Electronic Structure of Pyrochlore Iridates: From Topological Dirac Metal to Mott Insulator
Xiangang Wan, Ari Turner, Ashvin Vishwanath, Sergey Y. Savrasov
EElectronic Structure of Pyrochlore Iridates: From Topological Dirac Metal to MottInsulator
Xiangang Wan , Ari Turner , Ashvin Vishwanath , , Sergey Y. Savrasov , National Laboratory of Solid State Microstructures and Department of Physics, Nanjing University, Nanjing 210093, China Department of Physics, University of California, Berkeley, CA 94720 Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley CA 94720. Department of Physics, University of California, Davis, One Shields Avenue, Davis, CA 95616.
In 5 d transition metal oxides such as the iridates, novel properties arise from the interplay ofelectron correlations and spin-orbit interactions. We investigate the electronic structure of thepyrochlore iridates, (such as Y Ir O ) using density functional theory, LDA+U method, and effectivelow energy models. A remarkably rich phase diagram emerges on tuning the correlation strength U . The Ir magnetic moment are always found to be non-collinearly ordered. However, the groundstate changes from a magnetic metal at weak U , to a Mott insulator at large U . Most interestingly,the intermediate U regime is found to be a Dirac semi-metal, with vanishing density of states atthe Fermi energy. It also exhibits topological properties - manifested by special surface states in theform of Fermi arcs, that connect the bulk Dirac points. This Dirac phase, a three dimensional analogof graphene, is proposed as the ground state of Y Ir O and related compounds. A narrow windowof magnetic ‘axion’ insulator, with axion parameter θ = π , may also be present at intermediate U .An applied magnetic field induces ferromagnetic order and a metallic ground state. Previously, some of the most striking phenomena insolids, such as high temperature superconductivity[1] andcolossal magnetoresistance[2] were found in transitionmetal systems involving 3 d orbitals, with strong electroncorrelations. Now it has been realized that in 4 d and the5 d systems, whose orbitals are spatially more extended,a regime of intermediate correlation appears. More-over, they display significant spin-orbit coupling, whichmodifies their electronic structure as recently verified inSr IrO [3]. This is a largely unexplored domain, butalready tantalizing new phenomena have been glimpsed.For example, in the 5 d magnetic insulator, Na IrO [4], adisordered ground state persists down to the lowest mea-sured temperatures, making it a prime candidate for aquantum spin liquid[5].It is known that strong spin-orbit interactions can leadto a novel phase of matter, the topological insulator[6].However, the experimental candidates uncovered so farhave weak electron correlations. Recently, it was realizedthat the iridates are promising candidates to realize topo-logical insulators[7], and that Iridium based pyrochloresin particular [8], provide a unique opportunity to studythe interplay of Coulomb interactions, spin-orbit couplingand the band topology of solids.The pyrochlore iridates, with general formula A Ir O ,where A = Yttrium, or a Lanthanide element, will bethe main focus of this work. Both the A and Ir atomsare located on a network of corner sharing tetrahedra[9, 10]. Pioneering experiments[11] on the pyrochlore iri-dates, revealed an evolution of ground state propertieswith increasing radius of the A ion, which is believedto tune electron correlations. While A =Pr is metallic, A =Y is an insulator as low temperatures. Subsequently,it was shown that the insulating ground states evolvefrom a high temperature metallic phase, via a magnetic transition[12, 13]. The magnetism was shown to arisefrom the Ir sites, since it also occurs in A =Y, Lu, wherethe A sites are non-magnetic. While its precise natureremains unknown, the absence of a net moment rules outferromagnetism.We show that electronic structure calculations can nat-urally account for this evolution and points to a novelground state whose properties are described here. First,we find that magnetic moments order on the Ir sites ina non-colinear pattern with moment on a tetrahedronpointing all-in or all-out from the center. This structureretains inversion symmetry, a fact that greatly aids theelectronic structure analysis. While the magnetic pat-tern remains fixed, the electronic properties evolve withcorrelation strength. For weak correlations, or in theabsence of magnetic order, a metal is obtained, in con-trast to the interesting topological insulator scenario ofRef. [8]. With strong correlations we find a Mott insula-tor, with all-in/all-out magnetic order. However, for thecase of intermediate correlations, relevant to Y Ir O ,the electronic ground state is found to be an unusual Dirac semi-metal , with linearly dispersing Dirac nodes at the chemical potential. Indeed, this dispersion is anal-ogous to graphene[14], but occurs inside a three dimen-sional magnetic solid. The small density of states leads toa vanishing conductivity at low temperatures. The Diracfermions here are rather different from those in three di-mensional semi-metals such as elemental Bismuth, whichare inversion symmetric and non-magnetic. Here, theDirac fermions at a particular momentum are describedby a handedness (which is left or right handed), and atwo component wavefunction. They cannot be gappedunless they mix with a fermion of opposite handedness.In contrast, Dirac fermions in Bismuth have four com-ponent wavefunctions, no particular handedness, and are a r X i v : . [ c ond - m a t . s t r- e l ] J un typically gapped. Such a three dimensional electronicstructure has, to our knowledge, not been discussed be-fore.A key property of this Dirac semi-metal phase of twocomponent Dirac fermions, is unusual band topology,reminiscent of topological insulators. Since the bulkfermi surface only consists of a set of momentum points,surface states can be defined for nearly every surface mo-mentum, and are always found to occur on certain sur-faces. They take the shape of ‘Fermi arcs’ in the surfaceBrillouin zone, that stretch between Dirac points. Hencewe term this phase topological Dirac metal .We also mention the possibility of an exotic insulat-ing phase emerging when the Dirac points annihilate inpairs, as the correlation are reduced. This phase showsa topological magnetoelectric effect[15], captured by themagneto-electric parameter θ = π , whose value is pro-tected by the inversion symmetry. Since it is analogous tothe axion vacuum in particle physics[16], so we call it the θ = π Axion insulator. Although our LSDA+U+SO cal-culations find that a metallic phase intervenes before thispossibility is realized, we note that LDA systematicallyunderestimates gaps, so this scenario could well occur inreality. Finally, we mention that modest magnetic fieldscould induce a reorientation of the magnetic moments,leading to a metallic phase. Our results are summarizedin the phase diagram Figure 1. Previous studies consid-ered ferromagnetism [17], and structural distortion [18]in iridates.Our calculations suggest that new functionalities suchas controlling electrical properties via magnetic textures,and field induced metallic states can be realized in thesematerials, with implications for spintronics, magneto-electrical and magneto-optical devices.
METHOD
We perform our electronic structure calculations basedon local spin density approximation (LSDA) to densityfunctional theory (DFT) with the full–potential, all–electron, linear–muffin–tin–orbital (LMTO) method[19].We use LSDA+U scheme[20] to take into account theelectron–electron interaction between Ir 5 d electrons andvary parameter U between 0 and 3 eV for Ir 5d electronsto see what effects the on site Coulomb repulsion wouldbring to the electronic structure of Iridates. In general,we expect that U can be somewhere between 1 and 2eV for the extended 5d states. When the A site is arare earth element, we also add the Coulomb interac-tion for the localized 4 f electrons and use U = 6 eV.We use a 24 × ×
24 k–mesh to perform Brillouin zoneintegration, and switch off symmetry operations in orderto minimize possible numerical errors in studies of vari-ous (non–)collinear configurations. We use experimentallattice parameters[12] in all set ups.
FIG. 1: Sketch of the phase diagram for pyrochlore iridiatesfrom our microscopic electronic structure calculation: Hori-zontal axis corresponds to the increasing interaction amongIr 5d electrons (the scale is obtained using LSDA+SO+Umethod) while the vertical axis corresponds to external mag-netic field which aligns the moments and triggers a transi-tion out of the zero field non-collinear ”all-in/all-out” groundstate. We find normal magnetic metal, Mott insulator andDirac semi-metal phases, and also possibly an exotic insu-lator, a magnetic insulator with magneto-electric parameter θ = π , which we label here as axion insulator. Throughout, we exploit inversion symmetry which con-strains the phase diagram, by tracking wavefunction par-ities at time reversal invariant momenta. Near electronicphase transitions, a low energy k.p theory is developedto understand qualitative features of the neighboringphases. Finally, topological band theory based on mo-mentum space Berry connections is utilized in deducingthe physical properties of the phases.
MAGNETIC CONFIGURATION
We first study magnetic configuration and discuss ourresults for Y Ir O . Since the strength of the spin or-bit (SO) coupling is large for Ir 5 d electrons, and leadsto insulating behavior in Sr IrO [3], we perform theLSDA+U+SO calculations. There are four Ir atoms in-side the unit cell forming a tetrahedral network as shownin Fig.1which is geometrically frustrated. Thus, we carryout several calculations with the initial state to be (i)ferromagnetic, with moment along (100), (111), (110) or(120) directions (ii) antiferromagnetic with two sites ina tetrahedron along and other two pointed oppositely tothe directions above; non-colinear structures (iii) ”all–in/out” pattern (where all moments point to or awayfrom the centers of the tetrahedron see Fig 1), (iv) ”2–in/2–out” (two moments in a tetrahedron point to thecenter of this tetrahedron, while the other two momentspoint away from the center, i.e. the spin–ice[21] configu- FIG. 2: The pyrochlore crystal structure showing Ir cornersharing tetrahedral network and the magnetic configurationcorresponding to the ”all–in/all–out” alignment of moments.A degenerate state is obtained on reversing the moments. ration), and (v) ”3–in/1–out” magnetic structures.We find that the ”all–in/out” configuration is theground state. Different from other magnetic configura-tions, during the self–consistency the ”all–in/out” statewill retain their initial input direction; thus, there is nonet magnetic moment. This is consistent with the ab-sence of the magnetic hysteresis in experiments[12]. Theall-in/all-out magnetic configuration was predicted to oc-cur in pyrochlore antiferromagnets with Dzyaloshinsky-Moriya (D-M) interactions[22]. Symmetry dictates theform of D-M interactions except for the sign, which leadsto two cases, direct and indirect D-M. The all-in/all-outstate is the unique ground state for the former while theindirect D-M ground state is a coplanar state with thefour spins being either antiparallel of orthogonal[22]. Wefind that the indirect D-M pattern has higher energy thanthe all-in/all-out state. This provides further evidencethat the all-in/all-out spin configuration is the naturalground state.The next lowest energy configuration is the ferromag-netic state. Interestingly, the rotation of magnetizationdoes not cost much energy despite strong SO interac-tions. The (111) direction is found to be lowest ferro-magnetic state, but the energy difference between thisand the highest energy (001) state is just about 4.17meV per unit cell. Also, all of them produce a consider-able net magnetic moment in contrast to the experiment[11, 12, 23]. Our findings are summarized in Table I for atypical value of U=1.5 eV, and similar results are foundfor other values of U in the range from 0 to 3 eV. Wefind that the energy difference between the ground andseveral selected excited states with different orientationsof moments is small. Therefore, modest magnetic fieldsmay induce a transition into the ferromagnetic state..
TABLE I: The spin (cid:104) S (cid:105) and orbital (cid:104) O (cid:105) moment (in µ B ),the total energy E tot (in meV) for several selected magneticconfigurations of Y Ir O as calculated using LSDA+U+SOmethod with U=1.5 eV. The IDM is a coplanar configurationpredicted for one sign of D-M interactions in Ref. [22]Configuration: (001) (111) 2–in/2–out IDM all–in/out (cid:104) S (cid:105) (cid:104) O (cid:105) tot (meV) 5.47 1.30 3.02 2.90 0.00 ELECTRONIC PHASES AND WAVEFUNCTIONPARITIES
We now discuss electronic properties of Iridates thatemerge from our LSDA+U+SO calculations. A varietyof phases ranging from normal metal at small U to Diracsemi-metallic at intermediate U ∼ . U above 2 eV with non–collinear mag-netic ”all-in/out” ordering are predicted. Since pressureor chemical substitution may alter the screening and theelectronic bandwidth resulting in changes in U we ex-pect that these phases can be observed experimentally iniridates.The basic features of the electronic structure can beunderstood by considering each of four Ir atoms in py-rochlore lattice which is octahedrally coordinated by sixO atoms. This makes the Ir 5 d state split into doubly de-generate e g and triply degenerate t g states. Due to theextended nature of Ir 5 d orbital, the crystal–field split-ting between t g and e g is large with the e g band to be2 eV higher than the Fermi level. The bands near theFermi level are mainly contributed by Ir t g with somemixing with O 2 p states. SO coupling has a considerableeffect on these t g states: it lifts their degeneracy andproduces quadruplet with J eff = 3 / J eff = 1 / J = 3 / J = 5 / J = 3 / J = 5 / doublet andΓ quadruplets. Since Ir occurs in its 4+ valence, its 5electrons would fill completely J = 3 / doublet thought asthe state with J eff = 1 / doublets of four Ir atoms.The precise behavior of these electronic states dependson magnetic configuration. Our band structure calcula-tions for collinear alignments of moments show metallicbands regardless the value of U that we use in our sim-ulations. On the other hand, we find that the electronicstates for the non–collinear ”all-in/out” magnetic state FIG. 3: Evolution of electronic band structure of Y2Ir2O7shown along high symmetry directions, calculated usingLSDA+U+SO method with three different values of U equal(a) 0, (metallic) (b) 1.5 eV, and (c) 2 eV. (Insulator withsmall gap). The Dirac point that is present in case (b), is notvisible along high symmetry lines. depend strongly on the actual value of U used in the cal-culation. In particular, we predict that when U is lessthan 1 eV, the ground state is a normal metal while if U is about 1.8 eV or larger, we find the band structureto be insulating with an energy gap whose value dependson U . FIG. 4: Calculated energy bands in the vicinity of the Fermilevel using LSDA+U+SO method with U=1.5 eV. on left:corresponds to plane k z = 0 with band parities shown; (b)corresponds to plane k z = 0 . π/a where Dirac point is pre-dicted to exist. The shaded plane is at the Fermi level. Weak Correlations:
An interesting recent study pro-posed a tight–binding model for the non-magnetic phaseof the iridates, which was a topological insulator [8], anatural phase on the pyrochlore lattice [8, 24]. Our LDAstudies of the realistic electronic structures contradictthis, instead we find a metallic phase (see Fig. 3a). Onecan understand the discrepancy by analyzing the struc-ture of energy levels at the Γ point (Brillouin Zone center)for the low energy 8–band complex, composed of the four J eff = 1 / TABLE II: Calculated parities of states at Time Reversal In-variant Momenta (TRIMs) for several electronic phases of theiridates. Only the top four filled levels are shown in order ofincreasing energy.Phase Γ
X, Y, Z L (cid:48) L ( × Strong Correlations and the Mott Limit:
When
U > . U to large values, where a site localized mo-ment is expected, i.e. the Mott insulator. This canbe further verified by calculating the parity eigenvalues.Note that all the magnetic structures considered abovepreserve inversion (or parity) symmetry. Thus, if wepick e.g. an Iridium atom as the origin of our coor-dinate system, then inversion r → − r leaves the crys-tal structure and magnetic pattern invariant. This im-plies a relation between crystal momenta ± k . At specialmomenta, called TRIMs (Time Reversal Invariant Mo-menta), that are invariant under inversion, we can labelstates by parity eigenvalues ξ = ±
1. In the Brillouin zoneof the FCC lattice these correspond to the Γ = (0 , , X, Y, Z [=2 π (1 , ,
0) and permutations] and four L points [ π (1 , ,
1) and equivalent points]. These par-ities are very useful to study the evolution of the bandstructure. The TRIM parities of the top four occupiedbands, in order of increasing energy, are shown in Table2. Note, although by symmetry all L points are equiv-alent, the choice of inversion center at an Iridium sitesingles out one of them, L (cid:48) . With that choice the pari-ties at the two sets of L points are the opposite of oneanother. The parities of the all-in/out state remains un-changed above U > U c ∼ . U = 2 eV. It is readily seen that these paritiesarise also for a site localized picture of this phase, whereeach site has an electron with a fixed moment along theordering direction. Due to the possibility of such a lo-cal description of this magnetic insulator, we term it theMott phase. Intermediate Correlations:
For the same all-in/outmagnetic configuration, at smaller U = 1 . FIG. 5: Brillouin zone of the FCC lattice with locations ofDirac points (shown by + signs denoting their ”positive” chi-ral charges) as found by our LSDA+U+SO calculation withU=1.5 eV for Y2Ir2O7. extension of the Mott insulator. However, a closer look atthe parities reveals that a phase transition has occurred.A pair of levels with opposite parity are exchanged atthe L points. Near this crossing point it can readily beargued that only one of the two adjacent phases can beinsulating[30]. Since the large U phase is found to besmoothly connected to a gapped Mott phase, it is reason-able to assume the smaller U phase is the non–insulatingone. This is also borne out by the LSDA+U+SO bandstructure. A detailed analysis perturbing about this tran-sition point (the k.p expansion see [30]) allows us to showthat a Dirac semi-metal is expected for intermediate U ,with 6 Dirac nodes about every L point. Indeed, in theLSDA+U+SO band structure, we find a 3 dimensionalDirac crossing located within the ΓXL plane of the Bril-louin zone. This is illustrated in Fig.5 and correspondsto the k–vector (0 . , . , . π/a . There also are twoadditional Dirac points in the proximity of the point Lrelated by symmetry. When U increases, these pointsmove toward each other and annihilate all together atthe L point close to U = 1 . < U < Ir O ismost likely the semi-metallic state with the Fermi surfacecharacterized by a set of Dirac points but in proximityto a Mott insulating state. Both phases can be switchedto a normal metal if Ir moments are collinearly orderedby a magnetic field. Topological Dirac semi-metal
The effective Hamiltonian in the vicinity of a Diracmomentum k = k + q is: H D = (cid:88) i =1 v i · q σ i (1), where energy is measured from the chemical poten-tial, and σ i are the three Pauli matrices. The three ve-locity vectors v i are generically non-vanishing and lin-early independent. The energy dispersion is ∆ E = ± (cid:113)(cid:80) i =1 ( v i · q ) . Note, by inversion symmetry, theremust exist Dirac points at both k and − k , whose ve-locity vectors are reversed. One can assign a chirality (or chiral charge) c = ± c = sign( v · v × v ), so Dirac points related by in-version have opposite chirality. Note, since the 2 × L point thereare three Dirac points related by the three fold rotation,which have the same chiral charge. Fig.5 denotes thosepoints as ”+” dots. Another three Dirac points with op-posite chirality, related by inversion. Thus, there are 24Dirac points in the whole Brillouin zone. Since they areall related by symmetry, they are at the same energy.The chemical potential is fixed to be at the Dirac pointenergy as verified in the microscopic calculation. TheFermi velocities at the Dirac point are found to be typ-ically an order of magnitude smaller than in graphene.We briefly note that this Dirac semimetal is a criticalstate with power law forms for various properties, whichwill be described in more detail elsewhere. For example,the density of states N ( E ) ∝ E . The small density ofstates makes it an electrical insulator at zero tempera-ture. For a single node with isotropic velocity v , the a.c.conductivity in the free particle limit of the clean systemis σ (Ω) = e h | Ω | v tanh | Ω | / k B T . Surface States:
We now discuss surface states thatare associated with the presence of the two componentDirac fermions. We first note that they behave like’magnetic’ monopoles of the Berry flux whose chargeis given by the chirality.The Berry connection, a vec-tor potential in momentum space, is defined by A ( k ) = (cid:80) Nn =1 i (cid:104) u n k |∇ k | u n k (cid:105) where N is the number of occupiedbands. As usual, an analog of the magnetic field, theBerry flux, is defined as F = ∇ k × A . Now considerenergy eigenstates at the Fermi energy (taken to be at E = 0). In the bulk, this corresponds to the set of Diracpoints, hence the bulk Fermi surface is a collection of Surface State k z k x k y FIG. 6: Illustration of surface states arising from bulk Diracpoints. For simplicity, only a pair of Dirac points with oppo-site chirality are shown. The imaginary cylinder in momen-tum space has unit Chern number, due to the Berry monopoleat the Dirac point. Hence a surface state must arise, as shownschematically in the same plot. When the Fermi energy is atthe Dirac point, a Fermi arc is present which connects the sur-face momenta of the projected bulk Dirac points of oppositechirality.
Fermi points. However, in the presence of a surface (saythe plane z = 0), new low energy states may be gener-ated. We show that these will occur along a curve in thesurface Brillouin zone as is illustrated in Fig. 5. The endpoints of this curve occur at the bulk Fermi point mo-menta, projected onto the surface Brillouin Zone. Also,the curve connects Dirac nodes with opposite monopolecharge. If more than one Dirac node projects to the samesurface momentum, the sum of the monopole chargesshould be considered. This is argued by showing thatthere must be Fermi arcs on the surface Brillouin zoneemanating from the projection ( k x , k y ) of the monopoleas argued below. Origin of Surface States:
We now prove that the bandtopology associated with the Dirac point leads to sur-face states. Construct a curve in the surface Brillouinzone encircling the projection of the bulk Dirac point,which is traversed counterclockwise as we vary the pa-rameter λ : 0 → π ; k λ = ( k x ( λ ) , k y ( λ )) (see Fig. 6).We show that the energy (cid:15) λ of a surface state at momen-tum k λ crosses E = 0. Consider H ( λ, k z ) = H ( k λ , k z ),the gapped Hamiltonian of the two dimensional subsys-tem defined by this curve. The two periodic parameters λ, k z define the surface of a torus in momentum space(see Fig. 6). The Chern number of this two dimensionalband structure is given by the Berry curvature integra-tion: π (cid:82) F dk z dλ which, by Stokes theorem, simply cor-responds to the net monopole density enclosed withinthe torus. This is obtained by summing the chiralitiesof the enclosed Dirac nodes. Consider the case when the net chirality is unity, corresponding to a single enclosedDirac node. Then, the two dimensional subsystem de-fines a quantum Hall insulator with unit Chern number.When defined on the half space z <
0, this corresponds toputting the quantum Hall state on a cylinder, and hencewe expect a chiral edge state. Its energy (cid:15) λ spans theband gap of the subsystem, as λ is varied. Hence, thissurface state crosses zero energy somewhere on the sur-face Brillouin zone k λ . Such a state can be obtained forevery curve enclosing the Dirac point. Thus, at zero en-ergy, there is a Fermi line in the surface Brillouin zone,that terminates at the Dirac point momenta (see Fig.6). An arc beginning on a Dirac point of chirality c hasto terminate on a Dirac point of the opposite chirality.Clearly, the net chirality of the Dirac points within the( λ, k z ) torus was a key input in determining the numberof these states. If Dirac points of opposite chirality lineup along the k z direction, then there is a cancelation andno surface states are expected.For U = 1 . . , . , . π/a and equivalentpoints (see Fig.3). They can be thought of as occur-ing on the edges of a cube, with a pair of Dirac nodesof opposite chirality occupying each edge, as, e.g., thepoints (0 . , . , . π/a and (0 . , . , − . π/a. For the case of U = 1 . . π/a ). Thus, the (111) and (110) sur-faces would have surface states connecting the projectedDirac points. If, on the other hand we consider the sur-face orthogonal to the (001) direction, it would lead tothe Dirac points of opposite chirality being projected tothe same surface momentum, along the edges of the cube.Thus, no protected states are expected for this surface. Model Calculation:
To verify these theoretical con-siderations, we have constructed a tight binding modelwhich has features seen in our electronic structure cal-culations for YIr O . We consider only t g orbitals of Iratoms in the global coordinate system. Since Ir atomsform tetrahedral network (see Fig. 2), each pair ofnearest neighboring atoms forms a corresponding σ − likebond whose hopping integral is denoted as t and othertwo π − like bonds whose hopping integrals are denotedas t (cid:48) . To simulate the appearance of the Dirac point itis essential to include next–nearest neighbor interactionsbetween t g orbitals which are denoted as t (cid:48)(cid:48) . With theparameters t = 0 . , t (cid:48) = 0 . t, t (cid:48)(cid:48) = − . t , the value of theon–site spin–orbit coupling equal to 2 . t and the appliedon–site splitting between spin up and spin down statesequal to 0.1 referred to the local quantization axis whichsimulates our non–collinear ’all–in/out’ configuration wecan model both the bulk Dirac metal state and its sur-face. The calculated (110) surface band structure for theslab of 128 atoms together with the sketch of the obtainedFermi arcs is shown in Fig. 7. Notice that since the slabcalculation involves two surfaces, the corresponding sur-face states and Fermi arcs for both surfaces are generated. FIG. 7: Calculated surface energy bands corresponding to(110) surface of the pyrochlore iridate Y2Ir2O7. A tight bind-ing approximation has been used to simulate the bulk bandstructure with 3D Dirac point as found by our LSDA+U+SOcalculation. The plot corresponds to diagonalizing 128 atomsslab with two surfaces. The top inset shows the deduced Fermiarcs connecting projected bulk Dirac points of opposite chi-rality. The bottom inset shows a sketch how these Fermi arcsare expected to behave for the (111) surface.
We also display the expected surface states for the (111)surface. Note, no special surface states are expected forthe (001) surface.
OTHER TOPOLOGICAL PHASES
We recall that topological insulators are non-magneticband insulators with protected surface states [6]. Timereversal symmetry is required in the bulk to definethese phases. When the surface states are eliminatedby adding, for example magnetic moments only on thesurface, a quantized magneto-electric response is ob-tained, where a magnetic field induces a polarization: P = θ e πh B , with the coefficient | θ | [15] is only definedmodulo 2 π . Under time reversal, θ → − θ . Apart fromthe trivial solution θ = 0, the ambiguity in the defini-tion of θ allows also for θ = π . For topological insulators θ = π .In magnetic insulators, θ is in general no longer quan-tized [26]. However, when inversion symmetry is re-tained, θ is quantized again, since inversion also changesits sign. Thus again θ = 0 , π mod2 π , and an insulatorwith the latter value will be termed a θ = π axion insu-lator.Which is the appropriate description of the pyrochlore iridate phases we have described? For the Mott insulator,at large U , the charge physics must be trivial and so wemust have θ = 0. Next, since the Dirac semi metal phaseis gapless in the bulk, θ is ill defined. However, we notethat on reducing U , the location of the Dirac points shift,with nodes of opposite chirality approaching each other.If these meet and annihilate, then one recovers a gappedphase in the low U regime. However, in the process theresulting phase will have θ = π . Indeed, the presence ofthe intervening Dirac phase can be deduced from the re-quirement that θ has to change between these two quan-tized values. As described elsewhere[25], the conditionfor θ = π when deduced from the parities, turns out tobe the same as the Fu-Kane formula, for time reversalsymmetric insulators [27, 28], i.e. if the total numberof filled states of negative parity at all TRIMs taken to-gether is twice an odd integer, then θ = π . Otherwise θ = 0. The small U insulator has the same parities asthe Dirac semi-metal, since the Dirac points annihilateaway from a TRIM. From Table II we can see that indeedthis corresponds to θ = π , since there are 14 negative par-ity filled states, while the Mott insulator corresponds to θ = 0, having 12 negative parity filled states.Unfortunately, within our LSDA+U+SO calculation, ametallic phase intervenes on lowering U ≤ . U ‘Slater’ insulator,unlike in many other cases where they are smoothly con-nected to one another. Inversion symmetry is critical inpreserving this distinction.In summary, a theoretical phase diagram for the phys-ical system is shown in Figure 1 as a function of U andapplied magnetic field, which leads to a metallic statebeyond a critical field. The precise nature of these phasetransformations are not addressed in the present study ACKNOWLEDGEMENTS
A.V. thanks L. Balents, J. Orenstein and R. Rameshfor insightful discussions. X.W. acknowledges supportby National Key Project for Basic Research of China(Grant No. 2006CB921802, and 2010CB923404), NSFCunder Grant No. 10774067, and 10974082. S.S. acknowl-edges support by DOE SciDAC Grant No. SE-FC02-06ER25793 and thanks Nanjing University for the kindhospitality during his visit to China. X.W. and S.S alsoacknowledge support from Kavli Institute for Theoreti-cal Physics where this work has been initiated. The workat Berkeley was supported by the Office of Basic EnergySciences, Materials Sciences Division of the U.S. Depart-ment of Energy under contract No. DE-AC02-05CH1123.
APPENDIXI. Effective k.p Theory and Intervening Dirac MetalPhase
Consider a pair of states at the L point which haveopposite parity, and cross each other as we tune U . Wewant to understand what happens to the band structure.The L point has two symmetries which do not changeits crystal momentum. First of course is inversion, and wecan label states by the eigenvalues P = ±
1. The second is120 o rotations about a line joining L − Γ. There are threepossible eigenvalues which we call s = − / , / , / { P, s } . Nowconsider writing the effective 2 × At the L point: Since we have inversion symmetry,the two states taken to be eigenvalues of τ z = ± H ( L ) = ∆ τ z where the coefficient ∆ changes sign when the lev-els pass through each other. Note, the s quantumnumber of the two levels is irrelevant here.2. Along the Γ − L Direction:
We still have the quan-tum number s , but not P , since that inverts themomentum. Denoting by k z the momentum alongthis line deviating from the L point, we have twocases. If the s quantum number of the two levelsis different, they still cannot mix, so the effectiveHamiltonian is H = (∆ + k z ) τ z . Now, when ∆ < k z = ±√− ∆, where thereare nodes along this Γ − L line. You can see thisfor the s = 1 / s = 3 / same s quantumnumber they can mix, once you move away from L .Now the effective Hamiltonian is: H (Γ − L ) = (∆ + k z ) τ z + k z τ x where the second term arises since inversion is bro-ken on moving away from L allowing for mixing.Now, the spectrum is E = (cid:112) ( k z − | ∆ | ) + k z , for∆ <
0, so despite a level crossing there is no nodealong the Γ − L line.3. General Point in BZ:
In the latter case, does thismean there are no Dirac points? No - we just needto move off the Γ − L line. Let the deviation be k ⊥ , a 2 vector. The fact the 2 levels have oppositeparity means we need an odd function of k ⊥ to induce a matrix element between the levels. Andalso, since k ⊥ is a 2 vector, it transforms under therotation - the rotationally symmetric form allowedis ∆ H = k ⊥ cos 3 θτ x + k ⊥ sin 3 θτ y . Putting this alltogether we have the effective Dirac Hamiltoniannear the L point: H ( k ) = (∆ + k z ) τ z + ( k z + k ⊥ cos 3 θ ) τ x + k ⊥ sin 3 θτ y Note, this has the form A ( k ) τ z + B ( k ) τ x + C ( k ) τ y .For a node, A = B = C = 0. This occurs if: A = 0so k z = + (cid:112) −| m | , and C = 0 implies θ = pπ/ p is an integer; and finally k ⊥ = k z , when p = 1 , ,
5. Similarly for k z <
0, the nodes areinverted. In all we have 6 nodes for this L point,24 Nodes in all.Actually this is not the complete expansion. Strictlywe should write A ( k ) = m + k z + αk ⊥ . This is an ef-fective mass Hamiltonian near the L point. If α > θ = π magnetic axion insulator to Dirac metal onincreasing U). However , if α <
0, this completely changesthe conclusions, as discussed below. In fact, this turnsout to be the physically relevant case according to theelectronic structure calculations for Y Ir O .Let us assume α = − /m <
0. Then the effectiveHamiltonian: H ( k ) = (∆+ k z m − k ⊥ m ) τ z +( βk z + k ⊥ cos 3 θ ) τ x + k ⊥ sin 3 θτ y (2)where a few parameters have been labeled. The Diracnodes then are at: (i) C = 0 → θ = pπ/ B = 0 → k z = ± k ⊥ /β depending on whether welook at p =odd or p = even. Finally, using these relationswe have for A = 0 equation:∆ + k ⊥ β m − k ⊥ m = 0for small ∆ this has the solution k ⊥ = 2 m ∆. Note,this has a solution only for ∆ >
0, i.e. before the gap getsinverted at the L point on increasing U . Thus, in thisscenario, there is a Dirac point only in the small U phase.These Dirac points live along θ = pπ/ k x = k y plane that contains the points Γ − L − K and rotations thereof. II. Summary of Experiments
We now summarize the experimental facts about thepyrochlore iridates A Ir O . Early work revealed thatincreasing the A ion radius triggered a metal-insulatortransition in the ground state. Thus while A =Pr ismetallic, A =Y is insulating at low temperatures [11].Reducing the ionic radius is believed to narrow the band-width and increase correlations. Subsequent improve-ment in sample quality revealed that A =Eu, Sm, Ndalso displayed low temperature insulating states[13]. Inthese systems, a metal insulator transition is clearly ob-served on cooling (eg. at T MI = 120 K for A =Eu). Atthe same temperature, a signature in magnetic suscepti-bility is also observed, indicative of a magnetic transition[12]. This magnetic signature, wherein the field cooledand zero field cooled magnetic susceptibilities separatebelow the transition temperature, is reminiscent of a spinglass state. Since this signature is see in A =Y, Lu [12],with nonmagnetic A site atoms, it is associated with Irsite moments. In Y Ir O , no sharp resistivity signaturehas been reported at the magnetic transition, but theresistivity climbs steeply on cooling below this temper-ature. Moreover, light hole doping suppresses both theinsulating state and the magnetic transition[23]. Finally,we note that a thermodynamic signature of the magnetictransition, a bump in the specific heat, is observed inclean samples of A =Sm [13]. X-ray scattering did notobserved any structural change below the ordering tran-sition [13], although the presence of new lines in the Ra-man spectroscopy [29] has been attributed to lowering ofcubic symmetry in A = Eu, Sm but not in A =Nd.We now discuss how the present theoretical descriptionsits with these facts. We propose that the low tempera-ture state of Y Ir O (and also possibly of A = Eu, Smand Nd iridates) is the Dirac semi-metal, with all-in/all-out magnetic order. This is broadly consistent with theinterconnection between insulating behavior and mag-netism observed experimentally. It is also consistent withbeing proximate to a metallic phase on lowering the cor-relation strength, such as A =Pr. In the clean limit, athree dimensional Dirac semi-metal is an electrical insu-lator, and can potentially account for the observed elec-trical resistivity. The noncolinear magnetic order pro-posed has Ising symmetry and could undergo a contin-uous ordering transition. Since this configuration is notfrustrated, it is not clear how spin glass behavior wouldarise, but the observed magnetic signature could perhapsalso arise from defects like magnetic domain walls. A di-rect probe of magnetism is currently lacking and wouldshed light on this key question. [1] J. Orenstein and A. J. Millis, Science 288, 468 - 474(2000)[2] Y. Tokura and N. Nagaosa, Science 288, 462-467 (2000) [3] B.J. Kim, et al. , Phys. Rev. Lett. , 076402 (2008),B.J. Kim, H. Ohsumi, T. Komesu, S. Sakai, T. Morita,H. Takagi, and T. Arima, Science . 1329 (2009).[4] Y. Okamoto, M. Nohara, H. Aruga-Katori, and H. Tak-agi, Phys. Rev. Lett. , 137207 (2007).[5] P.A. Lee, Science , 1306 (2008); L. Balents, Nature(London) , 199 (2010).[6] M.Z. Hasan and C.L. Kane, arXiv:1002.3895 (2010). J.E. Moore, Nature 464(7286):194-8 (2010). X. Qi and S.C. Zhang, Physics Today 63, 33 (2010).[7] A. Shitade, H. Katsura, J. Kune˜s, X.-L. Qi, S.-C. Zhang,and N. Nagaosa, Phys. Rev. Lett. , 256403 (2009).[8] D. A. Pesin, L. Balents, Nature Physics 6, 376 - 381(2010).[9] M.A. Subramanian, G. Aravamudan and G.V. SubbaRao, Prog. Solid St. Chem. , 55 (1983); S.T. Bramwelland M.J.P. Gingras, Science , 1495 (2001); A.P.Ramirez, Ann. Rev. Mater. Sci. , 453 (1994).[10] J. S. Gardner, M. J. P. Gingras, J. E. Greedan, Rev.Mod. Phys. , 53 (2010).[11] D. Yanagishima, and Y. Maeno, J. Phys. Soc. Jpn. ,2880 (2001).[12] N. Taira, M. Wakeshima and Y. Hinatsu, J. Phys.: Con-dens. Matter , 5527 (2001).[13] K. Matsuhira, M. Wakeshima, R. Nakanishi, T. Yamada,A. Nakamura, W. Kawano, S. Takagi, and Y. Hinatsu, J.Phys. Soc. Jpn. , 043706 (2007).[14] A. K. Geim and K. S. Novoselov, Nature Materials 6, 183- 191 (2007).[15] X. Qi, Taylor Hughes and S. C. Zhang, Phys. Rev. B 78,195424 (2008).[16] Frank Wilczek, Phys. Rev. Lett., , 1799 (1987).[17] K. Maiti, Solid State Commun. , 1351 (2009).[18] B. J. Yang and Y. B. Kim, arxiv/1004.463 .[19] S. Y. Savrasov, Phys. Rev. B , 767 (1997).[21] R. Siddharthan, B.S. Shastry, A.P. Ramirez, A. Hayashi,R.J. Cava, and S. Rosenkranz, Phys. Rev. Lett. , 1854(1999); M.J. Harris, S.T. Bramwell, D.F. McMorrow, T.Zeiske, and K.W. Godfrey, Phys. Rev. Lett. , 2554(1997).[22] M. Elhajal, B. Canals, R. Sunyer and C. Lacroix, Phys.Rev. B , 2578(2002); M. Soda, N. Aito, Y. Kurahashi, Y. Kobayashi,M. Sato, Physica B , 1071 (2003).[24] H.-M. Guo and M. Franz, Phys. Rev. Lett103