betaFeedback

WorksPhysics and Astronomy

Emergent Anyons in a Weakly Interacting Gauged Chern Insulator

Abstract

We derive a microscopic relation between the many-body Chern number CC of the fermionic ground state and the braiding of fluxes in Z2\mathbb{Z}_{2} gauge theory coupled to matter: under a spectral gap condition, dressed flux loops obey W2W1=eiπ2CW1W2W_2W_1=e^{i\fracπ{2}C}W_1W_2 on the ground-state sector. We realize C=1C=1 by gauging charge parity in a weakly interacting Qi-Wu-Zhang insulator. Reflection positivity and chessboard estimates show that flux defects are massive, which yields a gapped, topologically ordered phase with four locally indistinguishable ground states.

Introduction. Topologically ordered phases provide one of the most striking manifestations of collective quantum behavior [1]. Unlike symmetry-breaking phases, they cannot be characterized by a local order parameter and are instead detected through robust ground-state degeneracy, local indistinguishability [2, 3], and quasiparticle excitations with nontrivial braiding statistics [4, 5, 6]. Exactly solvable lattice models have played a central role in establishing this paradigm [3, 7, 8], but proving topological order and anyonic statistics in realistic interacting, non exactly solvable microscopic systems remains considerably more difficult. The main challenges are first to rigorously establish the presence of a spectral gap, and second to connect the statistics of excitations to the geometry of the many-body ground-state space.

A natural route toward such systems is provided by gauged topological insulators. Gauging a subgroup of the global U(1)U(1) charge symmetry promotes symmetry fluxes to physical excitations whose statistics reflect the topological response of the fermions. At the level of effective field theory, this mechanism is captured by Chern–Simons theory [2, 9, 10, 11, 12, 13]. A microscopic derivation, however, requires control of the preferred gauge-flux sector, the energy cost of flux defects, the persistence of a topological ground-state manifold, and the operators implementing flux transport.

In [14], using reflection positivity and quasi-adiabatic tools, we established a microscopic relation between the algebra of dressed gauge-flux loops and the Hall response of the fermionic matter, identifying their commutator with a ground-state holonomy. The matter ground state has uniform π\pi flux and vanishing Chern number: the dressed flux loops commute, and the resulting anyon algebra is that of the toric code [3]. For related microscopic results connecting Hall response and flux statistics, see [15].

In this Letter we derive the general flux-loop relation and realize its nonzero-Chern case, in a microscopic setting. We consider the Qi–Wu–Zhang (QWZ) model [16] with a weak on-site Hubbard interaction and gauge the Z2\mathbb{Z}_2-subgroup of its U(1)U(1) charge symmetry. Reflection positivity selects a zero-flux magnetic vacuum. Combined with the chessboard estimate, it allows us to quantify the energy cost of flux defects, yielding a positive flux mass. The fermionic gap remains open under weak interactions [17], preserving the Chern number C=1C = 1 and the quantized Hall response [18]. A uniform gap for all boundary twists then allows us to construct quasi-adiabatic spectral flow, control the splitting of the four physical ground states, and establish their local indistinguishability.

The closest exactly solvable analogue is Kitaev’s honeycomb model [7]. There too, fermions hop in a static Z2\mathbb{Z}_2 gauge field with conserved plaquette fluxes, and the Chern number ν\nu of the fermion bands fixes the anyon content through the sixteenfold way. Our fermions are complex rather than Majorana, so that ν=2C\nu= 2C: the gauged C=1C = 1 insulator realizes the Abelian ν=2\nu= 2 entry of the sixteenfold way. The original honeycomb model, with ν∈{0,±1}\nu\in\{0, \pm1\}, does not reach it; exactly solvable generalizations do [19], but only in π\pi-flux or vison-crystal sectors whose stability is established numerically, since Lieb’s theorem no longer applies to them. The two settings also differ in what can be proven. In [7] each flux sector is solved exactly, but the vortex-free ground state is inferred from Lieb’s flux-phase theorem [20], and vortex energies are obtained numerically or perturbatively; to our knowledge no uniform vortex mass, and hence no uniform many-body gap, has been established. Here the matter is interacting and time reversal is broken from the outset, yet we prove that the zero-flux sector is the magnetic vacuum, give an explicit lower bound on the energy of every flux defect, and combine it with a uniform fermionic gap into a uniform physical gap.

The geometric content of the construction appears in the algebra of dressed flux loops. The underlying physical picture is that adiabatic flux insertion, followed by a gauge transformation, can connect distinct topological ground states [21, 22]. We implement this process using Hastings’ quasi-adiabatic continuation [23, 24], which realizes parallel transport in the many-body ground-state line bundle over the boundary-condition torus [25, 26]. Gauge covariance identifies the commutator of two noncontractible flux loops with the holonomy around the boundary of a quarter of this torus. Locality makes the Berry curvature asymptotically independent of the twists, fixing the commutation phase to eiπ2Ce^{i\frac{\pi}{2}C}, where CC is the many-body Chern number. For C=1C=1, the gauged interacting Chern insulator thus realizes a nontrivial vortex braiding algebra starting from an explicit microscopic Hamiltonian.

The model. Let ΛL=(Z/LZ)2\Lambda_L=(\mathbb{Z}/L\mathbb{Z})^2, L∈4NL\in4\mathbb{N}. Write cx,α±c_{x,\alpha}^{\pm} for the fermionic creation and annihilation operators, α=1,2\alpha=1,2, and place one gauge qubit on each nearest-neighbour bond. The bond operators σ^xy3\hat{\sigma}_{xy}^{3} represent the Z2\mathbb{Z}_2-gauge potential, while σ^xy1\hat{\sigma}_{xy}^{1} is the conjugate electric-field operator. In compact form,

Hλ=∑x,y,α,βcx,α+h(σ^)xα,yβcy,β−+λ∑x(nx,1−12)(nx,2−12)(1)H_\lambda=\sum_{x,y,\alpha,\beta}c_{x,\alpha}^{+}h(\hat{\sigma})_{x\alpha,y\beta}c_{y,\beta}^{-}+\lambda\sum_x\left(n_{x,1}-\frac{1}{2}\right)\left(n_{x,2}-\frac{1}{2}\right) \tag*{(1)}

with, for t>0t>0 and m∈Rm\in\mathbb{R},

h(σ^)=m1⊗τ3−t2∑μ=12[Tμ(σ^)⊗Dμ+h.c.],(2)h(\hat{\sigma})=m\mathbb{1}\otimes\tau^3-\frac{t}{2}\sum_{\mu=1}^{2}\left[T_\mu(\hat{\sigma})\otimes D_\mu+\mathrm{h.c.}\right], \tag*{(2)}

where τμ\tau^\mu are Pauli matrices acting on the spin degrees of freedom of the fermions and

(Tμ(σ^))xy=σ^x,x+eμ3δy,x+eμ,Dμ=τ3−iτμ.\left(T_\mu(\hat{\sigma})\right)_{xy}=\hat{\sigma}_{x,x+e_\mu}^{3}\delta_{y,x+e_\mu},\qquad D_\mu=\tau^3-i\tau^\mu.

Physical states satisfy the local Z2\mathbb{Z}_2-Gauss law,

Gx∣Ψ⟩=∣Ψ⟩,Gx=(−1)NxAx,(3)G_x|\Psi\rangle=|\Psi\rangle,\qquad G_x=(-1)^{N_x}A_x, \tag*{(3)}

where Nx=∑αnx,αN_x=\sum_\alpha n_{x,\alpha} and Ax=∏⟨x,y⟩∋xσ^xy1A_x=\prod_{\langle x,y\rangle\ni x}\hat{\sigma}_{xy}^{1}. In the absence of an electric-field term, the magnetic fluxes Bp=∏⟨x,y⟩∈∂pσ^xy3B_p=\prod_{\langle x,y\rangle\in\partial p}\hat{\sigma}_{xy}^{3} are conserved. A plaquette with Bp=+1B_p=+1 carries zero flux, whereas Bp=−1B_p=-1 corresponds to a magnetic vortex, or vison. We can therefore study the Hamiltonians Hλ(σ)H_\lambda(\sigma) in fixed bond backgrounds σxy=±1\sigma_{xy}=\pm1. Gauge-equivalent backgrounds yield unitarily equivalent Hamiltonians: their spectra depend only on the plaquette fluxes and the two noncontractible Z2\mathbb{Z}_2-holonomies on the torus, see [27, 28]. The physical spectrum in each gauge class is obtained by restricting Hλ(σ)H_\lambda(\sigma) to even fermion parity [29]. Finally, the torus identity ∏pBp=1\prod_p B_p=1 implies that magnetic vortices occur in even numbers.

Our analysis relies on a generalization of the reflection-positivity strategy of Lieb’s flux-phase theorem [20] and its Z2\mathbb{Z}_2 gauge-theory implementations [28] [14]. A main difference with respect to these works is that, for the QWZ model, the minimum of the half-filled matter ground-state energy over bond backgrounds is attained in a flat configuration Bp=+1B_p=+1, with antiperiodic boundary conditions in both directions. We first establish the fermionic gap and topology of the flat sectors.

under weak interactions. Chessboard estimates then reduce the comparison with nonflat backgrounds to a periodic-cell calculation, yielding a positive energy cost for the vison excitations. Together, these estimates isolate the four physical ground states, and show that charge and magnetic excitations are massive, see Figure 1.

Schematic energy spectrum in the gapped phase

Figure 1. Schematic energy spectrum in the gapped phase. The four low-energy states ∣Ωa,b⟩|\Omega_{a,b}\rangle, labelled by the holonomies a,b=±1a,b=\pm1, have energy spread ϵL=O(L−∞)\epsilon_L=O(L^{-\infty}). They are separated from the rest of the spectrum by a gap gphysg_{\mathrm{phys}}.

Gap and topology of the magnetic vacuum

The four flat gauge classes are labeled by two holonomies a,b=±1a,b=\pm1, corresponding to periodic or antiperiodic boundary conditions for the fermions. Allowed momenta k=(k1,k2)k=(k_1,k_2) in ΛL∗(a,b)\Lambda_L^*{}_{(a,b)} satisfy eikL1=ae^{ikL_1}=a and eikL2=be^{ikL_2}=b. At λ=0\lambda=0, the Bloch Hamiltonian reads

h(k)=v⃗(k)⋅τ⃗,(4)h(k)=\vec{v}(k)\cdot\vec{\tau}, \tag*{(4)}

where

v⃗(k)=(−tsin⁡k1,−tsin⁡k2,m−t(cos⁡k1+cos⁡k2)).(5)\vec{v}(k)=(-t\sin k_1,-t\sin k_2,m-t(\cos k_1+\cos k_2)). \tag*{(5)}

Since h(k)2=∣v⃗(k)∣21h(k)^2=|\vec{v}(k)|^2\mathbb{1}, the eigenvalues are ±∣v⃗(k)∣\pm|\vec{v}(k)|,

E(k)2=t2(sin⁡2k1+sin⁡2k2)+[m−t(cos⁡k1+cos⁡k2)]2(6)E(k)^2=t^2(\sin^2 k_1+\sin^2 k_2)+[m-t(\cos k_1+\cos k_2)]^2 \tag*{(6)}

The gap between the energy bands is

M∞(0)=min⁡k∈T2E(k)=min⁡{∣m−2t∣,∣m∣,∣m+2t∣}.(7)M_\infty(0)=\min_{k\in\mathbb{T}^2}E(k)=\min\{|m-2t|,|m|,|m+2t|\}. \tag*{(7)}

For M∞(0)>0M_\infty(0)>0, the free ground state fills the lower band and has N=L2N=L^2 particles. Since the Gauss projection restricts the matter states to even fermion parity, the gap within the physical flat sector is bounded below by 2M∞(0)2M_\infty(0) and converges to this value in the thermodynamic limit.

In a gapped phase, the one-body Berry curvature is given by

Ω(k)=12v^(k)⋅(∂1v^(k)×∂2v^(k)),v^=v⃗/∣v⃗∣,(8)\Omega(k)=\frac{1}{2}\hat{v}(k)\cdot(\partial_1\hat{v}(k)\times\partial_2\hat{v}(k)),\qquad\hat{v}=\vec{v}/|\vec{v}|, \tag*{(8)}

and the Chern number of the occupied band is:

κ=12π∫T2Ω(k) d2k=12[sgn⁡(m−2t)−2sgn⁡m+sgn⁡(m+2t)],(9)\kappa= \frac{1}{2\pi}\int_{\mathbb{T}^{2}}\Omega(k)\,d^{2}k = \frac{1}{2}\left[\operatorname{sgn}(m-2t)-2\operatorname{sgn}m+\operatorname{sgn}(m+2t)\right], \tag*{(9)}

which yields the free phase diagram of Figure 2.

Free phase diagram in the zero-flux sector

Figure 2. Free phase diagram in the zero-flux sector. The three boundary rays are gapless. Weak interactions preserve the Chern number on compact subsets of the open phases; the interacting transition curves are not determined here.

In order to describe flux insertion, we extend the matter Hamiltonian to continuous boundary twists φ∈T2\varphi\in\mathbb{T}^{2}. These insert flux through the noncontractible cycles while leaving the plaquette fluxes unchanged. In the free model, these fluxes can be introduced via a shift of the momentum grid by φj/L\varphi_j/L, preserving the lower bound M∞(0)M_{\infty}(0) for the Fock-space gap.

By [17], the gap remains open for sufficiently small ∣λ∣|\lambda|. In fact, the gap is bounded below by a twist-independent, volume independent constant M∞(λ)≥M∞(0)−c∣λ∣1/3>0M_{\infty}(\lambda) \ge M_{\infty}(0)-c|\lambda|^{1/3}>0. At periodic or antiperiodic boundary conditions, particle-number conservation, uniqueness of the ground state, and particle–hole symmetry imply N=L2N=L^{2} [30]. Since the ground state remains unique and gapped across the boundary-condition torus, its particle number cannot change as the twists vary. Consequently, the ground state is half-filled for every φ∈T2\varphi\in\mathbb{T}^{2}. For L∈4NL \in4\mathbb{N}, all these states have even fermion parity and are therefore admissible as physical states.

The uniform gap allows to define a fermionic ground-state line bundle over the torus of boundary twists. For the rank-one matter ground-state projector P(φ)P(\varphi) in a reference flat sector, the Berry curvature

Fλ(φ)=iTr⁡(P(φ)[∂1P(φ),∂2P(φ)]),(10)F_{\lambda}(\varphi)=i\operatorname{Tr}\left(P(\varphi)[\partial_{1}P(\varphi),\partial_{2}P(\varphi)]\right), \tag*{(10)}

yields the many-body Chern number [25, 26]

Cλ=12π∫T2Fλ(φ) d2φ,(11)C_{\lambda}=\frac{1}{2\pi}\int_{\mathbb{T}^{2}}F_{\lambda}(\varphi)\,d^{2}\varphi, \tag*{(11)}

where the trace is over the fermionic Fock space. Changing the flat-sector holonomies translates the twists by π\pi in the corresponding directions, leaving the integral over the full twist torus unchanged. At λ=0\lambda=0, the half-filled ground state is a Slater determinant, with:

F0(φ)=1L2∑n1,n2=0L−1Ω(2πn+φL).(12)F_{0}(\varphi)=\frac{1}{L^{2}}\sum_{n_{1},n_{2}=0}^{L-1}\Omega\left(\frac{2\pi n+\varphi}{L}\right). \tag*{(12)}

Integrating (12) over the twist torus one gets C0=κ∈ZC_{0}=\kappa\in\mathbb{Z}. As long as the gap stays open, the projector is smooth in the twists and λ\lambda, and hence its first Chern number is a constant along this interpolation, namely Cλ=κC_{\lambda}=\kappa. The uniform gap and locality further yield constant curvature, see [18] and [31, 32]:

sup⁡φ∣Fλ(φ)−κ2π∣=O(L−∞).(13)\sup_{\varphi}\left|F_{\lambda}(\varphi)-\frac{\kappa}{2\pi}\right|=O(L^{-\infty}). \tag*{(13)}

The matter ground states ∣ψa,b⟩|\psi_{a,b}\rangle in a flat gauge configuration σ\sigma determine the physical ground states ∣Ωa,b⟩|\Omega_{a,b}\rangle:

∣Ωa,b⟩=2(L2−1)/2PG(∣ψa,b⟩⊗∣σ⟩),(14)|\Omega_{a,b}\rangle=2^{(L^{2}-1)/2}P_{G}(|\psi_{a,b}\rangle\otimes|\sigma\rangle), \tag*{(14)}

where PG=∏x(1+Gx)/2P_{G}=\prod_{x}(1+G_{x})/2 is the Gauss law projection. It preserves the Berry connection of matter states, being a twist-independent isometry.

The four states ∣ψa,b⟩|\psi_{a,b}\rangle have exponentially close energies. For ∣λ∣|\lambda| small, locality and the uniform many-body gap give [33] [14], see also [34]:

∣Eref,λ(φ)−Eref,λ(φ′)∣≤O(L−∞),(φ,φ′∈T2),(15)\left|E_{\mathrm{ref},\lambda}(\varphi)-E_{\mathrm{ref},\lambda}(\varphi')\right|\le O(L^{-\infty}),\qquad(\varphi,\varphi'\in\mathbb{T}^{2}), \tag*{(15)}

where Eref,λ(φ)E_{\mathrm{ref},\lambda}(\varphi) is the matter ground-state energy with boundary twist φ\varphi upon a reference flat background. In particular, this shows that the four inequivalent ground states are spread by ϵL=O(L−∞)\epsilon_{L}=O(L^{-\infty}).

Reflection positivity and the chessboard bound. To prove that the ground state vectors lie in the flat sector and to control the energy cost of magnetic defects, we shall rely on reflection positivity for the matter Hamiltonians HΛ(σ)H_{\Lambda}(\sigma), via an extension of the approach of [20, 28] [14]. The matter partition function is ZΛ(σ)=Tr⁡e−βHΛ(σ)Z_{\Lambda}(\sigma)=\operatorname{Tr}e^{-\beta H_{\Lambda}(\sigma)} and the ground state energy is −lim⁡β→∞β−1log⁡ZΛ(σ)-\lim_{\beta\to\infty}\beta^{-1}\log Z_{\Lambda}(\sigma).

We cut the physical torus in two halves Λ±L\Lambda_{\pm}^{\mathrm{L}} mapped onto each other by a reflection ϑ\vartheta across a plane, see Figure 3. Let A±\mathcal{A}_{\pm} be the CAR algebras generated by the fermionic fields supported on Λ±L\Lambda_{\pm}^{\mathrm{L}}, respectively. The reflection Θ\Theta is a parity-preserving, antilinear ∗*-automorphism of A\mathcal{A} such that Θ2=id⁡\Theta^{2}=\operatorname{id} and Θ(A±)=A∓\Theta(\mathcal{A}_{\pm})=\mathcal{A}_{\mp}. For homogeneous A,B∈A+A,B\in\mathcal{A}_{+}, define Θ(A)∘B=ip(A)p(B)Θ(A)B\Theta(A)\circ B=i^{p(A)p(B)}\Theta(A)B, where p(A)=0,1p(A)=0,1 is the fermion parity of AA, extended by linearity. This twisted product accounts for the graded commutation of operators on opposite halves.

Reflection geometry on the torus, its periodic lattice representation, and the chessboard flux configuration

Figure 3. Reflection geometry on the torus (a), its periodic lattice representation (b), and the chessboard flux configuration (c). The reflection exchanges ΛL+\Lambda_L^{+} and ΛL−\Lambda_L^{-}. Shaded plaquettes carry a π\pi-flux.

Suppose that the Hamiltonian HH has the form:

H=H++Θ(H−)−∑jΘ(Cj)∘Dj(16)H=H_{+}+\Theta(H_{-})-\sum_{j}\Theta(C_{j})\circ D_{j} \tag*{(16)}

where H±∈A+H_{\pm}\in\mathcal{A}_{+} are even and self-adjoint and Cj,Dj∈A+C_{j},D_{j}\in\mathcal{A}_{+} are odd. Then with H(−)=H−+Θ(H−)−H^{(-)}=H_{-}+\Theta(H_{-})- ∑jΘ(Cj)∘Cj\sum_j \Theta(C_j)\circ C_j and H(+)=H++Θ(H+)−∑jΘ(Dj)∘DjH^{(+)}=H_{+}+\Theta(H_{+})-\sum_j\Theta(D_j)\circ D_j, reflection positivity means that for every β>0\beta>0,

(Tr⁡e−βH)2≤(Tr⁡e−βH(−))(Tr⁡e−βH(+)).(17)\left(\operatorname{Tr}e^{-\beta H}\right)^2\leq\left(\operatorname{Tr}e^{-\beta H^{(-)}}\right)\left(\operatorname{Tr}e^{-\beta H^{(+)}}\right). \tag*{(17)}

For the QWZ Hamiltonian, a bond reflection ϑν\vartheta_\nu perpendicular to an edge eνe_\nu (ν=1,2\nu=1,2) is implemented by

Θν(cx,α+)=∑β(Γν)αβcϑν(x),β−.(18)\Theta_\nu(c_{x,\alpha}^{+})=\sum_\beta(\Gamma^\nu)_{\alpha\beta}c_{\vartheta_\nu(x),\beta}^{-}. \tag*{(18)}

Observe that, compared to [20, 28] [14], the geometric reflection and particle-hole transformation have to be combined with a transformation acting on the internal degrees of freedom. Then, for an arbitrary bond background, we choose a gauge in which the bonds normal to the left boundary of ΛL+\Lambda_L^{+} have sign −1-1, whereas those normal to its right boundary have sign +1+1. The boundary-coupling matrices are 1±Sν\mathbb{1}\pm S_\nu, with Sν=−iτνS_\nu=-i\tau^\nu; their eigenvalues are 0 and 2. Diagonalization gives the boundary decomposition required for reflection positivity [34]. The on-site Hubbard term contributes only to the half-Hamiltonians and preserves this structure for every real λ\lambda. In this setting, (17) becomes

Zλ(σ)2≤Zλ(σ−)Zλ(σ+)(19)Z_\lambda(\sigma)^2\leq Z_\lambda(\sigma^{-})Z_\lambda(\sigma^{+}) \tag*{(19)}

where σ±\sigma^\pm are obtained by retaining and reflecting either half-background. In each reflected configuration, the normal bond signs occur in identical pairs, except at the seams where they remain unchanged. The holonomy perpendicular to the seams after reflection is therefore equal to −1-1: reflections in both directions yield antiperiodic boundary conditions. Moreover, Bp=+1B_p=+1 for each plaquette cut by the seam. This replaces the π\pi-flux seams of [20, 28] [14] with zero-flux seams.

Repeated reflections disseminate a chosen defect while inserting rows and columns of zero flux resulting in the chessboard pattern [35] of Figure 3. This observation and repeated applications of (19) yield a lower bound for the ground state energy Eλ(σ)E_\lambda(\sigma) of Hλ(σ)H_\lambda(\sigma) for any background σ\sigma:

Eλ(σ)≥Eref,λ+nv(σ)d∗,λ,(20)E_\lambda(\sigma)\geq E_{\mathrm{ref},\lambda}+n_v(\sigma)d_{*,\lambda}, \tag*{(20)}
d∗,λ=E∗,λ−Eref,λL2.(21)d_{*,\lambda}=\frac{E_{*,\lambda}-E_{\mathrm{ref},\lambda}}{L^2}. \tag*{(21)}

Here, nv(σ)n_v(\sigma) is the number of vortices in the configuration σ\sigma, Eref,λE_{\mathrm{ref},\lambda} and E∗,λE_{*,\lambda} are the ground state energies for the flat pattern and the chessboard pattern, with antiperiodic boundary conditions.

At λ=0\lambda=0, the periodicity of the chessboard configuration, with a 4×24\times2 unit cell, allows to compute explicitly d∗,0d_{*,0} for large LL [34]. At m=−tm=-t, the numerical value is a0t=0.066875ta_0t=0.066875t. Since the norm of the interaction is bounded above by ∣λ∣L2/4|\lambda|L^2/4, the interaction changes each ground-state energy density by at most ∣λ∣/4|\lambda|/4. Together with the dependence on mm, this gives

lim inf⁡L→∞d∗,λ≥a0t−2∣m+t∣−∣λ∣2.(22)\liminf_{L\to\infty}d_{*,\lambda}\geq a_0t-2|m+t|-\frac{|\lambda|}{2}. \tag*{(22)}

Hence flux defects have a positive energy cost in a neighbourhood of m=−t,λ=0m=-t,\lambda=0.

Quasi-adiabatic flow and the physical ground space. We now pick a flat background and denote Hλ(φ)H_\lambda(\varphi) the Hamiltonian in that background with boundary twists φ∈T2\varphi\in\mathbb{T}^2. For j=1,2j=1,2, the quasi-adiabatic generators Kj(φ)K_j(\varphi) [34] are such that ∂jP=i[Kj,P]\partial_jP=i[K_j,P] and PKjP=0PK_jP=0 [23, 24]. Integrating V˙γ(s)=iγ˙(s)⋅K(γ(s))Vγ(s)\dot V_\gamma(s)=i\dot\gamma(s)\cdot K(\gamma(s))V_\gamma(s) along a path γ\gamma in flux space with Vγ(0)=1V_\gamma(0)=\mathbb{1} yields a unitary quasi-adiabatic flow which realizes the Berry parallel transport

Vγ(s)P(0)=P(γ(s))Vγ(s).(23)V_\gamma(s)P(0)=P(\gamma(s))V_\gamma(s). \tag*{(23)}

It preserves the Gauss law, particle number, and the magnetic background.

The twist derivative ∂jHλ\partial_jH_\lambda is supported along the boundary. A Lieb–Robinson bound and the rapidly decay of the function fgf_g make KjK_j quasi-local near that cut. This locality depends crucially on the uniform gap in the flat sector, but not on the magnetic vortex gap.

The quasi-locality of KjK_j has two important consequences [34]. Firstly, it provides the ground state energy splitting estimates (15), as the following argument shows. Move the twist to a parallel cut at distance L/2L/2 from the current ∂jHλ\partial_jH_\lambda. Changing the twist then changes that current by O(L−∞)O(L^{-\infty}). Periodicity of the ground state energy forces its mean derivative to vanish, hence the energy variation itself is O(L−∞)O(L^{-\infty}). Secondly, the possibility of moving both cuts away from a local observable yields local indistinguishability. Consequently the four Gauss-projected flat ground states ∣Ωa,b⟩|\Omega_{a,b}\rangle have energy spread ϵL=O(L−∞)\epsilon_L=O(L^{-\infty}) and they are isolated by a spectral gap (taking into account the even number of vortices)

gphys≥min⁡{M∞(λ),2d∗,λ}−ϵL.(24)g_{\mathrm{phys}}\geq\min\{\mathcal{M}_\infty(\lambda),2d_{*,\lambda}\}-\epsilon_L. \tag*{(24)}

Flux operators and transport. A closed dual string X^j=∏b∩Cj∗σ^bx\hat{X}_j = \prod_{b\cap C_j^*}\hat{\sigma}_b^x flips one holonomy, without creating plaquette flux. Let us consider the operator VγjV_{\gamma_j} that implements the parallel transport (23) along the path γj(s)=πsej\gamma_j(s) = \pi s e_j, and define the operators:

Wj=X^jVγj.(25)W_j = \hat{X}_j V_{\gamma_j}. \tag*{(25)}

They connect the four different ground states of the gauge theory: W1∣Ωa,b⟩=c1(a,b)∣Ω−a,b⟩W_1|\Omega_{a,b}\rangle= c_1(a,b)|\Omega_{-a,b}\rangle and W2∣Ωa,b⟩=c2(a,b)∣Ωa,−b⟩W_2|\Omega_{a,b}\rangle= c_2(a,b)|\Omega_{a,-b}\rangle, where ∣cj∣=1|c_j| = 1, see Figure 4.

Two closed cuts on the lattice and sector changes

Figure 4. (a) Two closed cuts on the lattice. The colored bonds support the bare gauge flips X^j\hat{X}_j; VαjV_{\alpha_j} dresses them in a quasi-local ribbon. (b) Action of the dressed loops on the four physical ground states. The two routes from Ω+,+\Omega_{+,+} to Ω−,−\Omega_{-,-} differ by the braiding phase.

Consider the ground-state line bundle over the twist torus Tϕ2\mathbb{T}^2_\phi (Figure 5). A vector can be chosen smoothly on a square, but for C≠0C \ne0 there is no global smooth periodic ground-state vector. The curvature measures this obstruction. Gauge covariance gives [34]

Ground-state line bundle over the boundary-condition torus

Figure 5. Ground-state line bundle over the boundary-condition torus. One-dimensional complex fibres are depicted as real planes in local charts. The blue loop γ=∂Γ\gamma= \partial\Gamma has a horizontal lift γ~\widetilde{\gamma} under quasi-adiabatic transport, shown in brown. The transported vector returns to the initial fibre as eiθψ0e^{i\theta}\psi_0, with θ=πC/2+O(L−∞)\theta= \pi C/2 + O(L^{-\infty}).

W2∗W1∗W2W1=V∂Γ,Γ=[0,π]2.(26)W_2^*W_1^*W_2W_1 = V_{\partial\Gamma}, \qquad\Gamma= [0,\pi]^2. \tag*{(26)}

Products act from right to left: the path visits (0,0)(0,0), (π,0)(\pi,0), (π,π)(\pi,\pi), (0,π)(0,\pi), and returns to (0,0)(0,0). On a ground state, its holonomy is exp⁡(i∫ΓFλ dϕ)\exp\left(i\int_\Gamma F_\lambda\, d\phi\right). The asymptotically flat curvature (13) yields

W2∗W1∗W2W1Π=eiπ2CΠ+O(L−∞),(27)W_2^*W_1^*W_2W_1\Pi= e^{i\frac{\pi}{2}C}\Pi+ O(L^{-\infty}), \tag*{(27)}

where Π\Pi is the projector on the ground state sector.

Use of AI Tools. ChatGPT 5.6, 6 and Claude Opus 5.5 were used as research assistants. ChatGPT pointed out the QWZ model as a nearest-neighbour lattice model with a nontrivial topological phase and it performed the numerical calculation of the magnetic gap. ChatGPT and Claude have been used to produce the source code of the figures, for proofreading and for a literature search. The proof strategy has been developed by the authors, without the use of AI tools. The authors take full responsibility for the scientific content and conclusions of this work.

Acknowledgements. SB is supported by NSERC of Canada under Grant RGPIN-2024-03927. The work of MP has been carried out under the auspices of GNFM of INdAM. [29] Let PG=∏x(1+Gx)/2P_G=\prod_x(1+G_x)/2 and N=∑xNxN=\sum_xN_x. For a normalized bond state ∣σ⟩|\sigma\rangle and any fermionic state ∣ψ⟩|\psi\rangle, connectedness of the lattice gives

∥PG(∣ψ⟩⊗∣σ⟩)∥2=2−∣ΛL∣⟨ψ∣[1+(−1)N]∣ψ⟩.\left\|P_G(|\psi\rangle\otimes|\sigma\rangle)\right\|^2=2^{-|{\Lambda_L}|}\langle\psi|[1+(-1)^N]|\psi\rangle.

Thus odd-parity states are annihilated, while for even ∣ψ⟩|\psi\rangle the squared norm is 21−∣ΛL∣⟨ψ∣ψ⟩2^{1-|{\Lambda_L}|}\langle\psi|\psi\rangle. The normalized map ∣ψ⟩↦2(∣ΛL∣−1)/2PG(∣ψ⟩⊗∣σ⟩)|\psi\rangle\mapsto2^{(|\Lambda_L|-1)/2}P_G(|\psi\rangle\otimes|\sigma\rangle) is a unitary identification with the physical sector of the gauge orbit (for details see Proposition 2.7 and Remark 2.8 of [28]).

[30] The unitary particle–hole transformation Uph\mathcal{U}_{\mathrm{ph}} acts as

Uphcx,α±Uph−1=∑β(τ1)αβcx,β∓,\mathcal{U}_{\mathrm{ph}}c_{x,\alpha}^{\pm}\mathcal{U}_{\mathrm{ph}}^{-1}=\sum_{\beta}(\tau^1)_{\alpha\beta}c_{x,\beta}^{\mp},

and maps NN to 2L2−N2L^2-N. For real bond backgrounds it leaves Hλ(σ)H_\lambda(\sigma) invariant, including the centered Hubbard interaction. For continuous twists it relates Hλ(φ)H_\lambda(\varphi) to Hλ(−φ)H_\lambda(-\varphi).

[31] M. B. Hastings and S. Michalakis, Quantization of Hall conductance for interacting electrons without averaging assumptions, Communications in Mathematical Physics 334, 433 (2015), arXiv:1306.1258 [math-ph].

[32] S. Bachmann, A. Bols, W. De Roeck, and M. Fraas, Quantization of conductance in gapped interacting systems, Annales Henri Poincaré 19, 695 (2018), arXiv:1707.06491.

[33] H. Watanabe, Insensitivity of bulk properties to the twisted boundary condition, Physical Review B 98, 155137 (2018).

[34] See Supplemental Material for the details of the proof of the reflection-positivity bound, and for the Bloch diagonalization of the Hamiltonian in the chessboard background. There, we also give a review of the chessboard strategy of the quasi-adiabatic construction of the flux operators.

[35] For L∈4NL\in4\mathbb{N}, this pattern is preserved by reflections across its zero-flux seams; thus, it carries antiperiodic boundary conditions in both directions.

Fermionic reflection positivity: algebra and general estimates

Here we shall extend the reflection-positivity strategy of Lieb’s flux-phase theorem [20] to a setting that applies to the QWZ model. A main difference with respect to [20, 28] [14] is the definition of twisted product, Definition S1, which allows us to simplify the proof of the reflection bound in Theorem S4.

Fix a bond reflection ϑν\vartheta_\nu of the finite lattice torus ΛL\Lambda_L, with ν∈{1,2}\nu\in\{1,2\}, whose fixed hypersurface lies between lattice sites. On the torus, this hypersurface has two connected components. Let

ΛL=ΛL+⊔ΛL−,ϑν(ΛL±)=ΛL∓.\Lambda_L = \Lambda_L^+ \sqcup\Lambda_L^-, \qquad\vartheta_\nu(\Lambda_L^\pm) = \Lambda_L^\mp.

Let A\mathcal{A} be the full CAR algebra, and let A±\mathcal{A}_\pm be its subalgebras supported on ΛL±\Lambda_L^\pm. The associated Fock spaces are denoted by F\mathcal{F} and F±\mathcal{F}_\pm, respectively.

The reflection Θν:A→A\Theta_\nu:\mathcal{A}\to\mathcal{A} is a parity-preserving, antilinear ∗*-automorphism satisfying

Θν2=id⁡,Θν(A±)=A∓.\Theta_\nu^2 = \operatorname{id}, \qquad\Theta_\nu(\mathcal{A}_\pm) = \mathcal{A}_\mp.

We suppress the subscript ν\nu when no confusion can arise. An operator is homogeneous if it has a definite fermion parity p(A)∈{0,1}p(A) \in\{0,1\}. Operators supported on opposite halves satisfy

X−Y+=(−1)p(X−)p(Y+)Y+X−X_-Y_+ = (-1)^{p(X_-)p(Y_+)}Y_+X_-

whenever X−X_- and Y+Y_+ are homogeneous.

Definition S1 (Twisted product). For homogeneous X−∈A−X_- \in\mathcal{A}_- and Y+∈A+Y_+ \in\mathcal{A}_+, define

X−∘Y+≔ip(X−)p(Y+)X−Y+,X_- \mathbin{\circ} Y_+ \coloneqq i^{p(X_-)p(Y_+)}X_-Y_+,

and extend this operation bilinearly to A−×A+\mathcal{A}_- \times\mathcal{A}_+. Consequently,

Θ(A)∘B=ip(A)p(B)Θ(A)B\Theta(A) \mathbin{\circ} B = i^{p(A)p(B)}\Theta(A)B

for homogeneous A,B∈A+A,B \in\mathcal{A}_+. The map (A,B)↦Θ(A)∘B(A,B)\mapsto\Theta(A)\mathbin{\circ}B is antilinear in AA and linear in BB.

Lemma S2. For every homogeneous A∈A+A \in\mathcal{A}_+,

Tr⁡F(Θ(A)∘A)=∣Tr⁡F+(A)∣2≥0.\operatorname{Tr}_{\mathcal{F}}(\Theta(A)\mathbin{\circ}A)=\left|\operatorname{Tr}_{\mathcal{F}_+}(A)\right|^2\geq0.

Proof. The graded tensor product decomposition of the finite CAR algebra gives

Tr⁡F(X−Y+)=Tr⁡F−(X−)Tr⁡F+(Y+).\operatorname{Tr}_{\mathcal{F}}(X_-Y_+)=\operatorname{Tr}_{\mathcal{F}_-}(X_-)\operatorname{Tr}_{\mathcal{F}_+}(Y_+).

For even operators this is ordinary trace factorization. If either factor is odd, both sides vanish. Since Θ\Theta is an antilinear ∗*-isomorphism between the two half-algebras,

Tr⁡F−(Θ(A))=Tr⁡F+(A)‾.\operatorname{Tr}_{\mathcal{F}_-}(\Theta(A))=\overline{\operatorname{Tr}_{\mathcal{F}_+}(A)}.

Thus

Tr⁡F(Θ(A)∘A)=ip(A)∣Tr⁡F+(A)∣2.\operatorname{Tr}_{\mathcal{F}}(\Theta(A)\mathbin{\circ}A)=i^{p(A)}\left|\operatorname{Tr}_{\mathcal{F}_+}(A)\right|^2.

If AA is even, the phase equals one. If AA is odd, its trace vanishes: with P+≔(−1)N+P_+ \coloneqq(-1)^{N_+},

Tr⁡F+(A)=Tr⁡F+(P+AP+)=−Tr⁡F+(A).\operatorname{Tr}_{\mathcal{F}_+}(A)=\operatorname{Tr}_{\mathcal{F}_+}(P_+AP_+)=-\operatorname{Tr}_{\mathcal{F}_+}(A).

This proves the claim in both cases.

Lemma S3. Let A,B,C,D∈A+A, B, C, D \in\mathcal{A}_{+} be homogeneous, with p(A)=p(B)p(A)=p(B) and p(C)=p(D)p(C)=p(D). Then

(Θ(A)∘B)(Θ(C)∘D)=Θ(AC)∘BD.(\Theta(A)\circ B)(\Theta(C)\circ D)=\Theta(AC)\circ BD.

Proof. Set a=p(A)=p(B)a=p(A)=p(B) and c=p(C)=p(D)c=p(C)=p(D). By the fermionic anticommutation relations:

(Θ(A)∘B)(Θ(C)∘D)=ia+c(−1)acΘ(AC)BD.(\Theta(A)\circ B)(\Theta(C)\circ D)=i^{a+c}(-1)^{ac}\Theta(AC)BD.

This concludes the proof since p(AC)=p(BD)=a+c−2acp(AC)=p(BD)=a+c-2ac and p(AC)2=p(AC)p(AC)^2=p(AC).

Theorem S4 (Reflection-positive Hamiltonians). Let HH be self-adjoint and suppose that

−H=−H+−Θ(H+)+∑αΘ(Cα)∘Cα(28)-H=-H_{+}-\Theta(H_{+})+\sum_{\alpha}\Theta(C_{\alpha})\circ C_{\alpha} \tag*{(28)}

where H+=H+∗∈A+H_{+}=H_{+}^{*}\in\mathcal{A}_{+} is even and the finite family Cα∈A+C_{\alpha}\in\mathcal{A}_{+} consists of odd operators. Then, for every A∈A+A\in\mathcal{A}_{+},

Tr⁡F(e−H(Θ(A)∘A))≥0.(29)\operatorname{Tr}_{\mathcal{F}}\left(e^{-H}(\Theta(A)\circ A)\right)\geq0. \tag*{(29)}

Remark S5. The rewriting in Eq. (28) should be compared with [20, 28] [14], where the circle product is replaced by the usual product (and the set of CαC_{\alpha} is defined such that the Hamiltonian is Hermitian). Our representation of the Hamiltonian can be viewed as a different gauge choice with respect to [20, 28] [14], which simplifies the proof of the reflection bound (S2), thanks to Lemma S3.

Proof. We denote Q(A)=Θ(A)∘AQ(A)=\Theta(A)\circ A for any A∈A+A\in\mathcal{A}_{+} and note that Lemma S3 reads Q(A)Q(B)=Q(AB)Q(A)Q(B)=Q(AB). Set H0=H++Θ(H+)H_{0}=H_{+}+\Theta(H_{+}) and V=∑αQ(Cα)V=\sum_{\alpha}Q(C_{\alpha}), so that −H=−H0+V-H=-H_{0}+V. Letting Δn={(t1,…,tn):0≤t1≤⋯≤tn≤1}\Delta_{n}=\{(t_{1},\ldots,t_{n}):0\leq t_{1}\leq\cdots\leq t_{n}\leq1\}, the norm-convergent Duhamel expansion is

e−H=∑n≥0∑α1,…,αn∫Δndnt e−t1H0Q(Cα1)e−(t2−t1)H0⋯Q(Cαn)e−(1−tn)H0.e^{-H}=\sum_{n\geq0}\sum_{\alpha_{1},\ldots,\alpha_{n}}\int_{\Delta_{n}}d^{n}t\,e^{-t_{1}H_{0}}Q(C_{\alpha_{1}})e^{-(t_{2}-t_{1})H_{0}}\cdots Q(C_{\alpha_{n}})e^{-(1-t_{n})H_{0}}.

The n=0n=0 term is understood to be e−H0e^{-H_{0}}. The two even half-Hamiltonians commute, and hence

e−sH0=Θ(e−sH+)∘e−sH+=Q(e−sH+).e^{-sH_{0}}=\Theta(e^{-sH_{+}})\circ e^{-sH_{+}}=Q(e^{-sH_{+}}).

Define

Bα,t=e−t1H+Cα1e−(t2−t1)H+⋯Cαne−(1−tn)H+,B_{\boldsymbol{\alpha},t}=e^{-t_{1}H_{+}}C_{\alpha_{1}}e^{-(t_{2}-t_{1})H_{+}}\cdots C_{\alpha_{n}}e^{-(1-t_{n})H_{+}},

with B=e−H+B=e^{-H_{+}} when n=0n=0. Repeated application of Lemma S3 gives

e−H=∑n≥0∑α∫Δndnt Q(Bα,t).e^{-H}=\sum_{n\geq0}\sum_{\alpha}\int_{\Delta_{n}}d^{n}t\,Q(B_{\alpha,t}).

For homogeneous AA, Lemmas S3 and S2 imply

Tr⁡F(e−HQ(A))=∑n≥0∑α∫Δndnt ∣Tr⁡F+(Bα,tA)∣2≥0.\operatorname{Tr}_{\mathcal{F}}(e^{-H}Q(A))=\sum_{n\geq0}\sum_{\alpha}\int_{\Delta_{n}}d^{n}t\,\left|\operatorname{Tr}_{\mathcal{F}_{+}}(B_{\alpha,t}A)\right|^{2}\geq0.

For arbitrary A=A0+A1A=A_{0}+A_{1}, with p(Aj)=jp(A_{j})=j, the two mixed terms in Q(A)Q(A) are odd. Since e−He^{-H} is even, their traces vanish, and

Tr⁡F(e−HQ(A))=Tr⁡F(e−HQ(A0))+Tr⁡F(e−HQ(A1))≥0,\operatorname{Tr}_{\mathcal{F}}(e^{-H}Q(A))=\operatorname{Tr}_{\mathcal{F}}(e^{-H}Q(A_{0}))+\operatorname{Tr}_{\mathcal{F}}(e^{-H}Q(A_{1}))\geq0,

concluding the proof.

Remark S6 (Self-adjointness and inverse temperature). For odd CC,

(Θ(C)∘C)∗=Θ(C∗)∘C∗.(\Theta(C)\circ C)^{*}=\Theta(C^{*})\circ C^{*}.

Thus closure of the family {Cα}\{C_{\alpha}\} under adjoints, with equal weights for each adjoint pair, is sufficient to make the interaction self-adjoint. This is different from [20], where the family {Cα}\{C_{\alpha}\} only includes annihilation operators.

Theorem S7 (Reflection bound). Let H+,H−∈A+H_{+},H_{-}\in\mathcal{A}_{+} be even and self-adjoint. Suppose that

−H=−H+−Θ(H−)+∑αΘ(Cα)∘Dα(30)-H=-H_{+}-\Theta(H_{-})+\sum_{\alpha}\Theta(C_{\alpha})\circ D_{\alpha} \tag*{(30)}

where Cα,Dα∈A+C_{\alpha},D_{\alpha}\in\mathcal{A}_{+} are odd. Define

−H(−)=−H−−Θ(H−)+∑αΘ(Cα)∘Cα,-H^{(-)}=-H_{-}-\Theta(H_{-})+\sum_{\alpha}\Theta(C_{\alpha})\circ C_{\alpha},
−H(+)=−H+−Θ(H+)+∑αΘ(Dα)∘Dα,-H^{(+)}=-H_{+}-\Theta(H_{+})+\sum_{\alpha}\Theta(D_{\alpha})\circ D_{\alpha},

and Z=Tr⁡F(e−H)Z=\operatorname{Tr}_{\mathcal{F}}(e^{-H}), Z(±)=Tr⁡F(e−H(±))Z^{(\pm)}=\operatorname{Tr}_{\mathcal{F}}(e^{-H^{(\pm)}}). Then

∣Z∣2≤Z(−)Z(+).\lvert Z\rvert^{2}\leq Z^{(-)}Z^{(+)}.

Proof. For each (n,α,t)(n,\alpha,\mathbf{t}), define

B−=e−t1H−Cα1e−(t2−t1)H−⋯Cαne−(1−tn)H−,B_{-}=e^{-t_{1}H_{-}}C_{\alpha_{1}}e^{-(t_{2}-t_{1})H_{-}}\cdots C_{\alpha_{n}}e^{-(1-t_{n})H_{-}},
B+=e−t1H+Dα1e−(t2−t1)H+⋯Dαne−(1−tn)H+.B_{+}=e^{-t_{1}H_{+}}D_{\alpha_{1}}e^{-(t_{2}-t_{1})H_{+}}\cdots D_{\alpha_{n}}e^{-(1-t_{n})H_{+}}.

For n=0n=0, set B±=e−H±B_{\pm}=e^{-H_{\pm}}. Duhamel expansion and Lemmas S3 and S2 yield

Z=∑n≥0∑α∫Δndnt Tr⁡F+(B−)‾Tr⁡F+(B+)Z=\sum_{n\geq0}\sum_{\alpha}\int_{\Delta_{n}}d^{n}t\ \overline{\operatorname{Tr}_{\mathcal{F}_{+}}(B_{-})}\operatorname{Tr}_{\mathcal{F}_{+}}(B_{+})

as in the proof of Theorem S4. Here odd B±B_{\pm} give zero, while even B±B_{\pm} allow ordinary trace factorization. By Cauchy–Schwarz inequality

∣Z∣2≤(∑n≥0∑α∫Δndnt ∣Tr⁡F+(B−)∣2)(∑n≥0∑α∫Δndnt ∣Tr⁡F+(B+)∣2).(31)\lvert Z\rvert^{2}\leq\left(\sum_{n\geq0}\sum_{\alpha}\int_{\Delta_{n}}d^{n}t\ \lvert\operatorname{Tr}_{\mathcal{F}_{+}}(B_{-})\rvert^{2}\right)\left(\sum_{n\geq0}\sum_{\alpha}\int_{\Delta_{n}}d^{n}t\ \lvert\operatorname{Tr}_{\mathcal{F}_{+}}(B_{+})\rvert^{2}\right). \tag*{(31)}

The two factors are the Duhamel expansions of Z(−)Z^{(-)} and Z(+)Z^{(+)}.

Application to the QWZ Hamiltonian

Here we will show how to recast the Hamiltonian of the QWZ model as in Eq. (28). As discussed below, this will be possible after a suitable gauge fixing, different from the one of [20, 28] [14], and a unitary transformation.

Let τ1,τ2,τ3\tau^{1},\tau^{2},\tau^{3} be the standard Pauli matrices, and let cx,α+=(cx,α−)∗c_{x,\alpha}^{+}=(c_{x,\alpha}^{-})^{*}, with α∈{1,2}\alpha\in\{1,2\}. The QWZ Hamiltonian without interactions is given in explicit form by

H0(σ)=m∑x,α,βcx,α+(τ3)αβcx,β−−t2∑x,μ,α,βσx,x+eμ[cx,α+(Dμ)αβcx+eμ,β−+cx+eμ,α+(Dμ∗)αβcx,β−],H_{0}(\sigma)=m\sum_{x,\alpha,\beta}c_{x,\alpha}^{+}(\tau^{3})_{\alpha\beta}c_{x,\beta}^{-}-\frac{t}{2}\sum_{x,\mu,\alpha,\beta}\sigma_{x,x+e_{\mu}}\left[c_{x,\alpha}^{+}(D_{\mu})_{\alpha\beta}c_{x+e_{\mu},\beta}^{-}+c_{x+e_{\mu},\alpha}^{+}(D_{\mu}^{*})_{\alpha\beta}c_{x,\beta}^{-}\right],

where Dμ=τ3−iτμD_{\mu}=\tau^{3}-i\tau^{\mu}, the sums run over x,y∈ΛLx,y\in\Lambda_{L} and α,β,μ∈{1,2}\alpha,\beta,\mu\in\{1,2\}, m∈Rm\in\mathbb{R}, t≥0t\geq0, and σxy=σyx∈{±1}\sigma_{xy}=\sigma_{yx}\in\{\pm1\} is a fixed background. We also recall the definition (18) of the antilinear reflection automorphism

Θν(cx,α+)=∑βταβνcϑν(x),β−,Θν(cx,α−)=∑βcϑν(x),β+τβαν.(32)\Theta_{\nu}(c_{x,\alpha}^{+})=\sum_{\beta}\tau_{\alpha\beta}^{\nu}c_{\vartheta_{\nu}(x),\beta}^{-},\qquad\Theta_{\nu}(c_{x,\alpha}^{-})=\sum_{\beta}c_{\vartheta_{\nu}(x),\beta}^{+}\tau_{\beta\alpha}^{\nu}. \tag*{(32)}

extended antilinearly and multiplicatively to the whole algebra. Hermiticity and unitarity of τν\tau^{\nu} ensure that this extension preserves adjoints and the CAR. Moreover,

Θν2(cx,α+)=∑β,γταβν‾τβγνcx,γ+=cx,α+,\Theta_{\nu}^{2}(c_{x,\alpha}^{+})=\sum_{\beta,\gamma}\overline{\tau_{\alpha\beta}^{\nu}}\tau_{\beta\gamma}^{\nu}c_{x,\gamma}^{+}=c_{x,\alpha}^{+},

and similarly Θν2(cx,α−)=cx,α−\Theta_{\nu}^{2}(c_{x,\alpha}^{-})=c_{x,\alpha}^{-}. Thus Θν\Theta_{\nu} is an involutive, parity-preserving antilinear ∗*-automorphism exchanging A+\mathcal{A}_{+} and A−\mathcal{A}_{-}.

Proposition S8. Fix ν∈{1,2}\nu\in\{1,2\}. Assume that the background field is reflection invariant on unoriented bonds:

σϑν(x),ϑν(y)=σxy.\sigma_{\vartheta_\nu(x),\vartheta_\nu(y)} = \sigma_{xy}.

Assume that σ=1\sigma= 1 on all edges across the right boundary of ΛL+\Lambda_L^+, while σ=−1\sigma= -1 across its left boundary, see Figure S1. Then H0(σ)H_0(\sigma) has a decomposition as required by Theorem S4 with reflection Θν\Theta_\nu.

Proof. Write H0(σ)=H++H−phys+H∂H_0(\sigma) = H_+ + H_-^{\mathrm{phys}} + H_{\partial}, where

H+=m∑x∈ΛL+α,βcx,α+(τ3)αβcx,β−−t2∑x∈ΛL+x+eμ∈ΛL+∑μ=1,2∑α,βσx,x+eμ[cx,α+(Dμ)αβcx+eμ,β−+cx+eμ,α+(Dμ∗)αβcx,β−],H_+ = m \sum_{\substack{x\in\Lambda_L^+\\ \alpha,\beta}} c_{x,\alpha}^{+}(\tau^3)_{\alpha\beta}c_{x,\beta}^{-} - \frac{t}{2}\sum_{\substack{x\in\Lambda_L^+\\ x+e_\mu\in\Lambda_L^+}}\sum_{\mu=1,2}\sum_{\alpha,\beta}\sigma_{x,x+e_\mu}\left[c_{x,\alpha}^{+}(D_\mu)_{\alpha\beta}c_{x+e_\mu,\beta}^{-}+c_{x+e_\mu,\alpha}^{+}(D_\mu^{*})_{\alpha\beta}c_{x,\beta}^{-}\right],

and H−physH_-^{\mathrm{phys}} is the analogous restriction to ΛL−\Lambda_L^-. The notation H−physH_-^{\mathrm{phys}} distinguishes this actual negative-half operator from H−H_- used in Theorem S7. For any 2×22 \times2 matrix DD, a direct CAR calculation gives

Θν(∑α,βcx,α+Dαβcy,β−)=δxytr⁡D‾−∑α,βcϑν(y),α+(τνD∗τν)αβcϑν(x),β−.(33)\Theta_\nu\left(\sum_{\alpha,\beta}c_{x,\alpha}^{+}D_{\alpha\beta}c_{y,\beta}^{-}\right)=\delta_{xy}\operatorname{tr}\overline{D}-\sum_{\alpha,\beta}c_{\vartheta_\nu(y),\alpha}^{+}(\tau^\nu D^{*}\tau^\nu)_{\alpha\beta}c_{\vartheta_\nu(x),\beta}^{-}. \tag*{(33)}

Here D∗D^{*} denotes the conjugate transpose. With D=τ3D=\tau^3, the trace term vanishes and −τντ3τν=τ3-\tau^\nu\tau^3\tau^\nu=\tau^3, so the mass term is mapped to the mass term on the other half. For the hopping terms, set sμν=(−1)δμνs_{\mu\nu}=(-1)^{\delta_{\mu\nu}}. Then

ϑν(x+eμ)=ϑν(x)+sμνeμ,−τν(τ3+iτμ)τν=τ3+isμντμ.\vartheta_\nu(x+e_\mu)=\vartheta_\nu(x)+s_{\mu\nu}e_\mu,\qquad-\tau^\nu(\tau^3+i\tau^\mu)\tau^\nu=\tau^3+is_{\mu\nu}\tau^\mu.

Consequently, for y=x+eμy=x+e_\mu,

Θν(∑α,βcx,α+(Dμ)αβcy,β−)=∑α,βcϑν(y),α+(τ3+isμντμ)αβcϑν(x),β−.\Theta_\nu\left(\sum_{\alpha,\beta}c_{x,\alpha}^{+}(D_\mu)_{\alpha\beta}c_{y,\beta}^{-}\right)=\sum_{\alpha,\beta}c_{\vartheta_\nu(y),\alpha}^{+}(\tau^3+is_{\mu\nu}\tau^\mu)_{\alpha\beta}c_{\vartheta_\nu(x),\beta}^{-}.

For μ=ν\mu=\nu, this is a forward hopping term on the reflected bond; for μ≠ν\mu\ne\nu, it is the adjoint hopping term. Reflection invariance of σ\sigma therefore gives

Θν(H+)=H−phys.\Theta_\nu(H_+)=H_-^{\mathrm{phys}}.

It remains to cast the boundary terms in the required form. Set

Rν={z∈ΛL+:z+eν∈ΛL−},Lν={z∈ΛL+:z−eν∈ΛL−}.\mathcal{R}_\nu=\{z\in\Lambda_L^+:z+e_\nu\in\Lambda_L^-\},\qquad\mathcal{L}_\nu=\{z\in\Lambda_L^+:z-e_\nu\in\Lambda_L^-\}.

For these boundary sites, ϑν(z)=z+eν\vartheta_\nu(z)=z+e_\nu on Rν\mathcal{R}_\nu and ϑν(z)=z−eν\vartheta_\nu(z)=z-e_\nu on Lν\mathcal{L}_\nu. The inverse generator relations are

cϑν(z),α−=∑βταβνΘν(cz,β+),cϑν(z),α+=∑βτβανΘν(cz,β−).c_{\vartheta_\nu(z),\alpha}^{-}=\sum_\beta\tau^\nu_{\alpha\beta}\Theta_\nu(c_{z,\beta}^{+}),\qquad c_{\vartheta_\nu(z),\alpha}^{+}=\sum_\beta\tau^\nu_{\beta\alpha}\Theta_\nu(c_{z,\beta}^{-}).

Figure S1. Reflection on the torus. The reflection plane is shaded in brown. In the lattice diagram, thin bonds carry σ=1\sigma= 1 and thick bonds carry σ=−1\sigma= -1. The thick bonds form twisting lines realizing nontrivial Z2\mathbb{Z}_2-holonomies, or equivalently antiperiodic boundary conditions, in the two coordinate directions. For the vertical reflection, the shaded strip represents ΛL+\Lambda_L^+ and the vertical twisting line lies along its left boundary. A twisting line can be moved by a unitary gauge transformation.

Let

Sν=−iτν3=∑r=13ϵν3rτr,AνR=1+Sν,AνL=1−Sν,Bνb=(Aνb)T(b=R,L).S_\nu=-i\tau_\nu^3=\sum_{r=1}^{3}\epsilon_{\nu3r}\tau^r,\qquad A_\nu^R=\mathbb{1}+S_\nu,\qquad A_\nu^L=\mathbb{1}-S_\nu,\qquad B_\nu^b=(A_\nu^b)^T\quad(b=R,L).

Both SνS_\nu and AνbA_\nu^b are Hermitian; hence Bνb=Aνb‾B_\nu^b=\overline{A_\nu^b}. For z∈Rνz\in\mathcal{R}_\nu, the contribution to −H∂-H_\partial is

t2∑α,β[cz,α+(Dν)αβcϑν(z),β−+cϑν(z),α+(Dν∗)αβcz,β−]=t2∑α,β[(AνR)αβΘν(cz,α−)∘cz,β−+(BνR)αβΘν(cz,α+)∘cz,β+].\begin{aligned} \frac{t}{2}\sum_{\alpha,\beta}\left[c_{z,\alpha}^+(D_\nu)_{\alpha\beta}c_{\vartheta_\nu(z),\beta}^-+c_{\vartheta_\nu(z),\alpha}^+(D_\nu^*)_{\alpha\beta}c_{z,\beta}^-\right] \\ &=\frac{t}{2}\sum_{\alpha,\beta}\left[(A_\nu^R)_{\alpha\beta}\Theta_\nu(c_{z,\alpha}^-)\circ c_{z,\beta}^-+(B_\nu^R)_{\alpha\beta}\Theta_\nu(c_{z,\alpha}^+)\circ c_{z,\beta}^+\right]. \end{aligned}

For z∈Lνz\in\mathcal{L}_\nu, the oriented bond starts at ϑν(z)\vartheta_\nu(z) and carries σ=−1\sigma=-1. Its contribution is

−t2∑α,β[cϑν(z),α+(Dν)αβcz,β−+cz,α+(Dν∗)αβcϑν(z),β−]=t2∑α,β[(AνL)αβΘν(cz,α−)∘cz,β−+(BνL)αβΘν(cz,α+)∘cz,β+].\begin{aligned} -\frac{t}{2}\sum_{\alpha,\beta}\left[c_{\vartheta_\nu(z),\alpha}^+(D_\nu)_{\alpha\beta}c_{z,\beta}^-+c_{z,\alpha}^+(D_\nu^*)_{\alpha\beta}c_{\vartheta_\nu(z),\beta}^-\right] \\ &=\frac{t}{2}\sum_{\alpha,\beta}\left[(A_\nu^L)_{\alpha\beta}\Theta_\nu(c_{z,\alpha}^-)\circ c_{z,\beta}^-+(B_\nu^L)_{\alpha\beta}\Theta_\nu(c_{z,\alpha}^+)\circ c_{z,\beta}^+\right]. \end{aligned}

To check the phases explicitly, the annihilation matrices are

AνR=−iτr(τ3+iτr)=1+Sν,AνL=iτr(τ3−iτr)=1−Sν,A_\nu^R=-i\tau^r(\tau^3+i\tau^r)=\mathbb{1}+S_\nu,\qquad A_\nu^L=i\tau^r(\tau^3-i\tau^r)=\mathbb{1}-S_\nu,

and the corresponding creation matrices are their transposes. In particular,

A1R=(1i−i1),A1L=(1−ii1),A2R=(1111),A2L=(1−1−11).A_1^R=\begin{pmatrix}1&i\\-i&1\end{pmatrix},\qquad A_1^L=\begin{pmatrix}1&-i\\i&1\end{pmatrix},\qquad A_2^R=\begin{pmatrix}1&1\\1&1\end{pmatrix},\qquad A_2^L=\begin{pmatrix}1&-1\\-1&1\end{pmatrix}.

Since Sν2=1S_\nu^2=\mathbb{1} and tr⁡Sν=0\operatorname{tr}S_\nu=0, all four AA matrices and their transposes are positive semidefinite, with spectrum {0,2}\{0,2\}. For each boundary b∈{R,L}b\in\{R,L\} choose a unitary UνbU_\nu^b satisfying

UνbAνb(Uνb)†=diag⁡(0,2),U_\nu^b A_\nu^b(U_\nu^b)^\dagger=\operatorname{diag}(0,2),

and define one consistent CAR basis by

dz,a−,b=∑α(Uνb)aαcz,α−,dz,a+,b=∑α(Uνb)aα‾cz,α+=(dz,a−,b)∗.d_{z,a}^{-,b}=\sum_\alpha(U_\nu^b)_{a\alpha}c_{z,\alpha}^-,\qquad d_{z,a}^{+,b}=\sum_\alpha\overline{(U_\nu^b)_{a\alpha}}c_{z,\alpha}^+=(d_{z,a}^{-,b})^*.

Unitarity preserves the CAR. Antilinearity of Θν\Theta_\nu gives

∑α,β(Aνb)αβΘν(cz,α−)∘cz,β−=∑aλaΘν(dz,a−,b)∘dz,a−,b,\sum_{\alpha,\beta}(A_\nu^b)_{\alpha\beta}\Theta_\nu(c_{z,\alpha}^-)\circ c_{z,\beta}^-=\sum_a\lambda_a\Theta_\nu(d_{z,a}^{-,b})\circ d_{z,a}^{-,b},
∑α,β(Bνb)αβΘν(cz,α+)∘cz,β+=∑aλaΘν(dz,a+,b)∘dz,a+,b,\sum_{\alpha,\beta}(B_\nu^b)_{\alpha\beta}\Theta_\nu(c_{z,\alpha}^+)\circ c_{z,\beta}^+=\sum_a\lambda_a\Theta_\nu(d_{z,a}^{+,b})\circ d_{z,a}^{+,b},

where (λ1,λ2)=(0,2)(\lambda_1,\lambda_2)=(0,2). For a=1,2a=1,2, now define

Cz,a±,b=tλa2 dz,a±,b.C_{z,a}^{\pm,b}=\sqrt{\frac{t\lambda_a}{2}}\,d_{z,a}^{\pm,b}.

With BR=Rν\mathcal{B}_R=\mathcal{R}_\nu and BL=Lν\mathcal{B}_L=\mathcal{L}_\nu, the full boundary interaction becomes

−H∂=∑b=R,L∑z∈Bb∑a=12[Θν(Cz,a−,b)∘Cz,a−,b+Θν(Cz,a+,b)∘Cz,a+,b].-H_\partial=\sum_{b=R,L}\sum_{z\in\mathcal{B}_b}\sum_{a=1}^{2}\left[\Theta_\nu(C_{z,a}^{-,b})\circ C_{z,a}^{-,b}+\Theta_\nu(C_{z,a}^{+,b})\circ C_{z,a}^{+,b}\right].

Thus the interaction is in the required reflection-positive form, and Theorem S4 applies.

For an arbitrary bond background σ\sigma, let ϵx∈{±1}\epsilon_x\in\{\pm1\}, supported on Rν∪Lν\mathcal{R}_\nu\cup\mathcal{L}_\nu, be such that the gauge-transformed field σxyϵ=ϵxσxyϵy\sigma^\epsilon_{xy}=\epsilon_x\sigma_{xy}\epsilon_y equals +1+1 on the bonds leaving Rν\mathcal{R}_\nu and −1-1 on the bonds leaving Lν\mathcal{L}_\nu. Then, writing again H+H_+ and H−physH_-^{\mathrm{phys}} for the restrictions of H0(σϵ)H_0(\sigma^\epsilon) to ΛL±\Lambda_L^\pm and H−=Θν(H−phys)∈A+H_-=\Theta_\nu(H_-^{\mathrm{phys}})\in\mathcal{A}_+, the boundary computation in the proof of Proposition S8 applies unchanged and gives a representation of the Hamiltonian as in Eq. (S3),

−H0(σϵ)=−H+−Θν(H−)+∑jΘν(Cj)∘Cj(34)-H_0(\sigma^\epsilon)=-H_+-\Theta_\nu(H_-)+\sum_j\Theta_\nu(C_j)\circ C_j \tag*{(34)}

with the σ\sigma-independent operators CjC_j of Proposition S8. Thus, Theorem S7 with Dj=CjD_j=C_j and gauge invariance of the trace yield Zλ(σ)2≤Zλ(σ−)Zλ(σ+)Z_\lambda(\sigma)^2 \le Z_\lambda(\sigma^-)Z_\lambda(\sigma^+), where σ±\sigma^\pm is the reflection-invariant background that agrees with σε\sigma^\varepsilon on ΛL±\Lambda^\pm_L.

Representatives of the four flat gauge classes on a torus

Figure S2. Representatives of the four flat gauge classes on a torus with L=4L=4. Every plaquette has Bp=+1B_p=+1. Thin and thick bonds denote σxy=+1\sigma_{xy}=+1 and σxy=−1\sigma_{xy}=-1, respectively. The labels (a,b)(a,b) give the holonomies in the horizontal and vertical directions.

Local Hubbard interaction and reflection positivity

The many-body interaction is

λV=λ∑x(nx,1−12)(nx,2−12)=λ2∑x(Nx−1)2−λ4L2.(35)\lambda V=\lambda\sum_x\left(n_{x,1}-\frac{1}{2}\right)\left(n_{x,2}-\frac{1}{2}\right)=\frac{\lambda}{2}\sum_x(N_x-1)^2-\frac{\lambda}{4}L^2. \tag*{(35)}

It is number conserving and invariant under an on-site change of orbital basis. Since τν\tau^\nu is off diagonal for ν=1,2\nu=1,2, the CAR and (S5) gives

Θν(nx,1)=1−nϑν(x),2,Θν(nx,2)=1−nϑν(x),1.(36)\Theta_\nu(n_{x,1})=1-n_{\vartheta_\nu(x),2},\qquad\Theta_\nu(n_{x,2})=1-n_{\vartheta_\nu(x),1}. \tag*{(36)}

Consequently Θν((nx,1−12)(nx,2−12))=(nϑν(x),1−12)(nϑν(x),2−12)\Theta_\nu\left((n_{x,1}-\frac{1}{2})(n_{x,2}-\frac{1}{2})\right)=\left(n_{\vartheta_\nu(x),1}-\frac{1}{2}\right)\left(n_{\vartheta_\nu(x),2}-\frac{1}{2}\right), since the two density operators commute. Write V+=∑x∈Λ+(nx,1−12)(nx,2−12)V_+=\sum_{x\in\Lambda_+}(n_{x,1}-\frac{1}{2})(n_{x,2}-\frac{1}{2}), and let H+(λ)=H++λV+H_+(\lambda)=H_++\lambda V_+. Adding λV+\lambda V_+ to the positive-half Hamiltonian in Proposition S8 leaves its boundary couplings unchanged and yields

−Hλ=−H+(λ)−Θν(H+(λ))+∑jΘν(Cj)∘Cj,(37)-H_\lambda=-H_+(\lambda)-\Theta_\nu(H_+(\lambda))+\sum_j\Theta_\nu(C_j)\circ C_j, \tag*{(37)}

which is again in RP form.

Reflection positivity and chessboard estimates

Let Zβ,L(Φ;a,b;λ)Z_{\beta,L}(\Phi;a,b;\lambda) be the partition function of the fermionic Hamiltonian, where Φ=(Bx,y)\Phi=(B_{x,y}) specifies the fluxes across all plaquettes and a,ba,b the two holonomies. Define

F(Φ)=−log⁡max⁡a,b=±1Zβ,L(Φ;a,b;λ).(38)\mathcal{F}(\Phi)=-\log\max_{a,b=\pm1}Z_{\beta,L}(\Phi;a,b;\lambda). \tag*{(38)}

On admissible configurations (∏pBp=1)(\prod_p B_p=1) this value is finite, and we extend F\mathcal{F} by +∞+\infty on inadmissible configurations.

Minimizing magnetic configuration

Without loss of generality, we assume that the spins associated to bonds intersecting the right boundary of ΛL+\Lambda^+_L are equal to +1+1, while spins associated to bonds intersecting the left boundary of ΛL+\Lambda^+_L are equal to −1-1. This configuration can be realized after a suitable gauge transformation, that does not change the partition function. Next, we apply Theorem S7 to these two halves.

Let us focus on the plaquettes at a seam. There, after the reflection, the two normal bonds have identical signs, while the two tangential bonds are reflected copies. Therefore

Bseam=(±1)2σtangential2=+1.(39)B_{\mathrm{seam}}=(\pm1)^2\sigma_{\mathrm{tangential}}^2=+1. \tag*{(39)}

This is different from [20, 28] [14], where the plaquettes at the seam have π\pi flux rather than 00 flux. Next, denote columns of plaquettes by Xj=(Bj,y)y=0L−1X^j = (B_{j,y})_{y=0}^{L-1} and let 1=(1,…,1)\mathbf{1} = (1,\ldots,1). By Theorem S7:

F(X1,…,XL)≥12{F(X1,…,XL/2−1,1,XL/2−1,…,X1,1)+F(XL,…,XL/2+1,1,XL/2+1,…,XL,1)}.(40)\mathcal{F}(X^1,\ldots,X^L) \geq\frac{1}{2}\{\mathcal{F}(X^1,\ldots,X^{L/2-1},\mathbf{1},X^{L/2-1},\ldots,X^1,\mathbf{1}) + \mathcal{F}(X^L,\ldots,X^{L/2+1},\mathbf{1},X^{L/2+1},\ldots,X^L,\mathbf{1})\}. \tag*{(40)}

After a reflection, the holonomies of the new configurations might differ from the original ones; taking the maximum over a,ba,b makes the inequality independent of these choices. Cyclic translation of columns or rows leaves F\mathcal{F} unchanged. Repeated application of this inequality implies that the 00-flux configuration maximizes the partition function. Also, applying both reflections to a flat configuration fixes both normal holonomies to −1-1; therefore,

max⁡a,bZβ,L(1;a,b;λ)=Zβ,L(1;−1,−1;λ)=Zref.(41)\max_{a,b} Z_{\beta,L}(\mathbf{1};a,b;\lambda) = Z_{\beta,L}(\mathbf{1};-1,-1;\lambda) = Z_{\mathrm{ref}}. \tag*{(41)}

Chessboard estimate

The chessboard lemma of [28], Sec. 3.3, Lemma 3.10, applies to (S12), with the constant seam column −1-1 of [28] replaced by +1+1. We have, following [28]:

F(Φ)≥1L2∑0≤j,y<LF(D(Bj,y)).(42)\mathcal{F}(\Phi) \geq\frac{1}{L^2}\sum_{0\leq j,y<L}\mathcal{F}\bigl(\mathcal{D}(B_{j,y})\bigr). \tag*{(42)}

Here D(q)\mathcal{D}(q) places qq at every even-even plaquette and +1+1 at every other plaquette. For q=+1q=+1 it is the flat pattern; for q=−1q=-1 it is a chessboard pattern, as in Figure 3. Every original plaquette occurs once in the sum; hence:

F(Φ)≥L2−nvL2F(1)+nvL2F(Φ∗).(43)\mathcal{F}(\Phi) \geq\frac{L^2-n_{\mathrm{v}}}{L^2}\mathcal{F}(\mathbf{1})+\frac{n_{\mathrm{v}}}{L^2}\mathcal{F}(\Phi^*). \tag*{(43)}

Dividing by β\beta and taking the limit β→∞\beta\to\infty, Eq. (20) follows, with d∗,λ=lim⁡β→∞d∗,β,λd_{*,\lambda}=\lim_{\beta\to\infty}d_{*,\beta,\lambda} and

d∗,β,λ=−1βL2log⁡max⁡a,bZβ,L(Φ∗;a,b;λ)Zref.(44)d_{*,\beta,\lambda}=-\frac{1}{\beta L^2}\log\frac{\max_{a,b}Z_{\beta,L}(\Phi^*;a,b;\lambda)}{Z_{\mathrm{ref}}}. \tag*{(44)}

The fermionic Hamiltonian in a chessboard flux pattern

Here we shall determine the Hamiltonian of the fermionic model associated with the chessboard flux configuration Φ∗\Phi^*, Figure 3. Let us introduce a periodic 4×24\times2 bond cell, with sites a,b,c,da,b,c,d on y=0y=0 and A,B,C,DA,B,C,D on y=1y=1, in increasing order of x=0,1,2,3x=0,1,2,3. See Figure S3. All horizontal signs are positive; vertically set

Bond gauge for the disseminated defect pattern and the sites of its magnetic cell

Figure S3. Bond gauge for the disseminated defect pattern and the sites of its magnetic cell. The boundary is periodically identified after translations by 4e14\mathbf{e}_1 and 2e22\mathbf{e}_2. Uppercase and lowercase letters label sites, not independent physical species.

σ2(x,0)=sx=(1,−1,−1,1)x,σ2(x,1)=1,(45)\sigma_2(x,0)=s_x=(1,-1,-1,1)_x,\qquad\sigma_2(x,1)=1, \tag*{(45)}

yields the fluxes depicted in Figure S3. Put Tμ=−tDμ/2T_\mu=-tD_\mu/2 and Q=mτ3Q=m\tau^3. Explicitly,

D1=(1−i−i−1),D2=(1−11−1),Tμ2=0,∥Tμ∥=t.(46)D_1=\begin{pmatrix}1&-i\\-i&-1\end{pmatrix},\qquad D_2=\begin{pmatrix}1&-1\\1&-1\end{pmatrix},\qquad T_\mu^2=0,\quad\lVert T_\mu\rVert=t. \tag*{(46)}

For cell positions R∈4Ze1+2Ze2R \in4\mathbb{Z}e_1+2\mathbb{Z}e_2, the free Hamiltonian is, with every bilinear summed over α,β\alpha,\beta,

H∗(0)=∑R,x,y,α,βcR+(x,y),α†QαβcR+(x,y),β+∑R,y,α,β[∑x=02cR+(x,y),α†(T1)αβcR+(x+1,y),β+cR+(3,y),α†(T1)αβcR+4e1+(0,y),β+h.c.]+∑R,x,α,β[sxcR+(x,0),α†(T2)αβcR+(x,1),β+cR+(x,1),α†(T2)αβcR+2e2+(x,0),β+h.c.].(47)\begin{aligned} H_\ast(0)={}&\sum_{R,x,y,\alpha,\beta}c^\dagger_{R+(x,y),\alpha}Q_{\alpha\beta}c_{R+(x,y),\beta}\\ &+\sum_{R,y,\alpha,\beta}\left[\sum_{x=0}^{2}c^\dagger_{R+(x,y),\alpha}(T_1)_{\alpha\beta}c_{R+(x+1,y),\beta}+c^\dagger_{R+(3,y),\alpha}(T_1)_{\alpha\beta}c_{R+4e_1+(0,y),\beta}+\mathrm{h.c.}\right]\\ &+\sum_{R,x,\alpha,\beta}\left[s_xc^\dagger_{R+(x,0),\alpha}(T_2)_{\alpha\beta}c_{R+(x,1),\beta}+c^\dagger_{R+(x,1),\alpha}(T_2)_{\alpha\beta}c_{R+2e_2+(x,0),\beta}+\mathrm{h.c.}\right]. \tag*{(47)} \end{aligned}

Here 0≤x<40 \le x < 4, 0≤y<20 \le y < 2 whenever these indices are summed. The interacting Hamiltonian is H∗(λ)=H∗(0)+λVH_\ast(\lambda)=H_\ast(0)+\lambda V, with VV the Hubbard interaction.

Let zj=eiκjz_j=e^{i\kappa_j}, κj∈[0,2π)\kappa_j\in[0,2\pi) denote cell translation phases and set Bx=sxT2+z~2−1T2∗B_x=s_xT_2+\tilde z_2^{-1}T_2^\ast. In the basis (a,b,c,d,A,B,C,D)(a,b,c,d,A,B,C,D), the requested Bloch matrix is

h∗(κ)=(QT10z~1−1T1∗B0000T1∗QT100B1000T1∗QT100B20z1T10T1∗Q000B3B0∗000QT10z~1−1T1∗0B1∗00T1∗QT1000B2∗00T1∗QT1000B3∗z1T10T1∗Q).(48)h_\ast(\kappa)= \begin{pmatrix} Q&T_1&0&\tilde z_1^{-1}T_1^\ast&B_0&0&0&0\\ T_1^\ast&Q&T_1&0&0&B_1&0&0\\ 0&T_1^\ast&Q&T_1&0&0&B_2&0\\ z_1T_1&0&T_1^\ast&Q&0&0&0&B_3\\ B_0^\ast&0&0&0&Q&T_1&0&\tilde z_1^{-1}T_1^\ast\\ 0&B_1^\ast&0&0&T_1^\ast&Q&T_1&0\\ 0&0&B_2^\ast&0&0&T_1^\ast&Q&T_1\\ 0&0&0&B_3^\ast&z_1T_1&0&T_1^\ast&Q \end{pmatrix}. \tag*{(48)}

Every entry is a 2×22\times2 matrix; in particular (h∗)rα,r′β(h_\ast)_{r\alpha,r'\beta} retains the indices α,β\alpha,\beta. The terms z~2−1T2∗\tilde z_2^{-1}T_2^\ast are the reverse hoppings from the neighbouring cell. Two useful checks are

tr⁡h∗=0,tr⁡h∗2=16(m2+2t2).(49)\operatorname{tr}h_\ast=0,\qquad\operatorname{tr}h_\ast^2=16(m^2+2t^2). \tag*{(49)}

The second follows by summing the squared hopping norms; interference terms vanish because Tμ2=0T_\mu^2=0.

Evaluation of the fermionic trace and zero-temperature limit

Diagonalize h∗(κ)h_\ast(\kappa) as Uκdiag⁡(ε1,…,ε16)Uκ∗U_\kappa\operatorname{diag}(\varepsilon_1,\ldots,\varepsilon_{16})U_\kappa^\ast. For λ=0\lambda=0, we get:

Z∗,β,L(0)=∏κ∏a=116(1+e−βεa(κ))=∏κdet⁡(1+e−βh∗(κ))(50)Z_{\ast,\beta,L}(0)=\prod_\kappa\prod_{a=1}^{16}\left(1+e^{-\beta\varepsilon_a(\kappa)}\right)=\prod_\kappa\det\left(1+e^{-\beta h_\ast(\kappa)}\right) \tag*{(50)}
=∏κ,a2cosh⁡(βεa(κ)2).(51)=\prod_{\kappa,a}2\cosh\left(\frac{\beta\varepsilon_a(\kappa)}{2}\right). \tag*{(51)}

The exponential prefactors disappear because ∑aεa=0\sum_a\varepsilon_a=0. There are L2/8L^2/8 cell momenta. For holonomies (eiϕ1,eiϕ2)(e^{i\phi_1},e^{i\phi_2}),

κ1=2πn1+ϕ1L/4,κ2=2πn2+ϕ2L/2,0≤n1<L/4,0≤n2<L/2.(52)\kappa_1=\frac{2\pi n_1+\phi_1}{L/4},\qquad\kappa_2=\frac{2\pi n_2+\phi_2}{L/2},\qquad0\le n_1<L/4,\quad0\le n_2<L/2. \tag*{(52)}

For a,b=±1a,b=\pm1, ϕj\phi_j is 00 or π\pi. The partition function of the flat sector is:

Z+,β,L(0)=∏k4cosh⁡2(βE(k)/2).(53)Z_{+,\beta,L}(0)=\prod_k4\cosh^2(\beta E(k)/2). \tag*{(53)}

The ground state energies of the chessboard background and of the flat background are, in the infinite volume limit:

e∗(m,t)=−116∫[0,2π]2d2κ(2π)2tr⁡∣h∗(κ)∣.(54)e_\ast(m,t)=-\frac{1}{16}\int_{[0,2\pi]^2}\frac{d^2\kappa}{(2\pi)^2}\operatorname{tr}|h_\ast(\kappa)|. \tag*{(54)}
e+(m,t)=−∫[0,2π]2d2k(2π)2E(k),d∗(m,t)=e∗(m,t)−e+(m,t).(55)e_+(m,t)=-\int_{[0,2\pi]^2}\frac{d^2k}{(2\pi)^2}E(k),\qquad d_\ast(m,t)=e_\ast(m,t)-e_+(m,t). \tag*{(55)}

These expressions are the starting point for the numerical evaluation of the magnetic gap (22).

Quasi-local control of the ground states

Almost-insensitivity of the ground state energy on magnetic twists

Let EL(φ)E_L(\varphi) be the non-degenerate flat-sector fermionic ground state energy, separated from the rest of the spectrum by a uniform gap. Let IC(φ)=∂jHC(φ)I_C(\varphi)=\partial_j H_C(\varphi) be the current associated with a twist φj\varphi_j on a cut CC. The twist can be translated to a parallel cut C′C' at distance L/2L/2 by a site-dependent number rotation R=R(φj)R=R(\varphi_j):

RHC(φ)R∗=HC′(φ),RIC(φ)R∗=JC,(56)R H_C(\varphi) R^*=H_{C'}(\varphi),\qquad R I_C(\varphi)R^*=J_C, \tag*{(56)}

where JCJ_C is the same bond current with its local twist phase removed, and hence is independent of φj\varphi_j. The on-site interaction is unchanged. Hellmann–Feynman and unitary covariance imply

∂jEL(φ)=⟨JC⟩C′,φ.(57)\partial_j E_L(\varphi)=\langle J_C\rangle_{C',\varphi}. \tag*{(57)}

The quasi-adiabatic generator for varying the twist of HC′H_{C'}, see Sec. S6 A below, is localized near C′C'. In particular, for every NN:

∣∂j⟨JC⟩C′,φ∣=∣⟨[JC,KC′]⟩C′,φ∣≤CNL2(L/2)−N=O(L−∞),(58)\left|\partial_j\langle J_C\rangle_{C',\varphi}\right|=\left|\langle[J_C,K_{C'}]\rangle_{C',\varphi}\right|\le C_N L^2(L/2)^{-N}=O(L^{-\infty}), \tag*{(58)}

where the constants are uniform in φ\varphi. Since ELE_L is 2π2\pi-periodic in φj\varphi_j, ∫02π∂jEL dφj=0\int_0^{2\pi}\partial_jE_L\,d\varphi_j=0. A periodic function with vanishing average and oscillation O(L−∞)O(L^{-\infty}) has supremum O(L−∞)O(L^{-\infty}). Thus

sup⁡φ∣∂jEL(φ)∣=O(L−∞),ϵL=sup⁡φ,φ′∣EL(φ)−EL(φ′)∣=O(L−∞),(59)\sup_{\varphi}\left|\partial_jE_L(\varphi)\right|=O(L^{-\infty}),\qquad\epsilon_L=\sup_{\varphi,\varphi'}\left|E_L(\varphi)-E_L(\varphi')\right|=O(L^{-\infty}), \tag*{(59)}

as claimed in (15).

Local indistinguishability

For a bounded local physical observable AA supported in a fixed-size region, choose representatives of the flat magnetic backgrounds whose nontrivial seams are at distance of order LL from that region. They agree on the support of the observable. By the quasi-adiabatic flow, we can connect the matter ground states by continuous twists on these seams. With P(φ)P(\varphi) the ground-state projector,

∂jTr⁡(P(φ)A)=iTr⁡(P(φ)[A,Kj(φ)]),∣∂jTr⁡(PA)∣≤CA,NL−N(60)\partial_j\operatorname{Tr}(P(\varphi)A)=i\operatorname{Tr}(P(\varphi)[A,K_j(\varphi)]),\qquad\left|\partial_j\operatorname{Tr}(PA)\right|\le C_{A,N}L^{-N} \tag*{(60)}

for every NN. Matrix elements of components that change plaquette flux vanish between flat states. Furthermore, off-diagonal matrix elements between distinct holonomies vanish exactly. Consequently

ΠLAΠL=aL(A)ΠL+OA(L−∞),(61)\Pi_L A\Pi_L=a_L(A)\Pi_L+O_A(L^{-\infty}), \tag*{(61)}

where aL(A)a_L(A) is the average of AA over any of the four matter ground states.

Quasi-adiabatic flow, the ground-state bundle, and the loop algebra

The quasi-adiabatic flow

Let H(φ)=H(λ;φ)H(\varphi)=H(\lambda;\varphi), with twists on two oriented dual cuts, and let P(φ)P(\varphi) be the projector onto the gapped ground state in a flat magnetic background. Choose a real odd integrable function fgf_g with superpolynomial decay, whose Fourier transform is such that

f^g(ω)=∫Re−iωufg(u) du,f^g(ω)=−iω(∣ω∣≥gf),f^g(ω)=0(∣ω∣≤gf/2).(62)\widehat f_g(\omega)=\int_{\mathbb{R}}e^{-i\omega u}f_g(u)\,du,\qquad\widehat f_g(\omega)=-\frac{i}{\omega}\quad(|\omega|\ge g_f),\qquad\widehat f_g(\omega)=0\quad(|\omega|\le g_f/2). \tag*{(62)}

For example take f^g=−i(1−χg)/ω\widehat f_g=-i(1-\chi_g)/\omega, where χg\chi_g is an even smooth cutoff equal to 1 near zero and zero outside the gap interval. The inverse is understood as an oscillatory integral at high frequency; it is an integrable real odd function. Its bounded jump at u=0u=0 causes no difficulty, and all time-tail moments are finite. This is the spectral-flow filter of Refs. [23, 24]. Define

Kj(φ)=∫Rfg(u)eiuH(φ)∂jH(φ)e−iuH(φ) du.(63)K_j(\varphi)=\int_{\mathbb{R}}f_g(u)e^{iuH(\varphi)}\partial_jH(\varphi)e^{-iuH(\varphi)}\,du. \tag*{(63)}

This operator is the generator of the quasi-adiabatic flow. By Lieb-Robinson bounds, it follows that it is a quasi-local operator. It generates the parallel transport of the many-body matter ground state on the flux torus. We have [23, 24]:

∂jP=i[Kj,P],PKjP=0.(64)\partial_j P = i[K_j,P], \qquad P K_j P = 0. \tag*{(64)}

For a piecewise smooth path γ:[0,1]→Tϕ2\gamma: [0,1] \to\mathbb{T}^2_\phi, set Kγ(s)=γ˙j(s)Kj(γ(s))K_\gamma(s) = \dot{\gamma}_j(s)K_j(\gamma(s)) and consider the solution of

V˙γ(s)=iKγ(s)Vγ(s),Vγ(0)=I.(65)\dot{V}_\gamma(s) = iK_\gamma(s)V_\gamma(s), \qquad V_\gamma(0) = I. \tag*{(65)}

Differentiating Vγ∗P(γ(s))VγV_\gamma^*P(\gamma(s))V_\gamma with (S36) gives zero, hence

Vγ(s)P(γ(0))Vγ(s)∗=P(γ(s)).(66)V_\gamma(s)P(\gamma(0))V_\gamma(s)^* = P(\gamma(s)). \tag*{(66)}

Parallel transport in a line bundle of boundary conditions

The ground-state fibres

Eφ=Ran⁡P(φ),E=∐φ∈Tϕ2Eφ(67)\mathcal{E}_\varphi= \operatorname{Ran} P(\varphi), \qquad\mathcal{E} = \coprod_{\varphi\in\mathbb{T}^2_\phi} \mathcal{E}_\varphi \tag*{(67)}

form a complex line bundle. HH and PP are smooth and periodic in cut twists, but a normalized periodic ground vector need not exist globally. Choose local frames ψa\psi_a on contractible patches UaU_a. On overlaps ψb=eiχabψa\psi_b = e^{i\chi_{ab}}\psi_a, so

Aa=i⟨ψa∣dψa⟩,Ab=Aa−dχab,F=dAa=iTr⁡P[∂1P,∂2P]d2φ.(68)A_a = i\langle\psi_a|\mathrm{d}\psi_a\rangle, \qquad A_b = A_a - \mathrm{d}\chi_{ab}, \qquad F = \mathrm{d}A_a = i\operatorname{Tr}P[\partial_1P,\partial_2P]\mathrm{d}^2\varphi. \tag*{(68)}

The curvature is globally defined and its integral is 2πC2\pi C. A nonzero CC obstructs a global smooth periodic frame, see Figure 5. On a patch write Vγ(s)ψ(γ(0))=c(s)ψ(γ(s))V_\gamma(s)\psi(\gamma(0)) = c(s)\psi(\gamma(s)). Since PKγP=0P K_\gamma P = 0,

ddslog⁡c(s)=−⟨ψ(γ(s))∣ddsψ(γ(s))⟩.(69)\frac{\mathrm{d}}{\mathrm{d}s}\log c(s) = -\langle\psi(\gamma(s))|\frac{\mathrm{d}}{\mathrm{d}s}\psi(\gamma(s))\rangle. \tag*{(69)}

Thus the transported vector satisfies the horizontal condition ⟨Vψ(0)∣∂s(Vψ(0))⟩=0\langle V\psi(0)|\partial_s(V\psi(0))\rangle= 0. For a loop bounding a contractible square, Stokes’ theorem gives

V∂ΓP(0)=exp⁡(−∮∂Γ⟨ψ∣dψ⟩)P(0)=exp⁡(i∫ΓFLd2φ)P(0).(70)V_{\partial\Gamma}P(0) = \exp\left(-\oint_{\partial\Gamma}\langle\psi|\mathrm{d}\psi\rangle\right)P(0) = \exp\left(i\int_\Gamma F_L\mathrm{d}^2\varphi\right)P(0). \tag*{(70)}

Switching the ground states of the gauge theory

Choose oriented cuts Cj∗C_j^* and set

X^j=∏b∩Cj∗≠∅σ^bx,X^jH(φ)X^j=H(φ+πej).(71)\widehat{X}_j = \prod_{b\cap C_j^* \ne\varnothing}\widehat{\sigma}_b^x, \qquad\widehat{X}_j H(\varphi)\widehat{X}_j = H(\varphi+\pi e_j). \tag*{(71)}

Conjugation by X^j\widehat{X}_j changes the sign of each hopping across the cut, which is equivalent to increasing its external twist by π\pi. Plaquette fluxes remain unchanged, because a closed dual cut crosses every plaquette boundary an even number of times. Obviously, the two X^j\widehat{X}_j commute. The generator of the quasi-adiabatic flow (S35) satisfies:

X^jKℓ(φ)X^j=Kℓ(φ+πej),X^jVγ=Vγ+πejX^j.(72)\widehat{X}_j K_\ell(\varphi)\widehat{X}_j = K_\ell(\varphi+\pi e_j), \qquad\widehat{X}_j V_\gamma= V_{\gamma+\pi e_j}\widehat{X}_j. \tag*{(72)}

Let α1(s)=(πs,0)\alpha_1(s) = (\pi s,0) and α2(s)=(0,πs)\alpha_2(s) = (0,\pi s) and define Wj=X^jVαjW_j = \widehat{X}_jV_{\alpha_j}. These operators intertwine the four ground states of the gauge theory:

W1∣Ωa,b⟩=c1(a,b)∣Ω−a,b⟩,(73)W_1|\Omega_{a,b}\rangle= c_1(a,b)|\Omega_{-a,b}\rangle, \tag*{(73)}
W2∣Ωa,b⟩=c2(a,b)∣Ωa,−b⟩,∣cj(a,b)∣=1.(74)W_2|\Omega_{a,b}\rangle= c_2(a,b)|\Omega_{a,-b}\rangle, \qquad|c_j(a,b)| = 1. \tag*{(74)}

Nontrivial braiding of magnetic excitations

Using (S44) we obtain:

W2∗W1∗W2W1=Vα2∗Vα1+πe2∗Vα2+πe1Vα1=V∂Γ,Γ=[0,π]2.(75)W_2^*W_1^*W_2W_1=V_{\alpha_2}^*V_{\alpha_1+\pi e_2}^*V_{\alpha_2+\pi e_1}V_{\alpha_1}=V_{\partial\Gamma},\qquad\Gamma=[0,\pi]^2. \tag*{(75)}

The path is (0,0)→(π,0)→(π,π)→(0,π)→(0,0)(0,0)\to(\pi,0)\to(\pi,\pi)\to(0,\pi)\to(0,0), positively oriented for dϕ1∧dϕ2\mathrm{d}\phi_1\wedge\mathrm{d}\phi_2. In sector (a,b)(a,b) the curvature integral is on Γ+φa,b\Gamma+\varphi_{a,b}, with φa,b,j=0\varphi_{a,b,j}=0 or π\pi. This is the geometrical construction of [14], Sec. 6.1.3, this time with nonzero curvature.

Under the uniform gap assumption, the curvature FL(φ)F_L(\varphi) turns out to be almost constant on the flux torus [32], in the sense of Eq. (13). Since each translated square has area π2\pi^2,

∫Γ+φa,bFLd2φ=π2C2π+O(L−∞)=πC2+O(L−∞).(76)\int_{\Gamma+\varphi_{a,b}}F_L\mathrm{d}^2\varphi=\pi^2\frac{C}{2\pi}+O(L^{-\infty})=\frac{\pi C}{2}+O(L^{-\infty}). \tag*{(76)}

Combining this with (S42) proves

ΠLW2∗W1∗W2W1ΠL=eiπ2CΠL+O(L−∞),W2W1ΠL=eiπ2CW1W2ΠL+O(L−∞).(77)\Pi_LW_2^*W_1^*W_2W_1\Pi_L=e^{\mathrm{i}\frac{\pi}{2}C}\Pi_L+O(L^{-\infty}),\qquad W_2W_1\Pi_L=e^{\mathrm{i}\frac{\pi}{2}C}W_1W_2\Pi_L+O(L^{-\infty}). \tag*{(77)}

Finally, for a Wilson loop ZC=∏b∈Cσ~bzZ_C=\prod_{b\in C}\tilde{\sigma}_b^z, proceeding as in [14] we also have:

ZCWC∗=(−1)I(C,C∗)WC∗ZC.(78)Z_CW_{C^*}=(-1)^{I(C,C^*)}W_{C^*}Z_C. \tag*{(78)}

References

  1. [1]X.-G. Wen, Topological orders in rigid states, International Journal of Modern Physics B 4, 239 (1990).DOI
  2. [2]X.-G. Wen and Q. Niu, Ground-state degeneracy of the fractional quantum Hall states in the presence of a random potential and on high-genus Riemann surfaces, Physical Review B 41, 9377 (1990).
  3. [3]A. Kitaev, Fault-tolerant quantum computation by anyons, Annals of Physics 303, 2 (2003).DOI
  4. [4]J. M. Leinaas and J. Myrheim, On the theory of identical particles, Il Nuovo Cimento B 37, 1 (1977).DOI
  5. [5]F. Wilczek, Magnetic flux, angular momentum, and statistics, Physical Review Letters 48, 1144 (1982).DOI
  6. [6]J. Fröhlich, New super-selection sectors (“soliton-states”) in two dimensional Bose quantum field models, Communications in Mathematical Physics 47, 269 (1976).
  7. [7]A. Kitaev, Anyons in an exactly solved model and beyond, Annals of Physics 321, 2 (2006).DOI
  8. [8]M. A. Levin and X.-G. Wen, String-net condensation: A physical mechanism for topological phases, Physical Review B 71, 045110 (2005).DOI
  9. [9]J. Fröhlich, U. Studer, and E. Thiran, Quantum theory of large systems of relativistic matter, in Session LXII: Fluctuating Geometries in Statistical Mechanics and Field Theory, Proceedings of Les Houches summer school (1994).
  10. [10]X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Topological field theory of time-reversal invariant insulators, Physical Review B 78, 195424 (2008).
  11. [11]M. Cheng and Z.-C. Gu, Topological response theory of Abelian symmetry-protected topological phases in two dimensions, Physical Review Letters 112, 14602 (2014).pubmed.ncbi.nlm.nih.gov/24765942
  12. [12]J. Fröhlich and A. Zee, Large scale physics of the quantum Hall fluid, Nuclear Physics B 364, 517 (1991).DOI
  13. [13]J. Fröhlich and U. M. Studer, Gauge invariance and current algebra in nonrelativistic many-body theory, Re- [14] S. Bachmann, L. Goller, and M. Porta, Anyons in the π-flux phase of fermionic matter coupled to a Z₂-gauge field (2026), version 3, 5 June 2026, to appear in Communications in Mathematical Physics, arXiv:2508.21502 [math-ph].
  14. [15]A. Kapustin and N. Sopenko, Hall conductance and the statistics of flux insertions in gapped interacting lattice systems, Journal of Mathematical Physics 61, 101901 (2020).DOI
  15. [16]X.-L. Qi, Y.-S. Wu, and S.-C. Zhang, Topological quantization of the spin Hall effect in two-dimensional paramagnetic semiconductors, Physical Review B 74, 085308 (2006).DOI
  16. [17]W. De Roeck and M. Salmhofer, Persistence of exponential decay and spectral gaps for interacting fermions, Communications in Mathematical Physics 365, 773 (2019).DOI
  17. [18]A. Giuliani, V. Mastropietro, and M. Porta, Universality of the Hall conductivity in interacting electron systems, Communications in Mathematical Physics 349, 1107 (2017).
  18. [19]S.-S. Zhang, G. B. Halász, and C. D. Batista, Toward Kitaev’s sixteenfold way in a honeycomb lattice model, Physical Review Research 2, 023334 (2020).
  19. [20]E. H. Lieb, Flux phase of the half-filled band, Physical Review Letters 73, 2158 (1994).
  20. [21]A. Paramekanti and A. Vishwanath, Extending Luttinger’s theorem to Z₂ fractionalized phases of matter, Physical Review B 70, 245118 (2004).DOI
  21. [22]M. Oshikawa and T. Senthil, Fractionalization, topological order, and quasiparticle statistics, Physical Review Letters 96, 060601 (2006).pubmed.ncbi.nlm.nih.gov/16605977
  22. [23]M. B. Hastings and X.-G. Wen, Quasiadiabatic continuation of quantum states: The stability of topological ground-state degeneracy and emergent gauge invariance, Physical Review B 72, 045141 (2005).DOI
  23. [24]S. Bachmann, S. Michalakis, B. Nachtergaele, and R. Sims, Automorphic equivalence within gapped phases of quantum lattice systems, Communications in Mathematical Physics 309, 835 (2012).DOI
  24. [25]Q. Niu, D. J. Thouless, and Y.-S. Wu, Quantized Hall conductance as a topological invariant, Physical Review B 31, 3372 (1985).
  25. [26]J. E. Avron and R. Seiler, Quantization of the Hall conductance for general, multiparticle Schrödinger Hamiltonians, Physical Review Letters 54, 259 (1985).pubmed.ncbi.nlm.nih.gov/10031462
  26. [27]E. H. Lieb and M. Loss, Fluxes, Laplacians, and Kasteleyn’s theorem, Duke Mathematical Journal 71, 337 (1993).
  27. [28]L. Goller and M. Porta, Stability of the π-flux phase for Z₂ lattice gauge theory coupled to fermionic matter, Journal of Statistical Physics 193, 60 (2026).arxiv.org/abs/2501.10065
  28. [29]Let P_G = ∏ₓ(1 + Gₓ)/2 and N = ∑ₓ Nₓ. For a normalized bond state |σ⟩ and any fermionic state |ψ⟩, connectedness of the lattice gives |P_G(|ψ⟩ ⊗ |σ⟩)|² = 2⁻|Λᶜ|⟨ψ|[1 + (−1)ᴺ]|ψ⟩. Thus odd-parity states are annihilated, while for even |ψ⟩ the squared norm is 2⁻¹|Λᶜ|⟨ψ|ψ⟩. The normalized map |ψ⟩ ↦ 2⁽|Λᶜ|−1⁾ᐟ²P_G(|ψ⟩ ⊗ |σ⟩) is a unitary identification with the physical sector of the gauge orbit (for details see Proposition 2.7 and Remark 2.8 of [28]).
  29. [30]The unitary particle–hole transformation 𝒰_ph acts as 𝒰_ph c̃ₓ,α⁺𝒰_ph⁻¹ = ∑ᵦ (τ¹)ᵅᵦ c̃ₓ,ᵦ⁺, and maps N to 2L² − N. For real bond backgrounds it leaves H_λ(σ) invariant, including the centered Hubbard interaction. For continuous twists it relates H_λ(φ) to H_λ(−φ).
  30. [31]M. B. Hastings and S. Michalakis, Quantization of Hall conductance for interacting electrons without averaging assumptions, Communications in Mathematical Physics 334, 433 (2015), arXiv:1306.1258 [math-ph].
  31. [32]S. Bachmann, A. Bols, W. De Roeck, and M. Fraas, Quantization of conductance in gapped interacting systems, Annales Henri Poincaré 19, 695 (2018), arXiv:1707.06491.arxiv.org/abs/1707.06491
  32. [33]H. Watanabe, Insensitivity of bulk properties to the twisted boundary condition, Physical Review B 98, 155137 (2018).DOI
  33. [34]See Supplemental Material for the details of the proof of the reflection-positivity bound, and for the Bloch diagonalization of the Hamiltonian in the chessboard background. There, we also give a review of the chessboard strategy of the quasi-adiabatic construction of the flux operators.
  34. [35]For L ∈ 4ℕ, this pattern is preserved by reflections across its zero-flux seams; thus, it carries antiperiodic boundary conditions in both directions.

Paper details

vlnc:W_mrsk3fbg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 7

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents