AI model aims to sharpen supernova measurements for dark energy studies

Researchers led by the Institute of Cosmos Sciences of the University of Barcelona have developed an AI-aided framework meant to improve how astronomers measure the universe’s expansion. The university says the method could matter for dark energy studies because upcoming surveys are expected to detect millions of Type Ia supernova candidates, most of them without detailed spectroscopic follow-up.

The framework, called CIGaRS, is described in a 2026 paper in Nature Astronomy by Konstantin Karchev, Roberto Trotta and Raúl Jiménez. According to the University of Barcelona, CIGaRS is designed to draw more information from images of Type Ia supernovae and the galaxies that host them, rather than relying mainly on spectra, which are costly to obtain at large scale.

Type Ia supernovae occur when white dwarf stars explode, the university said. Astronomers use them as distance markers because their intrinsic brightness is relatively consistent; by comparing that expected brightness with how bright the explosion appears from Earth, researchers can estimate how far away it is.

Those measurements helped establish that the universe’s expansion is accelerating, a finding that scientists connect to dark energy, according to the University of Barcelona. The difficulty is that Type Ia supernovae are not identical in practice. The university said observations over the past two decades show that the apparent brightness of these explosions can depend on the properties of their host galaxies, including age and mass.

CIGaRS addresses that problem by combining several pieces of the measurement problem in one statistical and physical model, according to the research team. The model includes the supernovae, their host galaxies, dust effects, changing supernova rates over cosmic time and the expansion of the universe.

Jiménez, of ICREA-ICCUB, said the approach uses Bayesian inference to simulate the universe from first principles on a computer. He said that lets researchers vary possible parameters together and examine unknown systematic effects, which he called a key missing ingredient in current cosmological modeling approaches.

The team used simulation-based inference to make the calculation practical, according to the university. Researchers generate many simulated universes from physical models, then train a neural network to connect simulated observations with the conditions that produced them. Once trained, the system can compare real observations with those simulations to infer likely physical and cosmological parameters.

According to the University of Barcelona, one result is that CIGaRS can estimate galaxy redshifts from imaging data with accuracy comparable to spectroscopic measurements. Redshift measures how much light has been stretched by cosmic expansion and is used to infer distance and lookback time.

That capability is aimed at the Vera C. Rubin Observatory, now being built in Chile. The university said Rubin’s planned decade-long survey is expected to discover an unprecedented number of supernovae, with roughly 99% observed only photometrically, through images in different colors rather than spectra.

Karchev, of ICCUB and SISSA Trieste and the paper’s lead author, said the end-to-end simulation-based approach is intended to extract cosmological and astrophysical information from Rubin data while reducing selection and modeling biases. The researchers estimate the method could improve cosmological constraints by as much as a factor of four compared with traditional methods that use a smaller set of spectroscopically observed supernovae.

The University of Barcelona said the framework may also help study how Type Ia supernovae form by reconstructing how their occurrence rates vary with stellar ages in different galaxies. The journal paper is titled “CIGaRS I: combined simulation-based inference from type Ia supernovae and host photometry.”

This story draws on original reporting from ScienceDaily.