arXiv:2610.10351
Measurement-Efficient Differentiable Quantum Architecture Search for Combinatorial Optimization
Lukas Theißinger, Thore Gerlach, Christian Bauckhage
Abstract
Differentiable quantum architecture search (DQAS) is a promising framework for the automated design of quantum circuits, particularly for variational quantum optimization algorithms. However, its practical deployment on quantum hardware is limited by the large number of circuit measurements required during optimization, making hardware execution costly. In this work, we show that for a broad class of combinatorial optimization problems and commonly used rotational gate parameterizations, the measurement cost of DQAS can be significantly reduced without changing the optimization objective. We derive the proposed measurement reduction scheme theoretically and validate it experimentally on 3-SAT and MaxCut benchmark problems. Our approach reduces the requested gradient measurement cost by about 39 to 41% while introducing only negligible classical post-processing overhead, lowering the practical cost of executing DQAS on quantum hardware.
AI Summary
Differentiable quantum architecture search (DQAS) jointly optimizes circuit structure and gate parameters, but parameter-shift gradients require many circuit evaluations. This paper asks when those evaluations can be reduced for computational-basis-diagonal combinatorial optimization Hamiltonians without changing the objective or architecture-update rule. For Pauli-string rotations, it derives commutation-based certificates for exactly zero gradients and a one-shift gradient identity involving only Hamiltonian terms that anticommute with the gate generator. When the circuit suffix commutes with the Hamiltonian, this restricted observable remains computational-basis measurable. A Clifford suffix also permits a one-shift estimator through classical Pauli propagation and selection of contributing terms. ME-DQAS checks these conditions for each parameter and otherwise uses ordinary two-shift parameter estimation. Validation uses paired, shot-budgeted simulations on 100 3-SAT and 100 MaxCut instances, with eight qubits, eight circuit positions, and gate pool $\{R_x,R_y,R_z,\mathrm{CZ}\}$. Under a global budget of 750,000 shots, requested gradient-shot costs fall to 58.6% and 60.6% of baseline, respectively. Completed updates increase from 8.99 to 14.05 for 3-SAT and from 8.56 to 13.54 for MaxCut, with lower final normalized best-so-far loss. Zero-gradient elimination accounts for approximately 31–33 percentage points of savings; the diagonal one-shift rule contributes about eight, while the Clifford branch is negligible. Savings depend on the Hamiltonian, gate pool, and suffix structure; non-Clifford mixing can prevent eligibility. The demonstrations are small-system simulations rather than hardware tests, and the reported cost reduction concerns requested parameter-gradient shots.
Error found: The abstract omits the circuit-structure conditions governing measurement savings, while its reported 39–41% reductions are established only in simulations of 8-qubit, length-8 circuits using the gate pool {Rx, Ry, Rz, CZ}, not across general DQAS architectures.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 6 pages, 2 figures. Accepted at the 2026 IEEE 2nd International Conference on Quantum Artificial Intelligence (QAI). Code and data: https://github.com/Newida/ME-DQAS