Re: Restarting at a max timestep rather than specified timestep

[Manohar Teja]

Dear all,

I just want to remind you about my last issue. Do you have an idea why this could be happening and how to make the simulation restart with the last time step and not maximum time step?

Regards

Manohar

CGAFD, Univ of Exeter

On Wed, Jun 24, 2026 at 2:36 PM Manohar Teja <manohartej...@gmail.com> wrote:

Hi all,

I am trying to restart a IVP simulation from a checkpoint. The initial condition is as follows:

if not restart:

    <some set of lines>

else:
    write, initial_timestep = solver.load_state('checkpoints/checkpoints_s2.h5')
    initial_timestep = 1e-4
    file_handler_mode = 'append'

So I a, asking the simulation to restart with dt = 1e-4, but it starts with 1e-3 (which is the max_timestep). Because of which the simulation is crashing, see below:

2026-06-24 14:25:28,450 subsystems 0/224 INFO :: Building subproblem matrices 19/19 (~100%) Elapsed: 2s, Remaining: 0s, Rate: 8.5e+00/s
2026-06-24 14:25:28,590 solvers 0/224 INFO :: Loading solver state from: checkpoints/checkpoints_s2.h5
2026-06-24 14:25:28,688 solvers 0/224 INFO :: Loading iteration: 35981
2026-06-24 14:25:28,688 solvers 0/224 INFO :: Loading write: 2
2026-06-24 14:25:28,688 solvers 0/224 INFO :: Loading sim time: 9.95776188021427
2026-06-24 14:25:28,688 solvers 0/224 INFO :: Loading timestep: 0.0002722075449691476
2026-06-24 14:25:46,439 __main__ 0/224 INFO :: Starting main loop
2026-06-24 14:27:31,404 __main__ 0/224 INFO :: Iteration=35991, Time=9.967762e+00, dt=1.000000e-03
/leonardo/home/userexternal/mkalluri/miniforge3/envs/dedalus3/lib/python3.12/site-packages/dedalus/core/arithmetic.py:866: RuntimeWarning: overflow encountered in multiply
  np.multiply(arg0_exp_data, arg1_exp_data, out=out.data)
/leonardo/home/userexternal/mkalluri/miniforge3/envs/dedalus3/lib/python3.12/site-packages/dedalus/core/arithmetic.py:251: RuntimeWarning: invalid value encountered in add
  np.add(arg0.data, arg1.data, out=out.data)

Did someone else have this problem? How did you fix it?

Regards.

Manohar Teja Kalluri

CGAFD, Univ of Exeter

[Keaton Burns]

Hi Manohar,

What does your CFL code look like? 

-Keaton

--
You received this message because you are subscribed to the Google Groups "Dedalus Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dedalus-user...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/dedalus-users/CAGuRpfSZZqXR%3D0F8NCg0dzoAo%2BFy6cWUtpfrGSCUSvmsaV%3Dnpg%40mail.gmail.com.

[Daniel Lecoanet]

Hi Manohar,

When you call “solver.step” you pass in the timestep size that you’d like. On your first timestep you should pass in 1e-4. If you’re picking timestep sizes using the CFL calculation, you could set the initial_dt parameter.

Daniel