Re: ITSOL interface

[Kai Torben Ohlhus]

On 22 July 2013 00:00, Nir Krakauer <address@hidden> wrote:

> Only the implementation of ILUC for the complex case (part of ZITSOL) has
> been removed for unknown reasons. Should I add this method to ZITSOL via an
> patch as a workaround for the moment? Basically, the difference between real
> and complex versions of the solvers is the data type (double vs. complex
> double).

Yes, please try that. 

[...] 

Nir

I "succeeded" with a complex ILUC (https://docs.google.com/file/d/0B4V9W1l15-xMRjBXLTFTOVUwZnM/edit?usp=sharing). The user gets warned at the complex function call, that this is an experimental function. This naive implementation doesn't last for too long hopefully. Compared to MATLAB my implementation seems less reliable. When I execute this code:

[code]

clear all

clc

setup.type = 'crout';

setup.milu = 'off';

setup.droptol = 1e-4;

A = sprand(200,200,0.01);

A = A + speye(200);

A(1,3) = 10 + 1i;

[L,U] = ilu(A, setup);

norm(L*U - A, 1)

[/code]

MATLAB produces absolute errors about 10^-3 to 10^-4 and my implementation 10^-1 to 10^4. For small test cases 5x5 the results differ in about 2 elements. I'm a bit unhappy about this, but I don't want to spend more time on complex ILUC. I'm sure Prof. Saad has already a better implementation. Now I continue with my other tasks.

Kai