arXiv:2609.29098
Evaluation-efficient quantum architecture search with ZX-calculus-based topological reuse
Chenlu Li, Hui Zeng, Dazhi Ding
Abstract
Variational quantum algorithm is a leading approach for quantum chemistry and many-body physics on noisy intermediate-scale quantum devices. The performance depends strongly on the structure of the parameterized quantum circuits. Quantum architecture search (QAS) can automate ansatz design. However, it requires repeated training and evaluation of many candidate circuits, leading to high evaluation cost. In this work, we propose a noise-aware quantum architecture search framework based on ZX-calculus topological reuse (ZX-QAS). The framework encodes the search space with a ternary Gray-code mapping and integrates a noise-aware quantum neural network with a ZX-calculus topological reuse mechanism. The effectiveness of the framework is validated through ground-state energy estimation tasks and one-dimensional transverse-field Ising model tasks under noisy conditions. The results show that the ZX-QAS exhibits stable convergence and remarkably reduces the cost of expensive evaluations under noisy conditions.
AI Summary
Quantum architecture search requires repeated parameter optimization of candidate circuits. This paper asks whether ZX-calculus-derived structural information can reduce those evaluations while retaining ground-state energy accuracy under noise. The proposed ZX-QAS framework combines on-demand ternary Gray-code encoding of rotation choices, sparse entangling layers with at most one CNOT, noise-aware parameter-sharing supernets, and variable-depth evolutionary search. Candidates are converted to simplified ZX graphs. Exact encodings and matching structural signatures receive cached fitness; candidates with sufficiently similar structural features can probabilistically inherit a previously evaluated neighbor’s fitness, with a redundancy penalty. The probability of real evaluation decreases during search but retains a specified floor. Other candidates undergo L-BFGS-B fine-tuning. Evidence comes from simulations of $H_2$, active-space LiH, $H_4$, and an eight-qubit transverse-field Ising chain, using depolarizing, bit-flip, and thermal-relaxation noise. Across 400 evolutionary samples per task, similarity-based reuse accounts for 40–48.25% of samples, in addition to exact-cache hits of 1.25–11%. Reported noisy energy errors include approximately 0.015 hartree for $H_2$ and 0.082 hartree for $H_4$, versus 0.078 hartree for the latter baseline. For the Ising chain, the final dimensionless energy error is approximately 0.54 versus 0.65 for the baseline. Separate encoding tests establish single-rotation locality between adjacent Gray-code configurations and reduced storage through implicit decoding. The paper explicitly treats structural similarity as a heuristic, not circuit equivalence. Reuse can introduce approximation errors and accuracy loss; the similarity threshold, reuse penalty, and minimum evaluation probability require task-dependent adjustment.
Error found: The reported reuse counts establish fewer candidate fine-tuning evaluations, but do not quantify total computational savings after pretraining and ZX-processing overhead. The numerical evidence is restricted to simulated 4–8-qubit benchmarks and includes a modest accuracy loss on noisy H4, supporting an empirical accuracy–evaluation-count tradeoff rather than uniform accuracy preservation. The combined layer encoding in Eqs. (5)–(6) can change both a rotation gate and the CNOT configuration at an adjacent-index rollover, contrary to the claimed single-gate locality. Equation (18) unconditionally copies fitness for matching reduced ideal ZX signatures even though noisy energy and the original-depth penalty need not be preserved by ideal ZX simplification.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.