potential issue with neutral diffusion diagnostics when multiple iterations per timestep occur

[klindsay28]

The post_data calls for neutral diffusion diagnostics occur within the subroutine neutral_diffusion. This subroutine is called from subroutine tracer_hordiff inside a sub-cycling iteration loop that is there to avoid violating diffusive CFL limits. If there are multiple iterations because of CFL limits, then there are multiple calls to post_data per timestep for these diagnostics.

Note that advection diagnostics are handled differently. While advection can be sub-cycled, the advection diagnostics are accumulated across these iterations, and the post_data calls are outside of the iteration loop.

The potentially multiple calls per timestep to post_data for neutral diffusion diagnostics result in multiple calls per timestep to FMS's send_data. Does FMS handle multiple send_data calls per timestep for individual diagnostics?

The same question applies to diagnostics for lateral boundary mixing, whose implementation closely follows the neutral diffusion implementation.

[Hallberg-NOAA]

No, FMS does not handle multiple post_data calls per timestep gracefully, so it is up to the MOM6 code to accumulate the net transports across multiple iterations. The construct noted with this issue is a bug and should be fixed.

[Hallberg-NOAA]

@klindsay28, are there ongoing efforts to fix this via the dev/ncar fork, or should we try to organize someone to address this bug?

[klindsay28]

@Hallberg-NOAA, I am working on this on the dev/ncar fork.
Thanks for asking.