Skip to content

Comparison with 170817

Hi folks, after substantive digging I'm close to having 170817 agree. Here is a comparison page of my best effort. LALInferenceMCMCO2 is the samples in the pe_event_repository done by Aaron. LALInferenceNestO3ROQ is a run using Rory's ROQ. Bilby is being run with the same ROQ and a number of changes (discussed below).

Partially resolved issues

I'll list here some things that I needed to change, which we will then need to change in master of bilby or make a decision on how to implement them (hence partially resolved).

  1. The bilby constants are based on astropy, which differ at some level to LAL/LALInference, this results in a bias in the chirp mass (constants are fixed to LALInference values in the comparison above). My vote is we just fix our constants to LAL values.

  2. The reflective boundaries cause differences for the mass ratio. For the run above I am using !589 (merged) and have hacked dynesty to stop it from using reflective boundaries for non-periodic parameters. I think we need to put in a MR to dynesty ASAP to get that solved on master.

Remaining issues

  1. The geocent_time is shifted. My guess is that this is a difference in definition. While pesummary isn't showing it at the moment, the bilby run gets an equal log_evidence to LALInferenceNest so I'm confident it has found the same solution with a different definition.

Action items

  • Update the bilby constants to LAL values (hardcoded, not by import): if anyone disagrees let me know.
  • Put a MR into dynesty to remove the reflective boundaries for non-periodic items (literally just delete the lines and explain to Josh why).
  • Investigate the geocent time shift.