Easiest way to zero negative values for some tracers?

[iuryt]

This a spin off from #2385 but I am creating a new discussion because I think that this is useful for other situations.
In my case, nutrients are consumed and sometimes could reach negative values.
Despite being O(1e-5) negative values, I would like to zero them at each timestep.

I am also open for other alternatives to avoid these negative values on a NP model.

[glwagner]

Maybe

zero_tracer(sim) = map!(c -> ifelse(c < 0, 0.0, c), parent(sim.model.tracers.c), parent(sim.model.tracers.c))
simulation.callbacks[:zero_tracer] = Callback(zero_tracer)

[iuryt]

zero_tracer(sim) = parent(sim.model.tracers.c) .= max.(0, parent(sim.model.tracers.c))

Do I need to specify that c comes from model.tracers?

[glwagner]

Well yes, c must be defined somehow, sorry, I just wrote it in shorthand to save space and make it clearer. You can also write

c = model.tracers.c
zero_tracer(sim) = parent(c) .= max.(0, parent(c))

[glwagner]

In the context of a model that can't handle negative values you can also write things like max(0, c) inside your kernels. That might be slightly more robust (and also avoids the extra callback computation).

[iuryt]

That would be great! How should I do that? By now I only have a kernel for computing mld and average light.

[iuryt]

Well yes, c must be defined somehow, sorry, I just wrote it in shorthand to save space and make it clearer. You can also write

c = model.tracers.c
zero_tracer(sim) = parent(c) .= max.(0, parent(c))

This works. Just tested.