A race is on to discover Planet Nine using classical astronomy and new computational techniques
Illustration: Simon C. Page
Update on 3 October 2018: The International Astronomical Union’s Minor Planet Center announced an observation by Scott Sheppard, Chad Trujillo, and the University of Hawaii’s David Tholen that adds new supporting evidence for the existence of Planet Nine. The team identified a distant object labeled 2015 TG387, nicknamed “The Goblin,” with an elongated orbit that is consistent with Planet Nine’s expected size and trajectory, first proposed by Caltech’s Konstantin Batygin and Michael Brown in 2016.
Michael E. Brown is often called “the guy who killed Pluto.” But he takes the moniker in stride. Sitting in his sunny Pasadena office at the California Institute of Technology, Brown jokes that Pluto, which was reclassified as a dwarf planet in 2006, had it coming. The year before, Brown had discovered Eris, a frosty dwarf in the outer solar system more massive than Pluto and named, fittingly, for the Greek goddess of strife.
Brown now has good reason to hope that history will remember him not for the Eris-instigated demotion of Pluto but as codiscoverer of an as yet unseen, true ninth planet—a Neptune-size world so massive that it may have tipped the entire solar system a few degrees sideways.
I meet Brown in the late afternoon, shortly after his breakfast. The 52-year-old, sporting a week-old beard and Converse sneakers, is shifting his sleep schedule to spend the coming nights remotely babysitting a giant telescope as it scans the heavens from the snowy summit of Mauna Kea, Hawaii. Calculations that Brown published last year with Konstantin Batygin, a former student of Brown’s who now occupies the faculty office next to his, suggest that Planet Nine is real. Somewhere out there, they are convinced, drifts a frozen world so distant from the sun—perhaps 5.5 light-days, or roughly 150 billion kilometers—that high noon on its surface is no brighter than a moonlit night on Earth.
Persuaded—or at least intrigued—by several converging lines of evidence, teams of astronomers around the world are now trying to answer the obvious next question: Where is Planet Nine? Although it is thought to be 8 to 10 times as massive as Earth and 2 to 4 times as wide, it seems to be maddeningly hard to spot.
Greg Laughlin, an astronomer at Yale University, says, “Our best estimate for its current position and brightness put it about 950 times farther than Earth from the sun.” As faint as the tiniest moons of Pluto, Planet Nine would be barely two pixels wide on the Hubble Space Telescope’s camera. Searchers could easily miss it among random speckles of sensor noise and the twinkling of distant and variable stars. And because the planet is so far from Earth, near the far end of a highly elliptical path that takes at least 15,000 years to complete, astronomers have to wait a day or more between successive photographs of the right patch of sky to see the planet shift its apparent position relative to the much more distant stars.
Huge telescopes on Earth have been scanning the skies for months now. Brown and Batygin have been observing on Japan’s Subaru telescope on Mauna Kea—as have veteran minor-planet hunters Chad Trujillo of Northern Arizona University and Scott Sheppard of the Carnegie Institution for Science—to exploit that observatory’s giant mirror (8.2 meters across) and its 3-metric-ton, 870-megapixel camera. Meanwhile other astronomers, both professional and amateur, are digging through archives of images in hopes of finding this needle in a hayfield.
Celestial Scouts: Michael E. Brown (left) and Konstantin Batygin of the California Institute of Technology are using the Subaru Telescope in Hawaii to search for a ninth planet. Photo: Patrick T. Fallon/The Washington Post/Getty Images
Any of them could get lucky. But the smart money is on software, either to deliver the quarry or reveal it to be an illusion. Simulations running on supercomputers and in the cloud are modeling billions of years of celestial mechanics to pin down Planet Nine’s likeliest path. Engineers at the Jet Propulsion Laboratory, in Pasadena, have been analyzing telemetry from the Cassini spacecraft for clues to the current position of the putative planet within its enormous orbit. And an ambitious pair of graduate students is preparing to deploy machine-learning software on a petaflop-scale Cray XC40 supercomputer. Their strategy aims to cleverly combine multiple images in which Planet Nine is hidden within the noise to yield one image in which it shines unmistakably.
Although many astronomers share Brown’s enthusiasm at the prospect of finding a planet bigger than Earth for the first time in 170 years, some worry about being fooled by subtle biases or simple coincidences in the data. “My instinct—completely unjustifiable—is that there’s a two-thirds chance it’s really there,” Laughlin says.
Galileo could have discovered Uranus, had he kept better records. He did spot Neptune in 1612 but mistook it for a star. It wasn’t until 1781 that the amateur stargazer William Herschel stumbled upon what he thought was a comet. He notified other astronomers, who eventually worked out a circular orbit that revealed it to be a planet, which they named Uranus.
Further observations revealed that Uranus sometimes deviated from its calculated orbit—a clue to yet another undiscovered planet out there, tugging it off course. In 1846, John Couch Adams and Urbain LeVerrier independently used those deviations to compute the mass of Neptune, the size and shape of its orbit, and its current position in the sky. Both got the numbers quite wrong—except for the crucial one of where to look, which LeVerrier predicted to within 1 degree. German astronomers pointed their telescope at that spot and found Neptune in less than an hour.
Neptune explained most of the anomalous motion of Uranus, but not all of it. In 1905, Percival Lowell, a rich and ambitious American mathematician, set up a project at his observatory in Flagstaff, Ariz., to search for a planet beyond Neptune, but he died before resident astronomer Clyde Tombaugh found Pluto—again, by happy accident. When Voyager 2 flew by Neptune half a century later, astronomers learned that they had overestimated Neptune’s mass by 0.5 percent. Correcting that error fully explained the strange movements of Uranus, which is oblivious to tiny Pluto.
This history of clumsy planetary detections hasn’t deterred Batygin and Brown. Since 2001, Brown has led in the discovery of three dozen trans-Neptunian objects (TNOs) in and beyond the Kuiper Belt, a huge ring of icy planetoids that lies outside the orbit of Neptune. Three of Brown’s finds—Eris, Haumea, and Makemake—have officially attained the rank of dwarf planet, alongside Pluto. Others, such as Sedna, Orcus, and Quaoar, are next in line for that honor. (By one definition, a planet demonstrates gravitational dominance, snaring nearby objects or flinging them away as it orbits a star. A dwarf planet, on the other hand, has a gravitational field too weak to affect nearby objects to the same degree.)
Now Brown is hunting the biggest prize of all. His quest began one day in the summer of 2014, when he walked into Batygin’s office brandishing a copy of Nature. “Have you seen how weird this is?” he asked. He was pointing to a chart in a recent paper by Trujillo and Sheppard reporting the discovery of 2012 VP113, an odd new TNO, suspected to be a dwarf ice planet.
Like Sedna, an icy dwarf that Brown and Trujillo had discovered a decade earlier, VP113 is an extreme TNO, one that mysteriously “detached” from the Kuiper Belt and now comes nowhere near Neptune. Also like Sedna, VP113 travels a wildly elongated orbit that is tipped at a curiously steep angle to the invariable plane in which all the planets (except chaotic Mercury) move.
In their chart, Trujillo and Sheppard had shown that all 12 extreme TNOs discovered so far have orbits whose long axes are roughly aligned, rather than spread out randomly as expected. “This suggests,” they wrote, “that a massive outer Solar System perturber”—perhaps an undetected planet—“may exist.” They floated several other possible explanations as well.
Unlike Brown, Trujillo, and Sheppard, who all specialize in observation, the 31-year-old Batygin has a reputation as a hotshot at celestial mechanics. Plugging numbers for the six most distant TNOs into quick calculations on the blackboard, Batygin realized that the “perturber” must be a giant planet, also on a highly elongated and inclined path. The repeated gravitational influence of that planet would keep the orbits of the TNOs from precessing around the sun into widely varying alignments.
For a year, he and Brown examined every other possible mechanism while also running weeks-long supercomputer simulations of the solar system with a ninth large planet added to the mix. Mere coincidence, they calculated, was exceedingly unlikely. “If we pick six objects at random from this distance, how often would we get clustering this tight?” Batygin recalls asking. “The answer is about 0.007 percent.”
Those odds are now steeper because the list of relevant oddball planetoids known to haunt the outer reaches of our solar system has lengthened: from 6 in early 2016 to 20, Trujillo says. About a dozen of these objects orbit within the same vertically tilted plane as Planet Nine does, but they sweep away from the sun in the opposite direction of the planet; a couple of others are aligned with the planet. Then there are a handful of planetoids circling crazily at almost right angles to everything else in the solar system; a couple of these even travel backward around the sun. “They all fit in beautifully,” Batygin says. “As time has gone on, the evidence has only increased.”
Last summer, Elizabeth Bailey, a Caltech graduate student, looked into the century-old puzzle of what caused the sun’s axis to tip by 6 degrees from perpendicular to the invariable plane. Could it be that the sun’s axis hasn’t shifted at all since the star was born inside its protoplanetary disk—that instead Planet Nine, orbiting at a high angle, has gradually dragged all the other planets upward by 6 degrees?
Bailey calculated what masses and orbits of Planet Nine could accomplish that feat. The numbers nicely overlap the ranges that Batygin and Brown prefer. Independently, a group of French and Brazilian astronomers published a similar result in December.
With the idea of a big but undiscovered planet in our cosmic backyard moving from possible to plausible, Planet Nine hunters now have to face their biggest challenge: deciding where to point their telescopes. “We don’t actually know where the planet is today in its orbit,” Batygin says. To narrow the search, his team and other astronomers are sifting clues from computer simulations that recapitulate billion-year segments of the solar system’s past or predict its far future.
Splashed across two 27-inch monitors on Brown’s desk, seven charts are cluttered with hundreds of red and white streaks. To the uninitiated, the abstract Mondrian print hanging on his office wall is easier to interpret. But to Brown, each streak represents an asteroid or planet, and each chart captures one of the seven crucial parameters that define Planet Nine’s mass, orbital shape, and current position.
“I’m running 12 integrations of the real objects in the outer solar system and how they would behave over the next billion years with Planet Nine, given different values for the seven parameters,” he says. The combinations of values are guesses, guided mainly by his intuition. “If one ever happens to work”—meaning that the virtual solar system keeps humming along for the next billion years without the new planet wreaking havoc—“I can jump up and down,” he smiles.
It takes his workstation just two days to model the celestial interactions of 200 tracer objects over a billion years, thanks to advances in technology. Moore’s Law has obviously helped. But the early 1990s also brought a big breakthrough in an algorithm, known as symplectic integration, that reduced computational times by an order of magnitude. Then came multicore and massively parallel computing systems, which are ideally suited for what Brown calls “embarrassingly parallel” problems like tracing how the orbits of many objects evolve over a wide range of starting conditions.
Unusual Orbits: Nearly all minor planets detected so far that travel well outside the orbit of Neptune follow highly elliptical orbits that cluster to one side of the sun and are tilted at an angle. Two such objects are aligned opposite to the others. A large but very distant major planet could explain this arrangement, which is highly unlikely to have occurred by chance. Data sources: Michele Bannister, Konstantin Batygin, Michael E. Brown, Sarah Millholland, Scott Sheppard. Illustration: IEEE Spectrum
Symplectic integration is so complicated that even Brown admits he doesn’t fully understand the math. “But the key idea,” he explains, “is to take advantage of the fact that you already know that any object circling the sun mostly follows a simple orbit,” as described by Kepler’s laws of planetary motion—plus some minor perturbations. Because symplectic integrators don’t waste time rediscovering Kepler’s laws over and over, they run orbital simulations hundreds of times as fast as older methods do. One of the most popular symplectic modeling platforms is called Mercury (not to be confused with the planet), and it has become the tool of choice for several of the planet-hunting teams, including Brown and Batygin.
At Yale, Laughlin and his graduate student Sarah Millholland enhanced Mercury last autumn with a Markov-chain Monte Carlo algorithm to home in more quickly on promising orbits. Using the 1,000-core supercomputing cluster at Yale, they were able in a month to simulate a total of 1019 years of orbital mechanics, tracking not only 11 extreme trans-Neptunian objects but also uncertainties in their observations.
“We got orbital parameters that agree well with Brown’s and Batygin’s values,” Laughlin says. “But our simulation gives a more precise place in the sky to look for it.” Their paper, published in February, as well as more recent supercomputer simulations presented in April by Trujillo, puts Planet Nine somewhere in the constellation Cetus (the whale) or Eridanus (the river), at about 28 times the current distance to Pluto. “It’s still a vast search area,” Trujillo says.
Meanwhile, at the Southwest Research Institute in Colorado, David Nesvorný has been modifying his far more detailed models of the formation of the Kuiper Belt from the early days of the solar system to see what happens when he plugs in a ninth planet. The simulation, built on a symplectic code known as SyMBA, starts with a million virtual TNOs as they might have existed in the nascent solar system. The system computes 4.5 billion years of evolution and then compares the outcome to what astronomers see today. Each run takes more than five weeks to complete on 500 CPU cores of NASA’s Pleiades supercomputer.
Initial results seemed encouraging: Extreme TNO orbits lined up just as others had found. “It showed that Planet Nine could be responsible for that,” Nesvorný says. But things didn’t work out as well when he then focused on how Planet Nine would affect a certain class of comets.
“My model nicely reproduces all orbital parameters for these comets—until I add Planet Nine,” he says. In the model, the new planet tilts the so-called scattered disk, where Jupiter-family comets originate, causing the virtual comets to enter the solar system more steeply than the real ones do.
More caveats to Planet Nine’s theorized existence come from the Cassini probe, which has orbited Saturn since 2004. From minute changes in the spacecraft’s speed and other telemetry, the Cassini team calculates the distance from Earth to Saturn to within 3 meters. Those range measurements could reveal even small deviations in Saturn’s orbit due to the pull from Planet Nine, but only if it is close or large enough. William Folkner, a principal engineer at JPL, says he and coworkers examined the data and saw no perceptible distortion of Saturn’s orbit. So, if Planet Nine exists and is 10 times Earth’s mass, it must be within 25 degrees of the farthest point in its hypothetical orbit, he says. A smaller Planet Nine—Brown now favors a mass eight times that of Earth—would have 40 degrees of wiggle room to hide in.
The results, positive and negative, aid the handful of observers now hunting for Planet Nine on telescopes. In addition to the groups working on Subaru, Sheppard and Trujillo are leading searches in the high desert of Chile, in case the planet is easier to see from the Southern Hemisphere. There, both the 570-megapixel Dark Energy Camera (DECam) on the 4-meter Blanco telescope and the 6.5-meter Magellan telescopes are contributing to the hunt.
Dark Light: A new search technique developed by a team at the University of California, Berkeley, processes successive telescope images of the same patch of night sky. In Step 1, software subtracts out known objects and noise, leaving only unknown objects too faint to see in any individual image. In Step 2, software combines the images, shifting each so the planet’s theorized position is stacked directly on top of its position in previous images. This approach brightens any light reflected from the planet, making it easier to spot. Illustration: Mark Montgomery
“I actually think we will not discover Planet Nine by scanning the sky,” Brown says. “We could, but I think somebody will find it first in archival data,” from surveys that have already photographed huge swathes of the heavens. After Uranus and Neptune were discovered, astronomers noticed that earlier stargazers had already recorded the two worlds many times but not recognized them for what they were. Now at least four efforts are under way to find a new planet in old photos.
David Gerdes of the University of Michigan has been combing through the archive of DECam’s survey observations to find images of the planet. By coincidence, Brown notes, “our predicted path for the planet goes right through the Dark Energy Survey’s field of view.”
An army of amateurs has jumped into the game as well. In February, Marc Kuchner of NASA’s Goddard Space Flight Center helped launch a crowdsourced effort to compare successive infrared images made by the Wide-field Infrared Survey Explorer space telescope of the same spot in the sky. By July, the project had recruited 40,000 volunteers, who had thoroughly reviewed over 125,000 chunks of space. A southern-sky version, launched in March with data from the Australian SkyMapper telescope, blew through 106,000 search regions in just three days. Laudable as these citizen-science projects are, their odds of success are low because the small telescopes involved typically cannot gather enough light to see something as dim and distant as Planet Nine is thought to be.
Michael Medford and Danny Goldstein, graduate students at the University of California, Berkeley, think they have a solution to that problem. Drawing on hundreds of thousands of images covering the search area for Planet Nine—all shot from 2009 to 2016 using a 1.2-meter telescope in the mountains north of San Diego—their system will combine multiple images in an ingenious way that should brighten the faint flickers of light from Planet Nine enough to distinguish them from background noise.
“Because the planet is moving with respect to the background stars, you can’t just add overlapping images together,” Medford points out. Instead, their software selects each of the many distinct plausible orbits for Planet Nine, projects the planet’s movement onto the relevant patch of sky, and then offsets successive images to superimpose—and brighten—any pixels corresponding to the planet. A pipeline of software written with Peter Nugent, their faculty advisor, performs the overlapping and subtracts known objects such as stars.
The computational task is enormous because the planet’s orbit is still so uncertain. To do a 98 percent complete search, Medford estimates, they will need to perform 10 billion image comparisons. Fortunately, Nugent has time allocated on the Cori supercomputer, a new Cray XC40 system that recently ranked as the fifth most powerful in the world.
The Prize: An artist’s rendering of Planet Nine shows the planet’s far side, as if the viewer were looking back toward the sun. Illustration: Robert Hurt/California Institute of Technology
False positives are unavoidable. “Even if we get only one false hit for every million searches, we’ll still get 10,000 fake planets,” Goldstein says. “So we will be passing all detections through a machine-learning system trained to catch and reject artifacts: satellite trails, hot pixels, cosmic rays, and other spurious sources.”
With the data already in hand, the two expect the system, running in parallel on hundreds of Cori’s CPU nodes and 278 hyperthreads per node, to finish the work in just a few days when they flip the switch in August. “We’ll be sitting on the edge of our seats,” Goldstein says. “And whether we find P9 or not, this method can be used to detect other TNOs.”
“I’m rooting for them,” Brown says. Though his own major finds have all been made by classic observation, “I’ve been doing that since 1998,” he says. “It’s boring—I’m tired of it.”
He harks back to the heady days of technology when his father, a NASA engineer, worked on the Apollo moon-landing missions. “Discovering a major planet through clever computational methods would be better,” he argues, because it would represent an impressive dual discovery: a new giant added to the celestial pantheon, plus a powerful new computational technique for uncovering mysterious objects hidden right in our little corner of the cosmos.
This article appears in the August 2017 print issue as “Where Is Planet Nine?”