Surfaces and slabs of fractional topological insulator heterostructures

Sharmistha Sahoo,1 Alexander Sirota,1 Gil Young Cho,2 and Jeffrey C. Y. Teo1, ∗ 1Department of Physics, University of Virginia, Charlottesville, VA22904, USA 2School of Physics, Korea Institute for Advanced Study, Seoul 02455, Korea Fractional topological insulators (FTI) are electronic topological phases in (3 + 1) dimensions enriched by time reversal (TR) and charge U(1) conservation symmetries. We focus on the simplest series of fermionic FTI, whose bulk quasiparticles consist of deconfined partons that carry fractional ∗ electric charges in integral units of e = e/(2n + 1) and couple to a discrete Z2n+1 gauge theory. We propose massive symmetry preserving or breaking FTI surface states. Combining the long- ranged entangled bulk with these topological surface states, we deduce the novel topological order of quasi-(2 + 1) dimensional FTI slabs as well as their corresponding edge conformal field theories.

I. INTRODUCTION tron. In particular, we consider FTI that support decon- fined fermionic parton excitations ψ in the bulk, each car- ∗ Conventional topological insulators (TI)1–4 are time re- rying a fractional electric charge of e = e/(2n + 1). The electronicQP decomposes as ψ ψ . . . ψ . The versal (TR) and charge U(1) symmetric electronic band el ∼ 1 2n+1 insulators in three dimensions that host massless sur- (3 + 1)-DTO is based on a discrete Z2n+1 gauge the- 11 face Dirac fermions. The topologically protected surface ory . The theory supports electrically neutral string- Dirac fermion can acquire a single-body ferromagnetic like gauge flux Φ, so that a monodromy quantum phase 2πig/(2n+1) or superconducting mass by breakingTR or charge U(1) of e is obtained each time ψ orbits around it. symmetry respectively. Alternatively it can acquire a In other words, ψ carries the gauge charge g. The inte- many-body interacting mass while preserving both sym- ger g and 2n + 1 are relatively prime so that all localQP metries, and exhibit long-ranged entangled surface topo- must be combinations of the electronicQP ψel and must logical order5–8. On the other hand, fractional topolog- carry integral electric charges and trivial gauge charges. ical insulators (FTI)9–15 are long-range entangled topo- Generalizing the surface state of a conventionalTI , the logically ordered electronic phases in (3 + 1) dimensions surface of a FTI hosts massless Dirac partons coupling outside of the single-body mean-field band theory de- with a Z2n+1 gauge theory. Unlike its non-interacting scription. They carryTR and charge U(1) symmetries, counterpart whose gapless Dirac surface state is symme- which enrich its topological order (TO) in the sense that try protected in the single-body picture, a FTI is strongly a symmetric surface must be anomalous and cannot be interacting to begin with and there is no topological rea- realized non-holographically by a true (2 + 1)-D system. son for its surface state to remain gapless. In this Rapid In this Rapid Communication, we describe the topologi- Communication we focus on three types of gapped sur- cal properties of various massive surface states and quasi- face states – ferromagnetic surfaces (FS) that breakTR, (2 + 1)-D slabs of a series of FTI . In particular, we focus superconducting surfaces (SCS) that break charge U(1), on the quasi-particle (QP) structure. and symmetric surfaces which generalize the -Pfaffian surface state of a conventionalTI and is denotedT by - ∗ T surface gauge TR breaking Pf . The topological order for FTI slab with these sur- charge ζa ferromagnetic surface ↑ FTI slab with generalized Pfaffian gauge flux Φ faces are discussed in Sec.II,III andIV respectively. In topological order Pf* { parton ψ with gauge charge g Sec.V, we discuss, using an anyon condensation picture, TR symmetric electric charge e* the gluing of a pair of -Pfaffian surfaces. We conclude e2 π2k2 T FTI slab with -Pf* surfaces σ = ν κ = c B T T h 3h fractional Chern in Sec.VI with remarks on a complementary way to un- FTI slab with topological order generalized Pfaffian magnetic monopole F derstand these topological order17. topological order Pf* { electric charge e*/2

ν e2 c π2k2 T TR breaking { σ = κ = B 2 h 2 3h ferromagnetic surface ↓ hc/e magnetic flux surface gauge tube / Dirac string charge ζa II. FERROMAGETIC HETEROSTRUCTURE

arXiv:1701.08828v2 [cond-mat.str-el] 23 Oct 2017 FIG. 1. Summary of the QP and gauge flux content in FTI slabs. A pair of Pf∗ FTI slabs are merged into a fractional We begin with a slab that has oppositeTR breakingFS Chern FTI slab F by gluing the two TR symmetric T -Pf∗ . In theFS , in addition to the single-body Dirac mass m surfaces. Directed bold lines on the front surface are chiral for the surface parton, the Z2n+1 gauge sector also shows ∗ edge modes of the Pf and F FTI slabs. TR breaking signature. The Z2n+1 gauge theory is only present inside the FTI , and when a flux line Φ termi- We focus on a series of fermionic FTI , labeled by inte- nates at the surface, theTR breaking boundary condi- gers n, whose magneto-electric response is characterized tion confines an electrically neutral surface gaugeQP, by the θ-angle θ = π/(2n + 1) (modulo 2π/(2n + 1)) that denoted by ζa, with gauge charge a at the flux-surface associates an electric charge of e∗/2 = e/2(2n+1) to each junction (see Fig.1). This gauge flux-charge composite, magnetic monopole16, for e the electric charge of the elec- referred to as a dyon δ = Φ ζa, carries fractional spin × 2 hδ = a/(2n + 1) because a 2π-rotation about the normal sector by combining each parton with a specific number 2 axis braids a gauge charges around Φ and results in the of dyons λ = ψ γ−n ug, where ua + v(2n + 1) = 1 monodromy quantum phase of e2πia/(2n+1).TR conju- for some integer ×u, v, so that the combination is local gates all quantum phases so, a 0 modulo 2n + 1 breaks (i.e. braids trivially) with any dyons γm. λ has frac- 6≡ ∗ TR. tional electric charge qλ = e and spin hλ = 1/2 + The one-dimensional interface between twoTR conju- n3ug2/(2n + 1) modulo 1. The charge sector con- h i gateFS domains hosts a fractional chiral channel. For sists of the fractional AbelianQP products λm, where 2n+1 2n+1 example, the interface between twoFS domains with op- λ ψ ψel corresponds to the local elec- ∼ ∼ posite ferromagnetic orientations on the surface of a con- tronicQP . In particular, when a = 1 and g = 2, − − ventionalTI bounds a chiral Dirac channel 18–20, where hλ = 1/2(2n + 1) and therefore λ behaves exactly like electrons propagate only in the forward direction. Al- the LaughlinQP of the FQH state U(1)(2n+1)/2 with fill- ternatively, aTI slab with oppositeTR breakingFS is ing fraction ν = 1/(2n + 1) and chiral central charge topologically identical to a quasi-(2 + 1)-D Chern insula- ccharge = 1. Combining the neutral and charge sectors, tor21,22 and supports a chiral Dirac edge mode. Sim- the FTI slab withTR breakingFS has the decoupled ilarly, in the FTI case, the low-energy content of the tensor productTO fractional chiral channel between a pair ofTR conju- (2a) = charge Z , (1) gateFS domains can be inferred by the edge mode F h i ⊗ 2n+1 of a FTI slab withTR breakingFS that is topologi- and in the special case when a = 1 and g = 2, it is − − cally identical to a quasi-(2 + 1)-D fractional Chern in- identical to the Abelian state U(1)(2n+1)/2 SU(2n + 23–26 27 ⊗ sulator or fractional quantum Hall (FQH) state . 1)1, which has a total central charge c = 2n + 1. In The chiral (1 + 1)-D channel is characterized by two re- general, the filling fraction and chiral central charge are sponse quantities28–36 – the differential electric conduc- not definite and are subject to surface reconstruction. tance σ = dI/dV = νe2/h that relates the changes of For instance, the addition of 2N electronic Dirac fermions electric current and potential, and the differential ther- per surface modifies the two response quantities by an 2 2 mal conductance κ = dIT /dT = c(π kB/3h)T that re- equal amount ν ν + 2N, c c + 2N. lates the changes of energy current and temperature. In → → the slab geometry, they are equivalent to the Hall con- ductance σ = σxy, κ = κxy. ν = Ne/Nφ is also referred III. SUPERCONDUCTING to as the filling fraction of the FTI slab and associates the HETEROSTRUCTURE gain of electric charge (in units of e) to the addition of a magnetic flux quantum hc/e. c = cR cL is the chiral Next we move on to superconducting heterostructures. − central charge of the conformal field theory (CFT)37 that We begin with the fractional Chern FTI slab in (1) effectively describes the low-energy degrees of freedom of and introduce weak superconducting pairing, perhapsF in- the fractional chiral channel. duced by proximity with a bulk superconductor, without Since the top and bottom surfaces of the FTI slab are closing the bulk energy gap. In the simplest scenario, TR conjugate, their parton Dirac masses m and gauge this condenses all parton pairs ψ2m, which form a La- flux-charge ratio a have opposite signs. The anyon con- grangian subgroup40 – a maximal set of mutually local 2 2(2n+1) tent is generated by the partons and gauge dyons. When bosons – containing the Cooper pair ψel = ψ . a gauge flux passes through the entire slab geometry Since the parton pair ψ2 carries gauge charge 2g, which from the bottom to the top surface, it associates with is relatively prime with 2n + 1, the condensate confines total 2a gauge charges at the two surface junctions. We all non-trivial dyons γm, which are non-local and have denote this dyon by γ = Φ ζ2a, which corresponds non-trivial monodromy with ψ2. As the neutral sector × (2a) to an electrically neutral anyon in the slab with spin Z2n+1 is killed by pairing, the superconducting FTI slab hγ = 2a/(2n + 1). If a is relatively prime with 2n + 1, withTR conjugateFS has a trivial fermionicTO . It the primitive dyon generates the chiral Abelian topolog- however still carries chiral fermionic edge modes with the (2a) 38,39 same chiral central charge c . On the other hand, these ical field theory Z2n+1 , which consists of the dyons F m 2 fermionic channels also live along the line interface be- γ , for m = 0,..., 2n, with spins hγm = 2am /(2n + 1) 0 0 modulo 1 and fusion rules γm γm = γm+m , γ2n+1 = tweenTR conjugate ferromagnetic domains on the sur- γ0 = 1. In particular, when a×= 1, γn now has spin face of a weakly superconducting FTI . When the line 2n2/(2n + 1) n/(2n + 1) modulo− 1, which is identical interface hits aTR symmetric SCS island (c.f. Fig.1 − ≡ by replacing the -Pf∗ surfaces by SCS ), these chiral to that of the fundamentalQP of the SU(2n + 1) Chern- T Simons theory at level 138,39. This identifies the Abelian channels split and divide along the pair of SCS-FS line interfaces. Both of these channels are electrically neutral theories (−2) (n) = SU(2n + 1) , which has chiral Z2n+1 ∼= Z2n+1 1 as charge U(1) symmetry is broken by the superconduc- central charge c = 2n. neutral tor, and each of them carries half of the energy current The FTI slab also supports fractionally charged par- of and has chiral central charge cF /2. For example, tons ψ, each carrying a gauge charge g. The electrically theF SCS-FS heterostructure on a conventionalTI sur- (2a) charged sector can be decoupled from the neutral Z2n+1 face holds a chiral Majorana channel with c = 1/2 along 3 the line tri-junction18,19. In the specific fractional case surface anyons are confined to theTR symmetric sur- 2j+1 when a = 1 and g = 2, each SCS-FS line interfaces face except the parton combinations ψ = Ψ8j+4 and − − 2j holds 2n + 1 chiral Majorana fermions and is described ψ = 118j. Moreover, theTR breaking boundary con- a by the Wess-Zumino-Witten SO(2n + 1)1 CFT with the dition confines a gaugeQP ζ per gauge flux Φ end- central charge c = (2n + 1)/2. Analogous to the conven- ing on theFS . On the other hand, there is no gauge tional superconductingTI surface 41, the SCS of the FTI charge associated with a gauge flux ending on the - supports a zero energy Majorana bound state (MBS) at Pf∗ surface because ofTR symmetry. Thus a gauge fluxT a vortex core. Now that the condensate consists of par- passing through the entire slab corresponds to the dyon a ton pairs, vortices are quantized with the magnetic flux δ = Φ ζ with spin hδ = a/(2n + 1) modulo 1. The - ∗ ∗ × T hc/2e = (2n + 1)hc/2e. Each pair of MBS forms a two- Pf state couples non-trivially to the Z2n+1 gauge theory level system distinguished by parton fermion parity. as the parton ψ = Ψ4 carries a gauge charge g. The gen- eral surface anyons Xj, for X = 11, Ψ, Σ, must carry the gauge charge z(j) n2gj modulo 2n + 1 and associate IV. GENERALIZED T -PFAFFIAN* SURFACE to the monodromy≡ quantum phase e2πiz(j)/(2n+1) when STATE orbiting around the dyon δ. For instance, as 2n 1 modulo 2n + 1, z(4j) gj counts the gauge charge≡ − of Lastly, we describe the -Pf∗ surface state that pre- the parton combination≡ψj. serves bothTR and charge TU(1) symmetries of the FTI. TheTO of this FTI slab is therefore generated by com- Generalizing the -Pfaffian symmetric gapped surface binations of the -Pf∗ anyons and the dyon δ. We denote state of a conventionalT TI described in Ref. 7, the FTI the composite anyonT by version – referred here as -Pfaffian∗ – consists of the T z+n3ugj Abelian surface anyons 11 and Ψ , for j even, and the X˜j,z = Xj δ , (4) j j ⊗ non-Abelian Ising-like anyons Σj, for j odd. The in- where X = 11, Ψ for j even or Σ for j odd, z = 0,..., 2n dex j corresponds to the fractional electric charge qj = je/4(2n + 1). The surface anyons satisfy the fusion rules modulo 2n + 1, and ua + v(2n + 1) = 1. They satisfy the fusion rules 11j 11j0 = Ψj Ψj0 = 11j+j0 , 11j Ψj0 = Ψj+j0 , × × × 1˜1j,z 1˜1j0,z0 = Ψ˜ j,z Ψ˜ j0,z0 = 1˜1j+j0,z+z0 , Ψ Σ 0 = Σ 0 , Σ Σ 0 = 11 0 + Ψ 0 , (2) × × j × j j+j j × j j+j j+j 1˜1j,z Ψ˜ j0,z0 = Ψ˜ j+j0,z+z0 , Ψ˜ j,z Σ˜ j0,z0 = Σ˜ j+j0,z+z0 , and the spin statistics × × Σ˜ Σ˜ 0 0 = 1˜1 0 0 + Ψ˜ 0 0 . (5) j,z × j ,z j+j ,z+z j+j ,z+z 1 j2 j2 1 h = hΨ = , hΣ = − modulo 1 (3) They follow the spin statistics 11j j − 2 16 j 16 1 1 so that 11 , Ψ are bosonic, fermionic or semionic, and Σ h(1˜1 ) = h(Ψ˜ ) = h(Σ˜ ) + j j j j,z j,z − 2 j,z 16 are bosonic or fermionic. The fermion Ψ is identical to 4 j2 az2 n6ug2j2 the super-selection sector of the bulk parton ψ, which is = + − modulo 1. (6) local with respect to all surface anyons and can escape 16 2n + 1 from the surface and move into the bulk.TR symme- The j, z indices in (4) are defined in a way so that X˜ try acts on the surface anyons the same way it acts on j,0 z ˜ those in the -Pfaffian state for conventionalTI 7,17. For are local with respect to the dyons δ = 110,z and de- T 2j+1 (a) ∗ example, the parton combinations ψ = Ψ8j+4 (and coupled from the dyon sector Z2n+1. The -Pf sur- 2j ˜ T ψ = 118j) are Kramers doublet fermions (respectively face anyons belong to the subset Xj = Xj,−n3ugj, which Kramers singlet bosons), while Ψ8j (118j+4) are Kramers is a maximal sub-category that admits aTR symmetry. singlet fermions (respectively Kramers doublet bosons). The electronicQP belongs to the super-selection sector ˜ Moreover, for identical reasons as in the conventional ψel = Ψ4(2n+1),0, which is local with respect to all anyons. ∗ TI case, the -Pf state is anomalous and can only be If one gauges fermion parity and includes anyons that as- T supported holographically on the surface of a topolog- sociate 1 monodromy phase with ψel, i.e. if one includes ∗ ical bulk. For instance, the bosonicTO of the -Pf 1˜1 , Ψ˜ − for j odd and Σ˜ for j even, the Ising sector T j,z j,z j,z h i state after gauging fermion parity would necessarily vio- generated by 1 = 1˜10,0, f = Ψ˜ 0,0, σ = Σ˜ 0,0 is local with lateTR symmetry. We notice in passing that there are and decoupled from the charge Pf∗ sector generated by 5,6 alternative surfaceTO that generalize those in Refs. . 1˜ h i ∗ 1j,0. TheTO of the FTI slab thus takes the decoupled However we will only focus on the -Pf state in this tensor product form after gauging fermion parity Rapid Communication. T ∗ ∗ (a) The FTI slab with aTR symmetric -Pf top sur- ∗ Pf = charge Pf Ising Z2n+1. (7) face and aTR breaking bottomFS carriesT a novel quasi- h i ⊗ h i ⊗ (2 + 1)-DTO . Its topological content consists of the Gauging fermion parity is not the focus of this Rapid fractional partons coupled with the Z2n+1 gauge theory Communication. Nevertheless, we notice in passing that in the bulk and the -Pf∗ surface state (see Fig.1). All there are inequivalent ways of fermion parity gauging, T 4 and in order for the Pf∗ theory to have the appropriate part, all combinations that involve only ΣA or only ΣB A B central charge, (7) needs to be modified by a neutral are confined by the Ψ0 Ψ0 condensate. Other confined 17 1 A 1 B A B 1 A B A 1 B Abelian SO(2n)1 sector . However, the tensor product anyons include 1ja 1jb ,Ψja Ψjb , 1ja+2Ψjb−2,Ψja+2 1jb−2 (7) is sufficient and correct to describe the fermionicTO A B and Σja±1Σjb∓1 for ja jb modulo 8. The remaining of the FTI slab (with global ungauged fermion parity) by deconfined Ising pair splits6≡ into simpler Abelian compo- restricting to super-selection sectors X˜j,z that are local nents with respect to the electronicQP ψel. We refer to this ΣA ΣB = S+ + S− , (10) fermionicTO as a generalized Pfaffian state. ja±1 jb∓1 ja±1,jb∓1 ja±1,jb∓1 where each S± carries the same spin as the original Ising pair but differs from the other by a unit fermion S± V. GLUING T-PFAFFIAN* SURFACES ΨA/B = S∓. In general the two Abelian components are× non-local with respect to each other. For instance, the The chiral channel in (1) between a pair ofTR conju- TR symmetric surface anyons S+ and S− are mutually F ∗ + gateFS domains divides into a pair of fermionic Pf in (7) semionic when ja = jb = 0. We choose to include S in at a junction where the twoFS domains sandwich aTR the condensate b in (9) while confining S−. Furthermore, ∗ symmetric -Pf surface domain (see Fig.1). Conser- the condensate identifies the deconfined anyons that are T vation of charge and energy requires the filling fractions different up to bosons in b. and chiral central charges to equally split, i.e. 2νPf∗ = νF 11A 11B ΨA ΨB ΨA 11B 11A ΨB and 2cPf∗ = cF . For instance, in the prototype case ja jb ≡ ja jb ≡ ja+2 jb−2 ≡ ja+2 jb−2 ± A B when a = 1 and g = 2, νPf∗ = 1/2(2n + 1) and S 11 11 (11) − − ja±1,jb∓1 ja+4 jb−4 c ∗ = (2n + 1)/2. Similar to the aforementioned case, ≡ ≡ Pf F these quantities are subjected to surface reconstruction for ja jb mod 8 and ja, jb both even. Equation (11) are ≡ A A B ν ν + N, c c + N. just parton combinations. For instance, ψ = Ψ4 110 → → A B B ≡ In addition to the response quantities, theTO of for 114 Ψ4 = ψ are now free to move inside both FTI slabs the FTI slab withTR conjugateFS is related to thatF of after gluing. TheTO after the gluing is generated by the the fermionic Pf∗ by a relative tensor product partons and dyons, which behave identically to those in of (1). This proves (8). The anyon condensation gluing ∗ ∗ F ∗ = Pf b Pf . (8) of the pair of -Pf states preserves symmetries for the F same reason itT does for the conventionalTI case 7,17. This can be understood by juxtaposing theTR symmet- It is worth noting that a magnetic monopole can be ric surfaces of a pair of Pf∗ FTI slabs and condensing mimicked by a magnetic flux tube / Dirac string (with surface bosonic anyon pairs on the two -Pf∗ surfaces. flux quantum hc/e) that originates at theTR symmet- This anyon condensation42–44 procedure effectivelyT glues ric surface interface and passes through one of the two the two FTI slabs together along theTR symmetric sur- FTI slab, say the A slab. In the prototype a = 2 − faces (see Fig.1). The relative tensor product involves and g = 1, the filling fraction ν ∗ = 1/2(2n + 1) of b − Pf first taking a decoupled tensor product when the two the quasi-two-dimensional slab ensures, according to the Pf∗ FTI slabs are put side by side. Among⊗ theTR sym- Laughlin argument28, that the monopole associates to metric surface anyons in ( -Pf∗)A ( -Pf∗)B where A, B the fractional charge q = 1/2(2n + 1), which is carried refers to the two slabs, weT condense⊗ theT collection of elec- by the confined -Pf∗ surface anyons 11A or ΨA. This T 2 2 trically neutral bosonic pairs surface condensation picture therefore provides a simple verification of the Witten effect16 for θ = π/(2n + 1).  A B A B A B  114j11−4j, Ψ4jΨ−4j, 114j+2Ψ−4j−2, Lastly, we noticed that in the band insulator case for b = A B A B . (9) Ψ4j+211−4j−2, Σ2j+1Σ−2j−1 n = 0, in (1) reduces to the Chern insulator or the lowest LandauF level (LLL), and Pf∗ in (7) is simply the All anyons that are non-local with respect to and braid particle-hole (PH) symmetric Pfaffian state45–47. The non-trivially around any of the bosons in b are confined. PH symmetry is captured by the relative tensor product ˜ A ˜ B (8), which can be formally rewritten into This includes all anyon combinations Xja,za Xjb,zb where the dyon numbers z + n3ugj and z + n3ugj disagree a a b b Pf∗ = Pf∗ (12) modulo 2n + 1. Physically, this ensures gauge fluxes F  must continue through both A and B slabs, or equiv- by putting Pf∗ on the other side of the equation. Here, alently all gauge monopoles at the interface are confined the tensor product is relative with respect to some col- as they signify imbalances of gauge fluxes through the lection of condensed bosonic pairs, and Pf∗ is theTR ˜A ˜B ∗ ∗ two slabs. The new deconfined dyon γ = 110,1110,1 con- conjugate of Pf . Equation (12) thus equates Pf with sists of a gauge flux that passes continuously across both itsPH conjugate, which is obtained by subtracting itself slabs with gaugeQP ζa on each of the remaining top from the LLL . In the fractional case with n > 0, (12) and bottomTR breaking surfaces. A deconfined anyon suggests a generalizedPH symmetry for Pf ∗, whosePH thus splits into a dyon component γz and a surface com- conjugate is the subtraction of itself from the FQH state ponent in ( -Pf∗)A ( -Pf∗)B. Within the surface . T ⊗ T F 5

VI. CONCLUSION modulo 1. The collection el : l = 0, 1,..., 2n is of 1-1 correspondence with { γm : m = 0, 1,..., 2}n . −2 2{n−1 } To conclude, we studied gapped FTI surface states For instance γ = e = e . At the same time, (n) l with (i)TR breaking order, (ii) charge U(1) breaking Z2n+1 = e : l = 0, 1,..., 2n is the anyon content of ∗ { } order, as well as (iii) symmetry preserving -Pf topo- the Abelian Chern-Simons SU(2n + 1)1 theory with La- T 1 R I J I logical order. We focused on FTI that supported frac- grangian density 2+1 = KIJ α dα , where α L 4π 2+1 ∧ tionally charged partons coupling with a discrete Z2n+1 for I = 1,..., 2n are U(1) gauge fields, and gauge theory. We characterized the fractional interface  2 1  channels sandwiched between different gapped surface −  1 2 1  domains by describing their charge and energy response,  − −   1 2  namely the differential electric and thermal conductance. K =  −  (A1) SU(2n+1)  ..  The low-energy CFT for these fractional interface chan-  .  nels corresponded to theTO of quasi-(2+1)-D FTI slabs  2 1  with the corresponding gapped top and bottom surfaces. 1− 2 In particular, a FTI slab withTR conjugate ferromag- − is the Cartan matrix of SU(2n + 1). netic surfaces behaved like a fractional Chern insulator withTO(1), and in the particular case when a = 1 − and g = 2, its charge sector was identical to that of Appendix B: Anyon Condensation the Laughlin− ν = 1/(2n + 1) FQH state. Combining the TR symmetric -Pf∗ surface with the FTI bulk as well as the oppositeTTR breaking surface, this FTI slab ex- Here we will elaborate how to glue the two TR sym- metric surfaces of a pair of Pf∗ FTI slabs and condense hibited a generalized PfaffianTO(7). Furthermore, we ∗ ∗ surface bosonic anyon pairs on the two -Pf surfaces. demonstrated the gluing of a pair of parallel -Pf sur- T faces, which are supported by two FTI on bothT sides. As before we take the decoupled tensor product of the anyons in two Pf∗ TO, denoted (Pf∗)A (Pf∗)B where It was captured by an anyon condensation picture that ⊗ killed the -Pf∗ TO and left behind deconfined partons A, B refers to the two slabs. Then we choose a set of T bosons that braid trivially around each other. and confined gauge and magnetic monopoles in the bulk. A B 17 ∗ First notice that dyon combinations γz 1˜1 1˜1 In Ref. we also construct the -Pf state of the FTI ≡ 0,z 0,z from the field theoretic duality approach.T are not confined. A particle with charge “j” has gauge charge n2gj, so our neutral pairs have gauge charge n2gj n2gj. Thus the braiding phase with these dyons 2×− 2 ACKNOWLEDGMENTS is zn gj zn gj = 0. Our parton− should continuously move from slab A to A B slab B, so we should condense Ψ4 Ψ−4, the parton cre- J.C.Y.T. is supported by NSF Grant No. DMR- ation annihilation operator. Anything that braids with 1653535. G.Y.C. acknowledges the support from it is confined. We can derive braiding statistics with the Korea Institute for Advanced Study (KIAS) grant ribbon formula, θA,B = hA×B hA hB. The braiding funded by the Korea government (MSIP) and Grant − −˜ A ˜ B phase from the anyon combination Xja,za Xjb,zb around No. 2016R1A5A1008184 under NRF of Korea. 3 3 A B A za+n ugja B zb+n ugjb Ψ4 Ψ−4 is is the same as (δ ) (δ ) A B around Ψ4 Ψ−4. The parton carries “g” gauge charge 3 3 Appendix A: Abelian Chern-Simons theory of dyons so this phase is g(za + n ugja zb n ugjb). This is zero if the dyon number z + n−3ugj−is equal on the A and B particle. This ensures gauge fluxes must continue The fractional topological insulator slab with time- through both A and B slabs, i.e., confines gauge mag- reversal conjugate surfaces has anyons which are dyons netic monopoles. This means that we are left with com- and partons. The neutral sector consists of only dyons. A A B z A B binations Xja Xjb γ . It also identifies Ψ4 Ψ−4 with the dyon γ is composed of a number of Z2n+1 gauge charge on vacuum, which identifies each surface associated with an unit gauge flux through m 1 A 1 B z A B z 1 A 1 B z the bulk. The dyons γ where m = 0, 1,..., 2n, with 1ja 1jb γ Ψja+4Ψjb−4γ 1ja+8 1jb−8γ , 0 ≡ ≡ 1 = γ being the vacuum, form the anyon content of an 1 A B z A 1 B z 1 A B z 1ja Ψjb γ Ψja+4 1jb−4γ 1ja+8Ψjb−8γ , Abelian topological state denoted as (2a) . They have ≡ ≡ Z2n+1 ΣA ΣB γz ΣA ΣB γz, 2am2 ja jb ja+4 jb−4 m ≡ spins hγ = 2n+1 modulo 1 and satisfy the Z2n+1 fusion A B z A B z A B z 0 0 11 Σ γ Ψ Σ γ 11 Σ γ rule γm γm = γ[m+m ], where [m + m0] is the remain- ja jb ≡ ja+4 jb−4 ≡ ja+8 jb−8 × 0 A B z der between 0 and 2n when dividing m + m by 2n + 1. Ψ Σ − γ . ≡ ja+12 jb 12 For the case when a = 1, the Abelian topological the- A B − − Next we choose the fermion pair Ψ0 Ψ0 . Notice Σ ( 2) (n) × ory becomes Z2n+1, which is actually identical to Z2n+1. braids with Ψ, so anything with just one Σ is confined. 2 This is because the dyon e = γn has spin −2n n This brings the identification to 2n+1 ≡ 2n+1 6

nents

A B + − Σja±1Σjb∓1 = Sja±1,jb∓1 + Sja±1,jb∓1, (B1) A B z A B z A B z 11j 11j γ 11j +4j11j −4jγ Ψj +4jΨj −4jγ , a b ≡ a b ≡ a b where each S± carries the same spin as the original 11A ΨB γz 11A ΨB γz ΨA 11B γz, ja jb ≡ ja+4j jb−4j ≡ ja+4j jb−4j Ising pair but differs from each other by a unit fermion A B z A B z S± ΨA/B = S∓. S+ and S− normally have non-trivial Σja Σjb γ Σja+4jΣjb−4jγ . ≡ mutual× monodromy. We choose to condense the elec- + trically neutral S1,−1 and its multiples, while confining − A B S1,−1. This means Σ1 Σ−1 is condensed. The Σ pair Next we can condense ΨA11B , which when braided A B 2 −2 around 11 11 gives a phase of 2(ja jb)/16 which is A B A B ja jb around 11 11 or Ψ 11 gives 4(ja jb)/16 which is not − A B ja jb ja jb zero if ja jb = 0 mod 8. The Σ pair around 11 Ψ − A B − ja jb confined if ja jb = 0 mod 4. For Σja Σj gives 4(ja gives a phase of 2(j j )/16 + 1/2 which is zero if − b − a − b jb)/16 + 1/2 which is not confined if ja jb = 2 mod 4. j j = 4 mod 8. The Σ pair around ΣA ΣB gives − a − b ja jb The identification is now a phase of 2(ja jb)/16 1/4 which is zero if ja jb = 2 or 6 mod 8. − ± − This then completes the full condensate, and we have the identification A B z A B z A B z 11 11 γ 11 11 γ Ψ Ψ γ A B z A B z A B z ja jb ja+4j jb−4j ja+4j jb−4j 11 11 γ Ψ Ψ γ Ψ 11 γ ≡ ≡ ja jb ja,z jb,z ja+2 jb−2 A B z A B z ≡ ≡ 11 Ψ γ 11 Ψ γ A B z ± z ja+2 jb−2 ja+2+4j jb−2−4j 11 Ψ γ S γ ≡ ≡ ja+2 jb−2,z ja±1,jb∓1 A B z ≡ ≡ Ψ 11 γ , A B z ja+2+4j jb−2−4j 11 11 γ (B2) ≡ ja+4 jb−4 A B z A B z ≡ Σja Σjb γ Σja+2jΣjb−2jγ . ≡ for ja jb mod 8 and ja, jb both even. This ends up ≡ A B being just the multiples of the parton 110 Ψ4 together with the dyons γz. Together they generate the theory Our ΣΣ pairs now split into simpler Abelian compo- of a FTI slab with two conjugate TR breaking surfaces.F

∗ [email protected] 15 P. Ye, M. Cheng, and E. Fradkin, Phys. Rev. B 96, 085125 1 L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, (2017). 106803 (2007). 16 E. Witten, Physics Letters B 86, 283 (1979). 2 R. Roy, Phys. Rev. B 79, 195322 (2009). 17 G. Y. Cho, J. C. Y. Teo, and E. Fradkin, ArXiv e-prints 3 J. E. Moore and L. Balents, Phys. Rev. B 75, 121306(R) (2017), arXiv:1706.00429 [cond-mat.str-el]. (2007). 18 J. C. Y. Teo and C. L. Kane, Phys. Rev. B 82, 115120 4 X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B (2010). 78, 195424 (2008). 19 L. Fu and C. L. Kane, Phys. Rev. Lett. 102, 216403 (2009). 5 C. Wang, A. C. Potter, and T. Senthil, Phys. Rev. B 88, 20 X.-L. Qi, E. Witten, and S.-C. Zhang, Phys. Rev. B 87, 115137 (2013). 134519 (2013). 6 M. A. Metlitski, C. L. Kane, and M. P. A. Fisher, Phys. 21 F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). Rev. B 92, 125111 (2015). 22 C.-X. Liu, S.-C. Zhang, and X.-L. Qi, Annual Review of 7 X. Chen, L. Fidkowski, and A. Vishwanath, Phys. Rev. B Condensed Matter Physics 7, 301 (2016). 89, 165132 (2014). 23 N. Regnault and B. A. Bernevig, Phys. Rev. X 1, 021014 8 P. Bonderson, C. Nayak, and X.-L. Qi, Journal of Sta- (2011). tistical Mechanics: Theory and Experiment 2013, P09016 24 T. Neupert, L. Santos, C. Chamon, and C. Mudry, Phys. (2013). Rev. Lett. 106, 236804 (2011). 9 J. Maciejko, X.-L. Qi, A. Karch, and S.-C. Zhang, Phys. 25 E. Tang, J.-W. Mei, and X.-G. Wen, Phys. Rev. Lett. 106, Rev. Lett. 105, 246809 (2010). 236802 (2011). 10 B. Swingle, M. Barkeshli, J. McGreevy, and T. Senthil, 26 D. N. Sheng, Z.-C. Gu, K. Sun, and L. Sheng, Nat. Com- Phys. Rev. B 83, 195139 (2011). mun. 2, 389 (2011). 11 J. Maciejko, X.-L. Qi, A. Karch, and S.-C. Zhang, Phys. 27 M. E. Cage, K. Klitzing, A. Chang, F. Duncan, M. Hal- Rev. B 86, 235128 (2012). dane, R. Laughlin, A. Pruisken, D. Thouless, R. E. Prange, 12 P. Ye, T. L. Hughes, J. Maciejko, and E. Fradkin, Phys. and S. M. Girvin, The Quantum Hall Effect (Springer Sci- Rev. B 94, 115104 (2016). ence & Business Media, Berlin, 2012). 13 J. Maciejko and G. A. Fiete, Nature Physics 11, 385 28 R. B. Laughlin, Phys. Rev. B 23, 5632 (1981). (2015). 29 B. I. Halperin, Phys. Rev. B 25, 2185 (1982). 14 A. Stern, Annual Review of Condensed Matter Physics 7, 30 Y. Hatsugai, Phys. Rev. Lett. 71, 3697 (1993). 349 (2016). 31 H. Schulz-Baldes, J. Kellendonk, and T. Richter, Journal 7

of Physics A: Mathematical and General 33, L27 (2000). 39 P. H. Bonderson, Non-Abelian Anyons and Interferometry, 32 G. E. Volovik, JETP Letters 55, 368 (1992). Ph.D. thesis, California Institute of Technology (2007). 33 C. L. Kane and M. P. A. Fisher, Phys. Rev. B 55, 15832 40 M. Levin, Phys. Rev. X 3, 021009 (2013). (1997). 41 L. Fu and C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008). 34 A. Cappelli, M. Huerta, and G. R. Zemba, Nuclear Physics 42 F. A. Bais and J. K. Slingerland, Phys. Rev. B 79, 045316 B 636, 568 (2002). (2009). 35 A. Kitaev, Annals of Physics 321, 2 (2006), january Spe- 43 L. Kong, Nucl. Phys. B 886, 436 (2014). cial Issue. 44 T. Neupert, H. He, C. von Keyserlingk, G. Sierra, and 36 J. M. Luttinger, Phys. Rev. 135, A1505 (1964). B. A. Bernevig, Phys. Rev. B 93, 115103 (2016). 37 P. Di Francesco, P. Mathieu, and D. Senechal, Conformal 45 D. T. Son, Phys. Rev. X 5, 031027 (2015). Field Theory (Springer, New York, 1999). 46 M. Barkeshli, M. Mulligan, and M. P. A. Fisher, Phys. 38 G. Moore and N. Seiberg, Comm. Math. Phys. 123, 177 Rev. B 92, 165125 (2015). (1989). 47 C. Wang and T. Senthil, Phys. Rev. B 93, 085110 (2016).