File that showcases the crash:
TestRun_AMXP_crash.py
To recap, my SRGA J1444 run was converging to a high compactness region and there the run got stuck. It got stuck because the likelihood could not be computed and returned an error to the likelihood value. However, this translated to returning a likelihood of around 0, which is actually a very good point (thus drowning out all previous points and throwing off the run). Multinest does not exit from that situation and is thus silently stuck.
I have now found that this is specific to high compactness points and fast rotating stars. So those of you working with those may need to be aware of this error. Compute rays throws the error, because when calculating deflection angles, it assumes that for increasing emission angles, deflection should also increase. Assumption breaks down and causes an error to be returned. I don't know yet the if this solution is strange but physical or if it a non-physical solution and therefore should be fixed with better math.
Anyhow, for my work, I will assume a stricter compactness limit, effectively avoiding this (hopefully). We still need to address this on the short term and find out why this happens and what to do with it long term.
Question: What do you think we should do with this, at least short term? I am unsure how to make multinest exit early ( if possible) and otherwise I think we need an informative warning?
File that showcases the crash:
TestRun_AMXP_crash.py
To recap, my SRGA J1444 run was converging to a high compactness region and there the run got stuck. It got stuck because the likelihood could not be computed and returned an error to the likelihood value. However, this translated to returning a likelihood of around 0, which is actually a very good point (thus drowning out all previous points and throwing off the run). Multinest does not exit from that situation and is thus silently stuck.
I have now found that this is specific to high compactness points and fast rotating stars. So those of you working with those may need to be aware of this error. Compute rays throws the error, because when calculating deflection angles, it assumes that for increasing emission angles, deflection should also increase. Assumption breaks down and causes an error to be returned. I don't know yet the if this solution is strange but physical or if it a non-physical solution and therefore should be fixed with better math.
Anyhow, for my work, I will assume a stricter compactness limit, effectively avoiding this (hopefully). We still need to address this on the short term and find out why this happens and what to do with it long term.
Question: What do you think we should do with this, at least short term? I am unsure how to make multinest exit early ( if possible) and otherwise I think we need an informative warning?