Nadarasa · release note

v0.4.3 — four white-paper tracks in one notebook

v0.4.2 was a multi-experiment release: QPDE on a 2-qubit ethylene active space, the same QPDE under H2-class noise, a zero-noise extrapolation, Laplacian moments separating a WL-indistinguishable graph pair, and a reading note on the Leviatan et al. heavy-hex Floquet result. v0.4.3 is the consolidation beat that ties those tracks together and clears the ground for the next experimental gate.

G16 — QPDE ethylene

20/20 cells PASS on the noiseless Selene Quest emulator. Recovered gap 0.8099 Ha vs 0.8000 Ha reference.

/nadarasa/g16

G17 — noise + ZNE

Depolarizing ladder from ideal down to 20× H2 rates. Raw gap error reaches 26% at 20×; Richardson quadratic ZNE recovers to 3.7%.

/nadarasa/g17

G18 — Laplacian TDA

C6 vs 2C3: 1-WL cannot tell them apart, but the measured propagator trace separates them at χ²/dof = 110.

/nadarasa/g18

G19 — Floquet validation

Reading note on the IBM/Quantinuum cross-platform Floquet result: noise-model → estimator agreement → classical check → cross-platform corroboration.

/nadarasa/g19

What ties the four together

G16 is the exact ideal baseline. G17 adds the noise ladder and asks how much of the signal survives. The ZNE step in G17 is the same extrapolation idea that QESEM uses in G19, just with a simpler depolarizing model. G18 adds the independent-estimator layer: a model-free curve check that does not assume a Taylor fit, analogous to QESEM-Unbiased vs QESEM-Extrapolated agreeing in the Floquet paper. The missing layer is cross-platform corroboration — Selene is the only platform currently in the wind tunnel.

Open frontiers for the next gate

PQP Frontier Step C-2q: completed — 90/90 cells PASS across 9 noise levels at H2-class rates; see /nadarasa/proofs/noise-2q.
Floquet native port: compile a small heavy-hex Floquet cycle in Guppy and run it on Selene to see whether the all-to-all topology keeps the subharmonic oscillation intact.
ADAPT-GQE composition: feed a generated transformer/RL circuit into the rule-(N/M/P) canonicalisation pipeline and verify the composed result before execution.