before, in all your logits the derivatvies get near 0 if
the value of the the magnitude of parameter is large. So add
a tiny penalty like

             f+=1.e-8*norm2(logitm2);
             f+=1.e-8*norm2(logitm3);
             f+=1.e-8*norm2(logits);

and the model runs real swell.