Reconstructed time looks odd

I did a run on GW150914 and recovered the following posterior:

I'm guessing this is just due to the choice of binning or so, any thoughts @colm.talbot @eric.thrane ?

added Bug label

Are you using explicit marginalisation over tc? It looks like it. Explicit marginalisation forces tc to be on a grid. If the SNR is high enough, you get behaviour like this where you see the discrete time bins. In order to fix this, you want to trick bilby into using a finer tc resolution. Since explicit tc marginalisation relies on an FFT, the default tc resolution is determined by the sampling frequency. You should be able to get better tc resolution by either increasing your sampling frequency, or by artificially zero-padding your frequency arrays in order to simulate the effect of a higher sampling frequency.

It's not a super-high-priority thing, but at some point, it would be nice if we added a feature to bilby so that it gave you a warning message if you tried to analyse an injection (or an event in the catalog) with insufficiently high frequency content for explicit tc marginalisation.

Eric's guess was right here. I was using a low sampling frequency. Closing as it isn't really an issue.

closed

mentioned in issue #336 (closed)