CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   OpenFOAM (http://www.cfd-online.com/Forums/openfoam/)
-   -   block solver already available in OpenFOAM? (http://www.cfd-online.com/Forums/openfoam/75984-block-solver-already-available-openfoam.html)

herbert May 11, 2010 06:39

block solver already available in OpenFOAM?
 
Hi FOAMers,

I've read a lot of times that a block matix solver was planned to be implemented. It should be something comparable to coupledMatrix but capable for multi-variable-coupling instead of multi-region-coupling.

Does anyone know about the proceeding for that. I'd like to use it for finite-element stress analysis, because when I'm solving the equations segregated, the solution converges far too slow (or even not at all).

Thanks,
Stefan

pbohorquez May 11, 2010 11:37

Let's see what is going on the "New Implementations in OpenFOAM" to be introduced at the "5th OpenFOAM Workshop". It seems that there are several talks planned on that topic.

Exciting!

hjasak May 12, 2010 16:17

Yes, it is already in use in a few projects (neutronics simulations, coupled multi-phase VOF) and to come out at the Workshop. There will be a presentation by Julia to show what Kathrin and Julia have done at the Summer School in Zagreb last year + I am seeing further work by Ivor Clifford in neutronics at Penn State.

In short: it is implemented, parts of it are due for release and we need to apply it to more cases.

Hrv

wang.zhy July 9, 2010 05:00

Hi Hrv
I downloaded your presentation "OpenFOAM: Year in Review"。and I found these big good news
New Features in Upcoming Release:
Block matrix implementation with parallelisation support
Re-meshing with tetrahedral edge swapping: Sandeep Menon
......

It may be rude ,but I can't help myself.
I really need it !
where and when could I find it ?

hjasak July 9, 2010 05:11

Well, I am finishing off the merge and 1.7-ext will follow up immediately afterwards. However, the block matrix is not so easy (have a look at the work Julia and Kathrin have done with it to have an idea of what to do).

As for Sandeep's contribution, this will slot right in, with tutorial examples etc. If we ask him nicely, Sandeep will give us some documentation as well :)

I will keep you posted,

Hrv

wang.zhy July 9, 2010 05:34

Thanks Hrv!

marupio July 9, 2010 18:01

Quote:

Originally Posted by wang.zhy (Post 266543)
Hi Hrv
I downloaded your presentation "OpenFOAM: Year in Review"。and ...

Really? I've been looking for it. Where can I find it? I was unable to go to the workshop, and was looking forward to reading that one.

-Dave

juho July 10, 2010 02:15

Quote:

Originally Posted by marupio (Post 266675)
Really? I've been looking for it. Where can I find it? I was unable to go to the workshop, and was looking forward to reading that one.

-Dave

You can find most of the presentations here:
http://web.student.chalmers.se/groups/ofw5/Program.htm

-Juho

alberto July 10, 2010 07:11

Quote:

Originally Posted by hjasak (Post 266545)
Well, I am finishing off the merge and 1.7-ext will follow up immediately afterwards. However, the block matrix is not so easy (have a look at the work Julia and Kathrin have done with it to have an idea of what to do).

I saw the slides. Very interesting work! If I understood it correctly, they
  • Estimate U as usual, with U = H/A.
  • Solve the coupled p, alpha1, alpha2 equations
  • Correct p and fluxes
  • "Re-evaluate" U
I have a couple of questions:
  1. Is the re-evaluation of U done reconstructing the flux (explicitly or implicitly?)?
  2. Would it be possible to deal with coupled U equations (like in FLUENT phase-coupled SIMPLE for Euler-Euler multi-fluid model)? In that algorithm, n equations for U are coupled to predict U, then a pressure equation based on total phase continuity is solved and phase velocities are corrected using a (point-implicit in FLUENT, but I'm not that worried about this point. I'm more interested in the coupling of U equations :-)) flux reconstruction.
Thanks

wang.zhy July 11, 2010 01:58

Be greedy ,It would be great if OpenFOAM-1.7-dev have a full coupled solver.There is paper "A coupled finite volume solver for the solution of incompressible flows on unstructured grids" in JCP .and I saw L. Mangani and C. Bianchini have accomplish this algrithm with a third party linear solver.In my view of their reslult ,both stablish and convergence have a huge enhancement compare to segregated solver.I guess computation cost will also be saving .

In my test,for a turbulence Flat plate case (like the case in http://www.lerc.nasa.gov/WWW/wind/valid/archive.html),

In general CFX 11 need about 60 iteration to achieve a 1e-6 convergence ,while OpenFOAM1.6 need 3000 or more iteration to reach the same convergence .

CFX11 use a full coupled solver,and OpenFOAM 1.6 use segragated solver ,maybe that is the main cause of the difference .

marupio September 29, 2010 11:43

Has any progress been made on the block matrix solvers? I have a biochemistry model with 30 scalars that need to be coupled with the transport equations. I'm currently using an ugly mix of an ODE solver with standard transport.

nileshjrane December 30, 2010 15:18

Hello FOAMers..

First of all,
Wishing you all a very Happy new year, 2011...May we all as the open source CFD community prosper in the coming year like non ever before...:)

My question to Hrv,

Where can i find the literature regarding the block matrix solver?? I am currently looking for the various options for coupled solvers...

Thank you,
Nilesh

hjasak December 30, 2010 15:21

M&C2009 paper by Ivor Clifford
ECCOMAS 2010 by Kathrin Kissling and Julia Springer

Enjoy,

Hrv

nileshjrane December 30, 2010 16:13

That was quick..:D

I have one more question. I have worked on a research code. The code is like, all the governing equations (including turbulence, species conservation equations, momentum and energy) written in vector form in transformed co-ordinates. Then a Jacobian is formed and inverted using a specialised (for parallel computation) algorithm. This is fully coupled solver system. Looking at the present state of OpenFoam, is it possible to model such solver in it?? i mean does it have the required machinery (building blocks) for that?? i am concerned with this because this would be the ideal way (may not be for industry but for research at least) of solving complex flows like hypersonic reactive flow etc. Please correct me if i am wrong.

And if i am not guessing incorrectly, the block matrix solvers have clear advantage over such implicit methods in terms of computational costs.


Regards,
Nilesh


All times are GMT -4. The time now is 20:04.