Arxiv:1112.3890V2 [Cond-Mat.Stat-Mech] 30 Jan 2012 Metry Be Uniform

BI-TP 2011/49

Spontaneous breaking of continuous translational invariance

Haruki Watanabe1, 2, ∗ and Tom´aˇsBrauner3, 4, † 1Department of Physics, University of Tokyo, Hongo, Tokyo 113-0033, Japan 2Department of Physics, University of California, Berkeley, California 94720, USA 3Faculty of Physics, University of Bielefeld, 33615 Bielefeld, Germany 4Department of Theoretical Physics, Nuclear Physics Institute ASCR, 25068 Reˇz,Czechˇ Republic Unbroken continuous translational invariance is often taken as a basic assumption in discussions of spontaneous symmetry breaking (SSB), which singles out SSB of translational invariance itself as an exceptional case. We present a framework which allows us to treat translational invariance on the same footing as other symmetries. It is shown that existing theorems on SSB can be straightfor- wardly extended to this general case. As a concrete application, we analyze the Nambu–Goldstone modes in a (ferromagnetic) supersolid. We prove on the ground of the general theorems that the Bo- goliubov mode stemming from a spontaneously broken internal U(1) symmetry and the longitudinal phonon due to a crystalline order are distinct physical modes.

PACS numbers: 11.30.Qc, 14.80.Va, 67.80.-s Keywords: Spontaneous symmetry breaking, Nambu–Goldstone boson, Supersolid

I. INTRODUCTION The objective of the present paper is to complement our previous study by a more detailed discussion of spon- The low-energy physics of systems with spontaneous taneous breaking of translational invariance. Since trans- symmetry breaking (SSB) is dominated by the associated lational invariance is the only uniform spacetime symme- soft modes, the Nambu–Goldstone (NG) bosons. The ex- try, this completes our analysis and shows that all uni- istence of at least one such a soft mode is guaranteed by form symmetries can be treated on the same footing. At the celebrated Goldstone theorem [1] (see also Ref. [2] the same time, it provides a generalization of the other for a comprehensive early review). The question of how existing results. What we do assume is an unbroken dis- many NG bosons there are given the symmetry-breaking crete translational invariance of the ground state though. pattern is easy to answer in Lorentz-invariant systems. This is a minimal necessary requirement for the notion of In the general case, it is however difficult; several par- NG bosons as soft quasiparticles to be meaningful. One tial solutions to the problem have been given over the can expect that even in the complete absence of transla- years [3–5] (see also Ref. [6] for a recent review). tional invariance, SSB will give rise to low-lying excited What is common to the general results existing in liter- states in the spectrum; such a generalization, however, ature is the separate treatment of internal and spacetime goes beyond the scope of this paper. symmetries. In particular, it is usually assumed that the We would hasten to emphasize that the issue of spon- ground state of the system has a continuous translational taneous breaking of translational invariance is, of course, invariance. SSB of translational invariance itself is then not of merely academic interest. Systems where an inho- treated as an exceptional case (an analysis based on ef- mogeneous state develops spontaneously have been ex- fective field theory was given in Ref. [7]). Since only a tensively discussed in the literature. A classic example discrete translational symmetry is needed to have well- is the inhomogeneous pairing in superconductors in ex- defined low-momentum quasiparticles, and thus to dis- ternal magnetic field, predicted by Larkin, Ovchinnikov, cuss the NG bosons, such separation seems to be merely Fulde, and Ferrell (LOFF) [9]. Similar type of pairing a matter of convenience. can occur in many other systems, ranging from ultracold In the recent paper [8], we argued that the distinction atomic gases to cold dense quark matter, see Ref. [10] for between internal and spacetime symmetries is somewhat a review and further references. artificial, at least as long as SSB is considered. Instead, In LOFF-type superconductors, the order parameter we pointed out that the requirement necessary for stan- is usually modulated in one dimension. On the contrary, dard arguments concerning SSB to apply is that the sym- complete breaking of continuous translational invariance

arXiv:1112.3890v2 [cond-mat.stat-mech] 30 Jan 2012 metry be uniform. This, roughly speaking, means that is featured by another prominent example, the super- the infinitesimal symmetry transformation does not de- solid [11]. A supersolid is a state which has both a crys- pend explicitly on the spacetime coordinates. Formally, talline order (ordinary long-range order) and a superfluid it can be encoded in the condition on the translational order (off-diagonal long-range order). In the present pa- property of the Noether charge density, Eq. (1). per, we will use it as an example to illustrate the general results concerning spontaneous breaking of translational invariance. Other examples of systems exhibiting spontaneous breaking of translational invariance include stripe ∗ [email protected] ordering in spinor Bose–Einstein condensates [12] and nu- † [email protected] merous systems in high-energy physics such as the charge 2

0 or chiral density wave [13] or the quarkyonic chiral spi- Rd−d , respectively. Likewise, the spatial coordinate vec- d0 ral [14]. tor x can be decomposed as x = (x1, x2) where x1 ∈ R d−d0 All in all, it is clearly desirable to have a systematic and x2 ∈ R . Moreover, the vector x1 can be uniquely framework that can be used to analyze the physics of NG represented as x1 = x¯1 + R where R is a lattice vector d0 modes in systems where translational invariance is bro- and x¯1 belongs to the Wigner–Seitz unit cell in R . We ken to a (nontrivial) discrete subgroup. Let us remark will sometimes for the sake of brevity refer to the set of that all our general arguments, presented in Section II, all spacetime coordinates, (t, x), using the corresponding are actually valid regardless of whether this breaking is regular-weight letter, x. spontaneous or explicit. As a byproduct of our inter- Apart from translational invariance, the system may est in spontaneous breaking of translational invariance, possess other, internal symmetries at the Lagrangian we therefore provide a formalism for description of NG level, which can be spontaneously broken. Following bosons of spontaneously broken internal uniform symme- the argument presented in Ref. [8], we always assume tries in systems with a discrete periodic structure such that such symmetries are uniform, that is, their Noether as crystalline solids or optical lattices [15]. charge density j0(x) satisfies the following condition [16], The plan of the paper is as follows. In Section II, we 0 discuss in detail the setup and generalize the existing the- j (t, x1, x2) orems on NG modes, in particular the Goldstone theo- = ei(Ht−P2·x2)/~T † j0(0, x¯ , 0) T e−i(Ht−P2·x2)/~. (1) rem [1], the theorem of Schäfer et al. [4], and the Nielsen– R 1 R Chadha theorem [3]. Readers not interested in general We also assume that the charge associated with a spon- properties of spontaneous symmetry breaking might pre- taneously broken symmetry (whether internal or trans- fer to skip this section and proceed directly to Section III lational) is time-independent and is given by an integral where we analyze in detail a model for a two-dimensional of a local charge density, Q = R ddx j0(t, x). (ferromagnetic) supersolid. As another nontrivial exam- Finally, we assume that there is a complete set ple, we argue how translational properties of the system {|ξ, ki} of simultaneous eigenstates of the Hamiltonian 0 get affected by an external uniform magnetic field and (H|ξ, ki = ε (k)|ξ, ki), discrete translations in d discuss phonons in a charged crystal under the magnetic ξ R (T |ξ, ki = eik1·R|ξ, ki), and the remaining momentum field (Section IV). In Section V, we summarize and con- R operators (P2|ξ, ki = ~k2|ξ, ki). The normalization of clude. 0 0 d d 0 the states is chosen as hξ, k|ξ , k i = (2π) δξξ0 δ (k − k ) so that the completeness of the basis is expressed by the relation II. GENERAL THEOREMS X Z ddk |ξ, kihξ, k| = 11. (2) (2π)d As a first step towards the generalization of the ex- ξ (FBZ) isting results on NG bosons, we carefully formulate our assumptions. This will in particular clarify the scope of The integration over k1 is performed within the first Bril- the validity of our arguments presented in the later parts louin zone (FBZ), that is, the Wigner–Seitz cell of the reciprocal space, while the k2 integration is done over of this section. 0 the whole Rd−d space. The index ξ may include discrete as well as continuous labels such as the band index, which together with the momentum uniquely iden- A. Assumptions tify each eigenstate. It is customary to characterize SSB by the number First of all, we should emphasize that when we speak of spontaneously broken generators Qa, denoted here as of breaking translational invariance, we always have in nBS. This can be usually determined as the difference of mind spatial translations. Continuous time translation the dimensions of the symmetry groups of the action and invariance, which guarantees energy conservation, will the ground state of the theory. However, for the sake be assumed throughout this paper without further men- of general arguments, it is more convenient to use the tioning it. As to the spatial translations, we assume dis- following formal definition: nBS symmetries are said to crete translational invariance in d0 directions and con- be spontaneously broken if there is as set of (quasi)local 0 nBS tinuous one in the remaining d − d directions. This operators {φa(x)}a=1 such that the matrix M given by requirement is imposed on both the Hamiltonian (or iMab ≡ h0|[Qa, φb(0)]|0i is nonsingular. Lagrangian) of the theory and its ground state. The (d0) (d−d0) translation symmetry group is thus GD × GC where (d0) d0 B. Goldstone theorem iP1·R/~ P GD = {TR ≡ e | R = i=1 niai (ni ∈ Z)} and d0 {ai}i=1 are the primitive lattice vectors. Furthermore, The Goldstone theorem [1] claims that whenever a uni- (d−d0) 0 iP2·∆x/~ d−d GC = {e | ∆x ∈ R } where P is the mo- form symmetry is spontaneously broken, there should be d0 mentum operator and P1, P2 its components in R and at least one massless bosonic state in the spectrum of 3 the theory. The proof proceeds essentially by finding the and k1 is equal to some of the vectors of the reciprocal spectral representation of the commutator of the broken lattice) [2, 17]. It then follows that there must be at 0 charge density ja and the interpolating field φb. Here we least one intermediate excited state |n, ki such that: (i) adapt the standard argument to account for the possi- h0|ja(k)|n, ki 6= 0 and hn, k|φb(0)|0i 6= 0 for some a, b in bility of discrete translational invariance and obtain the the vicinity of k = 0, and (ii) εn(0) = 0. Such inter- following spectral representation, mediate states are exactly the Nambu–Goldstone modes Z associated with spontaneous breaking of the charges Qa. d+1 i(ωt−k·x) 0 Denoting |ni ≡ |n, 0i and the number of these states as iSab(ω, k) ≡ d x e h0|[ja(x), φb(0)]|0i nNG, Eq. (6) becomes Xh 1 =2π δ(ω − εξ(k))h0|ja(k)|ξ, kihξ, k|φb(0)|0i n ~ XNG ξ M = 2 Im h0|j |nihn|φ (0)|0i . (7) i ab a b 1 n=1 − δ(ω + εξ(−k))h0|φb(0)|ξ, −kihξ, −k|ja |0i , ~ (k) (3) Before concluding the discussion of the Goldstone theorem, let us remark that in case there are long-range which holds provided k1 lies in FBZ. This expression is interactions in the system, the surface term appearing formally identical to one assuming full continuous trans- in the space integral of the continuity equation for the lational invariance; all effects of only discrete transla- Noether charge density and its current need not vanish. d0 tional symmetry in R are hidden in the definition of The integral charge then depends on time and massive the charge density averaged over the unit cell, modes can contribute to Eq. (6). This is why the longitudinal mode in a Coulomb-interacting Wigner solid (in 1 Z 0 d 0 −ik1·x¯1 ja(k) ≡ d x¯1 ja(0, x¯1, 0)e , (4) three spatial dimensions) acquires an energy gap, equal Ωuc uc to the plasma frequency [18]. Finally, note that the above presented proof of the where Ωuc is the volume of the Wigner–Seitz unit cell, over which the x integral is performed (this is indicated Goldstone theorem relies essentially on the operator for- 1 malism and the translational property (1) of the charge by the subscript “uc”). Note that in the limit of Ωuc → 0, 0 density as well as the time independence of the integral ja(k) → ja(0) which leads to the usual spectral representation with the full continuous translational invariance. charge. In concrete applications, it is often advantageous The proof of Eq. (3) is straightforward but somewhat to use an alternative proof based on the effective action tedious, and we thus refer the reader to Appendix A for formalism [1, 19], which provides a direct connection to details. Consider now the limit |k| → 0 of the commuta- the symmetries of the corresponding classical Lagrangian tor Fourier transformed only in the spatial coordinates, system. For the sake of completeness, we sketch a modifi- cation of this proof suitable for systems with only discrete Z d −ik·x 0 translational invariance in Appendix B. lim iSab(t, k) ≡ lim d x e h0|[ja(x), φb(0)]|0i |k|→0 |k|→0

= h0|[Qa, φb(0)]|0i = iMab. (5) C. Theorem of Sch¨afer et al.

Using the assumption that the charges Qa are time- independent (which is most naturally ensured when In the preceding subsection, we carefully formulated 0 the Goldstone theorem as showing the existence of at ja satisfies a continuity equation together with a spa- least one NG boson. The natural question to ask is, of tial current ja, and the corresponding surface term in the integration over space vanishes), we deduce that course, whether anything can be said in general about their actual number, n . A common lore in textbooks lim iSab(ω, k) ∝ δ(ω). Combining this with the spec- NG |k|→0 on relativistic field theory is that assuming Lorentz in- tral representation (3), we find variance, nNG equals the number of spontaneously broken generators, n . Nevertheless, there are numer- X BS Mab = 2 Im h0|ja|ξ, 0ihξ, 0|φb(0)|0i , (6) ous examples of Lorentz-noninvariant (both relativistic

ξ|εξ(0)=0 many-body as well as intrinsically nonrelativistic) systems where this simple relation does not hold (see Ref. [6] where ja ≡ ja(0). and references therein). The condition of SSB (det M 6= 0) ensures that the We start our discussion with the result of Schäfer et right-hand side of this equation is nonzero for some a, b. al. [4] who formulated the following theorem [20]: Pro- One might at first think that this is due to the contribu- vided that h0|[Qa,Qb]|0i = 0 for all a, b, the number of tion of degenerate ground states obtained by the action NG modes nNG is equal to the number of broken symme- of broken symmetry transformations on |0i. However, tries nBS. Here we generalize this theorem by taking into this contribution drops at large volume Ω asymptoti- account the possibility of a discrete translational symme- 0 cally as 1/Ω since such degenerate ground states only try in Rd (whether as a result of spontaneous breaking exist for isolated values of momentum (for which k2 = 0 of continuous translational invariance or of the intrinsic 4 setup of the system). We follow essentially the argument systems with only discrete translational symmetry, we of the review [6]. basically just need to replace the spectral representation Let us first show that nNG ≥ nBS. Based on the re- in their original paper by Eq. (3). We shall nevertheless maining translational symmetry of the ground state, we present some details of the argument. infer that h0|[Qa,Qb]|0i ∝ h0|[Qa, jb]|0i and, additionally, Let us define the vectors (An)b = h0|φb(0)|ni and va = using Eq. (7), Anhn|ja|0i. Let in addition p be the number of linearly independent vectors among the va’s (obviously p ≤ nNG), nNG X and set ∆n = nBS − p. By construction, there is a set of h0|[Qa, jb]|0i = 2i Im h0|ja|nihn|jb|0i . (8) α α coefficients C (α = 1,..., ∆n) such that C va = 0 and n=1 a a simultaneously rank C = ∆n. Using Eq. (3), we obtain Defining an nBS × nNG matrix A as Aan = h0|ja|ni, the iCαS (ω, k) ' assumption that h0|[Qa,Qb]|0i = 0 for all a, b together a ab with Eq. (8) implies that AA† is Hermitian as well as real. X 1 α 2π δ(ω − εn(k))h0|C ja |n, kihn, k|φb(0)|0i (9) ~ a (k) It can thus be diagonalized by a real orthogonal matrix n 0 0† † T O such that A A ≡ OAA O = diag(λ1, . . . , λn ), as- BS in the vicinity of k = 0. Here we dropped the contribu- suming without lack of generality 0 ≤ λ1 ≤ · · · ≤ λn . BS tion from massive modes and the second term of Eq. (3), Now if the number of NG bosons were smaller than nBS, the rank of A, and hence also the rank of AA†, be- since they both vanish at k = 0. Assuming along with Nielsen and Chadha the exponential decay of (expecta- ing at most nNG, would have to be smaller than nBS tion values of) commutators of local operators separated as well. The smallest eigenvalue λ1 would then necessarily vanish. However, since the orthogonal transfor- by a large spacelike interval, we infer that Sab(t, k) is mation by O can be interpreted as a change of basis of an analytic function of k. In other words, the support 0 of Sab(ω, k) should be differentiable with respect to k. the broken generators, Qa = OabQb, this would lead to 0 0 As a consequence, for any |n, ki that appears in Eq. (9) h0|j1 |ni = hn|j1 |0i = 0 for all n, in contradiction with the broken symmetry assumption, det M 6= 0. (Note that for some α and b, εn(k) must be an analytic function of M 0 = OM so that determinants of M and M 0 are equal k. We will call such NG modes type-C and denote their up to a sign.) Thus, the inequality n ≥ n follows. number as nC. NG BS As the next step, we observe that the ∆n vectors In order to prove the opposite inequality, nNG ≤ nBS, α∗ α∗ observe that A†A is positive semidefinite and Hermitian, Ca va = Anhn|Ca ja|0i must be linearly independent. and can therefore be diagonalized by a unitary matrix Indeed, if this were not true, there would be a set of coefficients β such that β Cα∗v =0 and thus β Cα∗M = 0, U as A00†A00 ≡ U †A†AU = diag(ρ , . . . , ρ ), again as- α α a a α a ab 1 nNG implying that M has a zero eigenvalue, in contradiction suming without lack of generality that 0 ≤ ρ ≤ · · · ≤ 1 with the assumption det M 6= 0. From the fact that the ρ . The transformation A → A00 = AU corresponds to nNG vectors Cα∗v are linearly independent, we can imme- a redefinition of the basis of states, |m00i = |niU , since a a nm diately conclude that the number of NG states |ni for A00 = A U = h0|j |m00i (summations over repeated am an nm a which h0|Cαj |ni 6= 0 for some α, must be at least ∆n, indices are implied). Now if the number of NG bosons a a hence n ≥ ∆n. were larger than n , we would analogously deduce that C BS The remaining NG modes, for which h0|Cαj |ni = 0 h0|j |m00i = 0 for all a and for m = 1, 2, . . . , n − n . a a a NG BS for all α, will be called type-N and their number denoted However, these zero modes do not couple to any broken accordingly as n . The above argument does not con- currents and thus do not satisfy the very definition of a N strain in any way the dispersion relation of type-N NG NG boson. bosons; in particular, it is allowed to be linear, ε(k) ∝ |k|. Putting both pieces together, we have proven that the Using the fact that, by construction, n + n = n , we number of NG modes is equal to the number of broken N C NG obtain the inequality symmetries if h0|[Qa,Qb]|0i = 0 for all a, b.

nN + 2nC ≥ nNG + ∆n = nBS + (nNG − p) ≥ nBS. (10) D. Nielsen–Chadha theorem Note that in Ref. [8], we actually proved the equality, nN + 2nC = nBS, using a slightly different definition of Nielsen and Chadha [3] discovered that the number the two types of the NG modes. of NG bosons is tightly related to their dispersion rela- In fact, the derivation of the inequality (10) does not tions. Concretely, they showed that under certain tech- make any use of rotational invariance which was implic- nical assumptions, the dispersion relation of a NG boson itly, and somewhat unnecessarily, assumed in Ref. [3]. is proportional to a positive integer power of momen- Once this constraint is released, the classification of NG tum in the long-wavelength limit. Classifying NG bosons bosons can no longer be based on a mere power of the with an odd (even) power of momentum in the dispersion momentum |k| in the dispersion relation. A convenient relation as type-I (type-II) and denoting their numbers generalization is to define as type-II those NG modes as nI and nII, respectively, they proved the inequality whose dispersion is analytic around k = 0, and as type- nI + 2nII ≥ nBS. In order to generalize their result to I the remaining ones; this classification reduces to that 5 of Nielsen and Chadha once rotational invariance is re- field B. Due to the explicit dependence of the vector stored. We have thus shown in other words that every potential on the coordinates, the action is no longer in- type-C NG boson is simultaneously type-II. Using the invariant under the usual translations. Instead, it is inequality (10), we then find nI + 2nII ≥ nN + 2nC ≥ nBS, variant under magnetic translations, that is, translations which reproduces the Nielsen–Chadha inequality. combined with a gauge transformation of the electromag- netic field. The momentum operator P as a symmetry generator is then replaced by the operator P B, which E. Implications for spontaneous breaking of satisfies translational invariance B B B [Pi ,Pj ] = −i~qijkBkQ0, [Pi ,Q0] = 0, (13) In the preceding subsections, we saw how the general theorems known in literature can be extended to account where q is the charge of the particles and Q0 the U(1) for the possibility that the ground state of the system Noether charge, that is, their total number. Note B B has only a discrete translational symmetry. We are par- that the rank of the matrix ρ defined by iρij ≡ ticularly interested in the situation when this arises as a −1 B B lim Ω h0|[Pi ,Pj ]|0i equals two if the magnetic field result of a spontaneous breakdown of a continuous trans- Ω→∞ lational invariance. We will therefore devote this subsec- as well as the expectation value of Q0 is nonzero (other- tion to some discussion of systems with the symmetry wise the rank is zero). Eq. (12) then suggests that in case breaking pattern magnetic translational invariance is spontaneously broken, there is one less NG boson than naively expected (d) 0 (d0) (d−d0) Gint × GC → Gint × GD × GC , (11) based on the number of broken translation generators. We will see an example of this in Section IV [21]. where Gint describes the internal symmetry group, which can also be (partially) spontaneously broken. First, note that the Noether charge of translational invariance, that is, the momentum operator, commutes III. NG MODES IN A SUPERSOLID with all other broken charges Qa. If we for a mo- ment assume, as in the theorem of Schäfer et al., that In this section, we will illustrate the general results h0|[Qa,Qb]|0i = 0 for all pairs of broken internal symme- obtained above on the example of NG modes in a (fer- try generators, we immediately conclude that the number romagnetic) supersolid in d spatial dimensions. We will of NG bosons equals the number of broken generators. employ the mean-field approximation at zero tempera- Therefore, there is one NG boson for each spontaneously ture. broken translation generator as well as for each spontaneously broken generator of an internal symmetry. In the general case, it is convenient to define the an- A. Formalism −1 tisymmetric matrix ρ by iρab ≡ lim Ω h0|[Qa,Qb]|0i Ω→∞ where a, b = 1, . . . , nBS. In the recent paper [8], we pro- 1. Definition of the model and its symmetry posed that the number of NG bosons is generally related to the rank of the matrix ρ by Let us consider the class of theories, defined by the 1 following Lagrangian, nBS − nNG = 2 rank ρ. (12) The fact that translation generators commute with ev- Z 2 2 d † ~ ∇ erything else now implies that rank ρ = rank ρ , where L[Ψ] = d x Ψ (x) i ∂t + + µ Ψi(x) int i ~ 2m ρ is the analogous matrix defined using the generators int Z (14) of broken internal symmetries only. We can thus infer 1 d d † † − d x d y Ψ (x)Ψi(x)V (x − y)Ψ (y)Ψj(y), that the counting of NG bosons of spontaneously broken 2 i j translational invariance is not affected by the presence of other spontaneously broken, internal symmetries: each where Ψ = (Ψ1,..., ΨN ) is an N-component complex spontaneously broken translation generator gives rise to scalar field, and the potential V (x) = V (−x) repre- exactly one NG boson. In Section III, we will discuss sents a finite-range interaction. The symmetry of the (d) examples of both types of systems where rank ρ is zero Lagrangian is obviously U(N) × GC . The simplest case and nonzero. N = 1 corresponds to the usual Bose–Einstein conden- So far, we have tacitly assumed that the generators sate (BEC) without internal degrees of freedom, while of space translations commute with one another as well the widely discussed N = 2 case can be understood as a as with other symmetry generators. Nevertheless, in spin-1/2 BEC (see, for instance, Ref. [22]). presence of external fields, this property can be violated It is convenient to analyze the model using dimension- while still maintaining well-defined translational symme- less variables; this makes the results universally applica- try. Let us consider, for instance, a system of (inter- ble to different physical systems, as long as they share acting) charged particles in a uniform external magnetic the same set of symmetries and degrees of freedom. We 6

first introduce a characteristic length scale of the poten- be overwhelmed by the energy gain from the nonlocal tial, a, in terms of which the strength of the interac- interaction. As a result, inhomogeneous field configuration can be measured by the parameter V0, defined by tions may become energetically favorable. d R d a V0 = d x V (x). Next, we rescale the spacetime coordinates as t → t~/µ and x → xa. (For conventional reasons we use the same symbols for the new, dimensionless coordinates; as we do so systematically from now on, no confusion can arise.) Finally, we introduce the 3. Spontaneous symmetry breaking dimensionless potential, v(x) = V (x)/V0, and the field p d ψ(x) = Ψ(x) V0a /µ. The dimensionless Lagrangian, L[ψ] = L[Ψ]V /µ2, thus becomes As soon as the classical field ψ0(x) assumes a nonzero 0 value, the symmetry of the Lagrangian is spontaneously Z 2 broken. Owing to the fact that the spinor ψ0(x) points d † ∇ L[ψ] = d x ψi (x) i∂t + + 1 ψi(x) in the same direction everywhere in space, ψ0i(x) = 2g ¯ Z (15) δi1ψ0(x), the internal U(N) symmetry is broken down to 1 d d † † its U(N − 1) subgroup; 2N − 1 of its generators are thus − d x d y ψi (x)ψi(x)v(x − y)ψj (y)ψj(y), 2 broken. Therefore, when ψ0(x) is homogeneous the resid- (d) where g is a dimensionless coupling parameter defined as ual symmetry is U(N − 1) × GC and nBS = 2N − 1. We 2 2 −1 will call this solution the superfluid phase. On the other g = µ(~ /ma ) . hand, if ψ0(x) is inhomogeneous and periodically modulated in all spatial directions (the supersolid phase), the (d) 2. Classical field symmetry of the ground state is merely U(N − 1) × GD and nBS = 2N − 1 + d. The first step in the mean-field analysis is to find the To predict the number of NG modes, we have to ground state, that is, the static classical field configura- evaluate the rank of the matrix ρ in Eq. (12). To tion ψ0(x) which minimizes the energy, that end, note that the 2N − 1 broken internal symmetry generators fall into three classes: the generator Z 1 E[ψ] = ddx ∇ψ∗(x) · ∇ψ (x) − ψ∗(x)ψ (x) of phase transformations of the first component of the 2g i i i i spinor, (T1)jk = δj1δk1/2; the N − 1 real generators Z R 1 d d ∗ ∗ (T` )jk = (δk1δj` + δj1δk`)/2; the N − 1 pure imag- + d x d y ψi (x)ψi(x)v(x − y)ψj (y)ψj(y). I 2 inary generators (T` )jk = i(δk1δj` − δj1δk`)/2, where (16) ` = 2, 3,...,N. Using these matrices, the broken generators can be represented on the Hilbert space of the To that end, we exploit the symmetry of the problem and R d † system by the operators Qa = d x ψi (x)(Ta)ijψj(x). cast the spinor ψ0(x) in a “canonical” form. One should Within the mean-field approximation, the fluctuations nevertheless be careful since the field in general depends are neglected and one easily derives on the coordinates. Note that it can always be written as ψ(x) = U(x)ψ0(x), where U(x) is a unitary matrix 0 ¯ ¯ 0 and ψ (x) = δi1ψ(x) with real ψ(x). As long as the field 1 R I iδ`` i h0|[Q ,Q 0 ]|0i = h0|Q |0i ψ(x) does not contain any vortices, both U(x) and ψ¯(x) Ω ` ` Ω 1 Z (18) are differentiable. It then follows that iδ``0 d ¯ 2 = d x ψ0(x) 2Ωuc uc N Z ¯ 2 2 X ψ(x) |∇Ui1(x)| E[ψ] − E[ψ0] = ddx ≥ 0. (17) 2g i=1 0 for `, ` = 2,...,N and h0|[Qa,Qb]|0i = 0 for other com- binations. We thus conclude that (1/2)rank ρ = N − 1. The equality occurs only if Ui1 is coordinate-independent. Thus, by a global U(N) rotation, ψ0(x) can always be In the case N = 1 (usual BEC), we can now use the ¯ ¯ chosen as ψ0i(x) = δi1ψ0(x) with a real field ψ0(x). Field theorem of Schäfer et al. to assert that nNG = nBS = configurations with vortices tend to have a higher energy 1 in the superfluid phase and nNG = nBS = 1 + d in and we therefore assume we can discard this possibility. the supersolid phase. For N ≥ 2, Eq. (12) implies that One (nontrivial) stationary point of the energy func- nNG = N in the superfluid phase and nNG = N + d in tional (16) is always given by a uniform field. Due to the supersolid phase. To be consistent with the Nielsen– our normalization, R ddx v(x) = 1, this corresponds to Chadha theorem, at least N −1 of the NG modes must be ¯ ψ0(x) = 1 with the energy density −1/2. The uniform type-II. This is in agreement with the result of Ref. [23] solution indeed minimizes energy for small g. However, which analyzed the relativistic version of the model with when g becomes large, the energy cost of creating spatial a local interaction, V (x) ∝ δd(x). The detailed nature modulation of ψ(x) is suppressed. Depending on the de- of the NG modes in our model will be investigated in the tailed structure of the potential v(x), this cost can then following subsection. 7

4. Elementary excitations (analogously for ϕi) with a band index n and a crystal momentum k ∈ FBZ. The solution ωn(k) represents the In the mean-field approximation, the excitation spec- allowed excitation spectrum. In particular, those modes trum is found simply by expanding the Lagrangian to sec- which satisfy ωn(0) = 0 are NG modes, assuming they ond order in fluctuations around the classical field. We couple to the spontaneously broken currents. parameterize the field as ψ(x) = ψ0(x) + φ(x) + iϕ(x) and plug this back into Eq. (15). Up to second order in the fluctuation fields φ, ϕ, the Lagrangian reads B. Numerical results L[ψ] = L[ψ0] + L2[φ, ϕ; ψ0] + ··· , where Z As a concrete illustration of the general setup, we d n L2[φ, ϕ; ψ0] = d x φi(x)Kxφi(x) + ϕi(x)Kxϕi(x) choose the simplest model for a supersolid, defined by the the step-function potential V (x) = (V0/π)θ(a − |x|) o + ϕi(x)∂tφi(x) − φi(x)∂tϕi(x) in two spatial dimensions [24]. Below, we summarize the Z results of our numerical computations. In order not to d d obscure the physics with technicalities, we describe some − 2 d x d y φ1(x)w(x, y)φ1(y) details of the methods we used in Appendix C. In partic- (19) ular, in Appendix C 1 we explain how the classical field ¯ up to a term which is a total time derivative. In the above ψ0(x) was found, and in Appendix C 2 how the band 2 structures presented in Figs. 2(a) and 2(b) were obtained. equation, Kx = (∇ /2g) + 1 − u(x) is the kinetic term, R d ¯ 2 u(x) = d y v(x − y)|ψ0(y)| is the effective periodic mean-field potential due to the background ψ0(x), and ¯ ¯ 1. Superfluid phase w(x, y) = ψ0(x)v(x−y)ψ0(y) captures the effects of the non-local interaction. N N Physically, the fluctuations {φ } and {ϕ } corre- For g < g0 ≈ 38.4, the energy (16) is minimized by the i i=2 i i=1 ¯ spond to NG fields of the spontaneously broken internal homogeneous solution ψ0(x) = 1 [25]. In this superfluid N phase, the Schrödingerequations (21) are solved straight- symmetry. Namely, the {φi}i=2 are obtained from the classical field ψ0(x) by real rotations, generated by the forwardly by going to the energy–momentum space, tak- I N T matrices Ti , while the {ϕi}i=1 are induced by symmetric ing the form Ai(k) φi(k), ϕi(k) = 0 (separately for R unitary transformations, generated by the matrices Ti i = 1,...,N), where (or T1 in case of i = 1). On the other hand, the φ1(x) 2 mode corresponds to displacements of the classical field (k /2g) + 2δi1v(k) −iω(k) Ai(k) = 2 . (23) configuration, ψ0(x + ξ(x)) − ψ0(x) ' ξ(x) · ∇ψ0(x), as iω(k) k /2g well as to density fluctuations. This can be seen by not- R 2 −ik·x+iω(k)t ing the role of φ1(x) in the density–density correlation Here φi(k) = d x φi(0, x)e [analogously R 2 −ik·x function, for ϕi(k)] and v(k) = d x v(x)e = 2J1(|k|)/|k| [J (x) is the Bessel function]. From the condition h0|T {δn(x)δn(y)}|0i n (20) det Ai(k) = 0 we find the quasiparticle dispersion re- ' 4ψ¯ (x)ψ¯ (y)h0|T {φ (x)φ (y)}|0i + ··· , p 2 2 0 0 1 1 lation, ωi(k) = (k /2g) [(k /2g) + 2δi1v(k)]. Around † † k = 0, the dispersions and eigenvectors corresponding to where δn(x) = ψi (x)ψi(x) − h0|ψi (x)ψi(x)|0i, and the ellipsis stands for terms of higher order in the fluctuation the individual modes are expanded as √ fields. ω (k) ' |k|/ g, φ , ϕ ' 0, 1 , (24) The dispersion relations of the various excitation 1 1 1 ω (k) ' k2/2g, φ , ϕ ' √1 1, −i , (j ≥ 2), branches are determined formally by a diagonalization of j j j 2 the Lagrangian (19). In practice, we accomplish this task by solving the Euler–Lagrange equations for the fluctua- to leading order in momentum. We illustrate these dis- tion fields, persion relations in Fig. 1(a) for the coupling set to the Z transition point, g = g0 − 0. In summary, there is one d ∂tϕi(x) = Kxφi(x) − 2δi1 d y w(x, y)φ1(y), type-I Bogoliubov mode and N − 1 type-II modes with (21) a quadratic dispersion and a circular polarization, which ∂tφi(x) = −Kxϕi(x). can be interpreted as ferromagnetic magnons (precession modes) [26]. This conclusion is in exact agreement with Note that the equations of motion for φi and ϕi are coupled; we will see the importance of this coupling later. the prediction we made in Section III A 3. It is worth emphasizing that while at k = 0 the Bogoli- Based on the residual discrete symmetry, G(d), we can D ubov mode corresponds to a pure phase fluctuation, ϕ1, assume the normal modes of φi, ϕi to have the Bloch at nonzero momentum it is given by a mixture of ϕ and form, that is, 1 φ1. In other words, the Bogoliubov mode generates not X i(k+G)·x−iωn(k)t φi,n(k, x) = φi,n,G(k) e , (22) only a phase modulation but also a density modulation of G the condensate ψ0. This is why we can associate it with a 8

ΩHkL

1.5

1.0

0.5

k 2 4 6 8 10 12 (b) (a)

FIG. 1. (color online). (a) The dispersion relations of NG (a) bosons in the superfluid phase at g = g0 − 0. The linear branch is the usual Bogoliubov mode, while the quadratic branch (with N − 1 fold degeneracy) can be interpreted as a ferromagnetic magnon. The dispersion of the Bogoliubov mode has a roton-like minimum around |k| ≈ 4.80, which corresponds to the magnitude of the reciprocal lattice vector at g = g0 + 0. (b) Three-dimensional plot of the classical field ψ¯0(x) in the supersolid phase. The energy functional is minimized by a hexagonal (triangular) lattice with the lattice constant approximately 1.50 at g = g0 + 0. Note that ψ¯0(x) does not have any nodes. density wave and detect it in Bragg spectroscopy experi- (b) ments [27]. However, this very fact has also caused some confusion regarding the independence of the Bogoliubov FIG. 2. The numerical result for dispersion relations in the su- mode and the longitudinal phonon originating from the persolid phase at g = g0 +0. Panel (a) displays the excitation crystalline order in the supersolid phase. We will shortly branches in the φ1, ϕ1 sector. We can see three linear disper- see that these two are, indeed, distinct modes. sions around the Γ point, which correspond to the Bogoliubov Finally, let us remark that the above discussed mix- mode and two phonons. The inset shows the irreducible part † of FBZ. Panel (b) displays the excitation branches in the φi, ing can be traced back to the term ψ ∂tψ in the La- ϕi (i ≥ 2) sector. The dispersion relation of the gapless mode grangian (15). A similar term also arises in the analogous is quadratic at low momentum, as can be seen in the inset relativistic model where it is induced by the chemical po- which zooms in the vertical axis. tential [23]. On the other hand, in a Lorentz-invariant model where SSB is triggered by a “wrong sign” of the mass term of the field, no such mixing would arise, and using the first 73 G vectors around the origin of the recip- the φi and ϕi modes would be completely decoupled. rocal lattice. The convergence of the result was checked by a comparison with one using only 61 G vectors. In the i = 1 sector, three of the excitation branches 2. Supersolid phase are linear around the Γ point (the origin of the k space), corresponding to a longitudinal phonon, a transverse For g > g0, periodic spatial modulation of the order phonon, and a Bogoliubov mode. On the other hand, ¯ parameter ψ0(x) is energetically favored. The transi- each i ≥ 2 sector contains just one massless mode with a tion at g = g0 is of first order within the mean-field quadratic dispersion relation—the magnon. In summary, approximation [24], and can be roughly understood as there are in total N + d NG modes, N − 1 of which are Bose–Einstein condensation of field modes with momen- type-II, exactly as predicted in Section III A 3. tum determined by the reciprocal lattice vectors. This In addition, we have checked numerically that the dis- interpretation is supported by the roton-like minimum in persion relations of all the gapless modes are isotropic the dispersion relation of the Bogoliubov mode shown in at the Γ point; the NG modes of the model will thus be Fig. 1(a). Numerical computation, following the strategy described by a rotationally invariant low-energy effective outlined in Appendix C 1, shows that the solution with field theory. Our findings are consistent with previous the lowest energy has the form of a hexagonal lattice, works for the case N = 1, based on the effective La- plotted in Fig. 1(b). grangian method [28]. A low energy effective field theory By solving the coupled Schrödingerequations (21) us- describing simultaneously the Bogoliubov mode and the ing the method explained in Appendix C 2, we found the phonons has recently also been worked out, in a different band structures in the supersolid phase at g = g0 + 0, context, in Ref. [29]. shown in Figs. 2(a) and 2(b). This result was obtained We would like to emphasize that while the specific re- 9 sults presented in Figs. 2(a) and 2(b) rely on the mean- lattice node, xi = Ri. The spectrum of oscillations field approximation, the qualitative conclusion that the of the crystal lattice can be determined by resorting Bogoliubov mode and the longitudinal phonon are dis- to the harmonic approximation, in which the Hamilto- tinct physical excitations is exact. Following the gen- PN 2 nian takes the form H = i=1[pi − qA(xi)] /(2m) + eral argument of Section II E, in the case of N = 1 it is PN αβ (m/2) W uiαujβ, where ui ≡ xi − Ri and based on the theorem of Schäfer et al. and thus can be i,j=1 ij considered rigorously proven. For N ≥ 2, it is strongly  supported by the conjectured equality (12). −∂α∂βv , (i 6= j),  x=Rij  mW αβ = N (26) ij X  ∂α∂βv , (i = j).  x=Ri` IV. PHONONS IN A CHARGED CRYSTAL IN `=1 EXTERNAL MAGNETIC FIELD `6=i

(We have also introduced the notation Rij = Ri − Rj.) As anticipated in the discussion in Section II E, the Let us now assume that the total particle number N ex- spectrum of NG modes can become more complicated tends to infinity or that periodic boundary conditions are once the generators of space translations do not commute imposed. Then the Heisenberg equations of motion— with each other as well as with generators of other, in- unlike the Hamiltonian itself—have the lattice transla- ternal symmetries. To illustrate this subtlety, we shall in tional symmetry, since only the magnetic field B ap- this section discuss briefly a simple quantum-mechanical pears in them. Thus, by using the Fourier compo- model for a charged crystal. αβ P αβ −ik·Rij nents, Wk = i Wij e for any fixed j, and P iω(k)t−ik·Ri ukα(t) = i uiαe , we simplify the equations 2 A. Quantum-mechanical model to the form Mαβ(k)ukβ = 0 with Mαβ(k) = δαβω(k) − γ αβ γ γ iαβγ ω(k)b −Wk and b = qB /m. As before, the condition det M(k) = 0 determines the dispersion relations The Hamiltonian for N particles with the common of the normal modes of the crystal lattice. charge q in non-relativistic quantum mechanics is given For the sake of brevity, we shall consider the simplest, by albeit unrealistic, model—the isotropic crystal in three

N 2 spatial dimensions, where X [pi − qA(xi)] 1 X H = + v(x − x ), (25) 2m 2 i j kαkβ kαkβ i=1 i6=j W αβ = δαβ − ω (|k|)2 + ω (|k|)2. (27) k k2 t k2 ` where A(xi) is a vector potential which determines the external magnetic field B(x) via ∇ × A(x) = B(x), and In the absence of the magnetic field, the functions ωt(|k|) v(x) is a repulsive interaction potential which is assumed and ω`(|k|) give the dispersion relations of transverse to be invariant under space inversion, v(x) = v(−x). and longitudinal phonons, respectively, and it is there- This model, with the Coulomb potential v(x) = e2/|x|, fore natural to assume both of them to be linear at low has been used in the context of the Wigner solid [30]. momentum. The determinant of M(k) is easily found, Nevertheless, since our goal here is to discuss the conse- 2 2 2 2 2 2 2 2 2 2 quences of SSB in the presence of a magnetic field, we det M =(ω − ωt ) (ω − ω` ) − ω (ω − ωt )b sin θ 2 2 2 2 2 will assume that the potential v(x) has a finite range. − ω (ω − ω` )b cos θ, (28) The canonical coordinate and momentum xiα, pjβ are required to satisfy the usual commutation relation. where θ is the angle between the vectors k and b, and (Within this section, we use Greek letters to denote we have for simplicity omitted the argument k where spatial indices.) Because of the vector potential, the necessary. At low momentum, the dispersion relations of PN the three normal modes can be evaluated analytically, operator of total momentum i=1 pi no longer commutes with the Hamiltonian. When the magnetic field ω ' |b|, gapped mode, is uniform, we can instead define the operator ρi ≡ B PN q pi − q [A(xi) − B × xi]. The sum, P = ρiα, is 2 2 2 2 α i=1 ω ' ωt sin θ + ω cos θ, type-I NG mode, (29) conserved and plays the role of the generator of spatial ` 2 B ω ω` translations ([iPα , xiβ] = ~δαβ). At the same time, it t ω ' q , type-II NG mode. satisfies the commutation relation (13) with Q0 = N. 2 2 2 2 |b| ωt sin θ + ω` cos θ

In order to determine the polarization of the modes, that B. Phonon spectrum is, the orientation of the vector uk, it is convenient to rewrite the equation of motion in a vector form Suppose that we have found a stable crystal config- 2 2 2 uration where the i-th particle is localized around the ω u − iωu × b − (ωt Ptu + ω` Pù) = 0, (30) 10 where Pt,` are projectors to transverse/longitudinal di- the angular velocity Ω). It then has formally the same rections with respect to momentum k, defined by the form as the Hamiltonian of charged particles in the exter- respective terms in Eq. (27). For the gapped mode, the nal magnetic field, provided the symmetric gauge is used 2 terms in Eq. (30) proportional to ωt,` can be neglected, and the replacement Ω → qB/(2m) made [31]. There- which leads to the condition u = iu×b/|b|. In words, this fore, the Hamiltonian of the rotating Bose gas has a two- mode is circularly polarized in the plane perpendicular dimensional continuous magnetic translational symme- to the magnetic field, in the opposite direction than the try, in addition to the U(1) global symmetry correspond- usual cyclotron motion. For the type-I NG mode, we like- ing to the conservation of particle number, as long as the wise neglect terms in Eq. (30) proportional to ω2 as well trapping potential (modified by Ω) can be neglected. 2 as ωt,` to arrive at the constraint u×b = 0, which simply The formation of a triangular vortex lattice has been means that this mode is always polarized along the di- experimentally observed [32], which implies spontaneous rection of the magnetic field. Finally, for the type-II NG breaking of the continuous magnetic translational invari- mode we neglect the term in Eq. (30) proportional to ω2 ance. The collective mode due to the lattice distortion, 2 2 and obtain the condition u × b = (i/ω)(ωt Ptu + ω` Pù). called the Tkachenko mode, has also been experimen- Let us for illustration consider some special cases. tally observed [33]. Similarly to our toy model, there When the momentum is parallel to the magnetic field is one gapless excitation with a quadratic dispersion re- (θ = 0), the energies of the modes given in Eq. (29) lation and one gapped mode with the cyclotron gap, 2 reduce to |b|, ω`, and ωt /|b|, respectively. The type-II qB/m = 2Ω [34]. The discussion in Section II E suggests NG excitation is circularly polarized in the plane perpen- there is still an independent Bogoliubov mode stemming dicular to b, in the opposite direction than the gapped from spontaneous breaking of the U(1) symmetry. We mode. This is in a nice analogy to the type-II NG boson believe that our general arguments based solely on sym- in ferromagnets—the magnon—where the circular polar- metry properties may lead to a new understanding of NG ization corresponds to the Larmor precession of the elec- modes in the vortex lattice. tron spin. Note that in the limit when ωt = ω`, the same conclusions hold regardless of the orientation of the momentum. V. SUMMARY AND CONCLUSIONS Second, when the momentum is perpendicular to the magnetic field (θ = π/2), the dispersion relations in In the present paper, we have developed a framework Eq. (29) become |b|, ωt, and ωtω`/|b|, respectively. The for the analysis of spontaneous symmetry breaking in the type-II NG mode is now polarized elliptically in the plane case that there is only a discrete translational invariance. transverse to the magnetic field. The ratio of the princi- While our prime motivation was to study spontaneous pal axes of the ellipse, oriented along and perpendicularly breaking of continuous translational invariance, our re- to the momentum, is given by ωt/ω`. sults apply equally well to systems where this is broken In the general case, as already mentioned above, the explicitly by construction. polarization of the gapped mode is circular and trans- We generalized existing theorems on NG bosons in verse to the magnetic field, while that of the type-I NG quantum many-body systems, thus completing the pro- mode is linear and aligned with the magnetic field, irre- gram initiated in our recent paper [8]. Our main message spective to the orientation of the momentum. Only the is that as long as SSB is concerned, the usual division of polarization of the type-II NG excitation is affected by symmetries into internal and spacetime ones is artificial. the momentum. Instead, we propose that all symmetries that are uni- In any case, there are only two (that is, d − 1) gapless form in the sense defined in Section II can, and should, phonons, and one of them has a quadratic dispersion re- be treated on the same footing. This naturally includes lation. This can be explained by the combination of the (uniform) internal symmetries as well as translational in- conjecture (12) and the Nielsen–Chadha theorem. The variance. non-commutativity of the components of the generator To illustrate our general arguments, we analyzed a of magnetic translations suggests that there is one less model for a (ferromagnetic) supersolid in two spatial phonon than one would naively expect. Then, to satisfy dimensions using the mean-field approximation at zero the Nielsen–Chadha theorem, at least one of the phonons temperature. Using numerical computations, we exam- must be type-II. ined the regime of weak coupling where the model fea- tures a homogeneous superfluid state, as well as the strong-coupling one where a supersolid crystalline order C. Possible application emerges. The properties of the spectrum of NG modes agree with our general conclusions in both phases. In par- The second-quantized version of the above model can ticular, we have demonstrated that the Bogoliubov mode be used to describe the rapidly rotating BEC of cold and the longitudinal phonon in the supersolid state are atoms. When viewed from the co-rotating frame, the distinct physical modes. Hamiltonian of the system picks an additional term, Finally, we discussed the effect of an external magnetic equal to −L·Ω (coupling of the angular momentum L to field on the translational properties of the system. Using 11 a simple quantum-mechanical toy model, we argued that Appendix A: Derivation of Eq. (3) this changes the spectrum of NG modes qualitatively by making the dispersion relation of some of them quadratic at low momentum, and giving a gap to others. This issue The derivation of Eq. (3) is straightforward, if some- would certainly deserve a more detailed investigation in what tedious. It bases solely on the partition of unity (2), a quantum-field-theoretic setting, with respect to its in- translational property (1), definition of the averaged trinsic interest as well as the potential application to the charge density in Eq. (4), and some elementary integrals. We would also like to first remind the reader of the Pois- physics of rotating Bose gases. 0 son identity for the periodic δ-function in Rd ,

ACKNOWLEDGMENTS Ωuc X X 0 eik1·R = δd (k − G), (A1) (2π)d0 1 R G We are grateful to H. Abuki, Y. Akamatsu, G. Anagama, H. Aoki, S. Endo, Y. Kato, Y. Kawaguchi, M. Kunimi, D.-H. Lee, and E. B. Sonin for useful dis- where G are vectors from the reciprocal lattice, G = Pd0 cussions. The work of T.B. was supported by the Sofja i=1 nibi (ni ∈ Z and ai · bj = 2πδij). Taking now the Kovalevskaja program of the Alexander von Humboldt ﬁrst term in the commutator in Eq. (3), we obtain by a Foundation. series of manipulations,

0 where we denoted q = (q1, k2). In the last step, we Appendix B: Goldstone theorem in the effective used the fact that, similarly to the coordinate decompo- action formalism sition x1 = x¯1 + R, any vector k1 from the reciprocal space can be uniquely expressed as a sum of a vector In Appendix B of the paper [8] we provided a proof of from FBZ and a vector G from the reciprocal lattice. the Goldstone theorem based on the quantum effective As long as k1 already lies in FBZ, the only solution to action in a form applicable to nonuniform symmetries. the condition q1 − k1 − G = 0 is q1 = k1 and G = 0. However, we for the sake of simplicity assumed contin- The second contribution to the commutator in Eq. (3), R d+1 0 i(ωt−k·x) uous translational invariance of the ground state. Our d x h0|φb(0)ja(x)|0i e , can be calculated in aim here is to relax this constraint. Like in the rest of the same way and we therefore skip the details. this paper, we will assume that the spontaneously broken symmetry is uniform; the extension of our argument to the fully general case is straightforward though. Consider a theory of a set of (not necessarily scalar) fields, φi(x), whose classical action is invariant under the 12

−1 infinitesimal transformation δφi(x) = θFi[φ(x)]. Here θ that the condition of G being inverse to G reads R d+1 −1 d+1 is a parameter of the transformation and Fi is a local d z Gik(x, z) Gkj (z, y) = δijδ (x − y), which takes functional of the fields that does not depend explicitly the following form in the frequency–momentum space, on the coordinate x. (This is equivalent to the require- Z d0 ment that the symmetry be uniform.) Provided that Fi d z¯1 −1 Gik(ω, k; x¯1, z¯1) Gkj (ω, k; z¯1, y¯1) is linear in the fields, the quantum effective action, Γ[φ], uc Ωuc (B5) shares the symmetry of the classical theory [19]. We then 0 = δ Ω δd (x¯ − y¯ ). obtain the invariance condition ij uc 1 1 Z 2 d+1 δ Γ[φ0] Multiplying this equation by Fj[φ0(y¯1)] and integrating d y Fj[φ0(y)] = 0, (B1) δφi(x)δφj(y) over y¯1, we assert with the help of Eq. (B3) and the where the integration measure involves time as well as spectral representation (B4) that as soon as Fj[φ0(x¯1)] is nonzero at some x¯1, the propagator (at the same point) d spatial coordinates and φ0i(x) denotes the vacuum expectation value of the field in accord with the nota- must have a pole such that εξ(0) = 0. tion introduced in Section III A 2. The second functional Finally, let us remark that the proof would have been derivative of the effective action here represents the exact technically much simpler had we considered the effective −1 action (and in turn the propagator) as a functional of inverse propagator of the theory, Gij (x, y). Assuming as in the rest of the paper a discrete translational invariance fields spatially averaged over the Wigner–Seitz unit cell, d0 d−d0 in and continuous one in as well as in time, this 0 R R Z dd x¯ can be Fourier transformed as φ (t, R, x ) ≡ 1 φ (t, x¯ + R, x ). (B6) i 2 Ω i 1 2 Z +∞ dω Z ddk uc uc G−1(x, y) = e−iω(tx−ty ) ij 2π (2π)d −∞ (FBZ) (B2) The order parameter φ0i would then be completely time i[k1·(Rx−Ry )+k2·(x2−y2)] −1 and coordinate independent and the left-hand sides of × e G (ω, k; x¯1, y¯1), ij Eqs. (B3) and (B5) would become mere products in mo- where k1 lies in FBZ. Note that the Fourier transform mentum space. The price for this simplification would be of the propagator still depends on the coordinates within that we could make no conclusions about the spectrum in the Wigner–Seitz unit cell as a result of the lack of full the cases that the symmetry is spontaneously broken but continuous translational invariance. Thanks to the as- the spatial average of the order parameter vanishes. The sumed periodicity of the ground state, φ0i(y) = φ0i(y¯1), above given proof is more general since it only assumes Eq. (B1) becomes that the order parameter is nonzero at some point. Z d0 d y¯1 −1 Gij (0, 0; x¯1, y¯1)Fj[φ0(y¯1)] = 0. (B3) uc Ωuc Appendix C: Calculational methods For a translationally invariant ground state, Fj[φ0(y¯1)] is a constant and it follows immediately that the inverse In this appendix, we collect some details of the tech- propagator must have a gapless pole at zero momen- niques used in the numerical computations in Section III. tum once the symmetry is spontaneously broken, that is, Fj[φ0] is nonzero. However, in presence of a spatially varying order parameter, the implications of Eq. (B3) for 1. Minimization of the energy the excitation spectrum are less straightforward. It is useful to first recall the spectral representation of To minimize the energy functional (16), we exploit the propagator. Using the definition of the propagator in the assumed periodicity of the classical field and per- the operator formalism, iGij(x, y) = h0|T {φi(x)φj(y)}|0i, form the Fourier decomposition of the field, ψG = and inserting the partition of unity (2), one arrives at the −1 R d −iG·x Ωuc uc d x ψ(x)e . The total energy divided by the following expression for the Fourier transformed propa- space volume is then equal to the energy density averaged gator, defined analogously to Eq. (B2), over the unit cell, which is given by X Gij(ω, k; x¯1, y¯1) = 2 ¯ X G ∗ ξ E[ψ] = − 1 ψiGψiG 2g h0|φi(0, x¯1, 0)|ξ, kihξ, k|φj(0, y¯1, 0)|0i G × (B4) (C1) 1 1 X ∗ ∗ ω − εξ(k) + i0 + ψ ψ v(q)ψ 0 ψ 0 . ~ 2 iG iG−q jG jG +q 0 h0|φj(0, y¯1, 0)|ξ, −kihξ, −k|φi(0, x¯1, 0)|0i G,G q − 1 . ω + εξ(−k) − i0 ~ Thanks to the parity invariance of the Lagrangian, which This ensures that even with just a discrete transla- is naturally assumed to be shared by the ground state, we tional invariance, one can still extract the quasipar- can restrict our attention to even-parity classical fields, ticle spectrum from the poles of a conveniently de- ψ0(x) = ψ0(−x). This, together with the reality of the ∗ fined propagator. To conclude our argument, we note classical field, implies the relation ψG = ψ−G = ψG, 13 which significantly reduces the number of variables in where the (infinite-dimensional) matrices K, T and W the minimization procedure. are defined by The minimization of the averaged energy density is accomplished by first fixing the primitive lattice vec- d ¯ 2 tors {ai}i=1 and minimizing E[ψ] with respect to the (k + G1) KG1G2 = − 1 δG1G2 + uG1−G2 , Fourier components {ψG}. This determines the function 2g E¯({ai}) ≡ E¯[ψ0({ai})], which is subsequently minimized TG1G2 = iω(k)δG1G2 , (C3) with respect to {ai}. The actual crystal form of the clas- X 0 ¯ 0 ¯ 0 sical field ψ0(x) is thus found. WG1G2 = ψ0G1−G ψ0G −G2 v(k + G ), G0

2. Calculation of the band structures P ¯ ¯ and uG = v(G) G0 ψ0G−G0 ψ0G0 . Because Eq. (C2) The Schrödinger equation (21) can be rewritten in represents a set of homogeneous linear equations for the Fourier space as Fourier components φiG and ϕiG (for each fixed i), the band structure can be found from the condition that the ∗ determinant of the matrix of coefficients in these equa- TG G ϕiG2 = −(KG1G2 + 2δi1WG1G2 )φiG2 , 1 2 (C2) tions be zero. TG1G2 φiG2 = −KG1G2 ϕiG2 ,

[1] J. Goldstone, Nuovo Cimento, 19, 154 (1961); J. Gold- [16] In fact, we should further distinguish which of the first d0 stone, A. Salam, and S. Weinberg, Phys. Rev., 127, 965 coordinates are associated with continuous translational (1962). invariance that is spontaneously broken, and which with [2] G. S. Guralnik, C. R. Hagen, and T. W. B. Kibble, in an intrinsically discrete translational symmetry. Since Advances in Particle Physics, Vol. II, edited by R. L. identities such as Eq. (1) only rely on the commutation Cool and R. E. Marshak (Wiley, New York, NY, USA, relations of the Noether charge(s), the former type of co- 1968) pp. 567–708. ordinates actually possess a stronger, “continuous” trans- [3] H. B. Nielsen and S. Chadha, Nucl. Phys. B, 105, 445 lational property. It is good to keep this in mind, it will (1976). nevertheless not be essential for the discussion to follow. [4] T. Schäfer,D. T. Son, M. A. Stephanov, D. Toublan, and [17] R. V. Lange, Phys. Rev. Lett., 14, 3 (1965); Phys. Rev., J. J. M. Verbaarschot, Phys. Lett. B, 522, 67 (2001). 146, 301 (1966). [5] H. Leutwyler, Phys. Rev. D, 49, 3033 (1994). [18] C. B. Clark, Phys. Rev., 109, 1133 (1958); R. Brout, [6] T. Brauner, Symmetry, 2, 609 (2010). 113, 43 (1959). [7] H. Leutwyler, Helv. Phys. Acta, 70, 275 (1997). [19] S. Weinberg, The Quantum Theory of Fields, Vol. II [8] H. Watanabe and T. Brauner, Phys. Rev. D, 84, 125013 (Cambridge University Press, Cambridge, UK, 1996). (2011). [20] Strictly speaking, what they proved is that the number of [9] P. Fulde and R. A. Ferrell, Phys. Rev., 135, A550 (1964); NG modes is equal to or greater than the number of bro- A. I. Larkin and Y. N. Ovchinnikov, Sov. Phys. JETP, ken symmetries. Here we supplement the theorem with 20, 762 (1965). the remaining half of the proof. [10] R. Casalbuoni and G. Nardulli, Rev. Mod. Phys., 76, 263 [21] As we observe in Section IV, the magnetic translations (2004). are actually not uniform and one should therefore be very [11] A. F. Andreev and I. M. Lifschitz, Sov. Phys. JETP, 29, careful trying to apply our conjecture (12) to this case. 1107 (1969); G. V. Chester, Phys. Rev. A, 2, 256 (1970); In fact, the magnetic translational symmetry is always E. Kim and M. H. W. Chan, Nature, 427, 225 (2004); broken for systems with nonzero charge density. Science, 305, 1941 (2004). [22] K. Kasamatsu, M. Tsubota, and M. Ueda, Int. J. Mod. [12] C. Wang, C. Gao, C.-M. Jian, and H. Zhai, Phys. Rev. Phys. B, 19, 1835 (2005). Lett., 105, 160403 (2010). [23] J. O. Andersen, (2005), arXiv:hep-ph/0501094. [13] A. W. Overhauser, Adv. Phys., 27, 343 (1978); D. Derya- [24] Y. Pomeau and S. Rica, Phys. Rev. Lett., 72, 2426 gin, D. Y. Grigoriev, and V. Rubakov, Int. J. Mod. Phys. (1994). A, 7, 659 (1992); E. Shuster and D. T. Son, Nucl. Phys. [25] In Ref. [35], a model equivalent to ours with N = 1 B, 573, 434 (2000). was analyzed and the critical coupling was found to be [14] T. Kojo, Y. Hidaka, L. McLerran, and R. D. Pisarski, approximately 41.6. The reason for the discrepancy is Nucl. Phys. A, 843, 37 (2010); T. Kojo, Y. Hidaka, twofold. First, the authors of Ref. [35] studied the model K. Fukushima, L. McLerran, and R. D. Pisarski, (2011), on a square domain with periodic boundary conditions, arXiv:1107.2124 [hep-ph]. which somewhat disfavors the supersolid state with a [15] These have the same translational properties as intrinsi- hexagonal structure and thus shifts the transition to- cally discrete systems defined on a (spatial) lattice. How- wards stronger couplings. Second, in Ref. [35] the energy ever, since the extension to such systems would require functional was minimized with a fixed particle number, further generalization of the notation, we exclude them whereas we fixed the chemical potential instead. As a for the sake of simplicity from our considerations. consequence, the average particle density has a tiny dis- 14

continuity at the transition point: while in the superﬂuid [30] H. Fukuyama, Solid State Commun., 17, 1323 (1975). phase it is exactly one by normalization, in the supersolid [31] C. Pethick and H. Smith, Bose-Einstein condensation in phase it is approximately 1.06. dilute gases, 2nd ed. (Cambridge University Press, 2008). [26] T.-L. Ho, Phys. Rev. Lett., 81, 742 (1998); T. Ohmi and [32] J. R. Abo-Shaeer, C. Raman, J. M. Vogels, and W. Ket- K. Machida, J. Phys. Soc. Jpn., 67, 1822 (1998). terle, Science, 292, 476 (2001). [27] J. Steinhauer, R. Ozeri, N. Katz, and N. Davidson, Phys. [33] I. Coddington, P. Engels, V. Schweikhard, and E. A. Rev. Lett., 88, 120407 (2002). Cornell, Phys. Rev. Lett., 91, 100402 (2003). [28] D. T. Son, Phys. Rev. Lett., 94, 175301 (2005); [34] E. B. Sonin, Rev. Mod. Phys., 59, 87 (1987); G. Baym, C. Josserand, Y. Pomeau, and S. Rica, 98, 195301 Phys. Rev. Lett., 91, 110402 (2003). (2007); J. Ye, Europhys. Lett., 82, 16001 (2008). [35] N. Sep´ulveda, C. Josserand, and S. Rica, Eur. Phys. J. [29] V. Cirigliano, S. Reddy, and R. Sharma, Phys. Rev. C, B, 78, 439 (2010). 84, 045809 (2011).