Independent Research August 2026

Checking a quantum optimization across compiler targets

This independent research used a held-out 64-bit circuit benchmark. The protocol fixed the inputs, comparison rules, workspace budget, and compiler settings before evaluation.

The question

Would a proposed quantum-circuit optimization still reduce resource use after compilation for different target layouts?

Check circuit behavior and compare resource use across different compiler targets without changing the acceptance rules after seeing the results.

See the result

Method

Define the test. Preserve the evidence.

The study compared the prototype with Reqomp, an existing optimization tool. It checked circuit identity and phase behavior, ran 60 Qiskit compilation jobs across three target layouts, and used LibLSQECC for a separate logical-resource estimate. Control and ablation comparisons helped distinguish the effects of scheduling and relative-phase transformations.

What the work produced

A frozen benchmark package, compiler comparison results, integrity records, and a report that retained both the favorable and unfavorable findings.

Tools used Qiskit 2.5.1 · LibLSQECC v2 · Reqomp · OpenQASM

Result

Evidence for the next decision.

Under the tested LibLSQECC settings, the prototype’s estimated total logical-instruction volume was about 11.0% lower than the Reqomp baseline: 6,575,175 versus 7,389,105. The Qiskit comparison met the prototype’s acceptance rule on one of three targets, short of the required two. The study therefore recorded a mixed overall result.

Decision: Keep the resource-benefit claim specific to the tested conditions; the study did not establish a general advantage across compiler targets.

Evaluation

How the result was checked.

The held-out benchmark used fixed targets, routing seeds, versions, and settings. The Qiskit acceptance rule required the prototype to pass on at least two of three targets. A separate LibLSQECC comparison evaluated estimated logical-instruction volume. Integrity checks covered circuit identity, phase behavior, and frozen input hashes.

Review the evidence checks

The completed August 5, 2026 holdout study supplied the compiler results and overall mixed classification.
A September 5 evidence review matched both result-publication files and all three preregistered file hashes. It recalculated the resource ratio and checked the target pass count against the saved summary.
This review checked stored evidence. It did not rerun compiler jobs, execute quantum hardware, or provide independent expert review.

Scope

Where these findings apply.

These are compiler and resource-model results for a defined circuit family, fixed targets, seeds, versions, and settings.
They do not establish faster execution on quantum hardware, general quantum advantage, or customer-system performance.
The software toolchains supplied external implementations; their use is not independent expert review or certification of the research.

Back to Work