Copyright © 2007 by The National Academy of Sciences of the USA Physics Superfluid transition of homogeneous and trapped two-dimensional Bose gases §To whom correspondence should be addressed. E-mail:
gbaym/at/uiuc.edu Contributed by Gordon Baym, November 15, 2006. Author contributions: M.H., G.B., J.-P.B., and F.L. performed research and wrote the paper. Received August 29, 2006. | |||||||||||||||||||||||||||||||||||||||||
Abstract Current experiments on atomic gases in highly anisotropic traps present the opportunity to study in detail the low temperature phases of two-dimensional inhomogeneous systems. Although, in an ideal gas, the trapping potential favors Bose–Einstein condensation at finite temperature, interactions tend to destabilize the condensate, leading to a superfluid Kosterlitz–Thouless–Berezinskii phase with a finite superfluid mass density but no long-range order, as in homogeneous fluids. The transition in homogeneous systems is conveniently described in terms of dissociation of topological defects (vortex–antivortex pairs). However, trapped two-dimensional gases are more directly approached by generalizing the microscopic theory of the homogeneous gas. In this paper, we first derive, via a diagrammatic expansion, the scaling structure near the phase transition in a homogeneous system, and then study the effects of a trapping potential in the local density approximation. We find that a weakly interacting trapped gas undergoes a Kosterlitz–Thouless–Berezinskii transition from the normal state at a temperature slightly below the Bose–Einstein transition temperature of the ideal gas. The characteristic finite superfluid mass density of a homogeneous system just below the transition becomes strongly suppressed in a trapped gas. Keywords: two dimensions, phase transitions, trapped atoms | |||||||||||||||||||||||||||||||||||||||||
The ability to produce two-dimensional atomic gases trapped in optical potentials has stimulated considerable interest in the Kosterlitz–Thouless–Berezinskii (KTB) transition in such systems (1–7); recently, the transition has been observed in a quasi two-dimensional system of trapped rubidium atoms (8, 9). A homogeneous Bose gas in two dimensions undergoes Bose–Einstein condensation (BEC) only at zero temperature, since long wavelength phase fluctuations destroy long range order (10–13); nonetheless, interparticle interactions drive a phase transition to a superfluid state at finite temperature, as first pointed out by Berezinskii (14, 15) and by Kosterlitz and Thouless (16, 17). The phase transition is characterized by an algebraic decay of the off-diagonal one-body density matrix (or single particle Green's function) in real space below the transition temperature, TKT. Furthermore, the superfluid mass density, ρs, jumps with falling temperature, T, from 0 just above TKT to a universal value ρs = 2m2 TKT/π just below (18), where m is the atomic mass. (We use units ħ = kB = 1.) A noninteracting homogeneous Bose gas in two dimensions does not undergo Bose–Einstein condensation at finite temperature. In such a system, the density is given in terms of the chemical potential μ by The question of whether an interacting trapped Bose gas when cooled from the normal state first undergoes an ordinary BEC or a KTB transition is not completely settled (20). In this paper, we show that in the thermodynamic limit, in which the trap frequency, ω, vanishes and the total particle number N grows, with the product Nω2 and hence the density at the origin, remaining constant, interactions at the mean-field level destroy ordinary BEC; instead, the system undergoes a KTB transition at a temperature slightly below the ideal condensation temperature, TBEC. We approach the phase transition of the homogeneous two-dimensional Bose gas by carrying out a scaling analysis similar to the one we used earlier to discuss the phase transition in a dilute three-dimensional homogeneous Bose gas (21–29). The phase below TKT is characterized by an algebraic decay in space of the single particle Green's function, G. As in three dimensions, the phase transition occurs when the single-particle spectrum in Fourier space becomes gapless, G−1(0, 0) = 0, where G(k, zν) is the Fourier component of G in space and (imaginary) time. Below, we first discuss the scaling structure of G as TKT is approached from above, and rederive the relation between the temperature and density at the transition Although the KTB transition was initially discussed in terms of unbinding of vortex pairs via an effective action, we present here an alternative microscopic self-consistent analysis of the transition in Fourier space, which does not explicitly introduce vortices.§§ We derive the principal thermodynamic features directly from the scaling structure of the Green's functions, obtained by summing perturbation theory to all orders. (However, such a diagrammatic analysis can never guarantee prima facie inclusion of all nonperturbative effects.) We turn then to the structure below TKT, working in terms of the Green's function in momentum space. For an infinite system below TKT, we expect G(r − r′, zν = 0) ~ 1/|r − r′|η as |r − r′| → ∞, where η depends on T (and equals 1/4 at TKT; ref. 16). Thus, G(k, 0) is not well defined in an infinite system; therefore, we start with a finite size system, of characteristic dimension L, to analyze the structure. The condensate density in a finite system is nonvanishing, n0 ~ G(r ~ L, 0) ~ 1/Lη, and vanishes in the limit L → ∞, because long-range order is prohibited in two dimensions; nonetheless, we show using Josephson's relation between the condensate density, the superfluid mass density, ρs, and the infrared behavior of G(k) (M.H. and G.B., unpublished, and refs. 38 and 39), In The Transition in Trapped Gas, we apply our results to a trapped gas within the local density approximation. We show that the KTB transition temperature, for a weakly interacting system, lies below the BEC temperature of the ideal trapped gas by terms of order α log2 α. Furthermore, the jump of the total superfluid mass at the transition is ~ α, and is thus highly suppressed compared with that in a homogeneous system. | |||||||||||||||||||||||||||||||||||||||||
Scaling Structure Above the Transition Here, we derive the KTB transition of a two-dimensional weakly interacting homogeneous Bose gas by studying, as in ref. 21, the scaling structure of G just above the transition. For wavevector k and complex frequency z, G(k, z) is given in terms of the self-energy, Σ(k, z), by The mean-field contribution, Σmf = 2gn, to the self-energy is independent of frequency and momentum, and can be absorbed in a shift of the chemical potential. We introduce the mean-field coherence length, ζ, by Using Eqs. 10 and 6, we may write the transition condition, Eq. 7, as To calculate the critical density at given temperature we use the mean-field density, Eq. 9, as a reference The true correlation length, ξ, above TKT is given by | |||||||||||||||||||||||||||||||||||||||||
Scaling Structure Below the Transition At the transition temperature, TTK, the single-particle Green's function decays algebraically in real space. Below TTK, the scaling structure is most readily analyzed in momentum space, as above; however, this approach is made difficult by the fact that below TTK the single particle Green's functions continues to decay algebraically in real space sufficiently slowly that its Fourier transform is not absolutely convergent. In order to avoid this problem, and to use the same approach as above, we adopt the strategy of working in a finite size system, of characteristic dimension L, in which the condensate density, n0, is nonzero. At the very end, we take the limit L → ∞, at fixed density, in which case n0 goes to zero, as required by the absence of long range order in two dimensions. In a finite size system, the condensate density is given by n0 = n − ñ, where Below TKT, the system is superfluid even though the condensate density is not extensive and vanishes in the thermodynamic limit. However, in two dimensions, an algebraically decaying correlation function is sufficient to yield a nonvanishing superfluid density, ρs, in the thermodynamic limit, as can be seen from the Josephson relation (M.H. and G.B., unpublished, and refs. 38 and 39) between ρs and n0 in a finite system¶¶ Let us now discuss the detailed structure of the Green's function below the transition. We basically follow the scaling approach used in ref. 43 in three dimensions. Our strategy is to expand the self-energies formally in powers of α, n0, and k0. In the infinite size system, k0 → 0, the self-energies diverge as n0 → 0; the point, k0 → 0, n0 → 0 is singular. However Josephson's relation constrains the limit n0 → 0 and k0 → 0 in terms of ρs. The particle density, n, in the condensed phase is a function of α, n0, and T, and has the form, n(α, n0, k0, T) = n0 + ñ(α, n0, T), where ñ(α, n0, T) is the density of noncondensed particles (with momentum k > k0). At the transition temperature, ñ(α, 0, Tc) = nc. We calculate ñ(α, n0, T) in terms of the matrix Green's function The lowest-order mean-field self-energies, Σ11 = Σ11mf = 2g(n0 + ñ), Σ12 = Σ12mf = gn0, are independent of momenta and Matsubara frequency, and, as above TKT, we absorb them in a mean-field coherence length, , Using power-counting, we can derive the scaling structure of the self-energies. As above the transition, we may neglect nonzero Matsubara contributions to leading order. Unlike in three-dimensions, there are no formal ultraviolet divergencies in the expansion beyond those in mean field, and therefore no need for renormalization. The expansion of the self-energies beyond mean field starts at order α22/λ2; furthermore, Σ12 is formally at least of order n0. Diagrams of order gκ with κ ≥ 3 in the formal expansion contain vertices with two Green's functions entering; similar to the structure at Tc, they involve the dimensionless combinations α2/λ2 P and n0λ2. The latter part originates from the dependence of mf on 2mΣ12mf ~ αn0. Any diagram with an explicit power, p, of n0 can be generated from a corresponding diagram of power p − 1, in which a line is replaced by at each of its ends. Thus, each power of n0 involves one fewer two-momentum loop to be integrated over. The explicit n0 dependence enters in two ways. Terms involving G11 − G12 lead to the combination P2n0λ2, which vanishes as n0 → 0. On the other hand, terms involving the combination G11 + G12, which in mean field diverges as n0 → 0 in the infrared limit, lead to divergences which are cutoff by k0 and thus produce an additional (k0)2/(n0λ2P) dependence. In the limit n0 → 0, k0 → 0, only the dependence on (k0)2/n0λ2 Q survives. Then, scaling all momenta k by 1/, we find the following scaling structure for the self-energies in this limit We now take the limit n0 → 0 and k0 → 0. Eqs. 26–28 imply that G(k0, 0) G11(k0, 0) = −ζ2(Q), and thus the Josephson relation implies that the superfluid mass density has the structure We can gain further insight into the structure below TKT by reformulating the analysis in terms of the correlation lengths, ζT and ζL, that control the infrared behavior of the transverse and longitudinal Green's functions, GT = G11 − G12, and GL = G11 + G12. We regularize the infrared divergent structure below TKT by assuming a finite condensate density, n0, and finite correlation lengths. In this description, the low-temperature phase of the KTB transition is characterized by the ratio of amplitude (L) and phase (T) fluctuations of the order parameter, even in the absence of long-range order in the thermodynamic limit in which n0 → 0. In the end, we take the limit, n0 → 0, ζT → ∞, and ζL → ∞. In this way, we do not have to introduce an explicit infrared cutoff, k0, as we had to above. We define The equilibrium state of the system is specified below the transition by ζT → ∞, as seen from the Hugenholtz–Pines relation (25), as well as n0 → 0. From Eqs. 31 and 32, we obtain the ratio | |||||||||||||||||||||||||||||||||||||||||
The Transition in a Trapped Gas We turn now to the behavior of a two-dimensional system trapped in an oscillator potential, of frequency ω. We consider for simplicity only the thermodynamic limit N → ∞, ω → 0, with Nω2 constant, where N is the total particle number. In the absence of interactions, the system undergoes a Bose–Einstein condensation at the critical temperature , as mentioned. In the thermodynamic limit, this result can be obtained by a local density approximation by integrating the density profile, nideal(r), Extending this argument to an interacting Bose gas, we can see, even at the mean field level, how interactions destroy simple Bose-Einstein condensation at finite temperature. The density profile calculated in the local density approximation is However, the system does undergo a KTB transition in the thermodynamic limit. Using the mean-field density profile, Eq. 38, we can calculate the transition temperature to leading order in α. The KTB transition occurs when the chemical potential reaches the critical value Σ(0, 0), calculated for a homogeneous system of the same density as in the center of the trap, or equivalently, when the central density reaches the critical value, Eq. 14. As we see from Eqs. 12 and 13 in the homogeneous case, the critical density is given to logarithmic accuracy by the mean-field density evaluated at the critical ζ given by Eq. 7; critical fluctuations produce corrections, which are, however, important only inside the critical region at small distances where 1/2mω2r2 |gn(0) − μ|, or In a KTB transition in a homogeneous system, the superfluid mass density, ρs, jumps discontinuously from zero to 2m2TKT/π as the temperature drops through TKT. In a trap, however, the transition first occurs in the center, extending over a region of size rc, Eq. 41. The total superfluid mass, Ms, is therefore A key indicator of superfluid behavior below the transition would be creation of a vortex at the center of the trap, where the system first becomes superfluid, e.g., by cooling a rotating system through TKT. To create a vortex at TKT, it is necessary that the vortex core, of radius ~ ζ, fit within the critical region of size ~ rc at the transition. From Eq. 41, ζ/rc ~ mωζ2, so that in the thermodynamic limit, . Thus, vortex formation within the critical region is possible for large N ~ α−2. The critical rotation frequency, Ωc, for creation of a vortex is of order (1/mrc2) log(rc/ζ), and therefore A further probe of the state of the system below TKT would be determination of the density correlations, as have been recently measured in Bose gases trapped in optical lattices (45). These correlations will depend strongly on the amplitude fluctuations, described by ζT (M.H. and G.B., unpublished data). | |||||||||||||||||||||||||||||||||||||||||
Acknowledgments M.H. and G.B. are grateful to the Aspen Center for Physics, where this work was completed. We thank Giuliano Orso for helpful comments. This research was supported in part by National Science Foundation Grants PHY0355014 and PHY0500914, and facilitated by the Projet de Collaboration Centre National de la Recherche Scientifique (CNRS)/University of Illinois at Urbana-Champaign. Laboratoire Kastler Brossel and Laboratoire de Physique Théorique de Matière Condensée are Unités Associées au CNRS Unité Mixte de Recherche (UMR) 8552 and UMR 7600. | |||||||||||||||||||||||||||||||||||||||||
Abbreviations | |||||||||||||||||||||||||||||||||||||||||
Footnotes The authors declare no conflict of interest. This article is a PNAS direct submission. ‡‡The scattering length in two dimensions is not a well defined concept because the scattering cross-section vanishes in the limit of low energies and low momentum. Nonetheless, for a singular potential such as a hard core, it is sufficient to sum particle–particle scattering processes in the T matrix and to replace the bare potential by the T matrix in the Hamiltonian (36, 37). Because the T matrix depends only logarithmically on energy and momentum, we can still work to leading order with a momentum- and energy-independent coupling constant, g = 2π/(m log 1/na2), where a corresponds to the hard core diameter. We further neglect the density dependence of the coupling constant in the following, because it does not enter in an essential way, and write g = 2πα/m. §§The expected phase transition has been calculated by path integral simulations (35) without introduction of explicit vortex degrees of freedom; the vorticity correlation function in these calculations showed no direct evidence of vortex unbinding at the transition. ¶¶Josephson's relation remains valid inside the critical region of a finite size system, with the limit of zero wavevector replaced by k → k0. | |||||||||||||||||||||||||||||||||||||||||
References 1. Gerlitz, A; Vogels, JM; Leanhardt, AE; Raman, C; Gustavson, TL; Abo-Shaeer, JR; Chikkatur, AR; Gupta, S; Inouye, S; Rosenband, T; Ketterle, W. Phys Rev Lett. 2001;87:130402. [PubMed] 2. Schweikhard, V; Coddington, I; Engels, P; Mogendorff, VP; Cornell, EA. Phys Rev Lett. 2004;92:040404. [PubMed] 3. Rychtarik, D; Engeser, B; Nagerl, H-C; Grimm, R. Phys Rev Lett. 2004;92:173003. [PubMed] 4. Smith, NL; Heathcote, WH; Hechenblaikner, G; Nugent, E; Foot, CJ. J Phys B. 2005;38:223–235. 5. Colombe, Y; Knyazchyan, E; Morizot, O; Mercier, B; Lorent, V; Perrin, H. Europhys Lett. 2004;67:593–599. 6. Trombettoni, A; Smerzi, A; Sodano, P. New J Phys. 2005;7:57. 7. Simula, TP; Lee, MD; Hutchinson, DAW. Phil Mag Lett. 2005;85:395–403. 8. Hadzibabic, Z; Kruger, P; Cheneau, M; Battelier, B; Dalibard, J. Nature. 2006;441:1118–1121. [PubMed] 9. Stock, S; Hadzibabic, Z; Battelier, B; Cheneau, M; Dalibard, J. Phys Rev Lett. 2005;95:190403. [PubMed] 10. Hohenberg, PC. Phys Rev. 1967;158:383–386. 11. Mermin, ND. Phys Rev. 1968;176:250–254. 12. Mermin, ND; Wagner, H. Phys Rev Lett. 1966;22:1133–1136. 13. Garrison, JC; Wong, J; Morrison, HL. J Math Phys. 1972;13:1735–1742. 14. Berezinskii, VL. Sov Phys JETP. 1971;32:493–500. 15. Berezinskii, VL. Sov Phys JETP. 1972;34:610–616. 16. Kosterlitz, JM; Thouless, DJ. J Phys C. 1973;6:1181–1203. 17. Kosterlitz, JM. J Phys C. 1974;7:1046–1060. 18. Nelson, DR; Kosterlitz, JM. Phys Rev Lett. 1977;39:1201–1205. 19. Bagnato, V; Kleppner, D. Phys Rev A. 1991;44:7439–7441. [PubMed] 20. Petrov, DS; Holzmann, M; Shlyapnikov, GV. Phys Rev Lett. 2000;84:2551–2554. [PubMed] 21. Baym, G; Blaizot, J-P; Holzmann, M; Laloe, F; Vautherin, D. Phys Rev Lett. 1999;83:1703–1706. 22. Baym, G; Blaizot, J-P; Holzmann, M; Laloe, F; Vautherin, D. Eur J Phys B. 2001;24:107–124. 23. Baym, G; Blaizot, J-P; Zinn-Justin, J. Europhys Lett. 2000;49:150–155. 24. Holzmann, M; Baym, G; Blaizot, J-P; Laloe, F. Phys Rev Lett. 2001;87:120403. [PubMed] 25. Mueller, E; Baym, G; Holzmann, M. J Phys B. 2001;34:4561–4570. 26. Holzmann, M; Fuchs, J-N; Baym, G; Blaizot, J-P; Laloe, F. C R Phys. 2004;5:21–37. 27. Holzmann, M; Krauth, W. Phys Rev Lett. 1999;83:2687–2690. 28. Arnold, P; Moore, G. Phys Rev Lett. 2001;87:120401. [PubMed] 29. Kashurnikov, VA; Prokof'ev, NV; Svistunov, BV. Phys Rev Lett. 2001;87:120402. [PubMed] 30. Fisher, DS; Hohenberg, PC. Phys Rev B. 1988;37:4936–4943. 31. Prokof'ev, N; Ruebenacker, O; Svistunov, B. Phys Rev Lett. 2001;87:270402. [PubMed] 32. Prokof'ev, N; Svistunov, B. Phys Rev A. 2002;66:043608. 33. Sachdev, S; Demler, E. Phys Rev B. 2004;69:144504. 34. Sachdev, S. Phys Rev B. 1999;59:14054–14073. 35. Ceperley, DM; Pollock, EL. Phys Rev B. 1989;39:2084–2093. 36. Schick, M. Phys Rev A. 1971;3:1067–1073. 37. Cherny, AY; Shaneko, AA. Phys Rev E. 2001;64:027105. 38. Josephson, BD. Phys Lett. 1966;21:608–609. 39. Baym, G. Mathematical Methods in Solid State and Superfluid Theory. Clark RC, Derrick GH. , editors. Edinburgh, UK: Oliver and Boyd; 1969. pp. 121–156. 40. Popov, VN. Theor Math Phys. 1972;11:565–573. 41. Popov, VN. Functional Integrals in Quantum Field Theory and Statistical Physics. Dordrecht, The Netherlands: Reidel; 1983. 42. Le Guillou, JC; Zinn-Justin, J. Phys Rev Lett. 1977;39:95–98. 43. Holzmann, M; Baym, G. Phys Rev Lett. 2003;90:040402. [PubMed] 44. Hugenholtz, NM; Pines, D. Phys Rev. 1959;116:489–506. 45. Felling, S; Gerbier, F; Widera, A; Mandel, O; Gericke, T; Bloch, J. Nature. 2005;434:481–484. [PubMed] | |||||||||||||||||||||||||||||||||||||||||