Each tracer should have its own diffusivity

[ali-ramadhan]

Not an immediate concern and but right now the diffusivity κ is shared between temperature and salinity whereas they should be different.

Perhaps this should be worked on at the same time as arbitrary tracer fields (#25).

[glwagner]

I agree that this is naturally addressed when arbitrary tracer fields are implemented.

Note that this applies not just to diffusivities (if you're referring to ConstantIsotropicDiffusivity), but to background diffusivities associated with LES closures, and also to the turbulent prandtl numbers associated with the Smagorinsky closures.

[glwagner]

My plan:

• When the user specifies a single diffusivity, this diffusivity is applied to all tracer fields
• Otherwise, diffusivities are specified by a named tuple, where names correspond to diffusivity names, eg

tracers = (:T, :S)

closure = ConstantIsotropicDiffusivity(ν=1e-6, κ=(T=1e-6, S=1e-7))

and

tracers = (:T, :S)

closure = ConstantIsotropicDiffusivity(ν=1e-6, κ=1e-6, tracers=tracers)

would be possible use patterns. The tracers keyword is only required when the diffusivity is not itself a named tuple.

Or, possibly we can require that tracers is supplied, eg

closure = ConstantIsotropicDiffusivity(tracers, ν=1e-6, κ=(T=1e-6, S=1e-7))

This last pattern would be useful if, for example, we want to do input validation on the diffusivity tuple.

We may also want to think about designs / syntactic sugar for making the handling of large numbers of passive tracers easier.

[ali-ramadhan]

Awesome! Yeah just reviewed PR #452 which looks great. I just opened the issue to keep track of things, but the arbitrary tracers thing got resolved pretty quickly.