Quantum grid algorithm fails to beat random sampling in simulations
A peer-reviewed simulation benchmark found no advantage for a hybrid quantum approach over random sampling in small power-grid problems. The method assigned generator on-off decisions to its quantum component and power-flow calculations to classical computing. Even excluding quantum execution time, it did not pull ahead across the tested systems of five to thirteen generators.
Science··Midday
Random search kept pace with the hybrid method
A hybrid quantum algorithm did not improve on uniform random sampling in a peer-reviewed power-grid simulation study. Nine datasets represented systems with five to thirteen generators, with 120 randomly constructed problems for each size. The task combined choosing which generators to switch on with finding power flows compatible with nonlinear alternating-current constraints. Randomly generated networks explored a broad parameter range rather than replicating a particular national electricity grid.[1]
Splitting the variables reduced the qubit burden
In the tested design, the quantum component sampled generator on-off choices, while a classical solver calculated the continuous power-flow variables for each choice. An alternative design encoded the entire problem in qubits, the information units of quantum computing. Its resource requirement was too large for the researchers to benchmark even small examples. The split design was compared with random sampling and classical optimisation software. No working electricity network or physical quantum processor formed part of the experiment.[1]
Quantum operations were given zero execution time
The timing comparison excluded the simulated quantum component’s execution time, giving the hybrid approach a favourable assumption. Even then, the tested sizes showed no useful lead over random sampling. The authors identify twenty-four or more generators as a scale requiring further investigation; they do not demonstrate an advantage there. That larger benchmark exceeded their computational capacity. Circuit design and training choices also bound the scope of the negative result.[1]
