lduMatrix and LduMatrix

fetc95 · December 11, 2021, 20:39

Hello, I tried to find the information in the web and in this forum but did not find anything.

In the OpenFOAM source code, there are two big classes: lduMatrix and LduMatrix. Does anybody know what is the difference between them?

HPE · December 12, 2021, 06:46

LduMatrix is a class template. The classes starting with a capital letter in OpenFOAM are usually class templates.

dlahaye · December 12, 2021, 09:53

@HPE: thank you for sharing your insight.

Does the Ldumatrix class template allow to implement complex-valued (i.e. with non-zero imaginary part) matrices?

olesen · December 12, 2021, 13:35

Quote:

Originally Posted by dlahaye

@HPE: thank you for sharing your insight.

Does the Ldumatrix class template allow to implement complex-valued (i.e. with non-zero imaginary part) matrices?

In theory yes, but I would expect it to break in several places elsewhere - at the latest when doing the linear solve. Will also fall apart without processor boundary handling for complex types. IMO need to treat this as a development topic at the moment.

dlahaye · December 12, 2021, 14:40

Aha! Is there any other use of templating Ldumatrix that you can think off. Would single precision real matrices work using this template?

olesen · December 13, 2021, 06:24

Quote:

Originally Posted by dlahaye

Aha! Is there any other use of templating Ldumatrix that you can think off. Would single precision real matrices work using this template?

Single or double precision are no problem per se, since this is what the various WM_PRECISION_OPTION values do (SP, DP, SPDP). However, the backend linear solvers also need to be available for the corresponding precision. If, for example, you compile OpenFOAM with WM_PRECISION_OPTION=SP (instead of the default DP), then all fields and solvers are single-precision and it all will work. You can even compile with SPDP, which means that all of the fields are handled in single-precision, but the linear solve is actually running in double-precision. This gives the benefit of lower memory bandwidth for field operations, but higher precision for the linear solve which will normally have better/faster convergence.
If you are trying to have a single-precision solve with double-precision for all of the fields (ie, a non-existent DPSP mode), you are out of luck.

I think it is time that you explain what you are trying to solve.

dlahaye · December 15, 2021, 02:30

Dear Mark,

Thank you for asking me to clarify my doubts.

I am trying to solve doubts I have related to the computational efficiency of laplacianFoam and scalarTransportFoam (to start with). These doubts are related to the efficiency of the iterative solvers (both sequential and parallel).

In short:

1/ in laplacianFoam I would to compare GAMG with genuine AMG on unstructured grids;

2/ in scalarTransportFoam I would like to compare current solvers with ILU with drop tolerance after AMD reordering.

You pointed me earlier to work done using PETSc.

I need to go back and read more documentation. I will get back to you here.

Again, thank you for thinking along. Cheers, Domenico.

December 11, 2021, 20:39	lduMatrix and LduMatrix	#1
fetc95 New Member Russian Federation Join Date: Apr 2020 Posts: 18 Rep Power: 6	Hello, I tried to find the information in the web and in this forum but did not find anything. In the OpenFOAM source code, there are two big classes: lduMatrix and LduMatrix. Does anybody know what is the difference between them?

December 12, 2021, 06:46		#2
HPE Senior Member Herpes Free Engineer Join Date: Sep 2019 Location: The Home Under The Ground with the Lost Boys Posts: 932 Rep Power: 12	LduMatrix is a class template. The classes starting with a capital letter in OpenFOAM are usually class templates. dlahaye and yueyun like this. __________________ The OpenFOAM community is the biggest contributor to OpenFOAM: User guide/Wiki-1/Wiki-2/Code guide/Code Wiki/Journal Nilsson/Guerrero/Holzinger/Holzmann/Nagy/Santos/Nozaki/Jasak/Primer Governance Bugs/Features: OpenFOAM (ESI-OpenCFD-Trademark) Bugs/Features: FOAM-Extend (Wikki-FSB) Bugs: OpenFOAM.org How to create a MWE New: Forkable OpenFOAM mirror

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
What is LduMatrix template class for in comparison to lduMatrix class?	doctorWho	OpenFOAM Programming & Development	2	August 14, 2018 05:41
Create new member variable for lduMatrix	FabianL	OpenFOAM	0	December 13, 2017 04:39
Why renumbering works for LduMatrix?	chengdi	OpenFOAM	4	July 31, 2017 18:54
lduMatrix problem in High order scheme	rxgrch	OpenFOAM Programming & Development	3	April 2, 2013 10:30
ldumatrix	zaratustra	OpenFOAM Programming & Development	6	August 20, 2009 05:48

December 12, 2021, 09:53		#3
dlahaye Senior Member Domenico Lahaye Join Date: Dec 2013 Posts: 723 Blog Entries: 1 Rep Power: 17	@HPE: thank you for sharing your insight. Does the Ldumatrix class template allow to implement complex-valued (i.e. with non-zero imaginary part) matrices?

December 12, 2021, 14:40		#5
dlahaye Senior Member Domenico Lahaye Join Date: Dec 2013 Posts: 723 Blog Entries: 1 Rep Power: 17	Aha! Is there any other use of templating Ldumatrix that you can think off. Would single precision real matrices work using this template?

December 15, 2021, 02:30		#7
dlahaye Senior Member Domenico Lahaye Join Date: Dec 2013 Posts: 723 Blog Entries: 1 Rep Power: 17	Dear Mark, Thank you for asking me to clarify my doubts. I am trying to solve doubts I have related to the computational efficiency of laplacianFoam and scalarTransportFoam (to start with). These doubts are related to the efficiency of the iterative solvers (both sequential and parallel). In short: 1/ in laplacianFoam I would to compare GAMG with genuine AMG on unstructured grids; 2/ in scalarTransportFoam I would like to compare current solvers with ILU with drop tolerance after AMD reordering. You pointed me earlier to work done using PETSc. I need to go back and read more documentation. I will get back to you here. Again, thank you for thinking along. Cheers, Domenico.