[petsc-users] "-ts_exact_final_time matchstep" leads to DIVERGED_STEP_REJECTED

Mon Dec 16 08:53:17 CST 2024

Good morning Barry,

Changing the initial timestep on the commandline didn't change anything. It's like there is something happening afterwards that sets the initial dt to 0 s that doesn't happen with TS_EXACTFINALTIME_STEPOVER.

Cheers,

Sophie
________________________________
From: Barry Smith <bsmith at petsc.dev>
Sent: Friday, December 13, 2024 22:27
To: Blondel, Sophie <sblondel at utk.edu>
Cc: Jed Brown <jed at jedbrown.org>; Zhang, Hong <hongzhang at anl.gov>; Emil Constantinescu <emconsta at anl.gov>; petsc-users at mcs.anl.gov <petsc-users at mcs.anl.gov>; xolotl-psi-development at lists.sourceforge.net <xolotl-psi-development at lists.sourceforge.net>
Subject: Re: [petsc-users] "-ts_exact_final_time matchstep" leads to DIVERGED_STEP_REJECTED

   Ok, the SNES reason DIVERGED_FNORM_NAN is likely due to the TS using a dt of zero, and hence, the evaluation of the TS causing a divide by zero.

   So the TS adapt needs to be understood better in this situation. My guess is still TS_EXACTFINALTIME_MATCHSTEP is buggy when one actually does get an exact match.

  Please try using a different initial timestep like 0.5e-12 s on the command line.

  Barry

On Dec 13, 2024, at 4:40 PM, Blondel, Sophie <sblondel at utk.edu> wrote:

Barry,

The short output is "SNESSolve has not converged due to Nan or Inf norm", the full one is attached.

Cheers,

Sophie
________________________________
From: Barry Smith <bsmith at petsc.dev<mailto:bsmith at petsc.dev>>
Sent: Friday, December 13, 2024 14:56
To: Blondel, Sophie <sblondel at utk.edu<mailto:sblondel at utk.edu>>
Cc: Jed Brown <jed at jedbrown.org<mailto:jed at jedbrown.org>>; Zhang, Hong <hongzhang at anl.gov<mailto:hongzhang at anl.gov>>; Emil Constantinescu <emconsta at anl.gov<mailto:emconsta at anl.gov>>; petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov> <petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>>; xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net> <xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net>>
Subject: Re: [petsc-users] "-ts_exact_final_time matchstep" leads to DIVERGED_STEP_REJECTED

    There is a bit of complicated logic to determine the "adjusted" timestep in TSAdaptChoose() when if (*accept && ts->exact_final_time == TS_EXACTFINALTIME_MATCHSTEP) {

    Is it possible that hmax = tmax - t; is exactly zero, and the logic below does not correctly handle that case?

0 TS dt 0. time 0.
0 TS dt 0. time 0.
0 TS dt 0. time 0.
0 TS dt 0. time 0.
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00
      TSAdapt basic step   0 stage rejected (SNES reason DIVERGED_FNORM_NAN) t=0          + 0.000e+00 retrying with dt=0.000e+00

  Sophie,

     Any idea why SNES reason DIVERGED_FNORM_NAN?   Could you run with -snes_error_if_not_converged?

On Dec 13, 2024, at 2:34 PM, Blondel, Sophie <sblondel at utk.edu<mailto:sblondel at utk.edu>> wrote:

Hi everyone,

The first max time it is trying to reach is 1.0e-12 s, and the initial dt is set to 1.0e-12 s from the commandline options. I believe it's not a formatting issue and that the dt is actually set somewhere to 0 s because that's why the step is rejected.

Best,

Sophie
________________________________
From: Barry Smith <bsmith at petsc.dev<mailto:bsmith at petsc.dev>>
Sent: Friday, December 13, 2024 14:21
To: Blondel, Sophie <sblondel at utk.edu<mailto:sblondel at utk.edu>>; Jed Brown <jed at jedbrown.org<mailto:jed at jedbrown.org>>; Zhang, Hong <hongzhang at anl.gov<mailto:hongzhang at anl.gov>>; Emil Constantinescu <emconsta at anl.gov<mailto:emconsta at anl.gov>>
Cc: petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov> <petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>>; xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net> <xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net>>
Subject: Re: [petsc-users] "-ts_exact_final_time matchstep" leads to DIVERGED_STEP_REJECTED

   Hm, what is the final time you are stepping towards in this run?

   There is something wrong with the adapt code since it seems to start with a dt of 0 but then tries "adapting" several times, but it could be the
monitor function does not correctly format numbers smaller than 1.e-12 and it is just using truly small dt.

   Jed, Hong, Emil?

   Barry

On Dec 10, 2024, at 11:08 AM, Blondel, Sophie <sblondel at utk.edu<mailto:sblondel at utk.edu>> wrote:

Good morning Barry,

Attached are the updated files, there is more useful information in them.

Cheers,

Sophie
________________________________

From: Blondel, Sophie via Xolotl-psi-development <xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net>>
Sent: Monday, December 9, 2024 17:29
To: Barry Smith <bsmith at petsc.dev<mailto:bsmith at petsc.dev>>
Cc: petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov> <petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>>; xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net> <xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net>>
Subject: Re: [Xolotl-psi-development] [petsc-users] "-ts_exact_final_time matchstep" leads to DIVERGED_STEP_REJECTED

Hi Barry,

I hope you are doing well.

Attached are the output. To give a little more context, this is a "new" way of running the code where multiple instances are created and communicate together every few time steps (like coupling the code with itself in memory). Here there are 3 instances that each have a separate TS object, plus one "main" instance that doesn't solve anything but compute rates to exchange between the other instances.

Cheers,

Sophie
________________________________

From: Barry Smith <bsmith at petsc.dev<mailto:bsmith at petsc.dev>>
Sent: Monday, December 9, 2024 15:12
To: Blondel, Sophie <sblondel at utk.edu<mailto:sblondel at utk.edu>>
Cc: petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov> <petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>>; xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net> <xolotl-psi-development at lists.sourceforge.net<mailto:xolotl-psi-development at lists.sourceforge.net>>
Subject: Re: [petsc-users] "-ts_exact_final_time matchstep" leads to DIVERGED_STEP_REJECTED

On Dec 9, 2024, at 2:56 PM, Blondel, Sophie via petsc-users <petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>> wrote:

Hi,

I am trying to understand a strange behavior I'm encountering: when running my application with "-ts_exact_final_time stepover" everything goes well, but when I switch to "matchstep" I get DIVERGED_STEP_REJECTED before the first time step is finished.

   This is in the very first time-step in TSSolve?

    Please run with -ts_monitor and send all the output (best for a short time interval and do it twice once with -ts_exact_final_time stepover and once with exact.

   Barry

I tried increasing the maximum number of rejections and it just takes longer to diverge, and if I set the value to "unlimited" it is basically an infinite loop.

Is there a way to check why is the step rejected? Could the "matchstep" option change tolerances somewhere that would cause that behavior?

Let me know if I should provide more information.

Best,

Sophie Blondel

<matchstep_reason.txt>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20241216/935847fd/attachment-0001.html>