SN Ia Residual Host-Mass Luminosity Step

• The residual host-mass luminosity step is a systematic difference in standardized SN Ia brightness between low- and high-mass host galaxies, occurring near log(M*/M⊙) ∼ 10.
• Empirical measurements report a brightness offset ranging from 0.04 to 0.15 mag using methodologies like discrete Heaviside or smooth sigmoid models with multi-parameter bias corrections.
• This phenomenon impacts cosmological parameter estimates, such as H₀ and dark energy properties, necessitating improved progenitor diagnostics and refined bias mitigation techniques.

A residual host-mass luminosity step is a robust, empirically established feature of Type Ia supernova (SN Ia) cosmology in which standardized SN Ia distances exhibit a systematic offset between supernovae occurring in low-mass and high-mass host galaxies, after correction for light-curve stretch and color. This phenomenon is manifest as a discrete change—commonly labeled the "mass step"—in Hubble residuals (the difference between observed and model-predicted distance modulus) across a threshold host stellar mass of $\log_{10} (M_*/M_\odot) \sim 10$, with SNe Ia in more massive hosts appearing brighter by $0.04$–$0.08$ mag following standardization with linear Tripp-formula models. The mass step is a focal point for SN cosmology systematics, as its origin, amplitude, and treatment directly affect inferences of Hubble constant $H_0$, dark energy properties, and the integrity of SN Ia standard candle methodology.

1. Formal Definition and Measurement of the Mass Step

The residual host-mass luminosity step, commonly parametrized as $\gamma$, is defined by splitting the SN Ia sample at a characteristic host galaxy mass, typically $10^{10} M_\odot$, and measuring the difference in mean Hubble residuals:

$$\Delta\mu_{\mathrm{mass}} = \gamma = \langle \Delta\mu \rangle_{M_* < 10^{10} M_\odot} - \langle \Delta\mu \rangle_{M_* > 10^{10} M_\odot}$$

where the Hubble residual $\Delta\mu$ for each SN is

$$\Delta\mu = \mu_{\mathrm{obs}} - \mu_{\mathrm{model}}(z)$$

with $\mu_{\mathrm{obs}}$ derived via Tripp standardization,

$$\mu_{\mathrm{obs}} = m_B - (M_B - \alpha x_1 + \beta c) + \Delta_{\mathrm{Bias}}$$

Here, $m_B$ is rest-frame peak B-band magnitude, $x_1$ the light-curve stretch, $c$ the color, and $\Delta_{\mathrm{Bias}}$ a term accounting for observational selection effects and survey-specific biases (Smith et al., 2020).

Mass step amplitudes reported across recent surveys, including DES, SNfactory, Pantheon+, and ZTF DR2, range from $\sim 0.04$ to $0.15$ mag depending on methodology and sample selection (Smith et al., 2020, Childress et al., 2013, Gonzalez et al., 23 Dec 2025, Burgaz et al., 2 Sep 2025). The step is most often modeled as a Heaviside (discrete) or sigmoid (smooth) function of $\log M_*$, and is generally robust to choices of SED fitting, photometric bandpasses, and analysis techniques (Hand et al., 2021, Childress et al., 2013).

2. Empirical Characterization: Magnitude, Functional Form, and Population Dependence

Survey data consistently identify two plateaus in Hubble residuals at low and high host masses with a rapid transition at $\log_{10} (M_*/M_\odot) \sim 10$ (Childress et al., 2013, Johansson et al., 2012). Snfactory’s composite binned analysis, for example, yields a step

$$\Delta\mu_{\mathrm{step}} = 0.077 \pm 0.014 ~\mathrm{mag}$$

with fits to an error function of the form

$$\Delta\mu(M_*) = \mu_{\mathrm{low}} + (\mu_{\mathrm{high}} - \mu_{\mathrm{low}})\, \frac{1}{2} \left[1 + \operatorname{erf} \left( \frac{\log M_* - m_t}{\sqrt{2}\sigma} \right)\right]$$

where $\mu_{\mathrm{low}}$ and $\mu_{\mathrm{high}}$ are the plateau values and $m_t$ the transition mass (Childress et al., 2013).

The step is not an artifact of analysis choices. Robustness tests show that neither variations in mass estimation method, UV photometric coverage, nor fitting approach (e.g., photometric vs. spectroscopic masses) can synthesize a spurious mass step or significantly alter its size (Hand et al., 2021). The step persists under partial or full Bayesian hierarchical models (Thorp et al., 2022), with best-fit values for $\gamma$ around $0.06$–$0.09$ mag.

3. Physical Origin: Underlying Causes and Progenitor Diagnostics

A broad consensus has emerged that the mass step encodes physical differences in SN Ia progenitor environments. Key lines of evidence include:

• Bimodal Age Distribution: The galaxy mass–mean age relation is highly nonlinear, creating a bimodal age distribution in SN Ia hosts. Empirically, Hubble residuals vary $\sim -0.035$ mag/Gyr with host age, and convolving this with the age–mass relation produces a step-like feature at $\log_{10} (M_*/M_\odot) \sim 10$ of amplitude $\sim 0.11$ mag (Chung et al., 2023).
• Progenitor Channel Mix: Simulations and observational data indicate the step is associated with a change in the fraction of prompt (young) versus tardy (old) SN Ia progenitors as a function of galaxy mass. In the SNfactory sample, a transition in the prompt fraction near $\log_{10} M_*\sim 10.5$ can reproduce both the amplitude and sharpness of the observed step (Childress et al., 2013).
• Metallicity Effects: Recent hierarchical analyses found that host metallicity correlates more strongly with Hubble residuals than mass per se; after controlling for metallicity (cutting low-metallicity hosts), the mass step disappears ($b|_{Z>-1} = -0.004\pm0.018$ mag/dex), implicating chemical environment as a primary driver (Gonzalez et al., 23 Dec 2025).
• Spectroscopic Subtypes: The step is significantly larger for normal-velocity SNe Ia ($\Delta M_{\rm NV} = 0.149\pm0.024$ mag) than for high-velocity subtypes ($0.046\pm0.041$ mag), particularly for events in central, apparently older and metal-rich regions (Burgaz et al., 2 Sep 2025).

The mass step is thus best conceptualized as a convolution of a nonlinear galaxy property (age, metallicity) distribution and approximately linear residual correlation with SN standard-candle correction.

4. Role of Bias Corrections, Light Curve Model Choices, and Environmental Proxies

Bias correction schemes underlie key differences in reported mass step amplitudes:

• 1D (redshift-only) vs 5D (multi-parameter) Bias Corrections: DES-SN finds that a 5D correction reduces the measured mass step by $\sim0.03$ mag compared to 1D, due to a correlation between host mass and $x_1$ (light-curve width); unmodeled, this can misattribute part of the effect to selection biases (Smith et al., 2020).
• Light-Curve Model Expansion: Augmenting the SALT3 SN Ia model to include explicit host-mass surfaces (e.g., M_host(p, λ)) removes $\sim$35% of the mass step by capturing luminosity-independent spectral differences, but most (65%) of the step persists and must be ascribed to true luminosity dependence (intrinsic or extinction-driven) (Jones et al., 2022).
• Custom Light Curve Models: Splitting SALT3 training by host mass and fitting subsamples independently absorbs the mass step into encoded zero-point differences, but does not physically remove an underlying luminosity offset (Taylor et al., 2024).

Metallicity- or age-based corrections are advocated as more physical alternatives to pure mass-split treatments, with the potential to further reduce systematics in precision cosmology (Gonzalez et al., 23 Dec 2025, Chung et al., 2023).

5. Spectral Features, Local vs. Global Environment, and NIR Results

• Spectral Indicators: Models with host-mass explicit surfaces report phase- and wavelength-dependent differences in Si II, Ca II line strengths, and B–V color evolution (e.g., $\Delta$\rm EW(Ca II) significant at 2.2–2.7$\sigma$), indicating that the impact of the host extends beyond simple brightness shifts (Jones et al., 2022, Taylor et al., 2024).
• Local Environment: Analyses separating SNe by projected galactocentric distance demonstrate a significant reduction of the mass step in outer (>1 DLR) regions ($0.036\pm0.018$ mag, $2\sigma$) compared to inner regions ($0.100\pm0.014$ mag, $6.9\sigma$) (Toy et al., 2024). In central regions, the mass step is driven primarily by NV SNe Ia, consistent with environmental modulation (Burgaz et al., 2 Sep 2025).
• Near-Infrared (NIR) Measurements: Several studies show that the mass step is absent or reduced in NIR data (e.g., $\Delta J = -0.021\pm0.033$ mag), particularly when correcting for individual extinction law $R_V$ per SN, suggesting dust contributes to the optical step, but that a residual may persist due to intrinsic or progenitor-related effects (Johansson et al., 2021, Peterson et al., 2024, Thorp et al., 2022).

6. Cosmological Implications and Correction Strategies

Uncorrected, the residual mass step introduces a systematic bias in SN Ia distance moduli at the $0.05$–$0.15$ mag level, corresponding to several percent in cosmological parameter inference (e.g., $H_0$, $w$). The step is redshift-dependent, as the underlying population properties (age, metallicity) evolve (Chung et al., 2023, Gonzalez et al., 23 Dec 2025, Smith et al., 2020).

Current and next-generation analyses employ a step correction term: $$\mu_{\mathrm{corr}} = m_B - M_B + \alpha x_1 - \beta c - \gamma H(\log M_* - 10)$$ where $\gamma\sim0.06$–$0.08$ mag. More sophisticated correction schemes, such as hierarchical Bayesian marginalization over host mass, age, or metallicity (and correlated dust parameters), achieve further improvements in bias mitigation and are necessary for LSST- and Roman-level precision (Thorp et al., 2022, Gonzalez et al., 23 Dec 2025).

7. Summary Table: Observed Mass Step Measurements

Survey / Method	Mass Step $\gamma$ (mag)	Host Mass Threshold	Notes
DES-SN 5D correction (Smith et al., 2020)	$0.040\pm0.019$	$\log M_*/M_\odot=10.0$	5D bias correction
DES-SN 1D correction (Smith et al., 2020)	$0.066\pm0.020$	$\log M_*/M_\odot=10.0$	1D bias correction
SNfactory + SNLS/SDSS (Childress et al., 2013)	$0.077\pm0.014$	$\log M_*/M_\odot=10.0$	Combined sample
Pantheon+ (Gonzalez et al., 23 Dec 2025)	$0.060\pm0.013$	$\log M_*/M_\odot=10.0$	Smooth sigmoid, linear
ZTF DR2 (NV subsample) (Burgaz et al., 2 Sep 2025)	$0.149\pm0.024$	$\log M_*/M_\odot=10.0$	NV (SiII<$12,000$ km/s)
NIR (iPTF) (Johansson et al., 2021)	$-0.021\pm0.033$	$\log M_*/M_\odot=10.0$	Individual $R_V$

Measured values are robust to mass estimation scheme, bandpass, and modeling approach, modulo the caveats detailed above.

• (Smith et al., 2020, Childress et al., 2013, Johansson et al., 2012, Gonzalez et al., 23 Dec 2025, Taylor et al., 2024, Chung et al., 2023, Hand et al., 2021, Johansson et al., 2021, Jones et al., 2022, Thorp et al., 2022, Peterson et al., 2024, Toy et al., 2024, Burgaz et al., 2 Sep 2025).