Implementing stratified case

[Adiram97]

Hello,

I did a neutrally buoyant simulation with ERF and Boussinesq; so basically, I set temperature to be constant everywhere in the domain

Now, I would like to start from that chk file and implement stratification. Is there a way to implement a linear temperature profile now; when I am restarting from the old neutrally buoyant constant temperature chk?

Thank you very much

[asalmgren]

One could hack something but there aren’t any ways to do this with current code.

[Adiram97]

Thank you for your response.

So basically I just want to start stratified case with an initial field that is turbulent and neutrally buoyant.

At the moment, I do this by providing boundary conditions such as it is cooling and switching on the term erf.gravity=true. But I am noticing that this process is very slow buildup of stratification.

Therefore what I was thinking is to start the neutrally buoyant case from scratch but with a stratified profile for temperature in input sounding ;but with erf.gravity=false. This way the temperature is simply a passive scalar and therefore will remain neutrally buoyant and at 1 point I switch it on with erf.gravity. Do you think this would work in ERF?

Also this would mean that every time I have to change stratification, I have to start from scratch without a turbulent field. So is there really no way of imposing stratification at a given time on a neutrally buoyant case? (I always stick with Boussinesq equations so base state density is 1 and erf.sounding_type=constant)

Thank you in advance

[asalmgren]

A question -- would you get the right scenario if you effectively started from scratch with the stratified fields that you want and read in ONLY the velocities from the checkpoint that you had written earlier?

[Adiram97]

Yes I think I would get the right scenario yes.

So basically, my goal is to see how stratification affects the turbulent structures in the neutral case I simulated. Therefore, If I can read my existing turbulent velocity field from neutral case and further impose a stratification profile, it works yes.

But is there really a way to read my velocity field from old simulation when I start from scratch? Because if I provide amr.restart, it will treat as a restart

[asalmgren]

There isn't currently a way to do that but I think it wouldn't be hard to initialize the velocities from a checkpoint file. Let me think about how best to do that.

[Adiram97]

Ok sure, thank you and please let me know in case it is doable.

just 1 additional information, that I dont know if it affects. I use static box refinement. So I have Level 0 and Level 1 grids that is fixed in time

[asalmgren]

I've just added the capability -- see #2947

I did verify that it seemed to work on a simple level 0 case -- please try it and let me know if it works for you! See the addition to the docs in the PR -- it should tell you how to use it

[Adiram97]

Hello,

Thank you very much

I tried and I get the following error: amrex::Abort::2::1D Arrays are hard-wired to only 256 high !!!
SIGABRT

I saw that compared to previous version, it does not call InitConstantDensity in ERF_Prob.H ; but then there is no such file in Source/Prob. Instead I saw a function in InitDensity.H but it aborts much before

Basically I am trying to remain in the Boussinesq framework and therefore the density being constant while having variation in potential temperature

[asalmgren]

My suggestion for right now would be to set erf.init_type = Uniform in your inputs file, then in the file Source/Prob/ERF_InitDensityHSE.H in routine erf_init_const_dens_and_th_hse, modify the code to do a 1d loop to set th_hse instead of setting it to constant T0.

That will allow you to keep moving forward while I think about how to label the options so that they are clear to all users going forward.

[Adiram97]

Hello,

I tried something like this: I replaced constant T0 with the following:

amrex::Real dthetadz = 0.016737;
amrex::Real dz = 23.23/854.0;

for ( amrex::MFIter mfi(th_hse, amrex::TilingIfNotGPU()); mfi.isValid(); ++mfi )
{
    amrex::Array4<amrex::Real> th_hse_arr = th_hse.array(mfi);
    const amrex::Box& gbx = mfi.growntilebox(1);
    ParallelFor(gbx, [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
    {
        amrex::Real z = (k + 0.5) * dz;
        th_hse_arr(i,j,k) = T_0 + dthetadz * z;
    });
}

Basically this is to set a linear temperature gradient and it compiles fine. If I run a simulation with this it gives me the following error: 9::Assertion `lbase < max_level' failed, file "../../Submodules/AMReX/Src/AmrCore/AMReX_AmrMesh.cpp", line 553 !!!
SIGABRT

[asalmgren]

Ok, let me try to debug with your exact case. First -- make sure you're running with current development -- I've been refactoring things. Second -- attach here the info of which directory you're building in and your inputs file. Thanks!

[Adiram97]

Hi,

Yes I am on pr-2947

I am building it from Exec/ABL

I have attached my input file and the terrain file

input.zip

[Adiram97]

Ah but I dont know if you need my chk file that I restart from to reproduce the error. Its just that its a bit too big around 70GB

But basically if you switch off erf.gravity and set uniform temperature that is my old configuration

[asalmgren]

Ok -- do a git pull of that PR -- the changes I just made seem to fix the problem you were seeing but I didn't run it out to confirm everything is correct. Let me know!

[Adiram97]

Hi,
This works!

I am able to restart from the velocity fields and it is perfectly fine. Regarding the temperature profile it did not set it in the initial field plt00000 and it is basically constant everywhere. But I'll check further in the lines of code I added in InitDensity if I made a mistake in implementing the gradient

Thank you very much