pith. sign in

arxiv: 2605.28922 · v1 · pith:4WM2ZY3Vnew · submitted 2026-05-27 · 🌌 astro-ph.IM · astro-ph.CO

Photometry is all you need: supernova classification as a mixing problem

Ana Sof\'ia M. Uzsoy , V. Ashley Villar This is my paper

Pith reviewed 2026-06-29 09:29 UTC · model grok-4.3

classification 🌌 astro-ph.IM astro-ph.CO
keywords supernova classificationphotometric classificationGaussian mixture modelmixing fractionType Ia supernovaeType Ibc supernovaelight curve fittingunsupervised classification
0
0 comments X

The pith

Supernovae Ia and Ibc can be classified with at least 90 percent accuracy from photometry alone by optimizing the mixing fraction in a Gaussian mixture model on light-curve fit parameters.

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

The paper treats classification of large supernova populations as an unsupervised mixing problem instead of a supervised task that needs labels. All light curves are fit with a semi-analytical model powered by radioactive decay; the resulting distributions of fit parameters are then modeled as a two-component Gaussian mixture whose shared mixing fraction is optimized directly from the photometry. This recovers the population ratio between Ia and Ibc events and assigns individual classifications at or above 90 percent accuracy with no spectroscopic training data required. The approach is tested across different mixing fractions and with varying amounts of redshift information or a few known labels. The result matters because complete spectroscopic follow-up is impossible for the millions of transients expected from future photometric surveys.

Core claim

Fitting all supernova light curves with a semi-analytical model powered by radioactive decay produces distributions of fit parameters that are adequately described by a two-component Gaussian mixture model; optimizing the shared mixing fraction between the Ia and Ibc populations from the photometry alone recovers the population ratio and classifies individual events with at least 90 percent accuracy without any labeled spectroscopic dataset.

What carries the argument

Two-component Gaussian mixture model on the distributions of fit parameters obtained from the semi-analytical radioactive-decay light-curve model, with the mixing fraction optimized to match the observed population.

If this is right

  • The ratio of Ia to Ibc populations can be reliably constrained from photometry across a range of mixing fractions.
  • Classification performance holds when redshift information is included, excluded, or replaced by photometric redshifts.
  • A small number of known labels can be incorporated without changing the core unsupervised procedure.
  • The method remains viable for fast population characterization in large photometric datasets.

Where Pith is reading between the lines

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

  • The same mixing-model strategy could be applied to other transient classes if suitable physical light-curve models exist for them.
  • Reliance on spectroscopic follow-up for classification could be substantially reduced in upcoming wide-field surveys.
  • Physical model fitting may serve as a general substitute for hand-crafted features in other unsupervised astronomical classification problems.

Load-bearing premise

The fit-parameter distributions produced by the semi-analytical model for Ia and Ibc events are adequately described by a two-component Gaussian mixture whose shared mixing fraction can be optimized from photometry alone.

What would settle it

A test set of spectroscopically confirmed supernovae where the parameter distributions deviate strongly from two Gaussians or where the optimized mixing fraction yields classification accuracy well below 90 percent when compared to the true labels.

Figures

Figures reproduced from arXiv: 2605.28922 by Ana Sof\'ia M. Uzsoy, V. Ashley Villar.

Figure 1
Figure 1. Figure 1: Example fits of SN Ia and Ibc light curves with the Arnett light curve model using sncosmo for the ZTF g band (blue) and r band (red). Points indicate observed data points while solid lines indicate model fits. Bottom panels indicate the “pull”, (data - model)/error. Shaded regions indicate a pull between -1 and 1. be classified as SN Ia or Ibc by comparing the weights γIa,s and γIbc,s. We optimize the par… view at source ↗
Figure 2
Figure 2. Figure 2: The distributions of fit parameters for the entire dataset, with 1737 SNe Ia in cyan and 193 SNe Ibc in magenta (with a Ibc fraction of 0.1), as SNe Ibc with the Arnett model. Y-axis is log10 scaled and shared for all parameters. In our GMM framework, each object s has two weights γIa,s and γIbc,s, which denote the probability of that object belonging to the component for SNe Ia or Ibc. We first calculate,… view at source ↗
Figure 3
Figure 3. Figure 3: Fitted mixing fraction (α) from our GMM fit based on the log10 fNi and log10 vej features as a function of the true SNe Ibc fraction in the population. Light curve fits were done using spectroscopic redshifts (blue), photomet￾ric redshift estimates (yellow), and no redshift information (green). Dashed lines denote accuracy assuming no labels are known, and solid lines denote accuracy with labels known for … view at source ↗
Figure 4
Figure 4. Figure 4: Overall classification accuracy our GMM fit based on the log10 fNi and log10 vej features as a function of the true SNe Ibc fraction in the population. Light curve fits were done using spectroscopic redshifts (blue), photometric red￾shift estimates (yellow), and no redshift information (green). Dashed lines denote accuracy assuming no labels are known, and solid lines denote accuracy with labels known for … view at source ↗
Figure 5
Figure 5. Figure 5: Purity (top row) and completeness (bottom row) for SNe Ia (left column) and SNe Ibc (right column) from our GMM fit based on the log10 fNi and log10 vej features as a function of the true SNe Ibc fraction in the population. Light curve fits were done using either spectroscopic redshifts (blue), photometric redshift estimates (yellow), and no redshift information (green). Dashed lines denote accuracy assumi… view at source ↗
Figure 6
Figure 6. Figure 6: Example fits of true SN Ia and Ibc light curves with the SALT2 SN Ia light curve model. Points indicate observed data points while solid lines indicate model fits. Bottom panels indicate the “pull”, (data - model)/error. Shaded regions indicate a pull between -1 and 1. parameters. It models the flux F in terms of phase p and wavelength λ for a given SN as: F(p, λ) = x0 × [M0(p, λ) + x1M1(p, λ) + ...] × exp… view at source ↗
Figure 7
Figure 7. Figure 7: The results of the GMM fitting with the four parameters from the SALT2 SN Ia model and the “Arnett” SN Ia model. The gray histograms denote all samples, the blue curve denotes the SN Ia component, the red curve denotes the SN Ibc component, and the black curve denotes the mixture of the two. Dashed lines denote the median for each distribution. The true Ibc fraction is 0.1, and fit mixing fraction is 0.275… view at source ↗
Figure 8
Figure 8. Figure 8: Example fits of our 2D GMM to log10 fNi and log10 vej. Each gray point is a SN in our sample, and the blue and red curves correspond to the fitted SNe Ia and SNe Ibc components, respectively. Marginal histograms are plotted in gray in the top and right panels of each subplot, and the marginal distributions for SNe Ia and Ibc, scaled by their respective mixing fractions, are plotted in blue and red. The lef… view at source ↗
Figure 9
Figure 9. Figure 9: Example fits of our 2D GMM to log10 fNi and log10 vej. Each gray point is a SN in our sample, and the blue and red curves correspond to the fitted SNe Ia and SNe Ibc components, respectively. Marginal histograms are plotted in gray in the top and right panels of each subplot, and the marginal distributions for SNe Ia and Ibc, scaled by their respective mixing fractions, are plotted in blue and red. The lef… view at source ↗
read the original abstract

In the era of large-scale photometric surveys, scalable and robust methods for classifying supernova (SN) populations are increasingly necessary. Often, spectroscopy is essential in addition to photometry to reliably classify SNe; however, complete spectroscopic follow-up is infeasible for all of the millions of transient light curves being collected by facilities such as the Vera C. Rubin Observatory. Using light curves of SNe Ia and Ibc observed with the Zwicky Transient Facility, we frame the classification of large SN populations as a mixing problem. We fit all objects using a semi-analytical SN model powered by radioactive decay, and we model the resulting distributions of fit parameters with a Gaussian Mixture model to optimize the shared population mixing fraction. This approach allows us to reliably constrain the ratio of the populations and classify SNe Ia and Ibc with $\geq$ 90% accuracy without any need for labeled training data, i.e., a spectroscopic dataset. We validate this method for varying population mixing fractions and explore the impact of including spectroscopic, photometric, or no redshift information, and a small amount of known labels. Overall, this method allows for fast and accurate SN classification and population characterization using only photometry.

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.

Referee Report

2 major / 1 minor

Summary. The manuscript frames supernova classification as an unsupervised mixing problem: light curves from ZTF are fit with a semi-analytical radioactive-decay model; the resulting parameter vectors are modeled as a two-component Gaussian mixture whose shared mixing fraction is optimized by maximum likelihood on the unlabeled photometry alone; the fitted fraction then yields per-object Ia/Ibc classifications claimed to reach ≥90% accuracy without spectroscopic labels. The approach is tested for varying mixing fractions and with/without redshift information.

Significance. If the GMM assumption on the fit-parameter distributions holds and the accuracy claim is quantitatively substantiated, the method would provide a scalable, label-free route to both population-ratio estimation and individual classification for the millions of transients expected from Rubin Observatory, reducing dependence on scarce spectroscopic resources.

major comments (2)
  1. [Abstract] Abstract: the central claim of ≥90% classification accuracy on ZTF data across mixing fractions is stated without any quantitative validation metrics, confusion matrices, error budgets, or tests against model misspecification; this leaves the performance assertion unsupported by evidence visible in the manuscript.
  2. [Methods (GMM fitting)] The weakest assumption (semi-analytical model produces parameter vectors whose Ia and Ibc marginals are each adequately described by a single Gaussian, with components sufficiently separated that a shared mixing fraction can be recovered by ML fitting to unlabeled data) is load-bearing for both the mixing-fraction recovery and the subsequent classification accuracy; no explicit check that the actual ZTF-derived distributions satisfy this (e.g., via skewness, multimodality, or overlap diagnostics) is provided.
minor comments (1)
  1. [Methods] Notation for the semi-analytical model parameters and the GMM covariance structure should be defined once in a dedicated subsection rather than introduced piecemeal.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which help clarify the presentation of our results. We respond to each major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of ≥90% classification accuracy on ZTF data across mixing fractions is stated without any quantitative validation metrics, confusion matrices, error budgets, or tests against model misspecification; this leaves the performance assertion unsupported by evidence visible in the manuscript.

    Authors: The quantitative validation, including accuracy as a function of mixing fraction, confusion matrices, and performance with/without redshift, appears in Section 4 and the associated figures. To ensure the abstract claim is immediately supported by visible evidence, we will revise the abstract to include a concise reference to the validation metrics and results. revision: yes

  2. Referee: [Methods (GMM fitting)] The weakest assumption (semi-analytical model produces parameter vectors whose Ia and Ibc marginals are each adequately described by a single Gaussian, with components sufficiently separated that a shared mixing fraction can be recovered by ML fitting to unlabeled data) is load-bearing for both the mixing-fraction recovery and the subsequent classification accuracy; no explicit check that the actual ZTF-derived distributions satisfy this (e.g., via skewness, multimodality, or overlap diagnostics) is provided.

    Authors: We agree that explicit diagnostics would strengthen the manuscript. In revision we will add a dedicated subsection (or appendix) containing skewness, multimodality, and overlap diagnostics on the ZTF-derived parameter distributions, together with Q-Q plots and component-separation metrics, to directly substantiate the single-Gaussian assumption per class. revision: yes

Circularity Check

0 steps flagged

No circularity; unsupervised GMM inference on model-fit parameters is independent of target labels

full rationale

The paper fits a semi-analytical radioactive-decay model to photometric light curves, extracts parameter vectors, and optimizes a shared mixing fraction via maximum-likelihood GMM on the unlabeled distribution. This procedure uses only the photometry-derived parameters and does not reduce to a fit of the target labels or a self-referential definition. Validation against known labels occurs after inference and does not enter the mixing-fraction optimization. No self-citation chains, uniqueness theorems, or ansatzes smuggled via prior work are invoked in the derivation. The GMM Gaussianity assumption is an explicit modeling choice whose validity can be tested externally rather than a definitional loop.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The method depends on two domain assumptions about the adequacy of the decay model and the Gaussianity of parameter distributions; no new entities are postulated and the mixing fraction is the only explicitly optimized free parameter.

free parameters (2)
  • population mixing fraction
    Shared fraction between the two Gaussian components is optimized to explain the observed parameter distribution.
  • semi-analytical model parameters
    Decay-powered model parameters are fitted individually to each light curve before GMM step.
axioms (2)
  • domain assumption Light curves of SNe Ia and Ibc are adequately captured by a single semi-analytical model powered by radioactive decay.
    Invoked to justify fitting every object with the same functional form.
  • domain assumption Fit-parameter distributions for each supernova type are Gaussian and separable by a two-component mixture model.
    Core modeling choice that enables recovery of the mixing fraction.

pith-pipeline@v0.9.1-grok · 5739 in / 1350 out tokens · 31355 ms · 2026-06-29T09:29:41.272175+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references · 30 canonical work pages · 2 internal anchors

  1. [1]

    R., Khatami, D

    Afsariardchi, N., Drout, M. R., Khatami, D. K., et al. 2021, The Astrophysical Journal, 918, 89

    2021

  2. [2]

    Arnett, W. D. 1982, ApJ, 253, 785, doi: 10.1086/159681 Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33, doi: 10.1051/0004-6361/201322068 Astropy Collaboration, Price-Whelan, A. M., Sip˝ ocz, B. M., et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al. 2...

    work page doi:10.1086/159681 1982

  3. [3]

    2025, SNCosmo, v2.12.1 Zenodo, doi: 10.5281/zenodo.15019859

    Barbary, K., Bailey, S., Barentsen, G., et al. 2025, SNCosmo, v2.12.1 Zenodo, doi: 10.5281/zenodo.15019859

    work page doi:10.5281/zenodo.15019859 2025

  4. [5]

    doi:10.1088/1538-3873/aaecbe , eprint =

    Bellm, E. C., Kulkarni, S. R., Graham, M. J., et al. 2019b, PASP, 131, 018002, doi: 10.1088/1538-3873/aaecbe

    work page doi:10.1088/1538-3873/aaecbe

  5. [6]

    A., Gagliano, A., Nugent, A., & Hsu, B

    Boesky, A., Villar, V. A., Gagliano, A., Nugent, A., & Hsu, B. 2026, ApJS, 283, 55, doi: 10.3847/1538-4365/ae4350

    work page doi:10.3847/1538-4365/ae4350 2026

  6. [7]

    2021, The Astronomical Journal, 162, 275

    Boone, K. 2021, The Astronomical Journal, 162, 275

    2021

  7. [8]

    Branch, D., & Wheeler, J. C. 2017, Supernova explosions (Springer)

    2017

  8. [9]

    The Pantheon+ Analysis: Cos- mological Constraints

    Brout, D., Scolnic, D., Popovic, B., et al. 2022, The Astrophysical Journal, 938, 110, doi: 10.3847/1538-4357/ac8e04

    work page doi:10.3847/1538-4357/ac8e04 2022

  9. [10]

    , keywords =

    Cardelli, J. A., Clayton, G. C., & Mathis, J. S. 1989, ApJ, 345, 245, doi: 10.1086/167900

    work page doi:10.1086/167900 1989

  10. [11]

    Chatzopoulos, E., Craig Wheeler, J., & Vinko, J. 2012, The Astrophysical Journal, 746, 121, doi: 10.1088/0004-637X/746/2/121 12 1.5 1.0 0.5 0.0 0.5 1.0 1.5 log10(fNi) 3.6 3.8 4.0 4.2 4.4 4.6 log10(vej) Fit = 0.14 True = 0.10 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 log10(fNi) 3.6 3.8 4.0 4.2 4.4 4.6 log10(vej) Fit = 0.30 True = 0.30 2.0 1.5 1.0 0.5 0.0 0.5 1.0 log...

    work page doi:10.1088/0004-637x/746/2/121 2012

  11. [12]

    LSST Science Book, Version 2.0

    Collaboration, L. S., Abell, P. A., Allison, J., et al. 2009, LSST Science Book, Version 2.0, https://arxiv.org/abs/0912.0201

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  12. [13]

    K., Kasliwal, M

    Das, K. K., Kasliwal, M. M., Fremling, C., et al. 2025, Publications of the Astronomical Society of the Pacific, 137, 044203 de Soto, K., Uzsoy, A. S., & Villar, V. A. 2024, in Machine Learning and the Physical Sciences Workshop @ NeurIPS 2024 de Soto, K. M., Villar, V. A., Berger, E., et al. 2024, The Astrophysical Journal, 974, 169, doi: 10.3847/1538-43...

    work page doi:10.3847/1538-4357/ad6a4f 2025

  13. [14]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , author =

    Dempster, A. P., Laird, N. M., & Rubin, D. B. 1977, Journal of the Royal Statistical Society: Series B (Methodological), 39, 1, doi: 10.1111/j.2517-6161.1977.tb01600.x

    work page doi:10.1111/j.2517-6161.1977.tb01600.x 1977

  14. [15]

    Filippenko, A. V. 1997, Annual Review of Astronomy and Astrophysics, 35, 309, doi: 10.1146/annurev.astro.35.1.309

    work page doi:10.1146/annurev.astro.35.1.309 1997

  15. [16]

    Modeling the probability distribution for cosmological analysis with photometrically classified samples

    Freaza, M. P., & Reis, R. R. R. 2026, Modeling the probability distribution for cosmological analysis with photometrically classified samples, arXiv, doi: 10.48550/arXiv.2605.16513

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.16513 2026

  16. [17]

    A., Sharma, Y., et al

    Fremling, C., Miller, A. A., Sharma, Y., et al. 2020, The Astrophysical Journal, 895, 32, doi: 10.3847/1538-4357/ab8943

    work page doi:10.3847/1538-4357/ab8943 2020

  17. [18]

    K., et al

    Gomez, S., Berger, E., Blanchard, P. K., et al. 2020, The Astrophysical Journal, 904, 74, doi: 10.3847/1538-4357/abbf49

    work page doi:10.3847/1538-4357/abbf49 2020

  18. [19]

    S., et al

    Grayling, M., Thorp, S., Mandel, K. S., et al. 2024, MNRAS, 531, 953, doi: 10.1093/mnras/stae1202

    work page doi:10.1093/mnras/stae1202 2024

  19. [20]

    2007, Astronomy & Astrophysics, 466, 11, doi: 10.1051/0004-6361:20066930

    Guy, J., Astier, P., Baumont, S., et al. 2007, Astronomy & Astrophysics, 466, 11, doi: 10.1051/0004-6361:20066930

    work page doi:10.1051/0004-6361:20066930 2007

  20. [21]

    R., Millman, K

    Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2

    work page doi:10.1038/s41586-020-2649-2 2020

  21. [22]

    A., et al

    Hosseinzadeh, G., Dauphin, F., Villar, V. A., et al. 2020, The Astrophysical Journal, 905, 93

    2020

  22. [23]

    Hunter, J. D. 2007, Computing In Science & Engineering, 9, 90

    2007

  23. [24]

    2019, PASP, 131, 094501, doi: 10.1088/1538-3873/ab26f1

    Kessler, R., Narayan, G., Avelino, A., et al. 2019, PASP, 131, 094501, doi: 10.1088/1538-3873/ab26f1

    work page doi:10.1088/1538-3873/ab26f1 2019

  24. [25]

    Kulkarni, S. R. 2020, Towards An Integrated Optical Transient Utility, https://arxiv.org/abs/2004.03511

    work page arXiv 2020

  25. [26]

    A., & Hlozek, R

    Kunz, M., Bassett, B. A., & Hlozek, R. A. 2007, PhRvD, 75, 103508, doi: 10.1103/PhysRevD.75.103508

    work page doi:10.1103/physrevd.75.103508 2007

  26. [27]

    2011, Journal of Machine Learning Research, 12, 2825

    Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825

    2011

  27. [28]

    A., Fremling, C., Sollerman, J., et al

    Perley, D. A., Fremling, C., Sollerman, J., et al. 2020, The Astrophysical Journal, 904, 35

    2020

  28. [29]

    A., Jegou Du Laz, T., et al

    Rehemtulla, N., Miller, A. A., Jegou Du Laz, T., et al. 2024, ApJ, 972, 7, doi: 10.3847/1538-4357/ad5666 S´ anchez-S´ aez, P., Reyes, I., Valenzuela, C., et al. 2021, AJ, 161, 141, doi: 10.3847/1538-3881/abd5c1

    work page doi:10.3847/1538-4357/ad5666 2024

  29. [30]

    2026, Lightcurve Modelling of 2,205 ZTF DR2 Type˜Ia Supernovae: Implications for SN Ia Physics and Cosmology, arXiv, doi: 10.48550/arXiv.2602.02677

    Sarin, N., Lindsj¨ o, E., Kelsey, L., et al. 2026, Lightcurve Modelling of 2,205 ZTF DR2 Type˜Ia Supernovae: Implications for SN Ia Physics and Cosmology, arXiv, doi: 10.48550/arXiv.2602.02677

    work page doi:10.48550/arxiv.2602.02677 2026

  30. [31]

    G., Gagliano, A., Malanchev, K., et al

    Shah, V. G., Gagliano, A., Malanchev, K., et al. 2025, The Astrophysical Journal, 995, 4

    2025

  31. [32]

    Uzsoy, A. S. M., Thorp, S., Grayling, M., & Mandel, K. S. 2024, MNRAS, 535, 2306, doi: 10.1093/mnras/stae2465

    work page doi:10.1093/mnras/stae2465 2024

  32. [33]

    P., Gagliano, A

    Vidal, E. P., Gagliano, A. T., & Cuesta-Lazaro, C. 2025, arXiv preprint arXiv:2510.14202

    work page arXiv 2025

  33. [34]

    Villar, V. A. 2022, arXiv preprint arXiv:2211.04480

    work page arXiv 2022

  34. [35]

    A., Berger, E., Miller, G., et al

    Villar, V. A., Berger, E., Miller, G., et al. 2019, ApJ, 884, 83, doi: 10.3847/1538-4357/ab418c

    work page doi:10.3847/1538-4357/ab418c 2019

  35. [36]

    A., Hosseinzadeh, G., Berger, E., et al

    Villar, V. A., Hosseinzadeh, G., Berger, E., et al. 2020, ApJ, 905, 94, doi: 10.3847/1538-4357/abc6fd

    work page doi:10.3847/1538-4357/abc6fd 2020

  36. [37]

    2020 , pages =

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261, doi: 10.1038/s41592-019-0686-2

    work page doi:10.1038/s41592-019-0686-2 2020

  37. [38]

    L., Perley, D

    Wise, J. L., Perley, D. A., Sarin, N., et al. 2026, MNRAS, 546, stag130, doi: 10.1093/mnras/stag130

    work page doi:10.1093/mnras/stag130 2026

  38. [39]

    T., Mishra-Sharma, S., & Ashley Villar, V

    Zhang, G., Helfer, T., Gagliano, A. T., Mishra-Sharma, S., & Ashley Villar, V. 2024, Machine Learning: Science and Technology, 5, 045069, doi: 10.1088/2632-2153/ad990d

    work page doi:10.1088/2632-2153/ad990d 2024