Autographs of Celestial Duets: DESI Unveils a Thousand Cataclysmic Variables ✛
Five thousand robotic fingers of the DESI spectroscopy reach toward the telescope's focus. In a single night, they record one hundred thousand light signatures. Hidden among these records are binary star systems with white dwarfs — cataclysmic variables. Their spectra are genuine signatures of accretion: matter falls onto the dwarf, swirling into an Accretion disk. Hydrogen and helium lines betray this process.
CVs are natural laboratories for accretion physics and stellar evolution. The most compact of them will serve as calibration sources for the LISA detector. However, until recently CVs were mainly found through their powerful outbursts. This created a bias toward bright systems. Spectroscopic surveys like SDSS provided the first unbiased sample. But their depth wasn't enough to uncover faint old CVs. DESI, aimed at dark energy, went further.
Scientists trained a Machine Learning on SDSS spectra. It made zero errors on the test sample. The algorithm ran through the entire dataset and flagged 293 670 suspicious signatures. Then filters removed distant galaxies by redshift and known stars. Astronomers sifted through the remainder, cross-checking with ZTF light curves and Gaia data.
Human and machine together confirmed 1029 cataclysmic variables. 221 are newcomers. For 84 systems, orbital periods were found: from 80 minutes to several hours. DESI's depth (22,5m) revealed faint short-period CVs. Of 171 new systems with light curves, only 43 had outbursts. The rest remained silent for years.
The period distribution agrees with stellar evolution: systems abound on short orbits. The 'period gap' blurs. Previously, it was invoked to explain the dearth of CVs with intermediate orbits. Now it’s clear: accretion rates are more varied. The space density of short-period subtypes has increased several times. Magnetic systems may be less common than they seemed.
The archive of celestial autographs will keep growing. New releases from DESI and LISA will fill in the picture. Night at Kitt Peak continues to collect the signatures of binary duets. And machine vision learns to read the faintest lines.
One of the new objects, J094708.84+233151.8, threw a strange long outburst, after which its brightness never returned to its previous level — as if the star switched to a new accretion regime and forgot how to turn off the lights. And sifting through 98,9 million spectra by hand would have taken 15 years of non-stop work — the neural network did it in hours.