Create GWTemplate plugin by OMarkwell · Pull Request #216 · sxs-collaboration/sxs

OMarkwell · 2026-10-13T09:09:19Z

GWTemplate is the DDPC waveform generator for LISA. It functions through plugins built into the waveform models, which can then be called by the waveform generator to get the waveforms. I have added such a plugin to sxs, to allow sxs waveforms to easily be used for LISA science. Additionally, I moved some of the FFT-related routines to the time-series class. This is just to avoid having to Fourier transform all the modes when computing the frequency-domain polarizations in a particular direction.

📚 Documentation preview 📚: https://sxs--216.org.readthedocs.build/en/216/

codecov-commenter · 2026-10-13T09:39:11Z

Codecov Report

❌ Patch coverage is 5.26316% with 36 lines in your changes missing coverage. Please review.
✅ Project coverage is 44.74%. Comparing base (c39aa9a) to head (6d1bfdf).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
sxs/time_series.py	5.26%	36 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #216      +/-   ##
==========================================
- Coverage   44.91%   44.74%   -0.18%     
==========================================
  Files          87       87              
  Lines        6773     6810      +37     
  Branches     1099     1103       +4     
==========================================
+ Hits         3042     3047       +5     
- Misses       3458     3490      +32     
  Partials      273      273

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

moble

In addition to the individual comments:

Tests needed, at least to make sure the output is sensible
sxs/plugins/ needs an __init__.py.

I now realize there are several places that seem to assume astropy(?) input types. We need to make sure that those inputs really have the expected types. (I think we're supposed to use .to_value(unit) instead of just .value?) In particular, that assumption shouldn't make it into the sxs code; it should just be in the plugin code, since sxs doesn't know about astropy, etc.

moble · 2026-10-28T14:58:45Z

-        return type(self)(data, **self._metadata.copy())
+        return type(self)(data, **self._metadata.copy())
+
+    def preprocess(self, t1=0, t2=500, t3=None, t4=None, dt=None, **kwargs):


A few things:

First, we need to figure out a better way to do this to minimize code duplication. There are several ways this might work. One possibility would be to just move the function here, and guard the two changes with a test for the type of self — possibly making the WaveformModes version call this one with different arguments. The other would be to extract a lot of the code into a helper function somewhere else, and just make the two preprocess methods call that code with their own specializations.

Second, it's not clear to me that other people will want our preprocessing step at all; it's rather specific to how I want to do preprocessing, including the specific parameters for the transition window(s) and the line-subtraction step to remove memory discontinuities.

Third, note that the defaults include t1=0, t2=500. When this is applied to data that's gone through to_lvk, I think the t=0 point moves to the peak of the waveform, so this zeros out everything before the peak, and is still modifying the data far into the ringdown.

moble · 2026-10-28T14:59:34Z

+        ##Fourier transform
+        h = WaveformModes(time = t,
+                          input_array = h_array,
+                          modes_axis = 1,


This should usually be 0. I think you don't need this argument at all.

moble · 2026-10-28T15:00:23Z

+        return result
+    def fourier_transform(self, total_mass=None, luminosity_distance=None):


Needs a space:

Suggested change

return result

def fourier_transform(self, total_mass=None, luminosity_distance=None):

return result

def fourier_transform(self, total_mass=None, luminosity_distance=None):

moble · 2026-10-28T15:09:43Z

+        h_tilde : TimeSeries
+            Unitful Fourier transform.
+        """
+        from . import m_sun_in_meters, m_sun_in_seconds, parsec_in_meters


It might be worth nitpicking about LISA's conventions compared to mine:

sxs/sxs/__init__.py

Lines 49 to 61 in b7d25d5

# The value of the solar mass parameter G*M_sun is known to higher accuracy than either of its

# factors. The value here is taken from the IAU Division I Working Group's publication on

# "Numerical Standards for Fundamental Astronomy"

# <https://iau-a3.gitlab.io/NSFA/NSFA_cbe.html#GMS2012>.

# In the TDB (Barycentric Dynamical Time) time scale — which seems to be the more relevant one and

# looks like the more standard one for LIGO — we have

solar_mass_parameter = 1.32712440041e20 # m^3/s^2

# Dividing by the speed of light squared, we get the mass of the sun in meters; dividing again, we

# get the mass of the sun in seconds. For consistency with the nominal value above, we retain more

# precision than is warranted by the measurement.

m_sun_in_meters = 1476.6250385063112526099633973363 # m

m_sun_in_seconds = 4.925490949162941425997992909269e-06 # s

moble · 2026-10-28T15:12:07Z

+        result = result.pad(pad_length, mode=pad_mode, **kwargs)
+        result = result.line_subtraction(treat_as_zero=treat_as_zero)
+        return result
+    def fourier_transform(self, total_mass=None, luminosity_distance=None):


This will mutate the data in place. Elsewhere, this is fine because where it's called from is right behind an evaluate call. But here, it could bite people.

moble · 2026-10-28T15:14:15Z

+                h_unitful *= h_factor.value
+                h_unitful.t *= t_factor.value


What are these .values for?

moble · 2026-10-28T15:17:02Z

+        if luminosity_distance is not None:
+            h_factor /= (luminosity_distance * 1e6 * parsec_in_meters)
+        h_unitful = self
+        if total_mass is not None and luminosity_distance is not None:


So no scaling will happen unless both total_mass and luminosity_distance are given?

moble · 2026-10-28T15:22:03Z

+from gwtemplate.mass_time import MassTime
+from gwtemplate import constants
+import astropy.units as u
+from astropy.constants import G, c ,M_sun


See my comment about m_sun_in_meters below.

moble · 2026-10-28T15:25:23Z

+        t_scaling = (self.params['total_mass'] / u.solMass * MTSUN_SI)
+        h_scaling = G * (self.params['total_mass']).to('kg') / (c**2 * self.params['distance'].to('m'))


Increasing my concern about units...

Added FFT and preprocessing to time_series

3cba50c

Created plugin for interface with DDPC gwtemplate

6d1bfdf

OMarkwell force-pushed the main branch from 1578e61 to 6d1bfdf Compare October 27, 2026 10:16

moble requested changes Oct 28, 2026

View reviewed changes

Navigation Menu

Repository navigation

Create GWTemplate plugin - #216

Create GWTemplate plugin#216
OMarkwell wants to merge 2 commits into
sxs-collaboration:mainfrom
OMarkwell:main

OMarkwell commented Oct 13, 2026 •

edited by github-actions Bot

Loading

Uh oh!

codecov-commenter commented Oct 13, 2026 •

edited

Loading

Uh oh!

moble left a comment

Uh oh!

moble Oct 28, 2026

Uh oh!

moble Oct 28, 2026

Uh oh!

moble Oct 28, 2026

Uh oh!

moble Oct 28, 2026

Uh oh!

moble Oct 28, 2026

Uh oh!

moble Oct 28, 2026

Uh oh!

moble Oct 28, 2026

Uh oh!

moble Oct 28, 2026

Uh oh!

moble Oct 28, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		return result
		def fourier_transform(self, total_mass=None, luminosity_distance=None):

	# The value of the solar mass parameter G*M_sun is known to higher accuracy than either of its
	# factors. The value here is taken from the IAU Division I Working Group's publication on
	# "Numerical Standards for Fundamental Astronomy"
	# <https://iau-a3.gitlab.io/NSFA/NSFA_cbe.html#GMS2012>.
	# In the TDB (Barycentric Dynamical Time) time scale — which seems to be the more relevant one and
	# looks like the more standard one for LIGO — we have
	solar_mass_parameter = 1.32712440041e20 # m^3/s^2

	# Dividing by the speed of light squared, we get the mass of the sun in meters; dividing again, we
	# get the mass of the sun in seconds. For consistency with the nominal value above, we retain more
	# precision than is warranted by the measurement.
	m_sun_in_meters = 1476.6250385063112526099633973363 # m
	m_sun_in_seconds = 4.925490949162941425997992909269e-06 # s

		t_scaling = (self.params['total_mass'] / u.solMass * MTSUN_SI)
		h_scaling = G * (self.params['total_mass']).to('kg') / (c*2 self.params['distance'].to('m'))

Conversation

OMarkwell commented Oct 13, 2026 • edited by github-actions Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov-commenter commented Oct 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

moble left a comment

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!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

OMarkwell commented Oct 13, 2026 •

edited by github-actions Bot

Loading

codecov-commenter commented Oct 13, 2026 •

edited

Loading