No evidence of spatially correlated or non-Markovian dephasing across ten couplers of ibm_marrakesh, and why a white-noise null is not enough to look for it (CORRELATED-NOISE-01 Stage 1)
Stage 1 of a preregistered hardware test for non-local dephasing on the 156-qubit IBM Heron r2 processor ibm_marrakesh (job dbtp67dav06s73cfbsm0; 244 PUBs, 499,712 circuit shots, 139 s QPU). Ten couplers ≥5 hops apart were measured simultaneously: Bell states under DD with 0–256 ns sublattice staggers, Φ+ vs Ψ+, active vs idle spectators, echo and DD pulse-count series, Ramsey arms. No Φ+/Ψ+ asymmetry (C_DFS = +0.43 pp, 95% CI [−0.13, +1.00]), no spectator effect, no memory contrast, no pulse-collision step. The analysis as locked selects a non-Markovian model over white noise by ΔBIC = 10,263 but fails its preregistered held-out test (p = 0.82): the transferable memory claim is falsified. Synthetic validation shows that analysis selects the memory model on 100% of purely local 1/f or OU noise; an amended analysis with local coloured-noise nulls (hashed before the job, published after it, as disclosed) selects the local model. Verdict under both: local noise only. Bounds fast correlated dephasing at about 1 pp; says nothing about correlations slower than about 1 µs. The record contains the paper, LaTeX source, preregistration, theory, amendments A-2/A-3 with validation, the lock (38ebfe7f…) and QPY (158de032…), raw counts, both analyses and the scripts that regenerate every number.
Authors
- Devin Phillip Davis
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24406434
- Primary Topic
- Quantum Computing Algorithms and Architecture
- Type
- preprint