Time-domain likelihood (with heterodyning) by neha2023sharma · Pull Request #1161 · bilby-dev/bilby

neha2023sharma · 2026-10-19T17:34:52Z

What it does

IMR analyses in the time domain: This works with any LAL time-domain approximant. The noise in each detector is modelled by an autocovariance function computed from its PSD.
Inspiral and post-inspiral analyses: We can choose a cut-off time relative to the merger and analyse only the truncated segment of data before it (inspiral) or after it (post-inspiral). The analysis window is set with set_analysis_windows(duration=..., analysis_segment="inspiral" or "post_inspiral", segment_cut_time=..., cut_reference_parameters=...), where segment_cut_time is the cut-off time relative to the merger, and cut_reference_parameters gives the reference merger time and sky location.
Three ways of placing the waveform on the data samples, chosen with placement:
- "nearest": the waveform is generated on its own time grid and then shifted to the data sample nearest to its arrival time in each detector. It is the simplest and fastest option, but the timing can be off by up to half a sample.
- "fd_shift": the waveform is Fourier transformed, the exact time shift is applied as a phase in the frequency domain, and it is transformed back to the time domain.
- "model_time": the waveform is evaluated directly at the model times of each detector, shifted by the arrival time, so no shifting or interpolation is needed afterwards. This option currently works with the IMRPhenomT/TP/THM/TPHM approximants.
A heterodyned version of the time-domain likelihood to make likelihood computation faster. (Works only with the IMRPhenomT family)

ColmTalbot

Thanks @neha2023sharma this looks in generally pretty good shape! Given the nature of the proposed change, there are a decent number of conceptual questions here. I haven't looked in detail through everything, but I looked at the interferometer, likelihood, and source additions which seem the most important.

There are also some things in the implementation that I think are points of tension with previous implementation choices. I appreciate you not changing existing code, but I think over the review process we may need to find a few changes where we can make changes to make the new functionality feel more natural.

ColmTalbot · 2026-10-19T18:20:35Z

+
+            >>> from bilby.gw.detector import PowerSpectralDensity
+            >>> psd = PowerSpectralDensity.from_aligo()
+            >>> acf = AutoCovarianceFunction.from_power_spectral_density_array(


This begs the question of whether the AutoCovarianceFunction should mirror the PowerSpectralDensity more closely. Or even if they should just be one class.

I think one class here would be better. The ACF is just the inverse Fourier transform of the PSD, so it makes sense to keep them together. I can add a method to PowerSpectralDensity that computes the ACF whenever needed.

Does this sound good? If so, I can make these changes.

ColmTalbot · 2026-10-19T18:22:27Z

+        frequency_array = np.asarray(frequency_array, dtype=float)
+        psd_array = np.asarray(psd_array, dtype=float)
+        delta_f = frequency_array[1] - frequency_array[0]
+        frequency_values_missing = int(round(frequency_array[0] / delta_f))
+        if frequency_values_missing > 0:
+            frequency_array = np.concatenate(
+                [np.arange(frequency_values_missing) * delta_f, frequency_array])
+            psd_array = np.concatenate([np.zeros(frequency_values_missing), psd_array])
+
+        psd_sampling_frequency = 2 * frequency_array[-1]
+        if sampling_frequency is not None and not np.isclose(
+                sampling_frequency, psd_sampling_frequency, rtol=1e-9, atol=0):
+            logger.warning(
+                f"The sampling frequency given ({sampling_frequency:g} Hz) is not equal to the one "
+                f"computed from the PSD ({psd_sampling_frequency:g} Hz); using "
+                f"{psd_sampling_frequency:g} Hz. An ACF computed with {sampling_frequency:g} Hz "
+                "would be wrongly scaled and its lags wrongly spaced, giving wrong likelihood "
+                "and SNR values.")
+
+        band = (frequency_array >= minimum_frequency) & (frequency_array <= maximum_frequency)
+        if fill_value is None:
+            fill_value = fill_multiplier * np.max(psd_array[band])
+        else:
+            fill_multiplier = None
+        psd_array = psd_array.copy()
+        psd_array[~band] = fill_value
+        acf_array = 0.5 * np.fft.irfft(psd_array) * psd_sampling_frequency


This feels significant and large enough to just make it a standalone function to convert between PSDs and ACFs.

I will move this to a standalone function power_spectral_density_to_autocovariance that converts PSD to ACF. The new ACF method in PowerSpectralDensity will call this function to compute the ACF.

ColmTalbot · 2026-10-19T18:23:39Z

+        start_time=times[0])
+
+
+__all__ = ["condition_strain_data"]


__all__ should live at the top of modules if we define it.

Thanks! I will move __all__ to the top of the modules.

ColmTalbot · 2026-10-19T18:30:34Z

+            parameters=parameters, injection_polarizations=injection_polarizations, placement=placement)
+        return injection_polarizations
+
+    def inject_signal_time_domain_from_waveform_polarizations(self, parameters, injection_polarizations,


Can this use the existing functionality https://github.com/bilby-dev/bilby/blob/main/bilby/gw/detector/__init__.py#L33? If not, what is missing?

ColmTalbot · 2026-10-19T18:48:24Z

+            polarizations_at = signal["polarizations_at"]
+
+            def rescaled(times):
+                polarizations = polarizations_at(times)


Wait, this is a function?

ColmTalbot · 2026-10-19T18:49:10Z

Skipping this file for this pass.

ColmTalbot · 2026-10-19T18:49:48Z

               result, source, utils, waveform_generator)
 from .waveform_generator import WaveformGenerator, LALCBCWaveformGenerator
-from .likelihood import GravitationalWaveTransient
+from .likelihood import GravitationalWaveTransient, TimeDomainGravitationalWaveTransient


I wouldn't add this extra top level import. I'd like to reduce the number we have over time.

I'll remove this.

ColmTalbot · 2026-10-19T18:50:15Z

    return output_sample


+def _signal_polarizations_for_likelihood(likelihood, parameters):


This looks like a workaround for the method I suggested adding to the GravitationalWaveTransient.

ColmTalbot · 2026-10-19T18:53:07Z

+
+    Returns
+    =======
+    dict: polarizations_at, a function of the times relative to the merger (in s)


I don't understand this use case yet, is it for cases where you want to apply detector-based time delays before evaluating the model?

Co-authored-by: Colm Talbot <talbotcolm@gmail.com>

neha2023sharma added 19 commits October 8, 2026 22:24

time-domain likelihood

6fce0ae

Analysis windows added

19deee0

Remove crop data

3c1fba8

A few functions removed

b0ba2a7

removed set_strain_from_psd_time_domain function

bab2155

waveforms at model times

5f2ab4e

Time domain heterodyning

996000f

Time domain heterodyning

6643243

Time domain heterodyning

932a703

Heterodyning cleaner

bdbc6b0

float64->float32 in eigenvalues computation

244d495

float64->float32 in eigenvalues computation

81f785b

f_start based on modes

0646790

Error message added

e61d436

Adding examples

3faf72c

ACF error fix

330e17c

examples sampler settings changed

ed32e61

Examples edited

c5c27d5

fix docstring error

3891b0a

ColmTalbot reviewed Oct 19, 2026

View reviewed changes

mj-will added likelihood >100 lines gw labels Oct 20, 2026

neha2023sharma and others added 2 commits October 20, 2026 14:09

Update bilby/gw/detector/acf.py

2d2fe4e

Co-authored-by: Colm Talbot <talbotcolm@gmail.com>

r1r1 terms included in heterodyned likelihood

f00357b

		return output_sample


		def _signal_polarizations_for_likelihood(likelihood, parameters):

Conversation

neha2023sharma commented Oct 19, 2026

What it does

Uh oh!

ColmTalbot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

neha2023sharma Oct 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

neha2023sharma Oct 23, 2026 •

edited

Loading