[SOSlib-discuss] sparse Jacobi matrix, DENSE_ELEM, removing 0s

[Rainer M. <ra...@tb...>]

Hi there,

For large models with a very sparse Jacobi matrix, it is much more 
efficient to NOT print out all the DENSE_ELEM[] = 0 commands for the gcc 
compiler.

So I have added this to construction of the forward and adjoint Jacobi 
evaluation function. From looking at the results and the CVODES examples 
I think that it is OK, to just not initialize or set the DENSE_ELEM 
things for the 0 entries.


However, I guess that could also be very useful for the hard-coded 
R.H.S. functions (used by both evaluateAST and the arithmeticCompiler).

So instead of looping through the neq*neq matrix, we could generate some 
sort of list, which contains the correct Jacobi matrix equation (only 
for non-0 entris) and the indices i,j to which Jacobi element it refers 
to. Then just loop through this list instead of through neq*neq.


What do you think? Worth a try?

Rainer