SALT2–B22 Model for SN Ia Light Curves

• SALT2–B22 Model is a retrained spectral energy distribution framework that characterizes Type Ia supernova light curves for precise cosmological distance measurements.
• It uses two-dimensional cubic B-splines and an updated cross-survey zeropoint calibration to robustly model flux evolution and propagate systematic uncertainties.
• The enhanced model reduces calibration systematics, resulting in modest shifts in dark energy and Hubble constant estimates while strengthening SN Ia cosmology analyses.

The SALT2–B22 Model (commonly referred to as SALT2-B21 or "B21" in accompanying documentation) is the retrained incarnation of the SALT2 spectral energy distribution (SED) model, designed for the empirical characterization of Type Ia supernova (SN Ia) light curves within the Pantheon+ cosmological analysis and related distance-ladder measurements. The model builds upon a critical re-calibration of photometric systems, integrating updated cross-survey zeropoint solutions and the propagation of calibration uncertainties, to tightly constrain both SN Ia distances and cosmological parameters such as the Hubble constant ($H_0$) and dark energy equation of state ($w$) (Brout et al., 2021).

1. Model Framework and Functional Form

The SALT2–B22 model parameterizes the observed rest-frame SN flux as a function of phase and wavelength as follows:

$$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$

where:

• $p$ is the rest-frame phase (days since $B$-band maximum),
• $\lambda$ is the rest-frame wavelength (in Å),
• $x_0$ is the overall flux normalization (proportional to distance$^{-2}$),
• $x_1$ is the "stretch" or shape parameter,
• $c$ is the color parameter (observed $w$0 at peak),
• $w$1 and $w$2 are the mean and first-order SED surfaces, respectively,
• $w$3 is the color-law curve in magnitudes.

The model thus generalizes the spectral evolution of SNe Ia across both photometric and spectroscopic domains, permitting robust light-curve fitting and cosmological distance estimation.

2. SED Surfaces and Color Law Construction

$w$4 and $w$5 are constructed as two-dimensional cubic B-splines over a grid with:

• $w$6 (56 nodes),
• $w$7 Å (69 nodes).

The knots, spline coefficients, and corresponding arrays (i.e., “M0[i] [j]” and “M1[i] [j]”) are contained in the HDF5-formatted B21 release files, making the surfaces machine-interpolable at arbitrary $w$8 values. The model is trained using a library of approximately 800 SNe Ia spanning $w$9 and incorporates data from diverse photometric systems (25 telescopes/cameras, 105 filters), including Landolt, Sloan, CSP, DES, PS1, SNLS, HST, and others.

The color law $$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$0 follows the formalism of the B14 release, with

$$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$1

and a break near $$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$2 Å. Explicitly,

$$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$3

The polynomial coefficients ($$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$4) and breakpoint parameter $$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$5 are provided in the release tables; in B21/B22, these coefficients differ from B14 by a few $$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$6, reflecting the impact of the retraining.

3. Photometric Cross-Calibration and Systematic Propagation

A salient improvement over previous releases is the integration of the SuperCal-Fragilistic ("SuperCal") cross-calibration, which simultaneously solves for zeropoint offsets $$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$7 across all 105 filters, producing a full $$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$8 zeropoint covariance matrix $$F(p, \lambda) = x_0 \left[ M_0(p, \lambda) + x_1 M_1(p, \lambda) \right] \exp\left(c \cdot CL(\lambda)\right)$$9. The calibration reference is Pan-STARRS stellar photometry, with recalibration encompassing new updates to the fundamental HST CALSPEC standards (systematic shift $p$0 over $p$1 Å).

Prior to retraining, each survey’s photometry is shifted by its best-fit $p$2. The full covariance in the zeropoints is propagated through the light-curve model retraining in the following manner:

• Perform Cholesky decomposition of $p$3;
• Generate $p$4 correlated zeropoint realizations;
• For each, shift training photometry and retrain SALT2 to derive new $p$5, $p$6, $p$7;
• Fit the reprocessed sample to extract distances $p$8;
• Estimate the calibration-induced SN distance covariance via

$p$9

This term is then added to the global cosmological covariance matrix, ensuring that calibration systematics are fundamentally coupled to cosmological parameter inference.

4. Differences from Previous SALT2 Releases and Performance Metrics

Compared to B14 (Betoule et al. 2014) and T21, key differences in SALT2–B22 include:

• Mean SED stellar surface ($B$0) shifts of $B$1–3% tilt over 3000–7000 Å, resulting in a redshift-dependent change in distance modulus ($B$2) of 0.04 mag over $B$3.
• The color-law change is subdominant, differing at the level of a few $B$4 magnitudes.
• When using the Tripp formula for SN Ia distances:
  • Calibration-only (no retraining): $B$5 mag low-$B$6/high-$B$7 offset,
  • Retraining-only (no calibration shift): $B$8 mag per $B$9,
  • Combined: $\lambda$0 mag net offset versus B14/JLA.
• Cosmological impact (flat $\lambda$1CDM + Planck $\lambda$2 prior):
  • $\lambda$3 (retraining only),
  • $\lambda$4 (calibration only),
  • $\lambda$5 (combined) when applied to the JLA subset.
• The systematic calibration uncertainty on $\lambda$6 is $\lambda$7 (approximately half the sample’s statistical uncertainty).
• Impact on the SH0ES $\lambda$8 measurement is negligible ($\lambda$9 km s$x_0$0 Mpc$x_0$1), subdominant to the statistical error budget.

5. Release Structure and Implementation Resources

Numerical coefficients, spline grids, knot vectors, color law parameters, and the zeropoint covariance matrix are provided in release HDF5 tables (publicly at https://github.com/PantheonPlusSH0ES/DataRelease/). The distribution enables reproduction and incorporation of the exact retrained model into any light-curve fitter or cosmology codebase. Wherever explicit numerical coefficients are not detailed in published literature, their unique values are available in these accompanying files (Brout et al., 2021).

6. Significance for Cosmological Analyses

By combining retrained light-curve models and contemporaneous cross-survey calibration solutions with explicit propagation of filter-by-filter systematic covariance, SALT2–B22 addresses major sources of uncertainty in SN Ia cosmology pipelines. The net result is a well-characterized, empirically calibrated distance indicator optimized for large multi-survey datasets. The reduction of calibration systematics to $x_0$2 and maintaining $x_0$3 systematic contributions below $x_0$4 km s$x_0$5 Mpc$x_0$6 exemplifies the efficacy of this approach given current sample sizes and calibration limits.

A plausible implication is that, according to the Pantheon+ analysis, SN Ia calibration systematics are now sufficiently controlled that they cannot resolve the extant Hubble tension between distance-ladder and CMB-based $x_0$7 determinations at the present level of statistical uncertainty (Brout et al., 2021).