The prior on is a probability measure supported on
ProvedBosonicLearn.priorMu_probThe prior of the lower-bound argument is well defined.
Let be the measure on the space of traceless Hermitian matrices with density
with respect to Lebesgue measure, being the total mass of the unnormalized density. Then , so is a probability measure, and -almost every lies in .
Why it holds. is compact with non-empty interior in and the density is continuous, positive and at most on it.
Role. Averaging the success probability of a learner over uses that has total mass and that almost every gives a state of the hard family (proof of Theorem 5).
Formalization Note priorMu n is (m univ)⁻¹ • m with m the restriction of Measure.addHaar on Herm0 n to Gsupp n with the Gaussian density; the statement is IsProbabilityMeasure (priorMu n) and ∀ᵐ X ∂(priorMu n), X ∈ Gsupp n. It holds for every , including where .
import Mathlib import Definitions.Def_BosonicLearn_priorMu open MeasureTheory open scoped ENNReal
namespace BosonicLearn
/-- **The prior `µ` is a probability measure supported on `G_supp`.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, §5.1, Eq. (32), p. 30: "Sample X ∼ GUE*(n) conditioned on X ∈ G_supp. … The density of X with respect to the
Lebesgue measure on Herm_0(n) is given by µ(X) = Z_µ^{−1} exp(−(n/4)‖X‖_F²) 𝟙(‖X‖_op ≤ 4)."
`priorMu n` is a probability measure on `Herm₀(n)`, and `X ∈ G_supp` for `µ`-almost every `X`.
**Formalization Note.** `priorMu n` is defined by normalizing the measure with density
`exp(−(n/4)‖X‖_F²)` on `G_supp` against a Haar measure; the statement says the normalizing constant `Z_µ` is
finite and positive (`G_supp` is compact with non-empty interior in `Herm₀(n)`), and that the measure is
concentrated on `G_supp`. -/
theorem priorMu_prob (n : ℕ) :
IsProbabilityMeasure (priorMu n) ∧ ∀ᵐ X ∂(priorMu n), X ∈ Gsupp n := by sorry
end BosonicLearn