betaFeedback

WorksPhysics and Astronomy

Intrinsic topological magnons in arrays of magnetic dipoles

Abstract

We study a simple magnetic system composed of periodically modulated magnetic dipoles with an easy axis. Upon adjusting the modulation amplitude alone, chains and two-dimensional stacked chains exhibit a rich magnon spectrum where frequency gaps and magnon speeds are easily manipulable. The blend of anisotropy due to dipolar interactions between magnets and geometrical modulation induces a magnetic phase with fractional Zak number in infinite chains and end states in open one-dimensional systems. In two dimensions it gives rise to topological modes at the edges of stripes. Tuning the amplitude in two-dimensional lattices causes a band touching, which triggers the exchange of the Chern numbers of the volume bands and switches the sign of the thermal conductivity.

Introduction

The latest experimental findings associated with twisted heterostructures[1], demonstrated that handling a single geometrical parameter in a system is a simple yet highly effective strategy for manipulating electric correlations and realizing diverse electronic, magnetic, and topological phases. In the realm of magnetic natural and artificial matter, the recent proposal on Moire magnets[2], magnetic counterpart to twistronic, promises to complement and enlarge the current possibilities for manipulation of the flow of spin angular momentum with and without an accompanying charge current[3, 4, 5]. Currents of spins can be produced by the transport of electrons or, without their aid, by the collective propagation of coupled precessing spins or spin waves[6]. In magnetically ordered systems, based on magnetic insulators, ferromagnetic metals or heterostructures, spin-wave excitations, and their quantized versions, magnons[7, 8] can be controlled by varying the layer thicknesses and applying external fields[9, 10, 11, 12]. In artificial magnonic crystals[13] the periodic modulation of the magnetic properties can be engineered, which makes possible the formation of magnonic bands separated by gaps in one-dimensional, two-dimensional, and bicomponent structures[14]. The band’s gaps in these structures can be manipulated by combining changes in the crystal parameters, modification of crystal materials, and the tuning of direction and amplitude of magnetic fields[11, 15, 16, 17, 18].

In addition to gap manipulation and the reduction of heat associated with dissipation, the unidirectional transmission of information using spin waves is another aspect being considered. In topological magnonics[10, 19, 20, 21], the bulk-edge correspondence[22] ensures that spin waves at the edges of a sample realize chiral propagation along the edges regardless of the specific device geometry. Non-trivial topology is usually accompanied by the emergence of exotic phenomena[23, 24, 25]. This is the case in Lu2_2V2_2O7_7, an insulating collinear ferromagnet pyrochlore where spin excitations give rise to the anomalous thermal Hall effect[24]. In Lu2_2V2_2O7_7 the propagation of the spin waves is influenced by the antisymmetric Dzyaloshinskii-

Moriya (DM) spin-orbit interaction, which plays the role of the vector potential. This is also the case in the compound YMn6_6Sn6_6, a metallic system consisting of ferromagnetic kagome planes[26], where the subsequent magnon band gap opening at the symmetry-protected K points is ascribed to DM interactions. Antiferromagnetic materials with honeycomb lattices add to the list of unique crystal structures which realize topological phases[27, 28, 29].

In two dimensions (2D), magnonic crystals can host topological magnons when the system under consideration breaks time and space inversion symmetries[30]. Current evidence shows that this is possible when the system exposes spin-orbit physics on superlattices. In these structures, the common elements usually consist of units cells with the triangular motif and interactions that include all or some of the following elements: DM, anisotropy, exchange interactions, dipolar interactions, and external magnetic fields[20, 21, 31, 32]. Control over the band gaps and topological features in these structures include variations in magnetic fields, DM, temperature, and exchange interactions[13, 33, 34].

In systems where dipolar interactions play a dominant role, several proposals of chiral spin-wave modes in dipolar magnetic films have been put forward[13, 30, 35, 36], when subjected to an external magnetic field. Topological magnons also arise in magnonic crystals with the dipolar coupling truncated after a few neighbor magnets and in the case of antiferromagnetic films in the long-wave limit[13, 35]. In these systems, control over the dipolar magnonic bands is achieved through the application of magnetic fields.

With all the notable breakthroughs associated to topological magnons, significant progress could be achieved if the shape and the topology of the magnonic bands could be manipulated by precise control of a single geometrical parameter and without the aid of external fields. Here we present a new magnetic system where engineering of the bandgap and Berry curvature is possible by tuning a single intrinsic geometrical parameter. The system consists of two-dimensional lattices in the xx - zz plane whose building blocks are one-dimensional (1D) lattices of dipoles that extend along the xx direction and stack regularly along the zz direction. The sites along chains are regularly arranged in a modulated periodic fashion. Dipoles have an easy-axis anisotropy and couple with each other through full anisotropic magnetic dipolar interaction. Equilibrium magnetic states, spin-wave spectrum, and topological features of modulated chains, stripes, and lattices can be explicitly controlled by tuning Λ\Lambda, which is defined as the amplitude of the spatial modulation of the chain’s sites. The study of the magnon spectrum of chains reveals two topological phase transitions in terms of Λ\Lambda. They come along with magnon modes localized at the edges of open systems and jump discontinuities in the Fermi velocity of magnons. The 1D topology is inherited by the 2D lattices, which manifest non-trivial topology at Λ>0\Lambda> 0 and realize chiral magnonic modes at the edge of stripes.

The paper is organized as follows. In section II we present the model and show the energy spectrum of chains and two-dimensional lattices in terms of Λ\Lambda. In section III we show results for the spin-wave spectrum using the Linearized Landau-Lifshitz equations. Section IV is dedicated to the study of topological aspects of 1D and 2D systems. We summarize our results in Section V.

Model

The classical hamiltonian consists of dipolar interactions and uniaxial anisotropy. In units of Joule [J] it reads

H=D2∑i≠j=1nm^i⋅m^j−3(m^i⋅e^ij)(m^j⋅e^ij)∣ri−rj∣3−K2∑i=1n(m^i⋅z^)2(1)\mathcal{H} = \frac{\mathcal{D}}{2}\sum_{i\ne j=1}^{n}\frac{\hat{\mathbf{m}}_i\cdot\hat{\mathbf{m}}_j-3(\hat{\mathbf{m}}_i\cdot\hat{\mathbf{e}}_{ij})(\hat{\mathbf{m}}_j\cdot\hat{\mathbf{e}}_{ij})}{\lvert\mathbf{r}_i-\mathbf{r}_j\rvert^3}-\frac{\mathcal{K}}{2}\sum_{i=1}^{n}(\hat{\mathbf{m}}_i\cdot\hat{\mathbf{z}})^2 \tag*{(1)}

where nn is the total number of sites and ri\mathbf{r}_i denotes the position of a point dipole in a two dimensional array in the x−zx-z plane. e^ij=(ri−rj)/∣ri−rj∣\hat{\mathbf{e}}_{ij}=(\mathbf{r}_i-\mathbf{r}_j)/\lvert\mathbf{r}_i-\mathbf{r}_j\rvert, and D=μ0m024πa03\mathcal{D}=\frac{\mu_0m_0^2}{4\pi a_0^3} has units of [Nm] and contains the physical parameters involved in the dipolar energy such as μ0\mu_0, the magnetic permeability, a0a_0, the distance among nearest neighbour dipoles along the x^\hat{x} direction, and m0m_0, the intensity of the magnetic moments with units [m2A\mathrm{m}^2\mathrm{A}]. K\mathcal{K} is the easy axis anisotropy along the z^\hat{z} axis and has units of [Nm]. Magnetic moments have unit vector m^i=(cos⁡θi,0,sin⁡θi)\hat{\mathbf{m}}_i=(\cos\theta_i,0,\sin\theta_i). Point dipoles are located at the vertices of modulated chains that extend along the x^\hat{x} direction as highlighted in Figure 1. Magnets can rotate in an easy x−zx-z plane described by an azimuthal angle θi\theta_i which is chosen with respect to the z^\hat{z} axis.

Sketches of stripes built out of modulated chains of dipoles

Figure 1. Sketches of stripes built out of modulated chains of dipoles. (a) Along each chain, point dipoles are located at Λsin⁡2(κrx)\Lambda\sin^2(\kappa r_x) with rxr_x the position of dipoles along x^\hat{x}, κ=π/2\kappa=\pi/2 the wave vector of the modulation, and Λ\Lambda the amplitude of the modulation as illustrated by the dotted (blue) curve. a0a_0 denotes the distance between two nearest neighbor dipoles along x^\hat{x}. Lattice constant along x^\hat{x} and z^\hat{z} are 2a02a_0 and 3a03a_0, respectively. The dotted square highlights the unit cell, where (1) and (2) denote the two sublattices. (a) and (b) show stripes in the antiferromagnetic and dimer parallel magnetic orders, respectively. (c) and (d) show stripes in the ferromagnetic and antiferromagnetic collinear magnetic orders, respectively. Single chains are highlighted in grey.

The periodic modulation of dipoles in the x−zx-z plane is set by Λsin⁡2(κrx)\Lambda\sin^2(\kappa r_x), where Λ\Lambda is the amplitude of the modulation (in units of a0a_0), rxr_x denotes the position of dipoles along x^\hat{x} and κ\kappa is the wave vector of the modulation. Modulated chains are one-dimensional lattices with a basis. Depending on κ\kappa, the unit cell contains more than one dipole (each dipole in the unit cell defines a sublattice). Periodic chains are stacked across the z^\hat{z} axis and are the building blocks of stripes and lattices. Hereafter, we focus on the relevant case of κ=π2\kappa=\frac{\pi}{2} that corresponds to a lattice with a unit cell containing two dipoles (lattice with a two-point basis). The two sublattices are denoted by (1) and (2), and the unit cell is highlighted by the dotted square in Figure 1(a). The lattice constants along x^\hat{x} and z^\hat{z} axes are 2a02a_0 and 3a03a_0 respectively, otherwise stated. Systems with κ=π4\kappa=\frac{\pi}{4} and κ=π8\kappa=\frac{\pi}{8} are briefly discussed in the supplementary material[37]. Neither magnon-phonon nor magnon-magnon interactions are considered in this paper.

Magnetic configurations that minimize Eq. (1).

Next, we examine the effect of K\mathcal{K} and Λ\Lambda on the energy spectrum of the dipolar systems. In the paper, Λ\Lambda is tuned in the domain 0<Λ<3a00<\Lambda<3a_0.

Energetics of finite systems.

Depending on Λ\Lambda and K\mathcal{K}, finite modulated chains of point dipoles minimize Eq. (1) in the magnetic configurations illustrated in Figure 1. For small anisotropy K<KcA\mathcal{K}<\mathcal{K}_{c}^{A}, the collinear magnetic order, Figure 1(c-Figure 1d) is favored. For K>KcΛ\mathcal{K}>\mathcal{K}^{\Lambda}_{c} the parallel magnetic states (Figure 1(a-Figure 1b)) minimize energy. From Eq. (1) one can find the critical anisotropy that sets the boundary between collinear and parallel configurations. For chains, this is found to be

KcΛ=2D(3ζ(3)4−12ψ(2)(Λ2+1))(2)\mathcal{K}^{\Lambda}_{c}=2D\left(\frac{3\zeta(3)}{4}-\frac{1}{2}\psi^{(2)}\left(\sqrt{\Lambda^2+1}\right)\right) \tag*{(2)}

where ψ\psi is the PolyGamma function and ζ\zeta the Riemann zeta function (details can be found in [37]). Thus KcΛ\mathcal{K}^{\Lambda}_{c} becomes smaller as Λ\Lambda grows. For 2D lattices (0<Λ<3a00<\Lambda<3a_0), the critical anisotropy becomes KcΛ,L∼KcΛ+D\mathcal{K}^{\Lambda,L}_{c}\sim\mathcal{K}^{\Lambda}_{c}+D.

Consider the case of low anisotropy (K<KcΛ\mathcal{K}<\mathcal{K}^{\Lambda}_{c}). At Λ>0\Lambda>0, chains relax into two possible collinear magnetic configurations along x^\hat{x}. One, at low Λ\Lambda (roughly Λ<1\Lambda<1) is the ferromagnetic state shown in Figure 1(c), the other at larger amplitudes is the collinear antiferromagnetic state shown in Figure 1(d). For high anisotropy, K>Kc\mathcal{K}>\mathcal{K}_{c} and Λ≲1\Lambda\lesssim1 the antiferromagnetic parallel configuration along z^\hat{z} is favored, Figure 1(a). At intermediate amplitudes 1≲Λ≲21\lesssim\Lambda\lesssim2 dimers consisting of nearest neighbour ferromagnetic parallel dipoles arranged in an antiferromagnetic fashion ‘down-down-up-up’ (and time reversal) along z^\hat{z} are favored (Figure 1(b)). A second dimer configuration, ‘up-down-down-up’ (and time-reversal) competes energetically with the first. For Λ>2\Lambda>2 the two sublattices behave like independent chains with twice the lattice constant. The dimerized configurations in finite chains are not perfect. They contain domain walls or kinks with dipoles pointing in the wrong direction, which disturb the perfect dimer order[38]. A few kinks also occur in the antiferromagnetic parallel state. However, they are practically absent in the collinear configurations. Snapshots product of the energy minimization of finite chains can be found in[37].

Minimization of the dipolar energy in 2D systems results in stable magnetic configurations that preserve the magnetic order of its building blocks, as illustrated in Figure 1 and shown in the next section. Hereafter, we focus on systems at large anisotropy K>KcΛ\mathcal{K}>\mathcal{K}^{\Lambda}_{c}. Their energy scale is set by DD.

Band spectrum in terms of Λ\Lambda.

In units of DD, the interacting dipolar hamiltonian can be written

Hdd=∑i≠jMj†J^ijMi=∑i≠jMj†(Jij(11)Jij(12)Jij(21)Jij(22))Mi.(3)\mathcal{H}^{\mathrm{dd}}=\sum_{i\ne j}\mathcal{M}_{j}^{\dagger}\hat{\mathcal{J}}_{ij}\mathcal{M}_{i}= \sum_{i\ne j}\mathcal{M}_{j}^{\dagger} \begin{pmatrix} J_{ij}^{(11)} & J_{ij}^{(12)}\\ J_{ij}^{(21)} & J_{ij}^{(22)} \end{pmatrix} \mathcal{M}_{i}. \tag*{(3)}

Here, Mj†=(m^j(1),m^j(2))\mathcal{M}_{j}^{\dagger}=(\hat{m}_{j}^{(1)},\hat{m}_{j}^{(2)}), where m^j(1)\hat{m}_{j}^{(1)} and m^j(2)\hat{m}_{j}^{(2)} denote the magnetic moment of the jj-th dipole belonging to sublattice (1) and the jj-th dipole belonging to sublattice (2) respectively, see Figure 1(a). J^ij\hat{\mathcal{J}}_{ij} denotes the interaction matrix between dipoles ii and jj. In the parallel configuration along z^\hat{z}, the interaction among all dipoles that belong to the same sublattice in a chain reads ∑i≠jJij(11)≡Jz(11)=Jz(22)=∑n=1∞18n3\sum_{i\ne j}J_{ij}^{(11)}\equiv J_{z}^{(11)}=J_{z}^{(22)}=\sum_{n=1}^{\infty}\frac{1}{8n^3} while for dipoles in different sublattices ∑i≠jJij(12)=Jz(12)=Jz(21)=∑n=1∞−1+3Λ2(n2+Λ2)3/2\sum_{i\ne j}J_{ij}^{(12)}=J_{z}^{(12)}=J_{z}^{(21)}=\sum_{n=1}^{\infty}\frac{-1+3\Lambda^2}{(n^2+\Lambda^2)^{3/2}}. In momentum space Mq=1n∑jMjeiq⋅rj\mathcal{M}_{q}=\frac{1}{\sqrt{n}}\sum_j\mathcal{M}_{j}e^{iq\cdot r_j} and Eq. (2) becomes

Hdd=∑qMq†J^qMq,(4)\mathcal{H}^{\mathrm{dd}}=\sum_q\mathcal{M}_{q}^{\dagger}\hat{\mathcal{J}}_{q}\mathcal{M}_{q}, \tag*{(4)}

where J^q=(Jq,z(11)Jq,z(12)Jq,z(12)∗Jq,z(11))\hat{\mathcal{J}}_q=\begin{pmatrix}J_{q,z}^{(11)}&J_{q,z}^{(12)}\\J_{q,z}^{(12)*}&J_{q,z}^{(11)}\end{pmatrix}. Diagonalization of the interaction matrix J^q\hat{\mathcal{J}}_q for chains results in two branches with energy ϵ1(qx)=Jqx(11)+Jqx(12)\epsilon_1(q_x)=J_{q_x}^{(11)}+J_{q_x}^{(12)} and ϵ2(qx)=Jqx(11)−Jqx(12)\epsilon_2(q_x)=J_{q_x}^{(11)}-J_{q_x}^{(12)} with Jqx(11)=−18s1J_{q_x}^{(11)}=-\frac{1}{8}s_1, Jqx(12)=s3+Λ2s4J_{q_x}^{(12)}=s_3+\Lambda^2s_4[37] and

s1=Φ3(e2iqx),s_1=\Phi_3\left(e^{2iq_x}\right),
s3=eiqxf(Λ)Φ(eiqx,3,f(Λ)),s_3=e^{iq_xf(\Lambda)}\Phi\left(e^{iq_x},3,f(\Lambda)\right),
s4=e−iqxf(Λ)Φ(e−iqx,5,f(Λ))s_4=e^{-iq_xf(\Lambda)}\Phi\left(e^{-iq_x},5,f(\Lambda)\right)

where f(Λ)=1+Λ2f(\Lambda)=\sqrt{1+\Lambda^2} and Φ\Phi is the Polylogarithmic function. Figure 2 shows the energy bands of infinite (with periodic boundary conditions) chains (left panel), stripes made out of 60 infinite chains stacked along z^\hat{z} (middle), and 2D lattices (right panel). In the figure, momenta along x^\hat{x} and z^\hat{z} are shown in units of the primitive vectors of the first Brillouin zone (1BZ), Gx=2πa0G_x=\frac{2\pi}{a_0} and Gz=2π3a0G_z=\frac{2\pi}{3a_0}. Bands are obtained from the exact diagonalization of interaction matrices J^q\hat{\mathcal{J}}_q at several values of Λ\Lambda. Qualitatively, the spectra of chains, stripes, and lattices look alike. The main features like the band crossing at Gx2\frac{G_x}{2} are preserved in the three types of arrays and remain invariant as Λ\Lambda grows. Examination of the wavevectors of the minimum eigenvalues of the interaction matrix reveals that for Λ≲0.7\Lambda\lesssim0.7 the two branches in the energy spectrum of chains correspond to the antiferromagnetic and dimer configurations shown in Figure 1(a-Figure 1b). At the Γ\Gamma point, the minimum energy corresponds to the antiferromagnetic mode. Increasing Λ\Lambda augments the relative interaction strength between dipoles in the same sublattice. Therefore, at the band crossing at Λ=0.7\Lambda= 0.7, the second dimerized configuration becomes energetically favorable compared to the antiferromagnetic state in the chains, and now each energy branch corresponds to one of the two dimer modes.

Band spectrum of chains, stripes, and lattices

Figure 2. Band spectrum of chains (left panel), stripes (middle panel) and lattices (right panel, side view at qz=0q_z=0) at K>KcΛ\mathcal{K} > \mathcal{K}^{\Lambda}_{c}, κ=π2\kappa=\frac{\pi}{2} and (a) Λ=0.5\Lambda=0.5, (b) Λ=0.7\Lambda=0.7 and (c) Λ=1\Lambda=1. Momenta along x^\hat{x} and z^\hat{z} are measured in units of GxG_x and GzG_z respectively. The minimum eigenvalues ϵ\epsilon are in units of DD.

The middle panel of Figure 2 shows that bands in stripes sort into bundles or groups of bands. These bundles, whose number equals the number of energy branches in the chains, resemble the bulk bands of the lattices shown in the right panel of Figure 2.

Overall, increasing Λ\Lambda reduces rather drastically the energy gap at the Γ\Gamma point of the 1BZ. In addition, larger amplitudes amplify the relative interaction strength between dipoles belonging to different chains. This fact manifests as a qualitative change in the dispersive character of the bands, which become flatter as Λ\Lambda grows.

Magnon spectrum

Next we examine the collective transverse excitations of the dipoles magnetization vector with respect to the parallel states found in § 0.2 A 1. Neglecting damping, the dynamics of the magnetization vector of a dipole jj, belonging to sublattice α\alpha, m^jα(t)\hat{\mathbf{m}}_j^\alpha(t) is described by the Landau-Lifshitz equation[39, 40, 41, 42, 43, 44, 45]

dm^jαdt=γ(Beff,jα×m^jα)(5)\frac{d\hat{\mathbf{m}}_j^\alpha}{dt}=\gamma\left(\mathbf{B}_{\mathrm{eff},j}^\alpha\times\hat{\mathbf{m}}_j^\alpha\right) \tag*{(5)}

where γ\gamma is the modulus of the gyromagnetic ratio, and the effective magnetic field Beff,jα=−∂Hdd∂m^jα\mathbf{B}_{\mathrm{eff},j}^\alpha=-\frac{\partial H_{\mathrm{dd}}}{\partial\hat{\mathbf{m}}_j^\alpha} consists of the dipolar field created by other magnetic dipoles,

Beff,jα=−μ0∑i,βJ^ijα,β⋅m^iβ(6)\mathbf{B}_{\mathrm{eff},j}^\alpha=-\mu_0\sum_{i,\beta}\hat{\mathcal{J}}_{ij}^{\alpha,\beta}\cdot\hat{\mathbf{m}}_i^\beta \tag*{(6)}

where J^ijα,β\hat{\mathcal{J}}_{ij}^{\alpha,\beta} is the interaction matrix containing the geometrical aspects of the interactions between all dipoles in the array as discussed in the previous section. Hereafter we drop the hat from unit vectors. Magnetization of the jjth dipole in the stationary ground state μjα=(0,0,μjα)\boldsymbol{\mu}_j^\alpha=(0,0,\mu_j^\alpha) is a unit vector in the direction of the static ground state. It points along a local axis which we call z^\hat{z} and satisfies the system of equations:

Bjαμjα=−μ0∑i,βJ^ijα,β⋅μiβ(7)B_j^\alpha\mu_j^\alpha=-\mu_0\sum_{i,\beta}\hat{\mathcal{J}}_{ij}^{\alpha,\beta}\cdot\boldsymbol{\mu}_i^\beta \tag*{(7)}

where BjαB_j^\alpha is the intrinsic scalar magnetic field acting on the jjth dipole belonging to the α\alpha sublattice and J^α,β\hat{\mathcal{J}}^{\alpha,\beta} is the interaction matrix of the system in the stationary ground state. To find the dynamical equations describing small (linear) transverse magnetization excitations, we use the following ansatz for the dipole magnetization:

mjα(t)=(μjα+m~jα(t))(8)\mathbf{m}_j^\alpha(t)=\left(\boldsymbol{\mu}_j^\alpha+\tilde{\mathbf{m}}_j^\alpha(t)\right) \tag*{(8)}

where m~jα(t)=(m~j,xα(t),m~j,yα(t))\tilde{\mathbf{m}}_j^\alpha(t)=\left(\tilde{m}_{j,x}^\alpha(t),\tilde{m}_{j,y}^\alpha(t)\right) is the small dimensionless deviation of the magnetization vector of the jjth dipole from the static equilibrium state. Conservation of the length of the magnetization vector in each magnet requires that μj⋅m~j=0\boldsymbol{\mu}_j\cdot\tilde{\mathbf{m}}_j=0. Eq. (8) in Eq. (5) yields:

dm~jαdt=γBeff,jα×m~jα(9)\frac{d\tilde{\mathbf{m}}_j^\alpha}{dt}=\gamma\mathbf{B}_{\mathrm{eff},j}^\alpha\times\tilde{\mathbf{m}}_j^\alpha \tag*{(9)}

The effective field in terms of the scalar field and the stationary magnetization reads:

Beff,jα=Bjαμjα−μ0∑i,βJ^ijα,β⋅m~iβ(10)\mathbf{B}_{\mathrm{eff},j}^\alpha=B_j^\alpha\boldsymbol{\mu}_j^\alpha-\mu_0\sum_{i,\beta}\hat{\mathcal{J}}_{ij}^{\alpha,\beta}\cdot\tilde{\mathbf{m}}_i^\beta \tag*{(10)}

which back into Eq. (9) gives rise to

dm~jαdt=γ(Bjαμjα×m~jα+μ0∑i,βμjα×J^ijα,β⋅m~iβ)(11)\frac{d\tilde{\mathbf{m}}_j^\alpha}{dt}=\gamma\left(B_j^\alpha\boldsymbol{\mu}_j^\alpha\times\tilde{\mathbf{m}}_j^\alpha+\mu_0\sum_{i,\beta}\boldsymbol{\mu}_j^\alpha\times\hat{\mathcal{J}}_{ij}^{\alpha,\beta}\cdot\tilde{\mathbf{m}}_i^\beta\right) \tag*{(11)}

Therefore equilibrium orientations of the uniform magnetization μjα\boldsymbol{\mu}_j^\alpha and internal fields BjαB_j^\alpha depend only on the index α\alpha. At each sublattice, the linear spin-wave excitations have the form of plane waves. Thus Fourier transforming the magnetic excitation vector in time and space yields,

m~jα=1n∑qm~qαe−i(q⋅rjα+ωt)(12)\tilde{\mathbf{m}}_j^\alpha=\frac{1}{\sqrt{n}}\sum_q\tilde{\mathbf{m}}_q^\alpha e^{-i(\mathbf{q}\cdot\mathbf{r}_j^\alpha+\omega t)} \tag*{(12)}

where rjα\mathbf{r}_j^\alpha is the vector position of a dipole belonging to sublattice α\alpha and located at the jjth unit cell. Substituting Eq. (12) back into Eq. (11) one obtains a finite dimensional eigenvalue problem:

−iωqm~qα=γμqα×∑β(Bαm~qα+μ0J^qα,β⋅m~qβ)-i\omega_q\tilde{\mathbf{m}}_q^\alpha=\gamma\boldsymbol{\mu}_q^\alpha\times\sum_\beta\left(B^\alpha\tilde{\mathbf{m}}_q^\alpha+\mu_0\hat{\mathcal{J}}_q^{\alpha,\beta}\cdot\tilde{\mathbf{m}}_q^\beta\right)
−iωqm~qα=γμqα×∑βΩ^qα,β⋅m~qβ(13)-i\omega_q\tilde{\mathbf{m}}_q^\alpha=\gamma\boldsymbol{\mu}_q^\alpha\times\sum_\beta\hat{\Omega}_q^{\alpha,\beta}\cdot\tilde{\mathbf{m}}_q^\beta \tag*{(13)}

where

Ω^qα,β=Bαδα,β+μ0J^qα,β(14)\hat{\Omega}_q^{\alpha,\beta}=B^\alpha\delta^{\alpha,\beta}+\mu_0\hat{\mathcal{J}}_q^{\alpha,\beta} \tag*{(14)}

Since the vector m~qβ=(m~q,xβ,m~q,yβ)\tilde{\mathbf{m}}_q^\beta=\left(\tilde{m}_{q,x}^\beta,\tilde{m}_{q,y}^\beta\right) one can separate Eq. (13) along the x^\hat{x} and y^\hat{y} (transverse) directions:

−iωqm~q,xα=γ[μqα×∑βΩ^qα,β⋅m~qβ]x(15)-i\omega_q\tilde{m}_{q,x}^\alpha=\gamma\left[\boldsymbol{\mu}_q^\alpha\times\sum_\beta\hat{\Omega}_q^{\alpha,\beta}\cdot\tilde{\mathbf{m}}_q^\beta\right]_x \tag*{(15)}
−iωqm~q,yα=γ[μqα×∑βΩ^qα,β⋅m~qβ]y(16)-i\omega_q\tilde{m}_{q,y}^\alpha=\gamma\left[\boldsymbol{\mu}_q^\alpha\times\sum_\beta\hat{\Omega}_q^{\alpha,\beta}\cdot\tilde{\mathbf{m}}_q^\beta\right]_y \tag*{(16)}

with

[μqα×∑βΩ^qα,β⋅m~qβ]x=−μqα∑β(Bαδα,β+J^q,yα,β)m~q,yα(17)\left[\mu_q^\alpha\times\sum_\beta\hat{\Omega}_q^{\alpha,\beta}\cdot\tilde{m}_q^\beta\right]_x=-\mu_q^\alpha\sum_\beta\left(B^\alpha\delta^{\alpha,\beta}+\hat{J}_{q,y}^{\alpha,\beta}\right)\tilde{m}_{q,y}^\alpha \tag*{(17)}

and

[μqα×∑βΩ^qα,β⋅m~qβ]y=μqα∑β(Bαδα,β+J^q,xα,β)m~q,xα(18)\left[\mu_q^\alpha\times\sum_\beta\hat{\Omega}_q^{\alpha,\beta}\cdot\tilde{m}_q^\beta\right]_y=\mu_q^\alpha\sum_\beta\left(B^\alpha\delta^{\alpha,\beta}+\hat{J}_{q,x}^{\alpha,\beta}\right)\tilde{m}_{q,x}^\alpha \tag*{(18)}

Reconfigurable magnon frequencies.

In terms of magnon creation m+=mx+imym_+ = m_x + i m_y and annihilation m−=mx−imym_- = m_x - i m_y fields, the matrix form of the equations of motion Eq. (17) and Eq. (18) becomes

ωσ^z(m+m−)=(A^B^B^A^)(m+m−)(19)\omega\hat{\sigma}_z \begin{pmatrix} m_+\\ m_- \end{pmatrix} = \begin{pmatrix} \hat{A} & \hat{B}\\ \hat{B} & \hat{A} \end{pmatrix} \begin{pmatrix} m_+\\ m_- \end{pmatrix} \tag*{(19)}

where the right hand side matrix constitutes the magnon hamiltonian. The Pauli matrix σ^z\hat{\sigma}_z, takes +1+1 for the creation field or particle space and −1-1 for the annihilation field or hole space. The 2×22 \times2 matrices A^=(a1a2a2∗a1)\hat{A} = \begin{pmatrix} a_1 & a_2 \\ a_2^* & a_1 \end{pmatrix} and B^=(b1b2b2∗b1)\hat{B} = \begin{pmatrix} b_1 & b_2 \\ b_2^* & b_1 \end{pmatrix} with a1=γϵ0+γ2(Jq,x(11)+Jq,y(11))a_1 = \gamma\epsilon_0 + \frac{\gamma}{2}(J_{q,x}^{(11)} + J_{q,y}^{(11)}), a2=γϵ0+γ2(Jq,x(12)+Jq,y(12))a_2 = \gamma\epsilon_0 + \frac{\gamma}{2}(J_{q,x}^{(12)} + J_{q,y}^{(12)}), b1=γ2(Jq,x(11)−Jq,y(11))b_1 = \frac{\gamma}{2}(J_{q,x}^{(11)} - J_{q,y}^{(11)}), b2=γ2(Jq,x(12)−Jq,y(12))b_2 = \frac{\gamma}{2}(J_{q,x}^{(12)} - J_{q,y}^{(12)}) and ϵ0\epsilon_0 is the energy of the stationary magnetic state (hereafter we set ϵ0=0\epsilon_0 = 0). Expressions for Jq,x11J_{q,x}^{11} and Jq,y12J_{q,y}^{12} are shown in [37]. The magnon hamiltonian can be written as:

H=σ^z⊗(t^0+t^1)+σ^x⊗(t^2+t^3)(20)\mathcal{H}=\hat{\sigma}_z\otimes(\hat{t}_0+\hat{t}_1)+\hat{\sigma}_x\otimes(\hat{t}_2+\hat{t}_3) \tag*{(20)}

where σ^j\hat{\sigma}_j is the jth Pauli matrix, ⊗\otimes denotes Kronecker product, t^0=(a100a1)\hat{t}_0=\begin{pmatrix}a_1&0\\0&a_1\end{pmatrix}, t^1=(0a2a2∗0)\hat{t}_1=\begin{pmatrix}0&a_2\\a_2^*&0\end{pmatrix}, t^2=(b100b1)\hat{t}_2=\begin{pmatrix}b_1&0\\0&b_1\end{pmatrix} and t^3=(0b2b2∗0)\hat{t}_3=\begin{pmatrix}0&b_2\\b_2^*&0\end{pmatrix}. In Eq. (20) the term multiplying σ^z\hat{\sigma}_z is a mass term responsible for the gap, while the term multiplying σ^x\hat{\sigma}_x is proportional to the group velocity of the spin waves or magnon speed. Eigenfrequencies for collective spin wave modes in the particle space can be written as

ω1,22=γ24(a12±2a1a2+a22+b12±2b1b2+b22)(21)\omega_{1,2}^2=\frac{\gamma^2}{4}\left(a_1^2\pm2a_1a_2+a_2^2+b_1^2\pm2b_1b_2+b_2^2\right) \tag*{(21)}

where a1a_1, a2a_2, b1b_1 and b2b_2 can be written in terms of s1s_1, s3s_3 and s4s_4[37]. Figure 3, Figure 4 and Figure 5 show the effect of Λ\Lambda in the magnon spectrum of chains, lattices and stripes respectively. They result from the exact diagonalization of Eq. (20). Similar to the stationary energy spectrum, the frequency dispersion in 2D systems resembles magnon bands in chains. Smaller values of Λ\Lambda allow for a wider range of frequencies for magnon excitation at the Γ\Gamma point in all the arrays.

Magnon spectrum of infinite chains for five values of $\Lambda$

Figure 3. Magnon spectrum of infinite chains (a) Λ=0.5\Lambda= 0.5, (b) Λ=1\Lambda= 1, (c) Λ=1.3\Lambda= 1.3, (d) Λ=1.5\Lambda= 1.5 and (e) Λ=1.8\Lambda= 1.8. Band touchings are encircled. In (d) note the band touching at the three encircled points. Frequency ω\omega is in units of γ\gamma and qxq_x is in units of GxG_x.

Magnon spectrum of 2D lattices

Figure 4. Magnon spectrum of 2D lattices (side view at qz=0q_z = 0) with (a) Λ=0.5\Lambda= 0.5, (b) Λ=1\Lambda= 1, (c) Λ=1.5\Lambda= 1.5, and (d) Λ=1.8\Lambda= 1.8. In (c) two band touchings like Dirac points are located at (Gx3,Gz2)\left(\frac{G_x}{3},\frac{G_z}{2}\right) and (2Gx3,Gz2)\left(\frac{2G_x}{3},\frac{G_z}{2}\right) (and the two equivalent at qz=3Gz2q_z = \frac{3G_z}{2}) when Λ=32\Lambda= \frac{3}{2}. They resemble the case of infinite chains. For larger Λ\Lambda the group velocity of magnons decreases. Frequency ω\omega is in units of γ\gamma and qxq_x is in units of GxG_x.

Magnon spectrum of stripes at (a) $\Lambda=0.8$, (b) $\Lambda=1.0$, (c) $\Lambda=1.3$ and (d) $\Lambda=1.6$

Figure 5. Magnon spectrum of stripes at (a) Λ=0.8\Lambda=0.8, (b) Λ=1.0\Lambda=1.0, (c) Λ=1.3\Lambda=1.3 and (d) Λ=1.6\Lambda=1.6. Edge modes crossing the gap and joining the two bulk bands are highlighted. Frequency ω\omega is in units of γ\gamma and qxq_x is in units of GxG_x.

Λ\Lambda has a strong effect on the group velocity of the spin waves: larger amplitudes yield flatter bands and therefore lower magnon speeds around the middle of the spectrum.

Compare, for instance, frequency slopes of systems with Λ=0.5\Lambda=0.5 and Λ=1.8\Lambda=1.8 in Figure 4(a) and (Figure 4d) respectively. Notable features are the band touchings in chains and lattices. In 1D, the first touching occurs in the middle of the spectrum q0=Gx2q_0=\frac{G_x}{2} at Λ=1≡Λ1\Lambda=1\equiv\Lambda_1, Figure 3(b). At Λ>Λ1\Lambda>\Lambda_1 the gap opens again until Λ=1.5≡Λ2\Lambda=1.5\equiv\Lambda_2 where the touching at q0q_0 reappears and two additional band touchings arise at q1∼Gx4q_1\sim\frac{G_x}{4} and q2∼3Gx4q_2\sim\frac{3G_x}{4}, Figure 3(d). For Λ>Λ2\Lambda>\Lambda_2 the gap at q1q_1 and q2q_2 remains closed while that at Gx2\frac{G_x}{2} reopens. One can estimate the critical amplitude for which the magnonic gap closes by equating the eigenfrequencies ω1\omega_1 and ω2\omega_2 in Eq. (21). In 1D, for qx=Gx2q_x=\frac{G_x}{2} this yields the equality

Λqx=π22=12Φ(3,3,1+Λ2)Φ(3,5,1+Λ2)(22)\Lambda^2_{q_x=\frac{\pi}{2}}=\frac{1}{2}\frac{\Phi\left(3,3,\sqrt{1+\Lambda^2}\right)}{\Phi\left(3,5,\sqrt{1+\Lambda^2}\right)} \tag*{(22)}

which is satisfied at Λ∼Λ1\Lambda\sim\Lambda_1.

In the case of lattices the band touching happens at Λ2\Lambda_2 at two points of the 1BZ, (Gx3,Gy2)\left(\frac{G_x}{3},\frac{G_y}{2}\right), (2Gx3,Gy2)\left(\frac{2G_x}{3},\frac{G_y}{2}\right) as shown in Figure 4(c). Here the inset shows a close up of the band spectrum about the band touching points (and the equivalent (Gx3,3Gy2)\left(\frac{G_x}{3},\frac{3G_y}{2}\right) and (2Gx3,3Gy2)\left(\frac{2G_x}{3},\frac{3G_y}{2}\right)) which resembles Dirac cones. At all other Λ\Lambda the spectra of the lattices are gapped.

In stripes, Figure 5 shows magnon bands that cross the gap between the two bulk bands (highlighted in orange).

Isolated bands located above and below the bulk bands arise as well for large Λ\Lambda, as the ones shown below the lowest frequency band in Figure 5(d).

Topological Bands and Edge Modes.

In previous sections, we showed that the spin-wave spectrum acquires forbidden frequency band gaps due to the periodic modulation of the dipolar arrays. In this section, we revisit the spin-wave spectrum of the stripes. We note in Figure 5 the onset of features in between the two bulk bands, which resemble edge modes. Aimed to unveil the nature of the in-gap modes of Figure 5 we computed the Chern number of the volume bands associated with them. A non-zero Chern integer, cnc_n, for spin-wave volume bands results in the emergence of chiral spin-wave edge modes. These topological edge bands have chiral dispersion that favors the unidirectional propagation of magnetic degrees of freedom for a frequency between the gap. In addition, they are robust to intrinsic and externally induced disorder[22]. It is expected that finite Chern integers, cnc_n, result from strong spin-orbit coupled interactions[36, 46]. With an inner product between magnetization and position vectors, the magnetic dipolar interaction locks the relative rotational angle between the spin space and orbital space, similar to what the relativistic spin-orbit interaction does in electron systems[36]. As a result of the spin-orbit locking, the complex-valued character in the spin space is transferred into wave functions in the orbital space.

Λ\Lambda0.20.50.711.21.41.51.61.82.0
c1c_11111111-1-1-1
c2c_2-1-1-1-1-1-1-1111

Table 1. Chern numbers, of the magnon volume bands in 2D lattices at several values of Λ\Lambda. Note the exchange of the Chern number between the two bands at Λ2=1.5\Lambda_2 = 1.5.

Gauge connection and the associated Berry curvature in the lattice.

The sense of motion and the number of chiral modes in a system is determined from the magnitude and sign of the topological number for volume mode bands below the bandgap. cnc_n, can be changed only by closing the gap[22, 36]. Here, frequency bands are computed on a discretized Brillouin zone. Thus, we follow the approach of reference[47] to compute the Berry phase[48] and the Chern integer using wave functions given on such discrete points. The Chern number assigned to the nnth band is the integral of fictitious magnetic fields: that is,

field strengths of the Berry connection. It is defined by

cn=12πi∫τ2d2q Fxz(q)(23)c_n = \frac{1}{2\pi\mathrm{i}}\int_{\tau^2} d^2q\,F_{xz}(q) \tag*{(23)}

where τ2\tau^2 denotes the Brillouin zone torus. The gauge connection (gauge field) Aμ(q)A_\mu(q) (μ=x,z\mu= x,z) and the associated field strength Fxz(q)F_{xz}(q) or Berry curvature are given by

Aμ=⟨n(q)∣∂μ∣n(q)⟩(24)A_\mu= \langle n(q)|\partial_\mu|n(q)\rangle \tag*{(24)}
Fxz=∂xAz(q)−∂zAx(q)F_{xz} = \partial_x A_z(q) - \partial_z A_x(q)

where ∣n(q)⟩|n(q)\rangle is the normalized wave function of the nnth (particle) Bloch band such that H(q)∣n(q)⟩=ωn(q)∣n(q)⟩H(q)|n(q)\rangle= \omega_n(q)|n(q)\rangle. On the discrete Brillouin zone with lattice points qlq_l, l=1,…nxnzl = 1,\ldots n_xn_z the lattice field strength is given by

F~xz(ql)≡ln⁡Ux(ql)Uz(ql+x^)Ux(ql+z^)−1Uz(ql)−1(25)\widetilde{F}_{xz}(q_l) \equiv\ln U_x(q_l)U_z(q_l+\hat{x})U_x(q_l+\hat{z})^{-1}U_z(q_l)^{-1} \tag*{(25)}
−π<1iF~xz(ql)≤π-\pi< \frac{1}{i}\widetilde{F}_{xz}(q_l) \leq\pi

where Uμ(ql)U_\mu(q_l) is a U(1) link variable from a nnth band,

Uμ(ql)≡⟨n(ql)∣n(ql+μ^)⟩∣⟨n(ql)∣n(ql+μ^)⟩∣(26)U_\mu(q_l) \equiv\frac{\langle n(q_l)|n(q_l+\hat{\mu})\rangle}{|\langle n(q_l)|n(q_l+\hat{\mu})\rangle|} \tag*{(26)}

and the gauge invariant Chern number on the lattice[47] associated to the nnth band is finally

c~n≡12πi∑lF~xz(ql)(27)\widetilde{c}_n \equiv\frac{1}{2\pi i}\sum_l\widetilde{F}_{xz}(q_l) \tag*{(27)}

Topological magnons at the edges of stripes

Using the previous approach, we computed the Chern numbers for lattices corresponding to the bulk bands of the stripes shown in Figure 5. The results are presented in Table I. For Λ≤Λ2\Lambda\leq\Lambda_2 the lowest frequency band has Chern number c1=1c_1 = 1 while the upper band has c2=−1c_2 = -1. This topological phase of the magnon dispersion is characterized in terms of c1c_1 and c2c_2 and denoted (1,-1). At Λ>Λ2\Lambda> \Lambda_2 the Chern numbers of the two bands are exchanged, the topological phase is denoted (-1,1). Going back to Figure 4(c), we note two band touchings for the case Λ=Λ2\Lambda= \Lambda_2, at points (Gx3,Gz2)\left(\frac{G_x}{3},\frac{G_z}{2}\right) and (2Gx3,Gz2)\left(\frac{2G_x}{3},\frac{G_z}{2}\right) of the 1BZ. There, the bands form approximated gapless Dirac spectra (inset in Figure 4(c)). A band touching point in the 3D parameter space (qx,qz,Λ)(q_x,q_z,\Lambda) plays the role of a dual magnetic monopole.

The corresponding dual magnetic field is a rotation of the three component gauge field An=(Anx,Anz,An,Λ)\mathbf{A}_n = (A_{nx}, A_{nz}, A_{n,\Lambda}), Bn=∇×An\mathbf{B}_n = \nabla\times\mathbf{A}_n, where ∇≡(∂qx,∂qz,∂Λ)\nabla\equiv(\partial_{q_x},\partial_{q_z},\partial_\Lambda) and nn specifies either of the magnonic bands which form the band touching. Following Eq. (25), An,Λ=⟨n(q)∣∂Λ∣n(q)⟩A_{n,\Lambda} = \langle n(q)|\partial_\Lambda|n(q)\rangle. At the band touching point, the dual magnetic field for the respective bands has a dual magnetic charge, whose strength is quantized to be 2π2\pi times an integer[22]. Because the Chern integer cnc_n can be regarded as the total dual magnetic flux penetrating through the constant Λ\Lambda plane, the Gauss theorem implies that when Λ\Lambda goes across the Λ=Λ2\Lambda= \Lambda_2 plane, the Chern integer for the lowest magnonic band c1c_1 changes by unit per each touching point. Hence, due to the two band touchings c1∣Λ>Λ2−c1∣Λ<Λ2=2c_1|_{\Lambda>\Lambda_2} - c_1|_{\Lambda<\Lambda_2} = 2, which explains the exchange of Chern numbers between bands at Λ2\Lambda_2. According to the bulk-edge correspondence, principle[22] the number of in gap one-way edge states is determined by the winding number of a given band nn, that is the sum of all the Chern numbers of the band up to band nn. Consequently, stripes should realize one topological edge mode at each edge. Figure 6(a) and (b), show the amplitude of the magnon Bloch wave function for the modes crossing the band’s gap in stripes made out of 60 chains, at Λ=0.8\Lambda= 0.8 and Λ=1.6\Lambda= 1.6 respectively. The localization of the in-gap modes at the edges of the stripes is apparent in both cases.

Color map with the amplitude of the magnon Bloch wave function of the in gap eigenfrequencies of Figs. 5(a) and (d) at each position of the stripes

Figure 6. Color map with the amplitude of the magnon Bloch wave function of the in gap eigenfrequencies of Figure 5(a) and (Figure 5d) at each position of the stripes. (stripes with Lz=60L_z = 60 rows along the z^\hat{z} axis) at (a) Λ=0.8\Lambda= 0.8 and (b) Λ=1.6\Lambda= 1.6. In both cases, edges states localized at z=0z = 0 and z=Lzz = L_z are apparent. The color code, shown underneath each figure spans from 0 in blue to 1 in red.

Thermomagnetic Hall transport

Upon applying a temperature gradient, the magnon Hall effect MHE allows a transverse heat current mediated by magnons in two dimensions. The MHE was discovered in the ferromagnetic insulator Lu2_2V2_2O7_7[49] and explained in terms of uncompensated magnon edge currents in two dimensions[50, 51]. The relevant quantity characterizing the MHE is the thermal Hall conductivity. Similar to electronic systems, the thermal Hall conductivity is related to the Berry curvature of the eigenstates. The intrinsic contribution to the transverse thermal conductivity

κxz=−kB2T4π2ℏ∑i∫BZθ(ρi)Fixz(q) dq(28)\kappa^{xz} = -\frac{k_{\mathrm{B}}^2T}{4\pi^2\hbar}\sum_i\int_{\mathrm{BZ}}\theta(\rho_i)F_i^{xz}(q)\,dq \tag*{(28)}

is intimately related to the Chern numbers defined in Eq. (23). The sum is over all bands ii in the magnon dispersion, and the integral is over the 1BZ. ρi\rho_i is the Bose distribution function and the function,

θ(x)≡(1+x)(ln⁡1+xx)2−(ln⁡x)2−2Li⁡2(−x)(29)\theta(x) \equiv(1+x)\left(\ln\frac{1+x}{x}\right)^2-(\ln x)^2-2\operatorname{Li}_2(-x) \tag*{(29)}

where Li⁡2\operatorname{Li}_2 is the dilogarithm, TT the temperature, and kBk_{\mathrm{B}} is the Boltzmann constant. The thermal Hall conductivity can be interpreted as the Berry curvature weighed by the θ\theta function[51]. The sign of κxz\kappa^{xz} depends on the topological phase of the bulk system. This dependence can be understood in terms of edge modes and their propagation direction[20]. The topological phases (1,−1)(1,-1) and (−1,1)(-1,1) produce one edge mode. They differ in the slope of their dispersion. In the first case, the nontrivial edge mode propagates to the right, while in the second, it does to the left. The sign of κxz\kappa^{xz}, and therefore the direction of the heat transport in a given topological phase, depends on the occupation probability of the edge magnons. When there is more than one edge mode in the same phase with different slopes, two propagation directions are possible depending on TT. Since κxz\kappa^{xz} is weighted by the function θ\theta, edge modes propagating in different directions may induce cancellation of the transverse thermal conductivity at high energies. If all nontrivial edge modes propagate in the same direction, the sign of the thermal Hall conductivity is fixed within the topological phase, and its sign does not depend on temperature. Here, phase (1,−1)(1,-1) has κxz>0\kappa^{xz}>0, while in phase (−1,1)(-1,1) κxz<0\kappa^{xz}<0. Figure 9 shows the thermal Hall conductivity in terms of temperature for several values of Λ\Lambda. As expected, κxz\kappa^{xz} has opposite sign for lattices with Λ<1.5\Lambda<1.5 and lattices with Λ>1.5\Lambda>1.5.

Thermal conductivity as a function of $k_{\mathrm{B}}T$ at different $\Lambda$

Figure 9. Thermal conductivity as a function of kBTk_{\mathrm{B}}T at different Λ\Lambda. Note the change in the sign of κxz\kappa^{xz} for Λ>1.5\Lambda> 1.5.

Zak phase in infinite chains and topological magnons in open chains.

The strong resemblance between the frequency spectrum of 1D and 2D systems suggests that modulated chains could manifest topological behavior[31]. To examine this possibility, we have computed the Berry phase of 1D systems over a non-contractible loop ℓ\ell of the 1BZ. This invariant is known as the Zak phase[48, 52] and is defined as:

Z=i∮ℓdq⋅An(q)=i∫−Gx/2Gx/2dq⋅An(q)(30)Z = i\oint_{\ell} d\boldsymbol{q}\cdot\boldsymbol{A}_{n}(\boldsymbol{q}) = i\int_{-G_x/2}^{G_x/2} d\boldsymbol{q}\cdot\boldsymbol{A}_{n}(\boldsymbol{q}) \tag*{(30)}

When the 1D system has inversion symmetry a non zero Zak phase indicates that the system is in a topological phase[19]. It is easy to verify, that the magnon hamiltonian in Eq. (20) has an inversion symmetry IH(qx)I−1=H(−qx)\mathcal{I}H(q_x)\mathcal{I}^{-1}=H(-q_x), with respect to the unitary operators I=σ0⊗σx\mathcal{I}=\sigma_0\otimes\sigma_x and I=σx⊗σx\mathcal{I}=\sigma_x\otimes\sigma_x, with σ0\sigma_0 the identity matrix.

Following the approach of Section IV A the gauge field[47] in the 1D lattice reads,

A~x(ql)=ln⁡Ux(ql)(31)\widetilde{A}_x(q_l)=\ln U_x(q_l) \tag*{(31)}
Z~n=i∑lA~x(ql)=i∑lln⁡Ux(ql)\widetilde{Z}_n=i\sum_l\widetilde{A}_x(q_l)=i\sum_l\ln U_x(q_l)

Figure 7(a) shows the result of the Zak number of infinite modulated chains in terms of Λ\Lambda. For modulation amplitudes in the intervals 0<Λ<Λ10<\Lambda<\Lambda_1 and Λ>Λ2\Lambda>\Lambda_2 the Zak phase is zero, while for Λ1<Λ<Λ2\Lambda_1<\Lambda<\Lambda_2 the phase is equal to π\pi and the system is in a topological phase. There is a link between this bulk topological invariant, and the presence of topologically protected end states[22]. Figure 8 shows a color map of the magnon wave function amplitude with largest eigenfrequency in chains with n=100n=100 dipoles at different values of Λ\Lambda. The histogram in each figure shows the multiplicities of the magnon eigenfrequencies of such chains. In the interval Λ1<Λ<Λ2\Lambda_1<\Lambda<\Lambda_2, Figure 8(b) the density map shows well-localized end states. However for Λ<Λ1\Lambda<\Lambda_1 and Λ>Λ2\Lambda>\Lambda_2, Figure 8(a) and (Figure 8c) the magnon mode is delocalized and distributed all over the chain. Consequently we conclude that the chain realizes protected topological end states in the interval Λ1<Λ<Λ2\Lambda_1<\Lambda<\Lambda_2[10].

Zak phase of infinite chains and group velocity of magnonic edge states

Figure 7. (a) Zak phase of infinite chains. Topological phase transitions occur at Λ1=1.0\Lambda_1 = 1.0 and Λ2=1.5\Lambda_2 = 1.5. (b) The group velocity of magnonic edge states ν\nu depicts jump discontinuities at Λ1\Lambda_1 and Λ2\Lambda_2.

Histograms show the multiplicities of the magnon eigenfrequencies of open chains

Figure 8. Histograms show the multiplicities of the magnon eigenfrequencies ω\omega of open chains with n=100n = 100 dipoles. Magnetic stable configurations have the largest frequency in these plots. The color maps underneath each histogram show the respective amplitude of the magnon Bloch wave function of the largest eigenfrequency as a function of the position in the chain (LxL_x denotes the length of the chains). The color code, shown underneath each figure spans from 0 in blue to 1 in red. Figures correspond to chains with (a) Λ=0.5\Lambda= 0.5, (b) Λ=1.3\Lambda= 1.3 and (c) Λ=1.8\Lambda= 1.8. The localization of the wave function at the edges is apparent in (b).

Going back to Figure 3, we note that the quantized jumps in the Zak phase shown in Figure 7(a) coincide with the two band touchings at Λ1\Lambda_1 and Λ2\Lambda_2, shown respectively in Figure 3(b) and Figure 3(d). The previous analysis of the energy spectrum showed that at Λ∼0.7\Lambda\sim0.7, dipoles modify their magnetic equilibrium state from an antiferromagnetic to a dimer configuration (Figure 1(b)). Thus, the band touching in the magnon spectrum and the change in the Zak phase at Λ1\Lambda_1 can be attributed to this fact. Furthermore, the second jump of the Zak number and simultaneous band touchings at Λ2\Lambda_2 coincides with the magnetic transition from the ‘down-down-up’ to the ‘up-down-down-up’ dimerized configurations[53].

Effective model near band touching points.

In 1D the band touching at p1=(q0,Λ1)p_1=(q_0,\Lambda_1) (Figure 3(b)) can be seen as a single Dirac point around which the frequency dispersion for both bands can be approximated by a linear function

ω1,2(p1)∼±ν∣q∣\omega_{1,2}^{(p_1)}\sim\pm\nu|q|

with ν\nu the speed of magnons in the chain. The singular structure of the frequency dispersion near the band touching can be studied using degenerate perturbation theory[36]. For the magnon hamiltonian studied above, it takes the form

Hp=H1+VpH_p=H_1+V_p

with H1=H(p1)H_1=\mathcal{H}(p_1) and Vp=Hp−H1V_p=H_p-H_1. At the touching point, H1H_1 has twofold degenerate eigenstates ∣dj⟩\lvert d_j\rangle (j=1,2j=1,2) with eigenfrequency ω0\omega_0 (>0>0), that satisfy H1∣dj⟩=ω0σz∣dj⟩H_1\lvert d_j\rangle=\omega_0\sigma_z\lvert d_j\rangle[37]. On introducing the perturbation VpV_p, the degeneracy is split into two frequency levels. The eigenstate for the respective eigenfrequency is determined on the zero order of p−p1p-p_1 as

Tp=T1Up+O(∣p−p1∣)T_p=T_1U_p+\mathcal{O}(|p-p_1|)

where the matrix T1T_1 diagonalizes H1H_1 and the unitary matrix UpU_p diagonalizes a 22 by 22 hamiltonian hph_p formed by the twofold degenerate eigenstates,

hp=(d1†Vpd1d1†Vpd2d2†Vpd1d2†Vpd2)h_p= \begin{pmatrix} d_1^\dagger V_p d_1 & d_1^\dagger V_p d_2\\ d_2^\dagger V_p d_1 & d_2^\dagger V_p d_2 \end{pmatrix}

In Fourier space this can be written as

hq=(f1(q,Λ)f2(q,Λ)f3(q,Λ)f4(q,Λ))h_q= \begin{pmatrix} f_1(q,\Lambda) & f_2(q,\Lambda)\\ f_3(q,\Lambda) & f_4(q,\Lambda) \end{pmatrix}

where we find that f1(q,Λ)=−f4(q,Λ)f_1(q,\Lambda)=-f_4(q,\Lambda) and f2(q,Λ)=f3(q,Λ)f_2(q,\Lambda)=f_3(q,\Lambda),[37] Expanding f1(q,Λ)f_1(q,\Lambda) and f2(q,Λ)f_2(q,\Lambda) near p1p_1 yields f1(q,Λ)∼β(Λ−Λ1)f_1(q,\Lambda)\sim\beta(\Lambda-\Lambda_1) and f2(q,Λ)∼−iν(q−q0)f_2(q,\Lambda)\sim-i\nu(q-q_0). Thus

hq∼−ν(q−q0)σx+β(Λ−Λ1)σz(32)h_q\sim-\nu(q-q_0)\sigma_x+\beta(\Lambda-\Lambda_1)\sigma_z \tag*{(32)}

with constants β>0\beta>0 and ν>0\nu>0[37]. Near the band touching point the effective hamiltonian becomes

Heff=ω0σ0−iν(Λ)(q−q0)σx+mσz(33)\mathcal{H}_{\mathrm{eff}}=\omega_0\sigma_0-i\nu(\Lambda)(q-q_0)\sigma_x+m\sigma_z \tag*{(33)}

We identify the mass term m=β(Λ−Λ1)m=\beta(\Lambda-\Lambda_1) which cancels out at Λ=Λ1\Lambda=\Lambda_1 at the band crossing point. In the presence of the mass term, the spectrum becomes gapped

ω1,2∼±(νq)2+m2\omega_{1,2}\sim\pm\sqrt{(\nu q)^2+m^2}

Figure 7(b) shows the magnon speed, (spin wave group velocity) ν=ωq\nu=\frac{\omega}{q} versus Λ\Lambda around the Dirac point. The jump discontinuities around Λ1\Lambda_1 and Λ2\Lambda_2 coincide with the dimerization of the chain, and the limits of the topological phase as shown in Figure 7(a). This result suggests that the Fermi speed of magnons could be interpreted as susceptibility to Λ\Lambda. Its discontinuity would indicate a topological phase transition.

Discussion

The purpose of this work has been to demonstrate the explicit control of energy bands, magnonic frequencies, and edges modes in arrays of magnetic dipoles. These tunable systems are made out of stacked modulated dipolar chains that interact via magnetic dipolar coupling. When the easy axis anisotropy is strong enough, dipoles settle into parallel magnetic configurations, with gapped 1D and 2D magnonic bands, which are highly tunable by Λ\Lambda, the amplitude of the periodic modulation along single chains. The tuning of Λ\Lambda in single chains, gives rise to two topological phase transitions with non-zero Zak phase, which host chiral edge states. In 2D lattices made of modulated chains, the spin-wave volume bands take non-zero Chern numbers 1 and −1-1, respectively. These values are exchanged at Λ=32\Lambda=\frac{3}{2} due to two band touchings that yield two monopoles with charge +1+1 each. Due to the monopoles, the Berry phase acquires divergence, which triggers the exchange of the Chern numbers between the bands. This topological phase transition is manifested in the Hall conductivity of lattices which changes sign at Λ=32\Lambda=\frac{3}{2}.

We study an effective model for the 1D systems that expose the chiral character of the speed of edge magnons ν\nu. The jump discontinuities in the response of ν\nu to changes in Λ\Lambda, coincide with the two topological phase transitions in chains. Hence, it is proposed that discontinuities in the response function of the magnons Fermi speed could serve as an indicator of topological phase transitions.

The influence of phonons in the band of magnons have been neglected in this study. At finite temperatures, phonons could contribute to the thermal Hall current due to the phonon Hall effect[54]. However, it has been argued that this contribution is negligible compared to the magnon Hall current because phonon angular momenta are proportional to the gyromagnetic ratio of charged atoms, which is orders of magnitude smaller than that of magnons[31]. In the presence of magnon-phonon coupling, the two types of quasiparticles could hybridize near band crossing. Because in two-dimensional materials, thermal agitation would favor the out-of-plane vibrations over the in-plane ones, we expect that the dominant contributions would stem from out-of-plane phonon modes. In our system, the equilibrium magnetic configurations lie in the plane of the sample, and therefore the coupling between magnon and phonon modes should be relatively small. Evidence[31, 55] indicates that in cases when magnon phonon hybridization is realized, it slightly affects the magnon bands in regions of avoided band crossing and that unless the magnon phonon coupling becomes significant, it does not influence the regions of the band where Berry curvature is concentrated.

The current approach for magnon manipulation leaves out the intervention of external fields and relies instead on tuning a single intrinsic geometrical parameter. The calibration of Λ\Lambda tunes the internal anisotropic magnetic fields in the lattices. Such internal fields originate from dipolar interactions between magnets and produce a spin-momentum locking, which constrains the dipole’s magnetic moment orientation, as happens in systems that allow the spin-orbit interaction. The control over these internal fields is accomplished by changing the lattice constant of the system along one single direction, Λ\Lambda. A thrilling consequence of increasing Λ\Lambda is the flattening of the lower magnon band. The effect of the band flattening on thermal conductivity has not been addressed here, but we believe deserves special attention. With an intrinsic knob being able to tune the flatness of the spin wavebands, the present dipolar system could open up a new avenue to study correlation-driven emergent phenomena on the background of topological magnons.

Possible realizations of the systems presented here could be accomplished by means of molecular magnets[56, 57], optical lattices[52] and nanomagnetic arrays made out of permalloy[9]. Especially suitable are state-of-the-art magnonic crystals fabricated out of epitaxially grown YIG films[58]. Recent work has shown that thickness and width modulated magnonic crystals comprising longitudinally magnetized periodically structured YIG-film waveguides can manipulate magnonic gaps with advantages such as a small magnetic damping and high group velocity of the spin waves[59].

Acknowledgments

We thank Fondecyt under Grant No. 1210083.

References

References

  1. [1]Y. Cao, V. Fatemi, S. Fang, K. Watanabe, T. Taniguchi, E. Kaxiras, and P. Jarillo-Herrero, Nature 556, 43 (2018).
  2. [2]K. Hejazi, Z.-X. Luo, and L. Balents, Proceedings of the National Academy of Sciences 117, 10721 (2020).
  3. [3]J. Hirsch, Physical review letters 83, 1834 (1999).
  4. [4]Q. Tong, F. Liu, J. Xiao, and W. Yao, Nano letters 18, 7194 (2018).
  5. [5]Y.-H. Li and R. Cheng, Physical Review B 102, 094404 (2020).
  6. [6]J. Van Kranendonk and J. Van Vleck, Reviews of Modern Physics 30, 1 (1958).
  7. [7]L. Cornelissen, J. Liu, R. Duine, J. B. Youssef, and B. Van Wees, Nature Physics 11, 1022 (2015).
  8. [8]A. V. Chumak, V. I. Vasyuchka, A. A. Serga, and B. Hillebrands, Nature Physics 11, 453 (2015).
  9. [9]J. C. Gartside, S. G. Jung, S. Y. Yoo, D. M. Arroо, A. Vanstone, T. Dion, K. D. Stenning, and W. R. Branford, Communications Physics 3, 1 (2020).
  10. [10]X. Wang, H. Zhang, and X. Wang, Physical Review Applied 9, 024029 (2018).
  11. [11]M. Krawczyk and D. Grundler, Journal of physics: Condensed matter 26, 123202 (2014).
  12. [12]Y. Fan, P. Quarterman, J. Finley, J. Han, P. Zhang, J. T. Hou, M. D. Stiles, A. J. Grutter, and L. Liu, Physical Review Applied 13, 061002 (2020).
  13. [13]F. Pirmoradian, B. Z. Rameshti, M. Miri, and S. Saeidian, Physical Review B 98, 224409 (2018).
  14. [14]E. Iacocca, S. Gliga, R. L. Stamps, and O. Heinonen, Physical Review B 93, 134420 (2016).
  15. [15]J. Zang, M. Mostovoy, J. H. Han, and N. Nagaosa, Physical review letters 107, 136804 (2011).
  16. [16]S. A. Díaz, J. Klinovaja, and D. Loss, Physical review letters 122, 187203 (2019).
  17. [17]S. A. Díaz, T. Hirosawa, J. Klinovaja, and D. Loss, Physical Review Research 2, 013231 (2020).
  18. [18]S. K. Kim, K. Nakata, D. Loss, and Y. Tserkovnyak, Physical review letters 122, 057204 (2019).
  19. [19]L. Zhang, J. Ren, J.-S. Wang, and B. Li, Physical Review B 87, 144101 (2013).
  20. [20]A. Mook, J. Henk, and I. Mertig, Physical Review B 90, 024412 (2014).
  21. [21]R. Chisnell, J. Helton, D. Freedman, D. Singh, R. Bewley, D. Nocera, and Y. Lee, Physical review letters 115, 147201 (2015).
  22. [22]S. M. Girvin and K. Yang, Modern condensed matter physics (Cambridge University Press, 2019).DOI
  23. [23]Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom, science 306, 1910 (2004).
  24. [24]N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Reviews of modern physics 82, 1539 (2010).
  25. [25]H. Katsura, N. Nagaosa, and P. A. Lee, Physical review letters 104, 066403 (2010).
  26. [26]M. Li, Q. Wang, G. Wang, Z. Yuan, W. Song, R. Lou, Z. Liu, Y. Huang, Z. Liu, H. Lei, et al., Nature communications 12, 1 (2021).
  27. [27]K. H. Lee, S. B. Chung, K. Park, and J.-G. Park, Physical Review B 97, 180401 (2018).
  28. [28]S. Owerrer, Journal of Physics: Condensed Matter 29, 385801 (2017).
  29. [29]M. J. Park, S. Lee, and Y. B. Kim, Phys. Rev. B 104, L060401 (2021).
  30. [30]R. Shindou, R. Matsumoto, S. Murakami, and J.-i. Ohe, Physical Review B 87, 174427 (2013).
  31. [31]Y.-H. Li and R. Cheng, Physical Review B 103, 014407 (2021).
  32. [32]P. Nikolić, Physical Review B 102, 075131 (2020).
  33. [33]S. K. Kim, H. Ochoa, R. Zarzuela, and Y. Tserkovnyak, Physical review letters 117, 227201 (2016).
  34. [34]J. G. Rau, E. K.-H. Lee, and H.-Y. Kee, Annual Review of Condensed Matter Physics 7, 195 (2016).
  35. [35]J. Liu, L. Wang, and K. Shen, Physical Review Research 2, 023282 (2020).
  36. [36]R. Shindou, J.-i. Ohe, R. Matsumoto, S. Murakami, and E. Saitoh, Physical Review B 87, 174402 (2013).
  37. [37]See supplemental material for details.
  38. [38]J. Cisternas, P. Mellado, F. Urbina, C. Portilla, M. Carrasco, and A. Concha, Physical Review B 103, 134443 (2021).
  39. [39]M. Lakshmanan, Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 369, 1280 (2011).
  40. [40]S. Osokin, A. Safin, Y. Barabanenkov, and S. Nikitov, Journal of Magnetism and Magnetic Materials 465, 519 (2018).
  41. [41]A. Y. Galkin, B. Ivanov, and C. Zaspel, Journal of magnetism and magnetic materials 286, 351 (2005).
  42. [42]P. Bondarenko, A. Y. Galkin, B. Ivanov, and C. Zaspel, Physical Review B 81, 224415 (2010).
  43. [43]R. Verba, G. Melkov, V. Tiberkevich, and A. Slavin, Physical Review B 85, 014427 (2012).
  44. [44]I. Lisenkov, V. Tiberkevych, A. Slavin, P. Bondarenko, B. A. Ivanov, E. Bankowski, T. Meitzler, and S. Nikitov, Physical Review B 90, 104417 (2014).
  45. [45]I. Lisenkov, V. Tiberkevych, S. Nikitov, and A. Slavin, Physical Review B 93, 144401 (2016).
  46. [46]D. Peter, N. Y. Yao, N. Lang, S. D. Huber, M. D. Lukin, and H. P. Büchler, Physical Review A 91, 053617 (2015).
  47. [47]T. Fukui, Y. Hatsugai, and H. Suzuki, Journal of the Physical Society of Japan 74, 1674 (2005).
  48. [48]J. Zak, Physical review letters 62, 2747 (1989).
  49. [49]Y. Onose, T. Ideue, H. Katsura, Y. Shioni, N. Nagaosa, and Y. Tokura, Science 329, 297 (2010).
  50. [50]R. Matsumoto and S. Murakami, Physical review letters 106, 197202 (2011).
  51. [51]R. Matsumoto and S. Murakami, Physical Review B 84, 184406 (2011).
  52. [52]M. Atala, M. Aidelsburger, J. T. Barreiro, D. Abanin, T. Kitagawa, E. Demler, and I. Bloch, Nature Physics 9, 795 (2013).
  53. [53]M. Martí-Sabaté and D. Torrent, arXiv preprint arXiv:2107.10144 (2021).
  54. [54]L. Zhang, J. Ren, J.-S. Wang, and B. Li, Physical review letters 105, 225901 (2010).
  55. [55]S. Park and B.-J. Yang, Physical Review B 99, 174435 (2019).
  56. [56]L. Bogani and W. Wernsdorfer, in Nanoscience and technology: a collection of reviews from nature journals (World Scientific, 2010), pp. 194–201.
  57. [57]S. V. Syzranov, M. L. Wall, V. Gurarie, and A. M. Rey, Nature communications 5, 1 (2014).
  58. [58]P. Frey, A. A. Nikitin, D. A. Bozhko, S. A. Bunyaev, G. N. Kakazei, A. B. Ustinov, B. A. Kalinikos, F. Ciubotaru, A. V. Chumak, Q. Wang, et al., Communications Physics 3, 1 (2020).
  59. [59]59 L. Mihalceanu, V. I. Vasyuchka, D. A. Bozhko, T. Langner, A. Y. Nechiporuk, V. F. Romanyuk, B. Hillebrands, and A. A. Serga, Physical Review B 97, 214405 (2018).

Paper details

vlnc:W_ygr4wjfk
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.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents