Particle splitting "a la EAGLE"

This implements a new small physics module: overly-massive particle splitting.

At a rebuild time that does not coincide with a repartition (to avoid creating unnecessary trouble), we loop over all the gas particles and identify which ones are above a given mass threshold. These are then split into two equivalent pieces and slightly offset from each other. We create a companion gpart as well and relink the particle. All this is done on a node-by-node basis and just before rebuild such that the reorganisation and possible exchange of particles is done correctly.

Additional subgrid-related fields that also need splitting are offset to the respective subgrid implementation.

The only bit that needs to be properly taken care of is the reallocation of the arrays in case we run out of space.

Note that this should happen rarely in any real simulation and is really only here to avoid dramatic situations. However, given that the low-z ICs we use were run with a code that does not split particles, the first few rebuilds trigger a lot of these events. Which makes it relatively easy to verify that the feature works.

I have run our usual test cases but if you can have a look at it that would be great.

added 1 commit

• 20ebce4a - When creating a new particle, preserve the odd/even nature of IDs.

Compare with previous version

added 21 commits

• 20ebce4a...220dbc37 - 20 commits from branch master
• be65a629 - Merge branch 'master' into particle_splitting

Compare with previous version

added 1 commit

• 303487b1 - Add the splitting of particles to the list of changes from original EAGLE in the README file.

Compare with previous version

changed milestone to %Paris

added feature request label

added 32 commits

• 303487b1...74145955 - 30 commits from branch master
• 99ffc5f3 - Merge branch 'master' into particle_splitting
• a2df5493 - Only duplicate the gpart in the particle splitting code when we are running with gravity.

Compare with previous version

Ah yes, indeed! Thanks for checking. I have protected the lines touching the gpart with a check for gravity policy.

resolved all discussions

If you resolve those minor issue it all looks ready to go, but we should make an issue to consider using the threadpool for the particle loops.

added 1 commit

• f33ee46c - Print the reallocation messages only in verbose mode.

Compare with previous version

Thanks! And, yes, I agree; an issue about threadpooling the whole thing would be good.

resolved all discussions

resolved all discussions

resolved all discussions