Hey Shawn,I was not too specific about what to change: you should set PreconditionerIterations to 3, not PreconditionerRelaxation. You can also set this in the input file, it overrides the settings of the problem file.
I attach my current working input file for your convenience.
Kind regards BerndOn Wed, 28 Mar 2012 12:42:13 -0700 (PDT)
Shawn Zhang <[email protected]> wrote:
Hi Bernd and All,Sorry if I appear to be sticky about the convergence issue, but it kept coming back and bite me.Coming back to this test problem, with the attached 1p project, I have the following issue.Standard test: worked.Increase grid size to 100x100, not converging. It worked after I, increase # iteration to 1000, reduce convergence criterion to 10e-6. Increase grid size to 1000x1000, not converging. I changed the linear solver from Dumux::BoxCGILU0Solver to Dumux::BoxCGSSORSolver, and PreconditionerRelaxation from 1.0 to 3.0 as suggested by Bernd, it is still not converging...I attached the project folder. Do you think this maybe related to the Oracle VBox based Debian installation?Thanks. -Shawn From: Bernd Flemisch <[email protected]>To: DuMuX User Forum <[email protected]> Sent: Tuesday, March 27, 2012 4:01 AM Subject: Re: [DuMuX] Permeability heterogeneity and grid resolutionHey Shawn,with the attached input file, it works for 1000x1000 and four orders difference. I increased the maximum number of iterations to 1000 and the residual reduction to 1e-6.Kind regards Bernd On Mon, 26 Mar 2012 11:44:21 -0700 (PDT) Shawn Zhang <[email protected]> wrote:Well, I celebrated too fast.My goal is to determine, in a system of two material with different perm, when one material can be treated as having negligible impact on pressure drop.The two parameters influencing this are, 1. Permeability difference between two materials. 2. Grid resolution.I use the 1p with lens test problem, and increasing the perm of the lens material.At 100x100 resolution, I derive the attached plot (pressure plot along y=0.5 line), which is saying that when the permeability difference is greater than 3 order of magnitude, the material with high permeability can be seen as open channel flow comparing to the low permeable phase, and the pressure drop is neglegible. Moreover, further increase of the perm difference will have a neglegible effect.At 10x10 resolution, we start seeing this when the perm difference is 2 order of magnitude.When I increase the resolution to 1000x1000, with the permeability difference 4 order of magnitude, the solver does not converge again.Is this the expected behavior, as I am trying to solve large discontinuity caused by the big perm difference?-Shawn From: Bernd Flemisch <[email protected]>To: DuMuX User Forum <[email protected]> Sent: Monday, March 26, 2012 12:45 PM Subject: Re: [DuMuX] Permeability heterogeneity and grid resolutionHey Shawn,here, it helps to change the linear solver (or rather the preconditioner from ILU to SSOR). This unfortunately is a compile-time choice done in the problem file. I attach my corresponding problem file, changes are in lines 83 and 86.Kind regards Bernd On Mon, 26 Mar 2012 09:00:18 -0700 (PDT) Shawn Zhang <[email protected]> wrote:Hi Bernd,I switched the permeability between lens and outside, which gives the non-convergence exception.I.e., permeability = 1e-12 permeabilityLens = 1e-10 Instead of From: Bernd Flemisch <[email protected]>To: DuMuX User Forum <[email protected]> Sent: Monday, March 26, 2012 11:08 AM Subject: Re: [DuMuX] Permeability heterogeneity and grid resolutionHey Shawn,ok, this is difficult since I cannot reproduce it. Can you please send the output of your run with 100x100 and the attached input file? I increased the verbosity of the linear solver, so maybe we can see more from there.Kind regards Bernd On Mon, 26 Mar 2012 07:53:03 -0700 (PDT) Shawn Zhang <[email protected]> wrote:Hi Bernd,Thanks a lot for your quick attention. However, with your input file, and grid resolution 100x100, I got the same exceptions. I also tried to increase ResidualReduction parameter even further without any luck.It puzzles me that the code runs at lower resolution but not at higher...-Shawn From: Bernd Flemisch <[email protected]>To: DuMuX User Forum <[email protected]> Sent: Saturday, March 24, 2012 6:07 AM Subject: Re: [DuMuX] Permeability heterogeneity and grid resolutionHey Shawn,thank you for reporting this. It helps to relax the threshold for the reduction of the residual that the linear solver is required to achieve. I attach a corresponding input file, where the parameter is adjusted in line 31.To me, the default threshold of 1e-12 seems to be very small. We will discuss whether the default value should be adjusted.Kind regards Bernd On Fri, 23 Mar 2012 12:36:05 -0700 (PDT) Shawn Zhang <[email protected]> wrote:Dear All,I ran into an issue of 1p model not converging. After a series of tests, I trace it to this canonical situation, For standard test/boxmodels/1p problem, if I increase the 2D grid resolution from 10x10 to 100x100, the solver will not converge, with the following error reported.This error, however, will go away if I reduce the perm difference between lens and exterior from 2 order of magnitude into 1 order of magnitude. This makes me speculate that the error is related to sharp gradient at lens/exterior interface caused by the heterogeneity at high resolution.I think the reproducer on standard test makes it likely a serious problem, though I will be very happily corrected.Truly appreciate your quick attention. Best, -Shawnroot@szdebian:/home/szhang/dumux/dumux-2.1.0/test/boxmodels/1p# ./test_1p -parameterFile test_1p.input Welcome aboard DuMuX airlines. Please fasten your seatbelts! Emergency exits are near the time integration.Initializing problem '1ptest' Writing result file for "1ptest" Newton iteration 1 done, relative error = 0.666666 Newton iteration 2 done, relative error = 5.45083e-06Solve: M deltax^k = rNewton: Caught exception: "NumericalProblem [newtonSolveLinear:../../../dumux/nonlinear/newtoncontroller.hh:398]: Linear solver did not converge" Newton solver did not converge with dt=1 seconds. Retrying with time step of 0.5 secondsNewton iteration 1 done, relative error = 0.666666 Newton iteration 2 done, relative error = 5.45083e-06Solve: M deltax^k = rNewton: Caught exception: "NumericalProblem [newtonSolveLinear:../../../dumux/nonlinear/newtoncontroller.hh:398]: Linear solver did not converge"....Newton solver did not converge with dt=0.0078125 seconds. Retrying with time step of 0.00390625 secondsNewton iteration 1 done, relative error = 0.666666 Newton iteration 2 done, relative error = 5.45083e-06Solve: M deltax^k = rNewton: Caught exception: "NumericalProblem [newtonSolveLinear:../../../dumux/nonlinear/newtoncontroller.hh:398]: Linear solver did not converge" Newton solver did not converge with dt=0.00390625 seconds. Retrying with time step of 0.00195312 secondsNewton iteration 1 done, relative error = 0.666666 Newton iteration 2 done, relative error = 5.45083e-06Solve: M deltax^k = rNewton: Caught exception: "NumericalProblem [newtonSolveLinear:../../../dumux/nonlinear/newtoncontroller.hh:398]: Linear solver did not converge" Newton solver did not converge with dt=0.00195312 seconds. Retrying with time step of 0.000976562 seconds Dune reported error: Dune::MathError [timeIntegration:../../../dumux/boxmodels/common/boxproblem.hh:485]: Newton solver didn't converge after 10 time-step divisions. dt=0.000976562___________________________________________________________Bernd Flemisch phone: +49 711 685 69162 IWS, Universitaet Stuttgart fax: +49 711 685 67020 Pfaffenwaldring 61 email: [email protected] D-70569 Stuttgart url: www.hydrosys.uni-stuttgart.de___________________________________________________________ _______________________________________________ Dumux mailing list [email protected] https://listserv.uni-stuttgart.de/mailman/listinfo/dumux___________________________________________________________Bernd Flemisch phone: +49 711 685 69162 IWS, Universitaet Stuttgart fax: +49 711 685 67020 Pfaffenwaldring 61 email: [email protected] D-70569 Stuttgart url: www.hydrosys.uni-stuttgart.de___________________________________________________________ _______________________________________________ Dumux mailing list [email protected] https://listserv.uni-stuttgart.de/mailman/listinfo/dumux permeability = 1e-10 permeabilityLens = 1e-12Sorry about the ignorance. I should notice this the first time. Can you give a try and see whether you can reproduce?-Shawn___________________________________________________________Bernd Flemisch phone: +49 711 685 69162 IWS, Universitaet Stuttgart fax: +49 711 685 67020 Pfaffenwaldring 61 email: [email protected] D-70569 Stuttgart url: www.hydrosys.uni-stuttgart.de___________________________________________________________ _______________________________________________ Dumux mailing list [email protected] https://listserv.uni-stuttgart.de/mailman/listinfo/dumux___________________________________________________________Bernd Flemisch phone: +49 711 685 69162 IWS, Universitaet Stuttgart fax: +49 711 685 67020 Pfaffenwaldring 61 email: [email protected] D-70569 Stuttgart url: www.hydrosys.uni-stuttgart.de___________________________________________________________ _______________________________________________ Dumux mailing list [email protected] https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
___________________________________________________________Bernd Flemisch phone: +49 711 685 69162 IWS, Universitaet Stuttgart fax: +49 711 685 67020 Pfaffenwaldring 61 email: [email protected] D-70569 Stuttgart url: www.hydrosys.uni-stuttgart.de
___________________________________________________________
test_1p.input
Description: Binary data
_______________________________________________ Dumux mailing list [email protected] https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
