lalsimutils: fix spurious (l,±m) asymmetry in hlmoft FD-mode conditioning

[oshaughnessy-junior]

Single-commit replacement for #162 (which accidentally carried the whole development-branch history; closing it in favor of this).

Summary

RIFT's hlmoft ChooseFDModes→TD conditioning introduced spurious (±m) mode asymmetry at the ~1% level for every ChooseFDModes-consumed model (IMRPhenomXHM/XPHM/XPNR/XO4a/HM): an exactly-symmetric aligned-spin IMRPhenomXHM input acquired 1.02% (2,±2) and 0.9% (3,±3) relative amplitude asymmetry, violating the conjugate-pair identity h_{l,-m}(f) = (-1)^l conj(h_{lm}(-f)).

Root cause: SimInspiralChooseFDModes returns modes on an ascending two-sided grid [−fNyq, +fNyq] with DC at the center index (odd length TDlen+1), but the FD high-pass window assigned frequencies via evaluate_fvals, which assumes RIFT's reversed packing and, at odd length, gives f_assumed = −f_true + deltaF/2. Since (l,m) and (l,−m) modes occupy opposite signs of f, the nominally |f|-symmetric window landed one bin apart on the two members of each pair — precisely in the taper band [0.964·fmin, fmin] containing the |h̃(f)| maximum.

Fix: build the window on the true frequency grid (P.deltaF*(np.arange(npts) - npts//2), exactly even in f), and zero the −fNyq bin left unpaired when the resize truncates its +fNyq partner, so all conditioning commutes with complex conjugation. All other conditioning stages (start/end tapers, centering roll, IFFT) are covariant already.

Verification

Nonprec control (m1=44, m2=36 Msun, s1z=0.5, s2z=−0.3, fmin=fref=20, deltaF=1/16, deltaT=1/4096), rerun against this branch (master + this commit):

• Raw ChooseFDModes XHM satisfies the conjugate-pair identity exactly (0.0); through hlmoft the mode-level residual drops 1.02e-2 → 2.5e-16. IMRPhenomHM and XPNR machine-exact; XPHM reduced to its intrinsic 4.4e-3 (2,±1) aligned-spin model asymmetry; precessing XPHM amplitude identity at 4.2e-16.
• U/V-level diagnostics: D2 1.04e-9 → 1.28e-17, D3 1.7e-9 → 1.4e-16. (The corrupted band lies mostly below fmin, so the U/V-level leakage was ~1e-9, not the mode-level 1e-2; the fix removes the interface floor entirely.)
• Conditioned-waveform change: 1.1e-2 full-band, 99% of it below fmin in the discarded turn-on band; 1.6e-3 in-band (corrected window-edge placement at fmin).

The parity QA scripts used for verification live on the fork's rift_O4d branch under MonteCarloMarginalizeCode/Code/test/waveforms/ (merged there via oshaughnessy-junior#52).

🤖 Generated with Claude Code

[oshaughnessy-junior]

Addressed in the amended commit (c17c93f): the -fNyq zeroing is now guarded by whether the resize actually truncated (npts_fd_pre_resize > TDlen) rather than by not(no_condition), so the raw diagnostic path is covariant too. Verified both paths at machine precision (worst conjugate-pair residual 3.3e-16 conditioned, 3.5e-16 with no_condition=True, XHM nonprec control).

[oshaughnessy-junior]

Style point addressed in the amended commit (06764da): the grid choice now lives in evaluate_fvals(lal_2sided_fseries, lal_convention=False/True) — default False keeps the existing reversed-RIFT packing byte-identical; True returns the LAL-generator packing f[k] = deltaF*(k - npts//2) (exact for even and odd lengths), with docstring notes on when each applies and what goes wrong if you use the default on ChooseFDModes output. hlmoft now calls evaluate_fvals(..., lal_convention=True) instead of building the grid inline. No behavior change: XHM nonprec conjugate-pair residual still 3.3e-16.