Quantum Algorithms: a survey Umesh V. Vazirani U. C. Berkeley Exponential Superposition + - + + - - + - Superposition of all 2n classical states: a x x x 2 | a | x 1 x Measurement: See |xi with probability |ax|2 Quantum Algorithms: tension between these two phenomena Why limit computers to electronics implemented on silicon? • Extended Church-Turing thesis - Any “reasonable” model of computation can be efficiently simulated by a probabilistic Turing Machine. - Circuits, Random access machines, cellular automata. - “Reasonable” = physically realizable in principle • Quantum computers only model that violate this thesis Quantum Algorithms – Exponential Speedups • Shor’s quantum factoring algorithm. Breaks modern cryptography. • Simulating quantum mechanics. • Symmetry - Discrete logarithm - Pell’s equation - Shifted Legendre Symbol - Gauss sums - Elliptic curve cryptography Young’s Double slit experiment P1(x) = |F1(x)|2 P2(x) = |F2(x)|2 F1,2 = F1(x) + F2(x) P1,2(x) = |F1,2(x)|2 n-slits • Etch n slits in a pattern based on the input. • Send photon through and measure. • Location at which photon detected gives provides information about solution to input. photon Input-based slit pattern screen Quantum Circuits UF F - Each Wire Carries a qubit of information - Controlled Not: ..\ a a b a©b - Single Bit Gates (Rotations) |1i |0’i q |0i |0i ! cosq |0i + sinq |1i |1i ! sinq |0i - cosq |1i |1’i Quantum Fourier Transform 1 1 b 0 1 2 b 1 1 . . . . . . . . b m 1 2 ( m 1) 1 m 1 . 1 a 0 m 1 . a1 . . . . . . . ( m 1)( m 1) a m 1 Classical: FFT O(m logm) Quantum: Input: |Fi = j=0m aj |ji O(logm) qubits Fourier transform: F|Fi = j=0m-1 bj |ji O(log2m) gates Limited Access: Measure: see |ji with probability |bj|2 Quantum Fourier Transform |1i One Qubit or Z2 |0’i H 1 2 |0i 1 2 |0i ! |0i + |1i |1i ! 1 |0i 2 - 1 |1i 2 |1’i Two Qubits or Z22 H H |00i ! |01i ! |10i ! |11i ! 1 (|0i + |1i) 2 1 (|0i + |1i) 2 1 (|0i - |1i) 2 1 (|0i - |1i) 2 1 (|0i 2+ |1i) = 1/2(|00i + |01i + |10i + |11i) 1 (|0i 2- |1i) = 1/2(|00i - |01i + |10i - |11i) 1 (|0i 2+ |1i) = 1/2(|00i + |01i - |10i - |11i) 1 (|0i 2- |1i) = 1/2(|00i - |01i - |10i + |11i) n Qubits or Z2n |ui H H H x1 H xn ux (1) | u | x n/2 x{0 ,1}n 2 Outline of Shor’s Factoring Algorithm • N ! exponential superposition x ax |xi • Factors of N encoded in global property of superposition – its period. • quantum fourier transform and measure to extract period. • Reconstruct factors of N from the period. 0 q 0 q Outline of Shor’s Factoring Algorithm (Example) • N = 15 = p¢ q • Randomly choose a = 7 (mod 15) • Consider sequence ax (mod 15) for x =0,1,2,… 1, 7, 4, 13, 1, 7, 4, 13, 1, 7, 4, 13, … • Period r = 4. • N | (ar-1)=(ar/2+1)(ar/2 -1) = (72 +1)(72 -1) = 50¢ 48 • p = gcd(15, 50), q = gcd(15, 48). • Create superposition x |xi |axi 0 q 0 q Outline of Shor’s Factoring Algorithm (Example) • N = 15 = p¢ q • Randomly choose a = 7 (mod 15) • Consider sequence ax (mod 15) for x =0,1,2,… 1, 7, 4, 13, 1, 7, 4, 13, 1, 7, 4, 13, … • Period r = 4. N | (r+1)(r-1) = (4+1)(4-1) = 5¢ 3 • p = gcd(15, 5), q = gcd(15, 3). • Create superposition x |xi |axi 0 q 0 q The Hidden Subgroup Problem Given f : G ! S, constant and distinct on cosets of subgroup H. Find H. G: Examples • Factoring N: G = • Discrete log: G = 1. State Preparation Given f : G ! S, constant and distinct on cosets of subgroup H. Find H. 1. Create Random Coset state h2 H |g + hi: FG f G: Measure 2. Fourier Sampling G: G: F.T. Measure a random element of H?. 3) (Classically) reconstruct H from polynomialy many samples. Example: Factoring To factorize N = P¢Q, sufficient to compute order of randomly chosen x mod N. i.e. smallest positive r: xr = 1 mod N. Let f: a ! xa mod N. Underlying group = ZM, where M = F(N) = (P-1)¢(Q1) Hidden subgroup = H = h r i = {0,r, 2r, …, M/r} H? = h M/ri = {0,M/r, 2M/r, …, M} Fourier sampling gives kM/r for random k: 0 · k · r-1 gcd(M, kM/r) = M/r if k,r relatively prime. Pell’s Equation • Given a positive non-square integer d, find integer solutions x,y of x2 – d¢ y2 = 1. • One of the oldest studied problem in algorithmic number theory. • Appears harder than factoring • [1989] Buchman-Williams cryptosystem • [Hallgren 2002]: 1) Quantum algorithm for Pell’s Equation 2) Breaks Buchman-Williams cyptosystem Abelian Hidden subgroup problem – but the group is not finitely generated Two Challenges Short vector in Lattice: • Basis v1, …, vn vectors in Rn. • The lattice is a1 v1 + … + an vn for all integers a1, … an. • Find shortest vector in lattice. Two Challenges 1. Short vector in Lattice: Finding short vector not easy! Regev: DN Dihedral group 2. Graph Isomorphism SN Symmetric group Ajtai-Dwork Cryptosystem. [GSVV] For random choice of basis, for sufficiently non-abelian groups (e.g. S_n), exponentially many samples necessary to distinguish |H|=2 from |H| =1. Dihedral HSP • Dihedral Group DN : Group of symmetries of a regular N-gon. • Generated by x, y: xN = 1, y2=1, xyxy = 1. • Assume N = 2n. • DN has 4 1-d irreps and (N-1)/2 2-d irreps. jl 0 l j ( x y) jl 0 jl 0 l j ( x ) jl 0 [Kuperberg ’03] 2O( n) algorithm for dihedral HSP. Dihedral HSP algorithm • Assume wlog H = {1, xhy} • Set up random coset state, fourier transform and sample irrep to get random j, jh j ( h k ) j ( h k ) jh • Can sample column superposition and do phase jh 2 estimation to get coin flip of bias cos N • O(log N) samples sufficient to determine h, but reconstruction problem hard. • Would like to sample particular irreps j. Dihedral HSP algorithm Claim: i j = i+j © i-j Algorithm: • Start with 22 n registers in random coset states, FT, sample irreps. • Sort irrep names, pair up successive registers • Apply above transformation, and retain iff i,j ! i-j • Number of bits reduced by 2 n per iteration. n iterations 2 • Number of irreps reduced by 4 per iteration. Non-abelian Hidden Subgroup Problem • Abelian quantum algorithm doesn’t generalize - Prepare random coset state - Measure in Fourier basis • Ettinger, Hoyer, Knill ’98: - Prepare several registers with random coset states - Perform appropriate joint measurement • Ip ’03: - Fourier transform & Measure irrep (character) for each register - Perform appropriate joint measurement on residual state. Adiabatic Quantum State Generation Aharonov, Ta-Shma ‘02 AharonovvanDamKempeLandauLloydRegev’03 Adiabatic Computation ≈ Quantum Computation Classical Simulation of Quantum Systems Vidal ’03 Polynomial time simulation of one dimensional spin chains with O(log n) entanglement length. AC = A C A 1 ABC = AB1 B2C B 2 C Summary • Quantum computation only model that violates extended Church-Turing thesis. • Exponential superposition vs limited access. • Exponential speedups appear to require symmetry. • Fast quantum algorithm for abelian hidden subgroup problem • Non-abelian case open.