Solver seems to diverge after re-run the simulation (parallel)

cryabroad · October 30, 2018, 03:52

Hi Guys,

As I understand, if we put latestTime in the startFrom entry in controlDict file, after a simulation is stopped and restarted, OpenFOAM should pick up the latest saved file and starts from there.

It only works for several iterations after re-run in my case. I ran a simulation using mpirun -np <#of processors> solver -parallel, then stopped the simulation, modified something in the fvScheme file (changing one of the divergence schemes), then restarted the simulation using the same command. However, the simulation always abort after several runs. Error messages are as following:

Code:

[2] --> FOAM FATAL ERROR: 
[2] PBiCG has failed to converge within the maximum number of iterations 1000
    Please try the more robust PBiCGStab solver.
[2] 
[2]     From function virtual Foam::solverPerformance Foam::PBiCG::solve(Foam::scalarField&, const scalarField&, Foam::direction) const
[2]     in file matrices/lduMatrix/solvers/PBiCG/PBiCG.C at line 210.
[2] 
FOAM parallel run exiting

In my case, the PBiCG solver is used for velocity and several user-defined scalars and based on the output the solver exits when solving the velocity. I have never had this type of error before and changing it to PBiCGStab doesn't solve the problem, I think there's something wrong I did when I restarted the simulation?

Thanks.

pete20r2 · October 30, 2018, 04:42

What precision is your results set to in controlDict?
Are you writing as ascii or binary?
Did the last timestep results write out clean? (Did you use writeNow to terminate)
Also, there is no need to stop the job just to change fvSchemes, fvSolutions and a number of other files. Just make sure runTimeModifiable is set to true in controlDict and OF will check these files for changes each timestep.(make sure there are no typos when you modify during a run or you will cause a crash though)

cryabroad · October 31, 2018, 00:55

Peter,

Thank you for the response. writePrecision is set to 6, and I'm writing as ascii. As for the last time step results, I didn't use the writeNow flag, and maybe that causes troubles. But I actually did try to restart the simulation using the previous saved time step results (by either deleting the last time step file or set startFrom to startTime and make startTime the time instant of the one before last time step) because those are definitely written out clean, it still fails after several iterations.

Is it possible that the problem comes from the fvScheme that I changed? I changed from Gauss linear to Gauss SFCD for the div(phi, U) divSchemes. I'm asking this because I tried a new simulation (starting from time 0) using the Gauss SFCD but it fails after one night run.

My runTimeModifiable is indeed set to true, but I thought that only affects changes you make in the controlDict file. If I make changes to the fvSchemes file during the middle of a run am I able to detect this change in the log file or somewhere else?

Thanks!

October 30, 2018, 03:52	Solver seems to diverge after re-run the simulation (parallel)	#1
cryabroad Senior Member Ruiyan Chen Join Date: Jul 2016 Location: Hangzhou, China Posts: 162 Rep Power: 10	Hi Guys, As I understand, if we put latestTime in the startFrom entry in controlDict file, after a simulation is stopped and restarted, OpenFOAM should pick up the latest saved file and starts from there. It only works for several iterations after re-run in my case. I ran a simulation using mpirun -np <#of processors> solver -parallel, then stopped the simulation, modified something in the fvScheme file (changing one of the divergence schemes), then restarted the simulation using the same command. However, the simulation always abort after several runs. Error messages are as following: Code: [2] --> FOAM FATAL ERROR: [2] PBiCG has failed to converge within the maximum number of iterations 1000 Please try the more robust PBiCGStab solver. [2] [2] From function virtual Foam::solverPerformance Foam::PBiCG::solve(Foam::scalarField&, const scalarField&, Foam::direction) const [2] in file matrices/lduMatrix/solvers/PBiCG/PBiCG.C at line 210. [2] FOAM parallel run exiting In my case, the PBiCG solver is used for velocity and several user-defined scalars and based on the output the solver exits when solving the velocity. I have never had this type of error before and changing it to PBiCGStab doesn't solve the problem, I think there's something wrong I did when I restarted the simulation? Thanks.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
SimpleFoam cannot open include file	Marija	OpenFOAM Running, Solving & CFD	1	October 28, 2020 11:35
Some questions about a multi region case run in parallel	zfaraday	OpenFOAM Running, Solving & CFD	5	February 23, 2017 11:25
Hybrid discretisation - blend factor	gcoopermax	CFX	5	September 23, 2016 09:05
Problem with an old Simulation	FrankW	CFX	3	February 8, 2016 05:28
simpleFoam in parallel issue	plucas	OpenFOAM Running, Solving & CFD	3	July 17, 2013 12:30

October 30, 2018, 04:42		#2
pete20r2 Senior Member Peter Baskovich Join Date: Jul 2014 Posts: 127 Rep Power: 12	What precision is your results set to in controlDict? Are you writing as ascii or binary? Did the last timestep results write out clean? (Did you use writeNow to terminate) Also, there is no need to stop the job just to change fvSchemes, fvSolutions and a number of other files. Just make sure runTimeModifiable is set to true in controlDict and OF will check these files for changes each timestep.(make sure there are no typos when you modify during a run or you will cause a crash though)

October 31, 2018, 00:55		#3
cryabroad Senior Member Ruiyan Chen Join Date: Jul 2016 Location: Hangzhou, China Posts: 162 Rep Power: 10	Peter, Thank you for the response. writePrecision is set to 6, and I'm writing as ascii. As for the last time step results, I didn't use the writeNow flag, and maybe that causes troubles. But I actually did try to restart the simulation using the previous saved time step results (by either deleting the last time step file or set startFrom to startTime and make startTime the time instant of the one before last time step) because those are definitely written out clean, it still fails after several iterations. Is it possible that the problem comes from the fvScheme that I changed? I changed from Gauss linear to Gauss SFCD for the div(phi, U) divSchemes. I'm asking this because I tried a new simulation (starting from time 0) using the Gauss SFCD but it fails after one night run. My runTimeModifiable is indeed set to true, but I thought that only affects changes you make in the controlDict file. If I make changes to the fvSchemes file during the middle of a run am I able to detect this change in the log file or somewhere else? Thanks!