arxiv: 2603.25262 · v1 · submitted 2026-03-26 · 🌌 astro-ph.GA · astro-ph.IM

Recognition: 2 theorem links

· Lean Theorem

Star-Galaxy Classification in Deep LSST Data with Random Forest: A Pilot study on the Data Preview 1 Release

M. Gatto , V. Ripepi , M. Bellazzini , C. Tortora , M. Dall'Ora

Authors on Pith no claims yet

Pith reviewed 2026-05-15 00:36 UTC · model grok-4.3

classification 🌌 astro-ph.GA astro-ph.IM

keywords star-galaxy separationLSSTRandom Forestmachine learningmulti-band photometryfaint magnitudesultra-faint dwarfsData Preview 1

0 comments

The pith

LSST multi-band photometry with a Random Forest classifier separates stars from galaxies reliably at faint magnitudes, outperforming morphology alone.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether supervised machine learning can deliver clean star samples from deep LSST photometry for studies of ultra-faint dwarf galaxies. A Random Forest is trained on a curated set of confirmed stars and galaxies drawn from the Extended Chandra Deep Field South in Data Preview 1. Multi-band colors alone, especially those involving the u filter, produce good separation that remains effective where morphology fails. Adding photometric uncertainties as explicit features improves results further. Galaxy contamination stays low across nearly the entire magnitude range examined.

Core claim

Applied to LSST Data Preview 1 observations of the Extended Chandra Deep Field South, a Random Forest classifier using all six LSST filters achieves effective star-galaxy separation. Multi-band photometry alone significantly outperforms classification based on the morphological parameter refExtendedness at faint magnitudes. Colors that include the u band prove essential for robust performance, and including photometric uncertainties as input features yields the best results overall. Galaxy contamination remains negligible across almost the full magnitude range probed, up to r < 27.5 mag.

What carries the argument

Random Forest classifier trained on combinations of LSST multi-band colors, the refExtendedness morphological parameter, and photometric uncertainties, using a curated training sample from spectroscopic and Gaia data.

If this is right

Stellar samples cleaned this way can support searches for ultra-faint dwarf galaxies in full LSST data.
u-band photometry is required for reliable separation at the faint end.
Explicit inclusion of photometric uncertainties improves classifier accuracy across configurations.
Multi-band photometry alone suffices for low contamination over most of the r < 27.5 range without needing morphological information.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same feature set could be applied to early LSST releases for quick catalog cleaning before full data processing.
Similar classifiers might reduce contamination in other wide-field surveys that lack deep morphology but have multi-band coverage.
Extending training to include even fainter spectroscopic anchors could test whether performance holds below the current limit.

Load-bearing premise

The curated sample of bona fide stars and galaxies remains representative and unbiased in the faint regime relevant for ultra-faint dwarf searches.

What would settle it

Applying the trained classifier to an independent faint sample at r approximately 27 and measuring galaxy contamination substantially above a few percent would show the claimed performance does not hold.

read the original abstract

The Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) will produce unprecedentedly deep and wide photometric catalogs, enabling transformative studies of faint stellar systems such as the research of ultra-faint dwarf galaxies (UFDs). A critical challenge for these studies is reliable star-galaxy separation at faint magnitudes, where compact background galaxies increasingly contaminate stellar samples. This work aims to assess the performance of supervised machine-learning techniques for star-galaxy separation in LSST-like data, quantify the relative importance of morphological and photometric information, and identify the most effective combinations of input features for minimizing galaxy contamination while preserving stellar completeness in the faint regime relevant for UFD searches. We apply a Random Forest classifier to observations of the Extended Chandra Deep Field South from LSST Data Preview 1 (DP1), the deepest field observed within the DP1. We construct a curated sample of bona fide stars and galaxies using spectroscopic data, Gaia DR3, and multi-band photometric catalogs. We train and validate the classifier using several configurations of LSST-based input features, including multi-band colors, the LSST morphological parameter refExtendedness, and photometric uncertainties. We find that LSST multi-band photometry alone delivers a good star-galaxy separation, significantly outperforming morphology-based classification at faint magnitudes. Colors involving the u-band are essential to provide a robust star galaxy separation. Furthermore, explicitly including photometric uncertainties as input features yields the best overall performance. Across all configurations that include all the six LSST filters, galaxy contamination remains negligible almost the whole magnitude range probed in this work (i.e. r < 27.5 mag). [abridged]

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Multi-band photometry plus uncertainties gives solid star-galaxy separation in LSST DP1 to r=27.5, but faint-end labels may inflate the performance.

read the letter

The paper applies a Random Forest classifier to LSST Data Preview 1 in the Extended Chandra Deep Field South. It shows that using multi-band colors alone already outperforms morphology-based separation at faint magnitudes, and adding photometric uncertainties improves results further. Colors with the u-band turn out to be particularly important. Across the configurations that use all six LSST filters, galaxy contamination stays low out to r=27.5. This is a practical pilot. The authors build their training set from spectroscopy, Gaia DR3, and multi-band catalogs, then run controlled tests on different feature combinations. The results give clear numbers on how much each type of information helps, which is directly useful for planning ultra-faint dwarf searches with LSST. The main limitation is the training and validation labels at the faint end. Spectroscopy becomes sparse, Gaia astrometry drops away, and the remaining labels rely on multi-band catalogs. If those catalogs already incorporate similar color or shape information, the validation set is not fully independent. The paper does not provide a breakdown of label sources by magnitude or test the classifier against an external faint tracer, so the contamination figures could be optimistic. This is a real but not catastrophic issue for a first look at the data. The work is for people who need star-galaxy separation in deep LSST fields or who are modeling selection effects for faint stellar populations. It is a clean empirical study on public data with explicit feature tests, so it deserves peer review. A referee can ask for more on label purity without rejecting the core result. I recommend sending it for review.

Referee Report

2 major / 2 minor

Summary. The manuscript presents a pilot study using Random Forest classifiers on LSST Data Preview 1 photometry in the Extended Chandra Deep Field South to perform star-galaxy separation. It claims that multi-band colors (especially those involving the u-band) combined with photometric uncertainties deliver good separation that significantly outperforms morphology-based methods at faint magnitudes, with galaxy contamination remaining negligible across nearly the full range r < 27.5.

Significance. If the label construction holds, the work is significant for LSST-era studies of ultra-faint dwarf galaxies, as it supplies a practical, photometry-driven approach that reduces reliance on morphological parameters which degrade at depth. The explicit demonstration that including photometric uncertainties improves performance and that u-band colors are essential provides actionable guidance for survey pipelines and future UFD searches.

major comments (2)

[§3] §3 (Sample Construction): The curated bona-fide star/galaxy labels at the faint end (r > 25) rely on sparse spectroscopy plus multi-band catalogs once Gaia DR3 astrometry drops out. No magnitude-binned table or figure shows the fractional contribution of each label source versus r, leaving open the possibility that faint labels share color or morphology priors with the classifier features and thereby produce optimistic contamination estimates.
[§4] §4 (Results and Validation): The central claim of negligible galaxy contamination to r < 27.5 is presented without a magnitude-binned purity/completeness table or cross-validation error bars on the held-out set. Because the validation sample is drawn from the same curated labels, an independent faint-end tracer (e.g., deeper HST or spectroscopic follow-up) is required to confirm that the reported rates are not lower bounds.

minor comments (2)

The abstract states performance gains but supplies no numerical metrics (accuracy, contamination fraction, or feature-importance ranks); adding one or two key numbers would improve readability.
Notation for the morphological parameter (refExtendedness) should be defined on first use and kept consistent with the LSST Data Release documentation.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their constructive comments, which have helped improve the clarity and robustness of our analysis. We have revised the manuscript to include additional details on sample construction and validation metrics as requested.

read point-by-point responses

Referee: §3 (Sample Construction): The curated bona-fide star/galaxy labels at the faint end (r > 25) rely on sparse spectroscopy plus multi-band catalogs once Gaia DR3 astrometry drops out. No magnitude-binned table or figure shows the fractional contribution of each label source versus r, leaving open the possibility that faint labels share color or morphology priors with the classifier features and thereby produce optimistic contamination estimates.

Authors: We agree that a magnitude-binned breakdown of label sources is necessary to assess potential biases. In the revised manuscript, we have added a new figure (Figure 3) displaying the fractional contribution of spectroscopic, Gaia, and multi-band catalog labels as a function of r-band magnitude. This shows that for r > 25, labels are predominantly from multi-band catalogs selected using criteria (such as color-color selections from external surveys) that are largely independent of the LSST photometric features used in the classifier. We have also added text in §3 discussing the construction to minimize overlap with classifier inputs. While some residual correlation cannot be entirely ruled out, the use of orthogonal data sources supports the reliability of our contamination estimates. revision: yes
Referee: §4 (Results and Validation): The central claim of negligible galaxy contamination to r < 27.5 is presented without a magnitude-binned purity/completeness table or cross-validation error bars on the held-out set. Because the validation sample is drawn from the same curated labels, an independent faint-end tracer (e.g., deeper HST or spectroscopic follow-up) is required to confirm that the reported rates are not lower bounds.

Authors: We have addressed the request for binned metrics by adding Table 2, which reports purity and completeness in 0.5-mag bins from r=20 to 27.5, including 1-sigma error bars obtained from 5-fold cross-validation on the held-out validation set. This provides a more quantitative view of performance across magnitudes. However, we acknowledge that an independent faint-end validation using deeper HST imaging or additional spectroscopy is not feasible within the current DP1 dataset and would require new observations. We have expanded the discussion in §5 to highlight this limitation and emphasize that the reported rates should be interpreted as internal validation metrics. revision: partial

standing simulated objections not resolved

The need for an independent faint-end tracer (deeper HST or spectroscopic follow-up) to confirm contamination rates, as this is beyond the scope of the available DP1 data in this pilot study.

Circularity Check

0 steps flagged

No circularity: purely empirical ML evaluation on held-out data

full rationale

The manuscript trains a Random Forest classifier on a curated sample of stars and galaxies (constructed from external spectroscopy, Gaia DR3, and multi-band catalogs) and evaluates performance metrics directly on held-out validation data. No equations, derivations, or predictions are present that reduce by construction to fitted parameters or self-referential definitions. All reported results (e.g., contamination rates to r < 27.5) are statistical measurements on data splits, not outputs forced by the model's own inputs or prior self-citations. The approach is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central performance claims rest on the quality and representativeness of the spectroscopically labeled training set and on the assumption that LSST photometric features remain informative at the faintest magnitudes.

free parameters (1)

Random Forest hyperparameters (n_estimators, max_depth, etc.)
Model parameters are tuned on the training data but not reported in the abstract.

axioms (1)

domain assumption Labels from spectroscopy and Gaia DR3 provide accurate, unbiased ground truth for stars and galaxies at the magnitudes used
These labels define the training and validation sets without further independent verification described.

pith-pipeline@v0.9.0 · 5627 in / 1204 out tokens · 40613 ms · 2026-05-15T00:36:49.367388+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We apply a Random Forest classifier to observations of the Extended Chandra Deep Field South from LSST Data Preview 1 (DP1)... multi-band colors, the LSST morphological parameter refExtendedness, and photometric uncertainties.
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Colors involving the u-band are essential... galaxy contamination remains negligible almost the whole magnitude range probed (r < 27.5 mag).

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.