GWSignal: fix TEOBResumSDALI global sign (psi was displaced by pi/2)

[oshaughnessy-junior]

What

RIFT/physics/GWSignal.py corrects the magnitude of gwsignal's
TEOBResumSDALI modes but not their sign. RIFT's TEOBResumSDALI templates are
therefore -1 x bilby's, which is exactly psi -> psi + pi/2.

One-line change: negate distance_rescaling. Nothing else, and no other
approximant is touched.

Why it was invisible

F+(psi+pi/2) = -F+(psi) and Fx(psi+pi/2) = -Fx(psi), in every detector, for
every mode. So a global sign on h is exactly a quarter turn in psi and
nothing else — sky location, distance, masses, inclination and peak lnL all
still agree with bilby and dingo. The fit looks perfect.

Flagship case (M=120, chi_eff 0, ecc 0), injected psi = pi/4 = 0.785:

code	psi circular mean	psi mod pi/2
dingo	0.777	0.743
bilby	0.747	0.717
rift	2.364	0.739

3pi/4 = 2.356. Across 18 comparisons psi_rift = psi_bilby + pi/2 wins for
TEOBResumSDALI in 7 of 9 informative cases and never for SEOBNRv5EHM, which
is identity in all 8. dingo-vs-bilby is identity everywhere, TEOB included.

Root cause

The codes reach the same plugin through different doors:

• bilby / dingo -> GenerateFDWaveform -> polarizations
• RIFT (--use-gwsignal) -> GenerateTDModes -> modes, then its own
  harmonic sum

For this one plugin the modes are minus the polarization convention (and are
dimensionless). Restricting to one (l,|m|) at a time via ModeArray makes
hp - i hc = c * sum_lm h_lm {-2}Y_lm(iota,0) exact for a single complex c,
so an azimuth convention (mode-dependent, e^{i m dphi}) cannot masquerade as
a global sign:

mode	|c|	arg(c)	residual
(2,2)	8.50741e-22	-180.000	0.0000
(2,1)	8.50741e-22	+180.000	0.0000
(3,3)	8.50741e-22	+180.000	0.0000
(4,4)	8.50741e-22	-180.000	0.0000

arg(c) = 180 deg for every m, not arg(c) ∝ m. And 8.50741e-22 is
nu M G/(c^2 D) for these parameters to six digits — precisely the
distance_rescaling already in the code. The magnitude was patched; the sign
was not.

RIFT is correct for SEOBNRv5EHM: fitting a free complex coefficient per
azimuthal index against the gwsignal polarizations gives 1.000 at 0.0 deg
for all eight m. The exp(i m pi/2) shift in hlmoft is right and necessary
there, and is left alone.

Verification

RIFT template held at psi_R = pi/4; gwsignal/bilby template scanned over
psi_T with phi_ref maximized over, signed overlap, all three detectors,
run against the patched tree itself (not a monkeypatch):

case	best psi_T - psi_R
SEOBNRv5EHM, before and after	<= 0.056 x pi/2
TEOBResumSDALI, before	+1.03 x pi/2
TEOBResumSDALI, after	<= 0.10 x pi/2

Same answer in H1, L1 and V1, at phi_ref = 0 and 0.3.

This also rules out a cross-polarization (hx) sign error, the obvious first
guess: flipping only hx is a reflection of (F+,Fx) rather than a
rotation, so no single psi could absorb it across three detectors and the sky
position would shift. It doesn't.

Deliberately not in this PR

TEOB's modes also do not want the exp(i m phi_shift) that hlmoft applies,
which leaves RIFT's reported phase for this approximant offset by pi/2
from the polarization path. That is a phase-convention difference rather than
a sign error; it exists today, is unchanged by this PR, and the shift is
load-bearing for every other approximant, so it is left for separate review.

Regression test

test/test_gwsignal_teob_mode_sign.py asserts the sign, not the
convention.

Fitting a free complex coefficient c_m per azimuthal index, the templates
are correct iff c_m = s exp(i m delta) with s = +1: a common delta is
only a relabelling of phi_ref, but s = -1 is the psi bug. A single m
cannot separate them — -exp(i m delta) is reabsorbed into delta, which is
precisely why the quadrupole alone never saw this. Two m values sharing one
delta do:

arg(c_4) - 2 arg(c_2)  ==  0 if s = +1,  ==  pi if s = -1

invariant under any choice of delta. It reads +178.7 deg on the
unpatched tree and passes after the fix — and will keep passing if the phase
convention is revisited later, so it does not prejudge that question.

The test skips where the TEOBResumSDALI plugin is absent.

This branch has no CI workflow, so run it manually where the plugin exists:

python -m pytest -q MonteCarloMarginalizeCode/Code/test/test_gwsignal_teob_mode_sign.py

Companion PR against rift_O4d: oshaughnessy-junior#92

🤖 Generated with Claude Code