# Ultra-cold bosons in zig-zag optical lattices

###### Abstract

Ultra-cold bosons in zig-zag optical lattices present a rich physics due to the interplay between frustration, induced by lattice geometry, two-body interaction and three-body constraint. Unconstrained bosons may develop chiral superfluidity and a Mott-insulator even at vanishingly small interactions. Bosons with a three-body constraint allow for a Haldane-insulator phase in non-polar gases, as well as pair-superfluidity and density wave phases for attractive interactions. These phases may be created and detected within the current state of the art techniques.

July 12, 2019

## I Introduction

Atoms in optical lattices offer extraordinary possibilities for the controlled emulation and analysis of lattice models and quantum magnetism Lewenstein2007 (); Yukalov2009 (). Various lattice geometries are attainable by means of proper laser arrangements, including triangular Becker2010 () and Kagome Jo2012 () lattices, opening fascinating possibilities for the study of geometric frustration, which may result in flat bands with the constrained mobility and largely enhanced role of interactions Huber2010 (). Moreover, the value and sign of inter-site hopping may be modified by means of shaking techniques Eckardt2005 (); Zenesini2009 (), allowing for the study of frustrated antiferromagnets with bosonic lattice gases Struck2011 ().

Interatomic interactions may be controlled basically at will by means of Feshbach resonances Chin2010 (). In particular, large on-site repulsion may allow for the suppression of double occupancy in bosonic gases at low fillings (hard-core regime). Interestingly, it has been recently suggested that, due to a Zeno-like effect, large three-body loss rates may result in an effective three-body constraint, in which no more than two bosons may occupy a given lattice site Daley2009 (). This constraint opens exciting novel scenarios, especially in what concerns stable Bose gases with attractive on-site interactions, including color superfluids in spinor Bose gases Titvinidze2011 () and pair-superfluid phases Daley2009 (); Bonnes2011 (); Chen2011 (). The suppression of three-body occupation has been hinted in recent experiments Mark2012 ().

Under proper conditions, lattice gases may resemble to a large extent effective spin models, e.g. hard-core bosons may be mapped into a spin- XY Heisenberg model Lewenstein2007 (). Lattice bosons at unit filling resemble to a large extent spin- chains DallaTorre2006 (), and in the presence of inter-site interactions, as it is the case of polar gases Lahaye2009 (), have been shown to present a gapped Haldane-like phase Haldane1983 () (dubbed Haldane-insulator (HI) DallaTorre2006 (); Berg2008 ()) characterized by a non-local string-order DenNijs1989 ().

In this work we analyze the physics of ultra-cold bosons in zig-zag optical lattices. We show that the interplay of frustration and interactions lead to a different physics for unconstrained and constrained (with up to two particles per site) bosons. For unconstrained bosons, geometric frustration induces chiral superfluidity, and allows for a Mott-insulator phase even at vanishingly small interactions. For constrained bosons, we show that a Haldane-insulator phase becomes possible even for non-polar gases. Moreover, pair-superfluid Daley2009 (); Bonnes2011 (); Chen2011 () and density-wave phases may occur for attractive on-site interactions. A direct first-order phase transition from Haldane-insulator to pair-superfluid is observed and explained. These phases may be realized and detected with existing state of the art techniques.

The structure of the paper is as follows. In Sec. II we introduce the zig-zag lattice model under consideration. Sec. III is devoted to the unconstrained case, whereas Sec. IV deals with bosons with a two-body constraint. Finally, in Sec. V we summarize our conclusions. Further details on analytical and numerical procedures are discussed in the Appendices.

## Ii Zig-zag lattices

In the following we consider bosons in zig-zag optical lattices. As shown in Fig. 1, this particular geometry may result from the incoherent superposition of a triangular lattice with elementary cell vectors and (formed by three laser beams of wavenumber oriented at degrees from each other, as discussed in Ref. Becker2010 ()) and a superlattice with lattice spacing oriented along . For a sufficiently strong superlattice, zig-zag ladders are formed, and the hopping between ladders may be neglected. We will hence concentrate in the following on the physics of bosons in a single zig-zag ladder, which is to a large extent given by the rates and characterizing the hopping along the two directions (Fig. 1). As shown in Ref. Struck2011 (), a periodic lattice shaking may be employed to control the value of and independently. Interestingly, their sign may be controlled as well. In the following we consider an inverted sign for both hoppings, which result in an anti-ferromagnetic coupling between sites Struck2011 ().

Ordering the sites as indicated in Fig. 1, the physics of the system is given by a Bose-Hubbard Hamiltonian with on-site interactions characterized by the coupling constant , nearest-neighbor hopping and next-nearest-neighbor hopping :

where are the bosonic creation/annihilation operators of particles at site , , and we have added the possibility of three-body interactions, characterized by the coupling constant . We assume below an average unit filling .

## Iii Unconstrained bosons

We discuss first the ground-state properties of unconstrained bosons (). At , the Hamiltonian (II) is diagonalized in quasi-momentum space , with the dispersion , with . Depending on the frustration we may distinguish two regimes. If , the dispersion presents a single minimum at , and hence small will introduce a superfluid (SF) phase, with quasi-condensate at . If , presents two non-equivalent minima at . As shown below, interactions favor the predominant population of one of these minima, and the system enters a chiral superfluid (CSF) phase with a non-zero local boson current characterized by a finite chirality , with . At , the Lifshitz point, the dispersion becomes quartic at the minimum, , the effective mass diverges, and even vanishingly small interactions become relevant.

To study the effect of interactions we combine numerical calculations based on the density matrix renormalization group (DMRG) method White () (with up to sites keeping per block on average states for gapped phases and states for gapless ones), and bosonization techniques to unveil the low-energy behavior of model (II). For , we employ standard bosonization transformations Giamarchi (), with an additional oscillating factor , to obtain the low-energy effective theory, which is given by the sine-Gordon model

(2) |

where and describe phase and density fluctuations of bosons respectively, , is the sound velocity and the Luttinger parameter. In the weak-coupling, , hydrodynamic relations are expected to hold: and , clearly showing that enhances correlations. At , diverges and the system enters a Mott-insulator (MI) even for vanishingly small (Fig. 2(a)). The SF-MI transition takes place however in the strong-coupling regime in which and must be determined numerically. We obtain from the single-particle correlations which in the SF decay as . The value marks the boundary between SF (, ) and MI (, , and ). The MI phase is characterized by a hidden parity order Berg2008 (), , which has been recently measured in site-resolved experiments Endres2011 ().

The case is best understood from bosonization in the regime. We may then introduce two pairs of bosonic fields () and (), describing, respectively, the subchains of even and odd sites. The effective model is governed by the Hamiltonian density

where , , , , and are phenomenological parameters (in the regimes displayed on Figure 2 (a)), . Note that the chirality is given by . In weak-coupling, , with , and . In this case only the term is relevant, resulting in Nersesyan (). Hence, a small is expected to favor a CSF for , as our numerical results confirm (Fig. 2(a)). The CSF phase is characterized by , where .

Moreover, depending on the values of bosonization opens the possibility of two consecutive phase transitions with increasing starting from the CSF phase (App. A), which we have confirmed with our DMRG calculations (Fig. 2(a)) detailed in App. B. First a KT transition occurs from CSF to chiral-Mott (CMI), a narrow Mott phase with finite chirality. Then an Ising transition is produced from CMI to non-chiral MI. At both KT transition lines in Fig. 2(a) (SF-MI and CSF-CMI), up to a logarithmic prefactor, , where in CSF .

## Iv Constrained bosons

As mentioned above, sufficiently large three-body losses may result in a three-body constraint () Daley2009 (). In that case, Model (II) may be mapped to a large extent onto a frustrated spin- chain model commentSpin1 (), which, presents the possibility of a gapped Haldane phase, characterized by a non-local string order. Hence, interestingly, constrained bosons in a zig-zag lattice may be expected to allow for the observation of the HI phase in the absence of polar interactions.

Indeed, a model with and finite shows that at the Lifshitz point, , a HI phase is stabilized for arbitrarily weak (Fig. 2(b)). The effective theory describing the HI is again the sine-Gordon model (2) with . However, now , which selects a hidden string order Berg2008 (). Resembling the case of Fig. 2(a), SF, HI, chiral-HI (CHI) and CSF phases occur (Fig. 2(b)). These phases are expected for from known results in frustrated spin- chains Kolezhuk (); Lecheminant (); Hikihara2000 (); Hikihara2002 (). Our DMRG simulations suggest that all these phases meet at for .

Figure 3 shows the phase diagram for constrained bosons (). Starting from the HI phase, increasing can induce a Gaussian HI-MI phase transition, characterized by a vanishing in (2), resembling the phase transition between Haldane and large-D phases induced by single-ion anisotropy in spin-1 chains Schulz (). The SF phase is separated from the MI and HI by KT transitions, whereas at the CSF boundary with the MI (HI) a CMI (CHI) occurs as mentioned above (these very narrow regions are not resolved in Fig. 3).

Interestingly, constrained bosons allow as well for the exploration of attractive two-body interactions, , without collapse. The phases are also depicted in Fig. 3. For sufficiently large , bosons tend to cluster in pairs and, as already discussed in Ref. Daley2009 (), for an Ising transition between a SF and a pair superfluid (PSF) occurscommentIsing (), analogous to the XY1 to XY2 phase transition in spin- chains induced by single-ion anisotropy Schulz () (this transition has been recently studied for 2D lattices as well Bonnes2011 (); Chen2011 ()). The PSF phase is characterized by an exponentially decaying but algebraically decaying pair-correlation function . Indeed a PSF occurs for sufficiently large , also for which is characterized in bosonization in Eq. (III) by a gapped antisymmetric sector, with pinned , and a gapless symmetric sector Vekua (). Though one may anticipate an Ising phase transition between the CSF (with broken discrete parity symmetry) and the PSF (with restored symmetry), the behavior of and (not shown) hints to a weakly first-order nature.

Small disfavors singly-occupied sites and thus enhances and the bulk excitation gap of the HI phase (see Figs. 4 and 5). However, since large removes singly occupied sites completely, just like strong nearest neighbour repulsion, it is expected that the HI phase eventually will transform for growing into a gapped density-wave (DW) phase via Ising phase transition DallaTorre2006 (), and string order will evolve into DW order (Fig. 4 shows how merges with for ). The DW phase is characterized by an exponential decay of both and though a finite . Our DMRG results confirm this scenario (see Fig. 4), showing that a DW phase is located between the above mentioned PSF regions (Fig. 3).

Interestingly the DW phase remains in between both PSF regions all the way into . In that regime, we may project out singly-occupied sites, and introduce a pseudo-spin-, identifying , and defining the spin operators , . The effective model to leading order in is a spin- chain:

(4) |

where . For , this is a symmetric chain, whereas the terms break the symmetry down to , moving the effective theory obtained after bosonization of towards the irrelevant direction (in the renormalization group sense). As a result of this, a gapless XY phase of the spin-1/2 chain is expected, i.e. a PSF phase. Higher order terms in (not shown explicitely) break, even for , the symmetry to in the irrelevant direction. However, interestingly, the ring exchange along the elementary triangle of the zig-zag chain, with amplitude , forces the effective theory towards the relevant direction, leading to a gapped Néel phase of the spin-1/2 chain, i.e. the DW phase. The competition between exchange along the lattice bonds and ring-exchange leads hence to two consecutive KT phase transitions induced by , for first from PSF to DW, followed by DW back to PSF. The width of the DW phase is , and it extends all the way into the limit.

Finally, our DMRG simulations show a narrow region where a direct, apparently first-order, HI-PSF transition occurs (see for details App. B), characterized by discontinuous jumps of (Fig. 5). This first-order nature is explained because on one hand increasing within the HI phase increases due to the suppression of singly-occupied sites, and on the other hand, for (Fig. 3), a growing destroys the insulating state in favour of a PSF phase, where string order cannot exist. On the contrary, diminishes for decreasing when approaching the HI-CSF boundary (Fig. 5).

## V Conclusions

In conclusion, the interplay between geometrical frustration and interactions leads to rich physics for ultra-cold bosons in zig-zag optical lattices. Unconstrained bosons may present chiral superfluidity, and Mott insulator for vanishingly small interactions. Constrained bosons may allow for the observation of Haldane-insulator without the necessity of polar interactions, as well as pair-superfluid and density wave phases at attractive interactions. All the predicted phases may be detected using state of the art techniques. The SF and CSF phases may be distinguished by means of time-of-flight (TOF) techniques, in a similar way as recently done for condensates in triangular optical lattices Struck2011 (). The DW and PSF phases are characterized by double or zero occupancy, which could be detected using parity measurements as those introduced in Refs Bakr2010 (); Sherson2010 (), and could be discerned from each other by the absence/presence of interference fringes in TOF Greiner2002 (). Finally, the string-order of the HI phase may be studied using similar site-resolved measurements as those recently reported for the measurement of non-local parity order in Mott insulators Endres2011 ().

###### Acknowledgements.

This work has been supported by the cluster of excellence QUEST (Center for Quantum Engineering and Space-Time Research). T.V. acknowledges SCOPES Grant IZ73Z0-128058.## Appendix A Bosonization analysis of the MI-CMI-CSF transition

In this appendix we provide additional details on our bosonization analysis. We take the limit , that allows to consider sub-chains formed by the even and odd sites, and introduce the symmetric and antisymmetric fields and . The interaction between these fields is given by the last two terms of Eq. (3) of the main text:

(5) |

where the first term supports chirality and the second one favors a MI phase.

Starting from , deep in the Mott phase of each sub-chain, then , and thus the fields are pinned in the Mott phase. In this case, to second order in , one can integrate out in the partition function from the first term of Eq.(5), obtaining the following contribution in the antisymmetric sector,

where the average is performed in the ground state of the MI phase, where is short ranged. Hence the leading contribution in the antisymmetric sector, after carrying out operator product expansion, is a term . Note that a contribution decreasing the value of is obtained as well in the antisymmetric sector. The competition between and (obtained using mean-field decoupling of the last term of Eq.(5) in the MI phase) is resolved with an Ising phase transition in the antisymmetric sector with increasing , leading to the pinning of in the new ground state , so that , driving the symmetric sector into a state with finite topological current, .

The simplest scenario to establish the Ising phase transition is for . In that case, performing a mean-field decoupling of the second term in Eq. (5), the antisymmetric sector is governed by the Hamiltonian density,

where , , , and . The antisymmetric sector can be hence described by two free massive Majorana fermions, with masses . At the Ising phase transition, , and the mass of one of the Majorana fermions vanishes Gogolin1998 ().

However, the Mottness of the ground state after the chirality gets long-range ordered, , does not necessarily disappear immediately, due to the possibility of a relevant contribution in the symmetric sector, , for which stems after integrating out the in the last term of Eq. (5) in the state with pinned . Note that in the CMI state , , and also . Further decreasing , at , the CMI phase () disappears at a KT phase transition in favor of the CSF phase ().

Thus, our bosonization analysis, for , suggests the possibility of two consecutive phase transitions with increasing starting from the CSF phase, first a KT transition from CSF to CMI, followed by an Ising transition from CMI to MI. Note that if an intermediate CMI phase between CSF and MI is absent, the direct transition between CSF and MI cannot be of Ising nature, since CSF is a gapless phase and MI is gapped. In the following section, we show provide a numerical proof of the existence of the CMI phase, showing that starting from the MI phase, the system experiences an Ising transition involving the growth of chirality.

## Appendix B Numerical analysis

Here we provide details on our numerical calculations, and in particular on how phase boundaries were determined and how error bars were estimated.

We investigate the phase diagrams by means of numerical simulations based on exact diagonalization and density-matrix renormalization group (DMRG) with up to lattice sites. Typically for DMRG-simulations we keep about matrix-states. The results have also been confirmed by infinite-system size algorithm (iDMRG) Schollwock2005 () with up to states. For our simulations of the unconstrained Bose-Hubbard-model we kept bosons per site for and for , which has been shown to be sufficient by comparing to simulations with higher . For open-boundary-conditions special care may be needed to take care of degenerate edge states in the Haldane-phase, i.e. by polarizing edges DallaTorre2006 ().

First we discuss phase transitions involving chiral order parameter , where is defined in Sec. III. Since chiral order is spontaneous, we study numerically the chirality-chirality correlation function Hikihara2000 (), which at large distances saturates to and we extract it by averaging over long distances. We study how the chirality vanishes with changing (for the MI-CMI and HI-CHI transitions; for the CSF-PSF transition we study instead the behavior of the parity order). The scaling of chirality close to MI-CMI and HI-CHI transitions unambiguously confirms that the corresponding phase transition is of Ising type, showing the correct scaling behavior. The critical value of for the corresponding Ising transition is located by extracting the intersection of curves for different system sizes (the width of the intersecting point provides the uncertainty of the procedure). Fig. (A1) illustrates our numerical results in the vicinity of the MI-CMI transition. The collapse of the data for different system sizes on a single curve (inset) confirms the Ising nature of the underlying phase transition. Similarly we confirm numerically the Ising nature of the HI-CHI transition (not shown).

The standard numerical procedure to locate the KT transition from superfluid to gapped phase is based on the Luttinger liquid parameter, which we extract from the single-particle correlation function . When the single particle correlations show incommensurate oscillations, we fit it to with including conformal correctionsCazalilla2004 (). To get a lower bound for the transition point one can apply a power law fit to shorter distances after dividing out incommensurate oscillations. The KT CMI-CSF transition (as well as the CHI-CSF transition) can be located in this way providing a strong hint on the existence of a finite CMI (CHI) region between the MI (HI) and CSF-phases. However, a much more accurate estimate of the KT transition point is provided by the analysis of the quasi-momentum distribution Dhar2011 (). Since with at the KT transition (up to logarithmic correction), has a maximum at and its height depends on the system size as . This behavior is illustrated in Fig. A2, which shows a clear intersection of the curves for different system sizes at a single point (the width of this point provides the error bar for the position of the KT transition). Hence for the case depicted in Figs. A1 and A2, a narrow but clearly determined CMI phase may be found between and .

In addition to chirality we have studied other order parameters, including parity, string, and DW orders (all defined in the main text). Applying finite-size-scaling analysis Ueda2008 () allows or an accurate
location of the corresponding phase transition lines depicted in Fig. 3 of the main text.

On Fig. A3 we depict the scaling of the parity order in the vicinity of PSF-SF and PSF-CSF transitions using the iDMRG algorithm. The correlation length for the quantum Ising model scales like , where denotes the matrix dimension Tagliacozzo2008 (). This result is in very good agreement with our results for the PSF-SF transition (Fig. A3, left). Indeed, using Ising critical exponents the parity order parameter behaves as

On the contrary, for the CSF-PSF transition the behavior is fundamentally different, and we instead observe a jump in the parity order (Fig. A3, right), indicating a first-order character of the transition. We observe as well a similar jump in the parity order across the PSF-HI transition. In addition string order shows an abrupt jump at the same transition as depicted on Fig. A4.

Besides monitoring order parameters we use a finite-size level crossing analysis (level spectroscopy) LevelSpectroscopy () to determine the location of different phase transitions. We use DMRG for longer chains, which compares well with our exact diagonalization results available only up to sites. The SF-PSF transition can be extrapolated very precisely from level crossing between the one-particle and two-particle excitation gaps. We determine in this way the SF-PSF boundary, which compares well with the boundary obtained from monitoring the parity order. The transitions from DW to PSF phases can also be determined very accurately by level spectroscopy. Finite-size extrapolation following law confirms the KT nature of those transitions. Finally, for twisted boundary conditions the Gaussian transition line between HI and MI can also be located by ground state level crossing Chen03 ().

## References

- (1) M. Lewenstein et al., Adv. in Phys. 56, 243 (2007).
- (2) V. I. Yukalov, Laser physics 19 (2009).
- (3) C. Becker et al., New J. Phys. 12, 065025 (2010).
- (4) G.-B. Jo et al., Phys. Rev. Lett. 108, 045305 (2012).
- (5) S. D. Huber and E. Altman, Phys. Rev. B 82, 184502 (2010).
- (6) A. Eckardt, C. Weiss, and M. Holthaus, Phys. Rev. Lett. 95, 260404 (2005).
- (7) A. Zenesini et al., Phys. Rev. Lett. 102, 100403 (2009).
- (8) J. Struck et al., Science 333, 996 (2011).
- (9) C. Chin et al., Rev. Mod. Phys. 82, 1225 (2010).
- (10) A. J. Daley et al., Phys. Rev. Lett. 102, 040402 (2009).
- (11) I. Titvinidze et al., New Journal of Physics 13, 035013 (2011).
- (12) L. Bonnes and S. Wessel et al., Phys. Rev. Lett. 106, 185302 (2011).
- (13) Y.-C. Chen, K.-K. Ng, and M.-F. Yang, Phys. Rev. B 84, 092503 (2011).
- (14) M. J. Mark et al., Phys. Rev. Lett. 108, 215302 (2012).
- (15) E. G. Dalla Torre, E. Berg, and E. Altman, Phys. Rev. Lett. 97, 260401 (2006).
- (16) T. Lahaye et al., Rep. Prog., Phys. 72, 126401 (2009).
- (17) F. D. M. Haldane, Phys. Lett. A 93, 464 (1983); Phys. Rev. Lett. 50, 1153 (1983).
- (18) E. Berg et al., Phys. Rev. B 77, 245119 (2008).
- (19) M. den Nijs, and K. Rommelse, Phys. Rev. B 40, 4709 (1989).
- (20) T. Giamarchi, Quantum Physics in One Dimension, Oxford University Press (2003).
- (21) M. Endres et al., Science 334, 200 (2011).
- (22) A. A. Nersesyan, A. O. Gogolin, and F.H.L. Essler, Phys. Rev. Lett. 81, 910 (1998).
- (23) One may map boson problem by means of the Holstein-Primakoff transformation , to frustrated spin-1 XY chain (up to particle-hole symmetry breaking terms that are irrelevant Berg2008 ()), where , and and characterize, respectively, the single-ion anisotropy and the next-nearest neighbor frustrating exchange.
- (24) Numerically we have unambiguously confirmed the Ising nature of SF-PSF transition by observing the correct scaling of close to transition point.
- (25) S. R. White, Phys. Rev. Lett. 69, 2863 (1992); Phys. Rev. B 48, 10345 (1993); U. Schollwöck, Rev. Mod. Phys. 77, 259 (2005).
- (26) It is notoriously hard to determine precisely the KT phase transition lines by this method alone. The precise location of the phase transition lines, and the determination of the nature of the multi-critical points close to are not pursuit in the current work.
- (27) A. K. Kolezhuk, Prog. Theor. Phys. Suppl. 145, 29 (2002); Phys. Rev. B 62, R6057 (2000).
- (28) P. Lecheminant et al., Phys. Rev. B 63, 174426 (2001).
- (29) T. Hikihara et al., J. Phys. Soc. Jpn. 69, 259 (2000).
- (30) T. Hikihara, J. Phys. Soc. Jpn. 71, 319 (2002).
- (31) H.J. Schulz, Phys. Rev. B 34, 6372 (1986).
- (32) T. Vekua, et al, Phys. Rev. B 76, 174420 (2007).
- (33) W. S. Bakr et al., Science 329, 547 (2010).
- (34) J. F. Sherson et al., Nature 467, 68 (2010).
- (35) M. Greiner et al., Nature 415, 39 (2002).
- (36) A.O. Gogolin, A. A. Nersesyan, and A. M. Tsvelik, Bosonization and Strongly Correlated Systems, Cambridge University Press (1998).
- (37) U. Schollwöck, Rev. Mod. Phys. 77, 259 (2005).
- (38) T. Hikihara et al., J. Phys. Soc. Jpn. 69, 259 (2000); T. Hikihara, J. Phys. Soc. Jpn. 71, 319 (2002).
- (39) M.A. Cazalilla, J. Phys. B 37 (2004).
- (40) M. Dalmonte et al., Phys. Rev. B 83, 155110 (2011).
- (41) A. Dhar et al., Phys. Rev. A 85, 041602 (2012).
- (42) H. Ueda, H. Nakano and K. Kusakabe, Phys. Rev. B 78, 224402 (2008).
- (43) L. Tagliacozzo et al., Phys. Rev. B 78, 024410 (2008).
- (44) K. Nomura, J. Phys. A 28, 5451 (1995); A. Kitazawa, J. Phys. A 30, L285 (1997); T. Murashima, K. Hijii, K. Nomura, and T. Tonegawa, J. Phys. Soc. Jpn. 74, 1544 (2005).
- (45) W. Chen, K. Hida, and B.C.T. Sanctuary, Phys. Rev. B 67, 104401 (2003).