Numerical approaches

[HendrikTolman-NOAA]

WW3 is built around Yanenko’s Fractional Step method, separating propagation in its various dimensions and source term computations in individual computation steps that are executed serially. This design is suitable for traditional explicit numerical schemes, but not for implicit schemes. For WW4, what is the preferred numerical approach (explicit versus implicit, details of fractional stepping). And do we intend to focus on a preferred numerical approach or allow options?

[JessicaMeixner-NOAA]

Implicit schemes need to be an option, particularly as many applications require higher resolution and therefore to stay reasonably fast, implicit schemes are beneficial.

Additionally for numerical methods -- having 2nd order propagation schemes and addressing numerical GSE is ideal.

[JessicaMeixner-NOAA]

@ukmo-kitstokes - What numerical methods will you want with the SMC grid, in terms of for the SMC grid spatial solvers, time-stepping, spectral space, etc?

@ardhuin, @mickaelaccensi - any thoughts?

@sbrus89 - any thoughts on numerics for a future WW4?

[HendrikTolman-NOAA]

@aronroland: You gave a pretty condensed presentation about all your triangular grid numerical considerations at WISE. Do you have this written up somewhere?

[aronroland]

@HendrikTolman-NOAA: The good news is that we are now finalizing the publication steps. Once the GMD paper is submitted, it will be immediately available online. For the limiter-free second-order time-space integration on unstructured grids, we are aiming for JCP, so this will likely take some more time.

The code release will accompany the GMD submission. My suggestion to NOAA is therefore to use the well-defined Triton-C interface and call Triton-C from WW4, rather than duplicating or modifying the internal implementation directly. This keeps the coupling clean, avoids unnecessary divergence, and allows future developments in Triton-C to remain available through a stable interface.

[HendrikTolman-NOAA]

@aronroland Thanks for the input. I will need to take a look at the actual Triton -C code and interface to see how well this fits into the WW4 high level architecture, which is much cleaner than the WW3 architecture.. Once we are ready for that, I will let you know if I agree with this approach in general. The three top questions to answer for this are

• Do the interfaces fit
• Is the Triton-C functionality consistent with the WW4 architecture regarding
  • Interface
  • Data structures and data abstraction
  • Code functionality other than pure compute
  • Open science governance

I am really looking forward to being able to see the code!