betaFeedback

WorksMathematics

A new butterfly in the meadow: magnetic lattice graphs with a circulant vertex coupling

Abstract

We analyze the spectrum of quantum square lattice graph exposed to a homogeneous magnetic field with a rational flux through plaquette, assuming all nontrivial cases of the vertex coupling which is circulant and scale-invariant, with a particular attention to those of them that give rise to a dynamics violating the time reversal invariance. Except for two cases with partially decoupled edges, the remaining twelve split into two spectral classes; the second one offers a hitherto unknown butterfly pattern.

To Uzy Smilansky, one of quantum graph prophets, on the occasion of his 85th birthday

Introduction

The idea of quantum mechanics with a graph as configuration space was proposed by Linus Pauling ninety years ago [37] but then happily forgotten for half a century. Its second birth came in the eighties, motivated primarily, but not exclusively, by the use of such operators as models of quantum nanostructures which then started entering the scene. It took time and vision, however, to reveal the true potential of the concept, and the birthday boy was one of the main movers in the field; to limit ourselves to a small sample one can mention, e.g., his work of quantum chaos on graphs [28], trace formulæ [29], inverse problems [19], irreversible dynamics [41], nodal domains on graphs [6], and many others.

While it is generally known that the self-adjointness requirement allows to match wave functions in the graph vertices by many different ways, most results in the literature concern the simplest possibility, usually called Kirchhoff, or at most the one-parameter family of the δ\delta-couplings [11 Sec. 1.4.4]. To look into other couplings, one needs a motivation. A possible one comes from the attempt to use quantum graphs to model the anomalous Hall effect [42, 43]. In order to get the result, one had there to impose a preferred direction of motion on the graphs edges, a step that cannot be justified from the first principles. One can, however, do the same by choosing a vertex coupling which violates the time-reversal invariance.

One of the examples we treated concerned a square lattice with a time asymmetric vertex coupling exposed to a homogeneous magnetic field [7]. It is of a particular interest because the vertex coupling and the magnetic field represent two competing mechanisms to violate the invariance: what was found was that the former dominates at low energies and the latter in the high-energy asymptotics exhibiting the well-known butterfly pattern. Recall that interesting spectral properties of systems having two characteristic lengths, possibly incommensurate, was foretold already by Azbel [5] but it was only the Hofstadter illustrative result [23] which managed to ignite the interest to this problem.

The fact that the Hofstadter butterfly dominates in the model of [7] at high energies is easy to explain: the time-reversal invariance violation is there energy dependent and it vanishes asymptotically. This naturally inspires the question how such a system would behave if the coupling would be scale-invariant which is the problem we address in this paper. It is well known [27] that the most general self-adjoint vertex coupling in a quantum graph exposed to a magnetic field (and free of other interactions) is given by the matching condition

(U−I)ψ(v)+i(U+I)(Dψ)(v)=0(1)(U-I)\psi(v)+i(U+I)(\mathcal{D}\psi)(v)=0 \tag*{(1)}

for the boundary values of the wave function ψ\psi at the vertex vv, where D=ddx−iAe\mathcal{D}=\frac{\mathrm{d}}{\mathrm{d}x}-iA_e with xx being the coordinate parametrizing an edge ee rooted in the vertex and AeA_e the component of the vector potential tangent to edge ee, and UU is an n×nn \times n unitary matrix with n=deg⁡vn=\deg v. The scale invariance means the absence of the Robin component of the coupling in the terminology of [11 Thm. 1.4.4], so that the matrix UU is at the same time Hermitean. To make the problem tractable, we also assume that the coupling is circulant, i.e. invariant with respect to a cyclic permutation of the edges which in case of a planar graph makes a clear sense.

We are going to list all the couplings with these properties, altogether fourteen cases, and show that, apart from the simple situation when the lattice edges are pairwise decoupled, the spectrum for magnetic fields, rational in terms of the flux per plaquette, splits into two classes. One is the classical Hofstadter butterfly, the other is a pattern which have not yet been observed, cf. Figure 1.

Figure 1. Spectrum of a lattice with class II coupling, cf. Theorem 4.1.

Preliminaries: a class of circulant matrices

In accordance with the motivation mentioned in the introduction, we are interested in a particular class of circulant matrices which are simultaneously Hermitean and unitary, which means that such a UU must be idempotent, U2=IU^2 = I. Let us recall some of their properties.

The general case

An n×nn \times n matrix UU is called circulant if it is of the form

U=(c1c2c3⋯cncnc1c2⋯cn−1⋮cnc1⋱⋮c3⋯⋱⋱c2c2c3⋯cnc1),U = \begin{pmatrix} c_1 & c_2 & c_3 & \cdots& c_n \\ c_n & c_1 & c_2 & \cdots& c_{n-1} \\ \vdots& c_n & c_1 & \ddots& \vdots\\ c_3 & \cdots& \ddots& \ddots& c_2 \\ c_2 & c_3 & \cdots& c_n & c_1 \end{pmatrix},

being uniquely determined by its generating vector (c1,c2,c3,…,cn)T(c_1,c_2,c_3,\ldots,c_n)^{\mathsf{T}}, cj∈Cc_j \in\mathbb{C}. Such matrices are diagonalized by the discrete Fourier transform, D=FCF∗D = FCF^*, where Frs=1nω−(r−1)(s−1)F_{rs} = \frac{1}{\sqrt{n}}\omega^{-(r-1)(s-1)}, r,s=1,…,nr,s = 1,\ldots,n and ω:=e2πi/n\omega:= e^{2\pi i/n} is the nnth complex root of unity. All n×nn \times n circulant matrices have the same eigenvectors,

χm=1n{1,ωm,ω2m,…,ω(n−1)m}T,m=0,1,2,…,n−1.(2)\chi_m = \frac{1}{\sqrt{n}}\{1,\omega^m,\omega^{2m},\ldots,\omega^{(n-1)m}\}^{\mathsf{T}}, \qquad m = 0,1,2,\ldots,n-1. \tag*{(2)}

The eigenvalues of such matrices are expressed through the generating vector, λj=∑a=1ncaωj(a−1)\lambda_j = \sum_{a=1}^{n} c_a\omega^{j(a-1)}, and vice versa, cj=1n∑a=1nλaω−a(j−1)c_j = \frac{1}{n}\sum_{a=1}^{n}\lambda_a\omega^{-a(j-1)}, where the indices are understood modulo nn, i.e. λn=λ0\lambda_n = \lambda_0, etc. The condition U2=IU^2 = I implies that the eigenvalues λj\lambda_j of UU can be only ±1\pm1, and we will write them as an ordered set D:=(λ0,λ1,λ2,…,λn−1)D := (\lambda_0,\lambda_1,\lambda_2,\ldots,\lambda_{n-1}); we also introduce the sets S±={m:λm=±1}S_{\pm} = \{m : \lambda_m = \pm1\}.

The relations between the generating vector and the eigenvalues show that each circulant matrix of this class in uniquely determined by the set DD, or equivalently by S:=S+S := S_+. This means that there are 2n2^n of them of which we will exclude the trivial situations where U=∓IU = \mp I for which the edges in (1) are fully decoupled with Dirichlet and (magnetic) Neumann endpoints, respectively. What is important is that the order of DD matters; a reordering yields a different circulant matrix. For instance, a cyclic permutation of DD given by λk′=λk+r(modn)\lambda'_k = \lambda_{k+r}\pmod n is easily seen to produce matrix U′U' the entries of which are Ujk′=ωr(k−j)UjkU'_{jk} = \omega^{r(k-j)}U_{jk}.

Let us further recall that according to [17] a quantum graph with vertex coupling given by (1) is time-reversal invariant if and only if the matrix UU is transpose invariant, which is further equivalent to the invariance with respect to the complex conjugation; for circulant matrices this means cj=cn−j+2c_j = c_{n-j+2}, j=2,…,⌊n+12⌋j = 2,\ldots,\left\lfloor\frac{n+1}{2}\right\rfloor.

The n=4n = 4 case. Let us now specify the above general claims to the 4×44 \times4 matrices which are of our interest here. It follows from the above discussion that there are fourteen nontrivial possibilities looking explicitly as follows:

U{0}=12(−11111−11111−11111−1),U{1,2,3}=12(1−1−1−1−11−1−1−1−11−1−1−1−11)(3)U_{\{0\}}=\frac{1}{2}\begin{pmatrix}-1&1&1&1\\1&-1&1&1\\1&1&-1&1\\1&1&1&-1\end{pmatrix},\qquad U_{\{1,2,3\}}=\frac{1}{2}\begin{pmatrix}1&-1&-1&-1\\-1&1&-1&-1\\-1&-1&1&-1\\-1&-1&-1&1\end{pmatrix} \tag*{(3)}
U{1}=12(−1−i−1ii−1−i−1−1i−1−i−i−1i−1),U{0,2,3}=12(1i1−i−i1i11−i1ii1−i1)(4)U_{\{1\}}=\frac{1}{2}\begin{pmatrix}-1&-\mathrm{i}&-1&\mathrm{i}\\\mathrm{i}&-1&-\mathrm{i}&-1\\-1&\mathrm{i}&-1&-\mathrm{i}\\-\mathrm{i}&-1&\mathrm{i}&-1\end{pmatrix},\qquad U_{\{0,2,3\}}=\frac{1}{2}\begin{pmatrix}1&\mathrm{i}&1&-\mathrm{i}\\-\mathrm{i}&1&\mathrm{i}&1\\1&-\mathrm{i}&1&\mathrm{i}\\\mathrm{i}&1&-\mathrm{i}&1\end{pmatrix} \tag*{(4)}
U{2}=12(−1−11−1−1−1−111−1−1−1−11−1−1),U{0,1,3}=12(11−11111−1−11111−111)(5)U_{\{2\}}=\frac{1}{2}\begin{pmatrix}-1&-1&1&-1\\-1&-1&-1&1\\1&-1&-1&-1\\-1&1&-1&-1\end{pmatrix},\qquad U_{\{0,1,3\}}=\frac{1}{2}\begin{pmatrix}1&1&-1&1\\1&1&1&-1\\-1&1&1&1\\1&-1&1&1\end{pmatrix} \tag*{(5)}
U{3}=12(−1i−1−i−i−1i−11−i−1ii−1−i−1),U{0,1,2}=12(1−i1ii1−i11i1−i−i1i1)(6)U_{\{3\}}=\frac{1}{2}\begin{pmatrix}-1&\mathrm{i}&-1&-\mathrm{i}\\-\mathrm{i}&-1&\mathrm{i}&-1\\1&-\mathrm{i}&-1&\mathrm{i}\\\mathrm{i}&-1&-\mathrm{i}&-1\end{pmatrix},\qquad U_{\{0,1,2\}}=\frac{1}{2}\begin{pmatrix}1&-\mathrm{i}&1&\mathrm{i}\\\mathrm{i}&1&-\mathrm{i}&1\\1&\mathrm{i}&1&-\mathrm{i}\\-\mathrm{i}&1&\mathrm{i}&1\end{pmatrix} \tag*{(6)}
U{0,1}=12(01−i01+i1+i01−i001+i01−i1−i01+i0),U{2,3}=12(0−1+i0−1−i−1−i0−1+i00−1−i0−1+i−1+i0−1−i0)(7)U_{\{0,1\}}=\frac{1}{2}\begin{pmatrix}0&1-\mathrm{i}&0&1+\mathrm{i}\\1+\mathrm{i}&0&1-\mathrm{i}&0\\0&1+\mathrm{i}&0&1-\mathrm{i}\\1-\mathrm{i}&0&1+\mathrm{i}&0\end{pmatrix},\qquad U_{\{2,3\}}=\frac{1}{2}\begin{pmatrix}0&-1+\mathrm{i}&0&-1-\mathrm{i}\\-1-\mathrm{i}&0&-1+\mathrm{i}&0\\0&-1-\mathrm{i}&0&-1+\mathrm{i}\\-1+\mathrm{i}&0&-1-\mathrm{i}&0\end{pmatrix} \tag*{(7)}
U{0,3}=12(01+i01−i1−i01+i001−i01+i1+i01−i0),U{1,2}=12(0−1−i0−1+i−1+i0−1−i00−1+i0−1−i−1−i0−1+i0)(8)U_{\{0,3\}}=\frac{1}{2}\begin{pmatrix}0&1+\mathrm{i}&0&1-\mathrm{i}\\1-\mathrm{i}&0&1+\mathrm{i}&0\\0&1-\mathrm{i}&0&1+\mathrm{i}\\1+\mathrm{i}&0&1-\mathrm{i}&0\end{pmatrix},\qquad U_{\{1,2\}}=\frac{1}{2}\begin{pmatrix}0&-1-\mathrm{i}&0&-1+\mathrm{i}\\-1+\mathrm{i}&0&-1-\mathrm{i}&0\\0&-1+\mathrm{i}&0&-1-\mathrm{i}\\-1-\mathrm{i}&0&-1+\mathrm{i}&0\end{pmatrix} \tag*{(8)}
U{0,2}=(0010000110000100),U{1,3}=(00−10000−1−10000−100)(9)U_{\{0,2\}}=\begin{pmatrix}0&0&1&0\\0&0&0&1\\1&0&0&0\\0&1&0&0\end{pmatrix},\qquad U_{\{1,3\}}=\begin{pmatrix}0&0&-1&0\\0&0&0&-1\\-1&0&0&0\\0&-1&0&0\end{pmatrix} \tag*{(9)}

As indicated, all of them have a common set of eigenvectors; relation (2.1) simplifies in this case to

χm=12{1,im,i2m,i3m}T,m=0,1,2,3;\chi_m=\frac{1}{2}\{1,\mathrm{i}^m,\mathrm{i}^{2m},\mathrm{i}^{3m}\}^{\mathrm{T}},\qquad m=0,1,2,3;

the spectral decomposition of a given USU_S can be then written explicitly as

US=∑m=03λm(χm,⋅)χm=E+−E−,U_S=\sum_{m=0}^{3}\lambda_m(\chi_m,\cdot)\chi_m=E_+-E_-,

where the E±E_{\pm} are projections onto span⁡{χm:λm=±1}\operatorname{span}\{\chi_m:\lambda_m=\pm1\}. In each pair (2.2)–(2.8) the matrices differ by sign, and moreover, the pairs (2.3) and (2.5), as well as (2.6) and (2.7), are complex conjugated.

Remark 2.1. Out of the fourteen matrices (2.2)–(2.8) eight are not invariant with respect to transposition, hence the corresponding vertex couplings give rise to the dynamics violating the time-reversal invariance. One may ask what is the measure of this violation. A natural characterization is given by the norm of the difference US−USTU_S-U_S^{\mathsf{T}}. It is easy to check that each of these matrices has eigenvalues zero of multiplicity two and ±2\pm2, meaning that ∥US−UST∥=2\lVert U_S-U_S^{\mathsf{T}}\rVert=2 so that the violation is maximal saturating the unitarity bound. In contrast to the general circulant vertex couplings considered in [17], the scale-invariant ones can be thus with respect to the time reversal extremal only, either such a coupling preserves the invariance or it violates it maximally.

Statement of the problem

The model

We consider an infinite square lattice graph; without loss of generality we may suppose that its vertices are the points of Z2\mathbb{Z}^2. The graph is the configuration space of a charged particle exposed to a homogeneous magnetic field of strength BB perpendicular to the lattice plane. The Hamiltonian thus acts as ψj↦D2ψj\psi_j\mapsto\mathcal{D}^2\psi_j on the jjth edge, where D=ddx−iA\mathcal{D}=\frac{\mathrm{d}}{\mathrm{d}x}-iA is the outward magnetic quasi-derivative; we denote it as −ΔB,U-\Delta_{B,U} with UU indicating the vertex coupling. We work in the Landau gauge choosing the vector potential as A⃗=(0,Bx,0)\vec{A}=(0,Bx,0), and use the rational units, ℏ=2m=e=c=1\hbar=2m=e=c=1, in which the magnetic flux quantum is Φ0=2π\Phi_0=2\pi; we assume that the flux per plaquette is a rational multiple of this quantity, Φ=2πpq\Phi=2\pi\frac{p}{q}, with coprime q≥2q\geq2 and p∈{1,…,q−1}p\in\{1,\ldots,q-1\}. In other words, the field strength is B=2πpqB=2\pi\frac{p}{q}, vertical edges in the mmth column carry the tangential potential equal to BmBm while the horizontal ones carry none, as in [7].

At the lattice nodes the wave functions satisfy the conditions (1.1) in which U=USU=U_S is one of the matrices (2.2)–(2.8). This is the crucial assumption; any vertex coupling of this type is scale-invariant which means that the matching condition decouple and can be written in the projection form,

E−ψ(v)=0,E+(Dψ)(v)=0(10)E_-\psi(v)=0,\qquad E_+(\mathcal{D}\psi)(v)=0 \tag*{(10)}

with the boundary values laying in E+C4E_+\mathbb{C}^4 and the magnetic quasi-derivatives values in E−C4E_-\mathbb{C}^4. Other consequences of the scale invariance are that the on-shell vertex scattering matrix [11 Sec. 2.1.1] is energy-independent, S(k)=US(k)=U for all kk, and that the negative spectrum of any −ΔB,US-\Delta_{B,U_S} is empty by [8 Thm. 2.6]. Our aim is to find the positive spectrum for all the indicated vertex couplings.

Duality

The scale-invariant character of the coupling allows us, in contrast to [7], to find an equivalent formulation of the problem in terms of an effective difference operator. This is a tool often used in such situations; we will comment on various related problems in § 5.2 below.

Let us denote by dd the dimension of the positive spectral subspace of the matrix determining the coupling, d:=∣S∣=dim⁡E+∈{1,2,3}d:=|S|=\dim E_+\in\{1,2,3\}. Given the splitting of the boundary values and quasi-derivative values at a vertex vv given by (3.1), the former can be written as ψ(v)=∑a∈Sφa(v)χa\psi(v)=\sum_{a\in S}\varphi_a(v)\chi_a with complex coefficients φa(v)\varphi_a(v). For further purposes we note that the entries of the eigenvectors χm\chi_m differ in the ‘left-right’ and ‘up-down’ pairing, j~:=j+2\tilde{j} := j + 2 (mod 4), at most by sign

(χm)j~=i2m(χm)j=(−1)m(χm)j.(11)(\chi_m)_{\tilde{j}} = i^{2m}(\chi_m)_j = (-1)^m(\chi_m)_j. \tag*{(11)}

Next we define for j=1,…,4j = 1, \ldots, 4 and a,b∈Sa, b \in S the d×dd \times d hopping matrices with entries

(Tj)ab:=(−1)bi(b−a)(j−1).(12)(T_j)_{ab} := (-1)^b i^{(b-a)(j-1)}. \tag*{(12)}

Each matrix TjT_j is rank one: to see it, take vectors vj∈Cdv_j \in\mathbb{C}^d with components (vj)a=ia(j−1)(v_j)_a = i^{a(j-1)}, then one can write the action of TjT_j as (Tjϕ)a=(vj‾)a∑b∈S(−1)b(vj)bϕb(T_j\phi)_a = (\overline{v_j})_a \sum_{b \in S} (-1)^b (v_j)_b \phi_b. Furthermore, it is straightforward to check that we have T3=T1∗T_3 = T_1^* and T4=T2∗T_4 = T_2^*.

Remarks 3.1. (a) The hopping matrices depend on the choice of UU through the indices contained in SS which enter in two different ways, through the differences b−ab-a (mod 4) and through the parities (−1)a(-1)^a. For d=2d=2 those are interconnected leading to two different direction types; note that the diagonal of TjT_j is jj-independent and the direction enters only through the off-diagonal factor i(b−a)(j−1)i^{(b-a)(j-1)}. For b−a=2b-a=2 this equals (−1)j−1(-1)^{j-1}, which depends on the axis orientation only, T1=T3T_1=T_3 and T2=T4T_2=T_4; one can say that the hopping cannot distinguish forward steps from backward ones along a line. On the other hand, for b−ab-a odd the factor runs through all four values 1,±i,−11,\pm i,-1, and all four TjT_j are distinct, being related only through the hermiticity pairing.

(b) To put it differently, it follows from (3.3) that tr⁡Tj=∑a∈S(−1)a\operatorname{tr} T_j = \sum_{a \in S}(-1)^a for all jj. Equal parities, S={0,2}S=\{0,2\} and {1,3}\{1,3\}, give trace ±2\pm2, hence the matrices are, modulo the sign, twice orthogonal spin projectors, T1,3=±(I+σ1)T_{1,3}=\pm(I+\sigma_1), T2,4=±(I−σ1)T_{2,4}=\pm(I-\sigma_1), which commute and are simultaneously diagonal, which suggests decoupling of horizontal and vertical motion. For opposite parities the trace vanishes, hence every TjT_j is nilpotent, Tj2=0T_j^2=0, the matrices TjT_j are non-Hermitean and none two of them commute; we will see that the dynamics is then irreducibly two-component.

Using the matrices (3.3) we can introduce the operator HeffSH_{\mathrm{eff}}^S on ℓ2(Z2;Cd)\ell^2(\mathbb{Z}^2;\mathbb{C}^d) by

(HeffSϕ)(m,n)=T1ϕ(m−1,n)+T3ϕ(m+1,n)+e−iBmT2ϕ(m,n+1)+eiBmT4ϕ(m,n−1)(13)(H_{\mathrm{eff}}^S\phi)(m,n) = T_1\phi(m-1,n) + T_3\phi(m+1,n) + e^{-iBm}T_2\phi(m,n+1) + e^{iBm}T_4\phi(m,n-1) \tag*{(13)}

and to state the following duality claim:

Theorem 3.2. Let USU_S be one of matrices (2.2)–(2.8) and B=2πpqB=2\pi\frac{p}{q} with coprime p,q∈Np,q\in\mathbb{N}, then for every k∈(0,∞)∖πZk\in(0,\infty)\setminus\pi\mathbb{Z} we have k2∈σ(−ΔB,US)k^2\in\sigma(-\Delta_{B,U_S}) if and only if 4cos⁡k∈σ(HeffS)4\cos k\in\sigma(H_{\mathrm{eff}}^S).

Proof. We will employ the appropriate Dirichlet-to-Neumann map. Identifying a lattice edge with the interval [0,1][0,1] we suppose that ψ\psi solves −D2ψ=k2ψ-D^2\psi=k^2\psi with constant (tangential component of the) potential AA there. The function is uniquely determined by its endpoint values ψ0,ψ1\psi_0,\psi_1, and the outward quasi-derivatives at the endpoints are

(Doutψ)(0)=ksin⁡k(e−iAψ1−ψ0cos⁡k),(Doutψ)(1)=ksin⁡k(eiAψ0−ψ1cos⁡k).(14)(\mathcal{D}_{\mathrm{out}}\psi)(0)=\frac{k}{\sin k}\left(e^{-iA}\psi_1-\psi_0\cos k\right),\qquad(\mathcal{D}_{\mathrm{out}}\psi)(1)=\frac{k}{\sin k}\left(e^{iA}\psi_0-\psi_1\cos k\right). \tag*{(14)}

Indeed, the solution can be written as ψ(x)=eiAx(ceikx+de−ikx)\psi(x)=e^{iAx}(c e^{ikx}+d e^{-ikx}) in which case Dψ=ikeiAx(ceikx−de−ikx)\mathcal{D}\psi=ik e^{iAx}(c e^{ikx}-d e^{-ikx}). Identifying the endpoint values, we get c+d=ψ0c+d=\psi_0 and ceik+de−ik=e−iAψ1c e^{ik}+d e^{-ik}=e^{-iA}\psi_1; this system is solvable having determinant −2isin⁡k-2i\sin k, nonzero by assumption. This yields the uniqueness of the correspondence, and using the obtained coefficients, we get (14).

We will also employ the expansion bearing the name of Floquet, Bloch, or Gelfand, depending on the community to which you belong. Since by a simple gauge transformation one can check that integer multiples of the flux quantum do not affect the spectrum, both operators commute with translations on one cell vertically and qq cells horizontally, cf. [7] for the picture of the elementary cell with qq vertices and 4q4q ‘half-edges’. Consequently, one has to compare the corresponding fibers, differential one on the lattice graphs cell and the difference one on Cdq\mathbb{C}^{dq}, both having a discrete spectrum, and to show that

dim⁡ker⁡(−ΔB,US(θ)−k2)=dim⁡ker⁡(HeffS(θ)−4cos⁡k)\dim\ker\left(-\Delta_{B,U_S}(\theta)-k^2\right)=\dim\ker\left(H_{\mathrm{eff}}^S(\theta)-4\cos k\right)

holds for any quasimomentum θ=(θ1,θ2)∈[0,2π)2\theta=(\theta_1,\theta_2)\in[0,2\pi)^2, see, e.g., [11 Sec. 4.2].

Let us begin with the map from the graph to the discrete lattice and consider a function ψ∈ker⁡(−ΔB,US(θ)−k2)\psi\in\ker\left(-\Delta_{B,U_S}(\theta)-k^2\right). By (10), the boundary values vector ψ(v)\psi(v) at any vertex belongs to E+C4E_+^{\mathbb{C}^4}; we associate with them the vector ϕ\phi with the components ϕa(v):=⟨χa,ψ(v)⟩\phi_a(v):=\langle\chi_a,\psi(v)\rangle. The map ψ↦ϕ\psi\mapsto\phi is obviously linear and injective: ϕ=0\phi=0 implies ψ=0\psi=0 by the Dirichlet-to-Neumann uniqueness mentioned above. We have to show that ϕ∈ker⁡(HeffS(θ)−4cos⁡k)\phi\in\ker(H_{\mathrm{eff}}^S(\theta)-4\cos k). Let us fix a vertex v=(m,n)v=(m,n) and index a∈Sa\in S. As a subset of an orthonormal basis in C4\mathbb{C}^4, {χb}b∈S\{\chi_b\}_{b\in S} is orthonormal, and ∑j(χa)j‾ψj(v)=ϕa(v)\sum_j\overline{(\chi_a)_j}\psi_j(v)=\phi_a(v) holds by the definition of ϕa(v)\phi_a(v), hence the identity ⟨χa,DΨ(v)⟩=0\langle\chi_a,\mathcal{D}\Psi(v)\rangle=0 can be using the first relation in (14) rewritten as

∑j=14(χa)j‾ωj(m)ψ(vjnext)=cos⁡k ϕa(v),ω1=ω3=1,ω2(m)=ω4(m)‾=e−iBm,(15)\sum_{j=1}^{4}\overline{(\chi_a)_j}\omega_j(m)\psi(v_j^{\mathrm{next}})=\cos k\,\phi_a(v),\qquad\omega_1=\omega_3=1,\quad\omega_2(m)=\overline{\omega_4(m)}=e^{-iBm}, \tag*{(15)}

where vjnextv_j^{\mathrm{next}} is the neighboring vertex at the jjth edge rooted in vv and where ωj(m)\omega_j(m) are the Peierls phases [38], written as e∓iAe^{\mp iA} in (14), coming from our choice of the Landau gauge. We observe that from the viewpoint of the vertex vjnextv_j^{\mathrm{next}}, the edge has the opposite orientation, hence ψ(vjnext)\psi(v_j^{\mathrm{next}}) equals ∑b∈Sϕb(vj)(χb)j‾\sum_{b\in S}\phi_b(v_j)(\chi_b)_{\overline{j}}. Using then the identity (11) together with (χa)j(χb)j=14i(b−a)(j−1)(\chi_a)_j(\chi_b)_j=\frac{1}{4}i^{(b-a)(j-1)}, the left-hand side of (15) becomes 14∑jωj(m)∑b(Tj)abϕb(vj)\frac{1}{4}\sum_j\omega_j(m)\sum_b(T_j)_{ab}\phi_b(v_j), and multiplying the obtained relation by 44 for a fixed a∈Sa\in S, we arrive at (HeffS(θ)ϕ)(v)=4cos⁡k ϕ(v)(H_{\mathrm{eff}}^S(\theta)\phi)(v)=4\cos k\,\phi(v), because the factors ωj(m)\omega_j(m) are nothing but the phases appearing in (13).

In the opposite direction, consider a ϕ∈ker⁡(HeffS(θ)−4cos⁡k)\phi\in\ker(H_{\mathrm{eff}}^{S}(\theta)-4\cos k) and define ψ(v):=∑a∈Sϕa(v)χa\psi(v):=\sum_{a\in S}\phi_a(v)\chi_a at every vertex; since sin⁡k≠0\sin k\ne0 by assumption, one can associate with these endpoint values the unique k2k^2-eigenfunction ψ\psi on each given edge. By construction, it satisfies E−Ψ(v)=0E_{-}\Psi(v)=0, and at the same time we have E+DΨ(v)=0E_{+}D\Psi(v)=0, because using the above argument in the reverse order step by step, we obtain the corresponding discrete equation at vv. Thus ψ∈ker⁡(−ΔB,US(θ)−k2)\psi\in\ker(-\Delta_{B,U_S}(\theta)-k^2), so that the two maps are mutually inverse which concludes the proof. □

Symmetries

An immediate consequence of Theorem 3.2 is that the spectrum of −ΔB,US-\Delta_{B,U_S} is 2π2\pi-periodic in the momentum variable kk. In fact, the spectrum is π\pi-periodic.

Proposition 3.3. σ(HeffS)=−σ(HeffS)\sigma(H_{\mathrm{eff}}^{S})=-\sigma(H_{\mathrm{eff}}^{S}) holds for every SS and every flux, and consequently, σ(−ΔB,US)\sigma(-\Delta_{B,U_S}) is π\pi-periodic in kk on (0,∞)∖πZ(0,\infty)\setminus\pi\mathbb{Z}.

Proof. For any fixed quasimomentum θ=(θ1,θ2)\theta=(\theta_1,\theta_2), the vertical terms in the fiber HeffS(θ)H_{\mathrm{eff}}^{S}(\theta) become e∓i(Bm−θ2)T2,4e^{\mp i(Bm-\theta_2)}T_{2,4} flipping thus sign under the shift θ2↦θ2+π\theta_2\mapsto\theta_2+\pi, while the horizontal ones acquire the sign factor, ϕm↦(−1)mϕm\phi_m\mapsto(-1)^m\phi_m, between the cell boundaries so that they remain the same for even qq and require the additional shift θ1↦θ1+π\theta_1\mapsto\theta_1+\pi for odd qq. This means that HeffS(θ1+πq,θ2+π)H_{\mathrm{eff}}^{S}(\theta_1+\pi q,\theta_2+\pi) is unitarily equivalent to −HeffS(θ1,θ2)-H_{\mathrm{eff}}^{S}(\theta_1,\theta_2), therefore the spectrum of HeffSH_{\mathrm{eff}}^{S} is symmetric with respect to zero. Finally, cos⁡(k+π)=−cos⁡k\cos(k+\pi)=-\cos k, which implies the π\pi-periodicity of all spectral bands of the original operator. □

Let us next look into further symmetries which would allow us to reduce the spectral analysis for the family (2.2)–(2.8) to a smaller number of cases.

Proposition 3.4. σ(−ΔB,−US)=σ(−ΔB,US)\sigma(-\Delta_{B,-U_S})=\sigma(-\Delta_{B,U_S}) holds for every SS and every flux.

Proof. The spectrum of −ΔB,US(θ)-\Delta_{B,U_S}(\theta) away from πZ\pi\mathbb{Z} can be determined using the vertex scattering matrix which is in the scale-invariant case independent of energy, S(k)=USS(k)=U_S. The eigenvalues then come from solutions of the secular equation [11 Sec. 2.2], [32 Sec. 16.2], which reads det⁡(I−eikLSP)=0\det(I-e^{ikL}\mathcal{S}\mathcal{P})=0, where S(k)\mathcal{S}(k) is the bond scattering matrix, LL has edge lengths on the diagonal and P\mathcal{P} carries the edge reversal (connecting two bonds of each edge) together with the Peierls and Floquet phase factors. In our case all edges are of length one, and replacing USU_S by −US-U_S one changes S\mathcal{S} to −S-\mathcal{S} and det⁡(I−eikLSP)\det(I-e^{ikL}\mathcal{S}\mathcal{P}) to det⁡(I+eikLSP)=det⁡(I−ei(k+π)LSP)\det(I+e^{ikL}\mathcal{S}\mathcal{P})=\det(I-e^{i(k+\pi)L}\mathcal{S}\mathcal{P}), thus the spectrum of −ΔB,−US(θ)-\Delta_{B,-U_S}(\theta) is shifted by π\pi in kk, however, by Proposition 3.3 it is π\pi-periodic, hence unchanged, and the conclusion carries over to the direct integral over θ\theta. The exceptional set πZ\pi\mathbb{Z} is invariant under the shift. □

This is not the only symmetry, another one comes from complex conjugation which is, in view of the unitarity, equivalent to transposition.

Proposition 3.5. σ(ΔB,US‾)=σ(−ΔB,US)\sigma(\Delta_{B,\overline{U_S}})=\sigma(-\Delta_{B,U_S}) holds for every SS and every flux.

Proof. Complex conjugation is an antiunitary equivalence between −ΔB,U-\Delta_{B,U} with the plaquette flux Φ\Phi and −Δ−B,U‾-\Delta_{-B,\overline{U}} with the flux −Φ-\Phi and we have U‾S=U−S  (mod⁡4)\overline{U}_S=U_{-S\;(\operatorname{mod} 4)}. We need to care about nonreal USU_S only, that is, the pairs (2.3) and (2.5) for d≠2d\ne2, and (2.6) and (2.7) for d=2d=2. Each of these two groups of four is invariant with respect to the swap S↔−S  (mod⁡4)S\leftrightarrow-S\;(\operatorname{mod} 4) and within each we have symmetry with respect to the flux reversal, p↔q−pp\leftrightarrow q-p (recall that integer multiples of the flux quantum do not matter). □\square

The main result

The described symmetries reduced the number of different situations from fourteen to five. It appears that from the spectral point of view the true number of isospectral classes is three:

  1. Eight couplings (2.2)–(2.5) with d=1d=1 and d=3d=3.

  2. Four couplings (2.6) and (2.7) with d=2d=2.

  3. The pair of couplings (2.8) with d=2d=2.

Theorem 4.1. The spectrum of −ΔB,US-\Delta_{B,U_S} for the coupling matrices USU_S of the indicated classes looks as follows:

I. For every rational value of the flux, Φ=2πpq\Phi=2\pi\frac{p}{q}, it coincides with that of the Kirchhoff lattice, i.e. k2∈σ(−ΔB,US)k^2\in\sigma(-\Delta_{B,U_S}) if and only if 4cos⁡k∈σ(Hp/q)4\cos k\in\sigma(H_{p/q}), where Hp/qH_{p/q} is the Harper operator, together with eigenvalues of infinite multiplicity (flat bands) at every k∈πZk\in\pi\mathbb{Z}.

II. For every rational flux, Φ=2πpq\Phi=2\pi\frac{p}{q}, the spectrum is by Theorem 3.2 dual that of the two-component operator with hopping matrices T1,3=σ3∓iσ2T_{1,3}=\sigma_3\mp i\sigma_2 and T2,4=σ3±iσ1T_{2,4}=\sigma_3\pm i\sigma_1 (for the given choice of the basis in E+C4E_+\mathbb{C}^4), or equivalently to that of the dimerized Harper chain

bm−1+sin⁡ξm am+cos⁡ξm bm=2cos⁡k am,am+1+cos⁡ξm am−sin⁡ξm bm=2cos⁡k bm,(16)\begin{aligned} b_{m-1}+\sin\xi_m\,a_m+\cos\xi_m\,b_m&=2\cos k\,a_m,\\ a_{m+1}+\cos\xi_m\,a_m-\sin\xi_m\,b_m&=2\cos k\,b_m, \tag*{(16)} \end{aligned}

where ξm=Bm−θ2\xi_m=Bm-\theta_2. There are no flat bands; the butterfly has three fractal clusters per period, centered at k≡π4,π2,3π4(modπ)k\equiv\frac{\pi}{4},\frac{\pi}{2},\frac{3\pi}{4}\pmod{\pi}.

III. The lattice graph decouples into independent horizontal and vertical one-dimensional lines, gauge equivalent to the non-magnetic situation; the spectrum is [0,∞)[0,\infty) independently of the flux, and there are no flat bands.

Proof. Class I. For d=1d=1 the hopping matrices are just numbers; if S={m}S=\{m\}, formula (3.3) gives Tj=(−1)mT_j=(-1)^m for all four directions, hence the dual operator is the (−1)m(-1)^m multiple of the well-known two-dimensional Harper operator HαH_\alpha, α=pq\alpha=\frac{p}{q}, cf. [24], also known under other names such as the discrete magnetic Laplacian [40]. In view of the chiral symmetry from Proposition 3.3, σ(HeffS)=σ(−HeffS)\sigma(H_{\mathrm{eff}}^S)=\sigma(-H_{\mathrm{eff}}^S), so that the spectrum is the same for both signs, identical with that of the Kirchhoff vertex coupling to which the first of the matrices (2.2) corresponds. For the other half of this class, d=3d=3, the coupling is of the form −U{m}-U_{\{m\}} so the result follows from Proposition 3.4. As for the flat bands, in the Kirchhoff case they are the usual Dirichlet loop states having for rational flux zeros at the vertices. For the other cases one can check their existence numerically, see Remark 4.2 below.

Class II. For S={0,3}S=\{0,3\} formula (12) gives T3=σ3+iσ2T_3=\sigma_3+i\sigma_2, T2=σ3+iσ1T_2=\sigma_3+i\sigma_1 and their adjoints. In the Landau gauge we use, HeffH_{\mathrm{eff}} is diagonalized by Bloch waves in the nn-direction: writing ϕ(m,n)=eiθ2nφm\phi(m,n)=e^{i\theta_2 n}\varphi_m and rotating the spinor by the Hadamard matrix W=(σ1+σ3)/2W=(\sigma_1+\sigma_3)/\sqrt{2}, the components φ~m=(am,bm)T=Wφm\widetilde{\varphi}_m=(a_m,b_m)^{\mathsf T}=W\varphi_m satisfy the chain equation (16). The other three members are obtained by conjugation and sign flip, hence they are isospectral by Propositions 3.4 and 3.5. The absence of flat bands and the shape of the spectrum in dependence on the flux shown in Figure 1 for q=1,…,30q=1,\ldots,30 have been obtained numerically. The clusters centers at λ=0,±22\lambda=0,\pm2\sqrt{2} are the loci of the band contacts of the flux-one-half line.

Class III. Consider S={0,2}S=\{0,2\} in which case E+C4=span⁡{χ0,χ2}={ψ:ψl=ψr, ψu=ψd}E_+\mathbb{C}^4=\operatorname{span}\{\chi_0,\chi_2\}=\{\psi:\psi_l=\psi_r,\ \psi_u=\psi_d\}; these coincidences of the boundary values mean that the wave function is continuous along each straight line through the vertex, with no relation between the two lines. The derivative conditions ⟨χ0,DΨ⟩=⟨χ2,DΨ⟩=0\langle\chi_0,\mathcal{D}\Psi\rangle=\langle\chi_2,\mathcal{D}\Psi\rangle=0 from (10) are equivalent to Dl+Dr=0\mathcal{D}_l+\mathcal{D}_r=0 and Du+Dd=0\mathcal{D}_u+\mathcal{D}_d=0, i.e. to Kirchhoff conditions separately on each line. In the operator sense, the lattice is thus a disjoint union of free one-dimensional ‘wires’; the tangential component vector potential on a straight wire is easily removed by a gauge transformation, so each wire contributes by [0,∞)[0,\infty) which is also the total spectrum, independent of the flux. Equivalently, on the dual side relation (12) gives T1,3=I+σ1T_{1,3}=I+\sigma_1, T2,4=I−σ1T_{2,4}=I-\sigma_1, which commute; in the σ1\sigma_1 eigenbasis the fibers decouple into the free horizontal chain with eigenvalues 4cos⁡θ1+2πlq4\cos\frac{\theta_1+2\pi l}{q} and the on-site vertical values 4cos⁡(Bm−θ2)4\cos(Bm-\theta_2), whose union over θ\theta is [−4,4][-4,4]. The other case, S={1,3}S=\{1,3\}, is covered by Proposition 3.4. □

Remarks 4.2. Except Kirchhoff, the existence of flat bands for class I has been checked numerically, see the disclosure claim below. Remarkably, for d=3d=3 there is a way to see their origin. Take. e.g., S={0,2,3}S=\{0,2,3\}, where formula (12) gives

T1=(11−111−111−1),T2=(1−1i−11−ii−i−1),T3=T1∗,T4=T2∗=T‾2,T_1=\begin{pmatrix}1&1&-1\\1&1&-1\\1&1&-1\end{pmatrix},\qquad T_2=\begin{pmatrix}1&-1&i\\-1&1&-i\\i&-i&-1\end{pmatrix},\qquad T_3=T_1^*,\qquad T_4=T_2^*=\overline{T}_2,

in the basis (χ0,χ2,χ3)(\chi_0,\chi_2,\chi_3); the other three d=3d=3 patterns are obtained analogously. By the symmetries, the above proof shows that spectrally these cases coincide with Harper’s operator, but on the level of wave functions they are different. The structure of hopping matrices shows that all four directions are now distinct; their traces are ∑a∈S(−1)a=−(−1)m′=±1\sum_{a\in S}(-1)^a=-(-1)^{m'}=\pm1, where m′m' is the index missing in SS and they satisfy Tj2=±TjT_j^2=\pm T_j being neither nilpotent as in class II nor Hermitian as in class III. Concerning the flat bands, each fiber component spectrum consists of 3q3q eigenvalues of which qq coincide with those of −σ(Hp/q(θ))-\sigma(H_{p/q}(\theta)) and the two others, each of multiplicity qq, are situated at ±4\pm4, the edges of the duality window.

(b) The collapse from five to three spectral classes has two sources which the symmetries alone do not reveal. The first is the exact π\pi-periodicity from Proposition 3.3, which converts the k↦k+πk \mapsto k+\pi shift of the sign flip into an equality of spectra, merging thereby each d=3d=3 coupling with its d=1d=1 partner; the second is the bipartite gauge equivalence of the four scalar duals, which merges the three class I symmetry cases.

Discussion

The present result

The three spectral classes are shown in Figure 2 in terms of the duality variable λ=4cos⁡k\lambda=4\cos k; in view of Theorem 3.2 and Proposition 3.3, each period [nπ,(n+1)π][n\pi,(n+1)\pi] of the momentum axis carries one copy of the respective diagram. Since the aim here is a comparison, the resolution is coarser than in Figure 1. The differences between the two nontrivial classes are obvious,

Figure 2. The three spectral classes of scale-invariant circulant couplings, in the duality variable λ=4cos⁡k\lambda=4\cos k, for all coprime p/qp/q with q≤12q\le12. In class I the heavy lines at λ=±4\lambda=\pm4 indicate the flat bands at k∈πZk\in\pi\mathbb{Z}, in class II the dashed lines at λ=±22\lambda=\pm2\sqrt{2} and λ=0\lambda=0 mark the three fractal clusters centers.

in the periodicity, shape of the fractal clusters, or the fact that class II is gapless at the half-of-the-quantum flux. To elucidate the mechanism behind this difference, it instructive to inspect the dispersion in the zero-field situation.

Corollary 5.1. For B=0B=0 the class-II lattice has the dispersion relation

cos⁡2k=cos⁡θ1cos⁡θ2.(17)\cos2k = \cos\theta_1\cos\theta_2. \tag*{(17)}

In particular, the spectrum is all of [0,∞)[0,\infty), with no flat bands, and the two dispersion surfaces of −Δ0,US-\Delta_{0,\mathrm{US}} touch conically at the quasimomenta (θ1,θ2)=(0,π),(π,0)(\theta_1,\theta_2)=(0,\pi),(\pi,0), that is, at the momenta k≡π2k\equiv\frac{\pi}{2} (mod π\pi).

Proof. For definiteness, consider S={0,3}S=\{0,3\}, the other cases are similar. If B=0B=0, the operator (13) is invariant under all lattice translations, so the Fourier Ansatz ϕ(m,n)=ei(θ1m+θ2n)φ\phi(m,n)=e^{i(\theta_1m+\theta_2n)}\varphi gives the fiber a 2×22\times2 matrix representation, namely

H(θ)=(σ3−iσ2)e−iθ1+(σ3+iσ2)eiθ1+(σ3+iσ1)eiθ2+(σ3−iσ1)e−iθ2=2(cos⁡θ1+cos⁡θ2)σ3−2sin⁡θ1σ2−2sin⁡θ2σ1=:h(θ)⋅σ.H(\theta)=(\sigma_3-i\sigma_2)e^{-i\theta_1}+(\sigma_3+i\sigma_2)e^{i\theta_1}+(\sigma_3+i\sigma_1)e^{i\theta_2}+(\sigma_3-i\sigma_1)e^{-i\theta_2} =2(\cos\theta_1+\cos\theta_2)\sigma_3-2\sin\theta_1\sigma_2-2\sin\theta_2\sigma_1=:\mathbf{h}(\theta)\cdot\sigma.

The matrix is traceless – recall that tr⁡Tj=0\operatorname{tr}T_j=0 for class II by Remark 3.1(b) – hence its eigenvalues are ±∣h(θ)∣\pm|\mathbf{h}(\theta)|, for which a simple manipulation gives

∣h(θ)∣2=4(cos⁡θ1+cos⁡θ2)2+4sin⁡2θ1+4sin⁡2θ2=8(1+cos⁡θ1cos⁡θ2).|\mathbf{h}(\theta)|^2=4(\cos\theta_1+\cos\theta_2)^2+4\sin^2\theta_1+4\sin^2\theta_2=8(1+\cos\theta_1\cos\theta_2).

The spectral duality from Theorem 3.2 gives 16cos⁡2k=8(1+cos⁡θ1cos⁡θ2)16\cos^2 k=8(1+\cos\theta_1\cos\theta_2), i.e. 2cos⁡2k−1=cos⁡θ1cos⁡θ22\cos^2 k-1=\cos\theta_1\cos\theta_2, which is (17). The right-hand side range is [−1,1][-1,1] as θ\theta runs over the Brillouin zone, so every k>0k>0 is attained and the spectrum is [0,∞)[0,\infty). The dispersion surfaces touch at the points where h(θ)=0\mathbf{h}(\theta)=0: expanding around (θ1,θ2)=(0,π)(\theta_1,\theta_2)=(0,\pi) with θ1=δ1\theta_1=\delta_1, θ2=π+δ2\theta_2=\pi+\delta_2, one finds ∣h(θ)∣=2δ12+δ22+O(δ3)|\mathbf{h}(\theta)|=2\sqrt{\delta_1^2+\delta_2^2}+O(\delta^3), a conical (Dirac) contact at λ=4cos⁡k=0\lambda=4\cos k=0, i.e. at k≡π2k\equiv\frac{\pi}{2} (mod π\pi). □

The dispersion (17) contrasts with 4cos⁡k=2cos⁡θ1+2cos⁡θ24\cos k=2\cos\theta_1+2\cos\theta_2, its Kirchhoff counterpart, which contains the sum of the quasimomentum cosines instead of their product; the two spectra are gapless but the Kirchhoff’s contains flat bands which are absent here. The conical contacts of Corollary 5.1 can be regarded as the seeds of the central cluster of the class-II butterfly; one can recall the central cluster of the graphene butterfly emanating from the Dirac points of the honeycomb dispersion. Here the cluster centered at λ=0\lambda=0 in Figure 2 grows out of the two zero-field Dirac contacts at k≡π2k\equiv\frac{\pi}{2}.

Remark 5.2. To indicate to what degree does the mechanism suggested by Corollary 5.1 extends to a nonzero flux, consider the square of the chain operator HcH_c acting on ψ=(ambm)\psi=\begin{pmatrix}a_m\\b_m\end{pmatrix} according to (16). A direct computation gives

(Hc2ψ)am=2am+cos⁡ξm am+1+cos⁡ξm−1 am−1+ρm bm−1,(Hc2ψ)bm=2bm+cos⁡ξm+1 bm+1+cos⁡ξm bm−1+ρm+1 am+1,(18)\begin{aligned} (H_c^2\psi)_{a_m}&=2a_m+\cos\xi_m\,a_{m+1}+\cos\xi_{m-1}\,a_{m-1}+\rho_m\,b_{m-1},\\ (H_c^2\psi)_{b_m}&=2b_m+\cos\xi_{m+1}\,b_{m+1}+\cos\xi_m\,b_{m-1}+\rho_{m+1}\,a_{m+1}, \tag*{(18)} \end{aligned}

where ρm:=sin⁡ξm−sin⁡ξm−1\rho_m:=\sin\xi_m-\sin\xi_{m-1}, because the mixed same-cell terms cancel. We note that the diagonal is independent of mm; putting thus ϵ=2cos⁡k\epsilon=2\cos k and using the identity ϵ2−2=2cos⁡2k\epsilon^2-2=2\cos2k, we see that the operator G:=12(Hc2−2)G:=\frac{1}{2}(H_c^2-2) satisfies the ‘doubled’ duality relation: k2∈σ(−ΔB,US)k^2\in\sigma(-\Delta_{B,\mathrm{US}}) if and only if cos⁡2k∈σ(G)\cos2k\in\sigma(G). Within each sublattice, GG is a purely off-diagonal Harper chain with no on-site term and hopping amplitudes 12cos⁡ξm\frac{1}{2}\cos\xi_m between consecutive aa’s, and similarly for the bb’s. The two sublattices are coupled by the bond ρm\rho_m which can be rewritten as

ρm=2sin⁡(πpq)cos⁡(ξm−πpq),\rho_m = 2\sin\left(\frac{\pi p}{q}\right)\cos\left(\xi_m-\frac{\pi p}{q}\right),

being a Harper-modulated amplitude of the same frequency and the global strength 2sin⁡(πp/q)2\sin(\pi p/q). It vanishes identically if and only if B=0B=0, and in that case the sublattices decouple and the Bloch reduction of the aa-chain reproduces dispersion relation (17). At a nonzero flux, this term prevents such a scalar reduction and generates the central cluster of the butterfly. Note that the amplitude is maximal at half-of-the-quantum flux when the spectrum is gapless.

On another topic, as the motivation to this paper we mentioned the spectral problem for the same lattice with a simple vertex coupling that violated the time-reversal invariance [7]. The said coupling was not scale-invariant so the obtained spectrum lacked the kk-periodicity; it was distorted at low energies while in the high-energy asymptotics it approached the Hofstadter butterfly pattern, except for narrow regions around the values where the Kirchhoff lattice had the flat bands. Here the result is significantly different, and not only because of the periodicity, natural in our case; the main, and to a degree surprising distinction is the way in which the time asymmetry, in the present model always maximal by Remark 2.1, enters. The new spectrum of class II is related to the time-reversal invariance violation as none of the matrices (2.6) and (2.7) is transpose invariant. On the other hand, the spectrum of class I has nothing to do with the time-reversal invariance, or the absence of it; one half of the couplings (2.2)–(2.5) gives rise to time-reversal invariant dynamics, the other to an irreversal one.

The result opens various questions. The most natural one concerns the nature of the spectrum for irrational values of the flux; one can conjecture that it would be a Cantor set in analogy with the answer to the classical Ten Martini Problem [3]. Of interest is also whether the Haussdorff dimension dichotomy known in the critical almost Mathieu situation [22, 33] carries over to the class II spectrum. At the same time, the present result certainly does not exhaust all questions about the rational flux case; one can ask, for instance, what would be a counterpart of the Thouless conjecture [21, 24, 44, 45] here.

Related problems in the literature

Our construction is connected with several areas beyond the familiar Kirchhoff and almost-Mathieu correspondence. Dualities: The differential-difference operator connection is the core element of our construction. The duality idea can be traced back to [12, 13] for weighted Kirchhoff on equilateral graphs and to [16] for δ\delta- and δ′\delta'-couplings; an abstract boundary-triple formulation is due to [35, 36]. Theorem 3.2 extends this to scale-invariant couplings, for which the dual operator is genuinely matrix-valued, of fiber dimension dim⁡ker⁡(U−I)\dim\ker(U-I). In this connection one can mention results on honeycomb lattices. Kuchment and Post [31] reduced the non-magnetic graphene quantum graph to a discrete two-component problem, Becker, Han, and Jitomirskaya [9] carried out the magnetic reduction and proved Cantor spectrum for irrational flux, see also [10]. The difference is that the honeycomb is an overlay of two lattices; in our case the matrix character comes from properties of a single vertex.

Discrete magnetic operators on periodic graphs: The operators (13) can be naturally regarded as magnetic Schrödinger operators on Z2\mathbb{Z}^{2} with a dd-dimensional fiber, d=1,2,3d = 1, 2, 3. A general band structure analysis for Zd\mathbb{Z}^{d} periodic discrete graphs was done by Korotyaev and Saburova, both in the non-magnetic [25] and magnetic [26] cases. From the similar reason as above, however, it does not apply directly to our present problem because the multidimensionality is not there a single-site matter but comes from the number of vertices in the elementary cell.

Harper-type models with modulated hopping: One-dimensional quasiperiodic operators in which the hopping exhibits a (typically incommensurate) modulation appear repeatedly in the literature; one can recall, for instance, the extended Harper model expressed in terms of off-diagonal cosine factors analyzed by Avila, Jitomirskaya, and Marx [4]. For comparison, squaring the chain (16) and projecting onto one sublattice produces, by (18), up to the coupling ρm\rho_m of strength 2sin⁡(πp/q)2\sin(\pi p/q) made explicit in Remark 5.2, a purely off-diagonal Harper chain with hopping cos⁡ξm\cos\xi_m and constant diagonal. However, the chain itself does not belong to the exactly solved extended-Harper family; the modulation appears both in the dimerized bond and the staggered site with the unitary constraint cos⁡2ξm+sin⁡2ξm=1\cos^{2}\xi_m + \sin^{2}\xi_m = 1, which recalls rather a structure appearing in the literature on the unitary side — in CMV matrices and the unitary almost-Mathieu operator [18], and in discrete-time quantum walks in electric and magnetic gauge fields [14, 15]. In contrast, here we deal with a Hermitean lattice operator having a unitary parametrization.

Hofstadter-type spectra. Butterfly patterns coming from multicomponent lattice models appear in various places; one can mention. e.g., Rammal’s spectrum of the honeycomb (graphene) tight-binding model in a field [1, 30, 39], butterflies of flat-band lattices [2], deformed square-lattice butterflies with next-nearest-neighbour hopping [20], or Hofstadter problems with non-Abelian gauge fields [34]. Our class II is distinct from all of these in that the magnetic field is Abelian and the matrix structure of the hopping is not a gauge degree of freedom but a fixed, rank-one nilpotent imprint of the vertex coupling — each bond transmits a single scalar channel, the hallmark of the quantum-graph origin.

AI usage disclosure. The text was prepared in dialogue with Anthropic Claude Fable used to check some steps in the reasoning, perform the numerical computations, and assist in literature search. The AI suggestions were verified and the text was written by the authors who take the full responsibility for its correctness.

Data availability statement. Data are available in the article.

Conflict of interest. The authors have no conflict of interest.

References

  1. [1]A. Agazzi, J.P. Eckmann, G.M. Graf: The colored Hofstadter butterfly for the honeycomb lattice, J. Stat. Phys. 156 (2014), 417–426.DOI
  2. [2]H. Aoki, M. Ando, H. Matsumura: Hofstadter butterflies for flat-band lattices, Phys. Rev. B54 (1996), R17296–R17299.
  3. [3]A. Avila, S. Jitomirskaya: The Ten Martini Problem, Ann. Math. 170 (2009), 303–342.
  4. [4]A. Avila, S. Jitomirskaya, C.A. Marx: Spectral theory of extended Harper’s model and a question by Erdős and Szekeres, Invent. Math. 210 (2017), 283–339.DOI
  5. [5]M.Ya. Azbel: Energy spectrum of a conduction electron in a magnetic field, Sov. Phys. JETP 19 (1964), 634–644.
  6. [6]R. Band, G. Berkolaiko, U. Smilansky: Dynamics of nodal points and the nodal count on a family of quantum graphs, Ann. H. Poincaré 13 (2012), 145–184.DOI
  7. [7]M. Baradaran, P. Exner, J. Lipovský: Magnetic square lattice with vertex coupling of a preferred orientation, Ann. Phys. 454 (2023), 169339.DOI
  8. [8]M. Baradaran, P. Exner, M. Tater: Spectrum of periodic chain graphs with time-reversal non-invariant vertex coupling, Ann. Phys. 443 (2022), 168992.arxiv.org/abs/2012.14344
  9. [9]S. Becker, R. Han, S. Jitomirskaya: Cantor spectrum of graphene in magnetic fields, Invent. Math 218 (2019), 979–1041.
  10. [10]S. Becker, M. Zworski: Magnetic oscillations in a model of graphene, Commun. Math. Phys. 367 (2019), 941–989.DOI
  11. [11]G. Berkolaiko, P. Kuchment: Introduction to Quantum Graphs, AMS, Providence, R.I. 2013.DOI
  12. [12]J. von Below: A characteristic equation associated to an eigenvalue problem on c²-networks, Lin. Alg. Appl. 71 (1985), 309–325.DOI
  13. [13]C. Cattaneo: The spectrum of the continuous Laplacian on a graph, Monats. Math. 124 (1997), 215–235.DOI
  14. [14]C. Cedzich, T. Geib, A.H. Werner, R.F. Werner: Quantum walks in external gauge fields, J. Math. Phys. 60 (2019), 012107.DOI
  15. [15]C. Cedzich, T. Rybár, A.H. Werner, A. Alberti, M. Genske, R.F. Werner: Propagation of quantum walks in electric fields, Phys. Rev. Lett. 111 (2013), 160601.pubmed.ncbi.nlm.nih.gov/24182244
  16. [16]P. Exner: A duality between Schrödinger operators on graphs and certain Jacobi matrices, Ann. Inst. H. Poincaré: Phys. Théor. 66 (1997), 359–371.
  17. [17]P. Exner, M. Tater: Quantum graphs: self-adjoint, and yet exhibiting a nontrivial PT-symmetry, Phys. Lett. A416 (2021), 127669DOI
  18. [18]J. Fillman, D.C. Ong, Z. Zhang: Spectral characteristics of the unitary critical almost-Mathieu operator, Commun. Math. Phys. 351 (2017), 525–561.DOI
  19. [19]B. Gutkin, S. Smilansky: Can one hear the shape of a graph?, J. Phys. A: Math. Gen. 34 (2001), 6061–6068.DOI
  20. [20]Y. Hatsugai, M. Kohmoto: Energy spectrum and the quantum Hall effect on the square lattice with next-nearest-neighbor hopping, Phys. Rev. B42 (1990), 8282–8294.
  21. [21]B. Helffer, P. Kerdelhue: On the total bandwidth for the rational Harper’s equation, Commun. Math. Phys. 173 (1995), 335–356.DOI
  22. [22]B. Helffer, Qinghui Liu, Yanhui Qu, Qi Zhou: Positive Hausdorff dimensional spectrum for the critical almost Mathieu operator, Commun. Math. Phys. 368 (2019), 369–382.
  23. [23]D.R. Hofstadter: Energy levels and wave functions of Bloch electrons in rational and irrational magnetic fields, Phys. Rev. B14 (1976), 2239–2249.
  24. [24]S. Jitomirskaya, L. Konstantinov, I. Krasovsky: On the spectrum of critical almost Mathieu operators in the rational case, J. Spectr. Theory 12 (2022), 11–21.arxiv.org/abs/2007.01005
  25. [25]E. Korotyaev, N. Saburova: Schrödinger operators on periodic discrete graphs, J. Math. Anal. Appl. 420 (2014), 576–611.arxiv.org/abs/1307.1841
  26. [26]E. Korotyaev, N. Saburova: Magnetic Schrödinger operators on periodic discrete graphs, J. Funct. Anal. 272 (2017), 1625–1660.arxiv.org/abs/1512.08207
  27. [27]V. Kostrykin, R. Schrader: Quantum wires with magnetic fluxes, Commun. Math. Phys. 237 (2003), 161–179.
  28. [28]T. Kottos, U. Smilansky: Quantum chaos on graphs, Phys. Rev. Lett 79 (1997), 4794–4797.DOI
  29. [29]T. Kottos, U. Smilansky: Periodic orbit theory and spectral statistics for quantum graphs, Ann. Phys. 274 (1999), 76–124.DOI
  30. [30]C. Kreft, R. Seiler: Models of the Hofstadter type, J. Math. Phys. 37 (1996), 5207–5243.DOI
  31. [31]P. Kuchment, O. Post: On the spectra of carbon nano-structures, Commun. Math. Phys. 275 (2007), 805–826.
  32. [32]P. Kurasov: Spectra Geometry of Graphs, Birkhäuser, Basel 2024.
  33. [33]Y. Last, M. Shamis: Zero Hausdorff dimension spectrum for the almost Mathieu operator, Commun. Math. Phys. 348 (2016), 729–750.DOI
  34. [34]K. Osterloh, M. Baig, L. Santos, P. Zoller, M. Lewenstein: Cold atoms in non-Abelian gauge potentials: from the Hofstadter “moth” to lattice gauge theory, Phys. Rev. Lett. 95 (2005), 010403.DOI
  35. [35]K. Pankrashkin: Spectra of Schrödinger operators on equilateral quantum graphs, Lett. Math. Phys. 77 (2006), 139–154.DOI
  36. [36]K. Pankrashkin: Unitary dimension reduction for a class of self-adjoint extensions with applications to graph-like structures, J. Math. Anal. Appl. 396 (2012), 640–655.DOI
  37. [37]L. Pauling: The diamagnetic anisotropy of aromatic molecules. J. Chem. Phys. 4 (1936), 673–677.DOI
  38. [38]R. Peierls: Zur Theorie des Diamagnetismus von Leitungselektronen, Zs. Phys. 80 (1933), 763–791.DOI
  39. [39]R. Rammal: Landau level spectrum of Bloch electrons in a honeycomb lattice, J. Physique 46 (1985), 1345–1354.DOI
  40. [40]M.A. Shubin: Discrete magnetic Laplacian, Commun. Math. Phys. 164 (1994), 259–275.DOI
  41. [41]U. Smilansky: Irreversible quantum graphs, Waves in Random Media 14 (2004), S143–S153.DOI
  42. [42]P. Středa, J. Kučera, Orbital momentum and topological phase transformation, Phys. Rev. B92 (2015), 235152.DOI
  43. [43]P. Středa, K. Výborný: Anomalous Hall conductivity and quantum friction, Phys. Rev. B107 (2023), 014425.DOI
  44. [44]D.J. Thouless: Bandwidths for a quasiperiodic tight binding model, Phys. Rev. B28 (1983), 4272–4276.DOI
  45. [45]D.J. Thouless: Scaling for the discrete Mathieu equation, Commun. Math. Phys. 127 (1990), 187–193.DOI

Paper details

vlnc:W_eu4xuxhg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents