Topological Quantum Computation II Eric Rowell October 2015 QuantumFest 2015 Topological Quantum Computation Math Physics TQC Computer Science Everything Else Fundamental Questions I Distinguish: indecomposable anyons I Classify: (models) by number of colors. I Detect: non-abelian anyons I Detect: universal anyons I 3-dimensional generalizations? Wilson Loops Define Sab for a, b ∈ L: C a b C I Sa,b ∈ C I det(S) 6= 0 I columns ⊥, distinguishes anyon types. Fusion rules The dimension N(a, b, c) of H(P; a, b, c): If b a = a, Na symmetric a b Example (Fibonacci) N0 = c 1 0 0 1 0 1 N1 = 1 1 provides fusion matrices: (Na )c,b := N(a, b, c) for particles a ∈ L Classify/Constraints Question How many fusion rules with r = |L|? Use constraints, for example: d d = m a b c I H(P; a, b, c) = CN(a,b,c) I Braiding: N(a, b, c) = N(b, a, c) I compute dimensions: gluing+ disjoint union I Na N c = Nc Na m a b c Rank-Finiteness Theorem (Bruillard, Ng, R, Wang: J. Amer. Math. Soc.) There are finitely many models with fixed r = |L|. Proof is by algebra and number theory. Complete classification known up to |L| = 5. |L| Models 1 Vec 2 Fib, semion 3 Z3 , PSU(2)7 , Ising 4 products, Z4 , PSU(2)9 5 Z5 , PSU(2)11 , SU(3)4 /Z3 , SU(2)4 Topological Spin Each anyon has a topological spin, which may distinguish them θa a a where θa = e 2πiha with ha ∈ Q. Bosonic: ha = 0; Fermionic: ha = 1/2; anyonic: any ha . Related to Dehn twist on torus. Examples Example Example Fibonacci: L = {1, f } f ×f =1+f. S= 1√ 1+ 5 2 and θf = e 4πi/5 . √ ! 1+ 5 2 −1 Ising: L = {1, σ, ψ} σ × σ = 1 + ψ, σ × ψ = σ, ψ × ψ = 1. √ 2 1 √1 √ S = 2 0 − 2 √ 1 − 2 1 and θσ = e πi/8 , θψ = −1. Algebraic Constraints Set Tij = δij θi , define dj := S0j , D 2 := (S, T ) satisfy P j dj2 , p± := P j dj2 θj±1 . t 1. S = S t , SS = D 2 Id, T diagonal, ord(T ) = N < ∞ 2N 2. (ST )3 = p+ S 2 , p+ p− = D 2 , pp−+ =1 P S S S 3. Nijk := a iaD 2jada ka ∈ N P 4. θi θj Sij = a Nik∗ j dk θk where Nii0∗ uniquely defines i ∗ . n P 5. νn (k) := D12 i,j Nijk di dj θθji satisfies: ν2 (k) ∈ {0, ±1} 6. Q(S) ⊂ Q(T ), AutQ (Q(S)) ⊂ Sr , AutQ(S) (Q(T )) ∼ = (Z2 )k . 7. Prime (ideal) divisors of hD 2 i and hNi coincide in Z[ζN ]. Braid group representations Bn acts on state spaces: I I Fix anyons a, b Braid group acts linearly: Bn y H(D 2 \ {zi }; a, · · · , a, b) by particle exchange i i+1 i Non-abelian Anyons Definition a ∈ L is a non-abelian anyon if particle exchange on H(D 2 ; a, a, a, i) (for some i) is a non-abelian group. a a a i Quantum Dimensions Definition Let dim(a) be the maximal eigenvalue of Na . Alternatively, dim(a) = S0,a . Fact 1. dim(a) ∈ R 2. dim(a) ≥ 1 3. dim(a) dim(b) = P c N(a, b, c) dim(c) 4. dim(a) > 1 implies Degeneracy: dim H(D 2 ; a, a, a, i) > 1 (for some i). Degeneracy implies Non-Abelian Statistics If dim(X ) > 1 there is a Y 6= 1 with N(X , X , Y ) 6= 0. Y IF σ1 σ2 σ1−1 σ2−1 = Id then X Y Y X X Y 6= 0 = α = γ X 1 = 0 1 X 1 X X X Universal Anyons Question When does an anyon a provide universal computation models? This means: simulate QCM. Mathematically: when does particle exchange on H(D 2 , a, . . . , a, i) simulate a universal gate set? Example √ Fibonacci dim(a) = 1+2 5 is universal: braid group Bn image is dense in SU(Fn ) × SU(Fn−1 ) Example √ Ising dim(a) = 2 is not universal: braid group Bn image is a finite group. Property F conjecture 2-dimensional B3 Fibonacci: " −4iπ/5 e σ1 7→ 0 " 3iπ/5 e σ2 7→ −e 3iπ/5 rep from e −4iπ/5 # e 3iπ/5 0 , # e −4iπ/5 Ising: localized with 1 0 0 0 1 1 R = √12 0 −1 1 −1 0 0 related to Bell states. 1 0 0 1 Conjecture Anyon a is universal if, and only if, dim(a)2 6∈ Z. Remark We expect: Universal anyons have “hard” classical computational complexity, whereas non-universal anyons have “easy” classical computational complexity. (assuming P 6= NP...) 3-dimensional materials I Point-like particles in R3 I Point-like particles in R3 loop-like particles? … Two operations: Loop interchange si : ↔ and Leapfrogging (read upwards): σi : ··· σi = 1 ··· i ··· si = 1 n i +1 ··· i i +1 n The Loop Braid Group LB n is generated by s1 , . . . , sn−1 , σ1 , . . . , σn−1 satisfying: Braid relations: (R1) σi σi+1 σi = σi+1 σi σi+1 (R2) σi σj = σj σi if |i − j| > 1 Symmetric Group relations: (S1) si si+1 si = si+1 si si+1 (S2) si sj = sj si if |i − j| > 1 (S3) si2 = 1 Mixed relations: (M1) σi σi+1 si = si+1 σi σi+1 (M2) si si+1 σi = σi+1 si si+1 (M3) σi sj = sj σi if |i − j| > 1 PRL 113, 080403 (2014) PHYSICAL REVIEW LETTERS week ending 22 AUGUST 2014 Braiding Statistics of Loop Excitations in Three Dimensions Chenjie Wang and Michael Levin James Franck Institute and Department of Physics, University of Chicago, Chicago, Illinois 60637, USA (Received 31 March 2014; published 19 August 2014) While it is well known that three dimensional quantum many-body systems can support nontrivial braiding statistics between particlelike and looplike excitations, or between two looplike excitations, we argue that a more fundamental quantity is the statistical phase associated with braiding one loop α around another loop β, while both are linked to a third loop γ. We study this three-loop braiding in the context of ðZN ÞK gauge theories which are obtained by gauging a gapped, short-range entangled lattice boson model with ðZN ÞK symmetry. We find that different short-range entangled bosonic states with the same ðZN ÞK symmetry (i.e., different symmetry-protected topological phases) can be distinguished by their three-loop braiding statistics. DOI: 10.1103/PhysRevLett.113.080403 Introduction.—A powerful way to characterize the topological properties of two dimensional gapped quantum many-body systems is to examine their quasiparticle braiding statistics [1]. Thus, it is natural to wonder: what is the analogous quantity that characterizes three dimensional (3D) systems? The simplest candidate—3D quasiparticle statistics—is of limited use since 3D systems can only support bosonic and fermionic quasiparticles. On the other hand, 3D systems can support much richer braiding statistics between particlelike excitations and looplike excitations [2–4] or between two looplike excitations [5–7]. Thus, one might guess that particle-loop and loop-loop braiding statistics are the natural generalizations of quasiparticle statistics to three dimensions. In this Letter, we argue that this guess is incorrect: particle-loop and loop-loop braiding statistics do not fully capture the topological structure of 3D many-body systems. Instead, more complete information can be obtained by considering a three-loop braiding process in which a loop α is braided around another loop β, while both are linked with a third loop γ (Fig. 1). We believe that three-loop braiding statistics is one of the basic pieces of topological data that describe 3D gapped many-body systems, and much of this Letter is devoted to understanding the general properties of this quantity. Also, as an application, we show that threeloop statistics can be used to distinguish different shortrange entangled many-body states with the same (unitary) symmetry—i.e., different symmetry-protected topological (SPT) phases [8–10]. The latter result shows that the braiding statistics approach to SPT phases, outlined in Ref. [11], can be extended to three dimensions. Discrete gauge theories.—For concreteness, we focus our analysis on a simple 3D system with looplike excitations, namely lattice ðZN ÞK gauge theory [12]. More specifically, we consider a 3D lattice boson model built out of K different species of bosons, where the number of PACS numbers: 05.30.Pr, 03.75.Lm, 11.15.Ha system has a ðZN ÞK symmetry. We suppose that the ground state of the boson model is gapped and short-range entangled—that is, it can be transformed into a product state by a local unitary transformation [13]. We then imagine coupling such a lattice boson model to a ðZN ÞK lattice gauge field [14]. In general, these gauge theories contain two types of excitations: pointlike “charge” excitations which carry gauge charge, and stringlike “vortex loop” excitations which carry gauge flux. The most general charge excitations can carry gauge charge q ¼ ðq1 ; …; qK Þ where each component qm is an integer defined modulo N. The most general vortex loop can carry gauge flux ϕ ¼ ðϕ1 ; …; ϕK Þ where ϕm is a multiple of 2π=N. In fact, since we can always attach a charge to a vortex loop to obtain another vortex loop, a general vortex loop excitation carries both flux and charge. Let us try to understand the braiding statistics of these excitations. In general, there are three types of braiding processes we can consider: processes involving two charges, processes involving a charge and a loop, and processes involving multiple loops. Clearly, the first type of process cannot give any statistical phase since the charges are excitations of the short-range entangled boson model and, therefore, must be bosons. On the other hand, the FIG. 1. (a) Three-loop braiding process. The gray curves show the paths of two points on the moving loop α. (b) A top view of the braiding process within the plane that γ lies in. (c) A torus Ωα is swept out by α during the braiding. Loop β (dashed circle) is Question Do such materials exist in nature? Maybe... Thank you! References I Chang,R.,Plavnik,Sun,Bruillard,Hong: 1508.00005 (J. Math. Phys.) I Kadar,Martin,R.,Wang: 1411.3768 (Glasgow J. Math.) I R.,Wang: 1508.04793 (preprint) I Bruillard,Ng,R.,Wang: 1310.7050 (J. Amer. Math. Soc.) I Naidu,R.: 0903.4157 (J. Alg. Rep. Theory)