A Fully Photometric Approach to Type Ia Supernova Cosmology in the LSST Era: Host Galaxy Redshifts and Supernova Classification

The Big Picture

Imagine trying to identify a million strangers from photographs alone, no names, no ID cards, just their faces and the neighborhoods they live in. That’s roughly the challenge astronomers face as the Vera C. Rubin Observatory prepares to come online and flood the world with cosmic explosions.

Type Ia supernovae are the universe’s gold standard for measuring cosmic distances. Decades of observation of these stellar detonations first revealed that the universe’s expansion is accelerating, earning the 2011 Nobel Prize in Physics.

Those early measurements used only a handful of events, each confirmed by spectroscopic follow-up: slow, expensive telescope time spent analyzing an explosion’s light spectrum to verify what type of event it is. When the Legacy Survey of Space and Time (LSST) begins cataloguing the sky from Chile, it will discover close to a million Type Ia supernovae. There is no way to take spectra of them all.

A new study from Mitra et al. develops and tests a fully photometric pipeline, one that relies on brightness measurements alone, to extract dark energy science from LSST’s data without a single spectrum.

Key Insight: By combining neural network classification, photometric redshifts, and Bayesian statistical frameworks, the pipeline achieves a dark energy Figure of Merit of ~150, roughly three times better than the current state of the art from the Dark Energy Survey.

How It Works

The team built their analysis on a simulation framework called ELAsTiCC (Extended LSST Astronomical Time Series Classification Challenge), which generates realistic mock observations mimicking what LSST will actually see. This is not a clean theoretical exercise. The simulations include contamination from other types of exploding stars, incorrect host galaxy associations, and realistic photometric redshift errors.

The pipeline has four steps:

1. Light curve fitting: Each transient’s brightness evolution across six optical filters is fitted using the SALT3 model, extracting peak brightness, color, and stretch parameters that standardize Type Ia supernovae as distance indicators.
2. Host galaxy association and photometric redshifts: Instead of taking a spectrum, the team performs a joint SN+host photometric redshift (photo-z) fit. This estimates distance from how an object’s colors shift with recession, using the host galaxy’s color-brightness profile alongside the supernova light curve.
3. Photometric classification with SCONE: A neural network classifier trained to distinguish true Type Ia supernovae from impostors (core-collapse supernovae, peculiar events) using only multi-band light curve shapes. SCONE assigns each event a probability of being a genuine Type Ia.
4. BEAMS with Bias Corrections (BBC): Rather than cutting the sample to only “clean” events, this statistical framework uses each event’s classification probability to weight its contribution to the final Hubble diagram, the plot of distance versus recession velocity that encodes dark energy’s influence on cosmic expansion.

The combined dataset covers two survey strategies: deep, narrow Deep Drilling Fields (DDF) reaching to high redshift, and the broader Wide Fast Deep (WFD) survey, which contributes a spectroscopically confirmed low-redshift anchor sample. Together they yield about 6,000 simulated events for cosmological analysis.

Why It Matters

The headline result: fitting the wCDM model, where dark energy has a constant but potentially non-standard equation of state, the team achieves a Figure of Merit (FoM) of approximately 150. The FoM measures how tightly an experiment can simultaneously pin down dark energy parameters. Higher is better.

For comparison, the DES-SN5YR analysis, currently among the most powerful supernova cosmology datasets ever published, achieved an FoM of 54. LSST analyzed with fully photometric methods could be roughly three times more constraining.

The team is not declaring victory. Averaging cosmological parameter fits across 25 independent simulation realizations, they find small but statistically significant biases: the recovered dark energy parameters don’t perfectly match the true values baked into the simulations.

These biases are a warning flag. The pipeline works, but it needs refinement before anyone should trust it at LSST’s precision. That’s the whole point of simulation-based stress-testing. The biases point to where development must focus next: better photo-z calibration, more careful contamination modeling, and tighter control of systematic uncertainties.

Neural network classifiers like SCONE can process light curve data at scales no human team could handle. LSST’s supernova sample will test whether dark energy truly behaves like a cosmological constant or whether it changes over time, as hinted by recent results from DESI and Pantheon+. A photometric pipeline that handles a million events without spectroscopic confirmation isn’t a convenience; it’s a scientific necessity.

The CPL w0wa model, which allows dark energy to evolve in time, also shows competitive FoM gains. This positions the framework to address key open questions in cosmology once real LSST data arrive.

Bottom Line: LSST can achieve world-leading dark energy constraints using a fully photometric analysis, no spectra required. The identified systematic biases set a clear agenda for what must be solved before the method is ready for real data.