Probing entanglement and parameter sensitivity in QAOA via Quantum Fisher Information
Published 12 March, 2026
Variational quantum algorithms are widely viewed as a route to near-term quantum advantage, but their performance often depends on a small subset of influential parameters and on how entanglement couples those parameters through the circuit. In a new study published in Quantum Review Letters, a team of researchers use Quantum Fisher Information (QFI) to quantify how the Quantum Approximate Optimization Algorithm (QAOA) states respond to parameter perturbations, and to separate individual parameter relevance from cross-parameter correlations.
In the study, the team used Quantum Fisher Information (QFI) to quantify how QAOA states respond to parameter perturbations, separating individual parameter relevance from cross-parameter correlations. "QFI gives us more than a single score — it exposes both the strength of parameter sensitivity and how strongly parameters become correlated as the circuit evolves," explains senior and corresponding author Prof Shi-Hai Dong.
The team analysed Max-Cut on cyclic and complete graphs, together with random Ising model instances, comparing RX-only and hybrid RX–RY mixers up to depth p=9p=9p=9. They also examined the effects of inserting entangling stages with different patterns. Across problem families, complete-graph Max-Cut consistently produced larger QFI eigenvalues and higher covariance fractions than cyclic graphs, exceeding the shot-noise scaling (4N)(4N)(4N) while remaining below the Heisenberg limit (4N2)(4N^2)(4N2). A saturation trend emerged: the first entangling stage accounted for most of the QFI growth, while additional stages often delivered diminishing benefits and could sometimes reduce the net gain.
"We computed QFI matrices averaged over random parameter samples, highlighting stable global trends and revealing strongly non-uniform parameter relevance that varied with circuit architecture and depth," adds Dong.
Building on these QFI signatures, the researchers introduced a QFI-Informed Mutation (QIm) heuristic that adapts mutation probabilities and step sizes using diagonal QFI entries. QIm improved convergence behaviour and run-to-run stability relative to uniform mutation and random-restart baselines, with the largest gains observed at greater depths and in random Ising model landscapes.
"In the NISQ era, optimisation is often the bottleneck. QFI offers a principled way to diagnose why a circuit is hard to train, and a practical handle for making optimisation more reliable," says Dong.
Overall, the results position QFI as both a structural probe of QAOA ansätze and a practical resource for shaping more reliable optimisation strategies under NISQ-era constraints.
Contact authors:
Shi-Hai Dong
View the author's ORCID record
Research Center for Quantum Physics, Huzhou University, Huzhou 313000, China
Centro de Investigación en Computación, Instituto Politécnico Nacional, UPALM, CDMX 07738, Mexico
Corresponding author at: Centro de Investigación en Computación, Instituto Politécnico Nacional, UPALM, CDMX 07738, Mexico.
Funder:
The authors acknowledge partial support from grants 20251089 and 20251109-SIP-IPN in Mexico. S.H. Dong started this work in China.
Conflict of interest:
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
See the article:
Probing entanglement and parameter sensitivity in QAOA via Quantum Fisher Information
Quantum Review Letters, Volume 2, Pages 1-20