Quantum Algorithms for Many-Body Systems: A chemistry and materials science perspective. Matthias Troyer

Size: px

Start display at page:

Download "Quantum Algorithms for Many-Body Systems: A chemistry and materials science perspective. Matthias Troyer"

Victor Flynn
5 years ago
Views:

1 Quantum Algorithms for Many-Body Systems: A chemistry and materials science perspective Matthias Troyer

2 International Journal of Theoretical Physics, VoL 21, Nos. 6/7, 1982 Simulating Physics with Computers Richard P. Feynman Department of Physics, California Institute of Technology, Pasadena, California Received May 7, 1981 Feynman proposed to use quantum computers to simulate quantum physics

3 Simulating quantum computers on classical computers Simulating a quantum gate acting on N qubits needs O(2 N ) memory and operations Qubits Memory Time for one operation kbyte microseconds on a smartwatch MByte milliseconds on smartphone GByte seconds on laptop TByte seconds on cluster PByte minutes on top supercomputers? EByte hours on exascale supercomputer? ZByte days on hypothetical future supercomputer? 250 size of visible universe age of the universe

4 Simulating a quantum gate acting on N qubits needs O(2 N ) memory and operations

5 First applications that reached a petaflop on ORNL Domain area Code name Institution # of cores Performance Notes Source: T. Schulthess Materials DCA++ ORNL 213, PF Materials WL-LSMS ORNL/ETH 223, PF Chemistry NWChem PNNL/ORNL 224, PF Materials DRC ETH/UTK 186, PF Nanoscience OMEN Duke 222,720 > 1 PF Biomedical MoBo GaTech 196, TF Chemistry MADNESS UT/ORNL 140, TF Materials LS3DF LBL 147, TF Seismology SPECFEM3D USA (multiple) 149, TF Combustion S3D SNL 147, TF 2008 Gordon Bell Prize Winner 2009 Gordon Bell Prize Winner 2008 Gordon Bell Prize Finalist 2010 Gordon Bell Prize Hon. Mention 2010 Gordon Bell Prize Finalist 2010 Gordon Bell Prize Winner 2008 Gordon Bell Prize Winner 2008 Gordon Bell Prize Finalist

.., " r N,t) Describes (almost) everything we encounter in daily life with a very simple Hamilton H

6 The Theory of Everything Laughlin & Pines PNAS 2000 The N-body Schrödinger equation i! Ψ(" r 1,..., " r N,t) t = HΨ( " r 1,..., " r N,t) Describes (almost) everything we encounter in daily life with a very simple Hamilton H = 1 Δ 2m i +V ext ( r! i ) + i i q i q j! r i! r j It is a simple linear partial differential equation (PDE) But is exponentially complex since it lives in 3N dimensions i,j

!! F[ρ]+ d 3 rv( r )ρ( r ) ( ) Successful in calculating properties many metals,

7 Density functional theory and quantum chemistry Approximates the N-body Schrödinger by a tractable 1-body problem E 0 = min ρ(! r )!!! F[ρ]+ d 3 rv( r )ρ( r ) ( ) Successful in calculating properties many metals, insulators, semiconductors Band structure of silicon Walter Kohn John A. Pople 1998 Nobel prize in chemistry J.R.Chelikowsky and M.L.Cohen, PRB (1974)

Cuprate high temperature superconductors Undoped materials: half-filled band and metal according to DFT but antiferromagnetic insulator in experiment!

8 Cuprate high temperature superconductors Undoped materials: half-filled band and metal according to DFT but antiferromagnetic insulator in experiment! Band structure calculation breaks down! Doped materials: high-temperature superconductors What causes superconductivity? Are there room temperature superconductors? Fermi level Band structure of La2CuO4

9 Preparing the ground state On a classical computer Imaginary time projection Power method or other iterative eigensolver Ψ GS Ψ GS = lim τ e τ H Ψ T = lim n (H Λ) n Ψ T On a quantum computer Imaginary time evolution Power method Unitary operations + measurements: prepare trial state projectively measure energy obtain the ground state if the ground state energy was measured Ψ T Ψ T φ n with H φ n = E n φ n φ n picked with propability φ n Ψ T 2

10 Quantum phase estimation Energy can be measured through the phase of a wave function after unitary time evolution U φ n e iht φ n = e ie n t φ n = e iφ φ n We can only measure relative phases, thus do a controlled evolution 0 φ n 1 2 ( ) φ n 1 2 ( 0 φ n +U 1 φ n ) = 1 2 ( 0 +e iφ 1 ) φ n 1 ( 2 (1+e iφ ) 0 +(1 e iφ ) 1 ) φ n Measure the ancilla qubit to obtain the phase

11 Solving quantum chemistry on a quantum computer 1. Select a finite (generally non-orthogonal) basis set 2. Perform a Hartree-Fock calculation to get an approximate solution get an orthogonal basis set 3. Find the true ground state of the Hamiltonian in this new basis set H = c p c q + pq t pq V pqrs exact classical approach: full-configuration interaction exponential complexity! pqrs c p c q c r c r

12 Solving quantum chemistry on a quantum computer 1. Select a finite (generally non-orthogonal) basis set 2. Perform a Hartree-Fock calculation to get an approximate solution get an orthogonal basis set 3. Find the true ground state of the Hamiltonian in this new basis set H = c p c q + 4. Prepare a good guess for the ground state 5. Perform quantum phase estimation to get the ground state wave function and energy pq t pq pqrs V pqrs c p c q c r c r

13 Representing fermion terms by quantum circuits Map the occupation of each spin-orbital to the states of one qubit 0 = 1 = Density operators get mapped to Pauli matrices n i = σ z i Hopping terms get mapped to spin flips with Jordan-Wigner strings c p c q =σ p ( ) Time evolution gets mapped to circuits built from unitary gates q 1 σ z + σ i p i=p+1 e iθ ( c c p q +c q c p ) =

14 Guessing a good trial state Mean field approximation Hartree-Fock Density functional theory Variational approaches: fit parameters in variational ansatz Configuration-Interaction Variational Monte Carlo Tensor network methods (e.g. DMRG) McMillan, Phys. Rev (1965) White, Phys. Rev. Lett. (1992) Neural network states Carleo & Troyer, Science (2017)

15 Preparing a good trial state Adiabatic evolution from a simple initial Hamiltonian H 0 with known ground state to the desired Hamiltonian H 1 Good starting Hamiltonians H(s)=(1-s)H 0 +sh 1 Mean-field approximation (with right broken symmetry) Non-interacting particles (in the right symmetry sector)

16 Cooling into a good trial state Evolve the system coupled to a heat bath that is being cooled See e.g. Kaplan, Klco, Roggero, arxiv: Monte Carlo sampling with the quantum Metropolis algorithm Use a unitary update + phase estimation to propose a new energy eigenstate Accept or reject according to the Metropolis algorithm Solves the sign problem of QMC by working with energy eigenstates Temme et al, Nature (2011)

Simulating time evolution on quantum computers There are O(N 4 ) interaction terms in an N-electron system H = t pq c p c q + V pqrs c p c q c r c r H m pq pqrs We need to evolve separately under

17 Simulating time evolution on quantum computers There are O(N 4 ) interaction terms in an N-electron system H = t pq c p c q + V pqrs c p c q c r c r H m pq pqrs We need to evolve separately under each of them Efficient circuits available for each of the N 4 terms M m=1 M = O(N 4 ) terms e iδth M e iδτ H m m=1 Runtime estimates turn out to be O(NM 2 ) = O(N 9 ) It s efficient since it s polynomial!

18 The polynomial time quantum shock Estimates for a benchmark molecule: Fe 2 S 2 with 118 spin-orbitals Gate count Parallel circuit depth Run 100ns gate time 300 years We cannot declare victory proving the existence of polynomial time algorithms We need quantum software engineers to develop better algorithms and implementations

19 The result of quantum software optimization Estimates for a benchmark molecule: Fe 2 S 2 with 118 spin-orbitals Gate count Parallel circuit depth Reduced gate count Parallel circuit depth Run 100ns gate time 300 years Run 100ns gate time 20 minutes Attempting to reduce the horrendous runtime estimates we achieved Wecker et al., PRA (2014), Hastings et al., QIC (2015), Poulin et al., QIC (2015) Reuse of computations: Parallelization of terms: Optimizing circuits: Smart interleaving of terms: Multi-resolution time evolution: Better phase estimation algorithms: O(N) reduction in gates O(N) reduction in circuit depth 4x reduction in gates 10x reduction in time steps 10x reduction in gates 4x reduction in rotation gates

20 Improvements in algorithms Original circuit Improved circuit

21 Reducing the number of terms Reduce the number of terms in the evolution from O(N 4 ) to O(N 2 ) by using real-space representations and quantum Fourier transforms First quantized formulation requires expensive quantum arithmetic Kivlichan, Wiebe, et al, J. Phys. A (2017) Plane wave basis Babbush, Wiebe, et al, arxiv: requires much larger number of basis states Can the asymptotic advantages be realized with reasonable runtimes?

22 Avoiding the Trotter breakup New quantum algorithms based on series expansions into sums of unitary operators have exponentially better error scaling K e ih U k M K α k U = β m V m m=1 Childs & Wiebe, QIC (2012) Berry et al, STOC 14 Low & Chuang, PRL (2017). Are these methods advantageous given the unavoidable errors due to finite basis sets/lattice spacings?

23 Nitrogen fixation Fertilizer production using Haber-Bosch process needs 3% of the world s natural gas production But bacteria can do it in the soil without a huge factory!

24 Understanding biological nitrogen fixation Intractable on classical supercomputers But a 200-qubit quantum computer will let us understand it

25 Simulation strategy

26 Quantum developer tools Learn quantum programming Develop quantum algorithms Estimate runtime and memory needs MICROSOFT.COM/QUANTUM

27 Total cost with error correction We want very good qubits! Better algorithms! Better error correction and gate synthesis

28 Steps in software development for quantum applications quantum software engineers 6. Embed into specific hardware and check runtime 5. Add error correction and check overhead and resources 4. Optimize code until logical circuit depth < Check for quantum speedup 2. Implement all oracles and subroutines 1. Find quantum algorithm with quantum speedup 7. Run the quantum algorithm to solve the problem

29 Coming soon: Quantum programming language Visual Studio extensions Quantum simulator Learn more & sign up at: MICROSOFT.COM/QUANTUM

Accelerating QMC on quantum computers. Matthias Troyer

Accelerating QMC on quantum computers Matthias Troyer International Journal of Theoretical Physics, VoL 21, Nos. 6/7, 1982 Simulating Physics with Computers Richard P. Feynman Department of Physics, California