Skip to content
Quantum.world

The Signal

Study finds hybrid quantum grid solver did not beat random sampling

A study published in Scientific Reports found that a simulated hybrid quantum-classical method did not outperform uniform random sampling on power-grid optimization tests. The comparison covered randomly generated cases with five to 13 generators and assumed quantum operations took no time.

Study finds hybrid quantum grid solver did not beat random sampling

A difficult grid scheduling problem

A hybrid quantum-classical approach did not outperform uniform random sampling in simulations of power-grid scheduling problems, according to a study published in Scientific Reports. The paper appeared on Nov. 11, 2026, and its simulations covered randomly generated cases with five to 13 generators.

Igor Gaidai and Rick Mukherjee tested a single-period alternating-current optimal power flow and unit commitment problem, or AC-OPF-UC. It combines decisions about which generators to commit, represented by binary variables, with continuous variables governed by nonlinear AC power-flow constraints.

Two approaches, one qubit-efficient

The researchers considered two ways to apply quantum optimization. One applied the quantum approximate optimization algorithm, or QAOA, to a fully discretized version of the problem, encoding the equality and inequality constraints with penalty terms and slack variables.

The authors say that full-encoding route would require hundreds of qubits even for small systems, so they did not benchmark its performance. Their second approach separated the problem’s binary and continuous variables: the quantum component handled generator-status choices, while classical optimization worked out the continuous power-flow variables for each sampled choice.

Compared with classical search

The team compared the qubit-efficient hybrid method with SCIP, a classical mixed-integer solver; SMAC, which optimizes binary variables through Bayesian optimization; and uniform random sampling. The latter samples generator-status combinations without optimizing their selection. The researchers assessed solution quality using an approximation ratio, defined as the optimal bitstring cost divided by the cost of the sampled bitstring, with higher values indicating better results.

In the five-generator dataset, the authors report that the hybrid method, SMAC and uniform sampling all found optimal solutions within seconds. Each had an approximation ratio of one across those instances, which had only 32 possible generator-status assignments. The team gave each instance a 30-minute budget; SCIP failed to find feasible solutions in most cases and was left out of later comparisons.

Simulation caveat and scale

For the 13-generator dataset, the paper says the hybrid, SMAC and uniform methods showed nearly identical performance, with SMAC falling somewhat behind the other two. Across tests with 10 to 13 generators, the researchers found no significant differences between hybrid and uniform sampling at intermediate run times.

The timing comparison made a favorable assumption for the quantum method: the quantum state preparation was simulated, and that simulation time was not counted. The authors instead treated quantum operations as if they took no time, so the result is a simulation benchmark under an idealized timing assumption, not a run on quantum hardware.

Gaidai and Mukherjee attribute the lack of separation between the methods partly to the tested problem sizes: random sampling could reach good bitstrings before the variational optimization had meaningfully adjusted its sampling distribution. They estimate that systems with 24 or more generators would need to be tested before the hybrid approach might outperform uniform sampling, but say doing so with their methodology would take roughly 10 million CPU hours, beyond their computational capacity.

Sources

  1. Scientific Reports
  2. arXiv

Related stories

Explore

More articles