Add cubic time interpolation for NoLoop likelihood

[oshaughnessy-junior]

Summary

• add a cubic fractional-time interpolation path for NoLoop Q_lm evaluation while preserving nearest-sample behavior by default
• wire ILE --interpolate-time to select the cubic NoLoop path
• add util_RIFT_pseudo_pipe.py --internal-ile-interpolate-time to pass the option into generated ILE args

Scope

Localized cherry-pick onto rift_O4c: 7 files changed, no O4d calmarg/pipeline material included.

Tests

• PYTHONPATH=MonteCarloMarginalizeCode/Code RIFT_LOWLATENCY=1 MPLCONFIGDIR=/private/tmp/mplconfig-rift-o4c python -m py_compile MonteCarloMarginalizeCode/Code/bin/integrate_likelihood_extrinsic_batchmode MonteCarloMarginalizeCode/Code/bin/integrate_likelihood_extrinsic MonteCarloMarginalizeCode/Code/bin/util_RIFT_pseudo_pipe.py MonteCarloMarginalizeCode/Code/RIFT/likelihood/factored_likelihood.py MonteCarloMarginalizeCode/Code/RIFT/likelihood/Q_inner_product.py
• RIFT_LOWLATENCY=1 PYTHONPATH=MonteCarloMarginalizeCode/Code MPLCONFIGDIR=/private/tmp/mplconfig-rift-o4c pytest -q MonteCarloMarginalizeCode/Code/test/test_noloop_time_interp.py

[oshaughnessy-junior]

Periodic auto-review baseline established for LVK-prep tracking.

Current review status:

• Local smoke checks passed from a clean PR worktree at 512c0012267572b78d33c681a5775170fd3948e7:
  • RIFT_LOWLATENCY=1 PYTHONPATH=MonteCarloMarginalizeCode/Code MPLCONFIGDIR=/private/tmp/mplconfig-rift-o4c pytest -q MonteCarloMarginalizeCode/Code/test/test_noloop_time_interp.py -> 3 passed
  • PYTHONPATH=MonteCarloMarginalizeCode/Code RIFT_LOWLATENCY=1 MPLCONFIGDIR=/private/tmp/mplconfig-rift-o4c python -m py_compile ... on the touched Python entry points/modules -> passed
  • git diff --check richard/rift_O4c...HEAD -> passed
• ldas-grid analysis/demo anchor exists at /home/richard.oshaughnessy/RIFT_roboto_paper.
• Mergeability is currently blocked. git merge-tree against current richard/rift_O4c shows both-branch edits in:
  • MonteCarloMarginalizeCode/Code/bin/integrate_likelihood_extrinsic_batchmode
  • MonteCarloMarginalizeCode/Code/bin/util_RIFT_pseudo_pipe.py

The visible pseudo-pipe conflict is in the parser option block: the conflict resolution should keep the newer rift_O4c --internal-ile-srate-time-resampling / --internal-ile-srate-internal options and also add this PR's --internal-ile-interpolate-time option and downstream --interpolate-time True propagation.

Recommended next gate before merge/LVK review: refresh the branch against rift_O4c, rerun the local checks above, then run or inspect the established ~/RIFT_roboto_paper ldas-grid demos/tests, including a real --interpolate-time True / --internal-ile-interpolate-time path and, if available, a GPU/CuPy parity check.

[oshaughnessy-junior]

Resolved the rift_O4c branch conflict and pushed the update to this PR branch.

Resolution commit:

• 3d0ca4ca5763afca434c40f2a5f92e4049e1c37e (Merge rift_O4c into NoLoop interpolation PR)

Preserved in util_RIFT_pseudo_pipe.py:

• this PR's --internal-ile-interpolate-time and downstream --interpolate-time True propagation
• current rift_O4c options --internal-ile-srate-time-resampling / --internal-ile-srate-internal and their downstream propagation

Post-resolution checks from /private/tmp/rift-pr153-resolve:

• RIFT_LOWLATENCY=1 PYTHONPATH=MonteCarloMarginalizeCode/Code MPLCONFIGDIR=/private/tmp/mplconfig-rift-o4c-resolve-post pytest -q MonteCarloMarginalizeCode/Code/test/test_noloop_time_interp.py -> 3 passed
• PYTHONPATH=MonteCarloMarginalizeCode/Code RIFT_LOWLATENCY=1 MPLCONFIGDIR=/private/tmp/mplconfig-rift-o4c-resolve-post python -m py_compile ... on the touched entry points/modules -> passed
• git diff --check richard/rift_O4c...HEAD -> passed

GitHub now reports this PR as MERGEABLE at head 3d0ca4ca5763afca434c40f2a5f92e4049e1c37e.

[oshaughnessy-junior]

Periodic auto-review at unchanged head 3d0ca4ca5763afca434c40f2a5f92e4049e1c37e / base a08e88a4ba5662faa123efca9b815d0f8362f9f7:

• GitHub reports MERGEABLE/CLEAN; git merge-tree --write-tree refs/codex-auto-review/pr153-base refs/codex-auto-review/pr153-head completed without conflicts (tree a63c2e12387b734e86639205bd414351e579f18c).
• Clean-worktree checks:
  • GW_SURROGATE='' RIFT_LOWLATENCY=1 PYTHONPATH=MonteCarloMarginalizeCode/Code MPLCONFIGDIR=/private/tmp/mplconfig-rift-pr153-current pytest -q MonteCarloMarginalizeCode/Code/test/test_noloop_time_interp.py -> 3 passed
  • baseline python -m py_compile ... set -> passed
  • git diff --check refs/codex-auto-review/pr153-base...refs/codex-auto-review/pr153-head -> passed
  • both ILE entry points accept bare --interpolate-time and explicit --interpolate-time False with --help (all exit 0); pseudo-pipe still registers --internal-ile-interpolate-time and emits --interpolate-time True.
• The established LDAS anchor remains /home/richard.oshaughnessy/RIFT_roboto_paper. Its analyses/bilby_v_rift_likelihood/ILE_MARGINAL_TIME_INTERP.md records real paired maintained-NoLoop ILE runs:
  • 3-template cubic-minus-nearest deltas: +0.0081, +0.0382, -0.0124, versus paired MC errors 0.0399, 0.0459, 0.0500
  • 60-point scan: mean -0.004, rms 0.033, max absolute 0.105
  • saved cubic command includes --interpolate-time True
• Important scope: the saved runtime says no cupy and Override --gpu (not available), so this is strong real CPU/numpy-path validation, not CUDA parity. The LDAS login node also lacks CuPy/NVIDIA access, though Condor advertises GPU slots.

Recommended remaining LVK gate: schedule a small Condor GPU job comparing Q_inner_product_cubic_cupy against the numpy helper at integer and fractional offsets, then run one paired real ILE case on CuPy. No new conflict or CPU regression found.