As beautiful as sunsets are on Earth, imagine a double sunset with stars of different colors, casting moving shadows of orange and red. For years the two of us wondered if paired, or “binary,” stars could support planets. Could worlds like the fictional Tatooine from Star Wars, where the sky is lit with the glow of two different suns, really exist?

Astronomers had reason to think such systems might exist, yet some theorists disagreed. The environment around a pair of stars, they argued, would be too chaotic for planets to form. Unlike a body circling a single star, a planet orbiting a pair of stars would have to contend with two gravitational fields. And because the stars themselves orbit each other, the strength of the gravitational forces would constantly change. Even if a planet could form in such a dynamic environment, its long-term stability would not be assured—the planet could wind up being ejected into deep space or crashing into one of the stars. Observations of binary star systems had shown some indirect evidence for these “circumbinary” planets, but direct evidence remained elusive.

Over two decades of effort by William Borucki and his collaborators to get an exoplanet-hunting spacecraft launched finally came to fruition in March 2009. nasa's Kepler Mission has since proved to be spectacularly successful, quickly revealing hundreds, then thousands, of planet candidates via the transit method, which searches for the mini eclipse that occurs when a planet orbits in front of the star, blocking some of its light. But after two years, no circumbinary planets had been detected. The frustrating lack of evidence began to take its toll. In a weekly Kepler telephone conference in the spring of 2011, one of us offered an attempt at black humor: “Maybe we should write a paper on why they don't exist.” Silence followed.

Our fears were misplaced. Within six months of that conversation we had a press conference to announce the discovery of the first transiting circumbinary planet. This planet was called Kepler-16b. Within months the Kepler Eclipsing Binary Working Group discovered two more circumbinary planets (Kepler-34b and Kepler-35b), showing that while exotic, such systems are not rare. A new class of planet system had been established. The current tally of Kepler circumbinary planets is seven, and that number could double in a short time. In fact, calculations suggest that tens of millions likely exist in the Milky Way.

Search Strategies
The quest for circumbinary planets began in the 1980s, even before astronomers found the first evidence of any “exoplanets” outside our solar system [see “Searching for Shadows of Other Earths,” by Laurance R. Doyle, Hans-Jörg Deeg and Timothy M. Brown; Scientific American, September 2000, and references therein]. Although transits can be much more complicated in a binary system, hopes for discovering such a system were fueled by a simple expectation: if a planet did orbit an eclipsing binary star system, we would expect it to travel in the same orbital plane as the stars themselves. In other words, if from our perspective on Earth the stars eclipsed each other, then the planet would be much more likely to eclipse one or both stars. This assumed that the planet and the stars had co-planar orbits, a reasonable hypothesis—and one that could be tested.

Eclipsing binary stars are in many ways the foundation on which stellar astrophysics is built. Their special orientation along our line of sight means that the stars pass in front of each other once per orbit, blocking some of the light. By precisely modeling how the light dims during the eclipses, we can learn the sizes and shapes of the stars and the geometry of their orbits. Coupled with other measurements, we can measure the stars' radii and masses. Eclipsing binary stars thus provide a fundamental calibration of stellar masses and radii, which in turn are used to estimate the stellar properties for noneclipsing and single stars.

If the two stars in a binary system are very far apart, with an orbital period of, say, hundreds of years, the stars hardly affect each other, and they act almost as if in isolation. Planets may orbit either one of the stars and in general will not be much influenced by the presence of the other star. These are known as circumstellar, or S-type, planets, and dozens of such planets have been discovered in the past decade.

Things get more interesting when stars are so close together that they take only weeks, or even days, to orbit each other. For a planet in such a binary to have a stable orbit, it would have to orbit both stars, not just one. Numerical calculations show that the planet's orbital separation from the stars has to be larger than a minimum critical distance; too close and the rotating binary system would destabilize the planet's orbit, either swallowing it up or ejecting it out into the galaxy. The minimum stable separation is roughly two to three times the size of the stars' separation. These kinds of planets are known as circumbinary, or P-type, planets. While planets around single stars and around individual stars in widely separated binaries are common, we wondered if nature could make planetary systems in the circumbinary configuration, where the planet orbits both stars.

In a simple one-star, one-planet system, transits will occur with a metronomic periodicity that greatly assists in their detection. Add another star, though, and the three-body system will start to display all manner of complicated effects. The complexity arises because the stars are quickly moving—in contrast to the single-star system where the star is effectively stationary. In fact, because the two stars are closer to each other than they are to the planet, they must orbit each other faster than the planet will orbit them—a manifestation of Johannes Kepler's famous laws of planetary motion. Thus, the planet will be transiting a quickly moving target, and sometimes it will cross the star early and sometimes late. While precisely predictable (if the masses and orbits are known), the transits will not be periodic. In addition, the duration of the transit will change depending on the relative motion of the planet to the star being crossed—if they are moving in the same direction, the transit will be longer in duration, but when the star is on the other half of its orbit and moving the other direction, the transit will be shorter. These variations make detecting circumbinary planets difficult, but they also offer an important benefit: once the binary star's orbit is deciphered, the pattern of changing transit times and durations can be used to unequivocally confirm the presence of an orbiting circumbinary body. No other astronomical phenomenon exhibits such a pattern. This is a unique characteristic of a circumbinary object—a smoking gun signature.

The First Detection
Until technical problems sidelined it earlier this year, Kepler had kept its eye trained on a single patch of sky, looking for the characteristic dimming caused by planets crossing in front of host stars. In its quest for planets, Kepler also discovered more than 2,000 new eclipsing binary star systems. Several exotic systems were discovered, including the first known eclipsing triple-star systems.

In 2011 one of us (Doyle), along with associate Robert Slawson, working with him at the SETI Institute in Mountain View, Calif., noticed extra eclipse events in the binary stars known as KIC 12644769. The two stars eclipsed each other every 41 days, but there were three other unexplained eclipse events. The first two occurred 230 days apart. The next occurred 221 days later—nine days earlier than expected. This was just the kind of signature one would get from a circumbinary planet.

These transits thus provided evidence of a third body orbiting the binary. But it could have been just a dim, small star grazing part of the large star—and as Kepler was showing us, such triple-star eclipsing systems are not exceptionally rare. The slight dimming indicated that the object could have a small radius, but starlike objects such as brown dwarf stars are also small, so we could not say for sure if the object was a planet. We had to measure its mass.

In a three-body system, an unseen companion to a binary can make its presence known in two main ways. Imagine two stars eclipsing each other, with a relatively large planet circling the pair farther away. The binary stars orbit each other, but in addition, the center of mass of the pair is orbiting the center of mass of the three-body system. As a consequence, sometimes the binary stars will be displaced a little bit closer to Earth; at other times, they will be farther away. When they are farther away, light from the stars will take longer to reach us and the eclipses will occur slightly late. When the stars are closer to us, the eclipses will be early. The larger the mass of the third body, the larger the change. Thus, this cyclic light-travel time effect allows one to infer the presence and estimate the mass of any unseen object. Also, the farther away the third body is from the binary, the greater the effect because the added distance will act as a lever, but the farther away, the longer the cycle time. In the case of our candidate circumbinary planet, there was no detectable cyclic change in the eclipse timing on the order of 230 days, implying that the hidden body had a low mass. But how low?

The other way for a third body to affect the binary is through direct gravitational interaction, called the dynamical effect. This method dominates the light-travel time effect for closer objects. The unseen companion slightly alters the orbits of the binary stars, and these changes can be picked up through variations in the occurrence times of the eclipses. Because the smaller star comes closer to the third body, its orbit will be perturbed more. Unlike the light-travel time effect, the dynamical effect alters the times of the eclipses in complex ways.

One of our colleagues on the Kepler Science Team, Daniel C. Fabrycky, now at the University of Chicago, noted that a stellar-mass object would strongly affect the eclipse times, whereas a planet would produce a much more subtle—but potentially measurable—signature. And for this system, the dynamical effects should be very much stronger than the light-travel time effect. We looked for and subsequently found the changes in eclipse timing, revealing that the tug on the stars was not anywhere near what a stellar-mass companion would produce.

The grand finale of the investigation was provided when Joshua A. Carter of the Harvard-Smithsonian Center for Astrophysics was able to create a sophisticated computer model of the system. It matched the complete data set perfectly for a planet with a mass similar to Saturn's. The excellent match between the observations and the modeling proved the existence of the planet and provided exquisitely precise values for the radii, masses and orbital characteristics of the system.

This planet was designated Kepler-16b and was the first transiting circumbinary planet discovered. The combination of the transits and the clear dynamical effects made this detection unambiguous. Because the binary stars would appear as sun-size disks as seen from this planet, Kepler-16b soon acquired the nickname “Tatooine” from the fictional planet in Star Wars and its iconic image of a double sunset. Science fiction had become science fact.

A New Class of Planet
Kepler-16b appeared, at first, to be a very strange planet. Its orbit is uncomfortably close to its host stars, being only 9 percent farther out than the minimum critical distance needed for orbital stability. And because this was the only transiting circumbinary planet at the time, we asked ourselves: Is Kepler-16b just a fluke?

Fortunately, the answer came quickly. Working with Jerome A. Orosz of San Diego State University, we had already been searching for circumbinary planets that do not transit their stars. These should be far more common than transiting cases, since the special alignment of the planet's orbit to create a transit is not required. As mentioned, small variations in eclipse timings should reveal such planets. We had been pursuing this line of research for a few months and had identified a few candidate systems. Then, on a Tuesday afternoon in August 2012, one of us (Welsh) noticed transits in one of the binary star systems. Within hours Fabrycky had created a computer model that reproduced the variable transit times and durations, confirming the transiting object as a planet. We had discovered Kepler-34b. Working feverishly, the very next day Orosz found transits in another eclipsing binary star system, and it, too, harbored a planet—Kepler-35b.

Over the next few months Orosz would go on to discover Kepler-38b, showing that smaller, Neptune-mass circumbinary planets also exist, and then the Kepler-47 planetary system with at least two planets, showing that binary stars can harbor multiple planets. The most recent circumbinary planet discovered, Kepler-64b (also known as PH1) was simultaneously and independently discovered by Johns Hopkins University graduate student Veselin Kostov and by amateur astronomers working as part of the Planet Hunters project. It is part of a quadruple-star system, further extending the diversity of places where planets can form.

The seven circumbinary planets found so far tell us that these objects are not extremely rare but rather that we have uncovered a whole new class of planetary system. For every transiting planetary system detected, geometry tells us that there are roughly five to 10 planets that we do not see because they do not have the correct orientation to pass in front of the binary stars as seen from our vantage point. Given that seven planets were found out of roughly 1,000 eclipsing binaries searched, we can conservatively estimate that the galaxy is home to tens of millions of such circumbinary planetary systems.

All the Kepler transiting circumbinary planets to date are gas-giant planets, worlds without the rocky crust that would allow an astronaut to stand on its surface and marvel at the double sunsets. The search continues for smaller rocky planets, although Earth-size circumbinary planets are going to be extremely difficult to detect.

But even with such a small sample of planets, a number of interesting questions arise. For instance, half of all the Kepler eclipsing binaries have an orbital period of less than 2.7 days, so we expected that half of the binaries with planets would also have periods less than 2.7 days. But none of them do; the shortest orbital period is 7.4 days. Why? We speculate that it might be related to the process that brought the stars so close together in the first place.

In addition, the planets tend to orbit their stars very closely. If they were in much closer, the planets' orbits would be unstable. What, then, causes them to live so dangerously? Understanding why the circumbinary planets orbit so close to their critical instability radius will help us improve theories about how planets form and how their orbits evolve over time.

Even though we do not know why these planets seem to prefer such precarious orbits, we can nonetheless infer something deep: the discovery that planets can live so near a chaotic environment is telling us that planet formation is vigorous and robust.

A Dynamic Habitable Zone
The tendency of the Kepler circumbinary planets to lie near the critical stability radius has an interesting consequence. For the Kepler sample of stars, the critical radius is generally close to the habitable zone—the region around a star (or in this case, around two stars) where the energy from that star makes the planet's temperature just right for water to persist in the liquid state. Too close to the star, and the planet's water boils; too far away, and the water freezes. And water is a prerequisite for life as we know it.

For a single star, the habitable zone is a spherical-shell region around that star. In a binary system, each star has its own habitable zone, which merge into a distorted spheroid if the stars are close enough together, as is the case for the Kepler circumbinary planets. As the stars orbit each other, the combined habitable zone also revolves with the stars. Because the stars orbit faster than the planet does, the habitable zone swings around more quickly than the planet orbits.

Unlike Earth, which is in a near-circular orbit around the sun, the distance a circumbinary planet has to each of its host stars can change radically over the course of the planet's orbital year. Thus, planetary seasons could wax and wane in only a few weeks as the stars whirl about each other. These climate changes could be large and only quasi regular—“It would be a wild ride,” Orosz notes.

Two of the seven known transiting circumbinary planets are in their system's habitable zone, a remarkably high percentage. Although being in the habitable zone does not guarantee conditions suitable for life—Earth's moon is in the sun's habitable zone and yet is as desolate as can be imagined because its small mass is too feeble to retain an atmosphere, for example—the high fraction of circumbinary planets in their habitable zones does cause one to pause and wonder. With its severe and rapidly changing seasons, what would life, and indeed a civilization, be like on a circumbinary world?