Post on 20-Apr-2020
Quantum Computation
and Statistical Mechanics
Maarten Van den Nest
Max Planck Institute for Quantum Optics
El Escorial, July 11th
2011
Recent investigations establish and exploit mappings betweenclassical statistical mechanics an quantum information & computation:
–
Quantum algorithms–
Measurement-based QC –
Strongly correlated systems & PEPS–
Quantum error-correction & fault tolerance–
…
Such mappings allow to
–
interchange techniques
between these two fields–
Formulate quantum algorithms
for problems in Stat Mech
Statistical Mechanics and QIT
Overview of connections between Stat Mech and Quantum Computation
Emphasis on:
–
Power of quantum versus classical computation–
Classical simulation of quantum computation–
Computational models: circuit model, measurement-based QC–
Complexity of classical spin systems
…Not so much: phase transitions, critical behavior, etc..
Scope of this talk
Outline
Part 0: Classical spin models
Part I: Stat Mech and Quantum Circuits
Part II: Stat Mech and Measurement-Based QC
0.
Classical spin models
Edge models
Ising model: 2-level spins sa = 1, -1
Potts model: q-level spins sa = 0, ..., q-1
With/without magnetic fields, e.g.
Spins located at vertices, interactions along edges
H({s}) = -
Σ
Jab
sa
sb
H({s}) = -
Σ
Jab
δ(sa
. sb
)
-
Σ
hab
sa
sa
sb haJab
Vertex models
Spins located at edges, interactions at vertices
Local energy H(s,t,u,v) associated to each configuration around vertex
Six-vertex model:
1 2
3 4
5 6
7 8
0 00 00 0
0 0
ϖ ϖϖ ϖϖ ϖ
ϖ ϖ
⎡ ⎤⎢ ⎥⎢ ⎥⎡ ⎤ =⎣ ⎦ ⎢ ⎥⎢ ⎥⎣ ⎦
stuvW
Eight-vertex model:
–
2-state spins on 2D lattice
–
Boltzmann weights Wst
= exp[-βH(s,t,u,v)]uv
2 70ϖ ϖ= =
s t
uw
Partition function
E.g. Ising model partition function
For each spin configuration {s} on the lattice, compute the product of all local Boltzmann weights
Partition function Z is the sum, over all spin configurations, of such products
Z contains info about thermodynamical properties of system
Z = Σ ∏
e-βJabsasb
{sa} edges
I.
Stat Mech
and Quantum circuits
I.A.
Mappings
Consider a poly-size quantum circuit C acting nearest-neighbor q-level quantum systems
Consider standard basis states |L⟩ and |R⟩ (e.g. |L⟩ = |R⟩ = |00…0⟩)
We will identify ⟨L|C|R⟩ with partition function of classical spin model
C =
Nearest-neighbor gate acting on (at most) two
systems
Quantum circuits
⟨L|C|R⟩ = tensor network
Graphical representation:
–
Each gate
becomes a vertex
with 4 incident edges:
–
Contraction
of two indices = gluing
together corresponding edges
abcdU
a c
b d
αα
α∑ ab b'
c c'd'U V
a c
b α 'c
'b 'd
Quantum circuits as tensor networks
For the whole circuit we find a 2D square lattice:
–
At each edge of 2D lattice sits an index variable which takes q values–
The variables left/right are fixed in boundary conditions L
and R–
Each configuration a, b, c, d
around a vertex is given a weight Uab
–
⟨L|C|R⟩
is given as sum, over all configurations, of products of Uabcd
cd
Quantum circuits as tensor networks
a c
b d
1Ls
2Ls
LNs
1Rs
2Rs
RNs
.∑
a,b,c,d,..⟨L|C|R⟩
=
Uabcd
We can interpret:
–
the variables a, b, c, d
as q-level classical spins and –
the weights Uab
= exp[-βH(a,b,c,d)]
as Boltzman
weights
Corresponds to vertex model:
⟨L|C|R⟩
= partition function of vertex model
on (tilted)
2D square lattice with left and right boundary conditions
cd
Tensor networks as partition functions
⟨L|C|R⟩
= ∑
a,b,c,d,...
∏vertices of 2D latt
abcd
ice
U
2-level classical spin at each vertex of 2D latticeSpins at the left/right are fixed in boundary conditions L and RTwo spins i, j at an edge have Boltzmann weight wij = exp[-βH(i,j)]
1Ls
2Ls
LNs
1Rs
2Rs
RNs
ijw
jkw
i j
k
What about edge models?
Corresponding quantum circuit C:–
Weights on horizontal
edges become single-qubit gates Σ
wij
|i ⟩⟨j|
–
Weights on vertical
edges become diagonal 2-qubit gates: Σ
wjk
|jk⟩⟨jk|–
Boundary conditions become computational basis states
Similar to before, ⟨L|C|R⟩ is partition function of edge model
1Ls
2Ls
LNs
2Rs
RNs
1Rs
ijw
jkw
1Ls
2Ls
LNs
1Rs
2Rs
RNs
jkw
i j
k
ijw
From edge models to quantum circuits
I.B.
Applications
Application 1: Quantum Algorithms
Immediate quantum algorithm for Z = ⟨L|C|R⟩ whenever Boltzmann weights are chosen such that C is unitary
circuit
–
standard method: Hadamard
test
This yields (additive) approximation: algo returns number c such that (with probability exponentially close to 1) one has
If Boltzmann weights correspond to universal quantum gate set, the corresponding partition function problem is BQP-complete
E.g. 6-vertex model and Ising model on 2D lattice are BQP-complete
≤1| - L C R | c
poly(N)
Quantum algorithms
Some remarks
Since quantum circuit is to be unitary, one generally needs complexBoltzmann weights (= unphysical)
This family of Q algo’s allows to efficiently approximate Z. However the exact evaluation of Z is #P-hard and thus generally intractable
–
also for quantum computers
–
Bonus:
using mappings to Q circuits one can in fact prove
#P-hardness of Z !
BQP completeness is strong indication that no classical algo exists
–
Note: BQP completeness e.g. includes Shor’s
factoring algo
Application 2: Classical simulations
Evaluation of ⟨L|C|R⟩ corresponds to evaluation of Z of lattice model
If lattice model is exactly solvable (i.e. Z can be computed efficiently)then ⟨L|C|R⟩
can be computed efficiently
However, watch out for
–
Translation invariance–
Complex couplings–
Finite lattice size–
Boundary conditions
Nevertheless, certain solvable models can be translated into classically simulatable quantum circuits
Classical simulations via solvable models
For example, the eight-vertex model
is solvable for finite dimensions, complex non-translation-invariant couplings, whenever the condition
is fulfilled. The solution is given by mapping to free fermions.
Correspondingly, Q circuits composed of such 2-qubit gates have efficient classical simulations
(quantum matchgate
circuits)
1 2
3 4
5 6
7 8
0 00 00 0
0 0
ϖ ϖϖ ϖϖ ϖ
ϖ ϖ
⎡ ⎤⎢ ⎥⎢ ⎥⎡ ⎤ =⎣ ⎦ ⎢ ⎥⎢ ⎥⎣ ⎦
abcdW
1 8 2 7 3 6 4 5ϖ ϖ ϖ ϖ ϖ ϖ ϖ ϖ− = −
Example: 8-vertex model and matchgates
II.
Stat Mech
and Measurement-based QC
Measurement-Based QC (= One-way QC)
Preparation of 2D cluster state: universal resource state Sequence of adaptive single-qubit measurementsRemaining qubits are (up to local basis change) in desired output state
Understanding the power of MBQC
2D cluster states are universal quantum computers via measurements
We want to understand where this exceptional computational power originates
–
Which other resource states are universal?–
Is there an essential feature that makes 2D cluster states universal?
–
Which states are efficiently simulatable?–
What is role of entanglement?–
…
?
?
?
?
? ?
?
?
?
?
1D cluster universal?
Mapping MBQC to Stat Mech
What will follow: map lattice models to entangled resource states à
la cluster states
Properties of lattice model are reflected in computational powerof resource state
II.A.
Mappings
Start with graph
Place qubit at each edge
Place qubit at each site
Define
Cluster-type states
Vertex
qubits
Edge
qubits
|ψ⟩
= ⊗|sa
⟩ ⊗ |sa
+ sb
⟩Σ{sa}
a ab
These states belong to familyof stabilizer states
–
i.e. eigenstates
of sets ofPauli operators
Relation to the stabilizer formalism
These states belong to familyof stabilizer states
–
i.e. eigenstates
of sets ofPauli operators
1 Pauli operator per vertex
Relation to the stabilizer formalism
These states belong to familyof stabilizer states
–
i.e. eigenstates
of sets ofPauli operators
1 Pauli operator per vertex
1 Pauli operator per edge
Relation to the stabilizer formalism
These states belong to familyof stabilizer states
–
i.e. eigenstates
of sets ofPauli operators
1 Pauli operator per vertex
1 Pauli operator per edge
|ψ⟩ is unique joint eigenstate
Relation to the stabilizer formalism
2D Ising model with external fields
Partition function = overlap
Z =
⟨α|ψ⟩
The mapping
Product state containing info of temperature β and couplings Jab and ha
Edges ab
Vertices a|α⟩
= ⊗
|αa
⟩ ⊗ |αab
⟩a ab
|αab
⟩
= e-βJab|0⟩
+ eβJab|1⟩
|αa
⟩
= e-βha|0⟩
+ eβha
|1⟩
2D Ising model with external fields
Partition function = overlap
Z =
⟨α|ψ⟩
The mapping
Product state containing info of temperature β and couplings Jab and ha
Edges ab
Vertices a|α⟩
= ⊗
|αa
⟩ ⊗ |αab
⟩a ab
|αab
⟩
= e-βJab|0⟩
+ eβJab|1⟩
|αa
⟩
= e-βha|0⟩
+ eβha
|1⟩
Similar mappings for arbitrary graphs, Ising without fields, q-state Potts model, etc
1D Ising
model with
fields 1D cluster state
1D Ising
model without
fields
–
Open boundary conditions
Product state
–
Periodic BCs
GHZ state
2D Ising
model with
fields 2D cluster state
2D Ising
model without
fields Planar code state
Examples
ISING MODEL QUANTUM STATE
II.B.
Applications
Application 1: Classical simulations & universality of MBQC
Mapping connects computational power of resource state with hardness of computing Z
of corresponding spin model
Exactly solvable model gives rise to classically simulatable resource state
Intractable model (probably) gives rise to powerful resource state
Understanding the power of MBQC
Partition function = overlap
Z =
⟨α|ψ⟩
1D Ising
model with
fields 1D cluster state
1D Ising
model without
fields
–
Open boundary conditions
Product state
–
Periodic BCs
GHZ state
2D Ising
model with fields 2D cluster state
2D Ising
model without
fields Planar code state
Exactly solvable models
ISING MODEL QUANTUM STATE
1D Ising
model with fields 1D cluster state
1D Ising
model without fields
–
Open boundary conditions
Product state
–
Periodic BCs
GHZ state
2D Ising
model with fields 2D cluster state
2D Ising
model without fields Planar code state
An NP-hard model
ISING MODEL QUANTUM STATE
1D Ising
model with fields 1D cluster state
1D Ising
model without fields
–
Open boundary conditions
Product state
–
Periodic BCs
GHZ state
2D Ising
model with fields 2D cluster state
2D Ising
model without fields Planar code state
An NP-hard model
ISING MODEL QUANTUM STATE
Universal
state
Application 2: Completeness
= overlap between cluster state |ϕ2D⟩ and product state
Consider |ψ⟩ associated with some classical model:
2D cluster state is universal resource for MQC. Therefore:
|ϕ2D
⟩
= 2D Cluster state on N qubits|β⟩
= product state on measured qubits
Therefore
Universal MQC and the Ising
model
{ }α ϕ⊗ 2Dψ = ' I
αGZ ({J}) = ψ
{ }'α α α ϕ IsingG 22D D= ψ = Z ({J}) Z ({J, = J'})
Ising2D Z
The 2D Ising
model is complete
Z of 2D Ising specializes to Z of any (reasonable) classical spin model
–
E.g. D-dimensional, q-level, vertex-
or edge model, etc..
Ising coupling strengths need to be complex. This problem can be resolved by considering instead Z2
lattice gauge theory in 4D
IsingG 2DZ ({J}) = Z ({J, J'})
- Initial model on n spins
- couplings {J}-
2D Ising
model on N = poly(n) spins
-
Couplings {J} on original spins
-
Couplings {J’} on additional spins
Thank you very much!