forum.alglib.net

Dear all,

I am trying to understand how it works the SKS format. Since I am obtaining incorrect results, I tried first to create a normal NxN sparse matrix, and then convert it to sks format. However, I am still getting incongruous results. Here is the piece of code (C#) that I am using for testing:

alglib.sparsematrix ss = new alglib.sparsematrix();
int nndof = 3;
alglib.sparsecreate(nndof, nndof,out ss);
alglib.sparseset(ss, 0, 0, 1.0);
alglib.sparseset(ss, 1, 1, 20.0);
alglib.sparseset(ss, 2, 2, 10.0);
alglib.sparseset(ss, 1, 2, -10.0);
alglib.sparseset(ss, 2, 1, -10.0);
alglib.sparsesolverreport report2 = new alglib.sparsesolverreport();
double[] force2 = new double[nndof];
force2[2] = -10.0;
double[] disp2 = new double[nndof];
alglib.sparsesolve(ss, force2, out disp2, out report2);
double[] disp3 = new double[nndof];
alglib.sparseconverttosks(ss);
alglib.sparsespdcholeskysolve(ss, true, force2, out disp3, out report2);

Results:
Disp2 = [0, -1, -2] (Correct)
Disp3 = [0, -0.05, -0.1] (Incorrect)

Can you please help me to understand what I am doing wrong? Many thanks.

Best regards,
Gabbia

Hi!

The reason is that spdcholeskysolve() family of functions expects its data to be not the matrix itself, but its Cholesky decomposition. That's why it gives wrong result - you passed original matrix which is used as if it was a Cholesky factor.

You should use:
* sparsespdsolvesks to solve SPD system in SKS format, usually low profile
* sparsespdsolve to solve general SPD system
* sparsesolve to solve general nonsymmetric system

Many thanks!