Machine Learning May Help Find Dark Matter

What will it take to detect dark matter—the invisible, nigh-intangible substance that might make up five-sixths of all matter in the universe? Dark matter should be all around us, exerting tiny effects on normal matter, but searches have come up empty so far. But a new study suggests that a strategy employing machine learning could help quantum sensors finally hunt it down. Such hyper-sensitive sensors may also have other applications, such as GPS-free navigation, the detection of underground bunkers, and the discovery of gravitational ripples in space-time from the moments after the Big Bang.

We know dark matter exists because of its gravitational impact on the motion of galaxies. But we don’t know what it’s made of, or how it might otherwise interact with the everyday particles that make up you or me. Although scientists have dreamt up dozens of models for dark matter’s potential composition and precise properties, most of these proposals predict infinitesimally small effects on normal particles. One possibility to detect these minuscule interactions is quantum sensors. Quantum effects are vulnerable to outside interference, and quantum sensors capitalize on that fragility in order to respond to the slightest disturbances in the environment—such as an interaction between dark matter and normal matter.

The new study focused on atom interferometers, a kind of quantum sensor that depends on an effect known as superposition, in which one atom can essentially exist in two or more places at once. The sensors have these Schrödinger’s cat–like states where an atom flies down separate paths and then is recombined. Due to particle-wave duality—the quantum phenomenon where particles can act like waves, and vice versa—these atoms interfere with each other, with their waves’ peaks and troughs suppressing or augmenting one other. Examining this interference can reveal the extent of the slightly different motions experienced on the different journeys.

One approach scientists use to boost the sensitivity of these interferometers relies on the laser pulses these sensors employ to split atom waves apart and reflect them back at each other. These laser pulses take the place of conventional mirrors in optical interferometry. Every time time the atoms’ waves bounce of one of these ephemeral mirrors, “the signal we’re looking for can get amplified, much like how light signals can get amplified when bouncing in a mirror-lined cavity,” says Timothy Kovachy, an assistant professor of physics and astronomy at Northwestern University in Evanston, Illinois.

However, the number of times an atom wave can experience such reflections depends on the quality of the atom mirror, and “it’s rather hard to make a good atom mirror,” Kovachy says.

Now, in the new study, Kovachy and his colleagues have revealed a strategy to boost the number of reflections possible off atom mirrors. Using machine learning, instead of reflecting atom waves off a sequence of roughly 10 laser pulses at most, the new approach enables a sequence of roughly 500 pulses.

More Sensitive Atom Interferometers

The new strategy “does not insist on making the perfect atom mirror,” Kovachy says. “Instead, it looks for a way to improve the collective net effort of many different atom mirrors, compensating for the imperfections of each individual atom mirror.” The result is a 50-fold improvement in an atom interferometer’s performance in lab tests.

“When we started this work, I hadn’t really imagined it would be possible to get this degree of improvement,” Kovachy says. “It’s always nice when there are pleasant surprises.”

The researchers now hope to implement their new technique with “the first major search campaigns for dark matter with atom interferometers, which are currently under construction,” Kovachy says. “We expect the first searches to come online in three to five years or so. We hope, in conjunction with better atom optics, to boost their sensitivity by potentially multiple orders of magnitude over what atom interferometers are capable of now.”

More precise atom interferometers may have other applications as well, such as GPS-free navigation, Kovachy says. The satellite links that help enable global navigation satellite systems do not work underwater or underground, and where they do work, they’re susceptible to jamming, spoofing, and the weather. A quantum motion sensor could help serve as the foundation of an inertial navigation system that does not rely on any outside signals.

Kovachy adds that atom interferometers can also help measure the strength of Earth’s gravitational field, which can vary across the planet’s surface depending on the amount of mass concentrated underneath it. Potential applications for such gravity sensors include seeing hidden underground structures, detecting subterranean natural resources, discovering underground archaeological sites, and monitoring volcanic activity and groundwater flows.

“I hadn’t really imagined it would be possible to get this degree of improvement.” —Timothy Kovachy, Northwestern University

Large-scale atom interferometers that are 100 meters tall or even large (in comparison, standard atom interferometers are only 1 or 2 meters tall) that are under development may one day even help detect ripples in spacetime known as gravitational waves. Scientists discovered the first direct evidence of these waves using the Laser Interferometer Gravitational-Wave Observatory (LIGO) in 2015, likely caused when two black holes collided. Atom interferometers could theoretically detect gravitational waves from significantly different events, such as the mysterious proposed epoch known as inflation, when the universe underwent a titanic growth spurt moments after the Big Bang, Kovachy says.

Future research should investigate this new technique with different types of atom interferometers. In the new study, the scientists experimented with a device based on strontium atoms, but “rubidium atoms are definitely the workhorse of atom interferometry,” Kovachy says.

The scientists detailed their findings 11 December in the journal Physical Review Letters.