Validation and performance benchmarks.

[ali-ramadhan]

We should incorporate the following validation tests which we can also use for performance benchmarking.

• Free convection (check heat budget, mixed layer depth, and turbulent kinetic energy)
• Rayleigh–Bénard convection (compare Nusselt and Péclet numbers)
• Deep convection (regression test)
• Rising thermal bubble?
• Stress-driven flow (not clear which experiment we run)

[ali-ramadhan]

Deep convection and rising thermal bubble regression tests (or golden master tests) have been implemented in PR #140.

@SandreOuza Do you have time to work on the Rayleigh–Bénard convection and free convection tests? I think you understand the physics that need to be tested better than anyone.

[sandreza]

I can work on some of these simulations on Friday

[ali-ramadhan]

That'll be awesome, we can work on them together.

[ali-ramadhan]

Some ideas for testing free convection:

• Check to see if the volume temperature drifts linearly as expected.

[glwagner]

I would also experiment with the 'molecular' diffusivity/viscosities. What happens when they are all set to the same number? Some mixing up of the constants could cause a bug.

Along the same lines, it could be good to try a simulation with Δx = Δy = Δz to make sure mixing those up isn't causing a bug.

Changing the reference density might be a good sanity check. It drops out of the incompressible Navier-Stokes equations so it's value shouldn't change results (ie it could be 1 or it could be κh)...

[ali-ramadhan]

Closing this as we have decided to go with the verification experiments outlined in #346.