Standard Candles, Sharper Light

On June 30, 2026, the Vera C. Rubin Observatory in Chile officially began the Legacy Survey of Space and Time, a ten-year project that will photograph the entire visible southern sky every few nights. Among the flood of data it starts generating is an uncomfortable statistical fact: roughly 99 percent of the exploding stars it discovers will never get a follow-up spectrum. There simply isn't enough telescope time on Earth to chase every one of the hundreds of thousands of supernovae Rubin is expected to find. For decades, that would have made most of them useless for precision cosmology. A framework published seven weeks earlier in Nature Astronomy suggests it no longer has to.

The framework is called CIGaRS, built by a team led by Konstantin Karchev at the Institute of Cosmos Sciences of the University of Barcelona, and its premise is almost impertinent in its ambition: measure the distance to an exploding star using only its picture, no spectrum required, and do it with a precision that rivals — the team's results suggest quadruples — what earlier photometry-only methods could manage. If it holds up at scale, CIGaRS may be the single piece of software best positioned to make Rubin's flood of supernova data scientifically usable rather than merely enormous.

Why an exploding star is a ruler

To understand why any of this matters, it helps to remember why astronomers care about exploding stars in the first place. A Type Ia supernova occurs when a dead star called a white dwarf either pulls enough material from a companion, or merges with another white dwarf, to cross a well-defined mass threshold and detonate in a runaway thermonuclear explosion. Because that threshold is close to the same everywhere in the universe, the explosions release, to a good first approximation, the same amount of energy. Measure how bright one looks from Earth, compare that to how bright it should intrinsically be, and you can calculate how far away it is — the same logic as judging the distance to a candle of known wattage by how dim it appears. That is why astronomers call Type Ia supernovae standard candles, and why a chain of them, at increasing distance, gave us the discovery that won the 2011 Nobel Prize in Physics: the expansion of the universe is accelerating, driven by something we still cannot directly detect and call, for lack of a better word, dark energy.

Getting that measurement right has always depended on knowing a supernova's precise brightness curve — how it flares and fades over weeks — which traditionally required not just images but spectra: light broken into its component wavelengths, revealing the chemical fingerprint that confirms a supernova's exact type and corrects for effects like dust reddening along the line of sight. Spectra are expensive. They require dedicated telescope time, often on some of the largest instruments in the world, pointed at one object for minutes to hours. A sky survey that finds a supernova every few minutes cannot possibly get a spectrum for each one.

Reading the star by reading its neighborhood

CIGaRS's central trick is to stop treating the supernova as an isolated object and instead model it jointly with the galaxy that hosts it. A supernova's host galaxy carries information — its age, its chemical composition, the population of stars living in it — that correlates with subtle variations in how the supernova itself explodes and fades. Earlier methods either ignored this connection or handled it with simplified corrections bolted on after the fact. CIGaRS instead builds a single, unified statistical model — what its authors describe as a combined inference framework — that reads the supernova's light curve and its host galaxy's properties together, using machine learning to extract far more information from an ordinary multi-color image than a human analyst, or a simpler pipeline, ever could.

The result, reported in the team's Nature Astronomy paper, is that photometric, image-only distance measurements can now approach the precision that used to require spectroscopy. That is not a marginal improvement; it is the difference between a supernova being scientifically usable or being discarded. Multiply that difference across the hundreds of thousands of supernovae Rubin is expected to catalog over its ten-year run, and the mathematics of dark energy research change: rather than compiling a precious, spectroscopically confirmed sample of a few thousand supernovae, cosmologists could soon draw on a photometric sample two to three orders of magnitude larger.

The mystery this is actually chasing

None of this would matter as much if dark energy were a settled question, but it is very much not. Recent supernova and galaxy-survey results have hinted that dark energy's strength might not be perfectly constant over cosmic time, as the simplest models assume — a possibility that, if confirmed, would upend the standard cosmological model that has held since the 1990s. Testing that possibility rigorously requires exactly what CIGaRS promises: very large numbers of precisely measured supernovae spread across a wide range of cosmic distances and epochs, so that any drift in dark energy's behavior over billions of years becomes statistically visible rather than lost in noise.

There is also the long-running Hubble tension to consider — the stubborn, still-unresolved mismatch between two different ways of measuring how fast the universe is expanding today, one based on the nearby universe (supernovae among them) and one based on the light left over from the Big Bang. A dramatically larger, more precise supernova catalog will not resolve that tension by itself, but it substantially narrows the room for the mismatch to be hiding in measurement error rather than in new physics.

A rivalry that started with a Nobel Prize

It is worth remembering how contested this territory has always been. When two competing teams — one led by Saul Perlmutter, the other jointly by Brian Schmidt and Adam Riess — independently measured distant Type Ia supernovae in the late 1990s and both arrived at the same startling conclusion, that the universe's expansion was speeding up rather than slowing down, the finding was so unexpected that both groups spent months trying to find the error in their own data before accepting it. The 2011 Nobel Prize in Physics that followed was not just recognition of a discovery; it was recognition that a class of exploding stars, studied carefully enough, could reveal something about the fundamental structure of reality that no other instrument at the time could touch.

Every advance in supernova cosmology since has been, in effect, an attempt to extend that original insight to larger samples with smaller errors. Digital sky surveys in the 2000s and 2010s pushed the supernova count from dozens to thousands. Rubin, paired with frameworks like CIGaRS, aims to push it from thousands to hundreds of thousands — and the Nancy Grace Roman Space Telescope, expected to begin its own wide-field survey work later this decade, is being designed from the outset around exactly this kind of imaging-heavy, spectroscopy-light supernova cosmology, on the assumption that tools like CIGaRS will exist to make sense of what it captures. The European Space Agency's Euclid mission, already returning data, sits in the same lineage: enormous imaging surveys, increasingly reliant on machine learning to extract cosmological signal from pictures rather than from painstaking one-object-at-a-time spectroscopy.

What still has to be proven

CIGaRS is a methods paper, not yet a finished cosmological result, and the honest caveats matter. The framework has been validated against existing datasets and simulations rather than against a full decade of live Rubin data, which does not yet exist. Systematic errors that are subtle at the scale of thousands of supernovae — small biases in how dust, galaxy type, or instrument calibration interact with the model's assumptions — can become significant once the sample grows a hundredfold, precisely because there is so much less room for a systematic error to average itself away. The astronomy community's next few years will be spent doing exactly the unglamorous work of stress-testing CIGaRS and frameworks like it against real Rubin data as it arrives, checking that the AI model's confidence in its own distance estimates is actually earned and not simply confidently wrong.

Still, the timing is hard to overstate. Rubin's first-light images, released earlier this year, were met with a mix of celebration and quiet panic among cosmologists who understood immediately that the survey would produce a torrent of data no existing analysis pipeline was built to handle. CIGaRS did not arrive because someone anticipated Rubin specifically; it arrived from a research trajectory already underway to solve photometric distance measurement generally. But the coincidence of timing — a decade-long survey capable of imaging-only supernova discovery at unprecedented scale, arriving just as a framework capable of extracting precision cosmology from exactly that kind of data reaches publication — is the sort of alignment that observational astronomy gets only rarely, and tends to remember for a long time afterward.

If CIGaRS holds up against real Rubin data over the next several years, the practical effect will be a supernova cosmology sample larger than everything astronomers have collected in the field's entire history, combined within a matter of a few survey seasons. Whether that sample finally pins down what dark energy is, or simply sharpens the mystery into a stranger and more precise shape, the tool that gets us there will not be a bigger mirror or a more sensitive spectrograph. It will be a model that learned to read a photograph of the sky the way a spectrograph reads light — one galaxy, and one exploding star, at a time.