Quantum Variational Algorithms

Variational Quantum Eigensolver (VQE)

The VQE is a hybrid quantum-classical algorithm for finding the ground state energy of a Hamiltonian . It uses the variational principle:

where is a parameterized quantum state and is the true ground state energy. The algorithm minimizes using classical optimization while evaluating the energy on a quantum computer.

Parameterized Quantum Circuits (PQCs)

A parameterized quantum circuit applies gates with tunable parameters :

where are parameterized rotations and are fixed entangling gates.

Common architectures:

Hardware-efficient ansatz: native gate set, low depth
UCCSD (Unitary Coupled Cluster): chemistry-inspired, systematic
ADAPT-VQE: grows the circuit dynamically

The key challenge is avoiding barren plateaus — exponentially vanishing gradients in high-dimensional parameter spaces.

QAOA (Quantum Approximate Optimization Algorithm)

QAOA solves combinatorial optimization problems by alternating between problem and mixer Hamiltonians:

where:

is the problem (cost) Hamiltonian
is the mixer Hamiltonian
is the QAOA depth

As , QAOA converges to the optimal solution. For MaxCut on 3-regular graphs, already gives a 0.6924 approximation ratio.

The Cost Function

For a combinatorial optimization problem with objective function , the cost Hamiltonian is:

The expectation value is estimated by sampling and averaged over many shots.

The optimization landscape can have local minima and saddle points, making classical optimizer choice critical.

Barren Plateaus

Barren plateaus are a fundamental challenge for variational algorithms. For random PQCs with qubits:

This means gradients vanish exponentially with system size.

Mitigation strategies:

Layerwise training: start shallow, grow deeper
Local cost functions: use local rather than global observables
Symmetry preservation: exploit problem symmetries
Noise-aware training: account for hardware noise in the circuit design

Python: VQE for H2 Molecule

import numpy as np

def vqe_h2():
    # Simplified VQE for H2 molecule (minimal basis).
    coeffs = {
        'I': -0.4804, 'Z0': 0.3435, 'Z1': -0.4347,
        'Z0Z1': 0.5716, 'X0X1': 0.0910, 'Y0Y1': 0.0910
    }

    def energy(theta):
        e = coeffs['I']
        e += coeffs['Z0'] * np.cos(2*theta)
        e += coeffs['Z1'] * np.cos(2*theta)
        e += coeffs['Z0Z1'] + coeffs['X0X1'] + coeffs['Y0Y1']
        return e

    from scipy.optimize import minimize
    result = minimize(energy, 0.0, method='COBYLA')
    print(f"Ground state energy: {result.fun:.6f} Ha")
    print(f"Exact FCI energy: -1.1372 Ha")

vqe_h2()

This demonstrates the core VQE loop: prepare state, measure energy, optimize classically.

Circuit Ansatz Design

The choice of ansatz circuit is critical for VQE performance:

Hardware-efficient ansatz: uses native gate set with minimal depth. Pros: fast execution. Cons: may not represent ground state well.

UCCSD (Unitary Coupled Cluster Singles and Doubles): chemistry-inspired ansatz that systematically includes excitations. Pros: physically motivated. Cons: deep circuits.

ADAPT-VQE: dynamically grows the circuit by selecting the most impactful operator at each step. Pros: optimal circuit depth. Cons: requires adaptive circuit compilation.

Parameter Optimization Landscape

The VQE energy landscape can have:

Local minima: suboptimal parameter configurations
Saddle points: flat directions in parameter space
Barren plateaus: exponentially vanishing gradients
Noise-induced local minima: hardware errors create artificial minima

Strategies to handle these challenges:

Multiple random initializations
Layerwise training (start shallow, grow deeper)
Gradient-free optimizers (COBYLA, Nelder-Mead)
Noise-aware training