Implement Tensile Instability Control (TIC)

[efaulhaber]

This is an undocumented feature until we figure out when and if it is useful.
So far, it is not useful alone:

But it might be useful in addition to PST.

Note that, just like PST, this does not work at free surfaces.

[codecov]

Codecov Report

Attention: Patch coverage is 33.33333% with 2 lines in your changes missing coverage. Please review.

Project coverage is 70.68%. Comparing base (13694fc) to head (087b500).

Files with missing lines	Patch %	Lines
src/schemes/fluid/pressure_acceleration.jl	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #769      +/-   ##
==========================================
- Coverage   70.70%   70.68%   -0.03%     
==========================================
  Files         106      106              
  Lines        6701     6703       +2     
==========================================
  Hits         4738     4738              
- Misses       1963     1965       +2     

Flag	Coverage Δ
unit	70.68% <33.33%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[svchb]

@efaulhaber Is there any point in adding that then? The function is not used anywhere.

[efaulhaber]

The point is that we can experiment with it. For example, Niklas can try using PST + TIC for the high-res vortex street he's having problems with.

[svchb]

But do we need to merge it than without documentation? He can try this like it is.

[efaulhaber]

It's easier to try it if you don't need an extra branch for it. I don't really see a downside of merging this, as it's a very small and non-invasive change.

[efaulhaber]

The point is that we can experiment with it. For example, Niklas can try using PST + TIC for the high-res vortex street he's having problems with.

I experimented with this and it is indeed solving the high resolution issues. Top is PST and bottom PST + TIC, left and right are the same setup, just a different resolution.

out_highres.mp4

I will try @svchb's symmetric version, and then I'll make this public API and add docs and tests.

[efaulhaber]

Update: The symmetric version (top panel) is not stable.

This is shortly before the symmetric version is crashing:

Here is the code again, so it doesn't get lost in the conversations:

@inline function tensile_instability_correction_sym(m_a, m_b, rho_a, rho_b, p_a, p_b, W_a)
    # Use absolute values of pressure so that negative pressures (tension)
    # also contribute repulsively, preventing unphysical clustering.
    p_abs = abs(p_a) + abs(p_b)
    return -m_b * p_abs / (rho_a * rho_b) * W_a
end

[svchb]

Yes that was the likely result probably if its conservative its not allowing some loss in energy necessary for it to work.

[efaulhaber]

# Note: This should only be applied for the fluid-fluid interaction for non-free surface flows!

@svchb where did you get the information from that this is not supposed to be used for fluid-boundary interaction? I don't find any mention of this in any of the papers.

[efaulhaber]

I just tried TIC only for fluid-fluid interaction and the regular (p_a + p_b) formulation for fluid-boundary interaction, and I got this. Top is no TIC for fluid-boundary, bottom is TIC for everything.
I would say that is clearly showing that we need TIC for both.

out_highres_nobnd.mp4

[svchb]

So TIC should only be applied for dummy boundaries and MK bnd need to be excluded.

[efaulhaber]

MK boundaries use a special "pressure acceleration" anyway, so we don't need to change anything.

[efaulhaber]

/run-gpu-tests

[svchb]

/run-gpu-tests

[LasNikas]

\run-gpu-tests

[svchb]

/run-gpu-tests

[svchb]

/run-gpu-tests