CencovPetz.Simplex #
The strictly positive probability simplex on a finite type, together with its canonical tangent submodule (sum-zero functions) and the classical Fisher information bilinear form.
This is groundwork for the finite/discrete Čencov (Chentsov) uniqueness story.
Main definitions #
CencovPetz.Simplex α: strictly positive distributionsp : α → ℝwith∑ p = 1.CencovPetz.tangentSpace α: the submodule{u : α → ℝ | ∑ u = 0}.CencovPetz.fisherBilin p: the Fisher bilinear form ontangentSpace α,∑ uᵢ vᵢ / pᵢ.
Main results #
CencovPetz.fisherBilin_comm: symmetry of the Fisher bilinear form.CencovPetz.fisherBilin_pos: positive-definiteness on nonzero tangent vectors.
@[reducible, inline]
Tangent space to the simplex, modeled as the sum-zero submodule of α → ℝ.
Equations
Instances For
The Fisher bilinear form on the simplex tangent space:
⟪u,v⟫_p = ∑ a, u a * v a / p a.
Equations
- LeanPool.CencovPetz.fisherBilin p = LinearMap.mk₂ ℝ (fun (u v : ↥(LeanPool.CencovPetz.tangentSpace α)) => ∑ a : α, ↑u a * ↑v a / p.p a) ⋯ ⋯ ⋯ ⋯
Instances For
@[simp]
theorem
LeanPool.CencovPetz.fisherBilin.apply
{α : Type u_1}
[Fintype α]
(p : Simplex α)
(u v : ↥(tangentSpace α))
:
theorem
LeanPool.CencovPetz.fisherBilin.comm
{α : Type u_1}
[Fintype α]
(p : Simplex α)
(u v : ↥(tangentSpace α))
:
theorem
LeanPool.CencovPetz.fisherBilin.pos
{α : Type u_1}
[Fintype α]
(p : Simplex α)
(u : ↥(tangentSpace α))
(hu : u ≠ 0)
: