Goddard Scientist Looks to AI, Lensing to Find Masses of Free-Floating Planets

This illustration shows a Jupiter-like planet alone in the dark of space, floating freely without a parent star. The planet survey, Microlensing Observations in Astrophysics, scanned the central bulge of our Milky Way galaxy using the microlensing technique CUμLUS seeks to employ. Image credit: NASA/JPL-Caltech

Exoplanet hunters have found thousands of planets orbiting close to their host stars, but relatively few of these alien worlds follow more distant orbits – and even fewer float freely through the galaxy, not bound to any star.

NASA’s Roman Space Telescope will discover many more planets by observing dense star fields to maximize the chances of detecting an intervening planet as it passes precisely in front of a distant star. These chance alignments cause background stars to brighten briefly. The planet’s gravity acts as a lens that magnifies light from the background star’s light. One drawback of this technique, called gravitational microlensing, is that the distance to the lensing planet is poorly known.

Goddard scientist Dr. Richard K. Barry is working to exploit parallax effects to pin down these distances. Parallax is the apparent shift in the position of a foreground object as seen by observers in slightly different locations. Our brains exploit the slightly different views of our eyes so we can see in 3D. Astronomers in the 19th century first measured distances to nearby stars using the same effect, measuring how their positions shifted relative to background stars in photographs taken when Earth was on opposite sides of its orbit.

It works a little differently with microlensing. In this case, two well-separated observers, each equipped with a precise clock, may observe the same microlensing event. The time delay between the two detections allows scientists to determine the distance to the lensing object.

Barry is developing a concept called the CUbesat MicroLensing Uniform Surveyor (CUμLUS, read: cumulus). It’s designed to make independent observations of some of the Roman lensing events by observing the same star fields. His goal is to develop a concept compatible with the new NASA Astrophysics Pioneers program, which aims to develop small low-cost missions) to gather information about these free-floating worlds and investigate their origins. CUμLUS will rely on gravitational microlensing, an effect first described by Albert Einstein in the 1930s and now used as a technique to detect planets or other objects that emit little or no light, such as black holes or neutron stars.

Microlensing works because the warped spacetime around a massive object moving between a distant star and an observer magnifies the source’s light. As the precise alignment between source and lens changes due to the object’s motion in space, so does the brightness of the source. How the event plays out provides information about the lensing object.

If only one observer views the lens’ passage, it is exceedingly difficult to determine its mass because its distance is unknown. If two observers view from different angles, however, they will see the event occur at slightly different times. Determining the amount of time that elapses between each detection makes it possible to determine the distance between the lens and the observer, yielding a more precise mass estimate of the lens. This technique is called microlensing parallax.

The CUμLUS project would support the Roman Space Telescope and PRime-focus Infrared Microlensing Experiment (PRIME), a terrestrial telescope currently in development using four detectors loaned by the Roman mission. While Roman and PRIME will detect many hundreds of microlensing planets, mass estimates for these objects will be significantly improved using simultaneous parallax observations provided by CUμLUS.

“CUμLUS would be at a great distance from the principal observatory, either Roman or a terrestrial telescope,” Barry said. “The parallax signal should then permit us to calculate quite precise masses for these objects, thereby increasing scientific return.”

CUμLUS would hitch a ride on a Mars-bound mission in four or five years and boost off past Mars around the Sun, where it could orbit at a sufficient distance from Earth to effectively measure the microlensing parallax signal and fill in this missing information.

Additionally, Barry said, if they are fortunate enough to detect a free-floating planet, the only way to confidently determine its mass, without resorting to galactic models to estimate a probable distance to the lens, will be through microlensing parallax.

Stela Ishitani Silva, research assistant at Goddard and Ph.D. student at the Catholic University of America in Washington, DC, said understanding these free-floating planets will help fill in some of the gaps in our knowledge of how planets form.

Some hypotheses predict there are many freefloating planets, while others predict only a few, she said. A survey in 2006 to 2007 found ten such planets through a microlensing technique from Mount John University Observatory in New Zealand.

“So we really need to know how many free-floating planets are there,” Ishitani Silva said. “Can we get a statistical analysis of it? Because it’s not just about finding one planet — we want to find multiple free-floating planets and try to obtain information about their masses, so we can understand what is common or not common at all.”

The project is currently receiving Internal Research and Development (IRAD) funding to accomplish the necessary preliminary work on orbit, optics, artificial intelligence and science arguments as well as operating a Mission Planning Lab run at NASA’s Wallops Flight Facility, in Chincoteague, Virginia.

“I’m really grateful for the IRAD support, because these very necessary steps simply wouldn’t be possible without it,” Barry said.

In order to efficiently find these planets, CUμLUS will use artificial intelligence. Dr. Greg Olmschenk, a postdoctoral researcher working with Barry at Goddard, has developed an AI, called RApid Machine learnEd Triage (RAMjET), for the mission.

“I work with certain kinds of artificial intelligence called neural networks,” Olmschenk said. “It’s a type of artificial intelligence that will learn through examples. So, you give it a bunch of examples of the thing you want to find, and the thing you want it to filter out, and then it will learn how to recognize patterns in that data to try to find the things that you want to keep and the things you want to throw away.”

Prior to beginning its mission, the AI is trained to know what to look for. This process typically begins by showing the AI, for example, a light curve — a graph that shows the brightness of an object over a particular period of time — and asking it to confirm or deny what it is seeing in the light curve. The AI starts out by searching for random patterns, and then small adjustments are made to the patterns to get it closer to the right answer. Then, the researcher gives the AI another example and the process starts over.

“And then you repeat this process many, many times,” Olmschenk said. “And this gets you progressively closer to a generalized answer that works well for any case. This itself is the training algorithm then — it’s a repetitive process where you’re giving the AI an example, asking what its prediction is, updating all the pieces of the AI, and then repeating.”

Eventually, the AI learns what it needs to identify and will only send back important information. In filtering this information, RAMjET will help save power and memory and keep costs down.

“CUμLUS will permit us to estimate many high precision masses for new planets detected by Roman and PRIME,” Barry said. “And it may allow us to capture or estimate the actual mass of a free floating planet for the first time — which has never been done before. So cool, and so exciting. Really, it’s a new golden age for astronomy right now, and I’m just very excited about it.”

CONTACT Richard.K.Barry@nasa.gov or 301-286-0664