Lecture 8.2 - Variational Quantum Eigensolver
Uploaded by
hu jackLecture 8.2 - Variational Quantum Eigensolver
Uploaded by
hu jackVariational Quantum
Eigensolver
Omar Shehab
Staff Research Scientist
IBM Quantum
Road Map
• Build up high-level description of VQE
• Discuss theoretical underpinnings
• Sample code walk through
Quantum Phase Estimation recap
Quantum phase estimation Iterative quantum phase estimation
VQE: Motivation
Goal: minimize the energy of a system
Optimizer Cost Function 𝐸: ℝ! ↦ ℝ
𝜃
𝐸 parameters
ℝ! ↦ ℝ ↓
𝐸(𝜃) real number
• Initialize params 𝜃"
• Repeat:
• Evaluate 𝐸(𝜃# )
• Choose 𝜃#$%
VQE: Motivation
Goal: minimize the energy of a system
Optimizer Cost Function 𝐸: ℝ! ↦ ℝ
𝜃
𝐸 parameters
ℝ! ↦ ℝ ↓
𝐸(𝜃) real number
• Initialize params 𝜃"
• Repeat:
Use a
• Evaluate 𝐸(𝜃# ) Quantum Computer
• Choose 𝜃#$% for this part
5
Cost Function: What is a Hamiltonian?
Def:
A Hamiltonian 𝐻 is a hermitian matrix describing the energy of a system
If the system is in state|𝜓⟩,
the energy of the system is: ⟨𝜓|𝐻|𝜓⟩
𝑚𝑖𝑛 ⟨𝜓|𝐻|𝜓⟩
|"⟩
We want to know the minimum energy of H.
In other words: the lowest eigenvalue, 𝜆! = ⟨𝜓" |𝐻|𝜓" ⟩
= 𝜆!
Trial States + The Variational Principle
Parameterize some continuous
subset 𝑀! of quantum states
"!
|𝜓(𝜃)⟩ ∈ 𝑀! ⊂ ℂ
Where k = dim( 𝜃 ) = 𝒪(poly)
Note: we are not guaranteed that 𝑀"
contains the ground state!
WHP: |𝜓$ ⟩ ∉ 𝑀%
Trial States + The Variational Principle
Trial States The Variational Principle
Parameterize some continuous
subset 𝑀! of quantum states ⟨𝜓(𝜃)|𝐻|𝜓(𝜃)⟩ ≥ ⟨𝜓!|𝐻|𝜓!⟩
"!
|𝜓(𝜃)⟩ ∈ 𝑀! ⊂ ℂ (Still holds if |𝜓(𝜃)⟩ isn’t pure)
Where k = dim( 𝜃 ) = 𝒪(poly)
⟨𝜓(𝜃)|𝐻|𝜓(𝜃)⟩
Note: we are not guaranteed that 𝑀" ≥ ⟨𝜓! |𝜓! ⟩
contains the ground state! ⟨𝜓(𝜃)|𝜓(𝜃)⟩
WHP: |𝜓$ ⟩ ∉ 𝑀% Proof attached after conclusion slide.
8
Cost Function Breakdown
Compile 𝜃 into Execute circuit Estimate
Measure Expectation
Combine
value of
𝜃 a quantum and prepare
trial state
Pauli Strings 𝐻 results to
circuit estimate
+
Generate pulse |𝜓(𝜃)⟩
sequences for ⟨𝜓(𝜃)|𝑃# |𝜓(𝜃)⟩ ⟨𝜓(𝜃)|𝐻|𝜓(𝜃)⟩
⟨𝜓(𝜃)|𝐻|𝜓(𝜃)⟩
quantum = 𝐸(𝜃)
circuit
𝐸(𝜃)
Cost Function Breakdown
Cost Function
Compile 𝜃 into Execute circuit Measure Combine
𝜃 a quantum and prepare
trial state
Pauli Strings results to
circuit estimate
+
Generate pulse |𝜓(𝜃)⟩ ⟨𝜓(𝜃)|𝑃! |𝜓(𝜃)⟩
sequences for ⟨𝜓(𝜃)|𝐻|𝜓(𝜃)⟩
quantum = 𝐸(𝜃)
circuit
𝐸(𝜃)
The whole VQE-nchillada
Start
Advantages:
Determine Problem Hamiltonian
▪ Uses shallow circuits
Decompose / Map Hamiltonian
▪ Results in an efficient
representation of the ground
#
𝜃 Choose new 𝜃
𝐸(𝜃) state
Optimizer
Robust to incoherent AND
Generate Prepare Measure Combine coherent noise
hardware trial state Pauli results
controls Strings into † Approximate representation &
|𝜓(𝜃)⟩ estimate only when successful.
𝐸(𝜃)
Cost Function
1
The whole VQE-nchillada
Start
Upcoming Talks:
Determine Problem Hamiltonian
1. Optimizer
Decompose / Map Hamiltonian 2. Hamiltonian Mapping
3. Hamiltonian Mapping & Reduction
4. Initial States + Variational Forms
𝜃 Choose new 𝜃
𝐸(𝜃) 5. Hardware Control
Optimizer 6. Error Mitigation Techniques
Generate Prepare Measure Combine
hardware trial state Pauli results
controls Strings into
|𝜓(𝜃)⟩ estimate
𝐸(𝜃)
Cost Function
1
VQE Application Areas
Chemistry Optimization
• Ground State Preparation • Constraint Satisfaction
• Excited state prep • Traveling Salesman
• Clustering
Computational Complexity of VQE
When should you try VQE?
Your problem is naturally expressed with a Hamiltonian
& the solution finding states with high ground-state overlap.
When does VQE perform well?
Not so easy to answer.
Computational Complexity of VQE
Traveling Salesman
QMA
Factoring* NP
(Open!)
BPP
BQP
Hard P
Easy for Quantum
Easy for Classical
Computational Complexity of VQE
VQE under the following conditions:
Traveling Salesman
QMA
Factoring* NP
(Open!)
BPP
BQP
Hard P
Easy for Quantum
Easy for Classical
Computational Complexity of VQE
VQE under the following conditions:
Traveling Salesman 𝐻 is k-local
QMA
Factoring* NP
(Open!)
BPP
BQP
Hard P
Easy for Quantum
Easy for Classical
Computational Complexity of VQE
VQE under the following conditions:
Traveling Salesman 𝐻 is k-local
QMA
𝐻 ⪰ 0‡
Factoring* NP
(Open!)
BPP
BQP
Hard P
Easy for Quantum ‡
Where depth and # VQE iterations is 𝒪(poly)
Easy for Classical
Computational Complexity of VQE
VQE under the following conditions:
Traveling Salesman 𝐻 is k-local
QMA
𝐻 ⪰ 0‡
Factoring* NP
Constant-depth
(Open!) /
2D circuits
BPP
BQP
Hard P
‡
Where depth and # VQE iterations is 𝒪(poly)
Easy for Quantum 0
The VQE ansatz has 2D connectivity
Easy for Classical
1
Computational Complexity of VQE
VQE under the following conditions:
Traveling Salesman 𝐻 is k-local
QMA
𝐻 ⪰ 0‡
Factoring* NP
Constant-depth
(Open!) /
2D circuits
BPP
BQP Gapped local 1D 𝐻
Hard P
‡
Where depth and # VQE iterations is 𝒪(poly)
Easy for Quantum 0
The VQE ansatz has 2D connectivity
Easy for Classical
2
Computational Complexity of VQE
VQE under the following conditions:
Traveling Salesman 𝐻 is k-local
QMA
Ising model in
Factoring* NP nonuniform mag. field
(Open!)
𝐻 ⪰ 0‡
BPP BQP Constant-depth
/
Hard P 2D circuits
Easy for Quantum Gapped local 1D 𝐻
Easy for Classical ‡
Where depth and # VQE iterations is 𝒪(poly)
0
2 The VQE ansatz has 2D connectivity
Computational Complexity of VQE
2
Key take aways
VQE is a heuristic algorithm that:
• Extends the reach of classical techniques
• Is suited to near term hardware
• Requires only shallow circuits
• Shows robustness to coherent and incoherent noise
The whole VQE-nchillada
Start
Upcoming Talks:
Determine Problem Hamiltonian
1. Optimizer
Decompose / Map Hamiltonian 2. Hamiltonian Mapping
3. Hamiltonian Mapping & Reduction
4. Initial States + Variational Forms
𝜃 Choose new 𝜃
𝐸(𝜃) 5. Hardware Control
Optimizer 6. Error Mitigation Techniques
Generate Prepare Measure Combine
hardware trial state Pauli results
controls Strings into
|𝜓(𝜃)⟩ estimate
𝐸(𝜃)
Cost Function
2
Check out Qiskit Textbook demo
https://learn.qiskit.org/course/ch-applications/simulating-molecules-using-vqe
The Variational Principle (quick proof)
Expand Hamiltonian
!
𝐻 = ∑ 𝜆$ |𝜓$ ⟩⟨𝜓$ |for eigenvalues𝜆' ≤ 𝜆& ≤ ⋯ ≤ 𝜆!
$%&
!
⟨𝜓(𝜃)|𝐻|𝜓(𝜃)⟩ = ∑ 𝜆$ ⟨𝜓(𝜃)|𝜓$ ⟩⟨𝜓$ |𝜓(𝜃)⟩
$%&
!
"
⟨𝜓(𝜃)|𝐻|𝜓(𝜃)⟩ = ∑ 𝜆$ ⟨𝜓(𝜃)|𝜓$ ⟩
$%&
𝑚𝑖𝑛⟨𝜓(𝜃)|𝐻|𝜓(𝜃)⟩ = 𝜆' → |𝜓(𝜃)⟩ = |𝜓' ⟩
Thank you
