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 The information provided may be out of date.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.
different solvers simultaneously for different time-dependent PDEs in one multiphysics model
Posted Feb 25, 2011, 12:05 p.m. EST Plasma Physics, Fluid & Heat, Heat Transfer & Phase Change, Computational Fluid Dynamics (CFD), Microfluidics, Chemical Reaction Engineering, Modeling Tools & Definitions, Parameters, Variables, & Functions, Studies & Solvers, Structural Mechanics Version 3.5a, Version 4.0, Version 4.0a, Version 4.1 5 Replies
Please login with a confirmed email address before reporting spam
I get a Multiphysics model consisting of
1). Time-dependent PDE (subdomain) in general form for incompressible Navier-Stokes.
2). Several time-dependent PDEs(subdomain) in coefficient form for phase field.
3). Time-dependent PDE(subdomain) in coefficient form for mass balance.
It is supposed to solve all of these coupled PDEs simultaneously in the same one subdomain. Now my problem is that I cannot define different solvers for different PDEs. For example, I am forced to choose GMRES+SSOR(Precondiction) to this model. How can I choose a special solver for Navier-Stokes and GMRES+SSOR for the rests.
In programming, according to my knowledge, based on equations splitting methods, different solvers could be assigned to different PDEs. I mean for one time step, first we calculate Navier-Stokes with projection methods, or “fractional step” methods, then move to phase field PDEs with GMRES solver and finally move to mass balance PDE with CG solver. loops for another time steps until some moment, of course there will be a limit for time step which is either decided by explicit scheme or implicit BDF(k) scheme as a example.
Instead of equations splitting method, there is full discretization method. For my problem as one order ODE, when BDF(1) is adopted, a nonlinear problem arises. By Newton iterative method(or others), this problem could be transfered to linear system. So as you can see for time dependent problem, you are forced to chose one(only one!) linear system solver in COMSOL. It is obvious in Comsol that this full discretization method is adopted.
Could you please tell me how to fix this problem, any suggestion with experience for this similar problem is warmly welcome.
Thank you
1). Time-dependent PDE (subdomain) in general form for incompressible Navier-Stokes.
2). Several time-dependent PDEs(subdomain) in coefficient form for phase field.
3). Time-dependent PDE(subdomain) in coefficient form for mass balance.
It is supposed to solve all of these coupled PDEs simultaneously in the same one subdomain. Now my problem is that I cannot define different solvers for different PDEs. For example, I am forced to choose GMRES+SSOR(Precondiction) to this model. How can I choose a special solver for Navier-Stokes and GMRES+SSOR for the rests.
In programming, according to my knowledge, based on equations splitting methods, different solvers could be assigned to different PDEs. I mean for one time step, first we calculate Navier-Stokes with projection methods, or “fractional step” methods, then move to phase field PDEs with GMRES solver and finally move to mass balance PDE with CG solver. loops for another time steps until some moment, of course there will be a limit for time step which is either decided by explicit scheme or implicit BDF(k) scheme as a example.
Instead of equations splitting method, there is full discretization method. For my problem as one order ODE, when BDF(1) is adopted, a nonlinear problem arises. By Newton iterative method(or others), this problem could be transfered to linear system. So as you can see for time dependent problem, you are forced to chose one(only one!) linear system solver in COMSOL. It is obvious in Comsol that this full discretization method is adopted.
Could you please tell me how to fix this problem, any suggestion with experience for this similar problem is warmly welcome.
Thank you
5 Replies Last Post Oct 17, 2011, 4:28 a.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
It seems time dependent segregated mode could handle this kind of problem. Different solver could be applied with separated group. I read a report in CFD with this mode, and test has been made with this mode. Although result seems agree with my expect but I am not sure is exact what I need.
Any suggestion is welcome. Should I supply more information about my model so that you can understand?
Any suggestion is welcome. Should I supply more information about my model so that you can understand?
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
I cannot come with many tips as you are really coupling many things together, but have you really played with the v4.1 solver settings, segregation does not mean you solve one by one fully separated, only one iteration in a sequence and then you loop around. In V4.1 you can really select only some physics to some solvers and then make your own loop, but you need to understand which is driving what and test it out.
Finally how to validate such a model, indeed COMSOL always gives you nice rainbow views, beautiful for a presentation, but if they are wrong, then what?
Your client/supervisor will only get suspicious if you cannot give some "hard" validation data to support some points in your analysis
--
Good luck
Ivar
I cannot come with many tips as you are really coupling many things together, but have you really played with the v4.1 solver settings, segregation does not mean you solve one by one fully separated, only one iteration in a sequence and then you loop around. In V4.1 you can really select only some physics to some solvers and then make your own loop, but you need to understand which is driving what and test it out.
Finally how to validate such a model, indeed COMSOL always gives you nice rainbow views, beautiful for a presentation, but if they are wrong, then what?
Your client/supervisor will only get suspicious if you cannot give some "hard" validation data to support some points in your analysis
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi, Ivar, Thanks a lot for your reply.
I finished my work with V3.5a and usually I will rerun it with V4.1 because I am more familiar with V3.5a. I will try it in V4.1 under your suggestion.
Additional remarks about equation splitting method and full discretization method are(for one time step):
1) Equation splitting method with explicit scheme is usually adopted in time dependent problem to avoid nonlinearity with the PDEs in model solved serially.
2) The PDEs in model will be solved simultaneously with Full discretization method in time dependent problem.
As the core of Comsol is Newton iteration, the nonlinear problem will always be transfered to linear equations set(FEM), that is the reason why we choose solvers in drop-down menu of 'linear system solver'. The time stepping method in Comsol has only two choices--BDF or Generalized alpha. Both of them are implicit scheme(I think) which ensure stability and accuracy. But on the other side, which prevents us to apply equation splitting method to the model with explicit scheme for some PDEs mixed with implicit scheme for the other PDEs.
I hope more friends join this topics which maybe deep our understanding of multiphysic problem, Comsol and FEM.
I finished my work with V3.5a and usually I will rerun it with V4.1 because I am more familiar with V3.5a. I will try it in V4.1 under your suggestion.
Additional remarks about equation splitting method and full discretization method are(for one time step):
1) Equation splitting method with explicit scheme is usually adopted in time dependent problem to avoid nonlinearity with the PDEs in model solved serially.
2) The PDEs in model will be solved simultaneously with Full discretization method in time dependent problem.
As the core of Comsol is Newton iteration, the nonlinear problem will always be transfered to linear equations set(FEM), that is the reason why we choose solvers in drop-down menu of 'linear system solver'. The time stepping method in Comsol has only two choices--BDF or Generalized alpha. Both of them are implicit scheme(I think) which ensure stability and accuracy. But on the other side, which prevents us to apply equation splitting method to the model with explicit scheme for some PDEs mixed with implicit scheme for the other PDEs.
I hope more friends join this topics which maybe deep our understanding of multiphysic problem, Comsol and FEM.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Zhu,
I am a new use of COMOSL Multiphysics and trying to build a model in Electromagnetic heating domain. I will heat the fluid by electromagnetic waves and then see the (flow) of fluid as a result of this heating.
Can you please let me know which modules of COMSOL would be required to build such a model?
In addition, what boundary conditions need to be defined?
The problem is how can I see the fluid flowing in COMSOL?
In addition, if I want to see the equations being used or EMPLOY NEW EQUATIONS IN comsol 4.0, how can I do that?
Many thanks in advance for any help!
I am a new use of COMOSL Multiphysics and trying to build a model in Electromagnetic heating domain. I will heat the fluid by electromagnetic waves and then see the (flow) of fluid as a result of this heating.
Can you please let me know which modules of COMSOL would be required to build such a model?
In addition, what boundary conditions need to be defined?
The problem is how can I see the fluid flowing in COMSOL?
In addition, if I want to see the equations being used or EMPLOY NEW EQUATIONS IN comsol 4.0, how can I do that?
Many thanks in advance for any help!
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
I'm afraid I could not answer your first problem cause Fluid flow with electromagnetic is not my field. But I'm quite sure that there is a lot of examples for your problem by comsol, not only in the comsol books but also online.
From my experience, what I can tell you is that be familiar with the equations of the model, at most of time the build-in module in comsol Software is not enough, some modification should be made. I prefer to use PDE modes cause there is not existed model in Comsol for my problem.
I think you can start fistly by going through an example(samples) in comsol, by doing this, you would be able to acquire the deep impression of how comsol run, how post-processing is performed. so all of the rest problem will be obviously easy to you.
From my experience, what I can tell you is that be familiar with the equations of the model, at most of time the build-in module in comsol Software is not enough, some modification should be made. I prefer to use PDE modes cause there is not existed model in Comsol for my problem.
I think you can start fistly by going through an example(samples) in comsol, by doing this, you would be able to acquire the deep impression of how comsol run, how post-processing is performed. so all of the rest problem will be obviously easy to you.