Deep convection example and golden master test should switch to BCs.

[ali-ramadhan]

Instead of a forcing function it should use a Flux boundary condition specified with an array.

[glwagner]

Could possibly nuke the deep convention test (the example was already nuked) and close this.

[christophernhill]

“Nuking” the test sounds like moving in the wrong direction - maybe I misunderstood?

…

On Wed, May 29, 2019 at 16:37 Gregory L. Wagner ***@***.***> wrote: Could possibly nuke the deep convention test (the example was already nuked) and close this. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#220>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA27DYAJVMIIMJQGL66LADDPX3SP5ANCNFSM4HM5MOYA> .

[glwagner]

Aha! The deep convection tests are currently marked as 'broken' and therefore ignored.

In the meantime, we have a new Rayleigh-Benard regression test that tests most of the same features that were tested by deep convention (and a few additional ones, as well).

We probably can't afford too many regression tests (since they are somewhat expensive), and we will want to add ones for new operators / LES closures / etc, so its probably the right decision in the long run to 'nuke' the deep convection test.

[christophernhill]

what about strategies for making regression tests quick? We have those too. Chris

…

On Wed, May 29, 2019 at 16:50 Gregory L. Wagner ***@***.***> wrote: Aha! The deep convection tests are currently marked as 'broken' and therefore ignored. In the meantime, we have a new Rayleigh-Benard regression test that tests most of the same features that were tested by deep convention (and a few additional ones, as well). We probably can't afford too many regression tests (since they are somewhat expensive), and we will want to add ones for new operators / LES closures / etc, so its probably the right decision in the long run to 'nuke' the deep convection test. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#220>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA27DYHMLKQ6ALTQSKSAPSDPX3UADANCNFSM4HM5MOYA> .

[ali-ramadhan]

Yeah I wanted to make deep convection one of the examples we maintain so it can also be a regression test to make sure the example always works as expected. It's not that expensive, but if we can make it faster that would be great. I suppose not using NetCDF would be a start.

[glwagner]

Ok, ok. If it's rotating, uses Flux bcs at the top, Gradient bcs at the bottom, and is non-trivially stratified in both salinity and temperature, it'll test a good number of features in addition to the Rayleigh-Benard test.

[ali-ramadhan]

Does linear stratification count as non-trivially stratified?

But yeah, adding a bit of salinity in there would be good for the test, maybe not so much for the example.

[johncmarshall54]

salinity could be put in as a passive tracer, perhaps, with zero flux at the top and bottom. Should be conserved.

…

On Wed, May 29, 2019 at 5:20 PM Ali Ramadhan ***@***.***> wrote: Does linear stratification count as *non-trivially stratified*? But yeah, adding a bit of salinity in there would be good for the test, maybe not so much for the example. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#220>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKXUEQTN73LJYCI4JDUXTYLPX3XTTANCNFSM4HM5MOYA> .

-- ========================================== John Marshall Cecil and Ida Green Professor of Oceanography Massachusetts Institute of Technology http://oceans.mit.edu/JohnMarshall/ ==========================================

[glwagner]

Aha, yes! By non-trivial I meant non-zero, I suppose. Just something to get some dynamics going.

The Rayleigh-Benard test uses salinity as a passive tracer (and forces it), so we actually capture passive tracer advection-diffusion with that regression test. By adding salinity-stratification, we ensure the modeled effects of salinity on buoyancy also do not regress.

[glwagner]

Moot since deep convection regression was nuked?

[ali-ramadhan]

Ah this was a very minor issue, technically resolved in PR #334, although relevant again as comparing deep convection with MITgcm is part of the verification experiments we agreed upon (see #346).