on complexity of the
play

On complexity of the quantum Ising model Sergey Bravyi (IBM) - PowerPoint PPT Presentation

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


  1. 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

  2. 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

  3. Transverse Ising Model ( TIM ) Qubits live at vertices of a graph β€’ Ising π‘Žπ‘Ž interactions between β€’ nearest neighbor qubits. Local magnetic fields along β€’ π‘Œ and π‘Ž axes. 𝐼 = 𝑕 𝑣 π‘Ž 𝑣 + β„Ž 𝑣 π‘Œ 𝑣 βˆ’ 𝐾 𝑣,𝑀 π‘Ž 𝑣 π‘Ž 𝑀 𝑣 (𝑣,𝑀) π‘Ž 𝑣 = 1 0 π‘Œ 𝑣 = 0 1 0 βˆ’1 1 0

  4. 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

  5. Quantum Annealing (Farhi et al 2001) 𝑗 πœ–| Ξ¨(𝑒) = 𝐼(𝑒/π‘ˆ)| Ξ¨(𝑒) πœ–π‘’ Hard Easy Unitary evolution 0 ≀ 𝑒 ≀ π‘ˆ Easy: 𝐼 0 = βˆ’ 𝑣 π‘Œ 𝑣 Hard: 𝐼 1 = (𝑣,𝑀) 𝐾 𝑣,𝑀 π‘Ž 𝑣 π‘Ž 𝑀 + 𝑣 𝑕 𝑣 π‘Ž 𝑣 𝐼(𝑑) interpolates between 𝐼 0 and 𝐼(1)

  6. Given an adiabatic path 𝐼 𝑑 , 0 ≀ 𝑑 ≀ 1, how large the evolution time π‘ˆ should be ? Adiabatic Theorem 𝐼 βˆ₯ 2 π‘ˆ~ βˆ₯ 𝐼 βˆ₯ + βˆ₯ + βˆ₯ 𝐼 βˆ₯ πœ€ 2 πœ€ 3 πœ€ 2 Here πœ€ is the minimum spectral gap above the ground state of 𝐼 𝑑 , 0 ≀ 𝑑 ≀ 1. Jansen, Seiler, Ruskai, JMP 48, 102111 (2007) We need a smooth path with a non-negligible spectral gap

  7. Big open question: what kind of problems can be efficiently solved by the quantum annealing (QA) ?

  8. 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 𝑁′ 𝑁

  9. 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 𝑁′ 𝑁 TIM Hamiltonians

  10. 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 𝑁′ 𝑁 TIM Hamiltonians Some fixed target class of Hamiltonians

  11. What does efficient simulation mean ? Target Simulator 𝐼 β€² 𝑑 , 0 ≀ 𝑑 ≀ 1 𝐼 𝑑 , 0 ≀ 𝑑 ≀ 1 Adiabatic path π‘œ β€² ≀ π‘žπ‘π‘šπ‘§(π‘œ) π‘œ Number of qubits πœ€ β€² β‰₯ πœ€ πœ€ Minimum spectral gap 𝐾 β€² ≀ π‘žπ‘π‘šπ‘§(π‘œ, πœ€ βˆ’1 , 𝐾) Maximum 𝐾 interaction strength All spins | All spins | + + Ground state at 𝑑 = 0 Ground state at 𝑑 = 1 | β‰ˆ π‘Š| πœ” πœ” Here π‘Š: 𝐃 2 β¨‚π‘œ β†’ 𝐃 2 β¨‚π‘œβ€² is a sufficiently simple encoding

  12. When efficient simulation is unlikely: target QA machine simulator QA machine 𝑁′ 𝑁 TIM Hamiltonians 2-local Hamiltonians

  13. When efficient simulation is unlikely: target QA machine simulator QA machine 𝑁′ 𝑁 TIM Hamiltonians 2-local Hamiltonians BQP Aharonov et al (2007) Oliveira and Terhal (2008)

  14. When efficient simulation is unlikely: target QA machine simulator QA machine 𝑁′ 𝑁 TIM Hamiltonians 2-local Hamiltonians BQP BQPβ‹‚postBPP SB, DiVincenzo, Oliveira, Aharonov et al (2007) Terhal (2007) Oliveira and Terhal (2008)

  15. When efficient simulation is unlikely: target QA machine simulator QA machine 𝑁′ 𝑁 TIM Hamiltonians 2-local Hamiltonians More Powerful β‰  BQP BQPβ‹‚postBPP SB, DiVincenzo, Oliveira, Aharonov et al (2007) Terhal (2007) Oliveira and Terhal (2008)

  16. Stoquastic k-local Hamiltonians System of π‘œ qubits with a Hamiltonian 𝐼 = 𝐼 𝛽 𝛽 Each term 𝐼 𝛽 acts on at most 𝑙 = 𝑃(1) qubits 1. Matrix elements of 𝐼 𝛽 in the standard basis are real. 2. Off-diagonal matrix elements of 𝐼 𝛽 are non-positive: 𝑦|𝐼 𝛽 |𝑧 ≀ 0 for all 𝑦 β‰  𝑧 ∈ 0,1 𝑙

  17. Building blocks for 2-local stoquastic Hamiltonians: Diagonal : Β±π‘Ž 𝑣 , Β±π‘Ž 𝑣 π‘Ž 𝑀 Transverse field: βˆ’π‘Œ 𝑣 0 0 , Elementary βˆ’π‘Œ βŠ— βˆ’π‘Œ βŠ— 1 1 interactions: βˆ’ π‘Œ βŠ— π‘Œ βˆ’ 𝑍 βŠ— 𝑍, βˆ’π‘Œ βŠ— π‘Œ + 𝑍 βŠ— 𝑍

  18. Result 1: universality of TIM for quantum annealing with 2-local stoquastic Hamiltonians simulator QA machine target QA machine = 𝑁′ 𝑁 TIM Hamiltonians Stoquastic 2-local Hamiltonians

  19. Result 1: universality of TIM for quantum annealing with 2-local stoquastic Hamiltonians simulator QA machine target QA machine = 𝑁′ 𝑁 TIM Hamiltonians Stoquastic 2-local Hamiltonians with k-local diagonal terms

  20. Result 1: universality of TIM for quantum annealing with 2-local stoquastic Hamiltonians simulator QA machine target QA machine = 𝑁′ 𝑁 TIM Hamiltonians Stoquastic 2-local Hamiltonians on degree-3 graphs with k-local diagonal terms

  21. Part II Computational hardness of estimating the ground state energy of TIM

  22. Ground state energy: 𝐹 0 = min πœ” 𝐼 πœ” Local Hamiltonian Problem (LHP): Input: (π‘œ, 𝐼 = 𝛽 𝐼 𝛽 , 𝐷 𝑧𝑓𝑑 < 𝐷 π‘œπ‘ ) Yes-instance: 𝐹 0 ≀ 𝐷 𝑧𝑓𝑑 No-instance: 𝐹 0 β‰₯ 𝐷 π‘œπ‘ Decide which one is the case. Promise: 𝐹 0 βˆ‰ 𝐷 𝑧𝑓𝑑 , 𝐷 π‘œπ‘ Normalization: 𝐼 𝛽 ≀ π‘žπ‘π‘šπ‘§ π‘œ , 𝐷 π‘œπ‘ βˆ’ 𝐷 𝑧𝑓𝑑 β‰₯ π‘žπ‘π‘šπ‘§ 1/π‘œ #terms ≀ π‘žπ‘π‘šπ‘§(π‘œ)

  23. Merlin-Arthur games (Babai 1985) I instance of yes/no problem P Merlin Arthur proof Polynomial-time Unlimited classical computer computational power reject accept

  24. A problem belongs to this class if … complexity class yes-instance : Arthur accepts some Merlin’s proof NP no-instance: Arthur rejects any Merlin’s proof

  25. A problem belongs to this class if … complexity class yes-instance : Arthur accepts some Merlin’s proof NP no-instance: Arthur rejects any Merlin’s proof Arthur is a quantum computer. Merlin’s proof can be a quantum state. QMA yes-instance: Arthur accepts some Merlin’s proof with high probability no-instance: Arthur rejects any Merlin’s proof with high probability

  26. A problem belongs to this class if … complexity class yes-instance : Arthur accepts some Merlin’s proof NP no-instance: Arthur rejects any Merlin’s proof Arthur is a quantum computer. Merlin’s proof can be a quantum state. QMA yes-instance: Arthur accepts some Merlin’s proof with high probability no-instance: Arthur rejects any Merlin’s 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€˜ +β€² . Arthur can use | 0 and | + ancillas. SB, Bessen, Terhal, arXiv:0611021

  27. 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

  28. Computing the minimum energy of the classical Ising model is NP -complete, even for the 2D geometry (with magnetic field) Barahona (1982)

  29. 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 𝑙 -local Hamiltonians is QMA -complete for any constant 𝑙 β‰₯ 2 Kitaev, Kempe, Regev (2006); QMA -complete for the 2D geometry Oliveira and Terhal (2008)

  30. 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 𝑙 -local Hamiltonians is QMA -complete for any constant 𝑙 β‰₯ 2 Kitaev, Kempe, Regev (2006); QMA -complete for the 2D geometry Oliveira and Terhal (2008) Local Hamiltonian Problem for 𝑙 -local stoquastic Hamiltonians is StoqMA -complete for any constant 𝑙 β‰₯ 2 SB, DiVincenzo, Oliveira, Terhal (2007)

  31. 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 𝑙 -local Hamiltonians is QMA -complete for any constant 𝑙 β‰₯ 2 Kitaev, Kempe, Regev (2006); QMA -complete for the 2D geometry Oliveira and Terhal (2008) Local Hamiltonian Problem for 𝑙 -local stoquastic Hamiltonians is StoqMA -complete for any constant 𝑙 β‰₯ 2 SB, DiVincenzo, Oliveira, Terhal (2007) Result 2: Local Hamiltonian Problem for TIM on degree-3 graphs is StoqMA- complete.

  32. Implications for Cubitt-Montanaro complexity classification of 2-local Hamiltonians (arxiv:1311.3161): 𝑇 - LHP : special case of the 2-Local Hamiltonian Problem. All terms in the Hamiltonian must belong to some fixed set 𝑇 (with arbitrary real coefficients). Example: 𝑇 = { π‘Žβ¨‚π‘Ž, π‘Žβ¨‚π½, π‘Œβ¨‚π½} describes TIM-LHP

Recommend


More recommend