RuntimeWarning: overflow encountered in matmul

[Manohar Teja]

Hi,

I am running a few test simulations of my 3D Rayleigh Taylor unstable problem with resolution 512x512x768. I have used Dedalus for RTI problems at the same resolution before using RK443 time stepping scheme. However, owing to memory constraints, I am moving to MultistepIMEX scheme. The simulations ran well, atleast in testing phase where I am running for 100 iterations, with SBDF3, but when I use SBDF4 it throws the following error: 

/home/e6a/mk737.e6a/miniforge3/envs/dedalus3/lib/python3.12/site-packages/numpy/_core/einsumfunc.py:1217: RuntimeWarning: overflow encountered in matmul
  ab = matmul(a, b, **kwargs)

Any clue why? and how this can be fixed? (without increasing resolution)

The time step was consistently dt = 0.5 for SBDF3 and drops to dt=2.860318e-03 for SBDF4, which I thought would avoid such overflow errors?

Regards.

Manohar Teja Kalluri

CGAFD, Univ of Exeter

[Daniel Lecoanet]

Hi Manohar,

The SBDF4 timestepping algorithm is less stable than SBDF3, so it is not a priori surprising that you might find numerical instability using that time stepper. It may be possible to avoid this by decreasing the safety factor further.

Daniel

--
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/CAGuRpfT2W77M9-vqoUQWnQSH2BvFY9_5orokw2bWxbkKpbeC2A%40mail.gmail.com.

[Manohar Teja]

Got it. Thank you Daniel

Regards

Manohar

CGAFD, Univ of Exeter

To view this discussion visit https://groups.google.com/d/msgid/dedalus-users/290D8003-08C4-4629-941E-8C5921309FAC%40northwestern.edu.