Time-domain approximant with Bilby (proposal of example)

[bilby-bot]

In GitLab by @git.ligo:ecm on Jan 17, 2022, 17:58

Dear Bilby developers,

together with Tom Colin (intern student), we have put together a Jupyter notebook based on Tuto 3.2 of the ODW 2021 [https://github.com/gw-odw/odw-2021/blob/master/Tutorials/Day_3/Tuto_3.2_Parameter_estimation_for_compact_object_mergers.ipynb] that does a simplified analysis of GW150914 using a time-domain approximant using the same model (SEOBNRv4). Our objective is to get consistent posterior distributions. Here is our notebook:

Notebook

We align the template in time so that it matches well that of the frequency domain approximant. The time alignment code is:

    max_index = np.argmax(h_plus**2 + h_cross**2)
    delta_index =  len(h_plus) - max_index
    h_plus = np.roll(h_plus, delta_index)
    h_cross = np.roll(h_cross, delta_index)

which leaves a small time shift with the time-domain waveform obtained from the frequency-domain generator (using the exact same model). See sanity check in the notebook.

Questions:

1. We obtain consistent posterior for the intrinsic parameters (chirp mass and mass ratio) but we see large discrepancies for geocent_time. See this corner plot

We conclude that the time alignment procedure is wrong. We struggle to understand how to align. Could you please give us a hint?

2. We believe this type of exercise could be useful to other users. We could come up with a standalone script that could be added to the example/ folder of the Bilby tree. Is this something you would value?

Notes:

For the time-domain approximants, we coded the two following functions (see cell [19]):

def lalsim_SimInspiralTD(...): def _base_lal_cbc_td_waveform(...): def lal_binary_black_hole_td(...):

which are the exact symmetric of the following frequency-domain functions already included Bilby:

def lalsim_SimInspiralFD(...): https://git.ligo.org/lscsoft/bilby/-/blob/master/bilby/gw/utils.py#L735

def _base_lal_cbc_fd_waveform(...): https://git.ligo.org/lscsoft/bilby/-/blob/master/bilby/gw/source.py#L247

def lal_binary_black_hole(...): https://git.ligo.org/lscsoft/bilby/-/blob/master/bilby/gw/source.py#L14

Those could be added to the Bilby code as well.

Best, Eric and Tom

[bilby-bot]

In GitLab by @git.ligo:colm.talbot on Jan 17, 2022, 20:28

Hi @git.ligo:ecm and Tom, this is interesting to see. I'm not sure exactly where such an example would be best to live, but I agree that it could be very informative.

I have a few minor comments:

• no tapering is done on the waveform here. It is common to taper the beginning and end of the template to avoid spectral artifacts after performing the window. Exactly what taper is used is pretty opaque to me, but I know that there are situations where neglecting it can have a pretty bad impact. It's possible that the conditioning done along with the taper is the reason for the difference in recovered time.
• there are faster time-domain waveforms available that might make prototyping easier. IMRPhenomT is fast and of a similar quality to SEOBNRv4.

[bilby-bot]

In GitLab by @git.ligo:ecm on Jan 19, 2022, 10:28

Hi @git.ligo:colm.talbot

thank you for the quick feedback!

no tapering is done on the waveform here

Tapering is included at template generation stage with SEOBNRv4. This can be checked in figure [14] of our notebook. The orange template is clearly tapered. Actually, the lack of tapering would probably affect the posterior of all parameters, not only geocent_time. So, this seems unlikely to us that tapering is the culprit.

Our question is: what is the time-alignment convention that should be followed by the template in Bilby?

I'm not sure exactly where such an example would be best to live

We were thinking to: https://git.ligo.org/lscsoft/bilby/-/tree/master/examples/gw_examples/injection_examples

Eric and Tom