Exponential improvement in precision for simulating sparse Hamiltonians Andrew Childs Department of Combinatorics & Optimization and Institute for Quantum Computing University of Waterloo based on joint work with Dominic Berry, Richard Cleve, Robin Kothari, and Rolando Somma arXiv:1312.1414 To appear in STOC 2014 “... nature isn’t classical, dammit, and if you want to make a simulation of nature, you’d better make it quantum mechanical, and by golly it’s a wonderful problem, because it doesn’t look so easy.” ! Richard Feynman Simulating physics with computers (1981) Why simulate quantum mechanics? Computational chemistry/physics • chemical reactions • properties of materials Implementing quantum algorithms • continuous-time quantum walk • adiabatic quantum computation • linear equations Quantum dynamics The dynamics of a quantum system are determined by its Hamiltonian. d i (t) = H (t) dt| i | i Quantum simulation problem: Given a description of the Hamiltonian H, an evolution time t, and an unknown initial state (0) , produce the final state ( t ) (approximately) | i | i A classical computer cannot even represent the state efficiently By performing measurements on the final state, a quantum computer can efficiently answer questions that (apparently) a classical computer cannot Quantum simulation Quantum simulation problem: Given a description of the Hamiltonian H, an evolution time t, and an unknown initial state (0) , produce the final state ( t ) (approximately) | i | i Equivalently, apply the unitary operator U(t) satisfying d i U(t)=H(t) U(t) dt iHt If H is independent of t, U(t)=e− More generally, H can be time-dependent Simulation should approximate U(t) to within error ² (say, with respect to the diamond norm) Local and sparse Hamiltonians Local Hamiltonians [Lloyd 96] m H = j=1 Hj where each H j acts on k = O(1) qubits Sparse HamiltoniansP [Aharonov, Ta-Shma 03] At most d nonzero entries per row, d = poly(log N) (where H is N N) £ In any given row, the H = location of the jth nonzero entry and its value can be computed efficiently (or is given by a black box) k Note: A k-local Hamiltonian with m terms is d-sparse with d = 2 m Previous simulation methods Product formulas • Decompose Hamiltonian into a sum of terms that are easy to simulate • Recombine the terms by alternating between them iAt/r iBt/r r i(A+B)t 2 e− e− = e− + O(t /r) iAt/2r iBt/r iAt/2r r i(A+B)t 3 2 e− e− e− = e− + O(t /r ) ... Quantum walk • Define an easy-to-implement unitary operation (a step of a quantum walk) whose spectrum is related to the Hamiltonian • Use phase estimation to obtain information about the spectrum • Introduce phases to give the desired evolution Complexity of previous simulation methods Parameters: dimension N sparsity d evolution time t allowed error ² [Lloyd 96]: poly(log N)( H t)2/✏ (for local Hamiltonians only) k k [Aharonov, Ta-Shma 02]: poly(d, log N)( H t)3/2/p✏ k k 4 4 1+δ δ [Childs 04]: O (d log N H t) /✏ (for any ± > 0) k k 4 1+δ δ [Berry, Ahokas, Cleve, Sanders 07]: O (d log⇤ N H t) /✏ k k 3 1+δ δ [Childs, Kothari 11]: O (d log⇤ N H t) /✏ k k [Childs10; Berry, Childs 12]: O(d H t/p✏) k kmax New result: O ⌧ log(⌧/✏) ⌧ := d2 H t log log(⌧/✏) k kmax Fractional-query simulation New approach: use tools for simulating the fractional-query model Two steps: • Reduce Hamiltonian simulation to fractional-query simulation • Improved algorithm for fractional-query simulation High-level idea of fractional-query simulation: • Decompose the evolution into terms that can be implemented in superposition • “Compress” the implementation • Unit-time evolution only succeeds with constant probability; boost this using oblivious amplitude amplification Strictly improves all methods based on product formulas Dependence on ² is exponentially improved! In fact, the improved dependence is optimal. Outline • Fractional-query model - Simulating fractional queries - Oblivious amplitude amplification • Reducing Hamiltonian simulation to fractional-query simulation • Features of the algorithm - Gate complexity - Local Hamiltonians - Time-dependent Hamiltonians • Optimality with respect to error • Comparison of simulation methods • Open questions Fractional- and continuous-query models Black box hides a string x 0, 1 n 2 { } Quantum query: Q i, b =( 1)bxi i, b | i − | i 00 ||||00ii ||||00ii ||||00ii QQ QQ ||||00ii ||||00ii UU00 UU11 ||||00ii ||||00ii |||| ii 00 ||||00ii ||||00ii 1/21/2 1/21/2 1/21/2 1/21/2 ||||00ii QQ QQ QQ QQ ||||00ii ||||00ii UU00 UU11 UU22 UU33 ||||00ii ||||00ii |||| ii 00 |||| ii 00 1/41/4 1/41/4 1/41/4 1/41/4 1/41/4 1/41/4 1/41/4 1/41/4 ||||00ii ||||00ii QQ QQ QQ QQ QQ QQ QQ QQ ||||00ii UU00 UU11 UU22 UU33 UU44 UU55 UU66 UU77 ||||00ii ||||00ii ||||00ii |||| ii 00 ||||00ii ||||00ii ||||00ii ||||00ii ii⇡⇡((HHQQ++HHDD))tt ||||00ii ||||00ii ee−− ||||00ii |||| ii Useful for designing algorithms [Farhi, Goldstone, Gutmann 07] More powerful than the discrete-query model? log t No: Can simulate a t-query fractional-query algorithm with O(t log log t ) discrete queries [Cleve, Gottesman, Mosca, Somma, Yonge-Mallo 09] Simulating fractional queries Fractional-query gadget: 0 R↵ P R↵ 0 | i • . Q . Q↵ | i . | i ⇡↵ 1 pc ps c = cos 2 10 R↵ = P = pc + s ps pc s =sin⇡↵ 0 i ✓ − ◆ 2 ✓ ◆ ↵m ↵1 “Segment” implementing U m Q U m 1 U 1 Q U 0 : − ··· 0 R↵ R↵ P | i 1 • ··· 1 . .. 0 R↵ R↵ P | i m ··· • m ··· . U0 Q U1 .. Um 1 Q Um | i − ··· Behavior of a segment ↵m ↵1 “Segment” implementing U m Q U m 1 U 1 Q U 0 : − ··· 0 R↵ R↵ P | i 1 • ··· 1 . .. 0 R↵ R↵ P | i m ··· • m ··· . U0 Q U1 .. Um 1 Q Um | i − ··· log(1/✏) Truncating the ancillas to Hamming weight k = O ( log log(1 / ✏ ) ) introduces error at most ² By rearranging the circuit, k queries suffice But this still only succeeds with constant probability Correcting faults success! ¾ ! ¾ ! ¾ ! segment 1! segment 2! …! segment t fault! ¼! ¾ ! ¼! ¾ ! t log(t/✏) undo! undo! Query complexity: O ✏ log log(t/✏) ¼! ¾ ! ¼! ¾ ! undo! undo! ¼! ¾ ! ¼! ¾ ! undo! undo! ¼! ¼! Rube Goldberg, Professor Butts and the Self-Operating Napkin [Cleve, Gottesman, Mosca, Somma, Yonge-Mallo 09] Oblivious amplitude amplification Suppose U implements V with amplitude sin µ: U 0 =sin✓ 0 V + cos ✓ 1 φ | i| i | i | i | i| i segment (without 1 ideal evolution final measurement) 2 To perform V with amplitude close to 1: use amplitude amplification? But the input state is unknown! Using ideas from [Marriott, Watrous 05], we can show that a - independent reflection suffices to do effective amplitude amplification.| i With this oblivious amplitude amplification, we can perform the ideal evolution exactly with only three segments (one backward). Hamiltonian simulation using fractional queries We reduce Hamiltonian simulation to fractional-query simulation. Suppose H = H1 + H2 where H1, H2 have eigenvalues 0 and ¼. i(H +H )t iH t/r iH t/r r Write e − 1 2 ( e − 1 e − 2 ) for very large r (increasing r does not affect the⇡ query complexity, and only weakly affects the gate complexity). iH iH This is a fractional-query algorithm with oracles e − 1 and e − 2 . iH iH Package them as a single oracle Q = 1 1 e − 1 + 2 2 e − 2 . | ih | ⌦ | ih | ⌦ (may not be diagonal in the standard basis, but the fractional-query simulation doesn’t require that) Decomposing sparse Hamiltonians To give a complete simulation, decompose the d-sparse Hamiltonian into a sum of terms, each with eigenvalues 0 and ¼ (up to an overall shift and rescaling). d2 • Edge coloring: H = j =1 H j where each Hj is 1-sparse new trick: H is bipartite wlog since it suffices to simulate H σ P x d2-coloring: color(`, r)=(idx(`, r), idx(r, `)) ⌦ • Approximately decompose into terms with all nonzero entries equal Ex: 010000 010000 000000 000000 100000 100000 000000 000000 00002001 00001001 00001001 00000001 = + + B002000C B001000C B001000C B000000C B C B C B C B C B000003C B000001C B000001C B000001C B C B C B C B C B000030C B000010C B000010C B000010C B C B C B C B C • Remove@ zero blocksA so@ that all termsA have@ two fixedA eigenvalues@ A Ex: 0000 1000 1000 0000 1 0100 1 −0 100 0 1 = 0 1 + 0 − 1 0001 2 0001 2 0001 B0010C B0010C B 0010C B C B C B C @ A @ A @ A Gate complexity log(⌧/✏) Query complexity of this approach: O ⌧ log log(⌧/✏) where ⌧ := d2 H t k kmax log(⌧/✏) Gate complexity is not much larger: O ⌧ log log(⌧/✏) (log(⌧/✏)+n) where H acts on n qubits Contributions to this gate complexity come from • Preparing the ancilla state • Performing simple operations between fractional queries • Implementing the fractional-query oracle using the sparse Hamiltonian oracle Using oblivious amplitude amplification instead of recursive fault correction considerably simplifies the analysis. Local Hamiltonians k Recall: A k-local Hamiltonian with m terms is d-sparse with d = 2 m Directly applying our main result gives gate complexity 2 O ⌧ log (⌧/✏) n ⌧ := d2 H t =4km2 H t log log(⌧/✏) k kmax k kmax Ex: Generic 2-local Hamiltonian acting for constant time k =2,m= n ,t, H = O(1) 2 k kmax gives complexity O˜ ( n 5 ) (cf. high-order product formulas: O ˜ ( n 4 ) ) But we can do better since we have an explicit decomposition into m k-local (and hence 2k-sparse) terms. Resulting gate complexity: 2 O ⌧˜ log (˜⌧/✏) n ⌧˜ := 2km H t log log(˜⌧/✏) k kmax Ex: Generic 2-local Hamiltonian acting for constant time: O˜(n3) Time-dependent Hamiltonians The query complexity of this approach depends only on the evolution time, not on the number of fractional-query steps ⇒ Query complexity of simulating sparse H(t) is independent of d H ( t ) (provided this is bounded) k dt k (cf.