When Do Entangling Gates Add Predictive Signal? A Preregistered Falsification Study of Quantum Reservoir Features for Combinatorial Optimization
David Vesterlund · Zenodo (CERN European Organization for Nuclear Research) · 2026
Choosing the best heuristic for a combinatorial optimization instance depends on landscape structure that cheap graph descriptors capture only partly. We test whether a small quantum circuit, used as a fixed feature map, adds predictive signal about solver performance beyond such descriptors. Graphs are spectrally compressed to eight supernodes and encoded in a single-layer circuit of RY rotations and RXX entangling gates; eight pooled statistics from Z- and X-basis expectation values feed a classical ridge readout. In a preregistered protocol on 10,000 simulated graphs, the features add ΔR² = 0.0208 over twelve descriptors. Flexible classical models with access to the same compressed graph input substantially outperform the quantum feature map. Removing entangling gates lowers the gain, and the Z-basis component peaks at intermediate entangling strength, consistent with a closed-form phase-accumulation mechanism. Hardware experiments on a 20-qubit trapped-ion processor show feature transfer from simulation, while noise-aware controls indicate that the apparent entanglement advantage is driven by differential noise robustness rather than additional predictive information. We claim no quantum advantage. The contribution is a falsification-tested, mechanistically interpretable, hardware-portable feature map and an evidence-first evaluation of when entangling gates add predictive signal.