LETTER Doi:10.1038/Nature10990

LETTER doi:10.1038/nature10990

An ultraviolet–optical flare from the tidal disruption of a helium-rich stellar core

S. Gezari1, R. Chornock2, A. Rest3, M. E. Huber4, K. Forster5, E. Berger2, P. J. Challis2, J. D. Neill5, D. C. Martin5, T. Heckman1, A. Lawrence6, C. Norman1, G. Narayan2, R. J. Foley2, G. H. Marion2, D. Scolnic1, L. Chomiuk2, A. Soderberg2, K. Smith7, R. P. Kirshner2, A. G. Riess1, S. J. Smartt7, C. W. Stubbs2, J. L. Tonry4, W. M. Wood-Vasey8, W. S. Burgett4, K. C. Chambers4, T. Grav9, J. N. Heasley4, N. Kaiser4, R.-P. Kudritzki4, E. A. Magnier4, J. S. Morgan4 & P. A. Price10

The flare of radiation from the tidal disruption and accretion of a on 2010 July 12.31 UT and its subsequent decay until 2011 September star can be used as a marker for supermassive black holes that 1.24 UT (Supplementary Table 1). PS1-10jh was discovered indepen- otherwise lie dormant and undetected in the centres of distant dently as a transient, near-ultraviolet (NUV) source at the 20s level by galaxies1. Previous candidate flares2–6 have had declining light curves the Galaxy Evolution Explorer15 (GALEX) Time Domain Survey in good agreement with expectations, but with poor constraints on (TDS) on 2010 June 17.68 UT within 2.5 6 3.0 arcsec of the PS1 loca- the time of disruption and the type of star disrupted, because the tion, and was detected in ten more epochs of TDS observations rising emission was not observed. Recently, two ‘relativistic’ candid- between then and 2011 June 10.68 UT (Supplementary Table 2). No ate tidal disruption events were discovered, each of whose extreme source is detected in a coaddition of all the TDS epochs in 2009, with a X-ray luminosity and synchrotron radio emission were interpreted 3s upper limit of .25.6 mag implying a peak amplitude of variability as the onset of emission from a relativistic jet7–10.Herewereporta in the NUV of .6.4 mag. See Supplementary Information for details luminous ultraviolet–optical flare from the nuclear region of an on the PS1 and GALEX photometry. inactive galaxy at a redshift of 0.1696. The observed continuum is PS1-10jh is coincident with the centre of a galaxy within the 3s cooler than expected for a simple accreting debris disk, but the well- positional uncertainty (0.036 arcsec; Supplementary Information), sampled rise and decay of the light curve follow the predicted mass with rest-frame u-, g-, r-, i- and z-band photometry from the Sloan accretion rate and can be modelled to determine the time of disrup- Digital Sky Survey16 and K-band photometry from the UK Infrared tion to an accuracy of two days. The black hole has a mass of about Telescope Infrared Sky Survey17 fitted with a galaxy template18 with 9 two million solar masses, modulo a factor dependent on the mass Mstars~(3:6+0:2)|10 M8 and Mr 5218.7 mag, where Mstars is the and radius of the star disrupted. On the basis of the spectroscopic galaxy stellar mass and Mr is the absolute r-band magnitude. The mass signature of ionized helium from the unbound debris, we determine of the central black hole as determined indirectly from locally established 19 z4| 6 that the disrupted star was a helium-rich stellar core. scaling relations is 4{2 10 M8.Weobtainedfiveepochsofoptical When the pericentre of a star’s orbit (Rp) passes within the tidal spectroscopy at the location of PS1-10jh between 2010 June 16.33 and 1=3 disruption radius of a massive black hole, RTstellar population with an age of 1.4–5.0 Gyr, depending on the the black hole and the rest being ejected at high velocity1. For black holes chosen metallicity, and a fading hot blackbody component with tem- 8 3=2 {1=2 4 above a critical mass, Mcrit<10 rÃ mÃ M8 (where rÃ~RÃ=R8, perature TBB < 3 3 10 K (Fig. 1). mÃ~MÃ=M8, R[ is the solar radius and M[ is the solar mass), the The spectra show no narrow emission lines that would be indicative star becomes trapped within the event horizon of the black hole before of star formation or an active galactic nucleus (AGN). We obtained a being disrupted. The mass accretion rate (M_ ) in a tidal disruption event 10-ks, 0.2–10-keV X-ray observation, using the Chandra X-ray can be calculated directly from the orbital return times of the bound Observatory, at the location of PS1-10jh on 2011 May 22.96 UT, and debris1,11,12. For the simplest case, of a star of uniform density, this detected no source above the background with a 3s upper limit of _ {5=3 41 21 yields M~(2=3)(fMÃ=tmin)(t=tmin) , where f is the fraction of the LX(0.2–10 keV) , 5.8 3 10 erg s for an unobscured AGN spec- star accreted and tmin is the orbital period of the most tightly bound trum. The X-ray faintness and extreme NUV variability amplitude debris and, therefore, the time delay between the time of disruption and of PS1-10jh strongly disfavour its origin in an AGN, and its prolonged 1=2 {1 3=2 the start of the flare, which scales as MBH MÃ RÃ for Rp 5 RT. The brightness in the ultraviolet strongly disfavours its origin in a super- radiative output of the accreted debris is less certain, and depends on nova (Supplementary Information). the ratio of the accretion rate to the Eddington rate13. The rise and decay of the light curve of PS1-10jh is well described by The optical transient, PS1-10jh (right ascension, aJ2000 5 16 h numerical simulations of the mass return rate from a star that is tidally 09 min 28.296 s; declination, dJ2000 5153u 409 23.5299), was discovered disrupted at Rp 5 RT and has an internal structure parameterized by a on 2010 May 31.45 UT (universal time) in the Pan-STARRS114 (PS1) polytropic exponent of 5/3 characteristic of a fully convective star or a Medium Deep Survey by our two independent image-differencing degenerate core20 (Fig. 2). The decay from the peak is too steep to be pipelines. The densely sampled (cadence, Dt < 3 d) optical light curves fitted by simulations of a more centrally concentrated stellar structure, of PS1-10jh in the gP1,rP1,iP1 and zP1 bands (Supplementary such as one that is characteristic of a solar-type star (Supplementary Information) follow the rise of the transient to its peak in the gP1 band Information). There are systematic differences between the light curve

1Department of Physics and Astronomy, Johns Hopkins University, 3400 North Charles Street, Baltimore, Maryland 21218, USA. 2Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, Massachusetts 02138, USA. 3Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, Maryland 21218, USA. 4Institute for Astronomy, University of Hawaii, 2680 Woodlawn Drive, Honolulu, Hawaii 96822, USA. 5California Institute of Technology, 1200 East California Boulevard, Pasadena, California 91125, USA. 6Institute for Astronomy, University of Edinburgh Scottish Universities Physics Alliance, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK. 7Astrophysics Research Centre, School of Mathematics and Physics, Queen’s University Belfast, Belfast BT7 1NN, UK. 8Pittsburgh Particle Physics, Astrophysics, and Cosmology Center, Department of Physics and Astronomy, University of Pittsburgh, 3941 O’Hara Street, Pittsburgh, Pennsylvania 15260, USA. 9Planetary Science Institute, 1700 East Fort Lowell, Tucson, Arizona 85719, USA. 10Department of Astrophysical Sciences, Princeton University, Princeton, New Jersey 08544, USA.

a Time since peak (rest-frame days)

) Day –22 –50 –20 0 200 500 –1 6 18 Å S SS S −1 NUV X s BB + 2.5-Gyr model g + 0.5 –2 n = 5/3 20 r + 2.0 4 i + 3.5 z + 5.0 erg cm

–17 22 2 T BB = 30,000 K (10 λ f 24 2.5-Gyr model 0

He II λ = 3,203 Å He II λ = 4,686 Å Magnitude (mag) 1 26

Difference 0 28

b 10 100 1.2 ) Day 254 Time since disruption (rest-frame days) –1

Å 1.0 Figure 2 | Ultraviolet–optical light curve. The GALEX NUV and PS1 g -,

−1 P1

s rP1-, iP1- and zP1-band light curves of PS1-10jh (with the flux from the host –2 0.8 galaxy removed), plotted against logarithmic time since the peak (top axis) and 0.6 since the disruption (bottom axis).The curves (shown with solid lines scaled to

erg cm the flux in the GALEX and PS1 bands) were determined from the best fit of the

–17 0.4 20 gP1-band light curve to a numerical model of the mass accretion rate of a

(10 0.2 tidally disrupted star with a polytropic exponent of 5/3. For each of the four λ f optical bands, we independently performed a least-squares fit of the model for a 0.0 106M8 black hole to the light curve from 236 to 58 rest-frame days from the 0.2 peak, with the time of disruption, a vertical scaling factor and a time stretch factor as free parameters. The GALEX and PS1 photometry at t . 240 rest- 0.0 frame days since the peak is shown binned in time to increase the signal-to- noise ratio. The dates of multiple epochs of MMT spectroscopy are marked Difference –0.2 with an S, and the date of the Chandra X-ray observation is marked with an X. 3,000 4,000 5,000 6,000 7,000 The grey line shows an n 5 5/3 power-law decay from the peak. Errors, 1s; Rest wavelength (Å) arrows, 3s upper limits.

Figure 1 | Optical spectrum. MMT optical spectra (black) of PS1-10jh 21 (full-width at half-maximum, 9,000 6 700 km s ) and 3,203 A˚ that obtained 222 (a) and 254 (b) rest-frame days from the peak, expressed in terms of flux density. Each continuum is fitted with a combination (magenta) of a fade in time along with the ultraviolet–optical continuum. The lack of stellar population 2.5 Gyr old (red) and a fading blackbody with a temperature Balmer line emission in the spectra requires an extremely low of ,3 3 104 K determined from the ultraviolet–optical spectral energy hydrogen mass fraction, of ,0.2 (Supplementary Information), distribution (SED) (blue). The difference between the black and magenta which cannot be found in the ambient interstellar medium or in a spectra is shown in the lower part of each panel. Helium II emission at passive accretion disk. This is the strongest evidence that PS1-10jh l 5 4,686 A˚ (Fowler series, n 5 4 R 3) is detected above the continuum model must be fuelled by the accretion of a star that has lost its hydrogen and fitted with a Gaussian with a full-width at half-maximum of 21 40 21 envelope, either through stellar winds or through tidal interactions 9,000 6 700 km s and luminosity L 5 (9 6 1) 3 10 erg s (plotted with a with the central supermassive black hole. The broad width of the green line in the early epoch (a)). Residual emission above the continuum line is also what is expected from the velocities of the most energetic model is also detected at ,3,200 A˚ , which is coincident with the location of the unbound stellar debris in a tidal disruption event21, that is He II l 5 3,203 A˚ (Fowler series, n 5 5 R 3) line, and confirms the | 4 6 1=6 {1=2 1=3 {1 identification of He II l 5 4,686 A˚ emission. The observed flux ratio of He II vmax<1 10 (MBH=10 M8) (RT=Rp)rÃ mÃ km s l 5 3,203 A˚ emission to He II l 5 4,686 A˚ emission is 0.50 6 0.10, measured We measure the SED of the flare over time from the nearly using a Gaussian fit to the l 5 3,203 A˚ line with a width fixed to that of the simultaneous PS1 optical and GALEX ultraviolet observations (with l 5 4,686 A˚ line, limits the internal extinction to E(B 2 V) , 0.08 mag the host galaxy flux removed; Fig. 3). The pre-peak SED is fitted by a II l 5 ˚ 4 (Supplementary Information). The He 4,686 A line is still evident as an blackbody with TBB 5 (2.9 6 0.2) 3 10 K, consistent with the black- excess above the model in the later epoch (b), but it has faded by a factor of ,10 body component seen in the spectra. However, the temperature fit is since 22 rest-frame days before the peak, the same factor by which the very sensitive to internal extinction. If we correct for the maximum ultraviolet continuum has faded during this time. The absolute flux scaling in internal extinction of E(B 2 V) 5 0.08 mag allowed by the ratio the later epoch is uncertain owing to obscuration by clouds on the date of the ˚ ˚ observation. between the observed He II l 5 3,203 A and l 5 4,686 A emission, the best-fit temperature increases to (5.5 6 0.4) 3 104 K. In fact, we know 4 that the photo-ionizing continuum must have TBB>5|10 K22rest- and the model during the early rise (more than 44 rest-frame days frame days before the peak to produce enough l , 228 A˚ photons to before the peak) and the late decay (more than 240 rest-frame days photo-ionize the He II l 5 4,686 A˚ line observed with a luminosity of after the peak), which could imply a stellar structure more complex (9 6 1) 3 1040 erg s21. The late-time SED can be fitted with the same than one described by a single polytrope. The mass of the black hole is temperature as the pre-peak SED. We note that the observed con- determined from the stretch factor of 1.38 6 0.03 applied to fit the tinuum temperature, and even the maximum temperature allowed by model of a 106M8 black hole to the light curve, which implies that the possible de-reddening, is considerably cooler than the temperature of 1=12 {1=2 {1=6 time of disruption was 76 6 2 d before the peak and that * : | 5( = 6 8) { 2 5 10 MBH 10 M rÃ mÃ K expected from material ~ | 6 2 3 13 MBH (1:9+0:1) 10 mÃrÃ M8. radiating at the Eddington limit at the tidal radius . This discrepancy The most constraining property of PS1-10jh is the detection of very is also seen in AGNs22 and may imply that the continuum we see is due broad high-ionization He II emission at wavelengths of l 5 4,686 A˚ to reprocessing of some kind22,23.

3. Komossa, S. et al. A huge drop in the X-ray luminosity of the nonactive galaxy RX GALEX Day –19 to –17 E B – V J1242.6–1119A, and the first postflare spectrum: testing the tidal disruption PS1 ( ) = 0.08 Day 242 to 247 scenario. Astrophys. J. 603, L17–L20 (2004). –14 10 LT Day 375 4. Esquej, P. et al. Evolution of tidal disruption candidates discovered by XMM- Chandra Newton. Astron. Astrophys. 489, 543–554 (2008). AGN 5. Gezari, S. et al. Luminous thermal flares from quiescent supermassive black holes. Astrophys. J. 698, 1367–1379 (2009). ) 10–15

–1 6. van Velzen, S. et al. Optical discovery of probable stellar tidal disruption flares.

Å Astrophys. J. 741, 73–96 (2011). –2 7. Bloom, J. S. et al. A possible relativistic jetted outburst from a massive black hole

cm –16 fed by a tidally disrupted star. Science 333, 203–206 (2011). –1 10 8. Burrows, D. N. et al. Relativistic jet activity from the tidal disruption of a star by a massive black hole. Nature 476, 421–424 (2011).

(erg s (erg 9. Zauderer, B. A. et al. Birth of a relativistic outflow in the unusual c-ray transient Swift λ f 10–17 J164449.31573451. Nature 476, 425–428 (2011). 10. Cenko, S. B. et al. Swift J2058.410516: discovery of a possible second relativistic E(B – V) = 0 tidal disruption flare. Preprint at Æhttp://arxiv.org/abs/1107.5307æ (2011). 11. Phinney, E. S. in The Center of the Galaxy (ed. Morris, M.) 543–553 (IAU Symp. 136, 10–18 Kluwer, 1989). 12. Evans, C. R. & Kochanek, C. S. The tidal disruption of a star by a massive black hole. 10 100 1,000 10,000 Astrophys. J. 346, L13–L16 (1989). 13. Ulmer, A. Flares from the tidal disruption of stars by massive black holes. Astrophys. Rest wavelength (Å) J. 514, 180–187 (1999). 14. Kaiser, N. et al. The Pan-STARRS wide-field optical/NIR imaging survey. Proc. SPIE Figure 3 | Spectral energy distribution. SED of PS1-10jh during nearly 7733, 77330E (2010). simultaneous GALEX ultraviolet and PS1 optical observations (with the flux 15. Martin, D. C. et al. The Galaxy Evolution Explorer: a space ultraviolet survey mission. from the host galaxy removed) at two epochs (rest-frame days 219 to 217 and Astrophys. J. 619, L1–L6 (2005). 242 to 247 from the peak of the flare). Flux densities have been corrected for 16. Aihara, H. et al. The eighth data release of the Sloan Digital Sky Survey: first data galactic extinction of E(B 2 V) 5 0.013 mag. The ultraviolet–optical SED from from SDSS-III. Astrophys. J. Suppl. Ser. 193, 29–45 (2011). 219 to 247 rest-frame days from the peak is fitted with a 2.9 3 104 K blackbody. 17. Lawrence, A. et al. The UKIRT Infrared Deep Sky Survey (UKIDSS). Mon. Not. R. Astron. Soc. 379, 1599–1617 (2007). Orange solid lines show blackbodies with this temperature scaled to the 18. Blanton, M. R. & Roweis, S. K-corrections and filter transformations in the respective NUV flux densities for the respective epochs. Open symbols show ultraviolet, optical, and near-infrared. Astron. J. 133, 734–754 (2007). the GALEX and PS1 flux densities corrected for an internal extinction of 19. Ha¨ring, N. & Rix, H.-W. On the black hole mass-bulge mass relation. Astrophys. J. E(B 2 V) 5 0.08 mag, and the dotted blue lines shows the 5.5 3 104 K 604, L89–L92 (2004). blackbody fit to the de-reddened flux densities for the respective epochs. The 20. Lodato, G., King, A. R. & Pringle, J. E. Stellar disruption by a supermassive black hole: is the light curve really proportional to t25/3? Mon. Not. R. Astron. Soc. 392, upper limit from the Chandra observation on 2011 May 22.96 UT assuming a 332–340 (2009). spectrum with a photon index of C 5 2, typical of an AGN, is plotted with a 21. Strubbe, L. E. & Quataert, E. Optical flares from the tidal disruption of stars by thick black line. The X-ray flux density expected from an unobscured AGN massive black holes. Mon. Not. R. Astron. Soc. 400, 2070–2084 (2009). with a comparable NUV flux is plotted for comparison with a thick grey line28. 22. Lawrence, A. The UV peak in active galactic nuclei: a false continuum from blurred Also shown are the u-, g- and r-band flux densities measured from aperture reflection? Preprint Æhttp://arxiv.org/abs/1110.0854æ (2011). 29 23. Loeb, A. & Ulmer, A. Optical appearance of the debris of a star disrupted by a photometry with the Liverpool Telescope (LT) on 2011 September 24.91 UT, massive black hole. Astrophys. J. 489, 573–578 (1997). after subtracting the flux from the host galaxy as measured by the Sloan Digital 24. Davies, M. B. & King, A. The stars of the galactic center. Astrophys. J. 624, L25–L27 Sky Survey. Errors, 1s. (2005). 25. Kobayashi, S., Laguna, P., Phinney, E. S. & Me´sza´ros, P. Gravitational waves and On the basis of the arguments above, we assume that the observed X-ray signals from stellar disruption by a massive black hole. Astrophys. J. 615, | 4 855–865 (2004). temperature is a lower limit, TBB>3 10 K. The peak bolometric 26. Maxted, P. F. L. et al. Discovery of a stripped red giant core in a bright eclipsing 44 {1 luminosity is thus >2:2|10 erg s and the total energy emitted binary system. Mon. Not. R. Astron. Soc. 418, 1156–1164 (2011). from integrating under the light-curve model is >2:1|1051 erg, cor- 27. Ayal, S., Livio, M. & Piran, T. Tidal disruption of a solar-type star by a supermassive {1 black hole. Astrophys. J. 545, 772–780 (2000). > : (e= : ) 8 responding to a total accreted mass (Macc)of 0 012 0 1 M , 28. Steffen, A. T. et al. The X-ray-to-optical properties of optically selected active where e is the efficiency of converting matter into radiation. galaxies over wide luminosity and redshift ranges. Astron. J. 131, 2826–2842 The internal structure and high helium abundance of the star (2006). 29. Steele, I. A. et al. The Liverpool Telescope: performance and first results. Proc. SPIE derived from the light curve and the spectra can be consistently 5389, 679 (2004). modelled as the tidally stripped core of a red giant (the precursor to Supplementary Information is linked to the online version of the paper at a helium white dwarf) that had a main-sequence mass of M >1M8 so Ã www.nature.com/nature. as to have evolved off the main sequence in less time than the age of the stellar population (,5 Gyr). This tidal stripping mechanism has been Acknowledgements We thank H. Tananbaum for approving our Chandra Director’s 24 Discretionary Time request. We are grateful to G. Lodato for providing the tidal invoked to explain the hot stars in the Galactic Centre , and the rate of disruption event models in tabular form, and to S. Moran for running software to tidal disruption of tidally stripped stars is likely to be higher than that calculate the host-galaxy K-corrections. We thank R. E. Williams for discussions on the of solar-type stars25. The mass of the black hole derived from the light- line emission in the spectra. S.G. was supported by NASA through a Hubble Fellowship grant awarded by the Space Telescope Science Institute, which is operated by AURA curve fit depends on the mass and radius of the star at the time of Inc. for NASA. Partial support for this work was provided by the National Science disruption. Using MÃ<0:23M8 and RÃ<0:33R8 (values measured Foundation. The PS1 survey has been made possible through contributions of the for a red giant core that was stripped in a binary system26), and assum- Institute for Astronomy, the University of Hawaii, the Pan-STARRS Project Office, the Max-Planck Society and its participating institutes, The Johns Hopkins University, ing that the evolution of the core is similar to one that is tidally Durham University, the University of Edinburgh, Queen’s University Belfast, the stripped, we find that f ~Macc=MÃ>0.058 (approaching f >0.1 as Harvard-Smithsonian Center for Astrophysics, the Las Cumbres Observatory Global 27 6 Telescope Network Inc. and the National Central University of Taiwan, and by NASA measured in simulations ), that MBH~(2:8+0:1)|10 M8 and that under a grant issued through the Planetary Science Division of the NASA Science the peak luminosity approaches the Eddington luminosity of the Mission Directorate. We acknowledge NASA’s support for construction, operation, and supermassive black hole (Lpeak>0:6LEdd). science analysis of the GALEX mission, which was developed in cooperation with Centre National d’Etudes Spatiales of France and the Korean Ministry of Science and Received 8 November 2011; accepted 23 February 2012. Technology. Some of the observations reported here were obtained at the MMT Observatory, which is a joint facility of the Smithsonian Institution and the University of Published online 2 May 2012. Arizona, and at the Liverpool Telescope, which is operated with financial support from the UK Science and Technology Facilities Council. The computations in this paper were 1. Rees, M. J. Tidal disruption of stars by black holes of 10 to the 6th-10 to the 8th run on the Odyssey cluster supported by the FAS Science Division Research solar masses in nearby galaxies. Nature 333, 523–528 (1988). Computing Group at Harvard University. R.J.F. is a Clay Fellow. 2. Komossa, S. & Bade, N. The giant X-ray outbursts in NGC 5905 and IC 3599: follow-up observations and outburst scenarios. Astron. Astrophys. 343, 775–787 Author Contributions S.G. designed the observations and the transient detection (1999). pipeline for the GALEX TDS, and measured the ultraviolet photometry of PS1-10jh. K.F.

10 MAY 2012 | VOL 485 | NATURE | 219 ©2012 Macmillan Publishers Limited. All rights reserved RESEARCH LETTER and J.D.N coordinated, and D.C.M. facilitated, the GALEX TDS observations. A.R. observation and analysed the data. A.L. obtained the Liverpool Telescope optical designed the PhotPipe transient detection pipeline hosted by Harvard/CfA for the PS1 imaging observations and analysed the data, and stimulated discussions on the nature Medium Deep Survey (MDS), and measured the optical photometry of PS1-10jh. R.C. of the SED of PS1-10jh. S.G. analysed and modelled the multicolour light curve and the designed, implemented and analysed the MMT optical spectroscopy observations, and SED of PS1-10jh. T.H. and C.N. stimulated discussions on the nature of the disrupted contributed to the operation of PhotPipe and the visual inspection of transient alerts. star. The paper was organized and written by S.G., and all authors provided feedback on E.B. proposed and facilitated the MMT observations. M.E.H., G.N., D.S. and R.J.F. the manuscript. contributed to the operation of PhotPipe and the visual inspection of transient alerts. P.J.C., R.J.F., G.H.M., L.C. and A.S. contributed to the MMT observations. S.J.S. designed, Author Information Reprints and permissions information is available at and K.S. operated, the transient pipeline for PS1 MDS hosted by Queen’s University www.nature.com/reprints. The authors declare no competing financial interests. Belfast. C.W.S., J.L.T. and W.M.W.-V. facilitated the transient pipelines for PS1 MDS. Readers are welcome to comment on the online version of this article at W.S.B., K.C.C., T.G., J.N.H., N.K., R.-P.K., E.A.M., J.S.M., P.A.P., C.W.S. and J.L.T. helped build www.nature.com/nature. Correspondence and requests for materials should be the PS1 system. S.G. requested the Director’s Discretionary Time Chandra X-ray addressed to S.G. ([email protected]).