Faraday RM-synthesis: Step 0 (rmsynth module + calibration) - #1

Open

christianhbye wants to merge 36 commits into

mainfrom

faraday-rmsynth

christianhbye commented Jun 9, 2026

Owner

Summary

First slice of the model-independent Faraday-detection pipeline (full spec: docs/superpowers/specs/2026-06-08-faraday-rmsynth-design.md).

src/lusee_faraday/rmsynth.py — RM synthesis on channelized Stokes Q/U: lambda2, faraday_resolution, max_scale, phi_grid (zero-centered), rmsf, faraday_spectrum. Pure NumPy, degenerate-input guards, 100% test coverage.
notebooks/rmsynth_calibration.py — Step-0 calibration: runs RM synthesis on the existing 10/30/50 MHz zoom spectra to size the eventual full-band grid.
Infra: pytest-cov added to the dev dependency-group so uv run pytest works; spec + step-0 plan under docs/superpowers/.

Key finding

The sparse 3-band comb (192 channels) produces catastrophic RMSF sidelobes — first sidelobe ~73% of peak, contamination across the whole ±100 rad/m² range. RM synthesis on the existing data alone cannot separate real RM from aliases, which quantitatively motivates continuous full-band λ² coverage in the next steps. RMSF resolution is ~0.004 rad/m² (huge λ² lever arm), so resolution is never the limit — SNR/sidelobes/beam-depolarization are.

Status

Step 0 of 5. Steps 1–4 (FrequencyPlan zoom/wide switch + noise + LST tagging; curated full-band sim; analysis with noise MC; repo attic + docs) will continue on this branch.

Test Plan

uv run pytest — 94 passed (18 new in tests/test_rmsynth.py)
uv run python notebooks/rmsynth_calibration.py runs headless and writes notebooks/results/rmsynth_calibration.png
Tests cover known-RM recovery (±sign), RMSF unity peak, weight handling, shape contracts, and degenerate-input guards

🤖 Generated with Claude Code

christianhbye and others added 30 commits

June 8, 2026 18:20


          update notebooks

bbb112b


          docs: Faraday RM-synthesis spec and step-0 plan

d54ed24

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          feat(rmsynth): lambda2, faraday_resolution, max_scale

06f26d5


          chore: add pytest-cov to dev dependency-group so 'uv run pytest' works

e7d8e23

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          docs: correct phi_grid spacing (round up) in step-0 plan

d276b23

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          feat(rmsynth): phi_grid

816311d


          feat(rmsynth): rmsf

7c063ee


          feat(rmsynth): faraday_spectrum

96144ba


          refactor(rmsynth): zero-centered phi grid, degenerate-input guards, s…

a1588b1

…tronger tests


          feat(rmsynth): expose rmsynth in package init

8e6aa31


          feat: step-0 RM-synthesis calibration on existing 3-band sims

0f8dfac


          fix(calibration): use 10 MHz band for galaxy-up time slice

6a4fbfe

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          feat(noise): radiometer_sigma and add_noise

624babb

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>


          feat(freqplan): FrequencyPlan sim_freqs and channelize

aab1539

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>


          docs: Faraday RM-synthesis step-1 plan

1903fa8

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          feat(rotations): topo_euler_angles for LST tagging

66526be

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>


          feat: expose noise and FrequencyPlan in package init

8b1485f

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>


          docs(rotations): NumPy-style docstring for topo_euler_angles

285cba0

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          perf(fast_sim): restrict per-frequency trig to above-horizon pixels

786b872

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>


          perf(fast_sim): parallelize compute_vis_fast over time steps

6e1d9d6

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>


          docs: Faraday RM-synthesis step-2a speedups plan

b37408b

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          feat(spectrometer): truncate response to significant support

1dd526a


          feat(freqplan): per-mode decimation and response truncation

ef9770e


          fix(fast_sim): forkserver context to avoid fork deadlock; hoist nkeep

010619e

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          polish(freqplan): document dict-decimation requirement; assert flat-s…

60c7b4d

…pectrum channelization

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          feat(pipeline): simulate_channelized FR (channelized) + no-FR (centers)

a8b1f63

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>


          feat: full-band grid-design script (RMSF + cost)


          feat: full-band Faraday sim driver (FrequencyPlan + parallel + LST)

76d6274


          docs: Faraday RM-synthesis step-2b full-band sim plan

3457cbd

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          fix(sim): pin BLAS threads to 1 and use one process per physical core

8b181f7

8 oversubscribed processes (each multi-threaded) on 4 physical cores
thrashed; single-threaded workers, one per physical core, run clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

christianhbye and others added 6 commits

June 8, 2026 23:36


          docs: Faraday RM-synthesis step-3 analysis plan

66b8106

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          feat(detection): Faraday-spectrum noise std and peak SNR

5e3b8f1


          feat: expose detection in package init

a92f762


          feat: step-3 RM-synthesis analysis + detection-significance figures

a1a8d64

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>


          fix(analysis): cap phi grid (avoid OOM); add Faraday-effect (FR-noFR)…

5de9d8e

… SNR

Peak-|F| SNR measures polarized-signal detection (huge); the Faraday
effect itself is the RM synthesis of the FR-noFR difference. Both are
noise-dominated-free at hours of integration -- the limit is the
intrinsic-sky degeneracy, not sensitivity.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>


          docs: document RM-synthesis detection layer in README and CLAUDE.md

e446fa6

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

christianhbye requested a review from Copilot

June 9, 2026 14:27

Copilot AI reviewed

Copilot AI left a comment

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

christianhbye mentioned this pull request

Fisher-forecast Faraday detection layer (sky-marginalized SNR) #2

Open

3 tasks

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

Labels

None yet