Planet formation is imperfectly understood, but many models involve the accumulation of solid bodies of up to several Earth masses while the hydrogen-rich solar nebula is still present1. These bodies form an envelope of nebula gas bound together by gravity. In the outer Solar System, they may continue to accrete gas, forming the giant planets, while the high ultraviolet output of the early Sun causes those bodies closer to it to lose their gaseous envelope. In one development of these ideas2, runaway accretion causes many embryos to form quickly, some of which may merge, while others may be scattered into escape trajectories by proto-Jupiter or proto-Saturn. The formation of planets may be quite inefficient in the sense that more solid material is ejected than retained. There are uncertainties and alternatives3,4,5 but, because solar systems may have been formed in diverse ways, the possibility of bodies of roughly Earth mass in interstellar space should be taken seriously.

The amount of nebular gas accumulated and retained depends strongly on planet mass, nebula temperature, opacity assumptions and accretion timescale. An Earth-mass body eliminating its energy of formation in a million years and with only pressure-induced opacity of hydrogen6 develops an atmosphere with Matm/M≈0.01, where M is planet mass and Matm is atmospheric mass. More opaque models7 yield atmospheric masses with Matm/M≈0.001, in agreement with detailed models1.

The retention of a major part of this atmosphere is difficult at Earth orbit once most of the nebula has cleared, but becomes increasingly likely at greater distances, especially once the atmosphere has cooled (so that the photosphere is no longer large compared with the solid body). The atmospheric escape time can be as short as a million years at one astronomical unit early in the Solar System1, but longer than the age of the Solar System in the interstellar medium. Sputtering (collision with interstellar molecular or atomic hydrogen at tens to hundreds of kilometres per second) can be important if denser interstellar regions are encountered, but the column density of hydrogen in the case of Matm/M≈0.001 to 0.01 is so large that removing such an atmosphere would correspond to much more mass being sputtered per unit area than the total mass per unit area of a comet in the Oort cloud.

At the present epoch (assumed to be around 4.6 Gyr after formation), an interstellar planet would have a luminosity derived from long-lived radionuclides of around 4×1020 χerg s−1 if it is like Earth8, where χis the planet mass in units of Earth masses. Assuming a thin atmosphere and an Earth-like density, the effective temperature Te of the planet is given by Te≈34 χ1/12K. From hydrostatic equilibrium, the surface pressure Ps≈106×Matm/M bars. However, optical-depth unity at relevant infrared wavelengths (about 100 μm) is achieved in such an atmosphere at a pressure of around 1 bar (refs 6, 9) and liquefaction at this pressure occurs at a temperature of around 22 K, below the actual atmospheric temperature. A convective gas adiabat must form at all greater depths (at a pressure between 1 bar and Ps), even when the heat flow is very low. An adequate estimate for this adiabat turns out to be TP0.36, which does not intercept the condensation curve for hydrogen. It follows that the surface temperature is given by Ts≈425χ1/12(Matm/M)/0.001)0.36 K.

The melting point of water is typically exceeded for basal pressures of the order of one kilobar. The atmosphere will have several cloud layers (methane, ammonia and perhaps water, like Uranus), but this has little influence on the temperature estimate.

It seems, then, that bodies with water oceans are possible in interstellar space. The ideal conditions are plausibly at an Earth mass or slightly less, similar to the expected masses of embryos ejected during the formation of giant planets. Bodies with Earthlike water reservoirs may have an ocean underlain with a rock core. Either way, these bodies are expected to have volcanism in the rocky component and a dynamo-generated magnetic field leading to a well developed (very large) magnetosphere. Despite thermal radiation at microwave frequencies that corresponds to the temperatures deep within their atmospheres (analogous to Uranus9), and despite the possibility of non-thermal radio emission, they will be very difficult to detect.

If life can develop and be sustained without sunlight (but with other energy sources, plausibly volcanism or lightning in this instance), these bodies may provide a long-lived, stable environment for life (albeit one where the temperatures slowly decline on a billion-year timescale). The complexity and biomass may be low because the energy source will be small, but it is conceivable that these are the most common sites of life in the Universe. Details of the above results are available from the author.