Arxiv:1005.2403V4 [Cond-Mat.Quant-Gas] 21 Sep 2010 Principle Numerical Simulations That Interaction Poten- Gorithm [12, 13]

Supersolid droplet crystal in a dipole-blockaded gas

F. Cinti,1 P. Jain,1 M. Boninsegni,1 A. Micheli,2 P. Zoller,2 and G. Pupillo2 1Department of Physics, University of Alberta, Edmonton, Alberta, Canada T6G 2J1 2IQOQI and Institute for Theoretical Physics, University of Innsbruck, 6020 Innsbruck, Austria. (Dated: October 31, 2018) A novel supersolid phase is predicted for an ensemble of Rydberg atoms in the dipole-blockade regime, interacting via a repulsive dipolar potential “softened” at short distances. Using exact numerical techniques, we study the low temperature phase diagram of this system, and observe an intriguing phase consisting of a crystal of mesoscopic superﬂuid droplets. At low temperature, phase coherence throughout the whole system, and the ensuing bulk superﬂuidity, are established through tunnelling of identical particles between neighbouring droplets.

PACS numbers: 67.80.K-, 32.80.Rm, 67.85.Hj, 67.85.Jk, 67.85.-d, 02.70.Ss

The search for novel phases of matter drives much of densates (droplets) organize into a single, global conden- the current research in condensed matter physics. Of sate. Specifically, we consider the following two-body particular interest are phases simultaneously displaying potential: different types of order. A chief example, of great cur-  3 rent interest, is the so-called supersolid, namely a phase  D/a if r ≤ a featuring crystalline order, and also capable of sustaining v(r) = , (1) dissipation-less flow. Attempts to observe experimentally  D/r3 if r > a a supersolid phase of matter, primarily in a crystal of solid helium, have spanned four decades since early the- D being the characteristic strength of the interaction. oretical predictions [1]. The most credible claim of such This kind of interaction potential can be realized with an observation to date [2, 3], has been subjected to in- cold dipole-blockaded Rydberg atoms [6–10]. The pa- depth scrutiny over the past few years, and it seems fair rameters D and a above can be controlled with external to state that agreement is lacking at the present time, as fields [5, 11] (we come back to this point below). to whether experimental findings indeed signal a super- Our system of interest comprises N identical bosons of solid phenomenon [4]. mass m, confined to two dimensions [21]. The many-body A new, fascinating avenue to the observation of super- Hamiltonian is the following (in dimensionless form): solid and other phases of matter not yet observed (or N even thought of), is now opened by advances in cold 1 X X H = − ∇2 + v(r ) (2) atom physics, providing not only remarkably clean and 2 i ij i=1 i>j controlled experimental systems, but also allowing one to “fashion” artificial inter-particle potentials, not aris- where rij = |ri − rj| is the distance between particles i ing in any known condensed matter system. This allows and j, and v is given by Eq. (1). All lengths are expressed one to address a key theoretical question, namely which 2 in terms of the characteristic length r◦ = mD/~ , and we two-body interaction potential(s), if any, can lead to the introduce a dimensionless cutoff Rc = a/r◦ for the poten- occurrence of a supersolid phase in free space (i.e., not tial (1). The system is enclosed in a square cell of area A, on a lattice). with periodic boundary conditions. The particle density is n = N/A, but we shall express our results in terms of In a recent article [5], Henkel et al. have proposed, p 2 the (dimensionless) inter-particle distance rs=1/ nr◦. based on a mean-field treatment, that a Bose conden- 3 2 2 The energy scale is ◦ = D/r◦ = ~ /mr◦. sate of particles interacting through an effective poten- The low-temperature phase diagram of such a system tial which flattens off at short distance, might support has been explored by means of first principles numerical a density modulation. In this Letter, we show by first simulations, based on the Continuous-space Worm Al- arXiv:1005.2403v4 [cond-mat.quant-gas] 21 Sep 2010 principle numerical simulations that interaction poten- gorithm [12, 13]. Numerical results shown here pertain tials which combine a long-distance repulsion with a to simulations with a number of particles N varying be- short-distance cutoff, lead in fact to the appearance of a tween 50 and 400, in order to carry out extrapolation novel self-assembled crystalline phase of mesoscopic su- of the results to the thermodynamic limit. Our ground perfluid droplets in a system of bosons. Furthermore, state estimates are obtained as extrapolations of results such a crystal can turn supersolid in the T → 0 limit, at finite temperature. Details of the simulations are stan- as tunneling of particles across neighbouring droplets dard, as the use of the potential (1) entails no particular takes place, and superfluid phase coherence is established technical difficulty. across the whole system, as individual separate Bose con- In the limit Rc rs, the truncation of the dipolar 2

(a) (b) 1.5 4 (a) 3 1 y g(r) 2

0.5 1 0 0 0.2 0.4 0.6 0.8 0 r (c) (d) 0.2 1.5 (b) s

1 ρ y 0.1

0.5 0 0 10 20 30 40 0 0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 T x x 1.0×10 (c) -2 FIG. 1: Snapshots of a system of bosons interacting via po- 1.0×10

tential (1), at the four diﬀerent temperatures 200 (a), 20 (b), P

-4 1.0 (c) and 0.1 (d), expressed in units of ◦. Points shown are 1.0×10 taken along individual particle world lines. The nominal value of rs in this case is 0.14, whereas the cutoﬀ of the potential -6 1.0×10 (1) is Rc=0.3. 0 20 40 60 80 100 L

FIG. 2: (Color online) Results shown are for rs=0.14 and potential at short distances does not play an important Rc=0.3. Temperature is in units of ◦. (a): Pair correla- role, and the low temperature phase diagram of (2) is tion function g(r) at a temperature T = 200 (triangles), 20 that of purely dipolar bosons in two dimensions, investi- (squares), 1.0 (diamonds) and 0.1 (circles). The simulated system comprises N=200 particles. (b): Superfluid density gated previously by several authors [14, 15]. It is known vs. T for systems with N=100 (square), and 200 (diamond) < C that for rs ∼ rs = 0.06 the ground state of the system particles. (c): Frequency of occurrence of permutation cycles > L is a triangular crystal, whereas for rs ∼ rs = 0.08 it is of length L at the same four temperatures reported in panel a uniform superfluid (in the intermediate density range (a). Longer permutation cycles occur at lower temperature. a more complex scenario is predicted [16]). As we show > below, a very different physics sets in when Rc ∼ rs, in the density ranges which correspond to either the crystalline or superfluid phase in the purely dipolar system. The formation of such droplets is a purely classical Fig. 1 shows typical configurations (i.e., particle world effect, that depends on the flattening off of the repul- lines) produced by Monte Carlo simulations of a system sive inter-particle potential below the cutoff distance. In of bosons interacting via the potential (1), at a nominal fact, a simple estimate of the number Nd of particles per droplet, can be obtained by considering a triangular lat- density corresponding to rs = 0.14, at different temperatures spanning three orders of magnitude. The value of tice of point-like dipoles, each one of strength ∝ Nd (as it comprises Nd particles), and by minimizing with re- the cutoff Rc in this case is 0.3. At the highest temperature, a simple classical gas phase is observed, as shown spect to Nd the potential energy per particle, for a fixed by the pair correlation function g(r), shown in Fig. 2 (a), density. The result is which is just a constant (note that g(r) does not vanish 2 Rc at the origin, owing to the flattening off of the potential Nd = γ (3) r at short distance). As T is decreased, an intriguing ef- s fect takes place, namely particles bunch into mesoscopic where γ ≈ 2.79. Eq. (3) furnishes a fairly accurate esti- droplets, in turn forming a regular (triangular) crystal. mate of Nd for the (wide) range of values of the param- This is shown qualitatively in the snapshots in Fig. 1, but eters rs and Rc explored here. For instance, using the also confirmed quantitatively by the structure of the g(r) parameters of Fig. 1, we find from (3) Nd ≈ 13, which as well (Fig. 2(a)), which displays pronounced, broad agrees quite well with our simulation result. It is worth maxima, as well as well-defined minima, where the func- noting that a similar sort of pattern formation, due to tion approaches zero. We henceforth refer to this phase competing interactions, has been previously established as the droplet-crystal phase. for classical colloidal systems [17, 18]. 3

2 IDC SDC SF (a) 1 L Rc > rs 1 IDC C SF y 0 0.5

C n(k) Rc < rs -1 rs -2 0 FIG. 3: Schematic ground state phase diagram of (2) as a -2 -1 0 1 2 -20 -10 0 10 20 2 function of rs. The superfluid droplet crystal (SDC) is sand- 1 wiched between an insulating droplet crystal (IDC) and a su- (b) < C 1 perfluid (SF). For Rc ∼ rs , an IDC, a single-particle crystal (C) and a superfluid phases are observed. The widths of the y 0 0.5 SDC and C regions depend on the value of Rc. n(k) -1

-2 0 -2 -1 0 1 2 -20 -10 0 10 20 In the T → 0 limit, long exchanges of identical parti- x cles can take place, as a result of particles tunneling from k one droplet to an adjacent one. Long exchanges of parti- FIG. 4: (Color online) Left panels: Monte Carlo snapshots of cles can result in a finite superfluid response throughout a system of N = 400 particles, interacting via the potential > L the whole system, and indeed for Rc ∼ rs we observe (1), confined in a harmonic trap of strength Γ = 500 ◦, at such a bulk superfluid signal, in a range of values of rs the two temperatures T =100 ◦ (a) and T =0.5 ◦ (b). Right in the vicinity of Rc/2. A typical result is shown in Fig. panels: corresponding momentum distributions, all normal- 2(b) [22]. Because superfluidity arises in concomitance ized to unity for comparison purposes. The value of Rc in this case is 0.3. The development of secondary peaks at low Tuesday, Maywith 11, 2010 the droplet-crystal structure, the denomination su- temperature signals the occurrence of a supersolid phase. persolid seems appropriate. In order to establish that droplets are individually superfluid, one may consider the statistics of permutation cycles. Fig. 2 (c) shows the frequency of occurrence of that supersolid behaviour in this system originates from exchange cycles involving a varying number L of parti- tunnelling of particles between droplets which are them- cles (1 ≤ L ≤ N), at three different temperatures, at selves individually superfluid, so that the individual su- the physical conditions of Fig. 1. As one can see, as the perfluid droplets connect to form a bulk superfluid. This temperature is lowered exchange cycles involving grow- is reminiscent of the phase-locking mechanism in a (self- ing numbers of particles occur. At low temperature they assembled) array of Josephson junctions. involve almost all the particles in the system; however, The results discussed so far pertain to numerical sim- even at a higher temperature (e.g., T =20 in Fig. 2(c)) ulation of the system described by Eq. (2) in its bulk one observes exchanges comprising a number of particles phase. However, in any experiment aimed at probing up to ∼ Nd, i.e., particles inside an individual droplet. the physics of such a system, the assembly of particles This is evidence that droplets are individually Bose con- must necessarily be finite (a few thousand particles is a densed and superfluid, even though the system as a whole typical number for current experiments with cold dipo- does not display superfluidity. That droplets should be lar atoms), confined by an external potential. In order superfluid at low T is not surprising, given that particles to enable a direct comparison with possible future ex- in a droplet are essentially non-interacting, due to the periments, we have performed simulations of the same flatness of the potential at short distance. However, that system spatially confined in-plane by a harmonic trap, P 2 2 droplets are themselves superfluid does not imply that i.e., the term Γ i ri is added to Eq. (2), Γ ≡ mω /2 a bulk supersolid phase will always occur in the T → 0 being the strength of the trap. limit, as discussed above. Fig. 4 shows typical many-particle configurations of At T = 0, the supersolid phase is sandwiched be- a trapped system comprising N=400 particles, at two tween an insulating droplet crystal at high density (i.e., different temperatures. Also shown are the associated lower rs) and a homogeneous superfluid phase at lower momentum distributions n(k), which are obtained by < C density. For Rc ∼ rs , only two insulating phases are Fourier transforming of the spherically and translation- observed, namely the insulating droplet crystal at high ally averaged one-body density matrix, computed by density and the crystal of single particles, already de- Monte Carlo. tected in Refs. [14, 15], as well as a superfluid phase at Here too, droplets with a well-defined average number lower density. All of this is summarized in the schematic of particles form, and organize themselves on a triangu- phase diagram shown in Fig. 3. It is important to stress lar lattice. Correspondingly, the momentum distribution, 4 which is directly observable experimentally by time-of- with n = 20, Ω/2π = 50MHz and |∆|/2π = 3GHz, flight measurements [19], develops a sharp central peak, we obtain d = 1450 Debye, a ' 400 nm, r◦ ' 210 with additional structure on its sides. The secondary nm, ◦/(hγ) ' 90, and ◦/kB ' 120 nK. Collective peaks correspond to oscillations in the one-body density many-body effects in the Rydberg-blockade regime [20] matrix, in turn reflecting particle tunnelling to adjacent not described by Eq. (2) should be negligible provided 2 2 droplets. They are therefore connected to the appearance (Rc/rs) (2∆/Ω) , which is readily satisfied for pa- of the supersolid phase, as explained above. rameters as in Fig. 1. Summarizing, accurate numerical simulations of a system of dipolar particles interacting via a potential soft- This work was supported in part by the Natural Sci- ened at short distance, reveal the existence of a low tem- ence and Engineering Research Council of Canada un- perature crystalline phase of superfluid droplets. This der research grant 121210893, and by the Alberta Infor- phase turns superfluid (supersolid) at T → 0 through matics Circle of Research Excellence (iCore), IQOQI, the a mechanism of tunnelling of particles between adjacent FWF, MURI, U.Md. PFC/JQI, EOARD (grant FA8655- droplets. At higher density this tunnelling is suppressed 10-1-3081), NAME-QUAM. Useful discussions with I. and an insulating droplet crystal occurs, a phase which Lesanovsky, N. Prokof’ev and T. Pohl are gratefully ac- has not previously been predicted. The interaction that knowledged. underlies such intriguing, until now unobserved physical behaviour, can be realized with dipolar atoms in the dipole-blockade regime. A comment is on order, concerning the dependence of [1] A. F. Andreev and I. M. Lifshitz, Sov. Phys. JETP. 29, the results on the particular form of potential utilized 1107 (1969). here, namely Eq. (1) with its abrupt, sharp cutoff at [2] E. Kim and M. H. W. Chan, Nature 427, 225 (2004). r = a. First off, the superfluid droplet crystal phase [3] E. Kim and M. H. W. Chan, Science 305, 1941 (2004). does not crucially depend on the dipolar form of the in- [4] N. Prokof’ev, Advances in Physics 56, 381 (2007). teraction at long distances. Indeed, it is also observed [5] N. Henkel, R. Nath, and T. Pohl, Phys. Rev. Lett. 104, 195302 (2010). in our simulations for a van der Waals-like potentials [6] M. Lukin et al., Phys. Rev. Lett. 87, 037901 (2001). −6 (i.e., v(r) ∼ r ). Secondly, we have obtained qualita- [7] D. Tong et al., Phys. Rev. Lett. 93, 063001 (2004). tively similar results with different model potentials, fea- [8] T. Vogt et al., Phys. Rev. Lett. 97, 083003 (2006). turing a more realistic “flat” region at short distances, [9] R. Heidemann et al., Phys. Rev. Lett. 100, 033601 as well as a smoother merge of long- and short-range (2008). behaviours. For example, we considered the potential [10] J. Pritchard et al., arXiv:1006.4087 (2010). V (r) = D/ a3 + r3, which is naturally realized in a [11] G. Pupillo et al., Phys. Rev. Lett. 104, 223202 (2010). [12] M. Boninsegni, N. Prokof’ev, and B. Svistunov, Phys. cold gas of alkali atoms by weakly dressing the ground- Rev. Lett. 96, 070601 (2006). state |gi of each atom with an excited Rydberg state |ri [13] M. Boninsegni, N. V. Prokof’ev, and B. V. Svistunov, with a large dipole moment d, in the kDebye range [11]. Phys. Rev. E 74, 036701 (2006). While several dressing schemes are possible [5, 20], here [14] H. P. Büchler et al., Phys. Rev. Lett. 98, 060404 (2007). we consider |ri as the lowest-energy state of a Rydberg [15] G. E. Astrakharchik, J. Boronat, I. L. Kurbakov, and manifold with principal quantum number n for an atom Y. E. Lozovik, Phys. Rev. Lett. 98, 060405 (2007). [16] B. Spivak and S. Kivelson, Phys. Rev. B 70, 155114 in the presence of a homogeneous electric field F < FIT 4 (2004). in the linear Stark regime, with FIT ∝ 1/n the Inglis- [17] Y. H. Liu, L. Y. Chew, and M. Y. Yu, Phys. Rev. E 78, Teller limit. For a laser with (effective) Rabi frequency 066405 (2008). Ω and red detuning |∆| Ω, the dressed groundstate [18] A. J. Archer, C. Ionescu, D. Pini, and L. Reatto, Journal reads |g˜i ∼ |gi − (Ω/2∆)|ri, and thus D ∝ (Ω/∆)4d2, of Physics: Condensed Matter 20, 415106 (2008). 2 2 1/3 [19] M. Greiner et al., Nature 415, 39 (2002). with d ∝ n , while the cutoff a ' (d /~|∆|0) arises because of the Rydberg-blockade mechanism [6]. Spon- [20] J. Honer, H. Weimer, T. Pfau, and H. P. Büchler, arXiv:1004.2499 (2010). taneous emission rates γr from |ri are strongly reduced 2 [21] Confinement to two dimensions can be achieved using a to values of at most γ ' (Ω/∆) γr. Observability of tight (magnetic or optical) trap along z. the phases above benefits from large values of ◦, and [22] We compute directly the bulk superfluid fraction through 2 3 in particular of the ratio ◦/(hγ) = [Ω /(4π|∆|γr)]RC, the usual winding number estimator for the bulk system. favoring comparatively small values of r◦ [23]. For exam- [23] 2D scattering requires r◦ > a⊥, with a⊥ the transverse ple, for 87Rb atoms in an electric field F = 25kV/m, harmonic oscillator length, in the tens of nm range.