On complexity of the quantum Ising model Sergey Bravyi (IBM) Matthew Hastings (Microsoft Research) Presenter: David Gosset Based on QIP arXiv: 1410.0703 Sydney arXiv: 1402.2295 January 16, 2015
Motivation Quantum annealing with >100 qubits Quantum Hamiltonian Complexity Boixo et al, Nature Phys. 10, 218 (2014) Cubitt & Montanaro, arxiv:1311.3161 QMA TIM Attempts to solve hard optimization NP problems such as QUBO P Basic model of phase transitions Onsager (1944) Understand computational hardness of estimating the ground state energy for quantum spin Hamiltonians
Transverse Ising Model ( TIM ) Qubits live at vertices of a graph β’ Ising ππ interactions between β’ nearest neighbor qubits. Local magnetic fields along β’ π and π axes. πΌ = π π£ π π£ + β π£ π π£ β πΎ π£,π€ π π£ π π€ π£ (π£,π€) π π£ = 1 0 π π£ = 0 1 0 β1 1 0
Part I Universality of TIM for quantum annealing Part II Computational hardness of estimating the ground state energy of TIM Part III Ferromagnetic TIM is easy
Quantum Annealing (Farhi et al 2001) π π| Ξ¨(π’) = πΌ(π’/π)| Ξ¨(π’) ππ’ Hard Easy Unitary evolution 0 β€ π’ β€ π Easy: πΌ 0 = β π£ π π£ Hard: πΌ 1 = (π£,π€) πΎ π£,π€ π π£ π π€ + π£ π π£ π π£ πΌ(π‘) interpolates between πΌ 0 and πΌ(1)
Given an adiabatic path πΌ π‘ , 0 β€ π‘ β€ 1, how large the evolution time π should be ? Adiabatic Theorem πΌ β₯ 2 π~ β₯ πΌ β₯ + β₯ + β₯ πΌ β₯ π 2 π 3 π 2 Here π is the minimum spectral gap above the ground state of πΌ π‘ , 0 β€ π‘ β€ 1. Jansen, Seiler, Ruskai, JMP 48, 102111 (2007) We need a smooth path with a non-negligible spectral gap
Big open question: what kind of problems can be efficiently solved by the quantum annealing (QA) ?
Big open question: what kind of problems can be efficiently solved by the quantum annealing (QA) ? Simpler question: can one QA machine efficiently simulate another QA machine ? target QA machine simulator QA machine πβ² π
Big open question: what kind of problems can be efficiently solved by the quantum annealing (QA) ? Simpler question: can one QA machine efficiently simulate another QA machine ? target QA machine simulator QA machine πβ² π TIM Hamiltonians
Big open question: what kind of problems can be efficiently solved by the quantum annealing (QA) ? Simpler question: can one QA machine efficiently simulate another QA machine ? target QA machine simulator QA machine πβ² π TIM Hamiltonians Some fixed target class of Hamiltonians
What does efficient simulation mean ? Target Simulator πΌ β² π‘ , 0 β€ π‘ β€ 1 πΌ π‘ , 0 β€ π‘ β€ 1 Adiabatic path π β² β€ ππππ§(π) π Number of qubits π β² β₯ π π Minimum spectral gap πΎ β² β€ ππππ§(π, π β1 , πΎ) Maximum πΎ interaction strength All spins | All spins | + + Ground state at π‘ = 0 Ground state at π‘ = 1 | β π| π π Here π: π 2 β¨π β π 2 β¨πβ² is a sufficiently simple encoding
When efficient simulation is unlikely: target QA machine simulator QA machine πβ² π TIM Hamiltonians 2-local Hamiltonians
When efficient simulation is unlikely: target QA machine simulator QA machine πβ² π TIM Hamiltonians 2-local Hamiltonians BQP Aharonov et al (2007) Oliveira and Terhal (2008)
When efficient simulation is unlikely: target QA machine simulator QA machine πβ² π TIM Hamiltonians 2-local Hamiltonians BQP BQPβpostBPP SB, DiVincenzo, Oliveira, Aharonov et al (2007) Terhal (2007) Oliveira and Terhal (2008)
When efficient simulation is unlikely: target QA machine simulator QA machine πβ² π TIM Hamiltonians 2-local Hamiltonians More Powerful β BQP BQPβpostBPP SB, DiVincenzo, Oliveira, Aharonov et al (2007) Terhal (2007) Oliveira and Terhal (2008)
Stoquastic k-local Hamiltonians System of π qubits with a Hamiltonian πΌ = πΌ π½ π½ Each term πΌ π½ acts on at most π = π(1) qubits 1. Matrix elements of πΌ π½ in the standard basis are real. 2. Off-diagonal matrix elements of πΌ π½ are non-positive: π¦|πΌ π½ |π§ β€ 0 for all π¦ β π§ β 0,1 π
Building blocks for 2-local stoquastic Hamiltonians: Diagonal : Β±π π£ , Β±π π£ π π€ Transverse field: βπ π£ 0 0 , Elementary βπ β βπ β 1 1 interactions: β π β π β π β π, βπ β π + π β π
Result 1: universality of TIM for quantum annealing with 2-local stoquastic Hamiltonians simulator QA machine target QA machine = πβ² π TIM Hamiltonians Stoquastic 2-local Hamiltonians
Result 1: universality of TIM for quantum annealing with 2-local stoquastic Hamiltonians simulator QA machine target QA machine = πβ² π TIM Hamiltonians Stoquastic 2-local Hamiltonians with k-local diagonal terms
Result 1: universality of TIM for quantum annealing with 2-local stoquastic Hamiltonians simulator QA machine target QA machine = πβ² π TIM Hamiltonians Stoquastic 2-local Hamiltonians on degree-3 graphs with k-local diagonal terms
Part II Computational hardness of estimating the ground state energy of TIM
Ground state energy: πΉ 0 = min π πΌ π Local Hamiltonian Problem (LHP): Input: (π, πΌ = π½ πΌ π½ , π· π§ππ‘ < π· ππ ) Yes-instance: πΉ 0 β€ π· π§ππ‘ No-instance: πΉ 0 β₯ π· ππ Decide which one is the case. Promise: πΉ 0 β π· π§ππ‘ , π· ππ Normalization: πΌ π½ β€ ππππ§ π , π· ππ β π· π§ππ‘ β₯ ππππ§ 1/π #terms β€ ππππ§(π)
Merlin-Arthur games (Babai 1985) I instance of yes/no problem P Merlin Arthur proof Polynomial-time Unlimited classical computer computational power reject accept
A problem belongs to this class if β¦ complexity class yes-instance : Arthur accepts some Merlinβs proof NP no-instance: Arthur rejects any Merlinβs proof
A problem belongs to this class if β¦ complexity class yes-instance : Arthur accepts some Merlinβs proof NP no-instance: Arthur rejects any Merlinβs proof Arthur is a quantum computer. Merlinβs proof can be a quantum state. QMA yes-instance: Arthur accepts some Merlinβs proof with high probability no-instance: Arthur rejects any Merlinβs proof with high probability
A problem belongs to this class if β¦ complexity class yes-instance : Arthur accepts some Merlinβs proof NP no-instance: Arthur rejects any Merlinβs proof Arthur is a quantum computer. Merlinβs proof can be a quantum state. QMA yes-instance: Arthur accepts some Merlinβs proof with high probability no-instance: Arthur rejects any Merlinβs proof with high probability Same as QMA but Arthur can apply only reversible StoqMA classical gates (CNOT, TOFFOLI) and measure some fixed output qubit in the X-basis. Arthur accepts the proof if the measurement outcome is β +β² . Arthur can use | 0 and | + ancillas. SB, Bessen, Terhal, arXiv:0611021
Ξ 2 PostBPP A 0 PP AM QMA SBP StoqMA MA - randomized analogue of MA NP NP AM=MA + shared randomness SBP approximate counting classes P A 0 PP
Computing the minimum energy of the classical Ising model is NP -complete, even for the 2D geometry (with magnetic field) Barahona (1982)
Computing the minimum energy of the classical Ising model is NP -complete, even for the 2D geometry (with magnetic field) Barahona (1982) Local Hamiltonian Problem for general π -local Hamiltonians is QMA -complete for any constant π β₯ 2 Kitaev, Kempe, Regev (2006); QMA -complete for the 2D geometry Oliveira and Terhal (2008)
Computing the minimum energy of the classical Ising model is NP -complete, even for the 2D geometry (with magnetic field) Barahona (1982) Local Hamiltonian Problem for general π -local Hamiltonians is QMA -complete for any constant π β₯ 2 Kitaev, Kempe, Regev (2006); QMA -complete for the 2D geometry Oliveira and Terhal (2008) Local Hamiltonian Problem for π -local stoquastic Hamiltonians is StoqMA -complete for any constant π β₯ 2 SB, DiVincenzo, Oliveira, Terhal (2007)
Computing the minimum energy of the classical Ising model is NP -complete, even for the 2D geometry (with magnetic field) Barahona (1982) Local Hamiltonian Problem for general π -local Hamiltonians is QMA -complete for any constant π β₯ 2 Kitaev, Kempe, Regev (2006); QMA -complete for the 2D geometry Oliveira and Terhal (2008) Local Hamiltonian Problem for π -local stoquastic Hamiltonians is StoqMA -complete for any constant π β₯ 2 SB, DiVincenzo, Oliveira, Terhal (2007) Result 2: Local Hamiltonian Problem for TIM on degree-3 graphs is StoqMA- complete.
Implications for Cubitt-Montanaro complexity classification of 2-local Hamiltonians (arxiv:1311.3161): π - LHP : special case of the 2-Local Hamiltonian Problem. All terms in the Hamiltonian must belong to some fixed set π (with arbitrary real coefficients). Example: π = { πβ¨π, πβ¨π½, πβ¨π½} describes TIM-LHP
Recommend
More recommend