Tentative: PSFEx models ~0.4% larger than stars on P3 tiles (worth exploring)

Offen
  • Stimmt etwas nicht?
#956 2 Kommentare 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen

Maintainer antworten meist innerhalb von 1 Tag

Dieses Issue hat noch niemand übernommen.

Bewertung

Schwierigkeit
5/5
Geschätzter Aufwand
Über eine Woche
Anfängerfreundlichkeit
18/100
Issue-Typ
Bug
Klarheit
Muss geklärt werden
Aktivitätsstatus
Aktiv
Tech-Stack
python
Bereich
data

Rechercherichtung

The payload points at the PSFEx model evaluation and the star catalogue full_starcat_smk-p3a.hdf5 for tiles 242.293, 243.293, 242.294 and 243.294; the open question is whether the PSF size residual of about -0.4% holds beyond these four tiles. Start by checking how positions are passed to PSFEx (the 0-based versus 1-based pixel frame) and by re-measuring with the PSFEx-fit stars held out. Done means a reproduced residual with a stated cause, or a documented null result, and this needs the PSF modelling maintainers.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Beschreibung

Tentative, and worth exploring. This is a first look on a small area, not a systematic study.

Summary

On four P3 tiles, the PSFEx models used by ShapePipe (smk-p3a products on nibi) appear to be slightly larger than the stars: the PSF size residual is δT/T ≈ −0.4% (star minus model), and ShapePipe's own star catalogue for the same exposures gives −0.58%. If this holds more widely, a PSF size error of this order biases the multiplicative shear calibration of any shape method by roughly m ~ (T_psf/T_gal)·δT/T, i.e. a few × 10⁻³ for typical UNIONS galaxies, which is above the ~0.1–0.2% PSF-size accuracy usually needed for |m| ≲ 2 × 10⁻³.

What was measured

  • Data: tiles 242.293, 243.293, 242.294, 243.294 (17 exposures, ~1 deg²), with the persisted PSFEx models (PSF_SAMP = 1, degree 2, median 44 accepted stars per CCD) evaluated at each star's CCD position and normalised to unit sum.
  • Stars: 1,615 Gaia DR3 sources (18 ≤ G ≤ 20.5, 5/6-parameter solutions, RUWE < 1.4), at Gaia positions with proper motion applied; core sample with no masked pixel within 8 px and raw peak below 50,000 ADU.
  • Method: HSM adaptive moments of each star stamp vs the model shifted (in Fourier space) to the star's sub-pixel centre, both in the CCD pixel frame.
G star epochs δT/T (robust mean, |δT/T| < 0.1) median
18–19 1,419 −0.34 ± 0.04% −0.32%
19–20 2,036 −0.43 ± 0.09% −0.48%
20–20.5 1,294 −0.43 ± 0.11% −0.45%
  • full_starcat_smk-p3a.hdf5 (the same 17 exposures, 7,835 stars, HSM star vs PSF) independently gives −0.58 ± 0.03%.
  • The plain (non-robust) mean over all stars is +0.13 ± 0.08%, pulled by ~1.5% outliers (binaries, neighbours), so the sign depends on outlier handling; the robust statistics and the ShapePipe star catalogue agree on a negative residual.
  • Per-exposure means span −0.50% to +0.95% (rms 0.33%), and the per-star rms is ~3% at G 18–19, above the measurement noise, which suggests spatial residuals that a degree-2 polynomial on ~44 stars per CCD may not follow.
  • An independent check points the same way: with a moments-based shape estimator, the shear response measured on the Gaia stars is slightly negative (R ≈ −0.003, about −1% of the galaxy response), as expected when the PSF model is larger than the true PSF. The stars' mean ellipticity is zero to 2 × 10⁻⁴, so this does not look like a WCS offset.

Candidate causes (none tested)

  • PSFEx's own resampling or interpolation of the star vignets;
  • the star selection used for the PSFEx fit (possibly related to #316 and #919);
  • a smaller observation, unverified: PSFEx defines its position polynomial in 1-based pixel coordinates, and ShapePipe appears to pass 0-based positions (a 1-px offset in where the model is evaluated). This is probably negligible for a degree-2 polynomial, but it may be worth checking.

Caveats

1 deg² only, a single star-selection choice, and HSM moments as the size measure. A check over more tiles, and with the stars actually used in the PSFEx fit held out, would show whether this is real and survey-wide.

The measurement code and the full write-up are in a separate project (genshear) and can be shared on request.

—
Posted by Claude (AI assistant, Anthropic) on behalf of Calum Murray (@calumhrmurray), who asked for this to be raised.

Vorherrschende Sprache
Python
Sterne
18
Forks
14
Ø Merge
3 T. 18 Std.
Gemergte PRs (30 T.)
27

Entwicklungsumgebung

Erste Schritte

  1. Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
  2. Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
  3. Forken Sie das Repository und arbeiten Sie in einem Branch.
  4. Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.

Mehr aus CosmoStat/shapepipe

Alle Issues in CosmoStat/shapepipe

Ähnliche Issues

Weitere Issues zu Python

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.