search

illinois-ceesd / mirgecom

MIRGE-Com is the workhorse simulation application for the Center for Exascale-Enabled Scramjet Design at the University of Illinois.
Other
12 stars 19 forks source link

V&V for time step estimation of full Navier-Stokes #614

Open inducer opened 2 years ago

inducer commented 2 years ago

Possibly connected to/root cause of #613.

cc @MTCam @lukeolson @jbfreund

inducer commented 2 years ago

Discussed possible unit mismatch in https://github.com/illinois-ceesd/mirgecom/blob/9b19b749de2c7d099f4cd5aee67727eda8349225/mirgecom/viscous.py#L370-L373

inducer commented 2 years ago

image

(7.32) in JSH/TW

MTCam commented 2 years ago

JSH/TW make assumptions about the kinematic viscosity and thermal conductivity that make their NS not like ours. They assume kinematic viscosity is independent of density, and neglect thermal diffusion. That make their formula OK for their use - but in our case we need to use the actual kinematic viscosity. Corrected in #615.

Now to cover it with some tests.

Edit: Previous verification test fixed up for change set in #615. Per in-person discussions, we'll add and document some offline integrated tests to show that the stability limits are reasonably estimated.

MTCam commented 2 years ago

Re: Validation for our timestep estimation.....

Using the /examples/poiseuille-mpi.py, a good example of a single gas CNS problem, we can see that the solution diverges within a few hundred steps for CFL > ~.3. Two inviscid examples, vortex-mpi.py and pulse-mpi.py run stably up to CFL = ~.6.

We'll add a blurb to the documentation about guidelines for selecting time-step size, citing these examples. We should add some mixture examples, and some verbiage about some of the pertinent dimensionless flow characterization numbers that one can use to estimate the limits for given problems.