§3, Theorem — reproducing kernels of finite-dimensional classes
ProvedAronszajnRK.Sum.finite_dimensional_classesLet be a set. A function is the reproducing kernel of a finite-dimensional complex Hilbert space of functions on if and only if it has the form
for some , a positive definite Hermitian matrix and linearly independent functions on . In that case the class with kernel is finite-dimensional and generated by the : its functions are exactly
and the norm of such an is
where is the inverse of the matrix .
The theorem describes completely the kernels of finite-dimensional spaces: is the Gram matrix of the generating functions.
Formalization Note "Positive definite" is Mathlib's Matrix.PosDef over ℂ (Hermitian with positive quadratic form); is the entrywise conjugate β.map star, not the conjugate transpose. The "only if" direction is stated for every finite-dimensional RKHS; the "if" direction asserts existence of a finite-dimensional RKHS with kernel (6) and describes every RKHS with that kernel. The value is compared as a complex number with the (real) right-hand side. Checked by hand at : .
import Mathlib import Definitions.Def_AronszajnRK_Sum_kernelFn open scoped ComplexOrder
namespace AronszajnRK.Sum
universe u v
/-- **Reproducing kernels of finite-dimensional classes** (Aronszajn, *Theory of Reproducing
Kernels*, Trans. Amer. Math. Soc. 68 (1950), §3, Theorem, p. 347 (PDF 11), with §3 (1), (2), (6),
p. 346 (PDF 10)). A function `K(x, y)` is the reproducing kernel of a finite-dimensional class of
functions if and only if it is of the form (6) `K(x, y) = ∑ᵢⱼ βᵢⱼ wᵢ(x) \overline{wⱼ(y)}` with a
positive definite matrix `{βᵢⱼ}` and linearly independent functions `wₖ(x)`. The corresponding class
`F` is then generated by the functions `wₖ(x)`, the functions `f ∈ F` given by (1)
`f = ∑ ζₖ wₖ`, and the corresponding norm given by (2) `‖f‖² = ∑ᵢⱼ αᵢⱼ ζᵢ ζ̄ⱼ`, where `{αᵢⱼ}` is
the inverse matrix of `{β̄ᵢⱼ}` (entrywise conjugate, no transpose). -/
theorem finite_dimensional_classes {X : Type u} :
(∀ (H : Type v) [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H]
[RKHS ℂ H X ℂ], FiniteDimensional ℂ H →
∃ (n : ℕ) (β : Matrix (Fin n) (Fin n) ℂ) (w : Fin n → X → ℂ),
β.PosDef ∧ LinearIndependent ℂ w ∧
kernelFn H = fun x y => ∑ i, ∑ j, β i j * w i x * star (w j y)) ∧
∀ (n : ℕ) (β : Matrix (Fin n) (Fin n) ℂ) (w : Fin n → X → ℂ),
β.PosDef → LinearIndependent ℂ w →
(∃ (H : Type u) (_ : NormedAddCommGroup H) (_ : InnerProductSpace ℂ H)
(_ : CompleteSpace H) (_ : RKHS ℂ H X ℂ), FiniteDimensional ℂ H ∧
kernelFn H = fun x y => ∑ i, ∑ j, β i j * w i x * star (w j y)) ∧
∀ (H : Type v) [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H]
[RKHS ℂ H X ℂ],
kernelFn H = (fun x y => ∑ i, ∑ j, β i j * w i x * star (w j y)) →
FiniteDimensional ℂ H ∧
Set.range (fun f : H => (f : X → ℂ)) =
{g : X → ℂ | ∃ ζ : Fin n → ℂ, g = ∑ k, ζ k • w k} ∧
∀ (f : H) (ζ : Fin n → ℂ), (f : X → ℂ) = ∑ k, ζ k • w k →
((‖f‖ ^ 2 : ℝ) : ℂ) =
∑ i, ∑ j, (β.map star)⁻¹ i j * ζ i * star (ζ j) := by sorry
end AronszajnRK.Sum
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.