Setting Different Vertical and Horizontal Eddy Viscosity with Smagorinsky-Lilly (or any other scheme than constant anisotropic)

[jakob-braga]

Hello,

@francispoulin and I had a question about using the Smagorinsky-Lilly closure scheme but specifying a different vertical and horizontal eddy viscosity separately where each would use its respective grid spacing in the scheme as the filter width.

Given that this is supported when specifying a constant eddy viscosity (constant anisotropic) I imagine it may be supported out of the box with any other closure and I just can not find an example of this.

If this is not supported out of the box would it be that difficult to implement myself? I am having a bit of difficulty finding the code I would need to edit, but I will hold off until I know this isn't already a feature.

Please let me know, Thanks!

[glwagner]

This is a substantial change because the viscosity is now assumed to be isotropic in 3D, eg

Oceananigans.jl/src/TurbulenceClosures/turbulence_closure_implementations/Smagorinskys/smagorinsky.jl

Line 101 in b5bce2c

@inbounds νₑ[i, j, k] = cˢ² * Δᶠ^2 * sqrt(2Σ²)

We don't have an anisotropic viscosity anymore (I don't think), instead our design us to use a closure tuple for that situation, in which there are separate horizontal and vertical viscosities.

You could edit the source code to achieve your goal though.

[jakob-braga]

Thank you so much for the quick reply,

I did some digging through the source last night to try and better understand what I need to edit and I think I am a little confused about a few things,

When you give a closure tuple to a model, does it just always assume that the first entry is for the horizontal eddy viscosity and the second for the vertical as in the case for Constant Anisotropic Diffusivity, or is it more complicated and related to the 'formulation' parameter in the Scalar Diffusivity constructor? I am having a hard time finding where this parameter is used.

This depends on the above but is all I would need to do make some overloads (or just out right new functions) for Smagorinsky that change the filter width to be just Δz or sqrt(ΔxΔy), I fear I may be greatly over simplifying.

Sorry for the super specific questions about editing source, I just think this would be a nice thing to have for my simulations as my Δx and Δz are quite different.

Please let me know, Thanks!