Variational Quantum Algorithm

A class of hybrid quantum-classical algorithms using parameterized circuits optimized by classical computers.


Variational Quantum Algorithms (VQAs) are the primary approach for near-term (NISQ) quantum computing. They combine quantum circuit evaluation with classical optimization.

The Structure

┌─────────────────────────────────────────────────────┐
│                Classical Computer                    │
│  ┌─────────────┐                    ┌────────────┐  │
│  │  Optimizer  │◄──── Cost ◄───────│   Measure  │  │
│  │   (θ→θ')    │                    │ expectation│  │
│  └──────┬──────┘                    └─────▲──────┘  │
│         │ New θ                           │         │
│         ▼                                 │         │
│  ┌──────────────────────────────────────────────┐  │
│  │              Quantum Computer                 │  │
│  │  |0⟩ ──[Ansatz U(θ)]──── Measure             │  │
│  └──────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────┘
  1. Ansatz: Parameterized quantum circuit
  2. Evaluate: Measure expectation value
  3. Optimize: Classical optimizer updates
  4. Repeat: Until convergence

Key Components

Ansatz (Circuit Structure)

Parameterized circuit that can represent the solution:

Ansatz TypeDescription
Hardware-efficientUses native gates
Problem-inspiredEncodes domain knowledge
UCCSDChemistry-motivated

Cost Function

What you’re minimizing:

  • Energy (for VQE)
  • Objective value (for QAOA)
  • Classification loss (for QML)

Classical Optimizer

Updates parameters based on cost:

  • Gradient-based: ADAM, L-BFGS
  • Gradient-free: COBYLA, Nelder-Mead
  • Quantum-aware: SPSA, QNG
AlgorithmApplication
VQEGround state energy
QAOACombinatorial optimization
VQLSLinear systems
VQCClassification
QGANGenerative models

Why Variational?

NISQ-Friendly

  • Short circuits (low depth)
  • Resilient to some noise
  • Flexible structure

Trainable

  • Classical optimizer handles complexity
  • Adapts to problem instance
  • No deep understanding needed

Challenges

ChallengeIssue
Barren plateausVanishing gradients
NoiseCorrupts cost landscape
OptimizationLocal minima, long training
ExpressibilityCan ansatz represent solution?
TrainabilityCan we find the solution?

Outlook

VQAs are pragmatic but limited:

  • Good for exploring NISQ capabilities
  • Unclear when they outperform classical
  • May not scale to large problems

Bridge to fault-tolerant era or dead end? Jury’s out.


See also: VQE, QAOA, NISQ