Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Solver Behavior Issue. Convergence Plot Advice.

Please login with a confirmed email address before reporting spam

Hi folks,

I have a simulation that is somewhat converging but I wonder about the behavior of the solver...

Please see attached picture for the convergence plot.

The segregated group keeps exhibiting the same behavior. Can anyone offer some advice or insight on this matter?

Is this mesh related?

Best Regards,

Rui Silva


9 Replies Last Post 6 lug 2012, 11:51 GMT-4
Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 18 apr 2012, 01:57 GMT-4
Hi

not really , looks like a large simulations with slow convergence, so perhaps ma following comment does not apply:

I have notice that often, when I have enough RAM, to use the direct solver (and no segregation steps) I often see COMSOL converge quicker

--
Good luck
Ivar
Hi not really , looks like a large simulations with slow convergence, so perhaps ma following comment does not apply: I have notice that often, when I have enough RAM, to use the direct solver (and no segregation steps) I often see COMSOL converge quicker -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 18 apr 2012, 06:50 GMT-4
Hello Ivar,

thank you for your reply.

Actually I also have noticed that. I have about 48 GB of ram on our cluster, and indeed, the Direct option is quicker most of the time.

I have a question, and hopefully you can answer: when we have a system that is slow converging or of difficult convergence is it better to use the "Fully Coupled" solver? Or is this option preferable for coupled multiphysics and PDEs?

Best Regards,

Rui Silva
Hello Ivar, thank you for your reply. Actually I also have noticed that. I have about 48 GB of ram on our cluster, and indeed, the Direct option is quicker most of the time. I have a question, and hopefully you can answer: when we have a system that is slow converging or of difficult convergence is it better to use the "Fully Coupled" solver? Or is this option preferable for coupled multiphysics and PDEs? Best Regards, Rui Silva

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 18 apr 2012, 16:05 GMT-4
Hi

I have no definitive answer there, sometimes you cannot avoid the segregated solver, => case clear.

I normally leave the default COMSOl settings, but If I feel I'm waiting" too long" and the segregated solver is turned on, without that I see any good reason why one should segregate, then I try a coupled case (setting it manually), either it solves or it stops rather quickly, then I revert back to segregated if nothing else works, and take a few Forum quiz the time COMSOL finishes ;)

--
Good luck
Ivar
Hi I have no definitive answer there, sometimes you cannot avoid the segregated solver, => case clear. I normally leave the default COMSOl settings, but If I feel I'm waiting" too long" and the segregated solver is turned on, without that I see any good reason why one should segregate, then I try a coupled case (setting it manually), either it solves or it stops rather quickly, then I revert back to segregated if nothing else works, and take a few Forum quiz the time COMSOL finishes ;) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 18 apr 2012, 16:21 GMT-4
Segregated solver has many more "knobs" to tune. The settings are more task-dependent as compared to direct solver.
First thing I check with segregated solver is the order in which the equations are solved. COMSOL defaults are sometimes very odd...
Segregated solver has many more "knobs" to tune. The settings are more task-dependent as compared to direct solver. First thing I check with segregated solver is the order in which the equations are solved. COMSOL defaults are sometimes very odd...

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 19 apr 2012, 06:06 GMT-4
Hi
Indeed a good point, specially when you have fiddled a lot with your model ;)

--
Good luck
Ivar
Hi Indeed a good point, specially when you have fiddled a lot with your model ;) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 5 lug 2012, 13:46 GMT-4
Hello Ivar,

I have been working on other issues ans I have returned to question in this post.

Thank you for your suggestions.

However, the problem persists (see attached pic1).

So I used the Fully Coupled + Direct Solver as you suggested (see attached pic2).

Is there a way to somehow give a numerical "nudge" to achieve proper convergence. I have refined the mesh, but
still nothing. I also changed the Damping Factor to a much lower value but I get the same behaviour...

Any thoughts?

Best Regards,

Rui
Hello Ivar, I have been working on other issues ans I have returned to question in this post. Thank you for your suggestions. However, the problem persists (see attached pic1). So I used the Fully Coupled + Direct Solver as you suggested (see attached pic2). Is there a way to somehow give a numerical "nudge" to achieve proper convergence. I have refined the mesh, but still nothing. I also changed the Damping Factor to a much lower value but I get the same behaviour... Any thoughts? Best Regards, Rui


Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 6 lug 2012, 02:53 GMT-4
Hi

well the only I can suggest is that looking at your segregated group convergence, one group is getting "rapidly" stale while the other is slower. This means perhaps that you could solve for part of the variables, and use that asa fixed value and solve only for the other dependent variables, this should make the solving go faster, but you might integrate away too, so its not always the best way, finsihing in a global run with all dependnet variables, using the last solver step is a way to check that you are a t a optimum result point

--
Good luck
Ivar
Hi well the only I can suggest is that looking at your segregated group convergence, one group is getting "rapidly" stale while the other is slower. This means perhaps that you could solve for part of the variables, and use that asa fixed value and solve only for the other dependent variables, this should make the solving go faster, but you might integrate away too, so its not always the best way, finsihing in a global run with all dependnet variables, using the last solver step is a way to check that you are a t a optimum result point -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 6 lug 2012, 03:07 GMT-4
Hi,

I assume that segregated group 2 is for turbulent parameters (k and eps)? If yes, try changing turbulent length scale from inlet. Good default values is 0.07*inlet diameter. If this doesn't work try to solve problem with very coarse mesh and use this as start value for finer mesh calculation.

Best regards

Tero
Hi, I assume that segregated group 2 is for turbulent parameters (k and eps)? If yes, try changing turbulent length scale from inlet. Good default values is 0.07*inlet diameter. If this doesn't work try to solve problem with very coarse mesh and use this as start value for finer mesh calculation. Best regards Tero

Nagi Elabbasi Facebook Reality Labs

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 6 lug 2012, 11:51 GMT-4
The lack of convergence is not always due to solver settings. It may be for example that you are using a stationary solver for a problem that does not have a stationary solution. That happens frequently in natural convection fluid/thermal problems and in solids problems involving contact.

Nagi Elabbasi
Veryst Engineering
The lack of convergence is not always due to solver settings. It may be for example that you are using a stationary solver for a problem that does not have a stationary solution. That happens frequently in natural convection fluid/thermal problems and in solids problems involving contact. Nagi Elabbasi Veryst Engineering

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.