p. 4 — the rate function is the Legendre transform
ProvedKWSpectrum.Estimator.legendrelarge-deviationslegendre-transformp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1relative-entropy
Let and let be the rate function (13), with and when for some . Then
the supremum being taken in over all .
This identifies the rate function produced by the Gärtner–Ellis theorem from in (25) with the relative entropy of with respect to .
Formalization Note The supremum is an EReal supremum; for , so the real logarithm is the true one.
Differences from the source: none.
Preamble
import Mathlib import Definitions.Def_KWSpectrum_Estimator_Model
Formal statement
namespace KWSpectrum.Estimator
open QAlgorithms QAlgorithms.Schur
/-- Keyl–Werner, arXiv:quant-ph/0102027v1, p. 4: the rate function (13) is the Legendre
transform of `c(η) = ln Σ_α r_α e^{η_α}`: `I(s) = sup_{η ∈ ℝ^d} (η · s − c(η))` for `r, s ∈ Σ`,
in `[0, ∞]`. -/
theorem legendre {d : ℕ} (r s : Fin d → ℝ) (hr : r ∈ SpectrumSet d) (hs : s ∈ SpectrumSet d) :
rate r s = ⨆ η : Fin d → ℝ,
(((∑ j : Fin d, η j * s j) - Real.log (∑ α : Fin d, r α * Real.exp (η α)) : ℝ) : EReal) := by sorry
end KWSpectrum.Estimator
Source
Keyl & Werner, Estimating the spectrum of a density operator, arXiv:quant-ph/0102027v1, p. 4 (sentence after eq. (25)), with eq. (13), p. 3
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.