@jeffriley pointed out two things about FastCosmicIntegration which I am afraid aren’t sufficiently clearly stated in the documentation (mea culpa) but explain some differences with the old cosmic integration that Jeff encountered:
FastCosmicIntegration, as currently written, expects metallicities to be distributed uniformly in the log, whether binned or stochastically sampled metallicities are used. (This is easy to extend, but that’s the current choice which should be documented.)
When the metallicity distribution extends beyond the range of COMPAS-simulated metallicities, we need to make a choice. Coen’s cosmic integration code folded the metallicities outside the simulated range into the edge bins. I am renormalising the metallicity distribution so that dP/dZ integrates to 1 between min_logZ_COMPAS & max_logZ_COMPAS — effectively, this means I am actually using a slightly different metallicity distribution than the input one. Or one could use the initial distribution without re-normalisation and simply assume that metallicities outside the COMPAS range have yields of zero. All choices can be justified, we again just need to be more explicit in documenting them (in code and paper).
@TomWagg — maybe you could include the documentation in your updates. @stellarphysicsandpopulations — just drawing your attention to this, because I know you have thought about this. ;)
ilyamandel
commented
2 years ago
@jeffriley pointed out two things about FastCosmicIntegration which I am afraid aren’t sufficiently clearly stated in the documentation (mea culpa) but explain some differences with the old cosmic integration that Jeff encountered:
FastCosmicIntegration, as currently written, expects metallicities to be distributed uniformly in the log, whether binned or stochastically sampled metallicities are used. (This is easy to extend, but that’s the current choice which should be documented.)
When the metallicity distribution extends beyond the range of COMPAS-simulated metallicities, we need to make a choice. Coen’s cosmic integration code folded the metallicities outside the simulated range into the edge bins. I am renormalising the metallicity distribution so that dP/dZ integrates to 1 between min_logZ_COMPAS & max_logZ_COMPAS — effectively, this means I am actually using a slightly different metallicity distribution than the input one. Or one could use the initial distribution without re-normalisation and simply assume that metallicities outside the COMPAS range have yields of zero. All choices can be justified, we again just need to be more explicit in documenting them (in code and paper).
@TomWagg — maybe you could include the documentation in your updates. @stellarphysicsandpopulations — just drawing your attention to this, because I know you have thought about this. ;)