PSF fourth-order moments in the sky frame (spin-2 + rho4)#859
Open
cailmdaley wants to merge 4 commits into
Open
PSF fourth-order moments in the sky frame (spin-2 + rho4)#859cailmdaley wants to merge 4 commits into
cailmdaley wants to merge 4 commits into
Conversation
Extend the sky-coordinate HSM measurement (PR #812) to also store fourth-order moments for PSF and star: - spin-2 combinations HSM_M4_1_* = M40 - M04 and HSM_M4_2_* = 2*(M13 + M31), ported from the PSFHOME/moment4 math, and - galsim's spin-0 HSM_RHO4_* (moments_rho4). Frame handling is the key change over the moment4 port. Whitening with the symmetric sqrt(inv(M)) leaves the whitened axes aligned with the frame in which M and the pixel grid are built, so the spin-2 combinations are defined relative to those axes. Building them in the pixel frame ties them to the CCD orientation. Instead _fourth_moments builds everything in the world frame: M from the use_sky_coords world ellipticity/size, and the grid via the local WCS Jacobian mapping pixel offsets to world offsets about the world centroid (moments_centroid is relative to the stamp true_center). The result is invariant to a re-orientation of the pixel frame viewing the same sky object. HSM failures fill the spin-2 terms with 0 and pass rho4 through (-1); the FLAG column stays the source of truth. Masked star pixels are zeroed before the moment sum. Columns are added to the single-epoch, validation and multi-epoch output paths. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K98geWHTuT62whqjKocfCm
…ments Drive the real _get_psfshapes / _get_starshapes on galsim stamps rendered through a nontrivial local WCS (scale + rotation + shear): - round and elliptical Gaussians: whitened spin-2 fourth moments ~ 0 and rho4 ~ 2 (whitening circularises any single sheared circular profile, so an elliptical Moffat would also vanish -- a two-Gaussian composite is used to get a genuine non-zero spin-2 signal), - frame invariance: one sky object rendered under WCS orientations of 0/28/63/-45/90 deg gives the same world-frame fourth moments, on both the PSF and star paths, - PSF/star agreement on a clean stamp, and HSM-failure sentinel handling. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K98geWHTuT62whqjKocfCm
MergeStarCatPSFEX read and wrote only HSM_G1/G2/T/FLAG for PSF and STAR, so the spin-2 fourth moments and rho4 written per-tile by psfex_interp were dropped at the merge step. The merged full_starcat is the catalogue PSF validation consumes, so the feature was inert past the per-tile stage. Read and write HSM_M4_1/HSM_M4_2/HSM_RHO4 for both PSF and STAR, matching the per-tile column grammar. Avoids the moment4 reference's copy-paste bug (m4_2_psf now reads the _2 column) and adds RHO4, which the reference omitted. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K98geWHTuT62whqjKocfCm
Two paths the summary claimed but the suite left unexercised: - Bad-pixel masking: every star fixture had an all-zero mask, so zeroing masked pixels before the fourth-moment sum was a no-op in the suite. Add a test that plants the -1e30 sentinel in outskirt pixels, asserts the masked vignet reproduces the clean result, and asserts the same sentinels fed through un-zeroed blow the fourth moments up by ~5 orders (mutation: removing the zeroing makes M4_1 go -0.03 -> -4278). - Centroid transform: every fixture was drawn centered, so moments_centroid ~ 0 and the sky-frame recentering was dead. Add a world-frame-shift case; the fourth moments stay invariant only because the transformed centroid is subtracted (mutation: dropping the subtraction fails both angles). Also tighten the frame-invariance tolerances from rtol=1e-3/atol=1e-5 to rtol=1e-5/atol=1e-6 (measured cross-orientation agreement is ~1e-8), so a subtle partial-frame error at the 1e-4 level now bites. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K98geWHTuT62whqjKocfCm
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #697. Supersedes #698. Follow-up to #812; based on its branch so the diff shows only the fourth-moment work — retarget to
developafter #812 merges.What this does
Adds fourth-order moment columns to the PSF/star measurement chain:
HSM_M4_1/2_{PSF,STAR}(spin-2 combinationsM40 − M04,2(M13 + M31), PSFHOME-style) andHSM_RHO4_{PSF,STAR}(galsim's spin-0 kurtosis, previously computed and discarded).How
All computation happens in the world frame. The pixel grid maps to sky offsets through the local WCS Jacobian. The whitening matrix comes from the sky-frame second moments returned by
use_sky_coords=True. This makes the spin-2 columns frame-invariant; the pixel-frame computation in #698 was not (verified numerically — it rotates with the CCD orientation).Columns flow through all write paths: single-epoch, validation, multi-epoch, and
merge_starcatinto the merged star catalogue.Two latent #698 issues do not carry over: the residual imaginary part from
scipy.linalg.sqrtmis dropped, and themerge_starcatcopy-paste bug (M_4_PSF_2reading the_1column) is avoided.Tests
23 pass (13 new + masked-pixel and centroid coverage from review, plus the #812 suite — no regression). Analytic guards: round Gaussian → spin-2 ≈ 0, rho4 ≈ 2; frame invariance across four WCS orientations at rtol=1e-5 (actual agreement ~1e-8);
-1e30mask sentinels; off-center sources. A physics note encoded in the tests: whitening circularises any sheared circular profile, so a single elliptical Moffat also gives spin-2 ≈ 0 — the spin-2 signal traces radial ellipticity gradients. The non-trivial fixture is a two-component coaxial Gaussian.Known follow-up
sp_validation's residual-star plot reads the oldM_4_STAR_1-style names; it needs 4 one-line renames to theHSM_M4_*grammar (separate repo, separate PR).🤖 Generated with Claude Code
https://claude.ai/code/session_01K98geWHTuT62whqjKocfCm