Published July 26, 2026 | Version v2

Compact physical Gaussian-ket models for homodyne quantum-state tomography

Authors/Creators

Description

  Continuous-variable quantum-state tomography usually reconstructs a truncated
  Fock-basis density matrix. We study a compact alternative: finite mixtures of
  displaced, squeezed Gaussian kets whose ρ = BB† construction is positive
  semidefinite by construction, fitted by per-sample homodyne likelihood with
  closed-form gradients, and composed with physical loss and noise channels. On
  the public propagating-light GKP dataset of Konno et al. (Science 2024), a
  rank-4 model with 92 real parameters matches the empirical full-rank
  maximum-likelihood frontier (255 parameters) at confidence-interval resolution
  on held-out likelihood. On a synthetic thermal-noise target that we prove lies
  outside the model family — no detection efficiency and no finite rank
  reproduces it exactly — the channel-composed model fitted blind exceeds a
  full-rank MLE run under the pre-declared 900-second baseline budget, a verdict
  that holds across all five pre-declared seed and noise configurations. We do
  not claim a universally superior method: comparisons on real data reuse
  observations across splits, the strongest baseline is test-selected, and the
  blind result covers one target class. The contribution is a compact,
  physically constrained model family together with a fully falsification-first
  research record — negative results, superseded scorings, and pre-declared
  protocols are all preserved in the accompanying repository.

Notes

v2 corrects two items in the body text, neither affecting the abstract or any headline result: (1) the best-approximation residuals in §3.3 are now quoted at the largest scored cutoff (≈2–3×10⁻³) and the "cutoff-stable" claim is withdrawn, as the series is still increasing; (2) §4 (C4) now cites the prior work defining the Gaussian rank and related measures (refs 16–18) and claims no novelty for the measure.

Files

preprint.pdf

Files (357.5 kB)

Name Size
md5:02579a7f110be7cb76e11126c9ca67e0
357.5 kB Preview Download

Additional details

Related works

Cites
Dataset: 10.5061/dryad.t76hdr86j (DOI)
Is supplemented by
Software: https://github.com/orangewk/wigner-splat (URL)
Software: 10.5281/zenodo.21387212 (DOI)