Variational Quantum Eigensolver Optimization Enters a New Era with the VQEzy Dataset
VQEzy Dataset
With the release of VQEzy, the first large-scale, open-source dataset for parameter initialization, a significant bottleneck restricting the practical implementation of Variational Quantum Eigensolvers (VQEs), a leading class of algorithms for the Noisy Intermediate-Scale Quantum (NISQ) era, has been overcome.
VQEzy, created by scholars Hui Min Leung and Fan Chen from Indiana University and Chi Zhang, Mengxin Zheng, and Qian Lou from the University of Central Florida, offers 12,110 examples of VQE specs and full optimization trajectories.
The choice of initial parameters has a significant impact on the performance of VQEs, which are used in many-body physics, quantum chemistry, and related domains. Enhancing trainability and reducing the possibility of convergent to suboptimal local minima depend on efficient parameter initialization. Although new machine learning based parameter initializers have demonstrated state-of-the-art performance, the lack of extensive datasets has severely limited their advancement.
You can also read AMD & IBM Join To Develop Quantum Centric Supercomputing
Overcoming Limitations of Prior Research
Three main limitations of the existing datasets, which were previously accessible to researchers, made them inadequate for reliable machine learning training: (1) they were limited to a single domain; (2) they were small in scale, usually consisting of only a few hundred instances; and (3) they lacked complete coverage, frequently leaving out ansatz circuits or complete optimisation trajectories.
VQEzy was specifically created to address these issues. Compared to earlier datasets, it is orders of magnitude larger and richer. The dataset includes seven typical jobs with different circuit implementations and qubit sizes, and it covers the three primary VQE application domains of quantum many-body physics, quantum chemistry, and random benchmarking.
Importantly, VQEzy offers a multitude of data features for each of the 12,110 cases, such as the optimized VQE parameter vector, comprehensive circuit specifications, issue Hamiltonians, and, most importantly, complete optimisation trajectories. VQEzy is a useful tool for theoretical research and real-world VQE optimisation because of this extensive data, which includes ground-state energy history, parameter dynamics, and barren plateau behaviour.
The dataset is openly accessible and is intended to be updated and expanded over time with community involvement.
You can also read Clifford Circuit Initialization Improves QAOA And VQE
Building a Diverse Quantum Resource
VQEzy was built using a methodical three-step process that included VQE optimisation, ansatz circuit selection, and problem Hamiltonian creation.
1.Diverse Hamiltonians
Applications from three key domains are included in VQEzy:
- Quantum Many-Body Physics: The one-dimensional Heisenberg XYZ (1D_XYZ) model, the one-dimensional Fermi–Hubbard (1D_FH) model, and the two-dimensional Transverse-Field Ising (2D_TFI) model are all included in the field of quantum many-body physics. For example, for both 4-qubit and 12-qubit scenarios, 1D_XYZ has 2000 distinct parameter tuples. Three thousand instances of 4, 6, and 8-qubit spin chains were contributed by the 1D_FH model.
- Quantum Chemistry: Three molecular Hamiltonians are included in quantum chemistry. Different bond lengths produce different configurations; for instance, one bond length can produce 1000 variants, while another can produce 150 and 160 configurations.
- Random VQE: Random half-integer Pauli string coefficients were used to produce 2800 four-qubit Hamiltonians in order to reduce structural bias.
2. Selected Ansatz Circuits
The choice of ansatz has a considerable impact on VQE performance. For many-body physics tasks, the researchers used the CZRXRY ansatz; for molecular Hamiltonians, they used the strongly entangling ansatz; and for random VQE benchmarking, they used the U3CU3 ansatz.
3. Standardized Optimization
Because of its ability to balance performance, computational cost, and GPU acceleration support, the Adam optimizer with a learning rate of was used in all VQE optimisation studies. It took more than 200 hours of wall-clock time to acquire the data for the first 12,110 instances utilizing an AMD Ryzen 5 1600 CPU and an NVIDIA RTX 3090 GPU.
You can also read UCR: University Of California, Riverside For Noisy Links
Parameter Landscapes and Insights
To obtain important insights into the optimized parameter space by characterizing the data using dimensionality reduction techniques such as Multidimensional Scaling (MDS) and t-distributed Stochastic Neighbour Embedding (t-SNE). The visualizations show that well-defined clusters formed by optimized parameters are adequate for differentiating between jobs and domains.
For instance, tasks such as 1D_XYZ, 1D_FH, and 2D_TFI show distinct parameter distributions in the quantum many-body domain. Additionally, study of the 1D_FH model’s optimized parameters showed distinct symmetries that mirrored those in QAOA. A more complicated environment results from the emergence of richer symmetries as the number of qubits rises, such as the O(2) symmetry shown in the 8-qubit case.
Important information was also obtained from the analysis of optimized ground-state energies. The random VQE application displays various modes originating from the discretized Hamiltonian structure rather than the number of qubits, whereas quantum many-body physics and quantum chemistry jobs display energy modes corresponding to the number of qubits.
You can also read SEEQC Quantum & IBM Boost DARPA Quantum Benchmarking
Future Applications and Expansion
The VQEzy dataset is well-positioned to offer significant advantages in a number of research areas:
- VQE Initialization and Optimization: It supports sophisticated ML-based initialisation tactics across a variety of domains by offering beginning points that reduce initial loss and speed up convergence.
- Transfer Learning: Previously limited by a lack of data, its broad scope and variety of tasks allow for methodical investigations of parameter transferability and model-agnostic meta-learning.
- VQE Architecture Design: VQEzy facilitates models that produce task-specific ansatz architectures by acting as a standard for investigating and creating ideal VQE circuits.
You can also read ORCA Computing Photonic Quantum System at UK’s NQCC
