A new planet-hunting algorithm suggests that at least 9 percent of nearby stars could host planets orbiting out of sight — and the stars’ chemistry could help find the worlds.
Planetary astrophysicist Natalie Hinkel of the Southwest Research Institute in San Antonio and colleagues trained a machine-learning algorithm on a catalog of thousands of stars and their chemical compositions (SN: 5/11/19, p. 34). In the dataset of stars located within about 500 light-years of the sun, 290 were known to host giant planets, while more than 4,200 didn’t — or so astronomers thought.
First, the algorithm analyzed the chemistry of the planet-hosting stars. Then, based on what it learned about those celestial objects, the program estimated the probability that each of the stars in the other group actually does host planets.
It works similarly to how online TV streaming companies like Netflix choose which TV shows to recommend to viewers, Hinkel says. “If I watch a bunch of movies, Netflix learns that I like science fiction, martial arts movies and British period movies,” she says. The program then uses that knowledge to identify other shows she might like — that is, the stars with planets not yet detected.
The new algorithm identified 368 additional stars — or about 9 percent of the stars thought to be lacking planets — that had a more than 90 percent probability of hosting a giant exoplanet, Hinkel will report June 26 in Seattle at the Astrobiology Science Conference. “That was way more than I was expecting,” she says.
The stellar elements that best predicted a potential planet’s presence were iron, carbon, oxygen and sodium. But the ratios of those elements to each other seemed to matter more than just having a lot of each one. The way the elements interact in a planet-forming disk around a star probably shapes planet formation, similar to how baking ingredients interact to make a cake rise, Hinkel says (SN: 5/12/18, p. 28).