[scilab-Users] code which does not work

[François Vogel]

Ivan Maximov said on 18/03/2009 17:52:
> krotov_csa.sci
> __________________________________________________________________________________________________________
> function [ef,u1,u2,u3,u4]=krotov_csa(x1,x2,x3,x4)
>  
>   u1=x1; u2=x2; u3=x3; u4=x4;
>   t1=x1; t2=x2; t3=x3; t4=x4;
>   
>   PROP=uu;
>   TE=uu;
>   pause;
>   
>   for i=1:N,
>     for j=1:csN,
>         PROP=expm(-%i*dt*(Hij(:,:,j)+Hamilt+H1(u1(i),u2(i),u3(i),u4(i))));
>         TE=PROP*U(:,:,j);
>         U(:,:,j)=TE;
>     end
>   end

This looks very much like a bug of Scilab when inserting a matrix in 
hypermatrices.

The problem is that after the first execution (i=1 and j=1) of:
>         PROP=expm(-%i*dt*(Hij(:,:,j)+Hamilt+H1(u1(i),u2(i),u3(i),u4(i))));
>         TE=PROP*U(:,:,j);
>         U(:,:,j)=TE;

something goes wrong in U: size(U) becomes 4x4, i.e. U becomes a 4x4 
matrix. It looses its third dimension.

The strange thing is that this does not happen whatever the context, 
and this explains why it works sometimes with the debugger (but I saw 
the problem also when debugging in Scipad). My guess is that global 
variables are involved, and perhaps execution contexts of your several 
exec instructions.

The problem can be easier seen when instrumenting the code a bit like 
this in function krotov_csa lines 10 and below:

for i=1:N,
  mprintf("i is: %d\n",i)
  for j=1:csN,
   mprintf("j is: %d\n",j)
   PROP=expm(-%i*dt*(Hij(:,:,j)+Hamilt+H1(u1(i),u2(i),u3(i),u4(i))));
   TE=PROP*U(:,:,j);
   disp("BEFORE");disp(size(U));
   U(:,:,j)=TE;
   disp("AFTER");disp(size(U))
  end
end


Now upon execution from Scilab shell:

-->exec main.sce;

  START GENERAL
i is: 1
j is: 1

  BEFORE

     4.    4.    155.

  AFTER

     4.    4.
j is: 2
  !--error 21
invalid index


My opinion is that Scilab finds out that U(:,:,j) is a matrix and no 
longer an hypermatrix (since the third dimension is exactly 1), and in 
this case it reduces the number of dimensions of U(:,:,j). The problem 
is that then Scilab erroneously affects U(:,:,j) into U, reducing the 
dimension of U in the process too.

I think that this deserves a bug report in the bugzilla:
http://bugzilla.scilab.org/

However it would help if you could trim down the problem to a simpler 
script. Thanks in advance.

Francois