On the impact of the supernova subsamples in reducing the Hubble tension

1. Problem Statement

The paper addresses the Hubble Tension, the persistent $4\text{--}6\sigma$ discrepancy between early-universe measurements of the Hubble constant ($H_0 \approx 67.4 \pm 0.5$ km s$^{-1}$ Mpc$^{-1}$ from Planck CMB data) and late-universe measurements ($H_0 \approx 73.04 \pm 1.04$ km s$^{-1}$ Mpc$^{-1}$ from the SH0ES collaboration using the cosmic distance ladder).

The authors hypothesize that this tension may be exacerbated by systematic biases arising from the mismatch between the Calibration sample (Type Ia Supernovae, SNe Ia, in nearby galaxies hosting Cepheids) and the Hubble Flow (HF) sample (SNe Ia in more distant galaxies). Specifically, they investigate whether differences in light-curve parameters (stretch $x_1$, color $c$) and host galaxy properties (stellar mass $M$, specific star formation rate sSFR) between these two samples lead to incorrect standardization of SN luminosities, thereby biasing the $H_0$ estimation. They also explore the existence of distinct SN Ia subpopulations (e.g., low-stretch vs. high-stretch) that may have different intrinsic properties and dust environments, which are currently treated as a single population in standard analyses.

2. Methodology

The study utilizes the Pantheon+SH0ES dataset, comprising 37 calibration SNe and 216 HF SNe (after filtering outliers).

• Data Processing: Repeated observations of the same SN were combined into single representative measurements using Gaussian distribution products to handle covariance. Outliers in sSFR were removed.
• Standardization Model: The authors use a modified Tripp formula:
  $$m_B^{corr} = m_B + \alpha x_1 - \beta c + \delta_{host}$$
  where $\delta_{host}$ is a mass-step correction. Notably, they exclude the bias correction term ($\delta_{bias}$) and intrinsic scatter floor ($\sigma_{floor}$) used in some other analyses to isolate the raw impact of sample mismatches and subpopulations.
• Statistical Approach:
  • Kolmogorov-Smirnov (K-S) Tests: Used to compare distributions of $x_1$, $c$, $\log(M)$, and $\log(\text{sSFR})$ between the calibration and HF samples.
  • Subsample Generation: 4,000 pairs of random subsamples were generated from the calibration and HF sets. The authors varied the size and composition to find pairs with the best distributional match (high multi-dimensional p-value, $p_{MD}$).
  • Bayesian Fitting: A Monte Carlo Markov Chain (MCMC) approach (using emcee) was used to estimate parameters ($H_0, M_B, \alpha, \beta, \Delta_{host}, \sigma_{int}$) for each subsample pair.
  • Binning Analysis: The full sample was split into "low-stretch" ($x_1 < \tilde{x}_1$) and "high-stretch" ($x_1 \geq \tilde{x}_1$) bins based on the median stretch value. Parameters were estimated independently for each bin.
  • Alternative Clustering: Gaussian Mixture Models (GMM) were applied to $x_1$ and $\log(M)$ to test robustness of subpopulation separation.

3. Key Contributions

1. Quantification of Sample Mismatch: Demonstrated statistically that the calibration sample is not representative of the HF sample regarding host stellar mass and sSFR, despite similarities in light-curve parameters.
2. Subpopulation Identification: Provided strong evidence that SNe Ia consist of distinct subpopulations (low-stretch vs. high-stretch) with significantly different cosmological parameters when analyzed separately.
3. Re-evaluation of Uncertainty: Argued that current $H_0$ uncertainties are underestimated because they fail to account for the dispersion introduced by mixing distinct SN subpopulations.
4. Origin of the Mass Step: Proposed that the "mass step" (the luminosity jump at a host mass threshold) is likely an artifact of over-correction when a single standardization model is applied to a mixture of subpopulations with different intrinsic properties and dust laws, rather than a single physical step.

4. Key Results

A. Sample Mismatch and Parameter Sensitivity

• Host Properties: The calibration sample lacks low-mass hosts ($\log(M/M_\odot) < 9$) and has higher sSFR values compared to the HF sample. The K-S test p-values for mass and sSFR are $< 0.05$, indicating different underlying distributions.
• Parameter Trends: As the match between calibration and HF subsamples improves (higher $p_{MD}$):
  • $H_0$ and $M_B$ decrease slightly.
  • The intrinsic scatter $\sigma_{int}$ decreases significantly.
  • The mass step ($\Delta_{host}$) becomes consistent with zero within $1\sigma$ for well-matched subsamples.

B. Impact of SN Subpopulations (Stretch Binning)

When the sample is divided by the median stretch ($\tilde{x}_1 = 0.087$):

• Low-Stretch Bin ($x_1 < \tilde{x}_1$): Yields $H_0 = 75.27 \pm 1.18$ km s$^{-1}$ Mpc$^{-1}$ and $M_B = -19.155$. This bin shows a tension of $6.07\sigma$ with Planck.
• High-Stretch Bin ($x_1 \geq \tilde{x}_1$): Yields $H_0 = 71.25 \pm 1.59$ km s$^{-1}$ Mpc$^{-1}$ and $M_B = -19.293$. This bin shows a tension of $2.52\sigma$ with Planck.
• Significance: The discrepancy between the two bins is $2.00\sigma$ for $H_0$ and $2.23\sigma$ for $M_B$. Random partitioning tests confirm these differences are unlikely to be due to chance ($P < 0.5\%$).
• Mass Step Resolution: When analyzed separately, the mass step $\Delta_{host}$ is consistent with zero for both bins ($\sim 1\sigma$), suggesting the mass step in the full sample arises from mixing these populations.

C. Revised Hubble Constant and Tension

By treating the two subpopulations as distinct and accounting for the additional dispersion in the full sample:

• The authors derive a revised $H_0$ for the full sample: $H_0 = 73.78 \pm 2.17$ km s$^{-1}$ Mpc$^{-1}$.
• The uncertainty ($\pm 2.17$) is roughly 2.3 times larger than the standard SH0ES uncertainty ($\pm 1.04$).
• Impact on Tension: This increased uncertainty reduces the tension with the Planck measurement from $5.87\sigma$ to approximately $2.86\sigma$.

D. Physical Interpretation

• Dust and Intrinsic Color: The high-stretch (younger) population appears to have dust properties similar to the Milky Way ($R_V \approx 4.3$), while the low-stretch (older) population has different dust properties ($R_V \approx 3$).
• Standardization Failure: The standard single $\beta$ (color-luminosity) correction assumes a universal dust law and intrinsic color relation. The authors argue this assumption fails for the mixed sample, leading to biased $H_0$ estimates, particularly for the low-stretch population which is over-luminous after standardization.

5. Significance

This paper challenges the assumption that the local distance ladder can be treated as a homogeneous population. Its primary significance lies in:

1. Reducing the Hubble Tension: By acknowledging the heterogeneity of SN Ia populations, the statistical tension with early-universe data drops from a critical $>5\sigma$ to a manageable $\sim 2.9\sigma$, suggesting the tension may be largely systematic rather than requiring new physics.
2. Methodological Shift: It highlights the need for subpopulation-aware standardization. Future analyses should likely separate SNe by stretch or host environment before applying corrections, or use multi-component models for color and stretch corrections.
3. Reinterpreting Systematics: It offers a compelling explanation for the "mass step" as a statistical artifact of population mixing rather than a fundamental astrophysical step, potentially resolving a long-standing issue in SN Ia cosmology.

In conclusion, the authors suggest that the Hubble tension is significantly alleviated when the distinct physical properties of SN Ia subpopulations are properly accounted for, implying that the current "best" value of $H_0$ may be an average of two distinct cosmological measurements with different systematic behaviors.