arXiv:2610.11710
Parameter-Dependent Noise Resilience in the Quantum Approximate Optimization Algorithm
Jorja Kirk, Ashley Montanaro
Abstract
The Quantum Approximate Optimization Algorithm (QAOA) is a variational quantum algorithm used for solving combinatorial optimization problems, which is thought to be a promising application of near-term quantum hardware. However, high rates of noise in these devices can obscure meaningful results. Therefore, it is crucial to understand and suppress the effects of noise on the algorithm's output. Here, we find that the stability of the algorithm in the presence of noise is dependent on its variational parameters. Using a Pauli noise model, we derive an upper bound on the total variation distance between the noisy and noiseless circuit output as a function of these parameters. Results from noisy simulations support this bound and give insight into the best methods to use for parameter generation in the presence of noise. We also introduce new methods of optimization using error penalties and truncated search based on this bound to manufacture error resilience. We extend this analysis to the quantum adiabatic algorithm (QAA) by taking the continuous limit of the QAOA.
AI Summary
The paper asks how QAOA’s variational parameters affect sensitivity to noise, focusing on unweighted Max-Cut with the standard $X$ mixer and stochastic single-qubit Pauli errors between layers. Propagating $Z$ errors toward computational-basis measurement and $X$ errors toward the initial $|+^n\rangle$ state converts errors into parameter-sign changes. Operator-norm and triangle inequalities then bound the total variation distance between noisy and noiseless outputs. For an error on degree-$d_j$ qubit $j$ after layer $s$, the linear bound is $2\sum_{i>s}|\beta_i|$ for $Z$, $2d_j\sum_{i\leq s}|\gamma_i|$ for $X$, and their sum for $Y$. Averaging the multiple-error bound gives $\mathcal E=2n\sum_{i=1}^p[\epsilon_{zy}i|\beta_i|+\epsilon_{xy}(p-i+1)d_{\rm avg}|\gamma_i|]$. This favors small angles and cost-increasing, mixer-decreasing schedules. Simulations on 6–10-qubit graphs qualitatively correlate this bound with output-distribution and energy changes. In six-node, depth-four tests, precomputed parameters become competitive with optimized parameters around error probability $0.01$ per qubit per layer. Penalty-based optimization and restricted parameter searches yield marginal, inconsistent improvements. A continuous-limit analysis lower-bounds this robustness surrogate for single-schedule adiabatic evolution. For two-schedule unstructured search, a local adiabatic-condition calculation finds that lowering the schedule crossing increases the runtime enough to worsen the surrogate. These are bounds on an error upper bound, not lower bounds on actual error. Stated limitations include loose bounds, small numerical instances, inconclusive optimization benefits, and unverified scaling and applicability to hardware noise.
Error found: The abstract's suggestion that the proposed optimization modifications manufacture resilience exceeds their numerical validation, which the paper itself describes as marginal and inconclusive. The proof of Theorem 1 incorrectly replaces a sum of maxima by a maximum of sums, although the stated sine-dependent TVD bound can be repaired using TVD <= 1. The expected-bound formula counts a pre-circuit error boundary absent from the specified after-each-layer noise model, making its claimed expectation equality incorrect while preserving a conservative upper bound. Lower-bounding an upper-bound certificate does not establish the paper's claimed physical limitation on adiabatic error resilience or exclude improvements from two schedules. The continuous-limit integral requires timestep scaling of both angles and error probabilities that is not specified, so it is not established as the limit of the stated discrete noise model.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 23 pages, 9 figures