Hydrostatic condition for open domain

[mameehan5]

Hi all,

I am looking to run a buoyancy-driven flow that is open to atmospheric conditions. However, when running a simulation with boundary conditions specified as "Outflow" normal to the direction of gravity, the hydrostatic condition induced by the gravitational field disappears.

To show this most clearly, starting from the HotBubble tutorial case, the following conditions

geometry.is_periodic = 1 0               # For each dir, 0: non-perio, 1: periodic
peleLM.lo_bc = Interior NoSlipWallAdiab
peleLM.hi_bc = Interior Outflow

produced an avg_pressure field (using yt)

as expected for the first snapshot with a hydrostatic condition.

Using

geometry.is_periodic = 0 0               # For each dir, 0: non-perio, 1: periodic
peleLM.lo_bc = NoSlipWallAdiab NoSlipWallAdiab
peleLM.hi_bc = NoSlipWallAdiab Outflow

produced

Again, this is expected.

And finally, using

geometry.is_periodic = 0 0               # For each dir, 0: non-perio, 1: periodic
peleLM.lo_bc = Outflow NoSlipWallAdiab
peleLM.hi_bc = Outflow Outflow

produced

This last image I would have expected to be quite similar to the other two. Is there a physical justification for this behavior or a bug? If the former, is there a way to introduce the hydrostatic condition?

Thank you very much!!

[mameehan5]

I realized now that I can just force the hydrostatic condition with gradP0 for fix my personal simulations. However, I still would like confirmation this is the expected behavior before closing, so please let me know!

[esclapez]

That rings a bell with an old issue of PeleLM that @nickwimer found a while back.
I'll see if the solution implemented in:
AMReX-Combustion/IAMR@9f5f52d
can fix the issue.

[mameehan5]

Sounds great! Those snippets do seem like they would do the trick.

[nickwimer]

Just checking in to see if this issue has been resolved?

[esclapez]

Not yet, the entire infrastructure to handle the phi BC isn't setup in LMeX. I'm thinking it could actually live into AMReX-Hydro projectors, but I need to discuss that with the LBNL team.

[Alex-jian522]

Is this problem solved now?

[mameehan5]

For myself, I just set gradP0 for the hydrostatic condition I need. This seems to work well, and in fact, may even be better than what I posted as the problem because this is no ambiguity as to what the ambient density would be. I will let the developers decide whether they want to pursue this further or close the issue.

[nickwimer]

I am in the middle of a rewrite to sections of the code to fix the underlying issue at play here. In the meantime, @mameehan5 would you be willing to share your solution involving the gradP0? It might be a good "band-aid" solution that we can patch for others facing the same type of issue in the short term.

[mameehan5]

Yep! In the input file, if using gravity, say peleLM.gravity = 0 0 -9.81, I define peleLM.gradP0 to be g*rhoinf where rhoinf is the ambient density. So then buoyancy is only generated where there is a density difference (i.e., density different than the ambient).

[drummerdoc]

Did this get resolved @nickwimer ?

[dmontgomeryNREL]

Adding another example case for this issue that was solved by setting $\vec{\nabla}p_0 = \rho_0\vec{g}$, where $\rho_0$ is the initial density at the inlet. The set up is as follows: we use embedded boundaries to define a 3D cylindrical pipe with TurbInflow to simulate turbulent flow in a pipe. The pipe has noSlipIsotherm walls that decrease in temperature from the inlet to the outlet, with inflow and outflow boundary conditions. The figure below demonstrates the effect of gravity with and without setting $\vec{\nabla}p_0 = \rho_0\vec{g}$.