Robustness of structured QAOA mixers under noise

Determine whether the structured QAOA mixers for minimum-weight vertex cover in protein–protein interaction network alignment retain their performance advantage over penalty-based mixers under realistic hardware noise.

Background

The paper compares seven QAOA formulations for protein–protein interaction network alignment, differing in how the minimum-weight vertex-cover constraint is enforced. Penalty-based variants impose feasibility through the cost Hamiltonian, whereas structured clique-cover and fully constrained mixers restrict the evolution toward feasible subspaces.

In noiseless simulations, the fully constrained mixer guarantees feasibility but requires substantially greater circuit depth, while cheaper structured mixers offer intermediate resource costs. Because deeper circuits are expected to accumulate more hardware errors, the paper explicitly identifies as unresolved whether the feasibility and performance benefits of structured mixers persist when noise is introduced.

References

The feasibility guarantee of the constrained mixer holds only for the ideal circuit, so whether the structured mixers keep their advantage under noise is open.

— Quantum Optimisation for Protein-Protein Interaction Network Alignment  (2609.05238 - Stahl et al., 4 Sep 2026) in Section 4, subsection “QAOA variants on small instances,” paragraph “Hardware cost and sampler behaviour”