Proceedings article of the 5th Conference on Cryocrystals and Quantum Crystals in Wroclaw, Poland,

submitted to J. Low. Temp. Phys. (2004).

Hydrogen-Helium Mixtures at High Pressure

Burkhard Militzer

Geophysical Laboratory, Carnegie Institution of Washington,

5251 Broad Branch Road, NW, Washington, DC 20015, USA

The properties of hydrogen-helium mixtures at high pressure are crucial to

address important questions about the interior of Giant planets e.g. whether

Jupiter has a rocky core and did it emerge via core accretion? Using path

integral Monte Carlo simulations, we study the properties of these mixtures

as a function of temperature, density and composition. The equation of

state is calculated and compared to chemical models. We probe the accuracy

of the ideal mixing approximation commonly used in such models. Finally,

we discuss the structure of the liquid in terms of pair correlation functions.

PACS numbers: 62.50.+p, 02.70.Lq, 64.30.+t

1. INTRODUCTION

Hydrogen and helium are the two most abundant elements in giant plan-ets. While Jupiter and Saturn are well characterized on the surface, manybasic questions about its interior have not been answered, e.g., Jupiter’scomposition has been measured in situ on the surface (H 73.6% by weight,He 24.9%, 0.015% heavier elements) but the amount of heavier elements inthe interior is not well constrained. So far one has also not been able to an-swer the fundamental question whether Jupiter has a core and did it emergevia core accretion? Alternatively, it could have been formed by gravitationalinstability of the hydrogen cloud in a similar way stars form.

Since there is no direct way to detect a core, one must refer to models forthe planet interior. Such models are constrained by the available observationdata, in particular, the properties at planet surface and the gravitational mo-ments measured through fly-by trajectories. All models rely on an equationof state (EOS) of hydrogen-helium mixtures. However, the uncertainties inthe available EOS are large and have, among many other questions, notallowed one to determine whether Jupiter has a core.

B. Militzer

In this article, we present results from path integral Monte Carlo simu-lations that enable us to study quantum many-body systems at finite tem-perature from first principles. In this simulation, hydrogen-helium mixturesare represented by an ensemble of electrons, protons and helium nuclei, eachdescribed by a path in imaginary time to incorporate quantum effects. Theelectrons are treated as fermions while exchange effects for the nuclei can beneglected for the considered thermodynamics conditions.

2. PATH INTEGRAL MONTE CARLO

The thermodynamic properties of a many-body quantum system at fi-nite temperature can be computed by averaging over the density matrix,

ρ = e−βH , β = 1/kbT . Path integral Monte Carlo (PIMC) is based on theidentity,

e−βH =[

e−βM

H]M

(1)

where M is a positive integer. Insertion of complete sets of states betweenthe M factors leads to the usual imaginary time path integral formulation,written here in real space,

ρ(R,R′;β) =

∫

. . .

∫

dR1 . . . dRM−1 ρ(R,R1; τ) . . . ρ(RM−1,R′; τ) (2)

where τ = β/M is the time step. Each of the M steps in the path now hasa high temperature density matrix ρ(Rk,Rk+1; τ) associated with it. Theintegrals are evaluated by Monte Carlo methods. In the results presentedhere, the high temperature density matrix was taken as a product of exactpair density matrices,

ρ(R,R′; τ)

ρ0(R,R′; τ)=

⟨

e−∫ τ

0dt∑

i<jV (rij)

⟩

R→R′

=

⟨

∏

i<j

e−∫ τ

0dtV (rij)

⟩

R→R′

(3)

≈∏

i<j

⟨

e−∫ τ

0dtV (rij)

⟩

rij→r′ij

≡ e−∑

i<ju(rij ,r

′ij ;τ)

, (4)

where ρ0(R,R′; τ) is the free particle density matrix and u(rij , r′ij ; τ) is the

pair action for paths initially separated by rij and finally at time τ by r′ij . Anapproximation is introduced by assuming that the different pair interactionscan be averaged by independent Brownian random walks that are denotedby brackets 〈. . .〉. This approach is efficient but not exact, and thereforeputs a limit on the imaginary time step τ in many-body simulations. Thepair action, u, can be computed by different methods.1–3

Hydrogen-Helium Mixtures at High Pressure

The density matrix for bosonic and fermionic systems can be obtainedby projecting out states of corresponding symmetry. In PIMC, one sums updifferent permutations P ,

ρb/f(R,R′;β) =1

N !

∑

P

(±1)Pρd(R,PR′;β) =1

N !

∑

P

(±1)P∫

R→PR′

dRt e−U [Rt].

(5)For bosons, this is essentially an exact numerical algorithm that has foundmany applications.4 For fermions, the cancellation of positive and negativecontributions leads to numerically unstable methods, which is known as thefermion sign problem. Ceperley showed that this problem can be solved byintroducing the restricted path approximation,5,6

ρf (R0,R′;β) ≈

1

N !

∑

P

(−1)P∫

R0 → PR′

ρT (R(t),R0; t) > 0

dRt e−U [R(t)] , (6)

where one only samples path R(t) that stay within the positive region ofa trial density matrix, ρT (R(t),R0; t) > 0. This procedure leads to anefficient algorithm for fermionic systems. All negative contributions to di-agonal matrix elements are eliminated.7 Contrary to the bosonic case, thealgorithm is no longer exact since it now depends on the approximations forthe trial density matrix. However, the method has worked very well in manyapplications.8,9 For ρT , one can use the free particle or a variational densitymatrix.10

The following simulation results were derived with Ne = 32 electronsand the corresponding number of protons, Np, and helium nuclei NHe to ob-tain a neutral system (Ne = Np+2NHe) with helium fraction x ≡ 2NHe/Ne.Periodic boundary conditions are applied. As imaginary time discretization,we employ τ = 0.079. (Atomic units of Bohr radii and Hartrees will be usedthroughout this article.)

The electrons are treated as fermions with fixed spin. We use variationalnodes to restrict the paths and have therefore extended the approach inRef. 10 to mixtures. The nuclei are also treated as paths but their exchangeeffects are not relevant here.

3. PHASE DIAGRAM OF HOT DENSE HYDROGEN

Figure 1 shows the high temperature phase diagram of dense hydro-gen beginning with the fluid and reaching up to a highly ionized plasmastate. The figure includes the isentropes for Jupiter and low mass stars12

B. Militzer

10−8

10−6

10−4

10−2

100

102

ρ (g/cc)

102

103

104

105

106

107

T (K)

Metallic Fluid

Plasma

Star 15 M sun

Atomic Fluid

Sun

Star 0.3 Msun

Jupiter

PIMC

Hu

go

nio

t

ICF

path

Molecular Fluid

100 10 1rs

Fig. 1. Density-temperature phase diagram of hot dense hydrogen. The bluedash-dotted lines separate the molecular, the atomic, the metallic, and theplasma regime. The green solid lines are isentropes for Jupiter and starswith 0.3, 1, and 15 solar masses. Single shock Hugoniot states as well as theinertial confinement fusion path11 are indicated by dashed lines. The thinsolid line show ρ-T conditions of PIMC simulations.

and indicates the thermodynamic conditions, at which PIMC simulationshave been applied.13,14,9 We are now going to use these simulation results tocharacterize hot dense hydrogen from a path integral perspective.

At low density and temperature, one finds a fluid of interacting hydrogenmolecules. A PIMC snapshot for T = 5000K and rs = 4.0 is shown in Fig. 2.The proton paths are very localized due to their high mass. Their spread canbe estimated from the de Broglie thermal wavelength, λ2

d = h2β/2πm. Theelectron paths are more spread out but they are localized to some extent sincetwo electrons of opposite spin establish the chemical bond in the hydrogenmolecule.

If the temperature is raised from 5000 to 106 K, hydrogen undergoesa smooth transition from a molecular fluid though an atomic regime andfinally to a two-component plasma of interacting electrons and protons.Many-body simulations at even higher temperatures are not needed since

Hydrogen-Helium Mixtures at High Pressure

Fig. 2. Snapshot from PIMC simulations of pure hydrogen in the molecularregime at T = 5000K and rs = 4.0. The pink spheres denote the protons.The bonds (white lines) were added as a guide to the eye. The electronpaths are shown in red and blue [light and dark gray] depending on theirspin state.

analytical methods like the Debye-Huckel theory work well. PIMC with ex-plicit treatment of the electrons can also be applied to temperatures below5 000K but groundstate methods are then more practical since excitationbecome less relevant and the computational cost scale with the length ofpaths like T−1.

A detailed analysis of the chemical species present in the low densityregime (rs ≥ 2.6) is given in Ref. 13. With decreasing density, one findsthat the degree of molecular dissociation increases since the atomic statehas higher entropy. For the same reason, one observes that the degree ofatomic ionization increases with decreasing density. All these low-densityeffects can be well characterized by analytical models based on approximatefree energy expressions for atoms, molecules and ionized particles.15–18 Inthis regime, one also finds reasonably good agreement between for the EOSderived from PIMC and chemical models.13

If the density is increased at T = 5000K, one finds an intermediateregime of strongly interacting molecules (Fig. 3). Some electron paths ex-change with neighboring molecules indicating the importance of fermionic

B. Militzer

Fig. 3. Deuterium in the dense molecular regime (T = 5000K and rS =1.86) Due to the density increase compared to Fig. 2 (see details there),the electron paths permute with a rising probability (shown as yellow [lightgray] lines) but are still localized enough to form a bond between the twoprotons in the molecule. The electron density average over many electronconfigurations is indicated in gray color on the blue rectangles.

effects. However, the electrons are still localized enough to provide a suffi-cient binding force for the protons.

If the density is increased further from rs = 1.86 to 1.60, this bindingforce is lost due to further delocalization of the electrons (Fig. 4). Almostall electrons are now involved in long exchange cycles indicating a highlyconducting, metallic state. No binding forces of the protons can be observed.If the density is increased further, the electrons form a rigid neutralizingbackground and one recovers the limit of a one-component plasma of protons.If the temperature is increased, the electron paths get shorter and shorter,the degree of electron degeneracy decreases gradually and one recovers thelimit of a two-component plasma at high temperature.

The nature of the molecular-metallic transition has not yet been deter-mined with certainty. A large number of analytical models19 predict a firstorder transition, others do not.16 PIMC simulation with free particle nodesby W. Magro et al.20 showed an abrupt transition characterized by a region

Hydrogen-Helium Mixtures at High Pressure

Fig. 4. Deuterium in the metallic regime characterized by unpaired protonsand a gas of degenerate electrons. The snapshot was taken from a PIMCsimulation at T = 6250K and rs = 1.60. The electron paths are delocalizedand exchange frequently (see description of Fig. 3).

dPdT

∣

∣

∣

V< 0 for rs=1.86. Later work by B. Militzer21 using more accurate

variational nodes did not show such a region for rs=1.86. Whether PIMCsimulations with variational nodes predict a gradual molecular-metallic tran-sition for all temperatures, or if the region of an abrupt transition has shiftedto lower, not yet accessible, temperatures has not been determined. How-ever, recent density functional molecular dynamics simulations by variousauthors predict a first order transition below 5 000K.22–25

There was a lot of recent interest focused on the hydrogen EOS becauseof the unexpectedly high compressibility inferred from the laser-driven shockwave experiment by Da Silva et al.26 and Collins et al.27 using the Novalaser at Lawrence Livermore National Laboratory. The measurements indi-cated that hydrogen could be compressed to about 6-fold the initial densityrather than 4-fold as indicated by the Sesame model.28,29 Even though thetemperatures and pressures reached in shock experiments are higher thanthose in the giant planets these measurements are crucial here since theyrepresent the only way to distinguish between different EOS models at hightemperature.

B. Militzer

The Nova results challenged the existing understanding of high P-Thydrogen and triggered many new experimental and theoretical efforts. Dif-ferent chemical models gave rise to very different predictions19,16,30 rangingfrom 4-fold compression as suggested in28,29 to 6-fold as predicted by theRoss model.31 While the accuracy of chemical models did not allow anyconclusive predictions, first principle simulations from PIMC13,32 as wellas from density functional molecular dynamics33,24,25 consistently predict alower compressibility of about 4.3.

Since then there have been many attempts to resolve this discrepancybut the most important contributions came from new experiments by Knud-son et al.34–36 at Sandia National Laboratory. Instead of a laser drive, theyused magnetically driven shock waves in combination with bigger samples.They found a significantly lower compressibility quite similar to predictionsfrom first principles methods. The new results are also supported by athird set of experiments by Russian investigators using spherically converg-ing shock waves.37,38

4. HYDROGEN-HELIUM MIXTURES

Due to importance for astrophysical applications, EOS models for hy-drogen-helium mixtures have been studied for quite some time. The mostwidely used EOS was derived by Saumon, Chabrier and van Horn (SCH).12

Like previous chemical models it is based on a hydrogen and a helium EOSthat combined using an ideal mixing rule.

Following the discussion of the hydrogen EOS above, we are now ana-lyzing the second ingredient: the EOS of helium. Given its low abundancein giant planets, helium is not expected to be present in very high concentra-tion. However, Stevenson39 has proposed that the hydrogen-helium mixturecould phase separate under certain high pressure conditions. As a result,helium droplets would form and fall as rain towards planet core. The associ-ated release of latent heat would delay the cooling process of the planet. Thiswas suggested as one possible explanation of why standard models predictSaturn to cool at a much faster rate than is observed.40 In order to detecthydrogen-helium phase separation one needs an EOS of pure helium that wewill now discuss.

Figure 5 shows an EOS comparison of our PIMC calculations with theSCH model. One finds that the pressure is underestimated by SCH and thatthe deviations increase with density. The internal energy at high tempera-ture is also underestimated. The analysis suggests a more careful treatmentof the helium ionization states is needed to improve the chemical model.

Hydrogen-Helium Mixtures at High Pressure

0 100 200 300 400 500T (1000 K)

0

1000

2000P

ress

ure

(GP

a)Saumon & ChabrierPIMC

rS=1.6

rS=2.0

rS=2.6

0 100 200 300 400 500T (1000 K)

−40

0

40

80

Inte

rnal

ene

rgy

(eV

per

ele

ctro

n) Saumon & ChabrierPIMC

rS=2.6

rS=1.6

Fig. 5. Equation of state comparison for pure helium showing the pressure(left) and the internal energy (right) for different density as derived frompath integral Monte Carlo simulations and the chemical model by Saumon,Chabrier, and van Horn.12

To conclude the comparison with chemical models, we test one more as-sumption that is generally made: the ideal mixing hypothesis. We performeda large number of PIMC calculations of fully interacting hydrogen-heliummixtures at a fixed density of rs = 1.86 for various temperatures and mixingratios. The resulting correction to ideal mixing is given by,

∆fmix = f(x)− (1− x) f(x = 0)− x f(x = 1) . (7)

Figure 6 shows that the corrections to pressure, ∆Pmix, increase withdecreasing temperature reaching 10% for 15 625K. In the considered temper-ature interval, the corrections to the internal energy are largest for 31 250Kwhich can be explained by the different ionization states of the two fluidsthat lead to larger error if ideal mixing is assumed.

Finally, we discuss pair correlation functions, g(r), for hydrogen-heliummixtures,

g(r) ≡Ω

N2

⟨

∑

i6=j

δ(r− rij)

⟩

. (8)

g(r) functions are a standard tool to characterize the short-range correlationof particles in liquids. In Fig. 7, we compare the correlation functions forpure hydrogen, a x=50% mixture, and pure helium at fixed temperature(15 625K) and density (rs = 1.86). Changes in the pair correlations arenow discussed as a function of helium concentration. The electron-protong(r) shows a strong peak at the origin due to the Coulomb attraction. Thepeak is not affected when 50% helium is added to a pure hydrogen sample.

B. Militzer

0 0.2 0.4 0.6 0.8 1Helium fraction x

0.00

0.05

0.10

0.15∆P

MIX

/ P

15625 K31250 K62500 K

0 0.2 0.4 0.6 0.8 1Helium fraction x

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

∆EM

IX (

eV)

15625 K31250 K62500 K

Fig. 6. Excess mixing pressure ∆Pmix and internal energy per electron ∆Emix

are shown for a hydrogen-helium mixture at rs = 1.86. Path integral MonteCarlo results are shown for several temperatures. The mixing is performedat constant density. ∆Emix is largest for T=31 250K because at this tem-perature the two end members are characterized by very different degrees ofionization.

Similarly, the correlation between electrons and helium nuclei is not alteredby the presence of protons.

The electron-electron correlation depends strongly on spin. For pairswith parallel spin, exchange effects lead to a strong repulsion. The resultingg(r) function does not depend much on whether helium is present or not. Thecorrelation function between pairs of antiparallel spins, on the other hand,is strongly affected by the presence of helium nuclei. In the helium atom,two electrons with opposite spin are attracted to the core, which indirectlyleads to the observed increase in the electron-electron g(r).

It is interesting to note that proton-proton g(r) changes with the heliumconcentration but the correlation of helium nuclei does not. With increasingpresence of helium, a peak in the proton-proton g(r) appears at r = 1.4 a0

which indicates the formation of H2 molecules. Adding helium nuclei leadsto the localization of a fraction of the electrons. The available space incombination with the reduced electronic exchange effects then leads to theformation of hydrogen molecules which are not present in pure hydrogen atthe same temperature and density.

To summarize, we discussed the phase diagram of hot dense hydrogenand presented the first PIMC results for hydrogen-helium mixtures. We an-alyzed the accuracy of existing helium EOS models as well as the commonlyused ideal mixing rule. Further work will include extending those calcula-tions to lower temperature and other densities in order to make construct

Hydrogen-Helium Mixtures at High Pressure

0 1 2 3 4 5r (a0)

0

2

4

6

8

10

g(r)

0

1

2

g(r)

0 1 2 3 4 5r (a0)

0

20

40

60

80

1000

1

2

0 1 2 3 4 5r (a0)

0

1

20

2

4

6

8

Pure hydrogen50% H−He mixturePure Helium

pp

pe

αα

αe ee ||

ee ||

T=15625Kρ=0.42 g/ccrS=1.86

p ... protonse ... electronsα ... helium nuclei

Fig. 7. Comparison of different pair correlation functions, g(r), from threePIMC simulations at rs = 1.86 and 15 625K: pure hydrogen (red dottedlines), a x = 50% mixture (black dashed lines) and pure helium (blue solidlines). The six graphs show g(r) for different pairs of protons (p), electrons(e) and helium nuclei (α). For electrons, pairs with antiparallel (upper right)and with parallel (lower right) spins are distinguished.

an accurate EOS for astrophysical applications.The author thanks W. Hubbard and D. Stevenson for useful discussions

and D. Saumon for providing us with his equation of state.

REFERENCES

1. R. G. Storer. J. Math. Phys., 9:964, 1968.2. E. L. Pollock. Comp. Phys. Comm., 52 :49, 1988.3. K. E. Schmidt and M. A. Lee. Phys. Rev. E, 51:5495, 1995.4. D. M. Ceperley. Rev. Mod. Phys., 67:279, 1995.5. D. M. Ceperley. J. Stat. Phys., 63:1237, 1991.6. D. M. Ceperley. In Ed. K. Binder and G. Ciccotti, editors, Monte Carlo

and Molecular Dynamics of Condensed Matter Systems. Editrice Compositori,Bologna, Italy, 1996.

7. B. Militzer, E.L. Pollock, and D. Ceperley, 2004.8. C. Pierleoni, D.M. Ceperley, B. Bernu, and W.R. Magro. Phys. Rev. Lett.,

73:2145, 1994.

B. Militzer

9. B. Militzer and E. L. Pollock. Phys. Rev. Lett., 89:280401, 2002.10. B. Militzer and E. L. Pollock. Phys. Rev. E, 61:3470, 2000.11. J. Lindl. Phys. Plasmas, 2 (11):3933, 1995.12. D. Saumon, G. Chabrier, and H. M. Van Horn. Astrophys. J., 99:713, 1995.13. B. Militzer and D. M. Ceperley. Phys. Rev. Lett., 85:1890, 2000.14. B. Militzer and D. M. Ceperley. Phys. Rev. E, 63:066404, 2001.15. D. Saumon and G. Chabrier. Phys. Rev. A, 46:2084, 1992.16. F.J. Rogers. Astrophys. J., 352:689, 1990.17. W. Ebeling, W.D. Kraeft, and D. Kremp. Theory of Bound States and Ionisa-

tion Equilibrium in Plasma and Solids. Ergebnisse der Plasmaphysik und derGaselektronik, Band 5. Akademie-Verlag, Berlin, 1976.

18. H. Juranek, R. Redmer, and Y. Rosenfeld. J. Chem. Phys., 117:1768, 2002.19. D. Beule, Werner Ebeling, and Andreas Forster. Phys. Rev. B, 59:14177, 1999.20. W. R. Magro, D. M. Ceperley, C. Pierleoni, and B. Bernu. Phys. Rev. Lett.,

76:1240, 1996.21. B. Militzer. PhD thesis, University of Illinois at Urbana-Champaign, 2000.22. S. Banier et al. Phys. Rev.E, 63:015301, 2000.23. S. Scandolo et al. PNAS, 100:3051, 2003.24. M. P. Desjarlais. Phys. Rev. B, 68:064204, 2003.25. G. Galli S. A. Bonev, B. Militzer. Phys. Rev. B, 69:014101, 2004.26. I. B. Da Silva, P. Celliers, G. W. Collins, K. S. Budil, N. C. Holmes, W. T. Jr.

Barbee, B. A. Hammel, J. D. Kilkenny, R. J. Wallace, M. Ross, R. Cauble,A. Ng, and G. Chiu. Phys. Rev. Lett., 78:783, 1997.

27. G. W. Collins, L. B. Da Silva, P. Celliers, D. M. Gold, M. E. Foord, R. J.Wallace, A. Ng, S. V. Weber, K. S. Budil, and R. Cauble. Science, 281:1178,1998.

28. G. I. Kerley. In J. M. Haili and G. A. Mansoori, editors, Molecular Based Study

of Fluids, page 107. ACS, Washington DC, 1983.29. G. I. Kerley. Equations of state of hydrogen and deuterium, 2003.30. H. Juranek and R. Redmer. J. Chem. Phys., 112:3780, 2000.31. M. Ross. Phys. Rev. B, 58:669, 1998.32. B. Militzer, D. M. Ceperley, J. D. Kress, J. D. Johnson, L. A. Collins, and

S. Mazevet. Phys. Rev. Lett., 87:275502, 2001.33. T. J. Lenosky, S. R. Bickham, J. D. Kress, and L. A. Collins. Phys. Rev. B,

61:1, 2000.34. M. D. Knudson, D. L. Hanson, J. E. Bailey, C. A. Hall, J. R. Asay, and W. W.

Anderson. Phys. Rev. Lett., 87:225501, 2001.35. M. D. Knudson, D. L. Hanson, J. E. Bailey, C. A. Hall, and J. R. Asay. Phys.

Rev. Lett., 90:035505, 2003.36. M. D. Knudson, D. L. Hanson, J. E. Bailey, C. A. Hall, J. R. Asay, and

C. Deeney, 2004. submitted to Phys. Rev. B.37. S. I. Belov et al. Sov. Phys. JETP Lett., 76:433, 2002.38. G. V. Boriskov et al., 2003. submitted to Phys. Rev. B.39. D.J. Stevenson and E.E Salpeter. Astrophys. J. Suppl., 35:239, 1977.40. J. J. Fortney and W. B. Hubbard. Astrophys. J., 2004. In press.