HamQASBench: A Hamiltonian-Informed Diagnostic Benchmark for Evaluating Quantum Architecture Search
2026-07-06 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors created HamQASBench, a new way to test quantum circuit designs that takes the structure of the problem (the Hamiltonian) into account, rather than just molecule type or number of qubits. They group molecules based on structural features and also check if circuits are too complex or fail in subtle ways that energy accuracy alone misses. By testing different quantum design methods, they found problems like unnecessary complexity, issues with degenerate states, and difficulties with circuit layout and size. Their work helps better understand where current quantum circuit search methods struggle.
Quantum Architecture SearchParameterized Quantum CircuitsHamiltonianVariational Quantum AlgorithmsEntanglementPauli Operator BasisGround StateCircuit Over-parameterizationDegeneracyScalability Bottleneck
Authors
Jiayang Niu, Akib Karim, Yan Wang, Jie Li, Ke Deng, Azadeh Alavi, Muhammad Usman, Yongli Ren
Abstract
Quantum Architecture Search (QAS) automates the design of parameterized quantum circuits for variational quantum algorithms, yet existing benchmarks organize instances by molecular identity or qubit count -- criteria agnostic to Hamiltonian structure -- and rely solely on energy accuracy, which cannot detect structural failures such as over-parameterization on near-product ground states. We introduce HamQASBench, a Hamiltonian-informed diagnostic benchmark organizing 11 molecules into five structural tiers via fingerprints derived from the Pauli operator basis, computational basis representation, and ground-state entanglement. A post-hoc critical-structure extraction procedure identifies minimal circuits consistent with each tier's requirements, complementing energy-based evaluation with per-qubit entanglement analysis and pairwise state fidelity. Benchmarking five QAS methods across four paradigms reveals failure modes invisible to conventional metrics: over-parameterization in the minimalism regime, eigenstate commitment under degeneracy, a representation bottleneck in strongly correlated systems, topology-induced routing failure, and circuit search space growth as a scalability bottleneck.