| VQE | |
|---|---|
| Name | Variational Quantum Eigensolver |
| Introduced | 2014 |
| Domain | Quantum computing, Quantum chemistry |
| Authors | Alán Aspuru-Guzik group, Peter J. Love (early contributors) |
| Implemented | IBM Quantum, Rigetti Computing, Google Quantum AI, IonQ |
| Input | Hamiltonian, parameterized quantum circuit |
| Output | Approximate ground-state energy |
VQE
The Variational Quantum Eigensolver (VQE) is a hybrid quantum-classical algorithm for estimating eigenvalues of Hamiltonians, most notably ground-state energies of molecular and condensed-matter systems. VQE leverages parametrized quantum circuits and classical optimization to minimize an energy expectation value, making it a leading application for noisy intermediate-scale quantum (NISQ) devices. Its importance stems from potential quantum advantage for problems in quantum chemistry and materials science that are intractable on classical hardware.
VQE was proposed to address the limited coherence and gate fidelity of early quantum processors by shifting much of the computational burden to classical optimizers. The algorithm frames the ground-state search as a variational principle applied to a parameterized trial state (ansatz) prepared on a quantum processor; measurement of expectation values provides an objective for classical parameter updates. VQE is motivated by foundational results in quantum mechanics (the variational method) and draws conceptual links to techniques in computational chemistry such as Hartree–Fock and configuration interaction methods. Because it reduces circuit depth relative to some fault-tolerant quantum algorithms like quantum phase estimation, VQE is well suited to NISQ era hardware from vendors such as IBM Quantum and Rigetti Computing.
The core theoretical element is the Rayleigh–Ritz variational principle: for a Hamiltonian H, any trial state |ψ(θ)〉 yields E(θ)=〈ψ(θ)|H|ψ(θ)〉 ≥ E0. VQE constructs |ψ(θ)〉 using a parameterized unitary U(θ) applied to an initial state, and decomposes H into measurable Pauli terms (e.g., via a Jordan–Wigner transformation or Bravyi–Kitaev transformation mapping fermionic operators to qubits). The expectation value is estimated by repeated measurements on a quantum processor; a classical optimizer updates θ to minimize E(θ). Prominent theoretical analyses compare VQE to quantum phase estimation and discuss expressibility, trainability, and the influence of barren plateaus as characterized in recent work by researchers at institutions such as MIT and Caltech.
Ansatz design is central to VQE performance. Common families include hardware-efficient ansätze developed for limited coherence superconducting qubit systems, and chemically motivated ansätze such as the unitary coupled cluster with singles and doubles (UCCSD), which relates to classical coupled cluster theory. Hardware-efficient circuits often combine repeated parameterized single-qubit rotations and entangling gates provided by platforms like Google Quantum AI and IonQ. Problem-inspired circuit construction may exploit symmetries enforced by groups (e.g., particle-number conservation) and use techniques from tensor network representations. Trade-offs between expressibility, circuit depth, and trainability are active research topics in academic groups at University of Toronto, Harvard University, and University of Cambridge.
The classical optimizer is a key component of the hybrid loop. Gradient-free methods such as COBYLA and Nelder–Mead and gradient-based methods including parameter-shift rule gradients and stochastic gradient descent have been used. Quantum-aware optimizers (e.g., ADAM adaptations) and Bayesian optimization approaches address noisy objective evaluations. Efficient measurement allocation schemes, grouping of commuting Pauli terms, and shot-frugal strategies reduce sampling cost. Integration with software frameworks like Qiskit, Cirq, and PyQuil supports deployment on both simulators and hardware backends.
VQE has been applied to compute molecular energies for small molecules (e.g., H2, LiH, BeH2) and prototype models in condensed matter such as the Fermi–Hubbard model. Studies by teams at Harvard and Caltech demonstrated proof-of-principle experiments on superconducting and trapped-ion processors. VQE targets problems where electronic structure calculations (e.g., multi-reference correlation) exceed classical methods like density functional theory or full configuration interaction. Extensions include computing excited states (via subspace-search VQE), properties such as dipole moments, and integration with embedding techniques from dynamical mean field theory.
Because VQE runs on noisy hardware, error mitigation techniques are essential. Methods include readout error mitigation, zero-noise extrapolation, probabilistic error cancellation, symmetry verification, and virtual distillation. Researchers at IBM Research and Google have benchmarked mitigation strategies to extend useful circuit depth. Error-aware ansatz selection and noise-adaptive compilation reduce sensitivity to device-specific decoherence channels in superconducting qubits and trapped-ion systems.
Scaling VQE to chemically relevant systems faces challenges in qubit count, circuit depth, and measurement overhead. Mapping second-quantized Hamiltonians requires qubit resources that grow with basis set size; techniques such as active-space selection and qubit tapering via Z2 symmetries mitigate resource demands. Measurement scaling is addressed by advanced grouping and shadow tomography methods. Fundamental obstacles include barren plateaus impeding optimization in large parameter spaces, and the exponential growth of entanglement for strongly correlated systems, which may necessitate fault-tolerant approaches like quantum phase estimation for ultimate accuracy. Ongoing research at institutions including ETH Zurich, University of Waterloo, and companies like IonQ seeks practical pathways to scale VQE toward useful quantum advantage.