2026-10-08T03:34:40Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34024641 2026-10-01T05:55:23Z category_29227 item_type_3 month_year_10_2026

Tensorloc N A (25137249) Quantum computation code and data TensorLocTensorLoc is a research prototype for pass-boundary regression checking in quantum compilers. It captures adjacent Qiskit pass snapshots, cancels exact matches exposed at both per-wire DAG frontiers, projects the unmatched windows to their gate support, and invokes a three-valued semantic oracle.<h2 dir="ltr">Verdicts and scope</h2><ul><li><code>EQUIVALENT_CERTIFIED</code>: syntactic, bounded dense-operator, or Clifford-tableau checking established equivalence under the configured numerical semantics.</li><li><code>NON_EQUIVALENT_DETECTED</code>: a deterministic checker or a concrete state probe found a difference.</li><li><code>NO_DIFFERENCE_OBSERVED</code> / <code>UNKNOWN</code>: the invocation abstained. These results are never converted to equivalence or a fault.</li></ul>The current implementation handles unitary snapshots only when qubit count, ordered wire interpretation, and layout metadata are stable.<h2 dir="ltr">Reproduce</h2><pre><pre>python -m pip install -r requirements.txt python -m pytest -q python scripts/run_rq1.py python scripts/run_rq2.py python scripts/run_rq3.py python scripts/run_rq4.py </pre></pre>RQ1 records every compiler callback. RQ2 compares the same eligible adjacent boundaries with TensorLoc, MQT QCEC's default flow, a deterministic QCEC configuration, and QCEC on the extracted windows, using an external process-level wall-clock limit. RQ3 evaluates admitted gate-level mutations and pass-wise baselines. RQ2 discards one warm-up and retains three measured trials per method and boundary; varying repeated verdicts become <code>UNKNOWN</code>. RQ4 audits and reproduces 3 provenance-verified Qiskit historical compiler defects, reporting 3/3 buggy detection, 3/3 exact-pass localization, and 3/3 fixed-release certification.The Qiskit 2.5.1 robustness slice uses the same 18 circuits with one fixed grid/O2 configuration family and is recorded in the new <code>rq2_qiskit251_robustness_*</code> artifacts. All recorded experiments ran sequentially on an Intel Core i9-14900K with 128 GiB RAM under Windows 11 and CPython 3.12.10. The robustness slice has a separate exact dependency lock because the primary study remains on Qiskit 1.3.1.PowerShell reproduction of the Qiskit 2.5.1 slice:<pre><pre>py -3.12 -m venv .venv-qiskit251 .\.venv-qiskit251\Scripts\python.exe -m pip install -r requirements-qiskit251.txt $env:TENSORLOC_RQ2_ROBUSTNESS = 'qiskit251' .\.venv-qiskit251\Scripts\python.exe scripts\run_rq2.py </pre></pre>Bash reproduction of the same slice:<pre><pre>python3.12 -m venv .venv-qiskit251 .venv-qiskit251/bin/python -m pip install -r requirements-qiskit251.txt TENSORLOC_RQ2_ROBUSTNESS=qiskit251 .venv-qiskit251/bin/python scripts/run_rq2.py </pre></pre>Raw observations are stored under <code>data/raw</code>, processed tables under <code>data/processed</code>, and provenance manifests under <code>data/metadata</code>. 2026-10-01T05:55:23Z Dataset Dataset 10.6084/m9.figshare.34024641.v4 https://figshare.com/articles/dataset/Tensorloc/34024641 CC BY 4.0