![Arxiv:2103.05656V1 [Cond-Mat.Str-El] 9 Mar 2021 Terparts − Have Been the Subject of Intense Study in Elec- Inelastic Neutron Scattering](http://data.docslib.org/img/b5644bb8555387ebb02481b77559415b-1.webp)
Spin-Space Groups and Magnon Band Topology
A. Corticelli,1 R. Moessner,1 and P. A. McClarty1 1Max Planck Institute for the Physics of Complex Systems, N¨othnitzer Str. 38, 01187 Dresden, Germany Band topology is both constrained and enriched by the presence of symmetry. The importance of anti-unitary symmetries such as time reversal was recognized early on leading to the classification of topological band structures based on the ten-fold way. Since then, lattice point group and non- symmorphic symmetries have been seen to lead to a vast range of possible topologically nontrivial band structures many of which are realized in materials. In this paper we show that band topology is further enriched in many physically realizable instances where magnetic and lattice degrees of freedom are wholly or partially decoupled. The appropriate symmetry groups to describe general magnetic systems are the spin-space groups. Here we describe cases where spin-space groups are essential to understand the band topology in magnetic materials. We then focus on magnon band topology where the theory of spin-space groups has its simplest realization. We consider magnetic Hamiltonians with various types of coupling including Heisenberg and Kitaev couplings revealing a hierarchy of enhanced magnetic symmetry groups depending on the nature of the lattice and the couplings. We describe, in detail, the associated representation theory and compatibility relations thus characterizing symmetry-enforced constraints on the magnon bands revealing a proliferation of nodal points, lines, planes and volumes.
I. INTRODUCTION tions are wholly or partially decoupled. In this paper, we investigate these enhanced symmetry groups − so-called Most readers will be very familiar with the huge abun- spin-space groups − in relation to band topology. Begin- dance and diversity with which crystalline solids occur ning with a brief introduction to the spin-space groups, in nature. Underlying the richness of the chemistry and we give an account of their importance to the understand- details of the structure is the set of lattice symmetries ing of physically relevant condensed matter systems. We which fall into one of 17 wallpaper groups in 2D materi- then turn to the investigation of constraints on band als and 230 space groups in 3D materials [1]. The profu- topology arising from the spin-space symmetries. sion of structures and symmetries grows when we focus Our presentation concentrates on magnon band topol- on magnetic crystals as one is then forced to include the ogy which is the simplest context for studying symme- role of time reversal and its interplay with magnetic or- tries in magnetic materials, though many of our consid- der. Altogether there are 80 magnetic space groups in erations carry over to the electronic band structures of 2D and 1651 in 3D [1]. These symmetry groups and the itinerant magnetic materials. For magnons, in common point groups on which they are based form one of the with other bosonic excitations, particle-hole and chiral cornerstones of condensed matter physics as they place symmetries are subsumed by time reversal resulting in constraints on couplings, dispersion relations, wavefunc- a three-fold rather than a ten-fold table [11]. For this tions and matrix elements [2]. reason, crystalline symmetries and, more generally, spin- Symmetry is essential also to understand the variety space symmetries are especially important if magnons are of possible topological band structures [3]. While band to be imbued with interesting band topology. topology can be nontrivial in the complete absence of Magnon band structures emerge through symmetry symmetry it is greatly enriched by its presence as may breaking from an underlying lattice of interacting mag- be appreciated by inspecting the ten-fold classification netic moments. As we show, the symmetry of the band of band topology with anti-unitary symmetries [3, 4]. structure of coherent single magnon excitations is teth- The connectivity of bands in momentum space, includ- ered both to the nature of the magnetic exchange Hamil- ing the presence or absence of band touchings, is highly tonian as well as to the magnetic structure. Such exci- constrained by lattice symmetries. Such constraints − tations are routinely probed in bulk magnetic materials originating from space groups and their magnetic coun- in energy-momentum over the entire Brillouin zone using arXiv:2103.05656v1 [cond-mat.str-el] 9 Mar 2021 terparts − have been the subject of intense study in elec- inelastic neutron scattering. tronic band structures and underlie various partial clas- One notable case where spin-space groups arise is in the sification schemes of band topology [5–7]. In turn, these study of Kitaev-Heisenberg models [12] that are of con- classification tables are important for identifying topo- siderable current interest [13–17] owing to the rich phe- logical materials. nomenology in such models and their relevance to materi- In strongly spin-orbit coupled magnetic materials, the als [18–30]. Kitaev physics is known to range over quan- magnetic moments are typically locked to the lattice so tum spin liquid phases [31], complex ordered magnetic that transformations performed on the moments are per- structures [19, 24], rich magnetic field induced phase di- formed also in real space. However, as was noted long agrams [32], nontrivial magnon band topology [33, 34] ago by Brinkman and Elliott [8–10] there are physically and unusual heat transport properties [35]. In the follow- natural settings where the spin and space transforma- ing pages we show that spin-space groups are essential to 2 understand the magnon band structures of such models We illustrate this through a number of examples be- and conversely can be used as a means of establishing the ginning with the particularly rich case of the Heisenberg- importance of Kitaev-Heisenberg terms in real materials. Kitaev model on a hyperhoneycomb lattice with a sim- Spin-space groups also arise in the study of other phys- ple collinear antiferromagnetically ordered ground state. ically motivated magnetic models including Heisenberg We show (Section IV A), by direct calculation, that the models, certain single ion anisotropies and certain anti- magnon band structure has a nodal plane that is punc- symmetric exchange couplings as we discuss below. For tured by a number of nodal lines and that these fea- various models with spin-space symmetry we show direct tures survive a tuning of the Kitaev-Heisenberg couplings connections between the enhanced symmetry and a wide within this phase suggesting that they originate from the variety of richly degenerate band structures with Dirac Hamiltonian symmetries. points in 2D, Weyl points, 4-fold degenerate points, nodal We then enumerate all the symmetries of the magneti- lines and planes and 2-fold degenerate volumes across the cally ordered state. These include a combination of non- Brillouin zone [36–40]. In addition, in the course of this symmorphic symmetry elements, anti-unitary elements analysis we show in detail how to analyse band represen- and spin-space symmetries. The resulting group is not tations for spin-space groups including non-symmorphic isomorphic to any of the 1651 magnetic space groups and groups. is instead a concrete example of a spin-space group. Sec- tion IV B is a derivation of the nodal plane and other degeneracies in the richly featured band structure on the A. Outline of the paper basis of the band representation theory of this group. Having given one detailed calculation of the magnon Spin-space groups characterize the symmetries of a band structure from a spin-space group over the entire range of physically relevant interacting magnetic systems. Brillouin zone, we present a number of further examples. In such cases, the standard magnetic space groups are in- The first set of examples comes from Heisenberg models sufficient to capture all the symmetries of the problem. in Section V. We make general observations about the na- In the next section, we give a number of examples ture of the spin-space symmetries of collinear ferromag- where such enhanced magnetic symmetries arise. In fact, nets and antiferromagnets and their effect on magnon such models are very common as they include Heisenberg bands. In the latter case, we describe how these symme- models that appear, to an excellent approximation, in tries can lead to nodal volumes. many materials with weak spin-orbit coupling. In such In Section VI, we focus on the honeycomb lattice − one models, the spin space transformations are completely of the most symmetric lattices in two-dimensions. The decoupled from the lattice transformations and there- recent literature on topological magnons includes the dis- fore lie at the extreme end of possible spin-space sym- covery of magnon Chern bands in the Kitaev-Heisenberg metry groups. We show that spin-space groups appear honeycomb model and in the case of the ferromagnet with also in cases where spin-orbit coupling is important. For second-neighbor anti-symmetric exchange. In these mod- example, for various kinds of single ion anisotropy, for els, the band topology can arise in the complete absence Kitaev-Heisenberg models and for certain kinds of anti- of symmetry. We revisit them to show that their nontriv- symmetric or Dzyaloshinskii-Moriya exchange. ial spin-space symmetries allow one to tune the appear- In Section II C we briefly review spin wave theory ance of Dirac points in the magnon band structure. As a and show that spin-space symmetries are inherited by corollary, the symmetry analysis reveals that the known magnons. Then, we give a short account of the band rep- Chern band regimes have the necessary symmetry, or lack resentation theory of spin-space groups that is the tool of it, to allow for nontrivial Berry curvature. of choice to determine the symmetry constraints on the Section VII is a detailed study of the Kitaev- magnon band structure (Section III). Indeed, our analy- Heisenberg hyperhoneycomb ferromagnet. This model sis is grounded in the representation theory of spin-space beautifully illustrates some important features of magnon groups and the associated compatibility relations. In band topology arising from spin-space groups because the contrast to the representation theory of magnetic space symmetry of the model can be tuned by simply rotating groups, this has not been worked out in detail and our the direction of the applied magnetic field. We give the study of band topology relies on calculations of band rep- full (spin-space) symmetry group corresponding to each resentations and their decomposition into irreducible rep- of the symmetry-distinct moment directions revealing a resentations of spin-space groups from first principles. hierarchy of magnetic symmetries. We also give the sym- Symmetry constraints on band topology have been ex- metry group that one would naively anticipate purely tensively analysed for space groups and their magnetic based on the invariance of the magnetic structure. The analogues. Here we show that the further enhancement latter is, by definition, a magnetic space group and there- of symmetry in going to spin-space groups can lead to a fore generally has lower symmetry than the spin-space proliferation of band degeneracies including nodal points, group for the same moment direction. lines, planes and volumes. Here “nodal” refers to points, We then enumerate all the magnon band degeneracies curves, surfaces and volumes where pairs of bands be- one would expect for each of these symmetry groups. come degenerate. This reveals, at a glance, that the spin-space group has 3 various features including nodal lines and Weyl points II. SPIN AND SPACE SYMMETRIES OF THE that would be absent for the corresponding magnetic MAGNETIC HAMILTONIAN space group. For many of these cases we can give sim- ple criteria or informal arguments for the appearance of A. Spin-Space Groups and Magnetic Couplings nodal features thereby circumventing the detailed repre- sentation theory analysis. Given a magnetic Hamiltonian H, we shall identify all symmetry operations that leave the Hamiltonian invari- We further show the surprising feature that the spin- ant. This parent group will be denoted GH . In general, space groups arising for this model are frequently iso- the Hamiltonian will be invariant under lattice symme- morphic to some magnetic space group albeit of higher tries that form a space group G that includes the prim- symmetry than the one that merely leaves the magnetic itive translations T as a normal subgroup and the space structure invariant. group then admits a coset decomposition [ G = {Rα| tα} T (1) α
where a general coset representative of the space group is conventionally denoted {R| t} where R is a point group element and t is a non-Bravais translation. These act on a general position r in real space as {R| t} r = Rr + t and so:
{R1| t1}{R2| t2} = {R1R2| R1t2 + t1} (2) −1 −1 −1 {R| t} = R −R t . (3) 3 4 The identity is denoted {E| 0}. There are 17 such groups 1 2 in two dimensions − usually called the wallpaper groups − and 230 in three dimensions. In the presence of time reversal symmetry, we allow for the possibility of anti- unitary elements T{ˆ S| w} leading to magnetic space groups M that have coset decomposition [ [ M = {Rα| tα} T + Tˆ {Sα0 | wα0 } T. (4) α α0
There are 80 magnetic space groups in two dimensions and 1651 in three dimensions. Magnetic Hamiltonians may have higher symmetry still: in general GH is a direct product of a magnetic space group and a group acting only in spin space. For example, consider the canonical Heisenberg model on some lattice with, for concreteness, nearest neighbor couplings X FIG. 1. The upper figure shows the tri-coordinated network Hˆ = J Sˆi · Sˆj. (5) of sites on a hyperhoneycomb lattice. The primitive unit cell hi,ji contains four sites labeled from 1 to 4. The primitive unit vectors for the 4-site unit cell are given by ai. The color The symmetries of this Hamiltonian include: of the bonds reflects the direction of the Kitaev interaction: SxSx (red), SySy (green), and SzSz (blue) interactions re- 1. the primitive lattice translations forming group T, spectively. The lower panel shows the Brillouin zone of the hyperhoneycomb lattice. The perpendicular planes are the 2. the symmetry elements denoted {α| tα} for α = mirror plane of glides: d1 (red), d2 (yellow), d3 (grey). The 1,..., |G| belonging to space group G, high symmetry paths are: Γ → Y → T → Z → Γ → X → A1 → Y ; T → X1; X → A → Z and Γ → L. The coordinates 3. the time reversal symmetry operator Tˆ: the oper- of the points are given in Appendix A. ˆµ Tˆ ˆµ ator acts on spin as Si −→−Si
4. and the group of global spin rotations R =∼ SO(3). 4 and combinations of these. The parent group is therefore its spin-only character or is mixed with the orbital mo- GH = G ⊕ TˆG ⊗R where G⊕TˆG on its own forms a ment is dependent on the magnetic ion and the material in which it appears but we now use Jˆµ to denote the mo- magnetic space group of type II in the notation of Bradley ment operators. In general, the single anisotropy takes and Cracknell [1]. the form We use the notation [Bk {R|t}] to denote the general symmetry element of those symmetry groups that allow X X l l HˆSIA = ∆mOˆi,m (10) for decoupled spin-space and real-space elements where i l,m B acts on spin space and {R| t} is the ordinary space ˆl group element that does not act on spin-space [41]. Such where Oi,m is a Steven’s operator which is the operator symmetry groups were named spin-space groups in the l equivalent of spherical harmonic Ym(θ, φ) and is polyno- original papers of Brinkman and Elliott [8, 9]. The action mial in the spin operators with degree l ≤ 6 as fixed by of a spin-space group element on the lattice moments the site symmetry. Jˆµ r active ( ) is the transformation We now give some concrete examples of possible spin X groups in lattices of moments arising from the single ion Bk {R|t} Jˆµ r B Bµν Jˆν {R| t} r [ ] ( ) = det( ) ( ) (6) anisotropy. While the single ion anisotropy has the site ν symmetry of the magnetic ion, there may be a hierarchy where the determinant is present because magnetic mo- of scales. For example, in tetragonal K2CuF4, the cop- ments are pseudovectors − for example they are invariant per is almost isotropic with exchange scale J ≈ 1 meV under inversion. It follows that because the spin-orbit coupling is weak. Nevertheless, it does have a detectable easy plane anisotropy of about −2 2 z 2 [B1k {R1|t1}][B2k {R2|t2}] = [B1B2k {R1R2|R1t2 + t1}] 10 J with single ion Hamiltonian Oˆ0 = 3(Jˆ ) −J(J+1) 2 (7) and ∆0 > 0. There is a even weaker but detectable four- −1 −1 −1 −1 fold anisotropy [47] that can be captured by a term in [Bk {R|t}] = B k R |−R t . (8) ˆ4 ˆ+ 4 ˆ− 4 HSIA of the form O4 = (1/2)((Ji ) + (Ji ) ). Thus, the In the case where the spin-orbit coupling vanishes, the 3D rotation group is broken by the easy plane anisotropy Hamiltonian is Heisenberg-like and the group elements down to U(1)×Z2. The weaker terms break this down to acting on spin space are completely decoupled from the the site symmetry D4 in principle allowing for five non- 2 4 4 6 real space elements − the former being the 3D rotation vanishing Steven’s operator coefficients ∆0, ∆0, ∆4, ∆0, 6 group. When spin-orbit coupling is present and Hamil- ∆4. There are materials where the easy plane anisotropy tonian is not fine-tuned, the moments are usually locked is much greater. to the space group transformations. We then write Analogous symmetry considerations guide our under- standing of interactions between magnetic moments. We X [Rk {R|t}] Jˆµ(r) = det(R)Rµν Jˆν ({R| t} r). (9) mainly restrict our attention to couplings that are bilin- ν ear in the moments: Since the magnetic Hamiltonian must be invariant under ˆ X µν ˆµ ˆν H = Jij Ji Jj . (11) lattice symmetries such locking is always possible and i,j the resulting group is one of the magnetic space groups. µν µν There are cases where the moments transform under lat- The Heisenberg coupling is Jij = δ Jij but, in gen- tice transformations but where there is a residual spin eral, the exchange may have anisotropies that respect space invariance − nontrivial elements that act purely the lattice symmetries. On a single bond, in the ab- on spin space. sence of symmetry constraints, there are nine allowed The magnetic Hamiltonian in many insulating mag- couplings: three diagonal, three off-diagonal and anti- nets is well approximated by a Heisenberg model in those symmetric and three off-diagonal and symmetric. Sym- cases where the spin-orbit coupling is weak [42], where metry generally places constraints on these couplings. the orbital part of the moment is quenched or frozen out For example, consider a simple cubic lattice and a near- [43] or through the fortuitous cancellation of anisotropic est neighbor xˆ bond. The C4 about the axis through terms [44, 45]. Such couplings are allowed by symmetry the bond takes yˆ → zˆ fixing the Jyy = Jzz. The mir- on all lattices as well as in amorphous solids. For ex- rors in the planes of the cubic faces are equivalent to ample, the parent material La2CuO4 of one prominent inversion and a C2. Inversion leaves the magnetic mo- high Tc superconducting cuprate which has a Heisenberg ment invariant so only the C2 acts nontrivially thus forc- exchange scale of about 100 meV [42] while any mag- ing the off-diagonal components to zero. The resulting xx yy yy netic anisotropies, for example inferred from the small Jii+xˆ = diag(J , J , J ) and the other nearest neigh- spin wave gap, are at most a hundredth of this scale [46]. bor bond coupling can be obtained from this using lattice In instances where there is spin-orbit coupling, the symmetries. Since the most general exchange Hamilto- crystal field may lead to single ion anisotropies that break nian has only the lattice symmetries, the spatial and spin the spin rotation group from SO(3) to the local site sym- transformations can be thought of as being locked to one metry group. The degree to which the moment preserves another and this is the limit of strong spin-orbit coupling. 5
◦ In this case, the group GH is just the group G ⊕ T G as same group D2 appears also for 90 compass models such αα there are no residual spin-space transformations that are as the simple cubic lattice model with Jii+αˆ along the decoupled from the real-space transformations. A de- αˆ = xˆ, yˆ, zˆ bonds. tailed worked example of this kind of argument is given An immediate implication of the above remarks for in Appendix B. magnon spectra is that given spin-space group of the However, this restriction, while strictly true in prin- Hamiltonian GH , the pure spin rotational part R of GH ciple, does not allow for the existence of a hierarchy of can be used to find new spin orientations that give the exchange couplings. For example, as we noted above, same magnon spectrum. To take an almost trivial ex- there are materials in which the Heisenberg coupling is ample: in the Heisenberg case, spin rotation invariance overwhelmingly the largest coupling. In other materials, means that the magnon spectrum is completely invariant there are various well-understood mechanisms (as well as to changes in the moment orientation. cases with merely a degree of fine-tuning), that can lead So far we have discussed purely magnetic models with to certain anisotropic couplings being significantly larger the aim of studying magnon band topology. However, than others. We consider various examples. spin-space groups that we discuss extensively here may Cobalt(II) in an octahedral crystal field has a relatively play a role in electronic systems too. For example, we small spin-orbit coupling that, in the presence of trigonal may minimally couple any of the spin-space symmetric distortion, may lead to a single ion ground state doublet magnetic exchange models (with Hamiltonian Hˆmag) to with easy plane anisotropy with a level splitting on the electrons through a Kondo-like Hamiltonian order of the exchange scale. An example is CoTiO3 for g /g ≈ . ˆ X † X † ˆµ αβ ˆ which ⊥ k 1 7 [48]. It is natural to write down an ef- H = tijciαcjα + h.c. + ciαJi · σµ ciβ + Hmag fective spin one-half model to understand the magnetism hi,ji,α i,αβ in this material and hence single ion anisotropy terms are (15) trivial. The easy-plane anisotropy must therefore be in- where the electronic band structure now inherits the spin- cluded through the magnetic interactions. Indeed, in this space symmetry of the magnetic subsystem. material the leading order description of the magnetism is in terms of an XXZ model J⊥ B. Symmetries of the Magnetically Ordered ˆ X ij ˆ+ ˆ− ˆ− ˆ+ zz ˆz ˆz H = Ji Jj + Ji Jj + Jij Ji Jj . (12) Ground State i,j 2 In this section, we show how the considerations of sym- and further anisotropies are sub-leading. The spin-space metry in the previous section must be adjusted in the group is therefore U(1) × 2. The same group appears Z presence of magnetic order. Since we are ultimately inter- for Dzyaloshinskii-Moriya (or anti-symmetric) exchange ested in magnons, we suppose the magnetic ground state with collinear D vector which may appear, for example, µ is characterized by local order parameter hJ i. From the on second nearest neighbor bonds of the honeycomb lat- i parent group GH − the group of operations that leave the tice and which may be important in the CrX3 magnets: magnetic Hamiltonian invariant − we identify the sub- ˆ X ˆx ˆy ˆy ˆx group of transformations that leave the magnetic struc- H = D Ji Jj − Ji Jj . (13) ture invariant GM . Frequently, the onset of magnetic or- hhi,jii der enlarges the unit cell thus breaking down the group A further example is Kitaev exchange. Consider a hon- of primitive translations to a subgroup. The wavevector eycomb lattice of magnetic moments with Ising couplings associated to the magnetic order may even be incommen- along perpendicular directions on the three bonds origi- surate significantly lowering the symmetry. In addition nating from each lattice site. Thus: to the translation symmetries there will tend to be com- binations of translations, point group operators on the ˆ X ˆγ ˆγ H = K Ji Jj (14) lattice and spin transformations that leave the magnetic structure invariant. Unlike the parent group, GM is gen- hi,jiγ erally not a simple product group. Instead the spin and where γ runs over x, y, z and identically oriented bonds space transformations tend to be coupled. belong to the same Ising component. If we now perform In the case where GH = G ⊕ TˆG − in other words, a global spin-space rotation about any of the cubic spin- when the spin transformations are locked to the space space axes x, y, z, the Hamiltonian will be left invari- group transformations − the subgroup that leaves the ant. So the spin-space group is isomorphic to point group magnetic structure invariant is another magnetic space D2. Kitaev-Heisenberg models can arise on several lat- group. Since the magnetic order breaks physical time tices with edge-sharing octahedra that supply a superex- reversal symmetry and G contains the identity GM can- change mechanism to generate such couplings. Such lat- not be of the form G ⊕ TˆG but must instead be a type I tices include the honeycomb lattice, its three-dimensional group − one with no anti-unitary elements, or a type III generalizations including the hyperhoneycomb, the py- or IV magnetic space group of the form G ⊕ AˆH where rochlore lattice, the kagome lattice and so on [12]. The Aˆ = TˆUˆ is an antiunitary element and Uˆ is a nontrivial 6 D E H ˆα unitary and is a (unitary) space group. Jia , for the local order parameter with i running over In the other extreme case of Heisenberg models, mag- the N magnetic primitive cells and a running over the netic order breaks the decoupled spin space and real m magnetic sublattices on a finite lattice. We further space transformations down to a discrete subgroup. For suppose that the moments are written in a local quanti- example, if the magnetic structure is collinear then the zation frame defined with local z component zˆa along the space group transformations acting only on real space D E ordered moment direction as in Jˆα ≡ Sδαz and with leave the moments invariant there are pure spin space ia rotations about the axis of the moments as well as uniform ordered moment from site to site. In principle, Tˆ [C2⊥k {E|0}] where the C2 rotation is about an axis we could consider ordered structures with ferrimagnetic perpendicular to the ordered moment. textures or models with different types of magnetic ion via a straightforward extension of the methods described here. Concurrently, we introduce local transverse direc- tions, xˆa and yˆa that may be chosen arbitrarily - ob- C. Magnon Symmetries servable quantities should not depend on the choice of transverse axes - so there is a local phase invariance. Here we show in outline how to determine the sin- The angular momentum operators are bosonized gle magnon excitation spectrum and the relationship be- through the Holstein-Primakoff representation for spins tween the symmetries of the magnetic Hamiltonian and of size S those of the magnons. We consider the following general ˆz † exchange Hamiltonian for localized moments defined on J = S − aˆ aˆ (17) r some lattice and including a Zeeman term √ aˆ†aˆ √ aˆ†aˆ Jˆ+ = 2S 1 − aˆ = 2S 1 − aˆ + ... (18) 2S 4S 1 X αβ ˆα ˆβ X α ˆα r H = JiajbJiaJjb − B Jia. (16) √ † √ † 2 − † aˆ aˆ † aˆ aˆ ia,jb;α,β ia,α Jˆ = 2Saˆ 1 − = 2Saˆ 1 − + ... (19) 2S 4S
αβ βα The couplings have the symmetry property Jia;jb = Jjb;ia. where the bosons satisfy the usual commutation relations We are supposing that the moments have nonzero expec- [ˆa, aˆ†] = 1. tation values either through spontaneous or field-induced Expanding about the mean field ground state leads to magnetic ordering. And we use the following notation, the quadratic Hamiltonian
S X † A(k) B(k) S X † HSW = Υˆ (k) ? ? Υˆ (k) ≡ Υˆ (k)M(k)Υˆ (k) (20) 2 B (−k) A (−k) 2 k k
where The diagonalizing transformation on Eq. 20 to find the spin wave spectrum, aˆk1 . . U † k M k U k Λ k . ( ) ( ) ( ) = ( ) a † † † † ˆkm U(k)ηU (k) = η Υˆ (k) = aˆ ... aˆ aˆ−k1 ... aˆ−km Υˆ (k) = † k1 km aˆ −k1 . where Λ(k) is diagonal, must preserve the commutation . † relations aˆ−km h †i (21) Υa, Υb = ηab (24) and the Aab(k) and Bab(k) depend on the exchange cou- plings in the local frame as follows: where ab = 1 if a = b ≤ m and ab = −1 if a = b ≥ m+1 X and zero otherwise. A k ˜J+− k − δ ˜Jzz 0 ab( ) = ab ( ) ab ac( ) (22) It is straightforward to see that c 1 xx yy xy yx ηM(k)U(k) = U(k)ηΛ(k) (25) Bab(k) = ˜J (k) − ˜J (k) − i˜J (k) − i˜J (k) 2 ab ab ab ab ˜−− so the diagonalizing transformation can be found by solv- = Jab (k). (23) ing this non-Hermitian eigenvalue problem. This diago- Note that these expressions with the factor one-half de- nalizing transformation has important consequences for ˜αβ fine Jab for α, β = ±. magnon band topology [49]. In short, the ten-fold way 7 that classifies single particle fermion problems in the ab- magnon interactions are important their effect can be re- sence of lattice symmetries, is reduced to a three-fold duced, at least in principle, by applying a large enough way with only time reversal symmetry that can be either magnetic field. Although we shall not consider magnon absent, or present and squaring to ±1. This has the con- interactions further in this paper, it is worth mention- sequence of leaving only Chern insulators in 2D, or Z2 ing that the symmetries of the magnetic Hamiltonian are topological bands in 2D or 3D as gapped bands. symmetries of the magnon Hamiltonian order-by-order in Symmetries of the spin wave Hamiltonian, Eq. 20, the interactions. It follows that any symmetry protected are inherited from the symmetries of the full magnetic degeneracies in the single magnon spectrum cannot be Hamiltonian that leave the magnetic structure invari- broken by higher order terms. ant. In other words, the appropriate symmetry group Although spin-space symmetries survive magnon inter- is GM and a particular representation of the symmetry actions, it is possible for the symmetries of the linear spin elements is determined from the transformation proper- wave Hamiltonian to be higher than one would expect ties of the transverse spin components in the local quan- on the basis of the foregoing discussion. In other words, tization frame since the Holstein-Primakoff bosons are the quadratic magnon Hamiltonian may have accidental related to them through symmetries. These are most commonly discussed in rela- √ tion to order-by-disorder where an accidental symmetry ˆ+ Jka → 2Saˆka (26) appears at the mean field level and in linear spin wave √ Jˆ− → 2Saˆ† . (27) theory [48, 50–52]. One example is the U(1) symmetry ka −ka of strong spin-orbit coupled pyrochlore moments [53, 54]. Under a unitary element of GM , the transverse spin com- This U(1) is absent in the magnetic exchange but appears ponents will transform into one another via a permuta- in linear spin wave theory and is broken down to the dis- tion of magnetic sublattices, a rotation − that, in the ± crete lattice symmetries by magnon interactions. In the frame amounts to an overall phase − a transformation in case of order-by-disorder both the ground state and linear momentum space and a translation. More precisely, for spin wave spectrum have an accidental symmetry. There spin-space transformation [Bk {R|t}] we may write are also cases where there is no order-by-disorder but still the linear spin wave spectrum is more symmetric [55]. X Uˆ ˆ k Uˆ † e−iRk·t U ˆ Rk In this paper, we concentrate our attention on features gS Υam( ) gS = [ B]ab Υbm( ) (28) b that survive magnon interactions as they are protected by spin-space symmetries that appear to all orders of spin ab −iφS where [UB]ab is a matrix of phase factors e where wave theory. indices a and b are constrained by the real space transfor- mation i+a = R(j +b)+t. The boson spinor Υˆ am is in- dexed by sublattice a and particle-hole index m = 0, 1. In III. REPRESENTATION THEORY AND order to treat antiunitary elements we require the trans- MAGNON BAND TOPOLOGY formations under time reversal
TˆiTˆ −1 = −i (29) In the following sections, we work out spin-space sym- metry constraints on magnon band structures using rep- ˆ ˆ± ˆ −1 ˆ∓ T JiaT = −Jia (30) resentation theory. The approach we take is to work ˆ ˆ± ˆ −1 ˆ∓ from the atomic limit and build band representations in- T JkaT = −J−ka. (31) troduced by Zak in Ref. 56. In this section we outline In ensuing sections, we describe in detail how to construct how to construct band representations for magnons. The band representations of the full spin-space group. reader who is content to skip the details can see an out- Before that, we make a couple of further important re- line of the method in the next paragraph. marks. The first concerns the effect of higher order terms In a nutshell, the band representation ties together all in the spin wave expansion. Such magnon interaction symmetry information about the band structure at dif- terms in general couple single magnon to multi-magnon ferent momenta using group elements and local atomic states leading to damping and renormalization of the sin- orbitals as ingredients. From the band representation, gle magnon modes. In all of the following we assume one may extract the irreducible representations at dif- that the effect of magnon damping is sufficiently weak ferent symmetry-distinct momenta as well as informa- that the single magnon states have a lineshape that is tion about the connectivity of these irreps. The building negligible compared to the magnon bandwidth. In other blocks of the band representation for magnons are the words, we shall rely on a robust single particle picture on-site transverse spin components in the local quanti- throughout. In practice, this assumption is often very zation frame, Jˆ±. These components form a basis for a reasonable. Indeed most magnetic materials are in such representation of a group that leaves the lattice site in- a regime and magnon interaction effects tend to be se- variant. Since this site symmetry group is a subgroup of vere only in geometrically frustrated magnets with a high the full spin-space group one may carry out a well-defined density of low-lying states and especially those with non- induction procedure to obtain a representation of the full collinear moments or anisotropic exchange. Even where group that has dimension equal to the number of bands 8
and that is also a function of momentum. We now describe the process in more detail. Take a Kitaev limit point r in the primitive cell and act on it with elements of the spin-space group g ∈ GS. Those elements that leave the point invariant form a group Gr called the site symmetry group or stabilizer group − that may include translations. By construction, elements of the stabilizer group leave the magnetic moment invariant at the site. The stabilizer group will generally have both unitary and Heisenberg limit anti-unitary elements. Now act on the point r with elements g ∈ G that are not in the stabilizer group. The set of points thus defined 3.0 2.5
JS {ra = gar|ga ∈/ Gr} for a = 1, . . . , n (32) / 2.0 1.5 1.0 is associated to a Wyckoff position with multiplicity n Y Z L n wavevector where is the number of points generated in this way that live in the primitive cell. The stabilizer groups as- sociated to these points are isomorphic. FIG. 2. (Main Panel, top left) Magnon spectrum for the N´eel Starting from the atomic limit, we take a set of or- phase and ϕ = 0.35π (where ϕ is defined following Eq. 40) bitals − essentially maximally localized Wannier func- with the band degeneracies indicated. Nodal lines between tions Wia(r) − forming a representation ρ of the sta- bands (1, 2) and (3, 4) are shown as green lines and those be- bilizer group Gr. Suppose there are nr such functions. tween bands (2, 3) as red lines. There is a nodal plane (green h surface) and 4-fold degenerate points (black points). The in- Then for each element of the stabilizer group the rep- set at the bottom right shows the magnon dispersions along resentation ρij(h) has indices that run from 1 to nr. the high symmetry path Y → Z → L in the vicinity of the Now, given representation ρ of the stabilizer group, one 4-fold degenerate point showing the double degeneracy along may induce to a representation of the full space group [YZ] in the nodal plane. Section IV B accounts for all these that we write as ρ ↑ G. In real space, the dimension features through a symmetry analysis. (Insets) The insets at the upper right show two coupling limits − the Kitaev limit of the representation is (nr × n × N) × (nr × n × N) ϕ = 0.5π and Heisenberg AFM limit ϕ = 0. In the Kitaev where N is the number of primitive cells. In momentum limit the bands (1, 2) are degenerate everywhere in the zone, space, the translations are diagonalized and the momen- bands (2, 3) form two tilted nodal planes (red surfaces) and tum dependent representation is a matrix of dimension bands (3, 4) nodal planes on the high symmetry surface (green (nr × n) × (nr × n) that acts on the Fourier transformed surfaces). The two types of nodal planes meet along line [ΓZ] Wannier functions: (black line) which is indeed 4-fold degenerate, and can be seen as the continuous shrinking of the circular nodal loop at ϕ = 0.35π. In the Heisenberg limit the modes form a doubly X −ik·tµ aia(k, r) = e Wia(r − tµ). (33) degenerate nodal volume over the entire Brillouin zone be- µ tween bands (1, 2) and (3, 4) (green volume). In addition the bands are four-fold degenerate on some high symmetry lines (black). A formula for the induced representation for g ∈ G is
−i(gk)·tba −1 (ρG(g))ia;jb (k)ajb(k, r) = e ρji gb {E| −tba} gga ajb(hk, r) (34)
where crete momentum space points of distinct symmetry in- cluding constraints on the connectivity of the bands in tba = gra − rb (35) the Brillouin zone. is a Bravais lattice vector. To each site a there is exactly At each point in the Brillouin zone, k, the little group one b and these are related through ggb = {E| tba} gbh Gk consists of elements h ∈ G such that hk = k which where h ∈ Gr. This band representation contains com- is a momentum diagonal block in the band representa- plete symmetry information about the bands at the dis- tion. We obtain a representation of the little group at 9
k k, ρG = ρG ↓ Gk. By modding out translations we ob- Now consider two high symmetry points k1 and k2 tain the little co-group, G¯ k at this wavevector that is joined by a high symmetry line k1 +λ(k2 −k1). The sym- isomorphic to some spin point group. This is generally metry group along the line is a subgroup of the groups at composite and can be decomposed into irreps using the the two endpoints. It follows that, at each high symmetry (α) point, the symmetry group associated to that point corre- orthogonality of irreps ρk sponds to a set of irreducible representations Xa that, in k M (α) general, are reducible under the subgroup along the line ρG = nαρk . (36) α connecting the endpoints. In terms of the characters, for each irrep Xa at a high symmetry point, and irreps Yb α Thus the number of times the irrep occurs is given by along the high symmetry line, there is a compatibility h relation 1 X ∗ nα = χ (α) (m)χρk (m) (37) X h ρk G m=1 χ (Xa) = χ (Yb) (39) b where the sum is over distinct relevant classes m, χσ is the character of the representation σ, and h is the order of with a similar condition at the other endpoint of the line. the group. In this way, we obtain the symmetry distinct Representation theory therefore supplies a discrete no- modes at each wavevector. In practice, the decomposi- tion of band connectivity in momentum space. For, given tion requires the character table of the spin point group. the magnon group representation at each high symmetry Later on we give some examples of this decomposition. point and line in the zone, there is a set of energy levels at So far we have the representation theory for magnons each labelled by some irrep. Then the compatibility rela- in the case where the spin-space group is unitary. When tions constrain the ways in which these levels connect to there are anti-unitary elements there are some impor- one another to form a continuous band structure through tant new features. The 1651 conventional magnetic space the zone. Depending on the ordering of the irreps in en- groups appear in four types: (I) the ordinary space ergy − that is not fixed by symmetry − the compatibility groups (230 in all), (II) paramagnetic groups of the form relations may enforce crossings between bands. G ⊕ T G (230), (III) H ⊕ T (G − H) (674) and (IV) the black and white magnetic groups G ⊕ T {E| t} G (517). It will be useful to bear these in mind as, in later sections, IV. NODAL POINTS, LINES AND PLANES we show that certain spin-space groups are isomorphic FROM SPIN-SPACE SYMMETRY: AN EXAMPLE to magnetic space groups. In general, a magnetic space group takes the form G ⊕ AG where A is some anti- A. Model and Overview of Results unitary element. In principle one can construct the full band represen- In this section, we illustrate the proliferation of nodal tation for the spin-space group including magnetic ele- features in band structures caused by spin-space symme- ments. Then at a given momentum one can determine tries using the example of a Kitaev-Heisenberg model on the decomposition into irreducible co-representations (or a tri-coordinated lattice in three-dimensions: the hyper- coreps) of the magnetic little co-group. However, since honeycomb lattice (Fig. 1). The Hamiltonian is our principal focus is the symmetry-enforced degenera- ˆ X X γ γ X cies, it is possible to side-step this process and find the H = J Ji · Jj + K Ji Jj − h · Ji (40) band representation for the unitary part of the group
field: the collinear ferromagnet, a N´eelphase, and two along a plane. The magnon band dispersions through the further antiferromagnetic phases called skew-stripey and Z point in a nodal plane direction (Y → Z) and an out skew-zigzag. of plane direction (Z → L) are shown in the lower panel The hyperhoneycomb lattice is also the iridium Ir4+ of Fig. 2. The aforementioned nodal loop shrinks as the sublattice in β-Li2IrO3. The oxygen ions in this mate- Heisenberg coupling is reduced, forming a 4-fold degen- rial form a lattice of edge-sharing octahedra such that erate line along Γ → Z in the Kitaev limit. One further the Ir-O-Ir bond angle is 90◦ and this geometry provides symmetry-enforced nodal line runs along the line Z → A the basis for a microscopic mechanism leading to Kitaev- (and equivalently by reciprocal vector translation along Heisenberg couplings [18, 19]. The spiral ground state Y → A1). of this magnet in zero field is suggestive of the presence B. Representation Theory of significant off-diagonal symmetric, or Γ, exchange in this system and, in general, we expect materials to de- We saw in Section III how to proceed from localized or- viate from the ideal Kitaev-Heisenberg model. However, bitals to band structures using symmetry considerations when there is a family of magnetic materials with similar alone. Here we show that the representation approach crystal structures a certain degree of fine-tuning is com- allows us to account for the symmetry protected features patible with the existence of materials proximate to the of a magnon model and enumerate the kinds of magnon Kitaev-Heisenberg limit. band topology that can arise for the given symmetry. The particular example we take in this section is the x The full group is GNeel = HNeel + T C (s)HNeel with hyperhoneycomb Kitaev-Heisenberg N´eelantiferromag- 2 coset representatives net in zero applied magnetic field. We note that there is an order-by-disorder mechanism that fixes the moments H to lie along one of the Cartesian axes and, without loss Neel = a x,y b c z of generality, we choose this to be the [001] direction. E + d1 + d3 + C2 + C2 (s)(P + d2 + C2 + C2) + C2 (s) This example will turn out to have enhanced magnetic (41) symmetry described by a spin-space group that is not iso- morphic to a magnetic space group and, therefore, that giving 16 elements in all and then there are translations could not be inferred simply by requiring that the mag- in addition to these. Here we have used a notation for netic structure be left invariant by space group transfor- the symmetry elements where pure spin transformations mations. are labelled with s and the remaining elements are com- Fig. 2 illustrates features of the magnon spectrum bined spin and space transformations − they are locked within the N´eelphase for non-vanishing Kitaev coupling to one another. All symmetry elements are defined in (ϕ = 0.35π). The main panel shows the band degen- Appendix A and a short introduction to group notation eracies within the first Brillouin zone. There is one is given in Appendix C. plane (shown in green) where the four magnon bands are The hyperhoneycomb lattice belongs to Wyckoff posi- symmetry-enforced to pair up into two two-fold degen- k tions 16g. The magnon band representation (BR) ρG is erate bands. This plane is intersected by several nodal induced from the representation ρ16g of spin transverse loops that can be inferred from compatibility relations. S⊥ components (J +,J −) around the ordering vector. They are protected by glide symmetry on their respec- tive mirror planes. One of these, between bands two and If we take as representative of the orbit the position q16g r three, is confined to the plane through the Γ point per- 1 = 1, the spatial site-symmetry group is: pendicular to the nodal plane. The existence of this nodal ∼ G 16g = {E| 0} + {2001| −1/4, −1/4, 0} = C2 (42) loop implies that the nodal plane bands meet at a four- q1 fold degenerate nodal point (shown in black) at the Z point. The band structure in the vicinity of the Z point, If we now consider also the possible additional spin ro- plotted with one axis in the Γ −Z −Y plane and another tations we get the enhanced magnetic (spin-space) site- SS perpendicular to this plane, is therefore a double cone symmetry group G16g (noting that spin rotations leave emanating from a single point where the two cones touch the position invariant):
SS + − G 16g = [Ek {E|0}] + [2010k {E|0}] + 4010k {2001|−1/4, −1/4, 0} + 4010k {2001|−1/4, −1/4, 0} q1 0 0 0 0 + [2010k {E|0}] + [2101k {E|0}] + [2100k {2001|−1/4, −1/4, 0}] + [2001k {2001|−1/4, −1/4, 0}]
Note that the a, b, c coordinate system is used here and addition we separate out the spin and space transforma- in all the group theory calculation (see Appendix A). In tions using notation introduced in Section II A. The first 11
16g 16g line is the unitary part which is isomorphic to the C4 and specifically ρS+ = Γ3 and ρS− = Γ4. Here it is point group. In the second line there are anti-unitary important to note that these two reps are complex con- elements (prime sign) which always give [E||E] when jugates of each other, as required by the relationship be- squared (the translation part is nonzero because of the tween the transverse spin components S+ (that maps to choice of origin and cancels out). From the anti-unitary the Holstein-Primakoff a operator) and S− (that maps elements we obtain (a) coreps (see Eq. 38) and so no to a†). extra degeneracies in the irreps of C4 are expected. + − To induce the local representation to the full group The transverse components (J ,J ) therefore trans- ρ ρ16g ↑ GSS ( S⊥ = S⊥ ) we need to consider all the orbits of form as irreps of C4. The symmetries act on the global SS the Wyckoff position {qα = gαq1 | gα ∈ G }, α = 1, ..., n conventional frame, while the transverse components are with multiplicity n of the Wyckoff position. For 16g the around the ordering directions, so in the local frames. multiplicity inside the primitive cell is 4 (while 16 in the T The spin transformations in local frames fS are then ob- conventional cell), therefore we can choose representative T RT T R i tained as fS = i S i where index appears in both gα as: rotation matrices because the symmetries considered here are site preserving. In this case, the matrix representa- + − tion of the spin rotations in the local (J ,J ) basis is: g1 = [Ek {E|0}] , g2 = [2-101k {2010|−1/4, −1/4, 0}] g3 = [Ek {m010|1/4, 0, 1/4}] , g4 = [2-101k {−1|0}] − ∓i D 1 0 D ± 0 . (45) (2010) = − (4010) = ±i (43) 0 1 ± 0 ±
Comparing with the character table of C4 we see that the Now we have all the ingredients necessary to use the gen- representation decomposes to: eral formula for induction. Since the characters of the band representation include all the information we need, ρ16g S⊥ = Γ3 + Γ4 (44) we can simply note that:
P −1 SS α exp −i (h k) · tααχρ16g (gα [Ek {E|−tαα}] h gα) h ∈ G 16g S q1 χ k (h) = ⊥ (46) ρG,S SS ⊥ 0 h∈ / G 16g q1
where tαα = h qα − qα. The second line is always zero the Brillouin zone and, for symmorphic groups, also on h∈ / GSS since if q16g then the symmetry will permute the boundary points. Indeed in these cases we need only find 1 ¯ k sublattices giving an off-diagonal band representation the point group isomorphic to the little co-group GSS (lit- matrix. Therefore for the N´eelcase we will have: tle group without primitive and non-symmorphic transla- tions) and decompose into the irreducible representations
χρk ([2010k {E|0}]) = −4 (47) of that group. G,S± ± χρk ( 4010k {2001|−1/4, −1/4, 0} ) = 0 (48) The Nodal Surface − G,S± For example we can explicitly calculate the enforcement of nodal surface E = (0, u, w) Now we have the full band representations and we can (plane Γ − Z − T in conventional basis) for the N´eel an- ρk ρ ↓ Gk ± subduce it to different little groups S⊥ = S⊥ SS. tiferromagnet with [00 1] moments. The little group The representations of the enhanced magnetic little (factored out primitive translations) on this high sym- k groups GSS are straightforward to find for points k inside metry surface is:
E + − GSS/T = [Ek {E|0}] + [2010k {E|0}] + 4010k {m100|0, 1/4, 1/4} + 4010k {m100|0, 1/4, 1/4} 0 0 + 0 − 0 + [Ek {−1|0}] + [2010k {−1|0}] + 4010k {2100|0, −1/4, −1/4} + 4010k {2100|0, −1/4, −1/4} where we note that the spin and space transformations are coupled but distinct, highlighting the importance of the enhanced symmetry coming from the internal spin symmetry. The unitary part is isomorphic to C4 with character table I. The coreps will therefore be given by the test: ( ¯ E + + X E 2 E E 4 = |GSS| Type (a) if p = E1 ,E2 χp (hk0 ) = 2 (χp ([Ek {E|0}]) + χp ([2010k {E|0}])) = − − (49) 0 Type (c) if p = E1 ,E2 hk0 12
¯ Z∗ where hk0 are all the anti-unitary elements of irreps of GSS can be obtained by conjugating the ones + the little co-group (such that hk0 k = −k + of the subgroup D4h by the symmetry 4010k {m100|0} . gi). We obtain therefore a doubly-degenerate corep Of these irreps we are only interested in the ones with − − − DE (2) = (E1 ,E2 )(2) with χ([2010k {E|0}]) = −2 and ∆([Ek {E|0}] , 1) = − I. The table of relevant irreps ± χ( 4010k {m100|0, 1/4, 1/4} ) = 0. is given in Table IV and these are all two-dimensional. ρk We then return to the anti-unitary elements and look Since we know the band representation G,S⊥ for ev- ery k we can now subduce it to the surface E and from for additional degeneracy in the corepresentations. The its characters in Eqs. 46 and 47 we get (the number in standard test reveals that the magnon bands belong to parenthesis indicates the dimension of corep): corep with class (c) binding two two-dimensional irreps. The overall degeneracy is therefore four-fold as was to be ρE DE− shown. S± = 2 (2) (50) Other symmetry constraints on the band structure − So we conclude, on the basis of spin-space symmetry, that All the degeneracies of the Neel case can be seen in the magnons on plane E are two-fold degenerate where Fig. 2. Here the group theory enforced degeneracies are the anti-unitary symmetry and the resulting binding of the nodal plane, the 4-fold degenerate points and the irreps is responsible for the degeneracy. straight lines A = [ZA], [YA1]. All the other curved lines Four-fold Degenerate Nodal Point − To analyse the are given by compatibility relations and are protected by degeneracies of the E plane, we were able to rely on glide symmetries on mirror planes. The bands (2, 3) (red) the little co-group being isomorphic to a point group are degenerate on a chain of loops on mirror planes d1 and use standard tables to decompose the representa- and d3. The band (1, 2) (green) has a closed nodal line tion into irreps. But, when k is a boundary point and on mirror plane d3 intersecting the nodal plane. there are non-symmorphic elements, the situation is, in principle, more complicated and we may need to consider projective representations. A representation is said to be V. SPIN-SPACE GROUPS AND NODAL projective when ∆(hi)∆(hj) = µ(hi, hj)∆(hk), where ∆ VOLUMES k are matrix representations of group elements hi ∈ GSS and µ(hi, hj) = exp(−igi · wj) is an element of the fac- Heisenberg models have the property that the spin −1 tor system, with gi = hi k − k and wj the transla- space part of the symmetry group is completely decou- tion associated to hj. If µ(hi, hj) = 1 for all cases pled from the spatial part. In the paramagnetic phase, then we reduce to ordinary (non-projective) representa- this symmetry group is the three dimensional rotation tions. If this is not the case we proceed by studying the group. The effect of the spin group on the magnon band representations of the central extension of the little co- structure for Heisenberg models has some general fea- ¯ k∗ ¯ k group GSS = GSS ⊗ Zg with kernel Zg, the cyclic group tures that we discuss in this section. While this section of integers 0, 1, ..., (g − 1). The number g comes from largely reviews known results [8, 9], it is useful to re-visit the parametrization of the factor system as µ(hi, hj) = them and cement their spin-space origin before breaking exp(2πia(hi, hj)/g), where a(hi, hj) = 0, 1, ..., (g−1) and new ground. the group elements are of the kind (hi, α) with prod- The simplest case is that of the Heisenberg ferromagnet uct rule (hi, α)(hj, β) = (hihj, α + β + a(hi, hj)). Of on an arbitrary lattice. The ground state is collinear, the ¯ k∗ all the irreps of GSS we are interested only in the ones magnetic structure preserves the translational symme- giving the right factor system, that is the ones with try of the lattice and, because spin space and real space ∆(E, α) = exp(2πiα/g) I. Since the set of elements are decoupled, the other lattice symmetries are also pre- ¯ k (hi, 0) is isomorphic to GSS, we can now extrapolate the served. It follows that the symmetry group of the mag- character tables of those irreps and build the table of pro- netic Hamiltonian that preserves the magnetic structure ¯ k jective irreducible representations of GSS (and therefore is the space group of the underlying lattice times a spin- k the one of GSS, adding the right phase factors coming space group GM = G ⊗ GS where the elements of GS from translations). are: axial rotations through angle φ about the moment These considerations are relevant to the point Z = direction, TˆC2 where the C2 is about an axis perpen- (0, 0, −1) (in the conventional basis), which in the N´eel dicular to the moment direction − the choice of axis is [00 ± 1] case is four-fold degenerate. This point is highly unimportant as a change in the axis may be absorbed Z symmetric − the little group GSS is the full spin-space into an axial rotation. Because spin and real space have group. The details of the representation theory for this decoupled, we label the magnon eigenstates with irreps point can be found in Appendix D. Here we summarize coming from the space group and an irrep from the spin the chain of reasoning. First of all one can show that the group. This means that there is no interplay between the factor system at the Z point is nontrivial. In particular, spin and space parts of the symmetry group. µ([2−101k {−1|0}]) = −1. The unitary part of the spin group is the continuous ¯ Z∗ We then find the central extension group for GSS with group C∞. This is the only nontrivial element of the lit- g = 2. This is a group with 32 elements that is iso- tle cogroup at a general position in the zone for a general ¯ Z∗ ∼ + morphic to GSS = D4h + D4h × ( 4010k {m100|0} ). The ferromagnet. This symmetry element leads to an infinite 13
E + − GSS [Ek {E|0}] [2010k {E|0}] 4010k {m100|0, 1/4, 1/4} 4010k {m100|0, 1/4, 1/4} Type Coreps + E1 1 1 ξ ξ (a) + E2 1 1 - ξ - ξ (a) − E1 1 -1 ξ - ξ (c) − E2 1 -1 - ξ ξ (c)
E 1 1 π TABLE I. Character table of the unitary part of GSS. The phase factor is ξ = exp(i E · (0 4 4 )) = exp(i 2 (u + w)). number of irreps labelled by integers but the invariance elements of the magnetic moment under the group operations im- ± ∓iφ ± GGP Ck s C⊥k {E|τ } ∼ Dk s plies that the rotation is fixed to Jia → e Jia. In other AFMbw = ( n( ) + 2 ) = n( ) (53) words, the rotation of the transverse spin components ⊥ picks out irreps with n = ±1 and also these rotation The C2 k {E|τ } mixes the two sublattices thus binding operators are diagonal operators in the basis of trans- the 1D irreps of C∞ into 2D irreps of D∞. This accounts verse components. Invariance of the magnon Hamilto- for the double degeneracy of the modes. For certain lat- nian under such axial rotations (or otherwise, inspection tices of this sort including the cubic lattice, there is an of Eq. 23) shows that Bab(k) = 0 for collinear Heisenberg additional symmetry at the general position coming from ferromagnets. This ensures that magnons are eigenstates the lattice inversion symmetry that swaps the magnetic of the global transverse spin rotations. Diagonalizing the sublattices Hamiltonian unitarily then reveals that the upper A block GP k ⊥ G = (C (s) + C k {E|τ } )(E + TP) components and the lower A components are eigenstates AFMbw+I n 2 ∼ k of the Hamiltonian but with different irrep label n. It = Dn(s)(E + TP) (54) follows that the magnons are each labelled by a common 1D irrep of C∞. They also have an irrep label originating The additional PT symmetry, taking the group into one from the space group symmetry. For collinear Heisenberg isomorphic to a type II magnetic dihedral point group 0 ferromagnets, any degeneracy in the magnon spectrum Dn1 , does not further bind irreps at the general position. that is enforced by symmetry must come from the space In passing, we mention a case where a volume degeneracy group symmetries. can arise on a black-and-white lattice in the presence of a ˆ2 For collinear ferromagnets with inversion symmetry single ion anisotropy of the form O±2 thanks to inversion that maps magnetic sublattices into themselves, the non- symmetry [8]. Here the general position has trivial part of the group at a general position is larger h i GGP Cz s Cx/yk {E|τ } E T Cxy s P AFMbw+I = ( 2 ( ) + 2 )( + 2 ( ) ) GGP = Ck(s)(E + T C⊥(s)P) (51) FMI n 2 ∼ k xy ∼ 0k = D2(s)(E + T C2 (s)P) = D4 (s) (55) and, in particular, it contains an anti-unitary part. The ± ± k symmetries act on the moments like: P : Ja → Ja , The unitary part D2(s) has 1D irreps, but anti-unitary ± ∓ ⊥ ± ∓ x xy T : Ja → −Ja and C2 (s): Ja → −Ja (where we C4 are created by tilted C elements and T C . The y 2 2 ⊥ 0k choose C2 (s) = C2 (s)). Therefore the additional ele- ⊥ final isomorphism is D4 (s), a type III magnetic point ment T C (s)P acts like an identity and does not mix 0 0 2 group (explicitly 4 2 2, i.e. with anti-unitary C4 rota- the irreps coming from the axial rotations. tions) which has 2D irreps at the general position. The next simplest type of Heisenberg model is one with For collinear antiferromagnets that do not lie on black- a collinear antiferromagnetic ground state. The two mo- and-white lattices − lattices composed of a pair of sub- ment directions may be related by lattices separated by a translation − such as the honey- comb and hyperhoneycomb lattices, the magnon bands J x → −J x J y → J y J z → −J z a b a b a b (52) are doubly degenerate originating from inversion sym- metry. At the general position where z is the quantization direction. As in the ferromag- netic case, C∞ is an element at a general position and GP k GAFM = Cn(s)(E + TP). (56) the oppositely oriented moments transform under irreps I of this group with opposite signs +1 and −1. The key here is that the inversion symmetry swaps the ± Let us first consider the classic example of a collinear magnetic sublattices. The inversion acts as P : Ja = ∓ ± ± antiferromagnet on a black-and-white lattice that in- −Jb , therefore PT : Ja = Jb mixes the two magnetic cludes the cubic lattice with a simple N´eelground state sublattices, pairing the conjugate irreps into 2D irreps. GP and the rutile lattice [8]. Rather than enumerating all Indeed, in this case, GAFM is isomorphic to a type II I 0 the symmetry elements we observe that there is a simple magnetic cyclic point group Cn1 which gives 2D irreps translation τ that maps the magnetic sublattices into for n > 2 (for C2 there are no conjugate irreps, only a one another. Then the general position has symmetry single real one). When there is inversion symmetry that 14 does not swap the magnetic sublattices, the analysis is similar to the ferromagnetic case considered above. We have established that the hyperhoneycomb antifer- romagnet has a double degeneracy at the general posi- tion. In fact, it has more degeneracy still enforced by lattice symmetries at more symmetric wavevectors as we show in Appendix E.
VI. ANISOTROPIC HONEYCOMB LATTICE MODELS
We turn now to the honeycomb lattice in 2D with anisotropic exchange couplings of different types as in Fig. 3 and use k · p arguments to understand the role of (a) Kitaev spin space symmetries in the protection of Dirac points and how these can be gapped out leading to topological bands. The first model we consider is the much-studied nearest neighbor Kitaev-Heisenberg model. The relevant point group under which the exchange is left invariant is D3d with nontrivial symmetries including three-fold rotations about the hexagonal centers, three two-fold rotational symmetries about axes through opposite hexagonal ver- tices and inversion symmetry again about the hexagonal centers. In addition there is the pure spin space D2 sym- metry discussed above and time reversal symmetry. The Kitaev-Heisenberg model has a rich semi-classical (b) Second neighbor Dzyaloshinskii-Moriya exchange phase diagram in an applied magnetic field [32]. In the polarized phase − both the zero field ferromagnet and the continuously connected field-polarized regime − the two magnon bands are connected at Dirac points when the field direction is along meridians [xy0], [x0z] and [0yz] relative to the Cartesian frame that defines the Kitaev Ising directions. For other field directions, there is an energy gap between the magnon bands and they carry Chern number ±1 [33, 34]. The sign of the Chern num- ber swaps between the bands when crossing one of the meridians. These facts can be understood on the basis of the spin-space symmetries of the magnetically ordered configurations. As a reminder, the stability of Dirac points in hon- eycomb tight-binding models follows from the presence (c) Nearest neighbor Dzyaloshinskii-Moriya exchange of time-reversal symmetry T and inversion P. Inversion swaps sublattices and takes k → −k. A possible time FIG. 3. Honeycomb tiling: (a) with different coloured bonds reversal operation is complex conjugation and k → −k. indicating the pattern of distinct Ising couplings of the Kitaev The effective model in the vicinity of the Dirac point is: exchange, (b) showing the second neighbor bonds and arrows indicating the direction of the DMI vector where the bond orientation is always anti-clockwise on the triangles within Hk = dx(k)σx + dy(k)σy + dz(k)σz (57) each hexagon, (c) showing the DMI vectors for the nearest that maps to neighbor DMI for a single honeycomb layer on a substrate where, here, the bond orientation is from the A to the B sites. Hk = dx(k)σx + dy(k)σy − dz(k)σz (58) under PT so the mass term must vanish if PT is a sym- 0 metry. Furthermore, the Dirac points are fixed at the K [111] direction. The symmetries are then E, C3, C2T , 0 0 and K points if C3z is also a symmetry. and P. C2 is a lattice symmetry but it reverses the mo- Consider the highly symmetric situation where the mo- ment direction which can be restored under the action of ment points perpendicular to the honeycomb plane − the T . Note that there are no spin space symmetries. The 15 only anti-unitary symmetries are combined with lattice layer honeycomb magnets grown or mounted on a sub- transformations and, in particular, the combination of strate. In this case, there are no spin space symmetries one of these elements with inversion does not map k to and, with moments perpendicular to the plane, the only z itself. The result is there is no constraint that the gap nontrivial symmetry is C3 . The two bands are therefore close between the magnon bands and since pure time re- gapped and topological. As noted in Ref. [55], linear spin versal is broken, although inversion is present, the Berry wave theory fails to capture this feature because there are curvature may be non-vanishing. The model therefore no O(S) quadratic terms coming from the antisymmetric ± z lies in the bosonic Altland-Zirnbauer class A [11] and the Si Sj couplings. In this case, symmetry breaking due to bands may therefore be topologically nontrivial. the presence of the D coupling comes from higher order Now consider the case where moments are polarized corrections. along the meridian [xy0]. The symmetry elements are z E, T C2 (s), P and products of these. As in the case of [111] moments, the time reversal operator is a composite VII. NODAL LINES, WEYL POINTS, with a C2 symmetry. However, in this case, the C2 is SPIN-SPACE GROUPS AND MAGNETIC SPACE a pure spin space transformation that, therefore, does GROUPS not transform the momentum. The presence of inversion z symmetry and the T C2 (s) are sufficient to forbid the We now return to the case of the Kitaev-Heisenberg mass term so Dirac points are present albeit not at the 0 model on the hyperhoneycomb lattice. In Section IV we K, K points owing to the absence of C3. It follows that identified the N´eelphase of this model as having non- there are Dirac points whenever the field is aligned along trivial spin-space group symmetry with dramatic conse- any one of the cubic meridians with one spin coordinate quences for the magnon band structure including a nodal vanishing. plane, four-fold degenerate point and an abundance of As an aside, the [100] moment direction has Dirac nodal lines. We now consider the collinear ferromag- points but also more symmetry than a general point along netic phase. This offers some important lessons about x the cubic meridians. These elements include E, C2 (s), spin-space symmetry groups for magnons and their ef- y z 0x T C2 (s), T C2 (s), C2 T and P. fect on band topology. For, as we shall see, this phase For a general moment direction [xyz] with none of makes accessible a tower of different spin-space symme- these vanishing, the only symmetries are E, P so the try groups, by tuning the direction of the moments, with symmetry constraints are not sufficient to close the gap a variety of properties. For the several spin-space groups between the magnon bands. relevant to this model, we present the features of the Spin-space symmetry is also important to understand band structures that are imposed by symmetry, includ- magnetic models with Dzyaloshinskii-Moriya exchange ing Weyl points and nodal lines. For comparison we also report the features expected on the grounds of invariance D D · Jˆ × Jˆ where the vector in i j is collinear. In of the magnetic structure thereby giving several examples this case, there is a continuous spin space rotation sym- of how spin-space symmetry leads to richer band struc- metry about an axis parallel to D. We now consider tures. We shall also see that spin-space groups may be the honeycomb ferromagnet with second nearest neigh- isomorphic to ordinary magnetic space groups including bor Dzyaloshinskii-Moriya exchange with D perpendicu- type II groups − of the form G + T G − that ordinarily lar to the plane. When the moments are polarized along would not describe magnetically ordered systems. D, the only manifestation of the spin-space symmetry To begin, we add a Zeeman term to the Hamiltonian, z group is the C∞ that simply adds a quantum number ±1 Eq. 40. At sufficiently large fields, the moments form to the magnons as discussed for the case of Heisenberg ex- a collinear ferromagnetic state and both the symmetry change. The remaining symmetries are identical to those group associated with the magnetic order and the extent of the Heisenberg-Kitaev model as discussed above so of the polarized phase depend on the field direction. For there is a gap between the magnon bands and nonzero the [111] direction, Fig. 4 shows the stability region of the Berry curvature. In fact, once again the magnon bands field-polarized ferromagnet as obtained from the conden- have nonzero Chern number. sation of magnons within linear spin wave theory that If, instead, the moments are aligned in the honeycomb also gives the ordering wavevectors indicated. One ob- plane, all the honeycomb lattice symmetries are present serves that the polarized phase is continuously connected once we allow for rotations in spin space around the D to the zero field ferromagnetic phase over a broad swathe z axis. There is, in addition, a T C2 (s) symmetry that, of ϕ. The phase diagram is qualitatively similar for all taken together with the inversion symmetry, ensures the field directions. z presence of Dirac points that now, owing to the C3 , are For this case, Fig. 5 shows the magnon band structure 0 located at the K, K points. along high symmetry directions in the Brillouin zone for We contrast the model with second neighbor ϕ = −3π/5 and zero applied field showing Goldstone Dzyaloshinskii-Moriya exchange and parallel D vector modes with quadratic dispersion at Γ. At finite field, the with the model with nearest neighbor Dzyaloshinskii- spectrum is completely gapped and details of the band Moriya with in-plane D that may be present for single structure change. However, some features are robust to 16
7 Γ Now, in contrast, if we lock the spin and space transfor- mations and determine the symmetries of the magnetic 6 structure − in other words those elements of the group 5 G + T G that leave the magnetic structure invariant − we find 4 a h / S GM = E+P +T d2 +T C2 = (E+P)+T d2(E+P) (61) 3 Ordered phases which is the magnetic space group C20/c0. 2 The magnetic symmetry of the magnons is therefore E 1 Ev v higher than that of the underlying magnetic structure owing to the freedom to perform D2 transformations in 0 π π -π - 0 π spin space. This enhanced symmetry has important con- 2 2 sequences for the magnon band structure and band topol- φ ogy. In particular, one can show that the lower sym- metry magnetic group C20/c0 enforces degeneracy along FIG. 4. Phase diagram of the Kitaev-Heisenberg model in line ZA but not along other directions. But when the a [111] field. The phase boundary is obtained from the con- full symmetry group is taken into account, as we show densation of magnons in linear spin wave theory about the in Appendix F, all the aforementioned observed magnon collinear field-polarized ferromagnetic state. The condensa- band degeneracies are enforced by symmetry − the sym- 2 tion wavevector is indicated. The wavevector Ev = (− 3 , 0, 0) metry guarantees the presence of nodal lines. In addi- is the one characteristic of the vortex and AF vortex phase, tion to arguments based on representation theory we give as in Ref. [58], which is indeed associated with a second order some more direct justification of zone boundary nodal phase transition. line degeneracies for the [111] hyperhoneycomb ferromag- net protected by magnetic glides in Appendix G. In a similar way, the spin-space group may be found ϕ changes of parameter and the magnitude of the field: for the remaining symmetry-distinct field directions. Ta- Y these are the Weyl point along the line Γ and the double ble II summarizes the results for the collinear ferromag- degeneracy of the lower and upper magnon bands along net and for three zero field antiferromagnetic phases − AZ YT YA zone boundary lines: , and 1. These nodal simple N´eelorder, the skew-stripy phase and the skew- Y lines cross at point . zigzag phase. For the ferromagnetic phase, there are 10 We now examine the magnon band structure from the different groups. One of these is the Heisenberg ferro- point of view of symmetry. Once again, the spin space magnet for which the group is independent of the field G SO part of the group H consists of (3) for the Heisen- direction. The remainder are for the Kitaev-Heisenberg berg coupling alone. For the Kitaev coupling, the ex- model for different high symmetry directions and the gen- x change on the bond is eral direction [xyz]. There are several observations to make about the K 0 0 groups listed in the table. First of all, because the lat- 0 0 0 (59) tice has inversion symmetry and the moments are invari- 0 0 0 ant under this operation, all the ferromagnetic spin-space groups have inversion symmetry. They also all have spin- and the transformations that leave this invariant are the space elements and non-symmorphic elements. Inspec- x [0yz] axial rotations C∞ and perpendicular rotation C2 tion of the anti-unitary part of the group reveals that, that together make up D∞. Taking the three in- for a random magnetic field direction, the group has no equivalent bonds together leaves the symmetry elements magnetic elements. Then, of the more symmetric field di- x y z {C2 ,C2 ,C2 } = D2. The paramagnetic parent group is rections, two ([±1±1z] and [±1∓1z]) have time reversal GH = (G + T G) ⊗ D2. multiplying a glide and the rest have time reversal mul- We now find the subgroup GM ≤ GH that leaves the tiplying an element that acts on spin space. As we saw collinear ferromagnetic structure invariant using separate when discussing honeycomb lattice models, the T C2(s) allowed real space and spin space transformations for the elements behave like pure time reversal symmetry be- [11z] direction with z =6 0. This turns out to be cause they simply reverse the sign of k. Since inversion is also present, these cases have PT symmetry. Fig. 7 shows z b GM = E + P + C2 (s)(d1 + C2 ) the degeneracies in magnon band structures in instances a z c + T (d2 + C2 + C2 (s)(d3 + C2 )) (60) where there is effective PT . Both have nodal lines in the spectra protected by PT with the Heisenberg case (in The group contains distinct spin space and real space the left panel) having various zone boundary nodal lines transformations as well as anti-unitary elements. How- and one pinned to a bisecting plane in the Brillouin zone. ever, the group is isomorphic to the magnetic space group The less symmetric [x0z] case (right panel) has interior F d0d0d as we shall discuss in more detail below. nodal lines that are not pinned. 17
2 Heisenberg 3.5 paramagnetic Fddd1′ × SO(3) (s) 3.0 DY2 2.5 1  ∥ 
S Fddd1′ × C∞(s) Fddd1′ × D2(s) J 2.0 / 1.5 + ∥ 1.0 2 " Fddd1′ × C (s) " DY1 2 Heisenberg-Kitaev 0.5 paramagnetic + 0.0 1 ∥ Y T Z X A1 Y T X1 X A Z L   [0 0 1] Fddd1′ C2/c1′ × C (s) wavevector 2 [1 1 0]
Fd′d ′d C2/c1′
[x y 0] [1 1 z] C2/c C2′/c′ Z X A [x y z] [x 0 z]
L [1 0 0] Heisenberg [x y z] P-1 X1 Y T A1
(b) φ = −0.6π (c) φ = −0.4π FIG. 6. The hierarchy of spin-space groups for the FIG. 5. Example of a field direction without PT in the Heisenberg-Kitaev ferromagnet on the hyperhoneycomb lat- magnetic group isomorphism, therefore allowing Weyl points. tice. At the top, there are the paramagnetic parent groups of Magnon band structure in [111] field and ϕ = −3π/5 along Heisenberg and Heisenberg-Kitaev coupling (magenta) and, high symmetry paths in the Brillouin zone (top). Here the below, all the relevant subgroups for different ferromagnetic irrep labels are marked for points Γ and Y , showing how the directions. The paramagnetic spin-space group is enhanced Weyl node along ΓY arises from compatibility relations. On from the normal grey group F ddd10 (dashed arrow). For the bottom the degeneracies of the band structure are shown each ferromagnetic case, the magnetic group of the ground for case ϕ = −3π/5 (left) and ϕ = −2π/5 (right). The green state is given (dark blue) and its enhancement (arrow) by the lines are nodal lines between bands (1, 2) and/or (3, 4), while Hamiltonian spin-space symmetries to a group relevant for red lines are for nodal lines between (2, 3). The colored point the magnon spectra (cyan). The ground state symmetries are are Weyl points, between (1, 2) (blue) and (2, 3) (red). The enhanced in all cases as a result of the parent group pure spin- case ϕ = −3π/5 has a stable Weyl point along ΓY . Indeed space symmetry. For all groups we have used magnetic space moving to ϕ = −2π/5 this point is still present (closer to group notation. For most cases, the spin-space group is iso- Y ). At this fine-tuned point in parameter space, there is a morphic to a magnetic space group. The only exception is the proliferation of other Weyl points. There are also new nodal [001] case. However, for [001], the group F ddd10 ×Ck(s) (dot- lines that are protected by the glide d (not by PT ) and which 2 1 ted cyan) nevertheless correctly captures the magnon band are therefore constrained to the mirror plane. degeneracies.
A summary of the groups that appear by rotating the Table II illustrates that, for the ferromagnet, the spin- field direction is given in Fig. 6. This plot shows the space groups are isomorphic to magnetic space groups groups arranged by increasing symmetry with the most except for the [100] moment direction. In contrast, the symmetric group − that of the Heisenberg Hamiltonian spin-space groups for the antiferromagnetic phases do not − at the top. The decoupled spin group SO(3) for the coincide with any of the magnetic space groups. To un- Heisenberg exchange is broken down to D2 by the Kitaev derstand these coincidences better, we first note that for coupling. The figure includes the groups that leave the collinear ferromagnets the only possible net spin trans- magnetic structure invariant without spin-space transfor- formations are axial rotations. These commute with one mations allowed by the Hamiltonian (dark blue). The another and, when they match the real space point group figure also includes those groups that leave the mag- transformation, the group multiplication table matches netic structure invariant up to spin-space transformations that of a magnetic space group simply because the spin (cyan) and these are all indexed using isomorphic mag- space transformation is not distinct. netic space group notation. The arrows connecting dif- For example the FM [±1 ± 1 0] has unitary elements: ferent groups reveal the symmetry enhancement in going c z a b from dark blue to cyan groups. HSS = E +P +d3 +C2 +C2 (s)(d1 +d2 +C2 +C2 ). (62) 18
z c Elements like C2 (s)C2 = [2100k {2001|−1/4, −1/4, 0}] can be mapped to real space symmetry only elements [Ek {2001|−1/4, −1/4, 0}] preserving the multiplication table of the group. The isomorphic group is therefore an usual magnetic group with unitary elements: c a b HISO = E + P + d3 + C2 + d1 + d2 + C2 + C2 , (63) which corresponds to the unitary part of F ddd10. The [±1 ± 1 0] spin-space group, despite being isomorphic to a normal magnetic space group, has an enhancement with respect to the ground state magnetic group F d0d0d with fewer unitary elements: (a) Heisenberg (b) FM [x0z] c HGS = E + P + d3 + C2 (64) FIG. 7. Example of field directions with PT in the isomor- We further note that even where there is an isomor- phic magnetic space group with, therefore, nodal lines allowed phism to a magnetic space group it is important to keep everywhere in the zone. The degeneracies of the band struc- track of the spin space transformations in representa- ture are shown for the Heisenberg case (left) and for field tion computations. Secondly, not all ferromagnetic spin direction [x0z] with ϕ = −3π/5 (right). The green lines are space groups are isomorphic to magnetic space groups. nodal lines between bands (1, 2) and/or (3, 4), while red lines For example, for the direction [0 0 ± 1], if we look for are nodal lines between (2, 3). The highly symmetric Heisen- berg case has all possible enforced degeneracy boundary zone a magnetic space group that is closest in form to the lines (green) and the characteristic nodal line (red). The [x0z] 0 k spin-space group, we would find F ddd1 × C2 (s). But if has a much lower symmetry, but still exhibits a nodal line we inspect the elements of the spin-space group we find away from symmetric positions in the BZ, since it has pure x c + C2 (s)C2 = 4010k {2001|−1/4, −1/4, 0} which does not time reversal symmetry in the isomorphic magnetic group, respect the multiplication rule of the F ddd10 group (triv- exactly as in the Heisenberg case, see Table II. ially the square of the element is not identity). Even so, 0 k F ddd1 ×C2 (s) does correctly capture the band structure degeneracies. This is because in this specific case, the de- as in Fig. 5. In the AFM cases, since the symmetry is generacies are restricted to the Brillouin zone boundary, highly enhanced, the presence of closed nodal lines is the and there the central extension groups of the two groups norm, and, as we discussed, nodal planes may also arise. are isomorphic because only spatial elements are rele- vant. For AFM cases, spin space elements may include rotations about axes perpendicular to the moment. Since VIII. SUMMARY AND CONCLUSIONS these generally do not commute with other spin space op- erators one tends to find different multiplication tables to The study of band topology has been a gigantic enter- the usual magnetic space groups. prise in condensed matter physics over roughly the last So far we have commented on the types of spin-space fifteen years. During that time, important insights have groups that can arise in the Kitaev-Heisenberg ferromag- arisen as more symmetries have been considered. The net. We now examine the consequences of these groups first known topological band insulator was a Chern insu- for the magnon bands structures. Table III summarizes lator that was the inspiration for time reversal symmetric the symmetry enforced degeneracies for all high symme- gapped band topology. Later people devised topological try moment directions for the hyperhoneycomb Kitaev- band structures with particle-hole and chiral symmetries Heisenberg model deduced from the representation the- and, later still, lattice symmetries and their interplay ory of the groups given in Table II and, in particular, with time reversal symmetry. In this paper, we have ex- both the groups expected for the ground state alone and tended this programme further to include yet more sym- the full spin-space group. The spin-space group with its metric cases by including spin rotation symmetry and enhanced symmetry leads always to a more degenerate analysing the resulting spin-space groups. spectrum except for the [xy0] case. We have given a number of examples of magnetic cou- As we have already observed, some of the spin-space plings where spin-space groups are the appropriate sym- groups have a time reversal element multiplying a pure metry groups in the magnetically ordered phase. These spin space element. This is isomorphic to a pure time include Heisenberg models, Kitaev Heisenberg models, reversal element and the system then has PT symme- collinear Dzyaloshinskii-Moriya couplings and various try that imposes a reality condition on the Hamiltonian kinds of single ion anisotropy. that protects nodal lines and forbids the presence of Weyl For various cases, we have worked out the represen- points as in Fig. 7. In the other cases ([xyz] and [11z]) we tation theory of the relevant spin-space group thus pro- have the opposite scenario, with presence of Weyl nodes viding the underlying symmetry reason for topological and no closed nodal lines (apart from the nodal line pro- features in the magnetic excitations on top of magnet- tected by glide symmetry constrained to its mirror plane) ically ordered states. These calculations show in un- 19
Spin direction Representative of the unitary subgroup HSS Spin-Space Group GSS Isomorphism b a c k ⊥ ∼ 0 k Heisen. FM [x y z] (E + P + d1 + C2 + d2 + C2 + d3 + C2 )(r) + C∞(s) HSS + T C2 (s) HSS = F ddd1 × C∞(s) c ⊥ b c k ⊥ Heisen. N´eel[x y z] (E + d1 + d2 + C2 )(r) + C2 (s)[(P + d3 + C2 + C2 )(r)] + C∞(s) HSS + T C2 (s) HSS - z b ∼ [x y z] E + P + C2 (s)(d1 + C2 ) HSS = C2/c z b y/x ∼ 0 [x 0 z], [0 y z] E + P + C2 (s)(d1 + C2 ) HSS + T C2 (s) HSS = C2/c1 z b z ∼ 0 [x y 0] E + P + C2 (s)(d1 + C2 ) HSS + T C2 (s) HSS = C2/c1 z b x/y z ∼ 0 k [±1 0 0], [0 ± 1 0] E + P + C2 (s)(d1 + C2 ) + C2 (s) HSS + T C2 (s) HSS = C2/c1 × C2 (s) z b ∼ 0 0 [±1 ± 1 z] E + P + C2 (s)(d1 + C2 ) HSS + T d2 HSS = F d d d z b ∼ 0 0 [±1 ∓ 1 z] E + P + C2 (s)(d1 + C2 ) HSS + T d3 HSS = F d d d c z a b z ∼ 0 [±1 ± 1 0] E + P + d3 + C2 + C2 (s)(d1 + d2 + C2 + C2 ) HSS + T C2 (s) HSS = F ddd1 a z b c z ∼ 0 [±1 ∓ 1 0] E + P + d2 + C2 + C2 (s)(d1 + d3 + C2 + C2 ) HSS + T C2 (s) HSS = F ddd1 b x,y a c z x,y ∼ 0 k [0 0 ± 1] E + P + d1 + C2 + C2 (s)(d2 + d3 + C2 + C2 ) + C2 (s) HSS + T C2 (s) HSS = ”F ddd1 × C2 (s)” a x,y b c z x,y N´eel[0 0 ± 1] E + d1 + d3 + C2 + C2 (s)(P + d2 + C2 + C2 ) + C2 (s) HSS + T C2 (s) HSS - b x,y a c z x,y Skew-Stripy [0 0 ± 1] E + d2 + d3 + C2 + C2 (s)(P + d1 + C2 + C2 ) + C2 (s) HSS + T C2 (s) HSS - a x,y b c z x,y Skew-Zigzag [0 0 ± 1] E + P + d2 + C2 + C2 (s)(d1 + d3 + C2 + C2 ) + C2 (s) HSS + T C2 (s) HSS -
TABLE II. Spin-space groups GSS (and their relative unitary subgroups HSS) for various phases of the Heisenberg-Kitaev model on a hyperhoneycomb lattice. For the collinear ferromagnet, the symmetry groups for different moment directions are listed. The right-most column indicates whether there is an isomorphism between the spin-space group and a magnetic space group. The short-hand notation used for the symmetry elements is: pure spin transformations are labelled with s, pure real space with r and the combined spin-space (locked) symmetry without extra label. Appendix C has further information about the group theory notation.
H. N´eel[x y z] N´eel[0 0 ± 1] SZ [0 0 ± 1] SS [0 0 ± 1] H. FM [x y z] [x y z] [x 0 z] [x y 0] [±1 ± 1 z] [±1 ± 1 0] [0 0 ± 1] IBZ GS SW GS SW GS SW GS SW GS SW GS SW GS SW GS SW GS SW GS SW GS SW Γ X X X X Y (4) XX X X X X XX T (4) X X (4) XX X X X XX X XX XX Z (4) X (4) XX XX X X X XX XX XX XX L X ∆ = [ΓY ] X X X Λ = [ΓZ] X X X X Σ = [ΓX] X X X H = [YT ] (4) X XX X X X XX B = [ZT ] (4) X XX XX X X XX XX X A = [ZA] (4) XX X XX XX XX E = [ΓZT ] X X X J = [ΓXZ] X M = [ΓXY ] X GP X
TABLE III. Degeneracies in ordered Heisenberg-Kitaev models for the hyperhoneycomb lattice at all high symmetry points, lines and surfaces. The first 4 cases are antiferromagnets (where SZ = skew-zigzag and SS = skew-stripy) while the others are ferromagnets. For Heisenberg N´eeland FM ground states the symmetries are given for a generic spin direction [x y z]. The column GS lists the degeneracies expected on the basis of the magnetic space group that leaves the magnetic structure invariant (and are therefore independent of the coupling). The column SW lists those degeneracies coming from the spin-space group. The excitations have not only all the degeneracies coming from the ground state invariance but also a significant enhancement due to the presence of spin space symmetries (especially in the AFM cases thanks to the mixing of spin rotations with the perpendicular axis). A checkmark represents a double degeneracy, while a number (4) a 4-fold degeneracy. Nodal planes are present along E = [ΓZT ] for two AFM cases, while the nodal volume only in the Heisenberg N´eelcase (see general position GP).
precedented detail how to work out band degeneracies to any other spin-space group. In many of these cases from the relevant group − methods that are applicable we have contrasted our findings with the expected band 20 structures one would obtain with foreknowledge only of at least the absence of certain symmetry breaking terms. the magnetic structure and the corresponding magnetic Similarly, even if symmetries are violated that would oth- space group. erwise protect topological surface states, the boundary We have found that spin-space groups of various sorts states can remain when the couplings are proximate to can lead to nodal points (Dirac points in 2D, Weyl points symmetric surfaces in parameter space. and four-fold degenerate points in 3D), nodal lines (2- Our work also sheds some light on the phenomenon fold and 4-fold degenerate) protected by non-symmorphic of order-by-disorder where accidental mean field ground spin-space groups, nodal planes including those with in- state degeneracies are broken down by fluctuations. One tersecting nodal lines and, in some instances, degenerate symptom of order-by-disorder is the presence of spurious volumes. The rich band structures presented in this work Goldstone modes in linear spin wave theory that cannot provide a mere glimpse of the types of gapless band topol- be present in the full interacting model by symmetry. ogy that can arise from spin-space groups in the magnons One may ask whether linear spin wave theory may have of magnetic insulators and the electronic band structures other incongruous features in the spectrum that may be of itinerant magnetic materials. As we should expect of lifted by fluctuations. For the cases we have considered groups of higher symmetry, magnetic groups including in this paper where order-by-disorder is present − such spin rotations are particularly efficient at generating de- as the hyperhoneycomb N´eel state − the answer is that generacies in band structures. linear spin wave theory faithfully reflects the spin-space The magnetic space groups are a subset of spin-space symmetries of the magnetically ordered state up to the groups where the spin and real space point group ele- appearance of Goldstone modes. ments are locked. However, we have also shown, for var- This paper points to a number of interesting future ious ferromagnetic Kitaev-Heisenberg models, that cer- directions most notably the considerable task of carry- tain spin-space groups with nontrivial spin rotation ele- ing out a complete classification of physically relevant ments can also be isomorphic to magnetic space groups, spin-space groups, their representations and associated albeit groups of higher symmetry than one would expect band topology. More immediately, it would be inter- on the basis of the magnetic structure alone. esting to investigate in detail the implications of spin- All of the models we have considered are in some space symmetries for electronic systems such as spin- sense fine-tuned. That is to say that spin-orbit coupling orbit coupled magnetic semi-metals and to magnetic exci- is almost omnipresent in condensed matter systems so tons that are not magnons. More speculatively, one may that the magnetic Hamiltonian will tend to include all ask whether there are physically natural generalizations, terms allowed by symmetry and the moments will then say for SU(N) magnets or exotic order parameters, of the be locked to real space. Nevertheless, all the couplings spin-space symmetries considered here to higher symme- we have included are physically allowed couplings and a tries still such as the polychromatic groups. degree of fine-tuning is often feasible in condensed mat- Acknowledgments: − This work was in part supported ter systems because the richness of chemistry admits an by the Deutsche Forschungsgemeinschaft under grants exploration of possible couplings. So, for example, in SFB 1143 (project-id 247310070) and the cluster of ex- many first row transition metal magnets and elsewhere, cellence ct.qmat (EXC 2147, project-id 390858490). spin-orbit is weak compared to the principal exchange scale and then Heisenberg models may be an excellent approximation to the magnetism. Indeed Heisenberg ex- change and XY exchange both of which admit nontrivial spin-space symmetries have been the canonical models of magnetism for decades and satisfactorily account for the properties of a great many magnetic materials. Going beyond Heisenberg couplings, antisymmetric ex- change that often appears as the leading exchange con- tribution in the spin-orbit coupling can be associated to spin-space group symmetries. Also, Kitaev-Heisenberg exchange has been argued to be the dominant set of cou- plings in various magnets with magnetic ions in edge- shared octahedral cages. Even where the couplings depart from the spin-space symmetric surfaces in parameter space, one can imag- ine that there are materials where the magnon spectra contain near degeneracies that arise from a nearby par- ent Hamiltonian with spin-space symmetries and that these degeneracies would be otherwise mysterious. In fact, such near degeneracies could be used to diagnose the presence of dominant Kitaev terms for example or 21
Appendix A: Hyperhoneycomb Lattice Conventions c∗ (such that for example a · a∗ = 2π) are and Symmetries G1 = (−1, 1, −1), G , − , − , The hyperhoneycomb lattice is a three-dimensional lat- 2 = (1 1 1) tice (1) based around a face-centered orthorhombic cell G3 = (−1, −1, 1). (A6) with a four site basis belonging to the space group Fddd (#70). The primitive lattice vectors in x, y, z Kitaev The high symmetry points in the first Brillouin Zone coordinate system are in both coordinate systems reads:
Γ = (0, 0, 0)xyz = (0, 0, 0)abc R1 = (2, 4, 0), R2 = (3, 3, 2), R3 = (−1, 1, 2) (A1) π Y = 0, 0, − = (0, −1, 0)abc 2 xyz π π π and the basis is T = − , − , − = (0, −1, −1)abc 6 6 2 xyz π π Z = − , − , 0 = (0, 0, −1)abc r1 = (0, 0, 0), r2 = (1, 1, 0), r3 = (1, 2, 1), r4 = (2, 3, 1). 6 6 xyz (A2) 29π 29π X = , − , 0 = (−29/36, 0, 0)abc The orthorhombic conventional unit cell vectors are 72 72 xyz 11π 11π π A1 = , − , − = (−11/36, −1, 0)abc a = (−2, 2, 0), b = (0, 0, 4), c = (6, 6, 0) (A3) 72 72 2 xyz 19π 5π π X1 = − , − , − = (7/36, −1, −1)abc 72 72 2 xyz The reciprocal lattice vectors are 13π 37π A = , − , 0 = (−25/36, 0, −1)abc 72 72 xyz 1 1 1 π π π G1 = 2π , − , , L = , − , − = (−1/2, −1/2, −1/2)abc (A7) 6 3 4 6 3 4 xyz 1 1 1 G2 = 2π − , , − , The first Brillouin zone together with the high symmetry 3 6 4 1 1 1 directions and points are shown in Fig. 1. G3 = 2π , − , − . (A4) The lattice symmetries that constitute the space group 3 6 4 G of the hyperhoneycomb include:
This coordinate system is useful to highlight the effect • Primitive translations Ri of global spin rotations compatible with the Heisenberg- • Inversion P at the bond center of sublattices 2, 3 Kitaev Hamiltonian. and 1, 4 (green and red bonds in Fig. 1); Throughout the paper also another coordinate system has been used, the one used by the Bilbao Crystallo- • Three orthogonal C2 axes at the bond center of graphic Server − the coordinate system of the conven- bonds connecting sublattices 1, 2 and 3, 4 (blue tional unit cell a, b, c with origin centered at the inver- bonds). These axes are parallel to the face- sion point. The latter is simpler for dealing with lattice centered-orthorhombic lattice vectors a, b, and c. symmetries and has been used for all the group theory Bonds 2, 3 and 1, 4 are interchanged via these C2 calculations. In this coordinate system the primitive lat- axes; tice vectors are: • Glide planes, d1, d2, d3, with translation Ri/2 in- terchanges bonds 1, 2 and 3, 4. 1 1 1 1 1 1 R1 = , 0, , R2 = 0, , , R3 = , , 0 2 2 2 2 2 2 Therefore we have for the 7 nontrival symmetry ele- (A5) ments {R| t}, in x, y, z Kitaev coordinates with l sub- and the reciprocal lattice vectors in the dual basis a∗, b∗, lattice index and in a, b, c system:
(x, y, −z, 3) , l = 1 (x, y, −z, 4) , l = 2 1 1 d1 :(x, y, z, l) = = {m010| 4 , 0, 4 }abc (A8) (x + 2, y + 4, −z, 1) , l = 3 (x + 2, y + 4, −z, 2) , l = 4 . 22
(y, x, z, 3) , l = 1 (y, x, z, 4) , l = 2 1 1 d2 :(x, y, z, l) = = {m100|0, 4 , 4 }abc (A9) (y + 3, x + 3, z + 2, 1) , l = 3 (y + 3, x + 3, z + 2, 2) , l = 4 .
(−y, −x, z, 4) , l = 1 (−y, −x, z, 3) , l = 2 1 1 d3 :(x, y, z, l) = = {m001| 4 , 4 , 0}abc (A10) (−y − 1, −x + 1, z + 2, 2) , l = 3 (−y − 1, −x + 1, z + 2, 1) , l = 4 .
(−x, −y, −z, 4) , l = 1 −1 (−x, −y, −z, 3) , l = 2 P = d1d2 d3 :(x1, x2, x3, l) = = {−1|0, 0, 0}abc (A11) (−x, −y, −z, 2) , l = 3 (−x, −y, −z, 1) , l = 4 .
(−y, −x, −z, 2) , l = 1 a −1 (−y, −x, −z, 1) , l = 2 1 1 C2 = d3 d1 :(x, y, z, l) = = {2100|0, − 4 , − 4 }abc (A12) (−y − 3, −x − 3, −z − 2, 4) , l = 3 (−y − 3, −x − 3, −z − 2, 3) , l = 4 .
(−x, −y, z, 2) , l = 1 b −1 (−x, −y, z, 1) , l = 2 1 1 C2 = d3 d2 :(x, y, z, l) = = {2010| − 4 , 0, − 4 }abc (A13) (−x − 2, −y − 4, z, 4) , l = 3 (−x − 2, −y − 4, z, 3) , l = 4 .
(y, x, −z, 1) , l = 1 c −1 (y, x, −z, 2) , l = 2 1 1 C2 = d2 d1 :(x, y, z, l) = = {2001| − 4 , − 4 , 0}abc (A14) (y + 1, x − 1, −z − 2, 3) , l = 3 (y + 1, x − 1, −z − 2, 4) , l = 4 .
Appendix B: Hyperhoneycomb symmetry-allowed specified by 6 exchange 3×3 matrices, one for each bond. exchange Hamiltonian Applying the space group symmetries in Appendix A we obtain two sets of equivalent bonds. A set of equiva- Here, taking the example of the hyperhoneycomb lat- lent bonds contains bonds that can be transformed into tice, we give an explicit example of how spin-orbit con- each other by symmetry operations, and therefore bonds strains the exchange Hamiltonian of a spin system and belonging to different equivalence sets are said to be in- how additional global spin symmetries can arise beyond equivalent and have, in principle, independent exchange those allowed by spin-orbit coupling. Hamiltonians. We call set a the one with (1, 2)z, (3, 4)z The hyperhoneycomb unit cell has i, j = {1, 2, 3, 4} while set b the one (2, 3)x, (2, 3)y, (4, 1)x, (4, 1)y where the spins and it is tri-coordinated, so there are in total 6 subscript a posteriori provides the Kitaev bond label. bonds per unit-cell, (i, j). We consider for this system a generic bilinear exchange Hamiltonian In the absence of spin-orbit the application of the space group symmetries does not act on the spins but only on ˆ X µν ˆµ ˆν their positions. This would result in a generic bilinear H = Jij Ji Jj , (B1) i,j exchange Hamiltonian to that is completely specified by only two generic symmetric matrices, one for each set of which accounting for the primitive translations can be equivalent bonds. However free spins are SU(2) symmet- 23 ric and the only possible isotropic coupling is Heisenberg. also on spin-space as pseudo-vector transformations. Af- In case spin-orbit is present, the symmetries will act ter a proper parametrization the constraints for set a are:
Ja Γa −D Ja Γa D J J D J J −D (1,2)z = Γa a (3,4)z = Γa a (B2) D −D (Ja + Ka) −DD (Ja + Ka) and for set b:
0 00 0 00 (Jb(xy) + Kb)Γ Γ (Jb(xy) + Kb)Γ −Γ J 0 J J 0 J − (2,3)x = Γ b(xy) Γb (4,1)x = Γ b(xy) Γb (B3) 00 00 Γ Γb Jb(z) −Γ −Γb Jb(z)
0 0 Jb(xy) Γ −Γb Jb(xy) Γ Γb J 0 J K − 00 J 0 J K 00 (2,3)y = Γ ( b(xy) + b) Γ (4,1)y = Γ ( b(xy) + b)Γ (B4) 00 00 −Γb −Γ Jb(z) Γb Γ Jb(z)
In total we count 10 possible exchange couplings. the Bravais lattice type, for example F stands for face- We see that the lattice supports Heisenberg-Kitaev cou- centered lattice, C for single-face centered − on C faces plings, antisymmetric or Dzyaloshinski-Moriya (DM) only − and P for primitive. Right after the letter we coupling D for z bonds and different off-diagonal sym- have the symmetry elements of the corresponding point metric exchange Γα. For a given bond, the couplings group (the group that remains if one removes all trans- are invariant under higher symmetries than the lattice as lational components from the space group). The order a whole. For example, Γα → D2(s) and D → C∞(s). of the symmetry elements follows the hierarchy of axes However, only for a subset of the couplings, is there a if present, from the most symmetric to the lowest. The global residual symmetry: for the Heisenberg case (with possible symbols are: number n for rotations Cn, m for global spin rotation symmetry) and Heisenberg-Kitaev mirror planes, -n for improper rotations Sn. Moreover with D2 symmetry. when two elements refer to the same axis they are writ- In the main text, we make the standard approximation ten as n/m. In case the group is non-symmorphic some of that the Kitaev couplings on inequivalent bonds are the the point group elements will be replaced by the symbol same. for the screw axis (a number with a number subscript, i.e. 21) or glide planes (letter a, b, c, n, d). Taking again the examples above we have: F ddd having 3 glide planes Appendix C: Magnetic Group Notation with translation along a quarter of the 3 different face diagonals (”diamond” glide plane), C2/c having a C2 ro- tation and a perpendicular glide plane with a translation In the main text, unless otherwise indicated, group along half the lattice vector of face C, P -1 having only elements such as two-fold rotation C2 are assumed to inversion. The standard form uses a short notation which act both on spin and real space degrees of freedom. shows only the generators in the shortest unambiguous This substantially abbreviates the full spin-space nota- way, i.e. the symmetry elements which, when composed, tion [C2k {C2|0}]. Sometimes we also use an argument: gives all the others. As an example we observe that the C2(s) to denote an operation acting only on spins and glide planes d1, d2, d3 generate all the other operations in C2(r) acting only on real space. appendix A. The magnetic group notation used through the paper is the one adopted in the International Tables For Crys- For magnetic groups, the notation is extended by indi- tallography, i.e. Hermann–Mauguin notation. We do not cating with a prime an anti-unitary operation. For type explain this notation in detail here but only give a general II groups one has the symbol of the space group from overview so that the meaning of the symbols used in this which it is derived plus a 10 at the end, indicating the paper can be appreciated without a detailed knowledge pure time reversal operation. For type III groups the of these tables. space group symbols are modified adding a 0 to the sym- As an example we take the hyperhoneycomb space metry elements which become anti-unitary. Finally for group F ddd and its subgroups C2/c and P -1. The sym- type IV groups, the black and white Bravais lattice is bols always start with an uppercase letter describing represented by the original uppercase letter lattice sym- 24 bol with an additional subscript. For example, the group F ddd10 is type II with pure time reversal and 3 normal 0 0 glides, while the group F d d d is type III with one nor- µ( [2−101k {−1|0}] , [2010k {m010|1/4, 0, 1/4}]) mal glide and two magnetic glides. = exp (−2πi (0, 0, 2) · (1/4, 0, 1/4)) = −1. (D2)
Since the factor system is not trivial we need to find the Appendix D: Central Extension Method for projective irreducible representations. One way to do hyperhoneycomb N´eel Kitaev-Heisenberg Z Point this is by first finding the irreps of the unitary central ¯ Z∗ extension of the little co-group GSS with g = 2 which is The representation theory of spin-space groups mirrors a group with 32 elements. The element of this group are ¯ Z that of (magnetic) space groups. One lesson coming from of the kind (hi, α), where hi ∈ GSS and α ∈ {0, 1, ..., (g − ¯ Z∗ ∼ space groups is that, for most wavevectors, the decom- 1)}. The group has the isomorphism GSS = D4h +D4h × + position of the little co-group of that wavevector into ir- ( 4010k {m100|0} , 0) where we can identify the elements reps is simply a matter of identifying the associated point of D4h as: group at that point and using standard tables. An im- portant exception is for certain boundary points in non- 2001 → ([Ek {E|0}] , 1) symmorphic groups where the group composition rule is + 4001 → ([2100k {2100|0}] , 0) projective. Here we give an example of such a case: the Z − → k {E|0} , point in the N´eelphase of the Kitaev-Heisenberg model 1 ([2010 ] 0) on a hyperhoneycomb lattice. 2100 → ([Ek {m010|0}] , 0) First we must determine the factor system as explained 21-10 → ([2100k {m001|0}] , 0) (D3) in Section IV B. The extra reciprocal lattice translation ¯ Z∗ for each spatial symmetry (noting that the spin part does The irreps of GSS can be obtained by conjugat- not act in reciprocal space) is: ing the ones of the subgroup D4h by the symmetry + ( 4010k {m100|0} , 0). Of these irres we are only inter- g−1 = g2 = g2 = gm = (0, 0, 2) 100 010 001 ested in the one with ∆([Ek {E|0}] , 1) = − I (constrain ± g2001 = gm100 = gm010 = (0, 0, 0) (D1) to get the proper phase factors), therefore Z5a(b) where ± (a, b) label comes from the fact that Z5 of D4h are The factor system will then have µ(hi, hj) = ±1 (as ex- + self-conjugated under ( 4 k {m100|0} , 0). Considering pected for glide symmetry with non-symmorphic trans- 010 only the elements (hi, 0) and adding the right phase fac- lation equal to half of primitive vectors). For example, tors for translation we can build the character table of Z the unitary part of GSS as in Table IV. µ( [2−101k {−1|0}] , [2001k {m001|1/4, 1/4, 0}]) Finally we can add now the anti-unitary elements and = exp (−2πi (0, 0, 2) · (1/4, 1/4, 0)) = +1 compose the coreps. The test gives:
( ¯ Z∗ + + X Z 2 Z Z 16 = |GSS | Type (a) if p = Z5a,Z5b χp (hk0 ) = 4 (χp ([Ek {E|0}]) + χp ([2010k {E|0}])) = − − (D4) 0 Type (c) if p = Z5a,Z5b hk0
where hk0 are all the anti-unitary elements for Appendix E: Representation theory for Heisenberg which hk0 k = −k + gi, so for Z are all N´eel antiferromagnet the anti-unitary elements in the full spin space group. We get therefore a 4-degenerate coreps − − − DZ5 (4) = (Z5a,Z5b) with χ([2010k {E|0}]) = −4 and For the Heisenberg coupling, space group symmetry el- ± ements act only on real space (since we can always cancel χ( 4010k {m100|0, 1/4, 1/4} ) = 0. Since we know the characters of the band representa- out their effect on the spins with a SO(3) rotation). In ρk k the antiferromagnetic case we can then divide the space tions G,S⊥ for every we can now subduce it to the point Z, therefore from its characters in 46 and 47 we group symmetries into two groups, those that do not get (the number in parenthesis indicates the dimension swap the magnetic sublattice G↑↑ and those that do. The latter have to be coupled with an additional spin rotation of the corep): ⊥ to preserve the magnetic order C2 (s) G↑↓. In addition, ρZ = DZ−(4) (D5) we have to consider pure spin rotations about axes paral- S± 5 k lel to moment directions C∞(s) (rotations ϑz). The full ⊥ group is therefore GH−Neel = HH−Neel+T C2 (s)HH−Neel 25
Z + − GSS [Ek {E|0}] [2010k {E|0}] 4010k {m100|0, 1/4, 1/4} 4010k {m100|0, 1/4, 1/4} ... Type Coreps + Z5a 2 2 2 ξ 2 ξ 0 (a) + Z5b 2 2 -2 ξ -2 ξ 0 (a) − Z5a 2 -2 2 ξ -2 ξ 0 (c) − Z5b 2 -2 -2 ξ 2 ξ 0 (c)
Z 1 1 TABLE IV. Character table giving the relevant irreps of the unitary part of GSS. The phase factor is ξ = exp(i Z·(0 4 4 )) = −i and the dot (...) indicates irrelevant symmetries − those with trivial characters.
⊥ with elements: add the representative C2 (s)P which mixes spin and space transformations. In deducing the reps of this mixed k ⊥ k HH−Neel = C∞(s) × G↑↑(r) × C2 (s) G↑↓(r) = C∞(s)+ group we need to check the conjugation of elements with c ⊥ b c C⊥(s)P: (E + d1 + d2 + C2 )(r) + C2 (s)[(P + d3 + C2 + C2 )(r)] 2 (E1) ⊥ ⊥ −1 (C2 (s)P) S(r)(C2 (s)P) = S(r) ∀ S(r) ∈ G↑↑(r) excluding primitive translations. Here we have used a C⊥ s P ϑ s C⊥ s P −1 ϑ−1 s ∀ ϑ ∈ Ck s . short-hand notation for the symmetry elements where ( 2 ( ) ) z( )( 2 ( ) ) = z ( ) z ∞( ) pure spin transformations are labelled with s and pure (E4) real space with r. All symmetry elements are defined in Appendix A. Therefore the irreps will pair into two dimensional irreps + − The representations of the enhanced magnetic little (Γn , Γn )(2). The last step is to consider anti-unitary ele- k T C⊥ s groups GSS are straightforward to find for points k inside ments 2 ( ). Coreps types are assessed by computing: the Brillouin zone. GP = (u, v, w) − the least symmetric (general) posi- X Γ 2 X Γ X Γ χ (h 0 ) = 4 χ (2ϑ(s)) + 4 χ (E) tion has little group: p k p p hk0 ϑ 2⊥ GP k ∼ 0 Z 2π GSS /T = C∞(s)(E + TP) = C∞1 (E2) Γ = 4 2 cos 2ϑ + 4 n χp (E) 0 and therefore doubly degenerate modes as explained in n |HΓ | Section V. = 8 = SS Type (a) (E5) Γ = (0, 0, 0) − the most symmetric point has little group: where hk0 are all the anti-unitary elements, n → ∞ is ⊥ Γ the multiplicity of the class C2 (s) and χp (E) = 2 for 2D Γ k ⊥ GSS/T =[(G↑↑(r) + C∞(s)) × (E + C2 (s)P)] irreps. Since the coreps are type (a) there is no further ⊥ degeneracy. × (E + T C2 (s)) (E3) Since the least symmetric and most symmetric points ⊥ where we have chosen C2 (s)P as a means of swapping inside the Brillouin Zone have the same degeneracy, all the magnetic sublattices. We can build the final coreps the intermediate cases must have the same degeneracy. in 4 steps. First, we consider only the real space group Nevertheless the presence of non-symmorphic symmetries G↑↑(r) = F dd2 which has only 1D irreps Γn. Second, can lead to extra degeneracies corresponding to projec- multiply C∞(s) into the group, which self-conjugates tive representations on the zone boundaries. ± each irrep of G↑↑(r) to Γn (since transverse spin com- B = (0, u, −1) − Using spin-space group notation the ponents transform as m = ±1 reps of C∞). Third, we [ZT ] line has little group elements:
B 1 1 1 1 1 1 GSS/T = ([ϑzk {E|0}] + ϑzk m100|0, 4 , 4 + 2⊥k m001| 4 , 4 , 0 + 2⊥k 2010|− 4 , 0, − 4 ) 0 1 1 0 1 1 0 1 1 0 + ([ϑzk {−1|0}] + ϑzk 2100|0, − 4 , − 4 + 2⊥k 2001|− 4 , − 4 , 0 + 2⊥k m010| 4 , 0, 4 ) (E6)
where the elements of the kind [ϑzk {E|0}] (or dicular axes). Also the anti-unitary elements are denoted [2⊥k {E|0}])) represent all the rotations around the by a prime. Here the 2⊥ rotations intertwine spin and collinear axis (or π rotation around the infinite perpen- space transformations, making the little group not a sim- 26 ple direct product and therefore making it impossible to D∞h as: use the tabulated space group projective representations. ϑ → ϑ k {E|0} , To build the projective representations from scratch z ([ z ] 0) we must determine first the factor system as explained 2⊥ → ([2⊥k {2010|0}] , 0) in Section IV B. The extra reciprocal lattice translation -1 → ([Ek {E|0}] , 1). for each spatial symmetry (noting that the spin part does ¯ B∗ not act on reciprocal space) is: The irreps of GSS can be obtained by conjugat- ing those of the subgroup D∞h with the symmetry Ek {m |0} , g2010 = gm001 = (0, 0, 2) ([ 100 ] 0). Of these representations we are only interested in the ones with ∆([Ek {E|0}] , 1) = − (this gm = (0, 0, 0). (E7) I 100 constraint ensures the correct projective phase factors) and with ∆([ϑzk {E|0}] , 0) = 2 cos ϑ (coming from the The factor system will then have µ(hi, hj) = ±1 and transverse spin component band representation). This therefore gives nontrivial projective irreducible represen- a(b) points to B1u . The (a, b) label comes from the fact that tations. B1u of D∞h are self-conjugate under ([Ek {m100|0}] , 0). The unitary central extension of the little co-group Considering only the elements (hi, 0) and adding the ¯ B∗ GSS with g = 2 is a group with 8 n elements (n is, right phase factors for translation we can build the char- B again, the multiplicity of the perpendicular axis). The acter table of the unitary part of GSS as in Table V. ¯ B∗ ∼ group has the isomorphism GSS = D∞h + D∞h × Finally we consider the anti-unitary elements and com- ([Ek {m100|0}] , 0) where we can identify the elements of pose the coreps. The test gives:
X Γ 2 X B ± 1 1 X B χp (hk0 ) = 2 χp ( 2ϑz k {E|0} ) + (1 + exp(i B · ( 2 0 2 ))) χp ([Ek {E|0}])
hk0 ϑ 2⊥ Z 2π = 2 2 cos 2ϑ = 0 Type (c) (E8) 0
k h 0 ρ where k are all the anti-unitary elements for which For band representation G,S⊥ we obtain: hk0 k = −k + gi, so for B are the anti-unitary elements B with real space transformation {m010, 2001, 2100, −1}. ρ DB . S± = 1u(4) (E9) We get therefore a 4-fold degenerate co-representation a b ± DB1u(4) = (B1u,B1u)(4) with χ([ϑz k {E|0}]) = 4 cos ϑ The same argument carries through for the high sym- and χ([Ek {m100|0, 1/4, 1/4}]) = 0. metry line A = (u, 0, −1), which again is 4-degenerate (line [ZA]). H = (u, 0, −1) − Lastly the line [YT ] line has little group:
H 1 1 1 1 1 1 GSS/T = ([ϑzk {E|0}] + ϑzk m100|0, 4 , 4 + ϑzk 2001|− 4 , − 4 , 0 + ϑzk m010| 4 , 0, 4 ) 0 1 1 0 1 1 0 1 1 0 + ([ϑzk {−1|0}] + ϑzk 2100|0, − 4 , − 4 + ϑzk m001| 4 , 4 , 0 + ϑzk 2010|− 4 , 0, − 4 ) k = (C∞(s) × G↑↑(r)) × (E + PT ). (E10)
Here the little group is a simple direct product of spin m = ±1 irreps of C∞). Finally we can consider the anti- and space. The final coreps are therefore easier to build, unitary PT symmetry, which produces no extra degen- exploiting tabulated projective irreps of group G↑↑(r) = eracy on the space part of the direct product (indeed the 0 F dd2 which are the 2D H1(2). Adding the direct prod- irreps of F dd2 × (E + PT ) = F ddd are still 2D), while uct with C∞(s), the irreps H1(2) will self-conjugate un- pairing up the irreps m = ±1 of the spin part, therefore ± + − der these new elements, producing two new irreps H1 (2) giving a 4-fold degenerate line DH1(4) = (H1 ,H1 )(4) (since again the transverse spin component transform as as well. 27
B ± GSS [Ek {E|0}] ϑz k {E|0} ... ∞ [2⊥k {2010|−1/4, 0, −1/4}] [Ek {m100|0, 1/4, 1/4}] Type Coreps a B1u 2 2 cos ϑ ... 0 2 ξ (c) b B1u 2 2 cos ϑ ... 0 -2 ξ (c)
B 1 1 TABLE V. Character table giving the relevant irreps of the unitary part of GSS. The phase factor is ξ = exp(i B · (0 4 4 )) = π exp(i 2 (v − 1)) and the dot (...) indicates here the infinite class of axial rotations.
Appendix F: Symmetry Analysis for [111] JK model. There are two such Weyl points to be consistent Ferromagnet with fermion doubling on opposite sides of Γ and, since the ordering of the Γ point energies is crucial to the exis- To see this, we exploit the fact that the enhanced mag- tence of the point, it is evidently an accidental crossing. netic symmetry group is isomorphic to an ordinary mag- This calculation therefore exposes the symmetry origin of netic space group − which need not be the case as we the Weyl point between bands 2 and 3 shown in Fig. 5. shall see later − and use the tabulated character tables T = (0, −1, −1) − Like Y and Γ, the little co-group b at high symmetry points and lines to determine the en- at T contains all the elements E, d1,C2 , P and the char- forced degeneracies. The enhanced magnetic group iso- acters of the nontrivial elements are zero giving T M = 0 0 morphism for [111] is F d d d: 2(T1 + T1) where T1 is a 2D irrep. Inclusion of anti- unitary elements reveals that the coreps belong to class b a c GM = E + P + d1 + C2 + T (d2 + d3 + C2 + C2 ) (a) so there are two two-fold degenerate bands. (F1) b Z = (0, 0, −1) − From elements E, d1,C2 , P at (0, 0, −1) we obtain two copies of the 2D irrep Z1 and , , − Γ = (0 0 0) The little co-group elements of the uni- class (a) coreps so there are two two-fold degenerate G E, d ,Cb tary subgroup of M at Γ has all the elements 1 2 magnon modes at this point. and P. We may construct the matrix representation of L 1 , 1 , 1 − E P these elements for the magnons and find the characters. = ( 2 2 2 ) This point has only and elements, These are computed to be zero character for the inversion symmetry and type (a) coreps leading to four distinct modes. M M b M χΓ (d1) = 0 χΓ (C2 ) = 0 χΓ (P) = 0 B = (0, u, −1) − This is the line [ZT ] with unitary b b M + − + − symmetry elements E and C2 . The character for C2 is =⇒ Γ = 2 Γ1 + Γ1 + Γ2 + Γ2 (F2) M zero and therefore B = 2(B1 + B1 + B2 + B2). The where the factor of two comes about because the diago- coreps are of type (a) leading to four 1D modes. nalization gives modes at k and −k which are identical H = (0, −1, u) − We focus on the line [YT ]. The unitary symmetry elements along the line are E and d1 here because there is inversion symmetry. M We now include the effect of the anti-unitary elements and the character for d1 is zero. Then H = 2(H1 + by finding H1 + H2 + H2) and the coreps are of type (c) meaning that there is a two-fold degeneracy of the magnon modes X p 0 0 2 where each mode has bound (H1,H2) irreps. χΓ {Rα| tα} = 4 (F3) α0 A = (u, 0, −1) − We focus now on the line [ZA] (and equivalently on [YA1]). As with the [YT ] line, the uni- for each irrep in Eq. F2 which, according to the criterion tary elements along these lines are E and d1, there are Eq. 38, gives only class (a) coreps. We therefore find four four 1D irreps that are bound into two copies of (A1,A2) distinct bands at Γ. In order of increasing energy these so the lines both have two two-fold degenerate bands. + + − − are Γ1 , Γ2 , Γ1 , Γ2 . Σ = (u, 0, 0) − This is the line [ΓX] (and equivalently Y = (0, −1, 0) − Similarly to Γ the little co-group X T ZA YT b [ 1 ]), which is different to [ ] and [ ] although the has elements E, d1,C2 and P, the characters of the non- unitary element are identical leading to four 1D irreps. identity elements are all zero and the irreps are This is because the anti-unitary elements lead to corep criterion χ(E) + exp(ik · (0, 1/2, 1/2)) = 2 meaning that Y M Y + Y − Y + Y − = 2 1 + 1 + 2 + 2 (F4) the coreps are type (a) and the modes are therefore singly degenerate. Similar argument hold for line ∆ = [ΓY ] = P p 0 0 2 only now 0 χY {Rα| tα} = 0 so the coreps are of (0, u, 0) and Λ = [ΓZ] = (0, 0, u). type (c). The coreps therefore bind unitary irreps into We have accounted for all the degeneracies in the + − pairs and the pattern of pairing is DY1 = (Y1 ,Y1 ) and magnon spectrum from the enhanced symmetry group. If + − DY2 = (Y2 ,Y2 ) so there are two doubly degenerate we had instead taken the symmetry group of the under- bands at Y . lying magnetic structure neglecting the pure spin space It follows from the results at Γ and Y , the ordering of transformations available to the Kitaev-Heisenberg cou- the Γ irreps in energy and compatibility relations that pled spins then the representation theory would have pre- a crossing of two bands between Γ and Y occurs in this dicted a degeneracy along [ZA] and nowhere else. 28
Appendix G: Informal Arguments for Band more illuminating to give a direct correspondence be- Touching tween a magnetic symmetry and any degeneracy. For example, we have encountered time reversal glide sym- The analysis given in the main text based on rep- metries, T di, in various cases. The action of the glide resentation theory supplied the underlying symmetry is given in Eqs. A8, A9 and A10. For concreteness we constraints on the observed robust degeneracies in the consider, T d2 acting on a magnon state at a given mo- magnon spectra. It is occasionally possible and certainly mentum k = k1b1 + k2b2 + k3b3
∗ |k2 − k3, k2, k2 − k1, 3i , l = 1 ∗ |k2 − k3, k2, k2 − k1, 4i , l = 2 T d2 : |k1, k2, k3, l i = (G1) e2πik2 |k − k , k , k − k , i∗ , l 2 3 2 2 1 1 = 3 2πik2 ∗ e |k2 − k3, k2, k2 − k1, 2i , l = 4 .
2 From this (T d2) = exp(±2πik2) = −1 when k2 = ±1/2. and the presence of both relies on the existence of spin This has the implication that at invariant momenta sat- and space transformations. The above straightforward isfying k2 = ±1/2 and k1 − k3 = ±1/2 − along a line at argument leads to the presence of crossing zone bound- the Brillouin zone boundary − Kramers theorem enforces ary nodal lines. Similar arguments were employed to un- a degeneracy. For the Kitaev-Heisenberg ferromagnet in derstand magnon nodal lines in the canted zig-zag order [111] field, the magnetic symmetry includes T d2 and T d3 in a magnetic field [59].
[1] C. Bradley and A. Cracknell, The mathematical theory and the fate of zero modes, Phys. Rev. B 102, 125127 of symmetry in solids: representation theory for point (2020). groups and space groups (Oxford University Press, 2009). [12] I. Kimchi and A. Vishwanath, Kitaev-Heisenberg models [2] M. S. Dresselhaus, G. Dresselhaus, and A. Jorio, Group for iridates on the triangular, hyperkagome, kagome, fcc, theory: application to the physics of condensed matter and pyrochlore lattices, Phys. Rev. B 89, 014414 (2014). (Springer Science & Business Media, 2007). [13] J. G. Rau, E. K.-H. Lee, and H.-Y. Kee, Spin-orbit [3] C.-K. Chiu, J. C. Y. Teo, A. P. Schnyder, and S. Ryu, physics giving rise to novel phases in correlated systems: Classification of topological quantum matter with sym- Iridates and related materials, Annual Review of Con- metries, Rev. Mod. Phys. 88, 035005 (2016). densed Matter Physics 7, 195 (2016). [4] S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. W. Lud- [14] M. Hermanns, I. Kimchi, and J. Knolle, Physics of wig, Topological insulators and superconductors: tenfold the kitaev model: fractionalization, dynamical cor- way and dimensional hierarchy, New Journal of Physics relations, and material connections, arXiv preprint 12, 065010 (2010). arXiv:1705.01740 (2017). [5] B. Bradlyn, L. Elcoro, J. Cano, M. G. Vergniory, [15] S. Trebst, Kitaev materials, arXiv preprint Z. Wang, C. Felser, M. I. Aroyo, and B. A. Bernevig, arXiv:1701.07056 (2017). Topological quantum chemistry, Nature (London) 547, [16] S. M. Winter, A. A. Tsirlin, M. Daghofer, J. van den 298 (2017), arXiv:1703.02050 [cond-mat.mes-hall]. Brink, Y. Singh, P. Gegenwart, and R. Valenti, Models [6] H. C. Po, A. Vishwanath, and H. Watanabe, Complete and materials for generalized kitaev magnetism, Journal theory of symmetry-based indicators of band topology, of Physics: Condensed Matter 29, 493002 (2017). Nature Communications 8, 50 (2017), arXiv:1703.00911 [17] S. M. Winter, Y. Li, H. O. Jeschke, and R. Valenti, Chal- [cond-mat.str-el]. lenges in design of Kitaev materials: Magnetic interac- [7] J. Cano and B. Bradlyn, Band Representations tions from competing energy scales, Physical Review B and Topological Quantum Chemistry, arXiv e-prints 93, 214431 (2016). , arXiv:2006.04890 (2020), arXiv:2006.04890 [cond- [18] G. Jackeli and G. Khaliullin, Mott insulators in the mat.mes-hall]. strong spin-orbit coupling limit: from Heisenberg to a [8] W. F. Brinkman and R. J. Elliott, Theory of Spin-Space quantum compass and Kitaev models, Phys. Rev. Lett. Groups, Proceedings of the Royal Society of London Se- 102, 017205 (2009). ries A 294, 343 (1966). [19] J. Chaloupka, G. Jackeli, and G. Khaliullin, Kitaev- [9] W. Brinkman and R. J. Elliott, Space Group Theory for Heisenberg model on a honeycomb lattice: possible exotic Spin Waves, Journal of Applied Physics 37, 1457 (1966). phases in iridium oxides A2IrO3, Phys. Rev. Lett. 105, [10] W. Brinkman, Magnetic Symmetry and Spin Waves, 027204 (2010). Journal of Applied Physics 38, 939 (1967). [20] S. Choi, R. Coldea, A. Kolmogorov, T. Lancaster, [11] Q.-R. Xu, V. P. Flynn, A. Alase, E. Cobanera, L. Viola, I. Mazin, S. Blundell, P. Radaelli, Y. Singh, P. Gegen- and G. Ortiz, Squaring the fermion: The threefold way wart, K. Choi, et al., Spin waves and revised crystal 29
structure of honeycomb iridate Na2IrO3, Phys. Rev. Lett. S. Ma, K. Sugii, N. Kurita, H. Tanaka, J. Nasu, Y. Mo- 108, 127204 (2012). tome, T. Shibauchi, and Y. Matsuda, Majorana quanti- [21] K. Plumb, J. Clancy, L. Sandilands, V. V. Shankar, zation and half-integer thermal quantum Hall effect in Y. Hu, K. Burch, H.-Y. Kee, and Y.-J. Kim, α- RuCl3: a Kitaev spin liquid, Nature (London) 559, 227 (2018), A spin-orbit assisted Mott insulator on a honeycomb lat- arXiv:1805.05022 [cond-mat.str-el]. tice, Physical Review B 90, 041112 (2014). [36] X. Wan, A. M. Turner, A. Vishwanath, and S. Y. [22] J. Sears, M. Songvilay, K. Plumb, J. Clancy, Y. Qiu, Savrasov, Topological semimetal and fermi-arc surface Y. Zhao, D. Parshall, and Y.-J. Kim, Magnetic order in states in the electronic structure of pyrochlore iridates, α- RuCl3: A honeycomb-lattice quantum magnet with Phys. Rev. B 83, 205101 (2011). strong spin-orbit coupling, Physical Review B 91, 144420 [37] A. A. Burkov, M. D. Hook, and L. Balents, Topological (2015). nodal semimetals, Phys. Rev. B 84, 235126 (2011). [23] M. Majumder, M. Schmidt, H. Rosner, A. Tsirlin, H. Ya- [38] N. P. Armitage, E. J. Mele, and A. Vishwanath, Weyl suoka, and M. Baenitz, Anisotropic Ru3+ 4d5 magnetism and dirac semimetals in three-dimensional solids, Rev. in the α- RuCl3 honeycomb system: Susceptibility, spe- Mod. Phys. 90, 015001 (2018). cific heat, and zero-field nmr, Physical Review B 91, [39] C. Fang, H. Weng, X. Dai, and Z. Fang, Topological nodal 180401 (2015). line semimetals, Chinese Physics B 25, 117106 (2016). [24] S. Williams, R. Johnson, F. Freund, S. Choi, A. Jesche, [40] W. Wu, Y. Liu, S. Li, C. Zhong, Z.-M. Yu, X.-L. Sheng, I. Kimchi, S. Manni, A. Bombardi, P. Manuel, P. Gegen- Y. X. Zhao, and S. A. Yang, Nodal surface semimetals: wart, et al., Incommensurate counterrotating magnetic Theory and material realization, Phys. Rev. B 97, 115125 order stabilized by Kitaev interactions in the layered (2018). honeycomb α- Li2IrO3, Physical Review B 93, 195158 [41] D. B. Litvin and W. Opechowski, Spin groups, Physica (2016). 76, 538 (1974). [25] A. Banerjee, C. Bridges, J.-Q. Yan, A. Aczel, L. Li, [42] R. Coldea, S. M. Hayden, G. Aeppli, T. G. Perring, C. D. M. Stone, G. Granroth, M. Lumsden, Y. Yiu, J. Knolle, Frost, T. E. Mason, S.-W. Cheong, and Z. Fisk, Spin et al., Proximate Kitaev quantum spin liquid behaviour waves and electronic interactions in La2CuO4, Phys. Rev. in a honeycomb magnet, Nature materials 15, 733 (2016). Lett. 86, 5377 (2001). [26] A. Little, L. Wu, P. Lampen-Kelley, A. Banerjee, S. Pan- [43] H. Ichikawa, L. Kano, M. Saitoh, S. Miyahara, tankar, D. Rees, C. Bridges, J.-Q. Yan, D. Man- N. Furukawa, J. Akimitsu, T. Yokoo, T. Matsumura, drus, S. Nagler, et al., Antiferromagnetic resonance M. Takeda, and K. Hirota, Orbital Ordering in Ferro- and terahertz conductivity in α− RuCl3, arXiv preprint magnetic Lu2V2O7, Journal of the Physical Society of arXiv:1704.07357 (2017). Japan 74, 1020 (2005), arXiv:cond-mat/0502020 [cond- [27] A. Ponomaryov, E. Schulze, J. Wosnitza, P. Lampen- mat.str-el]. Kelley, A. Banerjee, J.-Q. Yan, C. Bridges, D. Man- [44] J. G. Rau and M. J. P. Gingras, Frustration and drus, S. Nagler, A. Kolezhuk, et al., Unconventional spin anisotropic exchange in ytterbium magnets with edge- dynamics in the honeycomb-lattice material α- RuCl3: shared octahedra, Phys. Rev. B 98, 054408 (2018). High-field electron spin resonance studies, Physical Re- [45] G. Sala, M. B. Stone, B. K. Rai, A. F. May, D. S. Parker, view B 96, 241107 (2017). G. B. Hal´asz,Y. Q. Cheng, G. Ehlers, V. O. Garlea, [28] K. Ran, J. Wang, W. Wang, Z.-Y. Dong, X. Ren, S. Bao, Q. Zhang, M. D. Lumsden, and A. D. Christianson, Crys- S. Li, Z. Ma, Y. Gan, Y. Zhang, et al., Spin-wave exci- tal field splitting, local anisotropy, and low-energy excita- tations evidencing the Kitaev interaction in single crys- tions in the quantum magnet YbCl3, Phys. Rev. B 100, talline α- RuCl3, Physical review letters 118, 107203 180406 (2019). (2017). [46] R. T. Collins, Z. Schlesinger, M. W. Shafer, and T. R. [29] J. Sears, Y. Zhao, Z. Xu, J. W. Lynn, and Y.-J. Kim, McGuire, Antiferromagnetic resonance in La2−xCuO4−y, Phase diagram of α-RuCl3 in an in-plane magnetic field, Phys. Rev. B 37, 5817 (1988). Physical Review B 95, 180411 (2017). [47] H. Yamazaki, Y. Morishige, and M. Chikamatsu, In- [30] A. Wolter, L. Corredor, L. Janssen, K. Nenkov, Plane Anisotropy of Ferromagnetic Resonance and Spin- S. Sch¨onecker, S.-H. Do, K.-Y. Choi, R. Albrecht, Wave Relaxation in K2CuF4, Journal of the Physical So- J. Hunger, T. Doert, et al., Field-induced quantum criti- ciety of Japan 50, 2872 (1981). cality in the kitaev system α- RuCl3, Physical Review B [48] M. Elliot, P. A. McClarty, D. Prabhakaran, R. D. John- 96, 041405 (2017). son, H. C. Walker, P. Manuel, and R. Coldea, Visual- [31] A. Kitaev, Anyons in an exactly solved model and be- ization of Isospin Momentum Texture of Dirac Magnons yond, Annals of Physics 321, 2 (2006). and Excitons in a Honeycomb Quantum Magnet, arXiv [32] L. Janssen, E. C. Andrade, and M. Vojta, Honeycomb- e-prints , arXiv:2007.04199 (2020), arXiv:2007.04199 lattice Heisenberg-Kitaev model in a magnetic field: Spin [cond-mat.str-el]. canting, metamagnetism, and vortex crystals, Physical [49] Q.-R. Xu, V. P. Flynn, A. Alase, E. Cobanera, L. Viola, review letters 117, 277202 (2016). and G. Ortiz, Squaring the fermion: The threefold way [33] P. A. McClarty, X.-Y. Dong, M. Gohlke, J. G. Rau, and the fate of zero modes, Phys. Rev. B 102, 125127 F. Pollmann, R. Moessner, and K. Penc, Topological (2020). magnons in Kitaev magnets at high fields, Phys. Rev. [50] Villain, J., Bidaux, R., Carton, J.-P., and Conte, R., Or- B 98, 060404 (2018). der as an effect of disorder, J. Phys. France 41, 1263 [34] D. G. Joshi, Topological excitations in the ferromag- (1980). netic Kitaev-Heisenberg model, Phys. Rev. B 98, 060405 [51] E. Shender, Antiferromagnetic garnets with fluctuation- (2018). ally interacting sublattices, JETP 56, 178 (1982). [35] Y. Kasahara, T. Ohnishi, Y. Mizukami, O. Tanaka, [52] C. L. Henley, Ordering due to disorder in a frustrated 30
vector antiferromagnet, Phys. Rev. Lett. 62, 2056 (1989). B 26, 3010 (1982). [53] L. Savary, K. A. Ross, B. D. Gaulin, J. P. C. Ruff, and [57] E. K.-H. Lee, R. Schaffer, S. Bhattacharjee, and Y. B. L. Balents, Order by Quantum Disorder in Er2Ti2O7, Kim, Heisenberg-Kitaev model on the hyperhoneycomb Phys. Rev. Lett. 109, 167201 (2012). lattice, Phys. Rev. B 89, 045117 (2014). [54] M. E. Zhitomirsky, M. V. Gvozdikova, P. C. W. [58] W. G. F. Kr¨uger, M. Vojta, and L. Janssen, Holdsworth, and R. Moessner, Quantum Order by Disor- Heisenberg-Kitaev models on hyperhoneycomb and der and Accidental Soft Mode in Er2Ti2O7, Phys. Rev. stripy-honeycomb lattices: 3d-2d equivalence of ordered Lett. 109, 077204 (2012). states and phase diagrams, Phys. Rev. Research 2, [55] A. Mook, K. Plekhanov, J. Klinovaja, and D. Loss, 012021 (2020). Interaction-stabilized topological magnon insulator in [59] W. Choi, T. Mizoguchi, and Y. B. Kim, Nonsymmorphic- ferromagnets (2020), arXiv:2011.06543 [cond-mat.str-el]. symmetry-protected topological magnons in three- [56] J. Zak, Band representations of space groups, Phys. Rev. dimensional Kitaev materials, Phys. Rev. Lett. 123, 227202 (2019).