LETTERS

Laser-shock compression of diamond and evidence of a negative-slope melting curve

STEPHANIE BRYGOO1 , EMERIC HENRY1 , PAUL LOUBEYRE1 *, JON EGGERT2 , MICHEL KOENIG3 , BERENICE LOUPIAS3 , ALESSANDRA BENUZZI-MOUNAIX3 AND MARC RABEC LE GLOAHEC1,3
1 2

` Commissariat a lEnergie Atomique, BP 12, 91680 Bruy` eres le Ch atel, France Lawrence Livermore National Laboratory, Livermore, California 94551, USA 3 Laboratoire pour lUtilisation des Lasers Intenses, 91128 Palaiseau, France * e-mail: paul.loubeyre@cea.fr

Published online: 25 March 2007; doi:10.1038/nmat1863

Diamond is the only known high-pressure structure of carbon. In spite of its fundamental and planetary importance, the stability domain of this strong covalent material is largely unknown. After decades of experimental eorts, evidence was obtained that the diamondliquid melting line has a positive slope above the graphitediamondliquid triple point1 . At higher pressure, theoretical studies have suggested that the melting curve of diamond should have a maximum25 , owing to a loss of stability of the sp3 hybridization in the uid phase. Accurate Hugoniot data of diamond exist up to 590 GPa (ref. 6). Higher-pressure measurements along the diamond Hugoniot have recently been achieved by laser shocks7,8 , showing that diamond probably melts to a conducting uid. We report here laser-shock Hugoniot data across the melting transition. The shocked diamond crystal begins to melt around 750 GPa. Furthermore, a negative volume discontinuity at melting is observed. This requires a negative melting slope and thus supports the existence of a maximum on the diamond melting curve. These melting data allow us to test various calculations of the phase diagram of carbon at very high pressure. Finally, the stability domain of the diamond crystal is now constrained in a relevant region for Uranus-like planetary interiors9 . The determination of the melting curve of elements and compounds is a subject of great current interest in high-pressure research. Experiments are making great progress by extending the pressure (P )temperature (T ) domain of measurements, statically with laser heating in the diamond anvil cell and dynamically with laser shocks. Also, new computational methods are being applied to calculate melting curves. Yet, reliably predicting or measuring a melting curve is dicult. Carbon is probably the system for which the largest number of calculations of the melting curve has been carried out. Theoretical predictions of the phase diagram of carbon are made on the basis of various levels of approximations, for example either ab initio density functional theory (DFT) calculations that depend only on the atomic number and the form of the functional24 or chemical models on the basis of the description of interactions between C atoms10,11 or a thermodynamical approach on the basis of tting models of Gibbs energy to the most recent experimental data5 . Ab initio calculations have disclosed an interesting phase diagram of carbon, with the transformation of diamond into the BC8 structure at 1,100 GPa and a maximum on the melting curve, in the range 450 GPa (ref. 4) to 630 GPa (ref. 3). The thermodynamical approach also
274

predicts a maximum on the melting curve at 300 GPa (ref. 5). However, calculations on the basis of the chemical approach nd no evidence of a maximum on the melting curve. Therefore, the accurate determination of the melting point of diamond at very high pressure should be a good test of these various calculations. Recently, two laser shock-wave experiments7,8 have been carried out on diamond across its melting along the Hugoniot. First, reectance data7 show that diamond undergoes a continuous transition from optically absorbing to reecting in the range 6001,000 GPa, probably associated with melting. Second, Hugoniot data up to 2 TPa do not show any discontinuity8 . The present measurements are made with a dierent geometry of the target, with quartz as the reference material. They aim at carrying out a more precise Hugoniot determination around melting and hence at clarifying the issue of a possible negative volume discontinuity associated with melting. Experiments were conducted at the LULI 2000 facility. We used two laser beams at 527 nm, in a 1.5 ns pulse, with energy up to 1 kJ. The focal spot had a diameter of 400 m, so that the maximum ux onto the target was 5 1014 W cm2 . The targets are shown in Fig. 1a. They consisted of a CH plastic ablator, an Al foil to reduce the quantity of high-energy X-rays produced at the ablation front, which can preheat the target, a plate of quartz, which operates as the pusher, and a plate of IIA diamond oriented along the (100) axis. The thickness of the diamond plates was 21.1 m, with an error of 0.1 m, and that of quartz plates between 26 and 28 m, as measured by white-light interference with an uncertainty of 1 m. The diamond plate was stuck on the quartz plate only by adhesion forces. The shock was transmitted from quartz into diamond. Quartz was used as the reference material, the measurements being made relative to it. In some test targets, the diamond plate was replaced by a fused silica plate. Two series of experiments have been carried out on diamond. In the rst series, shock velocities were measured by an interferometric VISAR diagnostic12 operating at 1,064 nm. The shock velocities in quartz and in diamond were determined by the fringe shifts detected by the interferometer and caused by a Doppler shift at the reecting shock front. The refractive index and initial density of quartz and of diamond, which are used to extract the shock velocity from the VISAR phase shift measurement, are very accurately known13 . A 1% precision in the velocity measurement could be achieved because the fringe position is determined to 5% of a fringe and the shock velocities measured here correspond typically to ve to seven fringes14 . Also,
nature materials VOL 6 APRIL 2007 www.nature.com/naturematerials

LETTERS
a
kJ laser C, 20 m VISAR laser Space Al Qz C

Breakout in diamond

CH 20 m

SiO2, 25 m Al, 40 m Breakout in quartz Time

2,000

1,500 Pressure (GPa)

Hugoniot Isentrope Reshock Symmetric Hugoniot

1,000

0 diamond Us diamond

P1, Up1

500

0 fused silica Us fused silica

0 0 5 10 15 Up (km s 1) 20 25 30

Figure 1 Experimental conguration for the determination of the shock velocity in diamond. a, Drawing of the target. The CH and Al layers are the coupling part to the kJ laser; quartz is the pusher; the diamond plate (or fused silica) is stuck on it. b, VISAR image of a shock. Before breakout of the shock-wave in quartz, the VISAR laser is reecting on the Al surface, giving a steady interference pattern. At breakout of the shock-wave in quartz and then in diamond, the laser is reected at the shock front in these materials. The phase shift of the interference pattern is then directly related to the shock velocity. The shock velocity is also measured by the transit time across the plates obtained from the breakout lines. c, The impedance match construction using quartz as the reference material is made in the (P, Up ) plane. The polar line (with the slope 0 Us ) of the diamond state intersects the reshock branch of quartz, whereas for fused silica it is the release branch. The differences between the symmetric Hugoniot rst-order approximation and the true thermodynamical off-Hugoniot states of quartz are seen to be small.

the line-imaging VISAR allows us to determine the shock-velocity measurement in space and time. Therefore, a good planarity of the shock front was observed over 250 m and an unsteadiness of the shock was quantied with a 2% velocity decrease across the quartz and diamond plates. In the second run, the shock velocities in quartz and diamond were determined by the transit time across the quartz and the diamond plates, corrected for the 2% velocity attenuation, because ghost fringes perturbed the VISAR images owing to a poor antireective coating on the diamond surface in the second series of targets. The error in the velocity measurement was then larger and arose mainly from the uncertainty in the quartz thickness and from the uncertainty in the breakout time at various interfaces, that is about 50 ps. The shock state of diamond is completely given by the velocity of the shock front, Us , that is measured and the velocity of the material behind the front, Up , that is obtained from the shock velocity in the reference material, here quartz. The impedance-matching construction, shown in Fig. 1c, allows us to determine Up in diamond from the measured Us in quartz and in diamond. It requires the knowledge of the quartz Hugoniot
nature materials VOL 6 APRIL 2007 www.nature.com/naturematerials

and its o-Hugoniot states, along either a release isentrope or a reshock. The Hugoniot of quartz is now very well determined15 . A well known practical approximation for the o-Hugoniot states is a graphical construction which approximates the reshock and release prole by the mirror reection of the principal Hugoniot in the pressure (P ) versus particle velocity (Up ) plane16 (see Fig. 1c). This approximation gives remarkably accurate results if the pressure variation of the standard along the o-Hugoniot state is not too large (which is the case here for diamond and fused silica, with a 20% variation at maximum). The correction to this approximation can then be estimated by using an equation of state of quartz. In all the shocks, the quartz was highly reecting with a shock velocity Us > 17 km s1 , corresponding to the transformation of quartz into a conducting dissociated uid. We have used a MieGr neisen equation of state (EoS) of quartz, with a Gr neisen u u parameter = 0.66, valid in this dense uid domain17 , and the correction was made with it. Similar corrections were obtained by using the Sesame EoS. However, the dierences in the impedance matching constructions by using either the rst-order symmetric Hugoniot construction or the isentrope release (for fused silica) or reshock construction (for diamond) are small, roughly the size of the symbols in Fig. 2, hence propagating an error in the pressuredensity plane much smaller than the error bars due to the measurement uncertainties. It should be said that, in both runs, the velocity of the shock breakout from diamond into vacuum, which should be equal roughly to 2Up (ref. 16), could be measured by the VISAR and that it is in agreement with the quartz impedance-match construction within the error bars of the second run. Finally, a one-dimensional hydro-radiative simulation (CHIVAS code developed by A. Decoster) was carried out to see the level of preheat in quartz and diamond, if any. For the maximum laser irradiation intensity of 5 1014 W cm2 , used in the present experiment, the preheat in quartz is around 0.2 eV and the preheat in diamond is 0.03 eV. Preheat can thus be ignored. The shock measurements on fused silica were used as a test case for systematic errors. First, exactly the same target geometry and similar laser uence were used for fused silica and diamond targets. Second, the Hugoniot curves of samples of SiO2 with various initial densities had been measured extensively17 , even up to the pressure range of the present experiment, that is when SiO2 should be a dissociated conducting uid. It has been observed that the Hugoniot curves corresponding to dierent initial densities of SiO2 represent a family of approximately parallel lines in the (Us ,Up ) plane. In fact, the parallel shift with initial density between these Hugoniot curves was explained in the framework of the Gr neisen model with a Gr neisen parameter of 0.60.66. The u u present data points of the Hugoniot of fused silica are shown in Fig. 2a. Although there are no literature data in the same pressure range, the data points fall on the predicted locus, that is on a line parallel to the Hugoniot of quartz and with a shift compatible with the Gr neisen model and in the extrapolation of the loweru pressure data18 . These results on fused silica validate the design of the targets, with quartz as the reference material, to obtain accurate measurements. Nine shots were carried out on diamond, four in the rst run on the basis of VISAR measurements and ve in the second run on the basis of transit-time measurements. The Us Up data are shown in Fig. 2b. Our error bars include all measurement uncertainties. The good agreement between the two sets of data gives condence in the reproducibility of the measurements. The three data points corresponding to a Up less than 10 km s1 are in agreement within their error bars with the linear t of previous diamond Hugoniot data6 , Us = 1.02Up + 12.1, and extend them to 720 GPa. The slope of 1.02 is in agreement with the expected value of the initial slope on the Hugoniot19 given by (B0 + 1)/4, where B0 is the
275

LETTERS
a
30 2,000

25 Us (km s 1)

Quartz data Fit to quartz data Fused silica data Our data Fit to silica data
Pressure (GPa)

1,500

With transit time With VISAR Ref. 6 Liquid

20

1,000

15

500 10 Diamond 5 5 10 Up (km s 1) 15 20 0 4 6 8 Density (g cm3) 10

28 26 24 Us (km s 1) 22 20 18 16 14 0 5 10 Up (km s 1) 15

With transit time With VISAR Ref. 6

Figure 3 Hugoniot measurements of diamond in the (density, pressure) plane. The same symbols as in Fig. 2b are used to represent the data. The error bars are propagated from the shock-velocity uncertainties in quartz and in diamond. The dotted and dashed curves are the Hugoniot curves calculated with the linear ts of U s versus Up in the diamond phase and in the uid phase, respectively. Melting should take place over the pressure range indicated by the hatched region.

Figure 2 Hugoniot measurements in the (Up , Us ) plane. a, Fused silica. The present data with error bars are compared with lower-pressure data for fused silica18 and with quartz measurements15 . b, Diamond. Our data are represented by two different symbols to differentiate the VISAR measurements of run 1 from the transit-time measurements of run 2. They are compared with previous measurements of the Hugoniot of diamond6 as squares. The dotted and dashed lines are ts of the data in the diamond phase below 10 km s1 and in the uid phase above, respectively. The error bars indicate uncertainty in the velocity measurement for Us and for Up the propagation of the uncertainty in the shock velocity measurement in quartz.

pressure derivative of the bulk modulus recently measured equal to 3.0 (ref. 20). A clear discontinuity in the Us Up evolution is observed above Up 10 km s1 . Such a change in the Hugoniot should very reasonably be ascribed to melting. The Us (Up ) data above 10 km s1 are tted by the linear relation, Us = 1.25Up + 7.03. Given the small number of points and their error bars, there is an uncertainty of 0.15 on the value of the slope. Yet, the slope is there more in agreement with the measured slope of the Hugoniot of elemental metals at high pressure (for particle velocities typically greater than 10 km s1 ), which was explained to be approximately equal to 1.2 (ref. 20). Consequently, the change in the Us (Up ) slope at about 10 km s1 could reect the change from the covalent diamond phase into a more compact carbon uid. Ab initio calculations24 have found that the character of liquid carbon changes from about fourfold to about sixfold coordination in the pressure range 4001,000 GPa. Below, the observation of a signicant negative volume discontinuity along the Hugoniot should be interpreted as experimental evidence for a strong increase of coordination at melting. It has even been speculated that the
276

increase of coordination in the uid could be discontinuous at a rst-order uiduid phase transition2 , but this is beyond the scope of the present study. Us Up measurements are converted to pressure and density through the RankineHugoniot equations, which are derived from the conservation of mass, momentum and energy across the shock front. The corresponding data are shown in Fig. 3. Melting is expected to occur in between our highest measured point on the Hugoniot of diamond, 720 GPa, and the lowest measured point on the Hugoniot of the uid, 900 GPa. A striking feature of Fig. 3 is that melting is associated with a negative volume discontinuity. By extrapolating the Hugoniot curves of the solid and the uid and taking into account the uncertainties, the dierence in volume at about 750 GPa between these two Hugoniot curves is estimated to be in between 5% and 14%. It is interesting to note that the Gibbs free energy model of carbon5 predicts a very similar volume discontinuity on the Hugoniot curve at melting, with V = 9.7%, yet at a lower pressure, 600 GPa. Although a correct prediction of melting along the Hugoniot is probably beyond the scope of this model because it does not take into account a possible increased coordination in the uid phase, the model is used here to obtain further insights into the data. In this model, the large volume discontinuity spreads the melting transition along the Hugoniot over 230 GPa. As we observe a similar V , a similar extension of melting is expected, and it should take place between 720 and 900 GPa. Consequently, with safe error bars, we can state that melting starts at 750 50 GPa on the Hugoniot of diamond. Also, the volume change on the Hugoniot associated with melting stems from two contributions: the intrinsic volume discontinuity at melting and the volume reduction due to the cooling eect along the Hugoniot associated with the enthalpy of melting. From the Gibbs free energy model5 , the temperature cooling due to the completion of melting along the Hugoniot should contribute to 1.5% in the V . This is certainly smaller than the observed volume variation between the solid Hugoniot and the uid Hugoniot. Consequently, the melting of diamond at 750 GPa is associated with a signicant negative volume discontinuity, greater than 3%. In most systems, a positive volume discontinuity is generally observed at melting.
nature materials VOL 6 APRIL 2007 www.nature.com/naturematerials

LETTERS
16 14 12 T (K) ( 103) 10 8 6 4 2 0 200 400 600 800 1,000 Pressure (GPa) Uranus Diamond Fluid Melting ab initio DFT (refs 24) Melting (ref. 11) Melting (ref. 5) Hugoniot

Figure 4 Melting of diamond along its Hugoniot. The melting curves of various calculations25,11 are compared. The melting of shocked diamond should occur at the intersection of these melting lines and the Hugoniot of diamond. The experimental determination of the beginning of melting on the Hugoniot is represented by a bold segment because of the pressure uncertainty. The isentrope of Uranus21 is entirely within the stability domain of the diamond phase.

missing the change of coordination number in the uid at very high pressure. Finally, the melting curve given by the thermodynamical model5 gives a too low melting pressure but a reasonable V . This model could now be extended by adding a new phase of liquid carbon with high coordination and by adjusting its properties on the present Hugoniot data. Finally, in Fig. 4, it is observed that the isentrope of Uranus21 does not cross the melting line of diamond. Hence, as has been claimed previously9 , diamond, which is believed to precipitate after the pressure-induced dissociation of CH4 (ref. 22), might be stable all the way down the planets centre. We report here the rst quantitative data on the melting curve of diamond at very high pressure. The evidence for a negative melting slope implies that the melting curve of diamond has a maximum, probably in the 300500 GPa pressure range. A parallel can be drawn with graphite, which also has a maximum on the melting curve23 at (5.6 GPa, 4,790 K). The existence of this maximum has been explained by the change of coordination in the uid24 , going from a two- and threefold coordination at low pressure to tetrahedral, through the stabilization of the diamondlike bonding at high pressure. The existence of the maximum on the melting curve of diamond is also the consequence of a change of the coordination number in the uid, but here due to the destabilization of the diamond-like bonding towards a more compact uid24 .
Received 11 July 2006; accepted 26 January 2007; published 25 March 2007.

According to the Clapeyron equation, a solid with a larger specic volume than the uid requires the melting slope, dTm /dP , to be negative. Because the melting slope of diamond is positive at low pressure, above 20 GPa, and negative around 750 GPa, this implies the existence of a maximum in between these two pressures on the melting curve of diamond. The prediction of the maximum on the melting curve of diamond was explained by a gradual increase of the coordination number in the uid phase from 4 to 5 and 6 (refs 24). The measured negative V could thus be a quantication of the loss of the sp3 bonding. Predictions on the basis of empirical potentials10,11 predict a positive discontinuity because they cannot reproduce the change in the electronic conguration. The ab initio calculations24 that explicitly compute this change of electronic conguration predict a too small (1%), albeit negative, discontinuity. Therefore, the ab initio calculations might underestimate the change of the electronic conguration at melting. The thermodynamic description5 seems to predict a reasonable volume discontinuity at melting. However, in this model, the uid is assumed to be of diamond type coordination and the smaller volume of the liquid is due to a pressure derivative of the bulk modulus lower for the liquid than for the diamond phase. We used a diagnostic collecting light emitted by the shocked sample to measure a black-body temperature. Unfortunately, this diagnostic is not sensitive enough for temperatures below 1 eV. Because we did not observe any pyrometric signal, we can only say that the melting temperature of diamond occurs below 10,000 K. The melting temperature is thus estimated from the Sesame EoS of diamond, that is T = 8,500 K at 750 GPa. In Fig. 4, the present melting determination is plotted as a bold segment to take account of the 50 GPa uncertainty and it is compared with various calculated melting curves. The melting pressure along the Hugoniot predicted by the ab initio calculations24 is in good agreement with experiment. The present experimental uncertainty is of the same order of magnitude as the numerical errors, which is given by the dispersion between the three calculations. However, it has been discussed above, from the measured volume discontinuity at melting, that their melting slopes are predicted to be too at. The melting curves predicted within the chemical approaches10,11 do not show re-entrant high-pressure behaviour. They are probably
nature materials VOL 6 APRIL 2007 www.nature.com/naturematerials

References
1. Bundy, F. P. et al. The pressuretemperature phase and transformation diagram for carbon updated through 1994. Carbon 34, 141153 (1996). 2. Grumbach, M. P. & Martin, R. M. Phase diagram of carbon at high pressures and temperatures. Phys. Rev. B 54, 1573015741 (1996). 3. Wang, X., Scandolo, S. & Car, R. Carbon phase diagram from ab initio molecular dynamics. Phys. Rev. Lett. 95, 18570 (2005). 4. Correa, A. F., Bonev, S. A. & Galli, G. Carbon under extreme conditions: Phase boundaries and electronic properties from rst-principle theory. Proc. Natl Acad. Sci. 103, 12041208 (2006). 5. Fried, L. E. & Howard, W. M. Explicit Gibbs free energy equation of state applied to the carbon phase diagram. Phys. Rev. B 61, 87348743 (2000). 6. Pavlovskii, M. N. Shock compression of diamond. Sov. Phys. - Solid State 13, 741742 (1971). 7. Bradley, D. K. et al. Shock compressing diamond to a conducting uid. Phys. Rev. Lett. 93, 195506 (2004). 8. Nagao, H. et al. Hugoniot measurement of diamond under laser shock compression up to 2 TPa. Phys. Plasma 13, 052705 (2006). 9. Ross, M. The ice layer in Uranus and Neptunediamonds in the sky? Nature 292, 435436 (1981). 10. Glosli, J. N. & Ree, F. H. The melting line of diamond determined via atomistic computer simulations. J. Chem. Phys. 110, 441446 (1999). 11. Ghiringhelli, L. M., Los, J. H., Meijer, E. J., Fasolino, A. & Frenkel, D. Modeling the phase diagram of carbon. Phys. Rev. Lett. 94, 145701 (2005). 12. Barker, L. M. & Hollenbach, R. E. Laser interferometer for measuring high velocities of any reecting surface. J. Appl. Phys. 43, 46694675 (1972). 13. Edwards, D. F. & Philipp, H. R. Handbook of Optical Constants of Solids 665673 (Academic, New York, 1985). 14. Celliers, P. et al. Line imaging velocimeter for shock diagnostics at the OMEGA laser facility. Rev. Sci. Instrum. 75, 49164929 (2004). 15. Hicks, D. G. et al. Shock compression of quartz in the high-pressure uid regime. Phys. Plasma 12, 082702 (2005). 16. Enig, J. A complete E, P, V, T, S thermodynamic description of metals based on the P,U mirror-image approximation. J. Appl. Phys. 34, 746754 (1962). 17. Trunin, R. F. Shock compressibility of condensed materials in strong shock wave generated by underground nuclear explosions. Phys. Uspekhi 37, 11231145 (1994). 18. Marsh, S. P. LASL Shock Hugoniot Data (Univ. California Press, Berkeley, 1980). 19. Johnson, J. D. in Shock Compression Condensed Matter1997, CP429 (eds Scmidt S. C., Dandekar D. P. & Forbes J. W.) 2730 (American Institute of Physics, New York, 1998). 20. Occelli, F., Loubeyre, P. & LeToullec, R. Properties of diamond under hydrostatic pressures up to 140 GPa. Nature Mater. 2, 151153 (2003). 21. Hubbard, W. B., Podolak, M. & Stevenson, D. J. in The Interior of Neptune (ed. Cruikshank, D. P.) 109138 (Univ. of Arizona Press, Tucson, Arizona, 1995). 22. Benedetti, L. R. et al. Dissociation of CH4 at high pressures and temperatures: Diamond formation in giant planet interiors? Science 286, 100102 (1999). 23. Togaya, M. Pressure dependences of the melting temperature of graphite and the electrical resistivity of liquid carbon. Phys. Rev. Lett. 79, 24742477 (1997). 24. Morris, J. R., Wang, C. Z. & Ho, K. M. Relationship between structure and conductivity in liquid carbon. Phys. Rev. B 52, 41384145 (1995).

Acknowledgements
The authors gratefully acknowledge the technical support of the LULI facility. Correspondence and requests for materials should be addressed to P.L.

Competing nancial interests

The authors declare no competing nancial interests.
Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

277