topic Re: oneMKL PARDISO complex matrix pivoting perturbation in Intel® oneAPI Math Kernel Library

oneMKL PARDISO complex matrix pivoting perturbation

AnttiR — Fri, 13 Oct 2023 10:17:42 GMT

I have been struggling with getting correct results with oneMKL PARDISO mtype=13 (complex and nonsymmetric matrix). The algorithm the solution is a part of is known to handle this type of matrixes, and the results are consistent with Python between numpy.linalg.solve and scipy.linalg.spsolve.

I had one specific problem with a small 4x4 matrix with the solve producing totally incorrect results. I accidentally discovered that setting iparm(10) (Fortran) to a very large value or 0 solved the problem as the number of perturbed pivots changed from one to zero (with the default value being 13).

My actual question is that is iparm(10)=0 a valid parameter for disabling the pivoting perturbation? It is not evident from the documentation (https://www.intel.com/content/www/us/en/docs/onemkl/developer-reference-fortran/2023-2/pardiso-iparm-parameter.html) if this is a valid entry or not.

Re:oneMKL PARDISO complex matrix pivoting perturbation

IntelSupport — Wed, 18 Oct 2023 13:56:21 GMT

Hi,

Thanks for posting in Intel Communities.

Could you please provide us a sample reproducer, MKL version and OS details, so that we can replicate and investigate more at our end. Please confirm, is it okay to use C as well as Fortran?

Regards,

Jilani

Re: oneMKL PARDISO complex matrix pivoting perturbation

JilaniS_Intel — Thu, 26 Oct 2023 02:47:24 GMT

Hi,

A gentle reminder:

Could you please provide us a sample reproducer, MKL version and OS details, so that we can replicate and investigate more at our end. Please confirm, is it okay to use C as well as Fortran?

Regards,

Jilani

Re:oneMKL PARDISO complex matrix pivoting perturbation

JilaniS_Intel — Wed, 01 Nov 2023 08:55:56 GMT

Hi,

A gentle reminder:

We have not heard back from you. Could you please provide us with an update?

Regards,

Jilani