Disorder by disorder and flat bands in the kagome transverse field Ising model
M. Powalski,1 K. Coester,1 R. Moessner,2 and K. P. Schmidt1, ∗ 1Lehrstuhl f¨urTheoretische Physik 1, TU Dortmund, Germany 2Max Planck Institut f¨urkomplexe Systeme, D-01187 Dresden, Germany (Dated: August 10, 2018) We study the transverse field Ising model on a kagome and a triangular lattice using high-order series expansions about the high-field limit. For the triangular lattice our results confirm a second- order quantum phase transition in the 3d XY universality class. Our findings for the kagome lattice indicate a notable instance of a disorder by disorder scenario in two dimensions. The latter follows from a combined analysis of the elementary gap in the high- and low-field limit which is shown to stay finite for all fields h. Furthermore, the lowest one-particle dispersion for the kagome lattice is extremely flat acquiring a dispersion only from order eight in the 1/h limit. This behaviour can be traced back to the existence of local modes and their breakdown which is understood intuitively via the linked cluster expansion.
PACS numbers: 75.10.Jm, 05.30.Rt, 05.50.+q,75.10.-b
I. INTRODUCTION for arbitrary transverse fields providing a novel instance of disorder by disorder in which quantum fluctuations The interplay of geometric frustration and quantum select a quantum disordered state out of the classically fluctuations in magnetic systems provides a fascinating disordered ground-state manifold. There is no conclu- field of research. It gives rise to a multitude of quan- sive evidence excluding a possible ordering of the kagome tum phases manifesting novel phenomena of quantum transverse field Ising model (TFIM) and its global phase order and disorder, featuring prominent examples such diagram remains unsettled. So far, the only known Ising as spin-ice, spin-glass and spin-liquid phases. This rich system featuring such an unusual disorder by disorder behaviour originates from the frustration of microscopi- scenario is the one-dimensional sawtooth chain [8]. As cally competing magnetic interactions. This can lead to a candidate for disorder by disorder in two dimensions a huge number of quasi-degenerate ground-state configu- the kagome TFIM hence is a subject of great theoretical rations and a suppression of magnetic order even at zero interest. temperature. The macroscopic ground-state manifold of The central objective of this paper is the analysis of simple model systems is extremely sensitive to perturba- the gapped paramagnetic phase for the kagome TFIM to tions. Arbitrary linear combinations of the degenerate check whether it indeed persists for arbitrary field val- states may form novel ground states, yielding the great ues providing an instance of a disorder by disorder sce- variety of unusual effects some of which are not fully un- nario in two dimensions. For this purpose, we investigate derstood to date. the energy gap by high-order series expansions about the Perhaps the simplest and hence most tractable mod- high-field limit. Additionally, we determine the gap of els featuring such behaviour are the fully-frustrated Ising the corresponding low-energy mode for an infinitesimal systems in a transverse field [1]. On lattices like the tri- transverse field by means of a Lanczos based algorithm angular and the kagome lattice, the frustration results performed on appropriate finite clusters. Our findings from the incompatibility of the antiferromagnetic bonds suggest that high- and low-field gaps are indeed adiabat- in a single triangular plaquette. These systems exhibit ically connected implying that the kagome TFIM realizes a classical ground-state manifold with a macroscopic de- a disorder by disorder scenario. generacy which is lifted in the presence of an infinitesimal The analoque problem on the triangular lattice is dif- transverse field inducing quantum fluctuations. ferent. Here, series expansion of the one-particle gap In the limit of high transverse fields all lattices realize a about the high-field limit clearly indicates a second-order phase transition in the 3d XY universality class between
arXiv:1212.0736v1 [cond-mat.str-el] 4 Dec 2012 non-degenerate polarized phase. The situation for weak transverse fields however is much less trivial. A large the polarized and an ordered phase realized at low fields. class of these frustrated transverse field Ising systems ex- Our results are in quantitative agreement with quantum hibits a nontrivial quantum ordered phase arising from Monte Carlo simulations [9]. classical disorder, a scenario which is also known as order Finally, we show that a major difference between the by disorder [2–4]. In contrast, the kagome lattice plays triangular and the kagome TFIM lies in the existence a distinguished role as it seems to be very reluctant to of a localized low-energy mode for the kagome lattice order [5–7]. It is believed to be a quantum paramagnet that is completely flat up to order eight perturbation theory.√ In this√ order a specific momentum corresponding to the 3 × 3 structure is energetically favored. The emergence of this local mode and its disappearance at re- ∗Electronic address: [email protected] markably high orders in perturbation theory can be un- 2 derstood via the linked cluster expansion. Furthermore, (a) (b) w[2h] we generalize our findings to a larger class of systems. 2p This paper is organized as follows. In Sec.II we briefly 1.5 introduce the TFIM and discuss the physics of the low- p
field and high-field regime. In Sec.III we present the e1 e2 k2 0 derivation of the effective one-particle Hamiltonian in the high-field limit using the method of perturbative contin- -p uous unitary transformations (pCUTs) which is imple- 0.8 mented via a linked graph expansion. The one-particle -2p -p 0 p 2p k1 gap for the triangular and the kagome TFIM is analyzed in Sec.IV and in Sec.V. The more technical part dis- cussing the existence of flat bands and their breakdown FIG. 1: (a) Countour plot of the one-particle dispersion ωtr(~k) is placed in Sec.VI. Finally, Sec.VII provides a conclud- as a function of the momentum ~k = (k1, k2) on the trian- ing summary. gular lattice for x = 0.1. The red (solid) line corresponds to the first Brillouin zone. (b) Illustration of the complex phase of the eigen state corresponding to the one-particle√ √ ~ 2 2 II. MODEL gap at kmin = ( 3 π, − 3 π). The state exhibits a 3 × 3- structure. Note that the structure for√ the√ second gap mo- ~ 2 2 We study the two-dimensional Ising model in a trans- mentum kmin = −( 3 π, − 3 π) is also a 3× 3-structure with verse magnetic field on kagome and triangular lattices. a different orientation. We stress that the displayed arrows The corresponding Hamiltonian is given by do not represent spins.
X z z X x H = J σi σj + h σi (1)
The Hamiltonian (1) can then be written as
H X X 1 = E + nˆ + x b†b + b†b† + h.c. [2,4] 2h 0 i i j i j [3,3] i hi,ji [3,4] 0.8 [3,5] [3,7] = H0 + x (T0 + T+2 + T−2) = H0 + xV (3) [4,2] 0.6 [4,3] [4,4] [4,5] where x = J/2h is the perturbation parameter, [2h] † tr 0.02 [4,6] ∆ E0 = −1/2, andn ˆi = bi bi is the local density on site i. 0.4 [5,2] [5,3] We identify the Ising term as the perturbation V which [5,4] [2h] [6,4] can be decomposed into a sum of Tm operators incre- 0.2 ∆ [7,3] bare menting (or decrementing) the number of quasi-particles 0 0.302 0.304 0.306 0.308 x by m = {+2, −2, 0}. The pCUT transforms the initial 0 0 0.1 0.2 0.3 0.4 Hamiltonian (3) into a quasi-particle conserving effective x Hamiltonian of the form
X n X Heff = H0 + x C(m1, . . . , mn)Tm1 ...Tmn (4) n m¯ =0 0.34 L-M=0 L-M=+1 where the second sum runs over all possible L-M=+2 0.33 L-M=+3 L-M=-1 {m1, m2, . . . , mn} with mi = {+2, −2, 0} and L-M=-2 P L-M=-3 m¯ ≡ i mi = 0. The latter condition reflects the 0.32 QMC
block diagonality of the effective Hamiltonian Heff in the c number of elementary excitations allowing to restrict x 0.31 the calculations to subspaces with a defined number of quasi-particles. The coefficients C(m1, . . . , mn) have 0.3 been calculated as exact fractions up to high order in perturbation [10]. The monomials of Tm operators in 0.29 order n generate virtual fluctuations that involve at 3 4 5 6 7 8 9 10 11 12 r maximum n different bonds. Due to the cluster addi- tivity of the effective Hamiltonian, only linked virtual processes (these are processes that comprise bonds on simply connected clusters) have a finite contribution.
Consequently, the matrix elements of the irreducible L-M=0 L-M=+1 n-particle blocks of the Hamiltonian do possess a linked 0.85 L-M=+2 L-M=-1 cluster expansion [11–13]. L-M=-2 3d XY We calculated the effective Hamiltonian on the kagome 0.8 (triangular) lattice in the one-quasi-particle (1QP) sector
ν 0.75 up to order n = 13 (n = 11) in the perturbative param- eter x. This is done via a full linked-cluster expansion 0.7 taking into accout the 1830 (1784) topologically differ- ent graphs. Note that the pCUT provides results for the 0.65 thermodynamic limit that are exact up to the computed 0.6 order. 3 4 5 6 7 8 9 10 11 12 r
IV. TRIANGULAR LATTICE FIG. 2: (a) Bare series and dlogPad´eextrapolants of the one- particle gap ∆tr as a function of x for the triangular TFIM. In this section we discuss the one-particle properties Inset represents a zoom close to the quantum critical point. for the triangular TFIM which is confirmed to realize an (b) Illustration of the convergence behaviour of the critical order by disorder scenario. The triangular lattice can be values xc resulting from the dlogPad´eextrapolants [L/M] in described by a single-site unit cell. As a consequence, dependence of the considered order r = L+M +1. Connected points of the same color correspond to extrapolants with con- the 1QP block of the effective Hamiltonian is diagonal- stant L − M. The dashed line represents the quantum Monte ized by Fourier transformation giving a single band for (qmc) Carlo result xc = 0.303 [9]. The horizontal dotted lines the one-particle excitations which is fully characterized indicate the corresponding margin of error ±0.009. (c) Crit- by its dispersion ωtr(~k). A contour plot of the resulting ical exponent ν extracted from dlogPad´eextrapolants shown dispersion for a small coupling x = 0.1 is depicted in in (b) as a function of r. Dashed horizontal line indicates Fig.1(a). The minima of the dispersion ωtr are located the critical exponent ν3dXY ≈ 0.67 of the 3d XY universality class. 4
~ 2 2 at kmin = ±( 3 π, − 3 π) (see Fig.1(a)) defining the one- V. KAGOME LATTICE tr tr ~ particle gap ∆ ≡ ω (kmin). Up to order n = 11, this gap is given by In this section we aim at determining the one-particle gap of the kagome TFIM on the whole parameter axis. 3 15 243 1671 ∆tr = 1 − 3x + x2 − x3 + x4 − x5 Since the kagome lattice has a three-site unit cell (see 2 2 8 8 Fig.4(a)), the effective one-particle Hamiltonian H1QP 22275 162855 9700617 eff + x6 − x7 + x8 is only reduced to an hermitian 3 × 3 matrix ωlm(k1, k2) 16 16 128 with l, m ∈ {1, 2, 3} per momentum ~k using Fourier 595490847 2010551941313 − x9 + x10 transformation. One therefore has three one-particle 1024 442368 bands for the kagome TFIM. Due to the lattice sym- 1777064899901 − x11 . (5) metries, it is sufficient to calculate just the two matrix 49152 elements ω11(k1, k2) and ω12(k1, k2). The remaining ma- trix elements can be obtained by the following relations It is instructive to plot the complex phase of the eigen state corresponding to the gap momenta ~k which is min √ √ ω22(k1, k2) = ω11(k2, k1) shown in Fig.1(b). For the triangular lattice a 3 × 3 ω (k , k ) = ω (−k , k − k ) structure is found and one expects that linear combina- 33 1 2 11 2 1 2 tions of such structures are the most preferable ordering ω13(k1, k2) = ω12(k1 − k2, −k2) patterns (see below). ω23(k1, k2) = ω12(k2 − k1, −k1) In order to detect a possible second-order quantum ∗ ωlm = (ωml) . (6) phase transition, we use dlogPad´etechniques to extrapo- tr late the one-particle gap ∆ . To this end, various extrap- There are two dispersive high-energy bands exhibiting a olants [L, M] are constructed where L denotes the order finite bandwidth already in first-order perturbation the- of the numerator and M the order of the denominator. ory. Most interestingly, the remaining low-energy band In Fig.2 the corresponding results for the one-particle ωlow is completely flat up to and including order n = 7 gap ∆tr are displayed. The gap shows a strong tendency to close in a narrow 243 2609 ω(7) (~k) = 1 − 2x + 2x2 − 3x5 + x6 − x7 . (7) window around xc ≈ 0.305. These results are in quan- low 8 8 titative agreement with quantum Monte Carlo (QMC) (qmc) simulations giving xc = 0.303 ± 0.009 [9]. The con- Note that in general the presence of such a flat one- vergence behaviour of the critical values xc for different particle band indicates the existence of a local mode. We (qmc) give a detailed discussion on the physics of such local extrapolants compared to xc is shown in Fig.2(b). Altogether, our findings suggest a second-order quan- modes in Sec.VI. Here we continue with the analysis of (qmc) the one-particle gap. As the low-energy band starts to tum phase transition in the vicinity of x = 0.303 and c disperse in order n = 8, quantum fluctuations occuring x = 0.305. c in this order lift the degeneracy in momentum-space and Furthermore, one can extract the critical exponent ν select a specific gap momentum which is energetically from the dlogPad´eextrapolation. The corresponding re- favored. The resulting dispersion ω(13)(~k) of order 13 is sults are shown in Fig.2(c). Averaging over all dlogPad´e low extrapolants [n, m] of order 11 with |n|, |m| ≥ 2 , one ob- shown for x = 0.1 in Fig.3. tains ν = 0.708 ± 0.012 which is slightly larger than the expected critical exponent zν ≈ 0.67 for the 3d XY uni- w [2h] 2p versality class [14]. Interestingly, if one biases the dlog- 0.819987 (qmc) Pad´eextrapolants to have a pole at xc = 0.303, the corresponding averaged exponent νbias = 0.658 ± 0.016 is p much closer to 3d XY. k The closing of the gap corresponds to a quantum phase 2 0 transition between an disordered paramagnetic phase for -p x < xc and a quantum ordered phase for x > xc. The symmetry√ √ of the ordered phase is reduced according to the 3 × 3 structure which is consistent with the bond 0.819986 -2p -p 0 p 2p ordered phase suggested in Ref.6. k1 The triangular TFIM therefore realizes an order by dis- order scenario. A second-order quantum phase transition (13) ~ in the 3d XY universality class is detected by the closing FIG. 3: Lowest one-particle energy band ωlow (k) as a func- of the high-field gap. This is different for the analoque tion of ~k = (k1, k2) for the kagome TFIM at x = 0.1. Red problem on the kagome lattice as we will elaborate on in solid line indicates the first Brillouin zone. Note the extreme the next section. smallness of the bandwidth. 5
~ The locations of the energy gap are kmin = 2 2 (a) 3 3 ±( 3 π, − 3 π). In order n = 13, one finds the following expression for the one-particle gap 1 2 1 2 3 3 3 2 5 243 6 2609 7 ∆ = 1 − 2x + 2x − 3x + x − x 1 2 e 8 8 1 2 1 2 2 e1 28531 3448807 149125883 3 3 + x8 − x9 + x10 16 512 6144 1 2 1 2 1340557475 168486121055 − x11 + x12 12288 294912 113062814254323109 − x13 . (8) (b) 36238786560 The form of the resulting eigenvector (up to a global phase shift) is given by
1 i 2 π −i 2 π ~v = √ 1, e 3 , e 3 , (9) 3 where the three coordinates of ~v correspond to the three sites of the unit cell (see Fig.4). If one plots the√ complex√ phase of the full lattice eigen state, one finds a 3 × 3- structure illustrated in Fig.4(b). In an analogue fashion FIG. 4: (a) The kagome lattice has a three-site unit cell. We to the triangular lattice, one would expect that, if the define unit vectors ~e1 and ~e2 as shown on the right-hand side of the upper panel. The position of a unit cell is then given kagome TFIM realizes√ an order√ by disorder scenario, the ordered phase has a 3 × 3 structure. Interestingly, by (n1, n2) with ni ∈ Z. The three sites of the unit cell are denoted by {1, 2, 3}. (b) Illustration of the eigen state corre- this is suggested as the ordering pattern in the bond- sponding to the one-particle gap. Arrows denote the√ complex√ ordered phase of the kagome Ising antiferromagnet in a phase of this eigenvector. The pattern realizes a 3 × 3- longitudinal field [5]. structure. Note that the structure for√ the√ second gap mo- Furthermore, a high-temperature expansion of the ~ 2 2 mentum kmin = −( 3 π, − 3 π) is also a 3× 3-structure with classical kagome Heisenberg model displays a striking a different orientation. We stress that the displayed arrows similarity to our findings: Fluctuations (in this case ther- do not represent spins. mal) up to order n = 7 in β = 1/T do not pick any specific momentum. It is only√ in order√ eight that the momentum corresponding to a 3 × 3-structure is fa- corresponds to u = 1. The resulting extrapolations of vored [15, 16]. Consequently, thermal fluctuations of the the high-field gap are shown in Fig.6. classical kagome Ising model behave not unlike quantum For u < 0.4, all extrapolations as well as the bare se- fluctuations in the kagome TFIM in the high-field limit. ries are well converged. A closing of the gap in this re- One might wonder whether the origin of this equivalence gion can therefore be excluded. Note that the bare series can be traced back to geometrical relations. This is in- of ∆(x) converges very slowly as the absolute value of deed true as we will discuss in Sec.VI. the coefficients grows exponentially with ascending order In the following we analyze the one-particle gap on the and, additionally, as the sign of the coefficients alter- whole field axis in order to clarify whether the kagome nate. But also far beyond the high-field regime most of TFIM realizes a disorder by disorder scenario. At this the extrapolants indicate a clear trend. Though isolated point we assume that binding effects are small for the extrapolants exhibit instabilities, there is no tendency TFIM and, consequently, multi-particle sectors play no that the gap closes at a certain critical value. More- role for the excitation gap (for any J/h). This assump- over, most of the extrapolations approach a finite value tion is reasonable since the leading contribution to the ∆ ≈ 0.20 ± 0.03 for u = 1. two-particle interaction is repulsive and of order n = 2 ∞ perturbation theory. From the pCUT result the lowest one-particle mode For the extrapolation it is convenient to perform an in the high-field limit is characterized by a momen- ~ 2 2 Euler transformation tum kgap = ( 3 π, − 3 π) whereas the ground state has u ~k = (0, 0). If the gap stays finite for arbitrary x with x =: (10) u − 1 ∆(x → ∞) ≥ 0, corresponding to a disorder by disorder scenario, it implies that the ground state and the lowest which maps the parameter axis x ∈ [0, ∞[ on the finite one-particle excitation in the high-field limit can be adi- interval u ∈ [0, 1]. One expects that the dlogPad´eex- abatically connected to the low-field regime. It would be trapolants in u are more stable within the finite interval therefore desirable to additionally analyze the low-field u ∈ [0, 1] allowing to estimate the gap for x → ∞ which limit of the kagome TFIM which we focus on next. 6
In the low-field limit, the kagome TFIM is best ex- N=9 N=18 pressed as follows 3 1 2 3 24
HTFIM X h X = σzσz + σx = H + yV , (11) 1 2 3 1 6 1 5 6 1 J i j J i 0 hi,ji i 3 1 2 4 2 3 4 2 where y = h/J. For zero field y = 0, the Hamiltonian 2 3 6 1 5 6 describes the classical kagome Ising antiferromagnet with 2 3 macroscopic ground-state degeneracy. An infinitesimal N=27 field y 1 lifts this degeneracy leading to an energy N=36 splitting between states with different momenta ~k. 6 4 1 2 3 We assume that the two lowest energy levels are char- 3 9 2 ~ ~ acterized by momenta k = 0 (ground state) and kgap 7 11 12 4 1 7 8 1 (lowest excited state) such that they are adiabatically 3 8 9 10 1 connected to the corresponding states in the high-field 6 4 5 6 4 limit. One can show that the energy gap in the low-field 12 4 5 6 7 11 9 limit ∆(˜ y) (in units of J) and the gap ∆(x) (in units of 2 3 9 10 1 2 3 8 2h) are related by the following equation 8 1 7
7 11 12 4 ∂ 6 4 ∆(˜ y) = 2 lim ∆(x) . (12) ∂y y=0 x→∞
Consequently, the gap ∆(x → ∞) can be determined FIG. 5: The finite-size samples of the kagome lattice with pe- by means of first-order degenerate perturbation theory riodic boundary conditions which are used for the first-order about the low-field regime. degenerate perturbation theory in the low-field limit. Here Due to the extensive ground-state degeneracy, one N denotes the total number of spins. The finite-size samples has already in order-one degenerate perturbation the- are built by the thick black triangles labeled by an integer ory infinitely many states in the thermodynamic limit. number. Thin triangles are displayed to visualize periodic We therefore approach this problem by applying first- boundary conditions. order degenerate perturbation theory on appropriate fi- nite clusters of different system sizes in order to obtain an estimate of the energy splitting in the thermodynamic limit by finite-size scaling. One expects that the energy splitting converges to the gap ∆(x → ∞) rather rapidly with increasing system size because a finite energy gap 1 implies a finite correlation length. [3,5] [3,7] In order to ensure conservation of momentum the clus- [4,4] 0.25 ters have to exhibit translational invariance which can 0.8 [4,5] [4,6] 0.2 be realized by periodic boundary conditions. Since finite [4,7] [4,8] [3,5] DlogPade [3,7] 0.15 ~ [5,6]DlogPade [4,4] systems only allow for discrete values of k depending on 0.6 DlogPade [4,5] 0.6 0.8 1 0 0.05 0.1 0.15 0.2
DlogPade [4,6] [5,7]DlogPade [4,7] 1/N DlogPade [4,8]
DlogPade [5,6] their sizes and geometrical shapes, the choice of the clus- [6,5]DlogPade [5,7]
[2h] DlogPade [6,5]
DlogPade [6,6]
[6,6]DlogPade [7,3] ∆ ters plays a crucial role. We are interested in the en- DlogPade [8,4] [8,3]DlogPade [5,5] 0.4 N=9
N=18 [8,4]N=27 ~ N=36 ergy having the gap momentum kgap which is realized in N=2 √ √ [5,5] systems allowing for a 3 × 3-structure. Their corre- N=9 0.2 N=18 sponding unit cell contains N = 9 spins. Fig.5 shows N=27 N=36 the appropriate clusters considered in this paper. N= ∞ 0 Technically, we use a Lanczos based algorithm which 0 0.2 0.4 0.6 0.8 1 allows to calculate the lowest first-order energy correc- u tions in subspaces with fixed momentum ~k. For the case of N = 36 spins (ground-state degeneracy g ∼ 1010) we need about n ≈ 100 Lanczos steps for an accuracy of FIG. 6: One-particle gap ∆/2h as a function of u. Lines cor- L respond to dlogPad´eextrapolants in u = x while symbols 10−6. x+1 represent the first-order degenerate perturbation theory re- The obtained energy gaps for the different system sizes sults of the low-field limit for the different finite-size samples. are displayed in the inset of Fig.6. Using a 1 /N- (Left Inset): Zoom around u = 1. (Right Inset): First-order scaling, one finds for the thermodynamic limit the es- degenerate perturbation theory results of the one-particle gap timate ∆x=∞/2h ≈ 0.18. This value is remarkably con- against 1/N. Solid black line represents a linear fit of the data. sistent with the value obtained from the high-field expan- sion. 7
As a further consistency check, the value of the high- (a) (b) j field extrapolants for u = 1 is biased to the value ∆ /2h = 0.18 calculated in the low-field limit. The ∞ p* p* ii' resulting extrapolants are shown in Fig.7. The en- p* 12 61 i i' hanced stability of the extrapolation indicates a reason- + - + - 1 2 p* (1) (2) able consistency between the high-field and the low-field 23 - 6 3 + - + results. Altogether, the results provide considerable ev- p* 56 5 4 idence that the gap stays finite for all values of x. We + - + - conclude that the kagome TFIM exhibits a disordered p* p* 34 paramagnetic phase on the whole parameter axis. 45
FIG. 8: (a) Localized mode |ψloci on a hexagon of the kagome 1 lattice. The notion ± denotes the sign in front of the one- bias[2,5] particle states contained in |ψloci. Filled dark circles corre- bias[3,5] spond to the six sites p∗ surrounding the central hexagon. (b) bias[3,6] ij 0.8 bias[4,4] Example of a destructive interference between the two virtual bias[4,5] hopping processes (1) (thick dashed bonds) and (2) (thick bias[4,6] solid bonds) exhibiting the same absolute value t(1) = t(2) of 0.6 bias[4,7] ij i0j bias[4,8] the hopping amplitude. The initial states are given by the bias[5,5] one-particle states |ii and |i0i while the final state of both [2h] bias[5,6] ∆ bias[5,7] processes is given by |ji. Due to the opposite prefactor of |ii 0.4 0 bias[6,5] and |i i in |ψloci, both processes interfere destructively. bias[6,6] bias[7,3] 0.2 bias[8,3] bias[8,4] bias[9,3] destroyed in this order for the kagome TFIM. The localized mode of the kagome TFIM is given by 0 0 0.2 0.4 0.6 0.8 1 a superposition of one-particle states located on a sin- u gle hexagon with alternating relative phases as shown in Fig.8(a). According to the displayed numbering a local- ized mode on a certain hexagon can be written as FIG. 7: Biased dlogPad´eextrapolations of the one-particle gap as a function of u. The extrapolants are biased to 1 |ψloci = √ (|1i − |2i + |3i − |4i + |5i − |6i) (13) ∆(u = 1)/2h = 0.18 obtained from the 1/N-scaling in the 6 low-field limit. where |ii denotes a one-particle state with a particle (spin flip) located at site i. Based on the quasi-particle picture the locality can be VI. FLAT BANDS explained by destructive intereferences of virtual hopping processes which prevents the particles from escaping the hexagon. One of the key differences between the kagome and We start by explicitly discussing the locality of |ψloci in the triangular TFIM is the existence of a completely flat leading order perturbation theory. The first-order con- low-energy band up to order eight perturbation theory tribution H(1) of the effective Hamiltonian (4) is given for the kagome lattice. It is therefore the purpose of this eff by section to analyze the physics of such flat bands in more detail. This is done first for the kagome TFIM and it is (1) X H = xT = x τ b , (14) later generalized to a larger class of systems. eff 0 0 b In general, the momentum independence of an one- particle energy band indicates the existence of a localized where the sum is taken over all bonds of the lattice b one-particle mode as any superposition of momentum ba- and the τ0 correspond to nearest-neighbour hopping pro- 1QP cesses on bond b. The state |ψ i is then obvisously an sis states provides an eigen state of Heff . A uniform loc (1) superposition in momentum space corresponds to a state eigen state of Heff , since hopping processes escaping the that is completely localized in the spatial domain. The hexagon interfere pairwise destructively on each of the 1QP ∗ effective Hamiltonian Heff of the kagome TFIM there- six sites pij surrounding the hexagon (see also Fig.8a). fore gives rise to a local one-particle mode which is an Such destructive interferences are the guiding principle exact eigen state of the Hamiltonian up to order 7 in to understand the locality of |ψloci also in higher orders, x. As the low-energy mode starts to disperse in order which is best understood via the linked cluster expansion. n = 8, quantum fluctuations occuring in this order lift Indeed, due to the linked cluster theorem, the general the degeneracy in momentum space and local modes are expression of the effective Hamiltonian in order n can be 8 decomposed as a sum of linked virtual hopping processes as follows (a)
(n) n X X b b b j j H = x C(m ¯ ) τ 1 τ 2 . . . τ n eff m1 m2 mn m¯ =0 b1,b2,...,bn X X X p* p* = xn C(m ¯ ) τ b1 τ b2 . . . τ bn i i' m1 m2 mn + - + - Sn Csub C = {b } m¯ =0 sub i=1 i - + - + n X (n) = x Γ , (15) + - + - Csub Csub where the last sum is taken over all connected subclusters Csub built by the collection of linked bonds {bi}. Note that the same bond can appear several times in the same operator product τ b1 τ b2 . . . τ bn where τ bi acts on bond m1 m2 mn mj bi and mj ∈ {2, −2, 0} refers to a pair creation, pair anni- hilation, or single-particle hopping process. The operator (b) Γ(n) represents all virtual fluctuations of the connected Csub j j subcluster Csub in order n. Here we are interested in the single-particle hopping p* p* amplitudes tij describing a hopping from site i to site j i i' + - + - tij = hj|Heff |ii − δi,jh0|Heff |0i , (16) - + - + - where |0i denotes the zero-particle state being equivalent + + - to the fully polarized state. Consequently, the contribu- tion of a connected subclusters Csub to such a hopping element is given by
C X (n) (n) t sub = hj|Γ |ii − δ h0|Γ |0i Csub ij Csub i,j Csub FIG. 9: (a) Left figure illustrates a hopping process tij n on a subcluster Csub visualized as thick solid bonds. The C˜ X Csub,(n) sub ˜ = t . (17) right part represents the counter process ti0j on Csub. The ij ∗ n dashed vertical line is the symmetry axis through site p when acting on |ψloci. (b) Example of a hopping process that is not Csub annihilated by a counter process. Any counter cluster has a In the following we visualize a contribution tij by col- different topology and therefore a different contribution. This oring the bonds bi of the subcluster Csub. Most impor- tantly, the tCsub only depend on the topology of the sub- leads to a finite hopping amplitude of a particle leaving the ij hexagon. cluster Csub. The latter property is crucial to pinpoint the properties of the local mode |ψloci. The dispersion emerging in order n = 8 is then related to non-trivial processes that do not second more complex example of such counter clusters is occur for n < 8. In essence, all processes up to order shown in Fig.9(a). n = 7 cancel out. This can be understood as follows. There are also subclusters that do not possess counter In order to escape the hexagon any particle has to tra- clusters of the same topology as exemplified in Fig.9(b). verse at least one of the sites p∗ enclosing the hexagon. These subclusters contain virtual fluctuations that are ij not completely annihilated by counter processes when Now, envision a process t(1) on a subcluster (1) as shown ij acting on |ψloci. Hence, they correspond to fluctuations in Fig.8(b). The initial state |ii is assumed to have a that lead to a breakdown of the local mode. prefactor +1 in |ψloci and one considers the particle to The next crucial step is to identify the lowest pertu- ∗ escape through the site pii0 to the final state |ji. A pos- bation order for which a hopping process exists which sible counter process on a subcluster (2) is then given by has at least one counter cluster with a different topol- mirroring only the part of subcluster (1) being ”below” ogy. In general those graphs are related to closed loops ∗ ∗ the site pii0 (one can always perform a global rotation of comprising two different p that do not lie on the same ∗ ij the lattice such that pij is located on top of the hexagon). symmetry line as any counter cluster cuts such a loop. In this case the topology of the subcluster (2) is equiva- Clearly, the smallest nontrivial process with a loop is of lent to the subcluster (1) and, as a consequence, one has order n = 6 as shown in Fig. 10(a). In this order, the (1) (2) 0 tij = ti0j . Since the prefactor in |ψloci of the state |i i is hopping process on this subcluster is always local (i = j) −1, the sum of both processes contained in Eq. 15 cancel for the TFIM and hence does not contribute to the dis- out when acting on the state |ψloci (see Fig.8(b)). A persion. For n = 7 the only process with a loop, where 9
∗ the final state |ji is not part of |ψloci (and therefore could structive interference every gate prr0 has to be connected lead to a dispersion), is displayed in Fig. 10(b). But, this to at least two sites belonging to L. This restricted area process has the special property that the site j is located provides a local mode up to a certain order n, if it allows on a symmetry line of the lattice (one finds j = j0 for the for a superposition of one-particles states with alternat- counter cluster). ing sign ±1 inside L, such that all hopping processes In compliance with our results, the first hopping pro- Csub,(m) tij on all subclusters Csub for m ≤ n have a topo- cess leading to a breakdown of |ψloci is indeed of or- logically equivalent counter cluster. For the triangular der n = 8. Figure 10 depicts one such hopping process lattice, it turns out that there is no local arrangment Csub,(8) tij on the corresponding Csub. Besides this subclus- that complies to the stated criteria. ter, there are also other processes leading to a dispersion in order n = 8. All of them are related to non-trivial General construction rules for localized magnon states processes with loops. are already well known from the intensive study in the context of macroscopic magnetization jumps [17]. Our geometrical observations are fully consistent with the (a) (b) general rules stated in these works. Beyond these con- struction rules, we are able to identify single processes that lead to a dispersion which also allows to determine the exact order of locality on the basis of a quasi-particle j'=j picture. The dispersion of a localized mode is affected by j=i i ∗ virtual processes that close a loop between different pij. + - + - ∗ The pij are sites enclosing the local mode located in L. ∗ - + - + Two sites pij are considered as different if they do not + - + - share the same sites in L. Let us illustrate these principles by means of another TFIM on a frustrated lattice. For instance, consider the n = 6 n = 7 checkerboard lattice shown in Fig. 11. Here the local- sq ized state |ψloci is given by an arrangement of particles (c) with alternating phase located at the sites of a square plaquette. The displayed process has a non-trivial loop j ∗ containing two different sites pij leading to a breakdown sq of |ψloci. In the case of the TFIM the corresponding hopping process first appears in order n = 5. Calcula- i + - tions for the checkerboard TFIM indeed show that the dispersion of the lowest one-particle mode is completely - + flat up to and including order n = 4 in accordance with + - our predictions.
n = 8
p*12 FIG. 10: (a) Illustration of the only looped process in or- j der n = 6 as thick bonds. The particle in the inital and final state is located at the same site (|ii = |ji). Hence, this p* + - process does not conrtibute to the dispersion. (b) Hopping 41 1 2 process contributing for orders n ≥ 7. The corresponding 34 counter process is represented by thick blue dashed bonds. - + (c) Hopping process contributing for orders n ≥ 8. For this case any counter cluster has a different topology leading to the breakdown of the local mode |ψloci.
Based on these geometrical observations for the kagome TFIM one finds the following criteria for the sq 1 FIG. 11: Localized mode |ψloci = 2 (|1i − |2i + |3i − |4i) on existence and the breakdown of local modes on a given the square plaquette of the checkerboard lattice. The dis- lattice. Consider an arrangement of one-particle states played hopping process from site 1 to site j involving 5 thick located inside a restricted area L. The sites enclosing solid bonds leads to a dispersion as it closes a loop between ∗ the two different sites p∗ and p∗ . this local area are denoted by prr0 . It is clear that any 41 12 ∗ subcluster leaving L has to pass through at least one prr0 which one can think of as a gate. In order to allow de- 10
VII. SUMMARY axis. Strong evidence is found that the low-field and high- field limit are adiabatically connected. Consequently, the We have studied the triangular and the kagome TFIM kagome TFIM realizes a disorder by disorder scenario. at zero temperature. Such highly-frustrated Ising models Let us also address similarities and differences between represent the simplest examples displaying an intriguing the impact of quantum fluctuations (in this work induced interplay of classical frustration and quantum fluctua- by the transverse field) and of thermal fluctuations on tions. Indeed, the tranverse magnetic field represents a classical magnets on the kagome lattice. Whereas ther- singular perturbation on the extensive number of ground mal fluctuations also do not induce order for Ising spins, states present for the classical Ising problem. the situation is different for the Heisenberg case. Here Generically, one expects an order by disorder sce- the high-temperature expansion has an analogous struc- nario for such problems [5,6]. The corresponding zero- ture, in particular with regard to the momentum space temperature phase diagram then consists of two phases, degeneracy up to order eight in J/T [15]. There, up to an ordered one at low fields and a polarized phase at Mermin-Wagner physics, dipolar, quadrupolar, and oc- high fields separated by a quantum phase transition. Al- tupolar orders appear to be induced by thermal fluctua- ternatively, a fascinating disorder by disorder scenario is tions [15, 16, 18–22]. possible. Here no symmetry-broken phase is realized and The momentum independence of the low-energy band a polarized phase is present on the whole parameter axis. in the first seven orders of perturbation theory for the In this work we give strong evidence that the kagome kagome TFIM can be traced back to the existence of local TFIM displays a notable first instance of an disorder by modes in the spatial domain. It is only in order eight disorder scenario in two dimensions. In contrast, the tri- that quantum fluctuations occur which do not interfere angular TFIM falls in the more conventional class of sys- desctructively and therefore lead to the breakdown of tems showing an order by disorder. the local modes. We have identified a mechanism for the Our findings are based essentially on high-order se- (dis)appearance of this degeneracy which readily applies ries expansions of the one-particle gap about the high- in a generalised setting. field limit. To this end a full linked graph expansion Our findings might inspire future research about sim- is performed using the pCUT method. For the trian- ilar questions in other classes of problems beyond trans- gular TFIM extrapolations of the one-particle gap give verse field Ising models. One fascinating idea would be to convincing evidence for a gap closing and therefore for establish a link between the importance of flat bands in a second-order quantum phase transition. The location magnetic systems discussed in this work and the physics of the critical point is in quantitative agreement with of fractional Chern insulators also based on the existence quantum Monte Carlo simulations [9]. Furthermore, the of flat bands on the one-particle level. obtained critical exponent is fully compatible with the suggested 3d XY universality class. In contrast, the extrapolation of the one-particle gap for the kagome TFIM displays no tendency to close. Most VIII. ACKNOWLEDGEMENTS of the extrapolations suggest a finite gap of the order ≈ 0.2 in units of 2h even in the low-field limit, which is KPS acknowledges ESF and EuroHorcs for funding consistent with first-order degenerate perturbation the- through his EURYI. RM thanks Premi Chandra, Sergei ory about the classical Ising antiferromagnet. The latter Isakov and Shivaji Sondhi for collaboration on closely is done for finite systems yielding a gap ≈ 0.18 in units related work. This work was in part supported by the of 2h in the thermodynamic limit. Put together, we have Helmholtz Virtual Institute ”New states of matter and determined the one-particle gap on the whole parameter their excitations”.
[1] R. Liebmann, Statistical mechanics of Periodic Frus- B 64, 134424 (2001) trated Ising Systems (Springer,Berlin, 1986). [9] S.V. Isakov and R. Moessner, Phys. Rev. B 68, 104409 [2] K. Kanˆoand S. Naya, Prog. Theor. Phys. 10, 158 (1953). (2003). [3] J. Villain, R. Bidaux, J.P. Carton, and R. Conte, J. Phys. [10] C. Knetter and G.S. Uhrig, Eur. J. B 13, 209 (2000). (Paris) 41, 1263 (1980). [11] C. Knetter, K.P. Schmidt and G.S. Uhrig, J. Phys. A 36, [4] E.F. Shender, Sov. Phys. JETP 56, 178 (1982). 7889 (2003). [5] R. Moessner and S.L. Sondhi, Phys. Rev. B 63, 224401 [12] M.P. Gelfand, R.R.P. Singh and D. Huse, J. Stat. Phys. (2001). 59, 1093 (1990). [6] R. Moessner, S.L. Sondhi and P. Chandra, Phys. Rev. [13] M.P. Gelfand, Solid State Commun. 98, 11 (1996). Lett. 84, 4457 (2000) [14] A.P. Gottlob and M. Hasenbusch, J. Stat. Phys. 77, 919 [7] P. Nikoli´c and T. Senthil, Phys. Rev. B 71, 024401 (1994). (2005). [15] A.B. Harris, C. Kallin and A.J. Berlinsky, Phys. Rev. B [8] D.J. Priour, M.P. Gelfand and S.L. Sondhi, Phys. Rev. 45, 2899 (1992). 11
[16] J.N. Reimers and A. Berlinsky, Phys. Rev. B 48, 9539 Phys. Rev. Lett. 68, 855 (1992). (1993). [20] I. Ritchey, P. Chandra and P. Coleman, Phys. Rev. B 47, [17] J. Schulenburg, A. Honecker, J. Schnack, J. Richter and 15342 (1993). H.-J. Schmidt, Phys. Rev. Lett. 88, 167207 (2002). [21] M. Zhitomirsky, Phys. Rev. B 78, 094423 (2008). [18] D.A. Huse and A.D. Rutenberg, Phys. Rev. B 45, 7536 [22] G.-W. Chern and R. Moessner, arXiv:1207.4752 (2012). (1992). [19] J.T. Chalker, P.C.W. Holdsworth and E.F. Shender,