betaFeedback

WorksPhysics and Astronomy

Spin Inertia as a Source of Topological Magnons: Chiral Edge States from Coupled Precession and Nutation

Abstract

Spin inertia has been demonstrated to give rise to high-frequency nutational excitations beyond the conventional low-frequency precessional modes. Here, we demonstrate that the hybridization between precessional and nutational magnons may give rise to topological phenomena in the spin-wave spectrum. This hybridization requires the presence of interactions breaking angular-momentum conservation, such as the pseudodipolar interaction. We show on the example of a honeycomb ferromagnet how topological gaps open between the precessional and nutational bands that host chiral edge states in slab geometries. Our work establishes a theoretical foundation for exploring inertial spin dynamics as a new route to engineer topological phases in magnetic materials.

The rapid development in the generation of ultrashort laser pulses over the last decades has stimulated studies of magnetization dynamics on femto- to picosecond time scales [1, 2, 3]. It has been proposed theoretically that the conventional Landau–Lifshitz–Gilbert (LLG) framework, which accounts for precession and damping, becomes incomplete at these time scales, and has to be extended by an inertial term [4, 5, 6, 7, 8]. This spin inertia induces a separation of the directions of the magnetic moment and the angular momentum, causing the former to perform a nutational motion around the latter. In ferromagnetic resonance spectra, spin inertia manifests as a secondary resonance peak appearing typically at terahertz frequencies [9], which has been recently identified experimentally [10, 11, 12]. Beyond modifying the linear response, inertial effects also influence spin transport [13], magnetization switching [14], auto-oscillations [15], and the spin-wave dispersion [16, 17, 18, 19]. In the latter case, spin inertia results in the doubling of the excitation bands by giving rise to high-frequency nutational spin waves.

The topology of the band structure of spin waves or magnons has also attracted considerable attention recently [20]. Topological magnon insulators with chiral edge modes have been proposed theoretically in multi-sublattice ferromagnets [21, 22, 23, 24], magnon crystals [25] and skyrmion lattices [26, 27, 28], and magnonic Weyl nodes [29, 30] have also been identified. The topology of the band structure also contributes to transport phenomena such as the magnon thermal Hall effect [31, 32, 33, 34]. A prototypical example of a topological magnon insulator is a ferromagnet on a honeycomb lattice, realized in, e.g., chromium trihalides and Cr(Si/Ge)Te3_3 compounds [35, 36, 37]. Two different microscopic mechanisms have been put forward that could explain the topologically non-trivial nature of the band structure in this geometry: the Dzyaloshinsky–Moriya interaction [35, 36, 38, 39] which is antisymmetric in the lattice sites, and various forms of the symmetric two-site exchange anisotropy including dipolar, pseudodipolar, and Kitaev terms [40, 41, 42]. It has been suggested that the Dzyaloshinsky–Moriya interaction may be more prominent in the opening of a gap in the band structure [43], since the gap size scales linearly with its strength but quadratically with the pseudodipolar term. Due to their different symmetry properties, it was also proposed that the two contributions may be distinguished by measuring the magnon spectrum for different orientations of the magnetization [44].

Here, we present how spin inertia influences the topology of the magnon band structure. As an example, we demonstrate that the precessional and nutational magnon bands may hybridize in a honeycomb ferromagnet, leading to an additional gap opening in the spectrum. This hybridization is enabled by the pseudodipolar interaction but not by the Dzyaloshinsky–Moriya interaction, enabling to distinguish between these two contributions, and the gap size is shown to scale linearly with the interaction strength. We confirm the topological nature of this gap by demonstrating the emergence of chiral edge modes in slab geometries, and connect this to the Chern numbers of the bands.

The dynamics of the spins is described by the inertial Landau-Lifshitz-Gilbert (iLLG) equation [4],

S˙i=Si×[−γBieff+αS˙i+ηS¨i].(1)\dot{\mathbf{S}}_i=\mathbf{S}_i\times\left[-\gamma\mathbf{B}^{\mathrm{eff}}_i+\alpha\dot{\mathbf{S}}_i+\eta\ddot{\mathbf{S}}_i\right]. \tag*{(1)}

Here, SiS_i denotes a unit vector at site ii corresponding to the direction of the magnetic moment, γ\gamma is the gyromagnetic ratio, α\alpha represents the Gilbert damping constant, and η\eta is the inertial relaxation time. The effective magnetic field is expressed as Bieff=−1Mi∂H∂Si\mathbf{B}^{\mathrm{eff}}_i=-\frac{1}{M_i}\frac{\partial\mathcal{H}}{\partial\mathbf{S}_i}, where H\mathcal{H} is the Hamiltonian of the system and MiM_i is the magnetic moment. The angular momentum at site ii may be defined as Li=Miγ(Si−ηSi×S˙i)L_i = \frac{M_i}{\gamma}\left(S_i-\eta S_i\times\dot{S}_i\right) [8], see the Supplemental Material (SM) [45]. In the absence of an external magnetic field and spin–orbit coupling, the system is invariant under global rotations, which leads to the conservation of the total angular momentum ∑iLi\sum_i L_i.

To calculate the magnon excitations within linear spin-wave theory, first an equilibrium spin configuration is sought which satisfies Si(0)×Bieff=0S_i^{(0)}\times B_i^{\mathrm{eff}}=0 at all lattice sites. Expanding Eq. (1) in small spin deviations Si⊥=[Si−,Si+]=[S~ix−iS~iy,S~ix+iS~iy]S_i^\perp=[S_i^-,S_i^+]=[\widetilde{S}_i^x-i\widetilde{S}_i^y,\widetilde{S}_i^x+i\widetilde{S}_i^y], where in equilibrium S~iz=1\widetilde{S}_i^z=1 at all sites in the local frames of reference, yields (cf. Ref. [18])

ω[S⊥V⊥]=[01γηM−1HSWσzη+iαη1][S⊥V⊥]=D[S⊥V⊥](2)\omega \begin{bmatrix} S^\perp\\ V^\perp \end{bmatrix} = \begin{bmatrix} 0 & 1 \\ \frac{\gamma}{\eta}M^{-1}\mathcal{H}_{\mathrm{SW}} & \frac{\sigma^z}{\eta}+i\frac{\alpha}{\eta}1 \end{bmatrix} \begin{bmatrix} S^\perp\\ V^\perp \end{bmatrix} = \mathcal{D} \begin{bmatrix} S^\perp\\ V^\perp \end{bmatrix} \tag*{(2)}

where V⊥=−iS˙⊥V^\perp=-i\dot{S}^\perp is the spin velocity, HSW\mathcal{H}_{\mathrm{SW}} is the spin-wave Hamiltonian, 11 is the unit matrix, MM is a diagonal matrix containing the magnetic moments, and σz\sigma^z acts as a Pauli matrix on the two components of S⊥S^\perp. We substituted the harmonic time dependence S⊥∝eiωtS^\perp\propto e^{i\omega t} and replaced the time derivatives with multiplication by iωi\omega. The spin waves correspond to the eigenmodes of the matrix D\mathcal{D} in Eq. (2). The magnon frequencies can also be determined by solving the secular equation

det⁡(−ηω21+ωσz+γM−1HSW)=0.(3)\det\left(-\eta\omega^2 1+\omega\sigma^z+\gamma M^{-1}\mathcal{H}_{\mathrm{SW}}\right)=0. \tag*{(3)}

If the calculations are performed around a stable equilibrium state, HSW\mathcal{H}_{\mathrm{SW}} is a positive definite operator, and the eigenvalue problem is quasi-Hermitian [46]. This means that all eigenvalues are real, but the left ⟨lq∣\langle l_q| and right ∣rq⟩|r_q\rangle eigenvectors belonging to the same eigenvalue are not adjoints of each other, but form a biorthogonal system instead, ⟨lq∣rq′⟩=δqq′\langle l_q|r_{q'}\rangle=\delta_{qq'}. Furthermore, the eigenvalues are found in ±ωq\pm\omega_q pairs due to the particle-hole symmetry, which is represented as C=diag⁡(σx,−σx)KC=\operatorname{diag}(\sigma^x,-\sigma^x)K in this system, where KK denotes complex conjugation. All eigenvectors have the form ∣rq⟩=[vq,ωqvq]|r_q\rangle=[v_q,\omega_qv_q], since the latter components are connected to the spin velocity.

In the following, we consider a ferromagnetic system magnetized along the global zz direction, and describe the undamped modes for α=0\alpha=0. The rotational symmetry around the zz axis is preserved in the absence of spin–orbit coupling, and the zz component of the total angular momentum, ∑iLiz\sum_i L_i^z, remains a conserved quantity. On the level of spin-wave theory, this is reflected by the absence of pairing terms in the Hamiltonian HSW\mathcal{H}_{\mathrm{SW}}, which would couple the first and second components of S⊥S^\perp together. Considering only the Si+S_i^+ sector, two types of excitations may be distinguished in inertial dynamics [47]: precessional excitations performing a counterclockwise rotation around the direction of the total angular momentum, which have ω>0\omega>0; and nutational excitations performing a clockwise rotation with ω<0\omega<0.

These are illustrated in Figure 1. Due to the different rotational senses, the angular momenta of the excited states also differ: ∑iLiz>∑iMiγSiz\sum_i L_i^z>\sum_i\frac{M_i}{\gamma}S_i^z for precessional and ∑iLiz<∑iMiγSiz\sum_i L_i^z<\sum_i\frac{M_i}{\gamma}S_i^z for nutational motion. To observe topological edge modes connecting the precessional and nutational magnon bands, this conservation of the total angular momentum has to be broken.

Sketch of the ferromagnet on the honeycomb lattice, with red and blue arrows showing the two sub-lattices and trajectories of precessional and nutational modes

Figure 1. Sketch of the ferromagnet on the honeycomb lattice, with red and blue arrows showing the two sub-lattices. Red and blue trajectories denote counterclockwise precessional and clockwise nutational modes, respectively.

As an example, we consider a ferromagnet on the honeycomb lattice. The Hamiltonian is given by

H=∑⟨i,j⟩[SiAJijSjB+F(SiA⋅r^ij)(SjB⋅r^ij)].(4)\mathcal{H}=\sum_{\langle i,j\rangle}\left[S_i^A J_{ij}S_j^B+F\left(S_i^A\cdot\hat{r}_{ij}\right)\left(S_j^B\cdot\hat{r}_{ij}\right)\right]. \tag*{(4)}

where AA and BB indicate the two sublattices denoted by red and blue spins in Figure 1, respectively. The exchange interaction tensor Jij\mathbf{J}_{ij} between nearest neighbors is diagonal here, with Jzz<Jxx=Jyy<0J^{zz}<J^{xx}=J^{yy}<0 preferring magnetization along z^\hat{z}. The pseudodipolar interaction FF connects the spin directions to the relative positions of the sites via the unit vector r^ij\hat{r}_{ij} connecting sites ii and jj, which is either δ^1=(−1,0)\hat{\delta}_1=(-1,0), δ^2=(12,−32)\hat{\delta}_2=\left(\frac{1}{2},-\frac{\sqrt{3}}{2}\right), or δ^3=(12,32)\hat{\delta}_3=\left(\frac{1}{2},\frac{\sqrt{3}}{2}\right). The parameter values used in the simulations are summarized in Table I. The exchange interaction Jxx=−3 meVJ^{xx}=-3\ \mathrm{meV} and the magnetic moments MA=MB=3μBM_A=M_B=3\mu_{\mathrm{B}} are comparable to experimentally determined values for honeycomb van der Waals materials [35, 36, 37, 48]. The inertial relaxation time was set to η=330 fs\eta=330\ \mathrm{fs}, which is similar to the values determined experimentally in transition metals [8, 10].

JxxJ^{xx} (meV)JzzJ^{zz} (meV)FF (meV)η\eta (fs)MA=MBM_A=M_B (μB\mu_{\mathrm{B}})
−3.0-3.0−3.3-3.31.51.533033033

Table I. Dimensionless model parameters used throughout the text.

The spin-wave Hamiltonian, after spatial Fourier transformation SA/B,⊥(k)=N−1/2∑i/je−ikRi/jSi/jA/B,⊥S^{A/B,\perp}(\boldsymbol{k}) = N^{-1/2}\sum_{i/j}e^{-i\boldsymbol{k}R_{i/j}}S^{A/B,\perp}_{i/j}, may be expressed as

HSW,k=(ΩH10F~1−iF~2H1∗ΩF~1∗−iF~2∗00F~1+iF~2ΩH1F~1∗+iF~2∗0H1∗Ω)(5)\mathcal{H}_{\mathrm{SW},\boldsymbol{k}} = \begin{pmatrix} \Omega& \mathcal{H}_{1} & 0 & \widetilde{F}_{1}-i\widetilde{F}_{2} \\ \mathcal{H}_{1}^{*} & \Omega& \widetilde{F}_{1}^{*}-i\widetilde{F}_{2}^{*} & 0 \\ 0 & \widetilde{F}_{1}+i\widetilde{F}_{2} & \Omega& \mathcal{H}_{1} \\ \widetilde{F}_{1}^{*}+i\widetilde{F}_{2}^{*} & 0 & \mathcal{H}_{1}^{*} & \Omega \end{pmatrix} \tag*{(5)}

where the components are written in the order [S−A,S−B,S+A,S+B][S_{-}^{A},S_{-}^{B},S_{+}^{A},S_{+}^{B}]. The matrix elements are given by

Ω=−3Jzz,(6)\Omega=-3J^{zz}, \tag*{(6)}
H1(k)=(Jxx+F2)(eikδ1+eikδ2+eikδ3),(7)\mathcal{H}_{1}(\boldsymbol{k})=\left(J^{xx}+\frac{F}{2}\right)\left(e^{i\boldsymbol{k}\boldsymbol{\delta}_{1}}+e^{i\boldsymbol{k}\boldsymbol{\delta}_{2}}+e^{i\boldsymbol{k}\boldsymbol{\delta}_{3}}\right), \tag*{(7)}
F~1(k)=F4(2eikδ1−eikδ2−eikδ3),(8)\widetilde{F}_{1}(\boldsymbol{k})=\frac{F}{4}\left(2e^{i\boldsymbol{k}\boldsymbol{\delta}_{1}}-e^{i\boldsymbol{k}\boldsymbol{\delta}_{2}}-e^{i\boldsymbol{k}\boldsymbol{\delta}_{3}}\right), \tag*{(8)}
F~2(k)=3F4(eikδ3−eikδ2).(9)\widetilde{F}_{2}(\boldsymbol{k})=\frac{\sqrt{3}F}{4}\left(e^{i\boldsymbol{k}\boldsymbol{\delta}_{3}}-e^{i\boldsymbol{k}\boldsymbol{\delta}_{2}}\right). \tag*{(9)}

The calculated magnon spectrum is shown in Figure 2 along the high-symmetry lines of the Brillouin zone. The exchange interaction forms two precessional bands in this two-sublattice structure, which touch in the KK and K′K^{\prime} points forming Dirac cones. Spin inertia leads to additional nutational modes, which have the same dispersion apart from being shifted up in frequency by η−1\eta^{-1} [17]; see the SM [45]. For sufficiently high values of ηγM−1∣Jxx∣\eta\gamma M^{-1}|J^{xx}|, first the top of the higher precessional band touches the bottom of the lower nutational band in the Γ\Gamma point, then a nodal line is formed between the bands along a contour encompassing the Γ\Gamma point. The pseudodipolar interaction FF breaks the continuous rotational symmetry around z^\hat{z}, thereby causing a hybridization between the precessional and nutational modes and opening gaps in the spectrum. In the Dirac points, the gap is opened for any finite value of FF, but its size scales with ∣F∣2|F|^{2}, making it difficult to identify in Figure 2; see the SM [45]. Along the nodal line, the gap size scales linearly with ∣F∣|F|, but this effect can only be observed if the inertial relaxation time is sufficiently high such that the bands overlap. The hybridization between the nutational and precessional bands may be captured by calculating the ellipticity of mode nn at wave vector k\boldsymbol{k},

Magnon band structure in momentum space

Figure 2. Magnon band structure in momentum space along high-symmetry points. The states are colored according to their ellipticity defined in Eq. (10).

en,k=2⟨ln,k∣P−∣rn,k⟩−1,(10)e_{n,\boldsymbol{k}}=2\langle l_{n,\boldsymbol{k}|}P_{-}|r_{n,\boldsymbol{k}}\rangle-1, \tag*{(10)}

where P−=diag⁡(1,1,0,0,1,1,0,0)P_{-}=\operatorname{diag}(1,1,0,0,1,1,0,0) is the projection onto the clockwise rotating S−−V−S_{-}-V_{-} subspace. The ellipticity takes the value of 1 for left circularly polarized, typically nutational, modes, and −1-1 for right circularly polarized, typically precessional, modes, while it vanishes for linearly polarized modes forming where the degeneracy along the nodal line is lifted. Alternatively, en,ke_{n,\boldsymbol{k}} may be thought of as the angular momentum carried by the mode, as discussed above. The avoided crossing may also be detected by studying the polarizations of the modes only in the high-symmetry points: in the Γ\Gamma point where the right-handed and left-handed modes are eigenstates, the sign of the ellipticity is alternating with increasing energy, while in the MM and KK points two predominantly right-handed modes are followed by two mainly left-handed modes; see the Supplemental Movies [45].

The gaps opened in the KK and K′K^{\prime} points have been demonstrated to be topologically non-trivial in this model [40], and this may also be the case for the gap between the precessional and nutational bands. A typical signature of topologically non-trivial gaps is the presence of chiral edge modes at the boundaries [23]. To confirm this bulk-edge correspondence, we recalculated the magnon spectrum in a slab geometry, as shown in Figure 3. The edge modes are identified by calculating the localization

Figure 3. Magnon band structure in a slab geometry. The system is 40 unit cells wide along the xx direction with open boundary conditions, while it is infinite along the yy direction with periodic boundary conditions, meaning that the eigenstates can be indexed by the wave vector component kyk_y. aa denotes the distance between nearest-neighbor atoms in the lattice, i.e., the lattice constant is 3a\sqrt{3}a. Dotted lines show the boundaries of the one-dimensional Brillouin zone. Coloring shows the localization defined in Eq. (11).

ℓn,k=⟨ln,k∣PLeft∣rn,k⟩−⟨ln,k∣PRight∣rn,k⟩,(11)\ell_{n,\boldsymbol{k}}=\langle l_{n,\boldsymbol{k}}|P_{\mathrm{Left}}|r_{n,\boldsymbol{k}}\rangle-\langle l_{n,\boldsymbol{k}}|P_{\mathrm{Right}}|r_{n,\boldsymbol{k}}\rangle, \tag*{(11)}

where PLeftP_{\mathrm{Left}} and PRightP_{\mathrm{Right}} are projections onto the sites on the left and right half of the slab, respectively. Close to the Dirac cones where the first and second, respectively the third and fourth bands almost touch, there is a single edge mode crossing the gap. These modes are localized at opposite edges of the system, and propagate in opposite directions based on their group velocity. Chiral edge modes may also be observed between the second and third bands where the avoided crossing between the precessional and nutational bands is found in Figure 2. While there is a single mode localized on each edge in this gap as well, note that the group velocity changes sign along these modes, with the edge modes having a frequency maximum at the edge of the one-dimensional Brillouin zone. This indicates that the gap between any adjacent pairs of bands is topologically non-trivial. Note that a pair of edge modes is also observable below the lowest magnon band. This is due to the reduced number of neighbors at the edge, which destabilizes the ferromagnetic alignment along the zz direction for higher values of the pseudodipolar interaction than what is considered here. However, these edge modes are of topologically trivial origin, and do not merge with the bulk bands as the wave vector varies.

To confirm the topologically non-trivial origin of the edge states, we calculate the topological invariants of the bands in the infinite system, corresponding to the Chern number in the considered case. The Chern number is the integral of the zz component of the Berry curvature Cn=(2π)−1∫Bnz(k) d2kC_n = (2\pi)^{-1}\int B_n^z(\mathbf{k})\,\mathrm{d}^2k, where the latter may be expressed with the left and right eigenvectors as (see the SM [45] for details)

Bnα(k)=iϵαβγ∑m≠n⟨ln,k∣∂kβDk∣rm,k⟩⟨lm,k∣∂kγDk∣rn,k⟩(ωn,k−ωm,k)2.(12)B_n^\alpha(\mathbf{k}) = \mathrm{i}\epsilon^{\alpha\beta\gamma}\sum_{m\ne n}\frac{\langle l_{n,\mathbf{k}}|\partial_{k_\beta}\mathcal{D}_{\mathbf{k}}|r_{m,\mathbf{k}}\rangle\langle l_{m,\mathbf{k}}|\partial_{k_\gamma}\mathcal{D}_{\mathbf{k}}|r_{n,\mathbf{k}}\rangle}{(\omega_{n,\mathbf{k}}-\omega_{m,\mathbf{k}})^2}. \tag*{(12)}

Here, Dk\mathcal{D}_{\mathbf{k}} is the matrix on the right-hand side of Eq. (2) in Fourier space, and α\alpha, β\beta, γ\gamma denote Cartesian indices. The distribution of BnzB_n^z for the four bands is shown in Figure 4. The sum of the Chern numbers ∣∑i=1nCi∣\left|\sum_{i=1}^{n}C_i\right| predicts the number of pair of chiral edge modes between bands nn and n+1n+1, which equals 1 between any two bands, in agreement with Figure 3. The Chern number C1=−1C_1=-1 of the lowest band in Figure 4(a) stems from the peaks in the Berry curvature close to the KK and K′K' points where the degeneracy is lifted.

Berry curvature of the four magnon bands

Figure 4. Berry curvature Bnz(k)B_n^z(\mathbf{k}) of the bands calculated from Eq. (12). (a)-(d) show the four magnon bands in increasing order of frequency. The Chern numbers CnC_n for each band are also given.

These peaks are also observable in the second band in Figure 4(b) with opposite sign, but they are cancelled out by the circular depression around the contour where the precessional and nutational bands overlap in the absence of spin–orbit coupling. The Berry curvatures in the third and fourth bands are qualitatively similar to the second and first bands, respectively, although with reversed signs.

In summary, we demonstrated that the hybridization between nutational and precessional spin waves in inertial spin dynamics may modify the topological properties of the bands. We illustrated this on the example of a honeycomb ferromagnet, which realizes a magnonic Chern insulator with an avoided crossing between the precessional and nutational bands in the bulk system, which is filled by chiral edge modes in a slab geometry.

Note that the width of the magnon bands is lower in the calculations than what was experimentally determined in van der Waals magnets [35, 36, 37, 48]; this is in part caused by the compression of the bands by the inertial term [17], as shown in the SM [45]. This means that the value of the exchange interaction may be higher than what was determined from the non-inertial model, and since the topological transition is governed by the parameter γM−1Jxzη\gamma M^{-1}J^{xz}\eta, it may be observed for lower inertial relaxation times than the value used in this work. The inertial relaxation time may be further enhanced in the presence of heavy elements with high spin–orbit coupling in these materials. Therefore, the proposed effects may be experimentally accessible by measuring the magnon spectrum in the whole Brillouin zone up to higher energies where the nutational bands are expected to emerge.

The nutational and precessional bands may be distinguished based on their different angular momentum, or opposite rotational sense. This means that methods sensitive to the rotational sense of the excitations may be particularly suitable for this purpose, such as spin-polarized electron-energy loss spectroscopy. Optical methods may also be applied in the vicinity of the Γ\Gamma point where the reversed rotational sense of the higher modes already provides an indication for the topologically non-trivial nature of the bands. Detecting the inertial dynamics may also shed further light on the role of the competing mechanisms of the Dzyaloshinsky–Moriya interaction and the pseudodipolar interaction in the gap opening in these systems [36, 44]: while the pseudodipolar interaction may open a gap between the precessional and nutational bands, the out-of-plane component of the Dzyaloshinsky–Moriya interaction has no such effect since it preserves the angular momenta of the excitations.

These calculations may be generalized to other systems, for example single-sublattice ferromagnets, where topologically non-trivial bands may only be formed if inertial effects are taken into account; see the SM [45] for an example. The generation of bulk and topological edge modes in ferromagnets at THz frequencies may also open further perspectives in controlling the angular-momentum transfer between magnons and phonons or electrons.

Acknowledgments

A.M. acknowledges funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through TRR 173-268565370 (Project B13), and Project No. 504261060 (Emmy Noether Programme), and from the Dynamics and Topology Center (Top-Dyn) funded by the State of Rhineland-Palatinate. R.M. and A.M. acknowledge the Indo-German Science Technology Centre (IGSTC) for the exchange visit via the PECFAR award No. IGSTC/PECFAR/Call 2024/IGSTC-02232/RM-AM/70/2024-25/74. M.C. acknowledges funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under the FOR 5844: ChiPS, Project-ID 541503763. L.R. gratefully acknowledges funding by the National Research, Development, and Innovation Office (NRDI) of Hungary under Project Nos. FK142601 and AD-VANCED 149745, and by the Hungarian Academy of Sciences via a János Bolyai Research Grant (Grant No. BO/00178/23/11).

References

Angular-momentum conservation

The Hamiltonian H\mathcal{H} is invariant under global rotations if H({Si})=H({RSi})\mathcal{H}(\{ \mathbf{S}_i \}) = \mathcal{H}(\{ R\mathbf{S}_i \}), where R∈SO(3)R \in\mathrm{SO}(3) is an arbitrary rotation matrix. This implies that the Hamiltonian only depends on scalar products of spins on different sites, H({Si})=H({Si⋅Sj})\mathcal{H}(\{ \mathbf{S}_i \}) = \mathcal{H}(\{ \mathbf{S}_i \cdot\mathbf{S}_j \}), similarly how from translational invariance in classical mechanics it follows that the Hamiltonian may only depend on relative positions. Consider a single scalar-product term H1=Si⋅Sj\mathcal{H}_1 = \mathbf{S}_i \cdot\mathbf{S}_j. From this it follows that

Si×∂H1∂Si+Sj×∂H1∂Sj=Si×Sj+Sj×Si=0(13)\mathbf{S}_i \times\frac{\partial\mathcal{H}_1}{\partial\mathbf{S}_i}+\mathbf{S}_j \times\frac{\partial\mathcal{H}_1}{\partial\mathbf{S}_j}=\mathbf{S}_i \times\mathbf{S}_j+\mathbf{S}_j \times\mathbf{S}_i=0 \tag*{(13)}

implying

∑kSk×∂H1∂Sk=0.(14)\sum_k \mathbf{S}_k \times\frac{\partial\mathcal{H}_1}{\partial\mathbf{S}_k}=0. \tag*{(14)}

This relation is straightforward to generalize to any rotationally symmetric Hamiltonian H\mathcal{H} via the chain rule. Equation (1) in the main text may be rearranged as

ddt(Si−ηSi×S˙i)=Si×[−γBieff+αS˙i],(15)\frac{\mathrm{d}}{\mathrm{d}t}\left(\mathbf{S}_i-\eta\mathbf{S}_i\times\dot{\mathbf{S}}_i\right)=\mathbf{S}_i\times\left[-\gamma\mathbf{B}^{\mathrm{eff}}_i+\alpha\dot{\mathbf{S}}_i\right], \tag*{(15)}

since S˙i×S˙i=0\dot{\mathbf{S}}_i\times\dot{\mathbf{S}}_i=0. Setting α=0\alpha=0 and multiplying the equation by Mi/γM_i/\gamma, then summing over the sites yields

ddt∑iMiγ(Si−ηSi×S˙i)=∑iSi×∂H∂Si.(16)\frac{\mathrm{d}}{\mathrm{d}t}\sum_i\frac{M_i}{\gamma}\left(\mathbf{S}_i-\eta\mathbf{S}_i\times\dot{\mathbf{S}}_i\right)=\sum_i\mathbf{S}_i\times\frac{\partial\mathcal{H}}{\partial\mathbf{S}_i}. \tag*{(16)}

For a rotationally invariant system, the right-hand side of the equation vanishes as shown in Eq. (14). This leads to the conservation of the sum of the quantities Li=Miγ(Si−ηSi×S˙i)L_i=\frac{M_i}{\gamma}\left(\mathbf{S}_i-\eta\mathbf{S}_i\times\dot{\mathbf{S}}_i\right), which is identified as the angular momentum, the conserved quantity traditionally associated with rotational invariance. For an alternative motivation of this expression as an angular momentum, see [1].

Linear spin-wave theory in the inertial regime

Equation (2) in the main text may be derived by rewriting Eq. (1) as a coupled set of first-order differential equations,

S~˙i=iVi,(17)\dot{\widetilde{\mathbf{S}}}_i=i\mathbf{V}_i, \tag*{(17)}
V˙i=−1ηSi×Vi+γη[Si(Si⋅Bieff)−Bieff]−αηVi−iSiVi2.(18)\dot{\mathbf{V}}_i=-\frac{1}{\eta}\mathbf{S}_i\times\mathbf{V}_i+\frac{\gamma}{\eta}\left[\mathbf{S}_i\left(\mathbf{S}_i\cdot\mathbf{B}^{\mathrm{eff}}_i\right)-\mathbf{B}^{\mathrm{eff}}_i\right]-\frac{\alpha}{\eta}\mathbf{V}_i-i\mathbf{S}_i\mathbf{V}_i^2. \tag*{(18)}

by introducing V=−iS˙\mathbf{V}=-i\dot{\mathbf{S}} as an additional variable. The equation has the same form when expressed in the local coordinate system with the spin vectors S~i\widetilde{\mathbf{S}}_i. The Hamiltonian H\mathcal{H} is expanded around the equilibrium state S~iz=1\widetilde{S}^z_i=1 up to second order in the small variables S⊥\mathbf{S}^{\perp} as

H≈E0+14(S⊥)†HSWS⊥.(19)\mathcal{H}\approx E_0+\frac{1}{4}\left(\mathbf{S}^{\perp}\right)^\dagger\mathcal{H}_{\mathrm{SW}}\mathbf{S}^{\perp}. \tag*{(19)}

Note that S⊥\mathbf{S}^{\perp} has S−\mathbf{S}^{-} and S+\mathbf{S}^{+} components, and is also a vector over the lattice sites. The spin-wave Hamiltonian is characterized by the particle-hole constraint σxKHSWKσx=HSW\sigma^x\mathcal{K}\mathcal{H}_{\mathrm{SW}}\mathcal{K}\sigma^x=\mathcal{H}_{\mathrm{SW}}, expressing that S−\mathbf{S}^{-} and S+\mathbf{S}^{+} are complex conjugates of each other as follows from their definition. In Eqs. (S5) and (S6), we only keep terms up to linear order in the small variables S⊥\mathbf{S}^{\perp}. This implies that the velocities V⊥\mathbf{V}^{\perp} are also treated as small variables, while we substitute S~iz=1\widetilde{S}^z_i=1 and consequently V~iz=0\widetilde{V}^z_i=0 for the third components. This simplifies the equation to

\begin{bmatrix} \mathbf{S}^{\perp}\\ \mathbf{V}^{\perp} \end{bmatrix} =i\begin{bmatrix} 0 & 1\\ \frac{1}{\eta}\mathcal{M}^{-1}\mathcal{H}_{\mathrm{SW}} & \frac{\sigma^z}{\eta}+\frac{i\alpha}{\eta} \end{bmatrix} \begin{bmatrix} \mathbf{S}^{\perp}\\ \mathbf{V}^{\perp} \end{bmatrix} =i\mathcal{D} \begin{bmatrix} \mathbf{S}^{\perp}\\ \mathbf{V}^{\perp} \end{bmatrix}. \tag*{(20)}\end{bmatrix}

and replacing the time derivative with iωi\omega yields the eigenvalue equation (2) in the main text. Note that the $-i prefactor in the definition of VV was introduced such that there is no explicit ii factor in the D\mathcal{D} matrix for α=0\alpha= 0, and the first and second parts of the eigenvectors are connected by multiplication via the real frequencies ω\omega. Introducing the notations

B=γηM−1,(21)\mathcal{B} = \frac{\gamma}{\eta}M^{-1}, \tag*{(21)}
A=[B12HSWB12]12,(22)\mathcal{A} = \left[B^{\frac{1}{2}}H_{\mathrm{SW}}B^{\frac{1}{2}}\right]^{\frac{1}{2}}, \tag*{(22)}
F=[AB−1200B−12],(23)\mathcal{F} = \begin{bmatrix} \mathcal{A}B^{-\frac{1}{2}} & 0 \\ 0 & B^{-\frac{1}{2}} \end{bmatrix}, \tag*{(23)}

one obtains the relation

D=F−1[0AAσzη]F.(24)\mathcal{D} = \mathcal{F}^{-1} \begin{bmatrix} 0 & \mathcal{A} \\ \mathcal{A} & \frac{\sigma_z}{\eta} \end{bmatrix} \mathcal{F}. \tag*{(24)}

This means that D\mathcal{D} is connected to a Hermitian matrix by a basis transformation F\mathcal{F}, i.e., it is quasi-Hermitian [2]. Note that the matrix B\mathcal{B} is always strictly positive definite, and A\mathcal{A} also needs to be strictly positive definite to be invertible. While the spin-wave Hamiltonian HSWH_{\mathrm{SW}} is necessarily positive semidefinite when the expansion is carried out around a stable equilibrium state, it may be necessary to project out its subspace belonging to zero eigenvalue to perform this transformation. The quasi-Hermiticity implies that the frequencies ω\omega are real valued, since they are not influenced by the basis transformation, while the particle-hole constraint of the spin-wave Hamiltonian enforces CDC−1=−D\mathcal{C}\mathcal{D}\mathcal{C}^{-1}=-\mathcal{D} for C=diag⁡(σx,−σx)K\mathcal{C}=\operatorname{diag}(\sigma^x,-\sigma^x)K, meaning that the non-zero frequencies are found in ±ω\pm\omega pairs.

The left and right eigenvectors of D\mathcal{D} are connected via

⟨lq∣=⟨rq∣†F2,(25)\langle l_q| = \langle r_q|^\dagger\mathcal{F}^2, \tag*{(25)}

and the biorthogonality relation leads to the normalization of the right eigenvectors as

⟨rq∣F2∣rq′⟩=δqq′.(26)\langle r_q|\mathcal{F}^2|r_{q'}\rangle= \delta_{qq'}. \tag*{(26)}

Berry phase

The calculation of the Berry phase follows the standard derivation [3], apart from a slight complication in the formalism due to the time-evolution operator being quasi-Hermitian. Equation Eq. (20) may be rewritten as

∂t∣v⟩=Dk∣v⟩,(27)\partial_t|v\rangle= \mathcal{D}_k|v\rangle, \tag*{(27)}

where the matrix D\mathcal{D} is now indexed with a parameter kk. In practice, in a translationally invariant system, D\mathcal{D} may be reshaped in a block-diagonal form, where each block is indexed by a wave vector kk, the possible values of which are determined by the system size and geometry. Assume that kk evolves in time according to a predefined function, i.e., a path is mapped out in the Brillouin zone. We assume that the state ∣v⟩|v\rangle is initialized in a right eigenstate of the system ∣rn,k(t=0)⟩|r_{n,k(t=0)}\rangle. In the adiabatic approach, we approximate the time evolution defined in Eq. (27) by assuming that the system stays in an eigenstate,

∣v⟩(t)≈cn(t)e∫Dk(t′)dt′∣rn,k(t)⟩.(28)|v\rangle(t) \approx c_n(t)e^{\int\mathcal{D}_{k(t')}dt'}|r_{n,k(t)}\rangle. \tag*{(28)}

Substituting Eq. (28) into Eq. (27) and using the biorthogonality relation, the phase factor cn(t)c_n(t) is explicitly given as

cn(t)=e−∫⟨ln,k(t′)∣(∂t′∣rn,k(t′)⟩)dt′=e−∫⟨ln,k∣(∂k∣rn,k⟩)dk.(29)c_n(t)=e^{-\int\langle l_{n,k(t')}|(\partial_{t'}|r_{n,k(t')}\rangle)dt'} =e^{-\int\langle l_{n,k}|(\partial_k|r_{n,k}\rangle)dk}. \tag*{(29)}

On the right-hand side, the phase is expressed by a line integral in parameter space, no longer requiring any assumptions about the time evolution. Due to the non-trivial topology of the parameter space, being a torus in the case of the Brillouin zone, the Berry phase factor cnc_n does not vanish over a closed curve. It may also be characterized by the Berry connection

Anα(k)=i⟨ln,k∣(∂kα∣rn,k⟩),(30)A_n^\alpha(\boldsymbol{k})=i\langle l_{n,k}|(\partial_{k_\alpha}|r_{n,k}\rangle), \tag*{(30)}

and the Berry curvature, which in a three-dimensional parameter space reads

Bnα(k)=iϵαβγ⟨(∂kβ∣ln,k⟩∣(∂kγ∣rn,k⟩).(31)B_n^\alpha(\boldsymbol{k})=i\epsilon^{\alpha\beta\gamma}\langle(\partial_{k_\beta}|l_{n,k}\rangle|(\partial_{k_\gamma}|r_{n,k}\rangle). \tag*{(31)}

In the two-dimensional Brillouin zone in the xyxy plane considered in the main text, only the zz component of the Berry curvature remains. Using the eigenvalue equation (2) and the biorthogonality relations yields

(ωn−ωm)⟨lm,k∣(∂kα∣rn,k⟩)=⟨lm,k∣(∂kαDk)∣rn,k⟩(32)(\omega_n-\omega_m)\langle l_{m,k}|(\partial_{k_\alpha}|r_{n,k}\rangle)=\langle l_{m,k}|(\partial_{k_\alpha}\mathcal{D}_k)|r_{n,k}\rangle \tag*{(32)}

for m≠nm\ne n, which leads to the alternative expression Eq. (12) in the main text. This has the advantage that the derivatives of the matrix Dk\mathcal{D}_k may often be calculated analytically, while calculating the derivatives of the eigenvectors numerically is difficult, because their phase may vary between different wave vectors. An alternative form of the Berry curvature may be derived by expressing the left eigenvectors via Eq. (25), substituting the expression ∣rn,k⟩=[vn,k,ωn,kvn,k]|r_{n,k}\rangle=[v_{n,k},\omega_{n,k}v_{n,k}] enforced by the matrix structure, and calculating the derivatives of Dk\mathcal{D}_k where only HSW,kH_{\mathrm{SW},k} depends on the wave vector. This results in

Bnα(k)=iϵαβγ∑m≠nωn,kωm,k⟨vn,k∣∂kβHSW,k∣vm,k⟩⟨vm,k∣∂kγHSW,k∣vn,k⟩(ωn,k−ωm,k)2.(33)B_n^\alpha(\boldsymbol{k})=i\epsilon^{\alpha\beta\gamma}\sum_{m\ne n}\omega_{n,k}\omega_{m,k} \frac{\langle v_{n,k}|\partial_{k_\beta}H_{\mathrm{SW},k}|v_{m,k}\rangle\langle v_{m,k}|\partial_{k_\gamma}H_{\mathrm{SW},k}|v_{n,k}\rangle}{(\omega_{n,k}-\omega_{m,k})^2}. \tag*{(33)}

This is similar to the formula in Ref. [4], with the caveat that the eigenvectors follow the normalization condition derived from Eq. (26),

⟨vn,k|[AB−12]2+B−1ωn,k2|vn′,k′⟩=δnn′δkk′.(34)\left\langle v_{n,k} \middle| \left[ A\mathcal{B}^{-\frac{1}{2}} \right]^2 + \mathcal{B}^{-1}\omega_{n,k}^{2} \middle| v_{n',k'} \right\rangle= \delta_{nn'}\delta_{kk'}. \tag*{(34)}

Alternatively, the Berry curvature may also be written as proposed in Ref. [5],

Bnα(k)=iϵαβγTr⁡[(1k−Pn,k)∂kβPn,k∂kγPn,k],(35)B_{n}^{\alpha}(\boldsymbol{k}) = i\epsilon^{\alpha\beta\gamma}\operatorname{Tr}\left[(\mathbb{1}_{k}-P_{n,k})\partial_{k_{\beta}}P_{n,k}\partial_{k_{\gamma}}P_{n,k}\right], \tag*{(35)}

where Pn,k=∣rn,k⟩⟨ln,k∣P_{n,k}=\lvert r_{n,k}\rangle\langle l_{n,k}\rvert is the (non-orthogonal) projection onto the subspace of the eigenstate.

Gap scaling with the pseudodipolar interaction

In the absence of spin inertia, the gap in the magnon spectrum of the ferromagnet on the honeycomb lattice is opened in the KK and K′K' points. In the K=2πa(0,133)K=\frac{2\pi}{a}\left(0,\frac{1}{3\sqrt{3}}\right) point, the spin-wave Hamiltonian from Eq. (5) in the main text reads

HSW,K=(−3Jzz0032F0−3Jzz0000−3Jzz032F00−3Jzz),(36)\mathcal{H}_{\mathrm{SW},K} = \begin{pmatrix} -3J^{zz} & 0 & 0 & \frac{3}{2}F \\ 0 & -3J^{zz} & 0 & 0 \\ 0 & 0 & -3J^{zz} & 0 \\ \frac{3}{2}F & 0 & 0 & -3J^{zz} \end{pmatrix}, \tag*{(36)}

The magnon frequencies are given by the eigenvalues of the matrix σzHSW\sigma^{z}\mathcal{H}_{\mathrm{SW}}. Treating the pseudodipolar interaction as a perturbation, the unperturbed frequencies are ∓3Jzz\mp3J^{zz}, both of them twice degenerate and of counterclockwise polarization. In first order of the perturbation theory, the matrix elements of the perturbation matrix proportional to FF have to be calculated on the degenerate subspaces. However, since these vectors only have matrix elements either in the upper left or the lower right block of the 4×44\times4 matrix, and the perturbation matrix is offdiagonal in the blocks (connecting counterclockwise modes with clockwise modes), these terms all vanish. The second-order corrections in perturbation theory will be finite, meaning that the gap scales with ∣F∣2\lvert F\rvert^{2}. This is different from the case of the next-nearest neighbor Dzyaloshinsky-Moriya interaction [6], which gives nonvanishing contributions to the diagonal terms of the matrix in the KK point, thereby opening a gap in first order of the perturbation theory.

In the presence of spin inertia, the precessional and nutational bands may become degenerate along a general contour in the Brillouin zone. Restricting the eigenvectors ∣rk⟩\lvert r_k\rangle to the spin components vkv_k, one of the degenerate eigenstates for F=0F=0 is restricted to the first two components (counterclockwise mode), while the other one is restricted to the second two components (clockwise mode). The pseudodipolar term breaking the conservation of total angular momentum introduces a matrix element between the counterclockwise and the clockwise mode already in first order of perturbation theory, thereby opening a gap which scales as ∣F∣\lvert F\rvert.

Topological phase transition in the honeycomb lattice

As described in the main text, two different types of gaps may be identified in the magnon spectrum of the ferromagnetic honeycomb lattice. The gap inside the precessional and nutational bands from the Dirac cones at the KK and K′K' points opens for arbitrarily small values of the pseudo-dipolar interaction FF, and is always topologically non-trivial. However, the precessional and nutational bands are separated by a gap for low values of the inertial relaxation time η\eta, as illustrated in Figure S1. Edge states are visible in the slab geometry below the nutational bands, but similarly to the edge states below the precessional bands, these bands do not connect two bands, and are of topologically trivial origin. As η\eta is increased, the gap between the bands closes and reopens. As shown in Figure S1, after the reopening the edge states connect the precessional to the nutational band. Increasing the inertial relaxation time further in Figure S1 increases the gap size, despite the general reduction in the band widths with increasing η\eta. This topological phase transition happens close to η=230 fs\eta=230\ \mathrm{fs}, as can be seen from the change in the Chern numbers in Figure S2. For lower values of η\eta, the Chern numbers of the bands are alternating between −1-1 and 11, meaning that their sum for the first two bands is zero, indicating no topologically non-trivial edge state between the second and third bands. As η\eta is increased, the Chern numbers of the second and third bands change to 00, explaining the formation of edge states between them. The topological character of the bands may also be modified by changing the strength of the pseudodipolar interaction strength, as shown in Figure S2. For higher values of FF, the width of the bands decreases, which increases the distance between the precessional and nutational bands, and transforms the gap between the second and third bands back to topologically trivial where the Chern numbers in all bands are +1+1 or −1-1.

Figure S1. Magnon band structure in a slab geometry. The value of η\eta is indicated in each panel, while the other parameters are the same as in Table I in the main text. The system is 40 unit cells wide along the xx direction with open boundary conditions, while it is infinite along the yy direction with periodic boundary conditions, meaning that the eigenstates can be indexed by the wave vector component kyk_y. aa denotes the distance between nearest-neighbor atoms in the lattice, i.e., the lattice constant is 3a\sqrt{3}a. Dotted lines show the boundaries of the one-dimensional Brillouin zone. Coloring shows the localization defined in Eq. (11) in the main text.

Figure S2. Chern numbers of the magnon bands. The parameters were the same as in Table I in the main text, except from the inertial relaxation time η\eta in panel (a) and the pseudodipolar interaction FF in panel (b) which were varied as indicated. Trivial and non-trivial denote the topological character of the gap between the second and third bands.

Magnon Chern insulator in a single-sublattice ferromagnet

Here, we consider a two-dimensional ferromagnet on a triangular lattice, which commonly occurs in epitaxially grown transition-metal layers on nonmagnetic substrates such as Co on Au(111) [7, 8]. The substrate breaks the inversion symmetry of the structure, leaving a C3vC_{3v} point-group symmetry with the direction perpendicular to the surface denoted by z^\hat{z}. Restricting the model to nearest-neighbor pair interactions, it can be described by the

H=12∑⟨i,j⟩[JSiSj−ΔJzz(2SizSjz−SixSjx−SiySjy)+F{(Sir^ij)(Sjr^ij)−[Si(z^×r^ij)][Sj(z^×r^ij)]}+(Dijzz^+Dij∥z^×r^ij)(Si×Sj)+ΔJ−{[Si(z^×r^ij)]Sj−+Si−[Sj(z^×r^ij)]}]+∑i[K(Siz)2+MBzSiz],(37)\mathcal{H} = \frac{1}{2}\sum_{\langle i,j\rangle}\left[J\mathbf{S}_i\mathbf{S}_j-\Delta J^{zz}\left(2S_i^zS_j^z-S_i^xS_j^x-S_i^yS_j^y\right) +F\left\{\left(\mathbf{S}_i\hat{\mathbf{r}}_{ij}\right)\left(\mathbf{S}_j\hat{\mathbf{r}}_{ij}\right)-\left[\mathbf{S}_i\left(\hat{\mathbf{z}}\times\hat{\mathbf{r}}_{ij}\right)\right]\left[\mathbf{S}_j\left(\hat{\mathbf{z}}\times\hat{\mathbf{r}}_{ij}\right)\right]\right\} +\left(D_{ij}^{z}\hat{\mathbf{z}}+D_{ij}^{\parallel}\hat{\mathbf{z}}\times\hat{\mathbf{r}}_{ij}\right)\left(\mathbf{S}_i\times\mathbf{S}_j\right) +\Delta J^{-}\left\{\left[\mathbf{S}_i\left(\hat{\mathbf{z}}\times\hat{\mathbf{r}}_{ij}\right)\right]S_j^{-}+S_i^{-}\left[\mathbf{S}_j\left(\hat{\mathbf{z}}\times\hat{\mathbf{r}}_{ij}\right)\right]\right\}\right] +\sum_i\left[K\left(S_i^z\right)^2+MB^zS_i^z\right], \tag*{(37)}

where JJ is the exchange interaction, ΔJzz\Delta J^{zz} is the uniaxial two-site anisotropy, FF is the pseudodipolar interaction, DijzD^{z}_{ij} and D∥D^{\parallel} are out-of-plane and in-plane components of the Dzyaloshinsky-Moriya interaction, respectively (note that DijzD^{z}_{ij} has opposite signs for opposite directions of r^ij\hat{r}_{ij}), ΔJ⊥\Delta J^{\perp} is an anisotropy term connecting the in-plane and out-of-plane components, KK is the uniaxial single-site anisotropy, and BzB^{z} is the external field applied along the symmetry axis. This includes all possible coefficients allowed by symmetry considerations.

Assuming a ferromagnetic ground state along the z^\hat{z} direction, the spin-wave Hamiltonian in Fourier space may be written as

HSW,k=[D0,k+Dnr,kDa,kDa,k∗D0,k−Dnr,k],(38)\mathcal{H}_{\mathrm{SW},k} = \begin{bmatrix} D_{0,k}+D_{\mathrm{nr},k} & D_{\mathrm{a},k} \\ D^{*}_{\mathrm{a},k} & D_{0,k}-D_{\mathrm{nr},k} \end{bmatrix}, \tag*{(38)}

with

D0,k=2(J+ΔJzz)[cos⁡(kxa)+2cos⁡(kx2a)cos⁡(3ky2a)]−6(J−2ΔJzz)−2K−MBz,(39)D_{0,k}=2\left(J+\Delta J^{zz}\right)\left[\cos(k_xa)+2\cos\left(\frac{k_x}{2}a\right)\cos\left(\frac{\sqrt{3}k_y}{2}a\right)\right]-6\left(J-2\Delta J^{zz}\right)-2K-MB^{z}, \tag*{(39)}
Dnr,k=−2Dz[sin⁡(kxa)−2sin⁡(kx2a)cos⁡(3ky2a)],(40)D_{\mathrm{nr},k}=-2D^{z}\left[\sin(k_xa)-2\sin\left(\frac{k_x}{2}a\right)\cos\left(\frac{\sqrt{3}k_y}{2}a\right)\right], \tag*{(40)}
Da,k=2F[cos⁡(kxa)−cos⁡(kx2a)cos⁡(3ky2a)]−i23Fsin⁡(kx2a)sin⁡(3ky2a),(41)D_{\mathrm{a},k}=2F\left[\cos(k_xa)-\cos\left(\frac{k_x}{2}a\right)\cos\left(\frac{\sqrt{3}k_y}{2}a\right)\right]-i2\sqrt{3}F\sin\left(\frac{k_x}{2}a\right)\sin\left(\frac{\sqrt{3}k_y}{2}a\right), \tag*{(41)}

where D0,kD_{0,k}, Dnr,kD_{\mathrm{nr},k}, and Da,kD_{\mathrm{a},k} are the normal (even in k\mathbf{k}), nonreciprocal (odd in k\mathbf{k}), and anomalous (pairing S+S^{+} with S+S^{+} and S−S^{-} with S−S^{-}) terms, respectively. Note that the D∥D^{\parallel} and ΔJ⊥\Delta J^{\perp} parameters do not influence the spectrum in linear spin-wave theory for this magnetization orientation, and the role of the two-site and single-site anisotropies as well as that of the external field are mostly interchangeable.

The magnon frequencies may be determined by solving Eq. (3) in the main text, which may be rearranged as

(−ωn,k2η+γMD0,k)2−(ωn,k+γMDnr,k)2−∣γMDa,k∣2=0.(42)\left(-\omega_{n,k}^{2}\eta+\frac{\gamma}{M}D_{0,k}\right)^{2} -\left(\omega_{n,k}+\frac{\gamma}{M}D_{\mathrm{nr},k}\right)^{2} -\left|\frac{\gamma}{M}D_{\mathrm{a},k}\right|^{2}=0. \tag*{(42)}

In the absence of inertia (η=0\eta=0), the single pair of solutions ω1/2,k=±ωk\omega_{1/2,k}=\pm\omega_{k} form particle-hole partners, which cannot be degenerate apart from the trivial case where HSW,k=0\mathcal{H}_{\mathrm{SW},k}=0. It is not possible to form topologically non-trivial bands in this case, since that would require changing the sign of one of the eigenvalues inside the Brillouin zone, which would destabilize the ferromagnetic ground state.

When spin inertia is taken into account, there are two pairs of particle-hole partners. The positive-frequency solutions may be approximately described by a two-level Hamiltonian

H2=d0,k1+dα,kσα,(43)\mathcal{H}_{2}=d_{0,k}\mathbf{1}+d_{\alpha,k}\sigma^{\alpha}, \tag*{(43)}

where d0,kd_{0,k} is the mean of the two eigenfrequencies, dx,k=γMRe⁡Da,kd_{x,k}=\frac{\gamma}{M}\operatorname{Re}D_{\mathrm{a},k}, dy,k=−γMIm⁡Da,kd_{y,k}=-\frac{\gamma}{M}\operatorname{Im}D_{\mathrm{a},k}, and dx,k=γMD0,k−η(γMDnr,k)2d_{x,k}=\frac{\gamma}{M}D_{0,k}-\eta\left(\frac{\gamma}{M}D_{\mathrm{nr},k}\right)^{2}. The two solutions are degenerate when dx,k=dy,k=dz,k=0d_{x,k}=d_{y,k}=d_{z,k}=0; note that this condition is exact also in Eq. (42), not only on the level of the two-level approximation. Since there are three conditions and the parameter space is two-dimensional, such a degeneracy only occurs for specific model parameter values where the gap closes.

The topology of the band structure may be characterized by how many times the direction of the vector (dx,k,dy,k,dz,k)(d_{x,k},d_{y,k},d_{z,k}) wraps the two-dimensional unit sphere. This corresponds to the Chern number of the considered bands up to a sign, with the two bands having opposite Chern numbers. The Chern number may only be nonzero if all three components change sign in the Brillouin zone. While this is always true for dx,kd_{x,k} and dy,kd_{y,k}, dz,kd_{z,k} is usually positive. Typically the exchange interaction is stronger than the Dzyaloshinsky-Moriya interaction, D0,k>∣Dnr,k∣D_{0,k}>|D_{\mathrm{nr},k}|, and D0,kD_{0,k} is required to be positive everywhere in the Brillouin zone in order for the ground state to be stable. D0,kD_{0,k} is smallest in the vicinity of k=0\mathbf{k}=0, but at this point ∣Dnr,k∣|D_{\mathrm{nr},k}| also vanishes since it is antisymmetric in k\mathbf{k}. Furthermore, the parameter ηγM∣Dnr,k∣\eta\frac{\gamma}{M}|D_{\mathrm{nr},k}| is often small since it scales with the strength of the Dzyaloshinsky-Moriya interaction and the inertial relaxation time, both of which emerge due to the spin-orbit coupling. While mathematically it is possible to select such a high value of η\eta that a sign change occurs and the Chern number becomes finite, this may be difficult to observe in actual materials.

Compared to the honeycomb lattice discussed in the main text, one can see that the pseudodipolar interaction FF is required for the formation of the topologically non-trivial phase in both cases, since this term breaks the conservation of angular momentum. While the Dzyaloshinsky-Moriya interaction played an unimportant role for the honeycomb lattice, it must assume very high values in the triangular lattice, possibly becoming stronger than the exchange interaction. In contrast, for the honeycomb lattice the distance between the precessional and nutational bands may be decreased by increasing the dimensionless parameter ηγM−1∣J∣\eta\gamma M^{-1}|J|, i.e., for higher values of either the inertial relaxation time or the exchange interaction.

References

References

  1. [1]H. C. Siegmann, E. L. Garwin, C. Y. Prescott, J. Heidmann, D. Mauri, D. Weller, R. Allenspach, and W. Weber, Magnetism with picosecond field pulses, J. Magn. Magn. Mater. 151, L8 (1995).DOI
  2. [2]E. Beaurepaire, J.-C. Merle, A. Daunois, and J.-Y. Bigot, Ultrafast spin dynamics in ferromagnetic nickel, Phys. Rev. Lett. 76, 4250 (1996).
  3. [3]T. Kampfrath, A. Kirilyuk, S. Mangin, S. Sharma, and M. Weinelt, Ultrafast and terahertz spintronics: Guest editorial, Applied Physics Letters 123, 050401 (2023).DOI
  4. [4]M.-C. Ciornei, J. M. Rubí, and J.-E. Wegrowe, Magnetization dynamics in the inertial regime: Nutation predicted at short time scales, Phys. Rev. B 83, 020410 (2011).DOI
  5. [5]H. Suhl, Theory of the magnetic damping constant, IEEE Trans. Magn. 34, 1834 (1998).
  6. [6]R. Mondal, M. Berritta, A. K. Nandy, and P. M. Oppe- neer, Relativistic theory of magnetic inertia in ultrafast spin dynamics, Phys. Rev. B 96, 024425 (2017).DOI
  7. [7]E. Olive, Y. Lansac, M. Meyer, M. Hayoun, and J.-E. Wegrowe, Deviation from the Landau-Lifshitz-Gilbert equation in the inertial regime of the magnetization, J. Appl. Phys. 117, 213904 (2015).arxiv.org/abs/1412.3783
  8. [8]R. Mondal, L. Rózsa, M. Farle, P. M. Oppeneer, U. Nowak, and M. Cherkasskii, Inertial effects in ultrafast spin dynamics, Journal of Magnetism and Magnetic Materials 579, 170830 (2023).DOI
  9. [9]E. Olive, Y. Lansac, and J.-E. Wegrowe, Beyond ferromagnetic resonance: The inertial regime of the magnetization, Appl. Phys. Lett. 100, 192407 (2012).
  10. [10]K. Neeraj, N. Awari, S. Kovalev, D. Polley, N. Zhou Hagström, S. P. K. Arekapudi, A. Semisalova, K. Lenz, B. Green, J.-C. Deinert, et al., Inertial spin dynamics in ferromagnets, Nature Physics 17, 245 (2021).DOI
  11. [11]V. Unikandanunni, R. Medapalli, M. Sa, E. Albisetti, D. Petti, R. Bertacco, E. E. Fullerton, and S. Bonetti, Inertial spin dynamics in epitaxial cobalt films, Phys. Rev. Lett. 129, 237201 (2022).DOI
  12. [12]A. De, J. Schlegel, A. Lentfert, L. Scheuer, B. Stadtmüller, P. Pirro, G. von Freymann, U. Nowak, and M. Aeschlimann, Magnetic nutation: Transient separation of magnetization from its angular momentum, Phys. Rev. B 111, 014432 (2025).DOI
  13. [13]R. Mondal and A. Kamra, Spin pumping at terahertz nutation resonances, Phys. Rev. B 104, 214426 (2021).DOI
  14. [14]L. Winter, S. Großenbach, U. Nowak, and L. Rózsa, Nutational switching in ferromagnets and antiferromagnets, Phys. Rev. B 106, 214403 (2022).DOI
  15. [15]R. Rodriguez, M. Cherkasskii, R. Jiang, R. Mondal, A. Etesamirad, A. Tossounian, B. A. Ivanov, and I. Barsukov, Spin inertia and auto-oscillations in ferromagnets, Phys. Rev. Lett. 132, 246701 (2024).DOI
  16. [16]T. Kikuchi and G. Tatara, Spin dynamics with inertia in metallic ferromagnets, Phys. Rev. B 92, 184410 (2015).DOI
  17. [17]R. Mondal and L. Rózsa, Inertial spin waves in ferromagnets and antiferromagnets, Phys. Rev. B 106, 134422 (2022).DOI
  18. [18]M. Cherkasskii, R. Mondal, and L. Rózsa, Inertial spin waves in spin spirals, Phys. Rev. B 109, 184424 (2024).DOI
  19. [19]M. Cherkasskii, M. Farle, and A. Semisalova, Dispersion relation of nutation surface spin waves in ferromagnets, Phys. Rev. B 103, 174435 (2021).
  20. [20]P. A. McClarty, Topological magnons: A review, Annual Review of Condensed Matter Physics 13, 171 (2022).DOI
  21. [21]L. Zhang, J. Ren, J.-S. Wang, and B. Li, Topological magnon insulator in insulating ferromagnet, Phys. Rev. B 87, 144101 (2013).DOI
  22. [22]A. Mook, J. Henk, and I. Mertig, Magnon hall effect and topology in kagome lattices: A theoretical investigation, Phys. Rev. B 89, 134409 (2014).DOI
  23. [23]A. Mook, J. Henk, and I. Mertig, Edge states in topological magnon insulators, Phys. Rev. B 90, 024412 (2014).DOI
  24. [24]B. Chisnell, J. S. Helton, D. E. Freedman, D. K. Singh, R. I. Bewley, D. G. Nocera, and Y. S. Lee, Topological magnon bands in a kagome lattice ferromagnet, Phys. Rev. Lett. 115, 147201 (2015).DOI
  25. [25]R. Shindou, R. Matsumoto, S. Murakami, and J.-i. Ohe, Topological chiral magnonic edge mode in a magnonic crystal, Phys. Rev. B 87, 174427 (2013).DOI
  26. [26]A. Roldán-Molina, A. S. Nuñez, and J. Fernández-Rossier, Topological spin waves in the atomic-scale magnetic skyrmion crystal, New Journal of Physics 18, 045015 (2016).
  27. [27]S. A. Díaz, J. Klinovaja, and D. Loss, Topological Magnons and Edge States in Antiferromagnetic Skyrmion Crystals, Phys. Rev. Lett. 122, 187203 (2019).DOI
  28. [28]T. Weber, D. M. Fobes, J. Waizner, P. Steffens, G. S. Tucker, M. Böhm, L. Beddrich, C. Franz, H. Gabold, R. Bewley, D. Voneshen, M. Skoulatos, R. Georgii, G. Ehlers, A. Bauer, C. Pfleiderer, P. Böni, M. Janoschek, and M. Garst, Topological magnon band structure of emergent Landau levels in a skyrmion lattice, Science 375, 1025–1030 (2022).DOI
  29. [29]F.-Y. Li, Y.-D. Li, Y. B. Kim, L. Balents, Y. Yu, and G. Chen, Weyl magnons in breathing pyrochlore antiferromagnets, Nature Communications 7, 12691 (2016).DOI
  30. [30]A. Mook, J. Henk, and I. Mertig, Tunable Magnon Weyl Points in Ferromagnetic Pyrochlores, Phys. Rev. Lett. 117, 157204 (2016).DOI
  31. [31]H. Katsura, N. Nagaosa, and P. A. Lee, Theory of the Thermal Hall Effect in Quantum Magnets, Phys. Rev. Lett. 104, 066403 (2010).pubmed.ncbi.nlm.nih.gov/20366838
  32. [32]R. Matsumoto and S. Murakami, Rotational Motion of Magnon Wave Packet and Berry Curvature, Phys. Rev. B 84, 184406 (2011).
  33. [33]R. Matsumoto and S. Murakami, Theoretical Prediction of a Rotating Magnon Wave Packet in Ferromagnets, Phys. Rev. Lett. 106, 197202 (2011).pubmed.ncbi.nlm.nih.gov/21668195
  34. [34]Y. Onose, T. Ideue, H. Katsura, Y. Shiomi, N. Nagaosa, and Y. Tokura, Observation of the Magnon Hall Effect, Science 329, 297 (2010).DOI
  35. [35]L. Chen, J.-H. Chung, B. Gao, T. Chen, M. B. Stone, A. I. Kolesnikov, Q. Huang, and P. Dai, Topological Spin Excitations in Honeycomb Ferromagnet CrI₃, Phys. Rev. X 8, 041028 (2018).DOI
  36. [36]L. Chen, J.-H. Chung, M. B. Stone, A. I. Kolesnikov, B. Winn, V. O. Garlea, D. L. Abernathy, B. Gao, M. Augustine, E. G. Santos, and P. Dai, Magnetic Field Effect on Topological Spin Excitations in CrI₃, Phys. Rev. X 11, 031047 (2021).DOI
  37. [37]F. Zhu, L. Zhang, X. Wang, F. J. dos Santos, J. Song, T. Mueller, K. Schmalzl, W. F. Schmidt, A. Ivanov, J. T. Park, J. Xu, J. Ma, S. Lounis, S. Blügel, Y. Mokrousov, Y. Su, and T. Brückel, Topological magnon insulators in two-dimensional van der Waals ferromagnets CrSiTe₃ and CrGeTe₃: Toward intrinsic gap-tunability, Science Advances 7, eabi7532 (2021).DOI
  38. [38]I. Dzyaloshinsky, A thermodynamic theory of “weak” ferromagnetism of antiferromagnetics, J. Phys. Chem. Solids 4, 241 (1958).DOI
  39. [39]T. Moriya, Anisotropic superexchange interaction and weak ferromagnetism, Phys. Rev. 120, 91 (1960).DOI
  40. [40]X. S. Wang, Y. Su, and X. R. Wang, Topologically protected unidirectional edge spin waves and beam splitter, Phys. Rev. B 95, 014435 (2017).DOI
  41. [41]P. A. McClarty, X.-Y. Dong, M. Gohlke, J. G. Rau, F. Pollmann, R. Moessner, and K. Penc, Topological magnons in Kitaev magnets at high fields, Phys. Rev. B 98, 060404 (2018).DOI
  42. [42]I. Lee, F. G. Utermohlen, D. Weber, K. Hwang, C. Zhang, J. van Tol, J. E. Goldberger, N. Trivedi, and P. C. Hammel, Fundamental Spin Interactions Underlying the Magnetic Anisotropy in the Kitaev Ferromagnet CrI₃, Phys. Rev. Lett. 124, 017201 (2020).DOI
  43. [43]R. Jaeschke-Ubiergo, E. Suárez Morell, and A. S. Nunez, Theory of magnetism in the van der Waals magnet CrI₃, Phys. Rev. B 103, 174410 (2022).
  44. [44]V. Brehm, P. Sobieszczyk, J. N. Kłogetvedt, R. F. L. Evans, E. J. G. Santos, and A. Qaiumzadeh, Topological magnon gap engineering in van der Waals CrI₃ ferromagnets, Phys. Rev. B 109, 174425 (2024).DOI
  45. [45]See Supplemental Material at [URL] for considerations about angular-momentum conservation, the derivation of linear spin-wave theory and the Berry curvature in the inertial regime, a discussion of the parameter dependence of the size of the topological gap and of the topological phase transition in the honeycomb lattice, and the formation of a magnonic Chern insulating state in a single-sublattice ferromagnet. It also contains Refs. [49, 50] and the Supplemental Movies.
  46. [46]F. Scholtz, H. Geyer, and F. Hahne, Quasi-Hermitian operators in quantum mechanics and the variational principle, Annals of Physics 213, 74–101 (1992).
  47. [47]R. Mondal, S. Großenbach, L. Rózsa, and U. Nowak, Nutation in antiferromagnetic resonance, Phys. Rev. B 103, 104404 (2021).DOI
  48. [48]D. Soriano, M. I. Katsnelson, and J. Fernández-Rossier, Magnetic two-dimensional chromium trihalides: A theoretical perspective, Nano Letters 20, 6225–6234 (2020).DOI

Paper details

vlnc:W_avuzz2tb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

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