VQEzy Dataset Unlocks New Variational Quantum Eigensolver
VQEzy Dataset Launches Variational Quantum Eigensolver Optimisation VQEzy Dataset
VQEzy, the first large-scale, open-source dataset for parameter initialisation, removes a major barrier to the practical implementation of Variational Quantum Eigensolvers (VQEs), a leading class of algorithms for the Noisy Intermediate-Scale Quantum (NISQ) era.
VQEzy, produced by Indiana University's Hui Min Leung and Fan Chen and the University of Central Florida's Chi Zhang, Mengxin Zheng, and Qian Lou, provides 12,110 VQE specs and full optimisation trajectories. In many-body physics, quantum chemistry, and related fields, VQE performance depends on initial parameters. Efficient parameter initialisation improves trainability and reduces suboptimal local minima. New machine learning-based parameter initialisers have shown groundbreaking performance, but a paucity of datasets has slowed their development. Overcoming Prior Research Limits The three primary drawbacks of the datasets available to academics make them unsuitable for machine learning training: (1) they were limited to a single domain; (2) they were modest, usually a few hundred instances; and (3) they often lacked complete coverage, missing ansatz circuits or optimisation trajectories. VQEzy was built to fix these difficulties. It is orders of magnitude larger and richer than previous datasets. Seven sample jobs with varying circuit implementations and qubit sizes cover the three core VQE application domains of quantum many-body physics, quantum chemistry, and random benchmarking. The optimised VQE parameter vector, thorough circuit specifications, problem Hamiltonians, and, most crucially, entire optimisation trajectories are available for each of the 12,110 examples in VQEzy. This rich data, including ground-state energy history, parameter dynamics, and barren plateau behaviour, makes VQEzy ideal for theoretical research and real-world VQE optimisation. The dataset is public and will be updated and expanded with user input. Diversifying Quantum Resources VQEzy was constructed using VQE optimisation, ansatz circuit selection, and problem Hamiltonian generation. Different Hamiltonians VQEzy contains applications from three main domains: Quantum many-body physics includes the one-dimensional Heisenberg XYZ (1D_XYZ), Fermi–Hubbard (1D_FH), and Transverse-Field Ising (2D_TFI) models. In all 4-qubit and 12-qubit cases, 1D_XYZ has 2000 parameter tuples. The 1D_FH model contributed 3,000 4, 6, and 8-qubit spin chains. Quantum chemistry has three molecular Hamiltonians. One bond length can yield 1000 combinations, while another can produce 150 and 160.
Random VQE: To avoid structural bias, 2800 four-qubit Hamiltonians were generated using random half-integer Pauli string coefficients. Selected Ansatz Circuits Ansatz selection affects VQE performance. The CZRXRY, highly entangling, and U3CU3 ansätze were employed for many-body physics, molecular Hamiltonians, and random VQE benchmarking, respectively. Normalised Optimisation The Adam optimiser with a learning rate of was employed in all VQE optimisation investigations since it balanced performance, computational cost, and GPU acceleration. Data for the first 12,110 instances was collected in over 200 hours using an AMD Ryzen 5 1600 CPU and an NVIDIA RTX 3090 GPU. Perspectives on Parameters Characterise the data using dimensionality reduction methods like Multidimensional Scaling (MDS) and t-distributed Stochastic Neighbour Embedding to provide insights into the optimised parameter space. Visualisations indicate that optimised parameter clusters may distinguish jobs and domains. The quantum many-body domain has separate parameter distributions for jobs like 1D_XYZ, 1D_FH, and 2D_TFI. Additionally, the 1D_FH model's optimised parameters displayed QAOA-like symmetry. As qubits increase, deeper symmetries like the 8-qubit O(2) symmetry create a more intricate environment. Analysis of optimised ground-state energies yielded important insights. The discretised Hamiltonian structure determines the energy modes of the random VQE application, while quantum many-body physics and quantum chemistry jobs use the qubit number. Future Uses and Growth In several research fields, the VQEzy dataset is advantageous: VQE Initialisation and Optimisation: It provides starting points that reduce initial loss and accelerate convergence for complex ML-based initialisation strategies across domains. Transfer Learning: Methodical parameter transferability and model-agnostic meta-learning investigations are now possible due to its large scope and variety of assignments.
The VQEzy standard helps models create task-specific ansatz architectures by analysing and developing optimum VQE circuits.