Changes default AMD poincare coefficient #4494
Conversation
|
You'll have to manually change the coefficient in the regression tests (once those pass we can merge right away). |
|
made the necessary change |
|
If you want extra credit, you can expand the discussion in the docstring. You could link the relevant issue or just write a simple independent explanation. |
|
I think the discussion in #4367 worths some mention in the docstring. Even if it's just a short comment + pointing people to the PR discussion? |
|
@jagoosw, can you update the docstring to explain why we chose 1/3 vs 1/12 with appropriate references? |
|
I'm not sure what to put since the references already there support 1/3 just the value was wrong? |
| By default: `C = Cν = Cκ = 1/3`, which is appropriate for a finite-volume method employing a | ||
| second-order advection scheme, and `Cb = nothing`, which turns off the buoyancy modification term. |
There was a problem hiding this comment.
@jagoosw please add text here from the issue, with appropriate citations, that explain why 1/3 is the right choice
There was a problem hiding this comment.
Also put in a reference to #4367 which has some relevant plots
There was a problem hiding this comment.
The discussion will be an elaboration of this remark:
"I think they're getting there a different way but that in equation 39 they're finding the equivalent of setting it to 1/3?"
| lines!(ax, [exp(1), exp(3)], x->(x/exp(1))^(-5/3)*exp(15), color = :red, linestyle = :dash) | ||
|
|
||
| text!(ax, exp(0.8), exp(13); text = L"$E(k)\sim k^{-5/3}$", color = :white) | ||
|
|
||
| Colorbar(fig[1, 2], colormap = :oslo, colorrange = (0, 10), label = "Time (s)") | ||
|
|
||
| k_filt = 1/sqrt(closure.Cν * 3 / (1/(2*minimum_xspacing(grid))^2 + 1/(2*minimum_yspacing(grid))^2 + 1/(2*minimum_zspacing(grid))^2)) | ||
|
|
||
| lines!(ax, ones(2) .* k_filt, [exp(0), exp(10)], color = :red, linestyle = :dash) | ||
|
|
||
| text!(ax, k_filt, exp(10); text = "1/δ") |
There was a problem hiding this comment.
| lines!(ax, [exp(1), exp(3)], x->(x/exp(1))^(-5/3)*exp(15), color = :red, linestyle = :dash) | |
| text!(ax, exp(0.8), exp(13); text = L"$E(k)\sim k^{-5/3}$", color = :white) | |
| Colorbar(fig[1, 2], colormap = :oslo, colorrange = (0, 10), label = "Time (s)") | |
| k_filt = 1/sqrt(closure.Cν * 3 / (1/(2*minimum_xspacing(grid))^2 + 1/(2*minimum_yspacing(grid))^2 + 1/(2*minimum_zspacing(grid))^2)) | |
| lines!(ax, ones(2) .* k_filt, [exp(0), exp(10)], color = :red, linestyle = :dash) | |
| text!(ax, k_filt, exp(10); text = "1/δ") | |
| lines!(ax, [exp(1), exp(3)], x->(x/exp(1))^(-5/3)*exp(15), color = :red, linestyle = :dash) | |
| text!(ax, exp(0.8), exp(13); text = L"$E(k)\sim k^{-5/3}$", color = :white) | |
| Colorbar(fig[1, 2], colormap = :oslo, colorrange = (0, 10), label = "Time (s)") | |
| k_filt = 1/sqrt(closure.Cν * 3 / (1/(2*minimum_xspacing(grid))^2 + 1/(2*minimum_yspacing(grid))^2 + 1/(2*minimum_zspacing(grid))^2)) | |
| lines!(ax, ones(2) .* k_filt, [exp(0), exp(10)], color = :red, linestyle = :dash) | |
| text!(ax, k_filt, exp(10); text = "1/δ") |
There was a problem hiding this comment.
@jagoosw can you address this?
src/TurbulenceClosures/turbulence_closure_implementations/anisotropic_minimum_dissipation.jl
Outdated
Show resolved
Hide resolved
…otropic_minimum_dissipation.jl Co-authored-by: Gregory L. Wagner <[email protected]>
|
Sorry I'm a bit late here. I understand why these changes were made. I'm still a bit confused though: what should this coefficient be for a WENO of order |
|
No, we don't know. Although I think that combination exhibits the lowest effective resolution (thus increasing computational expense) compared to 2nd order advection + AMD, or WENO only. @xkykai and @simone-silvestri might know more. |
Indeed. AMD + WENO seems to be the most dissipative when compared with 2nd order + AMD and WENO + no SGS closure. |
Following discussion in #4367