The Moore–Aronszajn theorem
ProvedFamousTheorems.moore_aronszajn_theoremThe Moore–Aronszajn theorem. Let be a positive semidefinite kernel on a set , with values in the bounded operators on a Hilbert space . Then there is a reproducing kernel Hilbert space of functions whose reproducing kernel is .
Moore (1916) and Aronszajn (1950) established this correspondence between positive semidefinite kernels and reproducing kernel Hilbert spaces, including uniqueness of the space. It is the foundation of kernel methods in machine learning, of Gaussian process regression, and of the theory of spaces of analytic functions such as the Hardy and Bergman spaces.
Formalization note. Mathlib's RKHS.OfKernel.kernel_ofKernel, with witness RKHS.OfKernel K, the completion of the pre-Hilbert space spanned by kernel functions. RKHS 𝕜 H X V says that embeds continuously and injectively into functions , so point evaluations are continuous. RKHS.kernel H is the matrix of operators built from the kernel functions. K.PosSemidef is positive semidefiniteness of the operator-valued matrix K : Matrix X X (V →L[𝕜] V). Universe levels are explicit because the statement quantifies over the type .
import Mathlib
namespace FamousTheorems
universe u v w
theorem moore_aronszajn_theorem {𝕜 : Type u} [RCLike 𝕜] {X : Type v} {V : Type w} [NormedAddCommGroup V] [InnerProductSpace 𝕜 V]
[CompleteSpace V] (K : Matrix X X (V →L[𝕜] V)) (hK : K.PosSemidef) :
∃ (H : Type (max v w u)) (_ : NormedAddCommGroup H) (_ : InnerProductSpace 𝕜 H) (_ : CompleteSpace H)
(_ : RKHS 𝕜 H X V), RKHS.kernel H = K := by sorry
end FamousTheorems