PHYSICAL REVIEW X 9, 021022 (2019)
Monte Carlo Study of Lattice Compact Quantum Electrodynamics with Fermionic Matter: The Parent State of Quantum Phases
– † ‡ Xiao Yan Xu,1,* Yang Qi,2 4, Long Zhang,5 Fakher F. Assaad,6 Cenke Xu,7 and Zi Yang Meng8,9,10,11, 1Department of Physics, Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong, China 2Center for Field Theory and Particle Physics, Department of Physics, Fudan University, Shanghai 200433, China 3State Key Laboratory of Surface Physics, Fudan University, Shanghai 200433, China 4Collaborative Innovation Center of Advanced Microstructures, Nanjing 210093, China 5Kavli Institute for Theoretical Sciences and CAS Center for Excellence in Topological Quantum Computation, University of Chinese Academy of Sciences, Beijing 100190, China 6Institut für Theoretische Physik und Astrophysik, Universität Würzburg, 97074, Würzburg, Germany 7Department of Physics, University of California, Santa Barbara, California 93106, USA 8Beijing National Laboratory for Condensed Matter Physics and Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China 9Department of Physics, The University of Hong Kong, China 10CAS Center of Excellence in Topological Quantum Computation and School of Physical Sciences, University of Chinese Academy of Sciences, Beijing 100190, China 11Songshan Lake Materials Laboratory, Dongguan, Guangdong 523808, China
(Received 1 August 2018; revised manuscript received 19 March 2019; published 2 May 2019)
The interplay between lattice gauge theories and fermionic matter accounts for fundamental physical phenomena ranging from the deconfinement of quarks in particle physics to quantum spin liquid with fractionalized anyons and emergent gauge structures in condensed matter physics. However, except for certain limits (for instance, a large number of flavors of matter fields), analytical methods can provide few concrete results. Here we show that the problem of compact Uð1Þ lattice gauge theory coupled to fermionic matter in ð2 þ 1ÞD is possible to access via sign-problem-free quantum Monte Carlo simulations. One can hence map out the phase diagram as a function of fermion flavors and the strength of gauge fluctuations. By increasing the coupling constant of the gauge field, gauge confinement in the form of various spontaneous-symmetry- breaking phases such as the valence-bond solid (VBS) and N´eel antiferromagnet emerge. Deconfined phases with algebraic spin and VBS correlation functions are also observed. Such deconfined phases are incarnations of exotic states of matter, i.e., the algebraic spin liquid, which is generally viewed as the parent state of various quantum phases. The phase transitions between the deconfined and confined phases, as well as that between the different confined phases provide various manifestations of deconfined quantum criticality. In particular, for four flavors Nf ¼ 4, our data suggest a continuous quantum phase transition between the VBS and N´eel order. We also provide preliminary theoretical analysis for these quantum phase transitions.
DOI: 10.1103/PhysRevX.9.021022 Subject Areas: Condensed Matter Physics, Particles and Fields
I. INTRODUCTION several decades [1–15]. This is because gauge fields coupled to matter fields is a fundamental concept in many The interplay between lattice gauge theories and fer- areas of physics. For example, in condensed matter, mionic matter has allured the imagination of physicists for ð2 þ 1ÞD field theories with a compact Uð1Þ gauge field coupled to gapless relativistic fermions often serve as the *[email protected] low-energy effective field theories in 2D strongly correlated † [email protected] electron systems including cuprate superconductors [1,2, ‡ [email protected] 5–7] and quantum spin liquids [11–14,16]. In high-energy physics, the mechanism of quark confinement in gauge Published by the American Physical Society under the terms of theories with dynamical fermions such as quantum the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to chromodynamics (QCD) is among the most elusive sub- the author(s) and the published article’s title, journal citation, jects, and the absence or presence of a deconfined phase in and DOI. 3D compact quantum electrodynamics (cQED3) coupled to
2160-3308=19=9(2)=021022(17) 021022-1 Published by the American Physical Society XU, QI, ZHANG, ASSAAD, XU, and MENG PHYS. REV. X 9, 021022 (2019)
(not necessarily large) Nf massless fermions has attracted a flavors of fermionic matter is particularly important to lot of attention [4,10–15] and remains unsolved to this day. condensed matter physics because these cases actually In recent years, collective efforts from both condensed correspond to the low-energy field theory description of matter and high-energy physics have started to generate many interesting strongly correlated electron systems, and promising outcomes [4,12,17–26]. There exist concrete therefore host the potential promise of establishing the new examples, by now, of discrete Z2 gauge field theories paradigms in condensed matter physics. Furthermore, coupled to fermionic matter at ð2 þ 1ÞD, deconfined phase perturbative renormalization-group calculations to higher with fractionalized fermionic excitations at weak gauge orders have recently been carried out in attempt to acquire fluctuation, as well as symmetry-breaking phase with the critical properties of the deconfinement-to-confinement gapped fermionic excitations at strong gauge fluctuation transition in the form of QED3-Gross-Neveu universality have been observed [18–20]. The apparently continuous classes [38–41]. transition between the deconfined and confined phases is Based on these considerations, in this work, we succeed highly nontrivial [20], as it is driven by the condensation of in performing large-scale quantum Monte Carlo (QMC) emergent fractionalized excitations and is hence beyond the simulations on the cQED3 coupled to Nf flavor of fermions scope of the Landau-Ginzburg-Wilson paradigm of critical and eventually map out the phase diagram (Fig. 1) in the phenomena in which symmetry breaking is described by a fermion flavor and gauge field fluctuations strength plane. local order parameter. Deconfined phases—Uð1Þ deconfined phase (Uð1ÞD here- The cQED3 is the simplest theory to discuss confinement after) to be more precise—are indeed found in the phase – and chiral symmetry breaking [27 29]. The pure cQED3 diagram for Nf ¼ 6 and 8, and even at Nf ¼ 2 and 4 there without a matter field is known to be always confining are very positive signatures of their existence. Various [10,27,30,31]. However, when there is fermionic matter, confined phases in the form of different symmetry break- the gapless fermionic fluctuations may drive the system ings, such as antiferromagnetic order (AFM) and valence- towards deconfinement. The large-Nf limit of cQED3 bond solid (VBS), are also discovered. Interesting quantum with fermionic matter is believed to belong to this case phase transitions between deconfined and confined phases [11,13,15,32], but the existence of the deconfined phase for and between different confined phases [18,42–46] are small Nf is still under debate [12,17,33–36]. Analytically, revealed as well. the perturbative calculation at small Nf is uncontrolled. For the sake of a smoother narrative, the rest of the paper Numerically, recent hybrid Monte Carlo (HMC) calcula- is organized in the following order. In Secs. II A and II B, tions in Ref. [17] faced difficulties caused by fermion zero we first start with a quantum rotor model coupled to modes. Though these difficulties may be cured by turning fermions, which can be formulated as cQED3 coupled to on a four-fermion interaction term, the scaling dimension of fermionic matter. Then in Sec. II C, we discuss the sign the four-fermion interaction will receive corrections from structure of this model, where we find that a pseudounitary gauge fluctuation at the order of 1=Nf, which may lead to a group can be used to avoid the phase problem at odd Nf relevant runaway flow in renormalization group (RG) and the sign problem at even Nf. In Secs. II D and II E,we calculation [37]. Thus, a combined RG flow of monopole explain the challenges in the QMC simulation even without and four-fermion interactions may be complicated, and a the sign problem and provide our solution with a fast deconfined phase could still exist in the phase diagram but update algorithm for simulating gauge fields with continu- evades the previous study of HMC. cQED3 with finite Nf ous symmetries. In Sec. III, we discuss the whole phase
(a)J (b) (c) c or c † 2 i i U(1)D AFM i c † te b i 4 U(1)D VBS AFM 6 U(1)D VBS cj Nf 8 U(1)D VBS
FIG. 1. (a) Phase diagram spanned by the Fermi flavors Nf and the strength of gauge field fluctuations J of the model shown in (b). Uð1ÞD stands for the Uð1Þ deconfined phase where the fermions dynamically form a Dirac system. This phase corresponds to the algebraic spin liquid where all correlation functions show slow power-law decay. VBS stands for valence-bond-solid phase and AFM stands for the antiferromagnetic long-range ordered phase (N´eel phase). (b) Sketch of the model of Eq. (1). The yellow circles represent the gauge field attached to each fermion hopping, and the yellow dashed lines stand for the flux term per plaquette. (c) The gauge- invariant propagator for fermions with a string of gauge fields attached.
021022-2 MONTE CARLO STUDY OF LATTICE COMPACT QUANTUM … PHYS. REV. X 9, 021022 (2019) X † ϕ diagram and then focus on the physical properties and ψ ∂ − μ δ − i ij ψ LF ¼ iα½ð τ Þ ij te jα þ H:c:; understanding of the Uð1ÞD phase, in particular, the reason hijiα for it being the parent state of various quantum phases, the 4 X 1 − ϕ τ 1 − ϕ τ deconfinement-confinement phase transitions, and VBS-to- Lϕ ¼ 2 f cos½ ijð þ Þ ijð Þ g JNfΔτ AFM phase transition at Nf ¼ 4. Preliminary theoretical hiji analysis of these transitions is also given in Sec. III. Finally, 1 X ϕ the discussion and conclusions are given in Sec. IV. þ 2 KNf cos ðcurl Þ; ð3Þ □
II. MODEL AND METHOD respectively, where μ is set to zero for the half-filled case. β ¼ð1=TÞ is the inverse temperature. The model in Eq. (1) A. Rotor model with fermions is now explicitly formulated as (unconstrained) cQED3 The system we are interested in can be most conveniently coupled to fermionic matter [11,13,17]. We now consider formulated as a 2D quantum rotor model coupled to the symmetries of the model and show that the Gauss law is fermions with Hamiltonian dynamically imposed in the low-temperature limit.
X X 1 1 2 † θˆ B. Symmetries and limiting cases ˆ − ˆ i ij ˆ H ¼ 2 JNf 4 Lij t ðciαe cjα þ H:c:Þ hi;ji hi;jiα Our model [see Eq. (1)] has global and local symmetries. 1 X It enjoys a manifest global SUðNfÞ spin symmetry as well θˆ þ 2 KNf cosðcurl Þ; ð1Þ as a particle-hole symmetry: □ ˆ −1 ˆ† ˆ ¯ −1 i ˆ P zciαP ¼ zð Þ ciα: ð4Þ ˆ ˆ where Lij and θij are canonical angular momentum, ˆ θˆ θˆ In the above, z is a complex number that makes it clear that ½L ;e i ij ¼ e i ij , and its coordinate operator of rotors ij the particle-hole symmetry is antiunitary, and ð−1Þi takes on each bond b ¼hiji of a 2D square lattice, as depicted in the value 1 (−1) on sublattice A (B). Fig. 1(b). The fermion flavor α runs from 1 to N , and the f The local Uð1Þ gauge transformation fermions are minimally coupled to the rotor via nearest- neighbor hopping on the square lattice. The flux term with φ ˆ → ˆ i i θˆ → θˆ φ − φ K>0 favors π flux in each elementary plaquette □, where ciα ciαe ; ij ij þ i j ð5Þ □ the magneticP flux of each plaquette is defined as θˆ θˆ θˆ is an invariant. The generator of this local symmetry curl ¼ b∈□ b, and the summation over b is taken in either clockwise or anticlockwise orientation around an corresponds to a local conserved charge (Gauss law) elementary plaquette. X X ˆ − ˆ ˆ† ˆ − 1 2 For the Monte Carlo simulations, it is convenient to Qi ¼ Lij þ ðciαciα = Þð6Þ ˆ j α work in a representation where θij is diagonal. That is, omitting the bond index θˆjϕi¼ϕjϕi with ϕ ∈ ½0; 2πÞ.In with ½Qˆ ; Hˆ ¼0. In our simulations, we sample over all Qˆ this representation, Lˆ ¼ −i½∂=ð∂ϕÞ with eigenvectors i i ˆ ˆ ϕ ϕ sectors, such that our Hamiltonian corresponds to an Lhϕjli ≡ Lei l ¼ lei l and l ∈ Z. With theseR defini- 2π ϕ ϕ unconstrained gauge theory. As a consequence, the corre- tions, theP resolution of unity reads 0 j ih j¼ lation functions of gauge-dependent quantities such as the 1 2π 1ˆ ½ =ð Þ l jlihlj¼ . To formulate the path integral, we single-particle operator are local in space but not in time: −Δτ ˆ 2 8 † † have to estimate the matrix element: ϕ0 e JNfL = ϕ .To ˆ τ ˆ δ ˆ τ ˆ h j j i hciαð Þcjαi¼ i;jhciαð Þciαi. Below, we argue that the this end, we insert resolution of the unit operator and use Gauss law constraint is dynamically imposed in the the Poisson summation formula to obtain zero-temperature limit. ˆ At J ¼ ∞, Lij vanishes and charges are completely −ΔτJN Lˆ 2=8 − 4= ΔτJN 1−cos ϕ−ϕ0 localized since hopping on a given bond involves excita- hϕ0je f jϕi ∼ e ½ ð fÞ ½ ð Þ ; ð2Þ tions of the rotor mode. In this limit, charge configurations ˆ corresponding to specific values of Qi are degenerate, and where the Villain approximation is used. With the at any finite value of J the degeneracy will be lifted. The above, the Hamiltonian in Eq. (1) can be formulated in dynamical generation of the term Ra coherent-state path integral with action S ¼ SF þ Sϕ ¼ X β ˆ ˆ ˆ 0 dτðLF þ LϕÞ, and the Lagrangian for the fermion and HQ ¼ Ki;jQiQj þ ð7Þ gauge field parts is i;j
021022-3 XU, QI, ZHANG, ASSAAD, XU, and MENG PHYS. REV. X 9, 021022 (2019) Z Z accounts for the lifting of this degeneracy. Note that since − − − ¯ ðSϕþSFÞ Sϕ SF ˆ ˆ Z ¼ Dðϕ; ψ; ψÞe ¼ Dϕe Trψ ½e : ð9Þ fP; Qig¼0, the terms containing products of odd numbers ˆ ’ of Qi s are forbidden. The above equation defines a R β classical model with a finite-temperature Kosterlitz- As the action of the gauge field part Sϕ ¼ 0 dτ Lϕ with Lϕ ˆ shown in Eq. (3) is always real (thus, its exponential is Thouless transition. At zero temperature, the Qi are frozen in a given pattern, and the Gauss law is imposed. always positive), the sign structure of the Monte Carlo For J → ∞, our model maps onto an SUðNfÞ quantum weight will come only from the trace over fermions. To antiferromagnetic. We again start from the J ¼ ∞ degen- trace out fermions, we first discretize the imaginary time τ Δτ 1 2 … erate case and consider t in second-order degenerate ¼ z (z ¼ ; ; ;Lτ) where Lτ is the total number of Δτ β perturbation theory. As we mention above, hopping of a time slices (Lτ ¼ ). Then, performing the fermion fermion with flavor index α from site i to nearest-neighbor trace, we have site j leaves the rotor in an excited state associated with Lτ Y Nf energy cost J. The only way to remove this excitation is for −S Trψ ½e F ¼ det I þ B ; ð10Þ a fermion with flavor index α0 to hop back from site j to z z¼1 site i. These processes are encoded in the SUðN Þ f R Heisenberg model β with SF ¼ 0 dτ LF and LF shown in Eq. (3). After the −Δτ β Vz 2 X discretization of , Bz ¼ e with Vz the coupling ˆ t ˆ † ˆ ˆ ˆ † H →∞ ∝− ðD D þ D D Þð8Þ matrix for each fermion flavor [we have Nf in total; J J ij ij ij ij hiji therefore, there is power Nf in Eq. (10)], which has only elements connecting the sites between different sublattices P ϕ N † − i ij ˆ f ˆ ˆ of the square lattice ðVzÞij ¼ te . We recognize that with Dij ¼ α¼1 ciαcjα. In our simulations, we have on such kinds of B matrices form a pseudounitary group average Nf=2 fermions per site such that the representation z SUðn; nÞ where 2n is a dimension of Bz (total number of of the SUðNfÞ group corresponds to the antisymmetric self- sites). As proved in the Appendix B, ∀ D ∈ SUðn; nÞ, adjoint representation (i.e., the Young tableau correspond- 1 ∈ ℜ ing to a column of N =2 boxes) [47]. On the square lattice detð þ DÞ holds. Therefore, the fermion weight will f always be real for all integers N and, most importantly, be and for even values of N where the negative sign problem f f semi-positive-definite for all even values of N such that is absent, this model has been considered in former f QMC simulations can be performed. Although the current auxiliary field QMC simulations [12].AtN ¼ 2, one f paper focuses only on the compact Uð1Þ gauge fields, the finds an antiferromagnetic state, and at and beyond N ¼ 6, f SUðn; nÞ group actually allows one to add extra non- a VBS state. In the large-Nf limit, we recover the Marston- Hermitian terms to the model, such as a staggered imagi- Affleck [2] saddle point accounting for dimerization. At nary chemical potential, that are also sign-problem-free for Nf ¼ 4 and in the absence of charge fluctuations, Ref. [12] even values of Nf. It will be very interesting to study the finds no compelling evidence of VBS and AFM orders non-Hermitian models and their properties in the presence 24 24 when considering lattices up to × . On the other hand, of gauge field fluctuations and electronic interactions in simulations of the corresponding SU(4) Hubbard model future investigations. [48] are consistent with an AFM state in the large-U limit albeit with decreasing value of the order parameter as a function of U. In our simulations, charge fluctuations are D. Difficulties of the QMC simulation present, and the phase diagram is consistent with AFM Although there is no sign problem for even Nf, the order in the large-J limit. simulation of Eq. (1) is by no means simple. Earlier ˆ At J ¼ 0, θij becomes a classical variable in the sense attempts in the high-energy community have been devoted that it has no imaginary time dynamics. Even in the absence to simulate similar models by means of the hybrid of the flux term (K term) the coupling to the fermions Monte Carlo [4,17,35,50] with (dynamical) mass term. favors, according to Lieb’s theorem [49],aπ flux per Mass terms are essential to avoid divergences when plaquette, with associated dynamically generated Dirac calculating forces in the realm of Hamiltonian [51] and dispersion relation of the cˆ fermions. The fate of this state Langevin [52] dynamics. Mass terms, however, introduce biases the severity of which have to be a posteriori at low values of Nf and when gauge fluctuations are accounted for is one of the central aims of our research. clarified. On the other hand, in the condensed matter community, the determinantal QMC (DQMC) is more popular, and it C. Absence of the sign problem for even Nf usually uses local updates and the mass terms are not To simulate the above model with the quantum essential here [53–56]. However, as far as we are aware, Monte Carlo method, we start with the partition function there exists no DQMC simulation on cQED3 coupled to
021022-4 MONTE CARLO STUDY OF LATTICE COMPACT QUANTUM … PHYS. REV. X 9, 021022 (2019) fermionic matter, and our work hence serves as a first YNb attempt. As we explain below, to be able to simulate the Bz ¼ Bz;b; ð12Þ 1 model in Eq. (1), there are several obstacles one needs to b¼ hz;b overcome. where Bz;b ¼ e and hz;b is an N × N matrix with only The most obvious obstacle is the computational com- nonzero elements Bz;bði; jÞ¼Bz;bðj; iÞ ¼ fðϕijÞ where plexity. For general models, the computation complexity θ ϕ fðϕ Þ¼ρðϕ Þei ð ijÞ is the complex function of auxiliary O β 4 ij ij of DQMC for one sweep is ð N Þ (here, N is the total field ϕ in polar form. It is obvious that only when number of sites) for models where a fast update is ij θðϕ Þ¼0 or π, B ði; jÞ will be diagonalized by an applicable, the complexity can be reduced to OðβN3Þ. ij z;b auxiliary-field-independent unitary transformation, and The most common example is the Hubbard model with then the aforementioned fast update scheme applies. on-site interaction; when we flip a single auxiliary field at There are many known models that belong to this case, time slice z and site i, it changes only one diagonal value in such as models with Heisenberg-type interaction [57–59], the Bz matrix, then the new equal-time Green’s function models with the Z2 (bosonic) gauge field coupled to G0ðτ; τÞ can be calculated as fermionic matter [18,19,60,61], etc. Novel physics has 0 … Δ … −1 been found in these models; e.g., in Ref. [58] a continuous G ¼½I þ B1B2; ;Bz−1ðI þ ÞBz; ;BLτ−1BLτ phase transition with fermion mass generation without Δ − −1 ¼ G½I þ ðI GÞ ; ð11Þ spontaneously breaking any symmetry was identified where G is the Green’s function at the previous step. (similar physics was also found in the lattice QCD community [62–65]). Unfortunately, our model in As Δii is the only nonzero element in the N × N matrix Δ, Δ 1 − Eq. (1) involves Uð1Þ gauge fields, and the auxiliary field ð GÞ can be written as the outer product of two ϕ θ ϕ ϕ vectors, and then the Sherman-Morrison formula can be ij is therefore a continuous variable and ð ijÞ¼ ij; used to reduce the complexity of calculating Eq. (11) from thus, our model does not belong to the cases discussed OðN3Þ to OðN2Þ [53,54]. Such an update scheme is above, and naively the fast update cannot be applied. referred as the fast update. For one sweep over all auxiliary U fields [scales as OðβNÞ], the scaling will be OðβN3) instead E. Fast update algorithm designed for ð1Þ gauge fields of OðβN4Þ. For models with off-site interaction, e.g., in our There is indeed an alternative way to design a fast update case due to the coupling between the fermion and gauge algorithm for the model in Eq. (1). It is based on the field, usually we need to make a further Suzuki-Trotter Woodbury matrix identity that effectively generalizes the decomposition over all bonds (assume bond b connects site Sherman-Morrison formula to higher-rank matrices. We 0 i and j, and the total number of bonds is Nb), and the Bz recognize that the new Bz;b after a single update can directly matrix is written as be factorized into ð1 þ ΔÞBz;b with