As recently as five years ago, the search for planetary systems beyond our own was the subject of a few painstaking surveys by a small but dedicated band of planet-hunters. Their goal was to find extrasolar counterparts of our own giant Jupiter, which circles the Sun once every 11 years. To do this they relied on the fact that the star around which any planet orbits also moves: for instance, the Sun circles its common centre of gravity with Jupiter at the leisurely pace of 12 metres per second. The aim of these early programmes therefore, was to develop techniques that could detect the periodic changes in the Doppler shift of the light from a star as it wobbled back and forth in response to the gravitational tug of an unseen Jupiter-like companion (figure 1).

Several teams achieved the required precision by the early 1990s. These efforts included a regular monitoring programme of 120 nearby stars that was started at the Lick Observatory in California in the late 1980s by Geoff Marcy and Paul Butler of San Francisco State University. However, the computational overheads needed to analyse the results were so great that much of their data remained unstudied in an archive. This strategy made good sense: by the time any candidate Jupiters had completed enough of their orbits to be clearly identifiable, processor speeds would have increased to the point where the data analysis could be carried out far more quickly. Other groups at the University of Victoria in Canada and the University of Texas at Austin adopted similar philosophies.

When Michel Mayor and Didier Queloz of the Geneva Observatory announced the first discovery of stellar reflex motion due to a planetary body in 1995, its form was so unexpected that it demanded independent confirmation. Their data showed the solar-like star 51 Pegasi to be wobbling back and forth at 56 m s^-1, completing one orbit every 4.2 days. The only plausible explanation for this wobble was the presence of an unseen body with at least half the mass of Jupiter in an orbit with a radius of 0.05 astronomical units (AU). An astronomical unit is defined as the mean distance between the Earth and the Sun, which is around 1.5 × 10⁸ km.

Fortunately Marcy and Butler had been monitoring the same star for several years, and confirmed almost immediately that the wobble was present in their data as well. At the same time, they found similar signatures of close-orbiting giant planets for three other stars: tau Bootis, 55 Cancri and upsilon Andromedae. A spate of similar discoveries followed. New monitoring programmes were established, and the total number of stars that are currently under observation stands at about 1000. Among these, the tally of nearby stars known to possess at least one planet has risen to over 50. Several of these stars show extra, slower wobbles superimposed on the main signal, which betray the presence of one or more additional planets in larger orbits. Most notable among these is upsilon Andromedae with a family of three giant planets.

Last year saw the record for the lowest mass extrasolar planet being broken several times. Mayor currently holds the record for detecting a planet weighing just 0.16 times the mass of Jupiter that is orbiting around HD 83443, a bright star in the constellation Vela some 141 light-years from Earth.

Meanwhile in August 2000, a team led by Bill Cochran of the University of Texas at Austin announced that it had detected the closest extrasolar planet to Earth. At a distance of just 10.5 light-years away, the star epsilon Eridani shows evidence of a seven-year wobble, although this result still needs to be confirmed.

Reflex orbit

The period and size of a star’s wobble encode important information about the planet’s mass. Usually it is only possible to determine a lower limit on the mass, because for most systems astronomers cannot measure the tilt of the orbit relative to the line of sight. Assuming that the orbit is edge-on to the line of sight gives the smallest possible value of the planet’s mass. This is not as serious a problem as it might appear. If the orbital axes of the planetary systems are oriented randomly in space, there is a natural statistical tendency for us to see many more orbits edge-on rather than face-on. (Try throwing a handful of coins in the air to convince yourself of this.) All the planets discovered so far are giants with masses ranging from slightly less than the mass of Saturn, which is 95 times heavier than Earth, to a dozen times the mass of Jupiter. Jupiter is the most massive planet in our solar system and is 320 times heavier than the Earth.

A planet in a circular orbit around its star produces a symmetric wobble that varies sinusoidally with time. The dozen or so known planets with periods of less than a week – the so-called hot Jupiters, deemed hot because they are substantially closer to their stars than our Jupiter is to the Sun – have orbits that are nearly circular, as expected (figure 2). The reason is that a close-orbiting planet in a highly elliptical orbit produces tides on the star, which move the star’s centre of gravity in such a way that the orbit gradually evolves into its lowest energy state, a circular orbit. The greater the distance between the planet and the star, the smaller the tidal effects and the longer it takes for the orbit to become circular. For planets that have orbital distances larger than a small fraction of an astronomical unit, the orbit will remain elliptical for longer than the star’s lifetime.

A planet in an elliptical orbit speeds up when it is close to the star and slows down when it is further away from it, giving the star a characteristic lopsided wobble. Astronomers have used this idea to deduce that the orbits of the longer-period planets are much more eccentric (i.e. more elliptical), in stark contrast to the near-circular orbits of the planets in our solar system. Indeed, nobody has yet found a well behaved Jupiter-like planet in a circular orbit with a radius of several astronomical units.

The high masses and short orbital periods of the planets discovered so far makes them easier to detect than conventional Jupiter-like planets with long periods. This is because the speed, v, of the parent star in its reflex orbit is given by the relation v = 12(M_P/M_J)(5.2M_Sun/aM_star)^1/2 m s^-1 where M_P, M_J, M_Sun, M_star are the masses of the extrasolar planet, Jupiter, the Sun and the star, respectively, a is the orbital distance of the planet and 5.2 AU is Jupiter’s distance from the Sun. This favours the discovery of massive planets in close short-period orbits, particularly as most of the monitoring programmes have only been running for a few years. However, it does not explain why so many of the orbits are so much more eccentric than those found in our solar system. To understand these differences, we need to look at the conditions under which planetary systems form.

How do Jupiters form?

Theories of giant-planet formation fall into two main categories. The “top-down” approach has planets forming from large-scale perturbations in the flattened, gaseous disc that surrounds a new-born star for the first few million years of its life. Meanwhile, the “bottom-up” approach requires dust grains with ice mantles to clump together to form bodies a few times the mass of the Earth. Once this critical mass is attained, the planet’s gravitational pull becomes strong enough for it to accrete large amounts of gas from the disc and grow rapidly into a gas giant.

As the planet grows, however, it causes slow-moving material outside its own orbit to speed up and faster-moving material with smaller orbits to slow down. This tidal effect sweeps the planet’s own orbit relatively free of material (figure 3). Computer simulations show that accretion can only occur along a pair of spiral shocks extending inward and outward from the planet. However, several things can go wrong. For example, the growth process itself may be self-limiting due to a lack of material in the region swept clear by the planet. And in many models, the angular momentum that is inevitably exchanged between the planet and the surrounding disc material causes the planet’s orbit to decay, spiralling in to be swallowed by its sun before it can attain a high enough mass.

Other models have shown that top-down planet formation occurs in local clumps in the wake of a growing giant planet, producing a system of several giant planets in dynamically unstable orbits. In this case, interactions between the planets may lead to some bodies being ejected from the system altogether, while others are left behind in eccentric orbits.

Tantalizing new evidence emerged recently that top-down formation of planetary-mass bodies may even occur in interstellar space. Maria Zapatero Osorio and co-workers at the Instituto de Astrofisica de Canarias in Spain and the California Institute of Technology in the US recently announced the discovery of several faint objects in the star-forming region around the massive star sigma Orionis (see further reading). The spectra of these “freely floating” objects, which appear unbound to the star, are a good match with theoretical models of gaseous bodies that weigh a few Jupiter masses and are between 1 and 5 million years old. At these very young ages, the objects still shine brightly as they contract by radiating gravitational energy. However, it is not yet clear from the observations whether the free-floaters have formed in isolation or have been ejected from nearby protoplanetary systems.

These new ideas – inspired largely by the properties of the exoplanetary systems discovered so far – paint a much more violent picture of the planet-formation process than is needed to explain our solar system. While the same problems of spiral-in and self-limiting growth were encountered as long ago as the mid-1980s, much of the fine-tuning of the models was carried out under the assumption that the end-product should look like our system, with well behaved giant planets in circular orbits at more or less the distances where they formed.

The orbits of the giant exoplanets suggest that many of the things that can go wrong in building a tidy system like our own, do go wrong elsewhere. Neither top-down nor bottom-up scenarios can produce Jupiter-like planets with four-day orbits, such as the planets around 51 Pegasi, tau Bootis, upsilon Andromedae and the star HD 187123 in the constellation Cygnus. If the cores of these exoplanets formed from rock-ice planetesimals, they must have done so several astronomical units from their stars, accreted their atmospheres, spiralled in and had their migration halted near 0.05 AU by some as yet unknown mechanism. The giants in eccentric orbits between a few tenths and a few AU must have undergone a similar formation and migration history, combined with violent dynamical interactions with other newly formed giants.

Hints from other worlds

The exoplanets themselves may give some clues about their formation history. The radius of a Jupiter-like planet depends weakly on its total mass and internal composition, and also on its age. Even today, Jupiter is shrinking, radiating 65% more energy than it absorbs from the Sun. Recent calculations by Adam Burrows at the University of Arizona in the US and Tristan Guillot at the Observatoire de la Côte d’Azur in France show that the shrinkage rate for a giant planet with a given mass can be slowed considerably if it is prevented from radiating efficiently. The extreme irradiation experienced by the hot Jupiters due to their close proximity to their stars has precisely this effect, as the planet can only radiate efficiently from its dark side.

Much of the shrinkage occurs early in a planet’s history when it is hot and has a large radiating surface area. The present-day radii of the hot Jupiters are therefore quite sensitive to the ages at which the planets reached their current orbits. For example, a planet that took a long time to form and spiral in will have plenty of opportunity to radiate and shrink. On the other hand, a planet that formed quickly and spiralled in rapidly would have arrived at its new orbit with a large radius, and would subsequently find it harder to cool.

Since the ages of the parent stars can generally be determined to within a billion years or so (from their luminosities, temperatures, heavy-element abundances and axial-spin rates), observational determinations of the radii of exoplanets with known masses can provide important insights into the interior compositions and histories of these planets.

Late last year, David Charbonneau, Tim Brown and others at Harvard University and the High-Altitude Observatory (HAO) in Boulder, Colorado discovered a planet in a 3.5-day orbit about the star HD 209458, which lies roughly 200 light-years from the Sun in the constellation of Pegasus. As the plane of the planetary orbit lies along our line of sight, the new planet passes across the face of its parent star once every orbit. The Harvard-HAO team, and others, quickly found that nearly 1.5% of the star’s light was blocked each time the planet crossed in front of it. This allowed the researchers to measure for the first time the relative sizes of the planet and the star directly, and so obtain the first confirmation that the hot Jupiters are indeed gas giants. With a radius 1.35 times that of Jupiter, HD 209458’s planet is over-sized for its age and mass, in agreement with recent models published by Burrows’ team.

Silicate clouds and the sodium greenhouse

The progress made so far in determining the properties of these other worlds is remarkable in that it has all been achieved without actually seeing the light directly from the planets themselves. Direct detection will help astronomers to pin down the properties of these planets’ atmospheres and to determine when the hot Jupiters arrived in their current, bizarre orbits.

A planet must balance the amount of radiation absorbed on its sunward hemisphere against the amount radiated back into space. As this balance determines the planetary radius, astronomers are keen to find out what fraction of the incident stellar radiation is absorbed so as to understand the sizes of these planets.

A star like our Sun or 51 Pegasi pumps out most of its power at optical and near-infrared wavelengths, so the reflectivity of a planet’s atmosphere at these wavelengths plays an important role in the overall energy balance. Indeed, the large radius of the planet orbiting around HD 209458 is best explained if the optical reflectivity is low, allowing the planet to absorb a large fraction of the radiation received from its star. The other side of this particular coin is that the planet’s atmosphere should then reflect relatively little starlight back into space.

Theoretical models offer several good reasons why this might be the case. A deep, cloudless atmosphere of molecular hydrogen should reflect a substantial proportion of incident radiation back into space, particularly at short wavelengths (i.e. blue light) where Rayleigh scattering is efficient. At longer wavelengths, however, the incident radiation can penetrate deeper into the atmosphere before being scattered. If the incoming photons are absorbed by other molecular or atomic species along the way, they may never re-emerge from the atmosphere. Instead, their energy is converted into heat, adding to the planet’s overall thermal energy.

As on Earth, water and methane are among the molecules that absorb strongly at red and near-infrared wavelengths, thereby trapping incoming stellar radiation. However, any resemblance to the Earth’s atmospheric chemistry ends there. As a rule of thumb, the temperature at the top of a planet’s atmosphere varies roughly as d^–1/2, where d is the distance from the star. A hot Jupiter orbiting at 1/20th of the Earth’s orbital distance around a Sun-like star should therefore be about four or five times hotter than the Earth’s cosy 300 K. Temperatures ranging from 1300 to 1500 K are hot enough for substantial amounts of the alkali metals to be present in the gaseous state.

In a high-pressure atmosphere, alkali-metal atoms constantly collide with the hydrogen molecules that dominate the gas. During these collisions, the atomic energy levels are perturbed, allowing the alkali-metal atoms to absorb light at wavelengths very different from the usual narrow ranges available to isolated atoms. As a result, the absorption signature of the familiar yellow sodium “D-lines” can become so broad that it absorbs photons over almost the entire optical spectrum. This is expected to give the planets a highly effective “stealth coating”, allowing very little starlight to be reflected back into space (figure 4).

The major uncertainty in this cosy picture of exoplanetary weather is the role of clouds. The pressure in any planetary atmosphere decreases with height, and so does the temperature. Clouds form if the temperature drops sharply enough with height to cross the condensation curve for any common molecule present in the atmosphere (figure 5). On Earth, the dominant cloud-forming molecule is water, and cloud systems appear brilliant white when viewed from above. Meanwhile, the temperatures in the upper atmosphere of Jupiter and Saturn are in the range where ammonia forms clouds. These cloud decks reflect large numbers of incoming solar photons back into space, even at long wavelengths that might otherwise be absorbed by methane. As a result, Jupiter and Saturn appear white in colour, while Uranus and Neptune have deep, cloudless atmospheres in which methane absorbs most of the red light, giving them a bluish appearance.

However, the atmospheric temperatures are so high in the giant exoplanets that the dominant cloud-forming species are expected to be silicates of magnesium, such as enstatite, and perhaps even iron. Current models indicate that the reflectivity of these planets’ atmospheres can increase drastically at visible wavelengths if the silicate clouds form high enough in the atmosphere so that they can scatter photons back into space before they are absorbed by sodium (figure 4). A team led by David Sudarsky of the University of Arizona in the US predicted recently that high-altitude cloud-forming conditions could be particularly favourable in the lowest mass and most strongly irradiated hot Jupiters.

tau Boo: now you see it, now you don’t

Like all forms of weather prediction, exoplanetary meteorology is a complex business in which unforeseen effects due to trace species can have a disproportionately large influence on the system as a whole. Given the importance of the optical reflectivity for determining the overall energy balance of planets, these models need to be guided by direct observation.

Two groups began searching for light reflected from exoplanets about three years ago. David Charbonneau, Bob Noyes and others at Harvard used the 10 m Keck telescope in Hawaii. Meanwhile, our team at the University of St Andrews – Keith Home, Dave James and myself – together with Alan Penny of the Rutherford Apple ton Laboratory used the 4.2 m William Herschel Telescope on La Palma to search for the faint, Doppler-shifted signature of starlight reflected from the massive planet orbiting tau Bootis.

Circling its star once every 3.3 days, tau Boo’s planet is the heaviest of the hot Jupiters, with a mass of at least 3.9 times – and more probably 7 or 8 times – that of Jupiter. Both teams selected tau Boo because the planet’s short orbital period and predicted large radius ensure that it intercepts more light from its star than any other of the hot Jupiters. We expected that the light reflected from this planet should therefore be brighter relative to its star than any of the other exoplanets known at the time.

Each team developed sophisticated data-analysis methods to disentangle the faint signature of the reflected starlight from that of the parent star (figure 6). The spectrum of the reflected light should contain copies of the thousands of narrow absorption lines produced by heavy elements in the star’s atmosphere. As the planet orbits the star, any light reflected from the planet towards the observer is Doppler-shifted by the planet’s orbital motion, so we expect to see a faint echo of the star’s absorption lines moving periodically back and forth with the planet’s orbital speed of 150 km s^–1. At the same time, the strength of the reflected signature rises and falls with the changing illumination of the planet by the star. The planet is brightest when it is on the far side of the star with its illuminated hemisphere facing towards us, but invisible when it is between us and the star.

Both groups independently developed methods for subtracting out a model of the direct starlight. We then searched deep in the resulting noise for statistical evidence that the known pattern of lines was wobbling back and forth while changing in brightness at a tempo dictated by the stellar-wobble measurements made by Marcy, Butler and co-workers.

If the orbit is tilted significantly to the line of sight, the component of the planet’s velocity toward the observer is lower, and the brightness variations are less pronounced. Asa result, we had to search for signatures over a plausible range of orbital tilts. Charbonneau and Noyes did not detect a measurable signal during three nights of observations at Keck. Instead, they established that the planet had to be at least 10,000 times fainter than the star if its illuminated hemisphere could be viewed face on.

Our observations in April 1998, and in April and May 1999, produced a weak but plausible signal that was about 30% brighter than Charbonneau’s upper limit. The result was controversial. It was hard to reconcile with the Harvard team’s results, and implied that if the planet had a Jupiter-like reflectivity, its radius had to be nearly twice as large as Jupiter’s. We estimated at the time that there was roughly a 5% possibility that a chance alignment of noise in our data could produce a spurious detection of this strength. In March, April and May 2000 we observed tau Boo for a further six nights, carefully targeting those points in the orbit at which the reflected-light signature would be strongest and the absorption lines in the reflected light would be Doppler-shifted well away from the direct starlight. This strategy allowed us to probe much more sensitively for reflected light coming from a planet in the orbit suggested by our earlier measurement. However, the new observations – when combined with the 1998 and 1999 data – indicated that our earlier result had been spurious.

Nevertheless, our findings provided a new insight into the exoplanet’s atmosphere. If the planet reflects light between 385 and 580 nm uniformly, it has to be at least 30,000 times fainter than the star. This means that if the planet’s radius is 20% greater man Jupiter’s – as the models of Burrows’ group predict – its atmosphere must be less than 40% as reflective as Jupiter. This is well below the reflectivity predicted for a high-altitude silicate cloud deck, and suggests that tau Boo’s planet may well have a deep cloud deck with an overlying stealth coating of sodium gas.

Next steps

Our efforts are currently devoted to the innermost of the three planets that orbit the Sun-like star upsilon Andromedae. This planet appears to be 10 times lighter than tau Boo’s planet, so it has a much lower surface gravity and a more distended atmosphere. The most recent models of cloud formation for planets near their stars suggest that the silicate cloud deck may form higher in the atmospheres of planets with low surface gravities. If there is less sodium above the clouds, this planet could be very much more reflective than the planet orbiting tau Boo. We went back to the William Herschel Telescope in October and November last year to search for starlight reflected from the planet closest to upsilon Andromedae using the same techniques we developed for tau Boo, and we are currently analysing the data.

Several groups worldwide also observed HD 209458 last summer, hoping to detect the faint spectral signature of sodium in its atmosphere at the times when the planet passes between us and the star. Rather than look for light reflected from the planet’s atmosphere, this method involves searching for evidence that light passing through the planet’s atmosphere has been absorbed at the wavelengths of the sodium lines. HD 209458 has a similar mass and surface gravity to the planet around upsilon Andromedae that we are studying. If no-one succeeds in detecting mis absorption, it will mean that the amount of sodium above the cloud deck is relatively small. This would augur well for a direct detection in upsilon Andromedae, so exoplanetary astronomers await these results with keen anticipation.

Ultimately, the holy grail for planet-hunters is an Earth-like planet orbiting another star. Several programmes that have the potential to detect such planets are just beginning. The first of these involves searching for the faint dips in light that would occur as an Earth-like planet passes between us and its parent star. Earth-like planets, however, are 10 times smaller in radius – and therefore 100 times smaller in area – than Jupiter-like objects. So astronomers have to be able to detect a dip that is only one part in 10,000 of the total light from the star. This accuracy is difficult to achieve from the Earth’s surface because of turbulence and the variable transparency of the Earth’s atmosphere. However, several space missions have been proposed that would be capable of making such precise measurements.

The likelihood of an Earth-like planet’s orbit being oriented so that we can see it cross in front of its star is very low (about 1 in 200), so these missions will have to survey hundreds or thousands of stars to have a reasonable chance of success. This is true even if half or more of all Sun-like stars possess Earth-sized planets at Earth-like distances.

In autumn last year, planet-hunters were delighted to learn that one of these missions – named Eddington in honour of the great early 20th century astrophysicist Sir Arthur Eddington – has been selected as a reserve for one of the European Space Agency’s (ESA) so-called flexi-missions to fly within the next decade.

Another more bizarre search technique uses the gravitational bending of light around a star to detect planets. If we look towards the densely packed star clouds at the centre of our galaxy, we find that at any given time, one star in a million has its light amplified by this “gravitational microlensing” effect, as a star (usually a faint red dwarf) passes in front of it (figure 7). The background star appears to grow brighter then fade over a few weeks. If the star in the foreground has a Jupiter-like planet, there is about a 20% chance that a second, briefer amplification will occur as planet passes across our line of sight. That probability falls to about 2% if the planet is Earth-like.

The duration of such a secondary lensing event tells us the mass of the planet. For a Jupiter-like planet it would last about a day, whereas for an Earth-like object it would last about an hour. Several groups are already monitoring micro-lensing events as intensively as existing telescopes permit. At the International Astronomical Union’s general assembly in Manchester last summer, Penny Sackett of the Kapteyn Institute in Groningen, the Netherlands, announced that the very fact that none of these groups has seen such an event yet suggests that less than 30% of red dwarfs possess Jupiter-like planets in Jupiter-like orbits.

Other groups are planning to build a global network of automated telescopes, capable of doing the intensive brightness monitoring without the need for human intervention. Although astronomers only get one shot at each planet by this method, it could provide a good census of just how common Earth- and Jupiter-like planets may be around other stars that are less massive than the Sun.

The search for extraterrestrial fife

If these relatively cheap methods produce evidence that Earth-like planets are reasonably common around solar-type stars, then the incentive to study them in detail will be over-whelming. Both NASA and ESA are looking at the possibility of using networks of infrared telescopes in space to image and obtain spectra of Earth-like planets orbiting stars up to 30 light-years away.

These two missions – known as Terrestrial Planet Finder (TPF) and Darwin – have similar aims. Both propose combining the starlight collected by four or five telescopes flying in formation about 100 metres apart to form an interference pattern. The telescopes will be positioned so that the crests and troughs of the wave trains coming from the central star via the different telescopes cancel each other out. This will allow astronomers to detect and study the light from any Earth-like planets, unobscured by the glare of the parent star.

If planets are detected, astronomers will be able to search for the thumbprints of gases like water, carbon dioxide and ozone in their infrared spectra. The presence of water would suggest a relatively benign environment for life, but finding ozone would be a clincher. Ozone – and by implication, oxygen – should not be present in a planetary atmosphere unless some mechanism is constantly renewing the supply of this highly reactive gas. On Earth, the name we give to this mechanism is “life”.

But TPF and Darwin will not be cheap. They will be the most technically challenging space-science missions either agency has ever attempted. NASA and ESA will probably combine the Terrestrial Planet Finder and Darwin to keep the price within their “large missions” budget. The anticipated launch date for Darwin/TPF is around 2014, so we are uniquely privileged to be living at a time when there is a realistic prospect of seeing such a long-standing and fundamental question answered within our lifetimes.