Thermalization Algorithms : Digital vs Analogue

Thermalization Algorithms: Digital vs Analogue

Fernando G.S.L. BrandãoUniversity College London

Joint work with

Michael KastoryanoFreie Universität Berlin

Discrete and analogue Quantum Simulators, Bad Honnef 2014

Dynamical Properties

Hij

Hamiltonian:

State at time t:

Expectation values:

Temporal correlations:

Quantum Simulators, Dynamical

Digital: Quantum Computer

Can simulate the dynamics of every multi-particle quantum system

(spin models, fermionic and bosonic models, topological quantum field theory, ϕ4 quantum field theory, …)

Analog: Optical Lattices, Ion Traps, Circuit cQED, Linear Optics, …

Can simulate the dynamics of particular models

(Bose-Hubbard, spin models, BEC-BCS, dissipative dynamics, quenched dynamics, …)

Static Properties

Hamiltonian:

Static Properties

Hij

Hamiltonian:

Groundstate:

Thermal state:

Compute: local expectation values (e.g. magnetization), correlation functions (e.g. ), …

Static Properties

Hij

Static PropertiesCan we prepare groundstates?

Warning: In general it’s impossible to prepare groundstates efficiently, even of one-dimensional translational-invariant models -- it’s a computational-hard problem (Gottesman-Irani ‘09)

Static PropertiesCan we prepare groundstates?

Warning: In general it’s impossible to prepare groundstates efficiently, even of one-dimensional translational-invariant models -- it’s a computational-hard problem

Analogue: adiabatic evolution; works if Δ ≥ n-c

Digital: Phase estimation*; works if can find a “simple” state |0>

such that

*

(Gottesman-Irani ‘09)

(Abrams, Lloyd ‘99)

H(si)ψi

H(s)ψs = E0,sψs

Δ := min Δ(s)H(s)ψs

H(sf)

Static Properties

Can we prepare thermal states?

Why not? Couple to a bath of the right temperature and wait.

But size of environment might be huge. Maybe not efficient

(Terhal and diVincenzo ’00, …)

S B

Static Properties





S B

Warning: In general it’s impossible to prepare thermal states efficiently, even at constant temperature and of classical models, but defined on general graphs

Warning 2: Spin glasses not expected to thermalize.

(PCP Theorem, Arora et al ‘98)

Static Properties





S B

Warning: In general it’s impossible to prepare thermal states efficiently, even at constant temperature and of classical models, but defined on general graphs

Warning 2: Spin glasses not expected to thermalize.

(PCP Theorem, Arora et al ‘98)

• When can we prepare thermal states efficiently?

• Digital vs analogue methods?

Summary

1. Glauber Dynamics and Metropolis Sampling

- Temporal vs Spatial Mixing

2. Quantum Master Equations (Davies Maps)

3. Quantum Metropolis Sampling

4. “Damped” Davies Maps

- Lieb-Robinson Bounds

5. Convergence Time of “Damped” Davies Maps

- Quantum Generalization of “Temporal vs Spatial Mixing” - 1D Systems

Metropolis SamplingConsider e.g. Ising model:

Coupling to bath modeled by stochastic map Q

The stationary state is the thermal (Gibbs) state:

Metropolis Update:

i j

Metropolis SamplingConsider e.g. Ising model:

Coupling to bath modeled by stochastic map Q

The stationary state is the thermal (Gibbs) state:

Metropolis Update:

• (Metropolis et al ’53) “We devised a general method to calculate the properties of any substance comprising individual molecules with classical statistics”

• Example of Markov Chain Monte Carlo method. Extremely useful algorithmic technique

i j

Glauber Dynamics

Metropolis Sampling is an example of Glauber dynamics:

Markov chains (discrete or continuous) on the space of configurations {0, 1}n that have the Gibbs state as the stationary distribution:

transition matrixafter t time steps

E.g. for Metropolis,

stationary distribution

Temporal Mixing

eigenvalueseigenprojectors

Convergence time given by the gap Δ = 1- λ1:

Time of equilibration ≈ n/Δ

We have fast temporal mixing if Δ = n-c

Spatial MixingLet be the Gibbs state for a model in the lattice V with boundary conditions τ, i.e.

blue: V, red: boundary

0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0

00000

00000

Ex. τ = (0, … 0)

Spatial MixingLet be the Gibbs state for a model in the lattice V with boundary conditions τ, i.e.

blue: V, red: boundary

0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0

00000

00000

Ex. τ = (0, … 0)def: The Gibbs state has correlation length ξ if for every f, g

fg

Temporal Mixing vs Spatial Mixing(Stroock, Zergalinski ’92; Martinelli, Olivieri ’94, …) For every 1D and 2D model, Gibbs state has constant correlation length if, and only, if the Glauber dynamics has a constant gap

constant: independent of the system size

Obs1: Same is true in any fixed dimension using a stronger notion of clustering (weak clustering vs strong clustering)




Obs2: Same is true for the log-Sobolev constant of the system





Obs3: For many models, when correlationlength diverges, gap is exponentially small in the system size (e.g. Ising model)





Obs3: For many models, when correlation length diverges, gap is exponentially small in the system size (e.g. Ising model)

Obs4: Any model in 1D, and any model in arbitrary dim. at high enough temperature, has a finite correlation length

(connected to uniqueness of the phase, e.g. Dobrushin’s condition)





Obs3: For many models, when correlation length diverges, gap is exponentially small in the system size (e.g. Ising model below critical β)

Obs4: Any model in 1D, and any model in arbitrary dim. at high enough temperature, has a finite correlation length

(connected to uniqueness of the phase, e.g. Dobrushin’s condition)

Does something similar hold in the quantum case?

1st step: Need a quantum version of Glauber dynamics…

Lindblad Equation:

(most general Markovian and time homogeneous q. master equation)

Quantum Master EquationsCanonical example: cavity QED

Lindblad Equation:

(most general Markovian and time homogeneous q. master equation)

Quantum Master Equations

completely positive trace-preserving map:

fixed point:

How fast does it converge? Determined by gap of of Lindbladian

Canonical example: cavity QED

Lindblad Equation:

Quantum Master EquationsCanonical example: cavity QED

Local master equations: L is k-local if all Ai act on at most k sites

(Kliesch et al ‘11) Time evolution of every k-local Lindbladian on n qubits can be simulated in time poly(n, 2^k) in the circuit model

Ai

Dissipative Quantum Engineering

Define a master equation whose fixed point is a desired quantum state

(Verstraete, Wolf, Cirac ‘09) Universal quantum computation with local Lindbladian

(Diehl et al ’09, Kraus et al ‘09) Dissipative preparation of entangled states

(Barreiro et al ‘11) Experiment on 5 trapped ions (prepared GHZ state)

Is there a master equation preparing thermal states of many-body Hamiltonians?

…

Davies MapsLindbladian:

Lindblad terms:

: spectral density


Lindblad terms:

Hij

Sα (Xα, Yα, Zα)

: spectral density

Thermal state is the unique fixed point:

(satisfies q. detailed balance: )

Davies Maps

(Davies ‘74) Rigorous derivation in the weak-coupling limit: Coarse grain over time t ≈ λ-2 >> max(1/ (Ei – Ej + Ek - El)) (Ei: eigenvalues of H)

Interacting Ham.

Davies Maps


But: for n spin Hamiltonain H: max(1/ (Ei – Ej + Ek - El)) = exp(O(n))

Consequence: Sα(ω) are non-local (act on n qubits);

cannot be efficiently simulated in the circuit model

(but for commuting Hamiltonian, it is local)

Interacting Ham.

O(n)

Energy

density

O(n1/2)

Davies Maps


But: for n spin Hamiltonain H: max(1/ (Ei – Ej + Ek - El)) = exp(O(n))

Consequence: Sα(ω) are non-local (act on n qubits);

cannot be efficiently simulated in the circuit model

(but for commuting Hamiltonian, it is local)

Interacting Ham.

O(n)

Energy

density

O(n1/2)

• Can we find a local master equation that prepares ρβ?

• Can we at least find a quantum channel (tpcp map) that can be efficiently implemented on a quantum computer whose fixed point is ρβ?

Digital: Quantum Metropolis Sampling(Temme, Osborne, Vollbrecht, Poulin, Verstraete ‘09)

Classical Metropolis:



Quantum Metropolis:random U

1. Prepare (phase estimation)

2.

3. Make the move with prob. (non trivial; done by Marriott-Watrous trick)

Gives map Λ s.t.



Quantum Metropolis:random U

1. Prepare (phase estimation)

2.

3. Make the move with prob. (non trivial; done by Marriott-Watrous trick)

Gives map Λ s.t.

What’s the convergence time? I.e. minimum k s.t.

Seems a hard question!

k


Lindblad terms:

Analogue: “Damped” Davies MapsLindbladian:

Lindblad terms:

Analogue: “Damped” Davies MapsLindbladian:

Lindblad terms:

Thermal state is the unique fixed point:

(satisfies q. detailed balance:

follows from: )

What is the locality of this Lindbladian?

Lieb-Robinson Bound

In non-relativistic quantum mechanics there is no strict speed of light limit. But there is an approximate version

(Lieb-Robinson ‘72) For local Hamiltonian H

HijX

Z

Lieb-Robinson Bound II

(another formulation) For local Hamiltonian H

l l

Lieb-Robinson Bound II

(another formulation) For local Hamiltonian H

time

l l

Applying Lieb-Robinson Bound to “Damped” Davies Maps

Consider:

fact:

proof:

LR bound Damping term

has the Gibbs state as its fixed point (up to error 1/poly(n)) and is O(logd(n))-locality for a Hamiltonian on a d-dimensional lattice.

Can be simulated on a quantum computer in time exp(O(logd(n)))

“Damped” Davies Maps are Approximately Local

Define

fact:

acts trivially on A

acts trivially on B

Mixing in Space vs Mixing in Time

thm If for every regions A and B and f acting on

then , for t = O( 2O(l) 2O(ξ) n), l = O(log(n/ε))

A B

Obs: Converse holds true for commuting Hamiltonians

AC : complement of A (yellow + blue)BC : complement of B (ref + blue)

Convergence Time in 1D

Def ρβ has correlation length ξ if for every f, g

fg



fg

Cor For a 1D Hamiltonian, ρβ has correlation length ξ, then

for t = O( 2O(l) 2O(ξ) n), l = O(log(n/ε))



fg

Cor For a 1D Hamiltonian, ρβ has correlation length ξ, then

for t = O( 2O(l) 2O(ξ) n), l = O(log(n/ε))

thm (Araki ‘69) For every 1D Hamiltonian, ρβ has ξ = O(β)

Thus: Can prepare 1D Gibbs states in time poly(2β, n)

No phase trans. in 1D

Conditional Expectation

Let Ll*A be the A sub-Lindbladian in Heisenberg picture

Note: ,

Conditional Expectation:

fact:

proof: commutes with all and thus with

all

Conditional Covariance and Variance

For a region C:

Ex. If C is the entire lattice,

Conditional Covariance

Conditional Variance

acts trivially on A

acts trivially on B




A B

acts trivially on A

acts trivially on B




A B

AC : complement of A (yellow + blue)BC : complement of B (ref + blue)

Proof Idea

The relevant gap is(Kastoryano, Temme ‘11, …)

We show that under the clustering condition:

A B

Getting:

V : entire latticeV0 : sublattice of size O(lξ)

Conclusions and Open Questions• “Davies like” master equations + Lieb-Robinson bound give interesting

approach for preparing thermal states efficiently.

• Connections between clustering properties of the thermal states (mixing in space) and fast convergence of the master equation (mixing in time), also in the quantum case.

Conclusions and Open Questions• “Davies like” master equations + Lieb-Robinson bound give interesting

approach for preparing thermal states efficiently.

• Connections between clustering properties of the thermal states (mixing in space) and fast convergence of the master equation (mixing in time), also in the quantum case.

Open questions: • Can we get O(log(n))-local Gibbs sampler in any dimension? (true in 2D if can improve Lieb-Robinson bound to Gaussian decay).

• How about really local samplers? Connected to stability question of “Damped Davies” maps.

• Can we prove in generality equivalence of spatial mixing vs temporal mixing? How about in 2D? (how to fix the boundary in the q. case?)

• What are the implications to self-correcting quantum memories?(Fannes, Werner ‘95)

Thermalization Algorithms : Digital vs Analogue

Documents

thermal states

analogue quantum simulators

quantum computer

quantum metropolis sampling

classical models

thermal gibbs state

bosonic models

shsshsf static propertiescan