[Developers] bug in file fvar_m41.cpp

[John Sibert]

Doh!
Fixed that in rev 417

dave fournier wrote:
> dave fournier wrote:
>   
>> at line 11
>> replace
>>  x.elem_value(imin)=v.elem_value(1)/m.elem_value(imin,imin);
>> by
>>  x.elem_value(imin)=v.elem_value(imin)/m.elem_value(imin,imin);
>>
>>   
>>     
> But the adjoint code needs to be fixed as well.
>
> void dfbltsolve(void)
> {
>   verify_identifier_string("ww");
>   dvar_vector_position xpos=restore_dvar_vector_position();
>   dvar_vector_position vpos=restore_dvar_vector_position();
>   dvector v=restore_dvar_vector_value(vpos);
>   dvar_matrix_position mpos=restore_dvar_matrix_position();
>   banded_lower_triangular_dmatrix m=
>     restore_banded_lower_triangular_dvar_matrix_value(mpos);
>   verify_identifier_string("rt");
>   dvector dfx=
>     restore_dvar_vector_derivatives(xpos);
>
>   int bw=m.bandwidth();
>   int imin=m.indexmin();
>   int imax=m.indexmax();
>
>   banded_lower_triangular_dmatrix dfm(imin,imax,bw);
>   dvector x(imin,imax);
>   dvector dfv(imin,imax);
>   dfm.initialize();
>   dfv.initialize();
>   x(imin)=v(imin)/m(imin,imin);
>   dvector dfsum(imin,imax);
>   dfsum.initialize();
>   dvector ssum(imin,imax);
>   ssum.initialize();
>   int i;
>   for (i=imin+1;i<=imax;i++)
>   {
>     int jmin=admax(imin,i-bw+1);
>     for (int j=jmin;j<=i-1;j++)
>     {
>       ssum(i)+=m(i,j)*x(j);
>     }
>     x(i)=(v(i)-ssum(i))/m(i,i);
>   }
>
>   for (i=imax;i>=imin+1;i--)
>   {
>     int jmin=admax(imin,i-bw+1);
>     //x(i)=(v(i)-ssum(i))/m(i,i);
>     dfv(i)+=dfx(i)/m(i,i);
>     dfsum(i)-=dfx(i)/m(i,i);
>     dfm(i,i)-=dfx(i)*(v(i)-ssum(i))/(m(i,i)*m(i,i));
>     dfx(i)=0.0;
>     for (int j=i-1;j>=jmin;j--)
>     {
>       //ssum(i)+=m(i,j)*x(j);
>       dfm(i,j)+=dfsum(i)*x(j);
>       dfx(j)+=dfsum(i)*m(i,j);
>     }
>   }
>   //x(imin)=v(1)/m(imin,imin);
>   // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>   //dfm(imax,imax)=0.0;
>   // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>   dfv(imin)+=dfx(imin)/m(imin,imin);
>   dfm(imin,imin)-=dfx(imin)*v(imin)/(m(imin,imin)*m(imin,imin));
>   dfx(imin)=0.0;
>
>   dfm.save_dmatrix_derivatives(mpos);
>   dfv.save_dvector_derivatives(vpos);
> }
>
>
>   

-- 
Visit the ADMB project http://admb-project.org/