Nadarasa · release note

v0.4.5 — Floquet native port on Selene

v0.4.5 is the corroboration beat: the mixed-field Ising Floquet magnet from G19 is recompiled in Guppy and run on the Selene Quest emulator, so the same subharmonic collapse appears on an all-to-all trapped-ion compiler with an independent noise model.

G20 — Floquet native port

384 simulated cells (2 lattices × 4 kick imperfections × 8 cycles × 6 noise levels). 512 shots per cell. Exact-diagonalization baseline vs ideal Selene vs H2 noise ladder, plus Richardson ZNE.

/nadarasa/g20

Verdicts

V1 structure: subharmonic peak collapses as kick imperfection grows.
V2 agreement: worst ideal deviation 0.0495 vs 0.1250 threshold.
V3 resilience: 93.8% peak retention at 1× H2; quadratic ZNE within 2.0%.
V4 corroboration: qualitative cross-platform mechanism match.

What the result means

G19 mapped the validation hierarchy for a 74-qubit heavy-hex superconducting result: noise-model calibration → estimator agreement → classical comparison → cross-platform corroboration. G20 closes the last layer on a small, classically simulable instance. The mechanism — a period-doubled staggered magnetization that loses amplitude with kick imperfection — reproduces on an all-to-all trapped-ion compiler.

The numerical values are not a match to the published data: six qubits over eight cycles is far from the 74-qubit regime. The depolarizing noise model is also not a device-characterized quasiprobability model, so the ZNE performance here is an upper bound on what probabilistic-error-cancellation-free extrapolation can achieve on hardware.

Next gate

The remaining deferred option from v0.4.3 is ADAPT-GQE composition: take a small generated/variational chemistry ansatz and run it through the rule-(N/M/P) canonicalisation pipeline before execution on Selene. This connects the AI-generated-chemistry literature to the existing QPDE and rewriter tooling.