search

CliMA / Oceananigans.jl

🌊 Julia software for fast, friendly, flexible, ocean-flavored fluid dynamics on CPUs and GPUs
https://clima.github.io/OceananigansDocumentation/stable
MIT License
1k stars 196 forks source link

Move barotropic mode calculation to `update_state!`? #2244

Closed glwagner closed 1 year ago

glwagner commented 2 years ago

We currently calculate the barotropic mode inside the AB2 step:

https://github.com/CliMA/Oceananigans.jl/blob/e27988039e65e9244a84d241b70c2dcdeac93309/src/Models/HydrostaticFreeSurfaceModels/hydrostatic_free_surface_ab2_step.jl#L18-L22

I'm wondering if it makes more sense to calculate this inside update_state!.

Our algorithm calculates update_state! at the end of a time-step, which means that after calling time_step!(model) the model auxiliary and prognostic state are all concurrent, which is useful for output.

cc @sandreza @simone-silvestri

francispoulin commented 2 years ago

If the barotropic mode is not used in the AB2 scheme, which I presume is the case, then I agree it makes sense to compute it elsewhere.

update_state! seems like a good place to me.

glwagner commented 2 years ago

If the barotropic mode is not used in the AB2 scheme, which I presume is the case, then I agree it makes sense to compute it elsewhere.

update_state! seems like a good place to me.

Basically we have to calculate the barotropic mode before doing the free surface update if we are using a SplitExplicitFreeSurface. But I think we can regard the barotropic mode as part of the "auxiliary state" of the problem. The rest of the auxiliary state is calculated in update_state!, so I think that's the appropriate place to calculate the barotropic mode as well.

francispoulin commented 2 years ago

I like that plan.

sandreza commented 2 years ago

I think the only potential issue would be the first time step. Otherwise this seems okay to me

glwagner commented 2 years ago

We always calculate update_state! before the first time-step!

glwagner commented 2 years ago

For reference, update_state!:

  1. Fills halo regions (this could include communication for distributed models, or passing between adjacent regions for multi-region models like the CubedSphere)
  2. Calculates any auxiliary fields (nonlinear diffusivities, hydrostatic pressure, vertical velocity for hydrostatic models)

update_state! is called:

  1. Before the first time-step (eg when invoking run!, or also time_step!(simulation) when simulation.initialized == false)
  2. After calling set!(model, ...)
  3. After every time-step
glwagner commented 1 year ago

@simone-silvestri ?

simone-silvestri commented 1 year ago

This part of the code has been removed. The initial barotropic mode is calculated now as part of the initialize_free_surface! function called at initializatiom