zetesis-puremath Blueprint

6.1 Bounded kernels

Definition 6.1 Bounded kernel

Lean: BoundedKernel

A BoundedKernel \(\mathcal{H}\) is a witness \(\langle C, h_{\ge 0}, h_{\le } \rangle \) with \(C \ge 0\) and \(\| k_x \| \le C\) for all \(x : X\). Bounded-feature kernels are the canonical setting for kernel mean embeddings [ 21 , § 4.2 ] .

Theorem 6.2 Evaluation bound

Lean: norm_apply_le_bound_mul_norm

\(|f(x)| \le C \cdot \| f\| \) for every \(f \in \mathcal{H}\) and \(x \in X\) [ 21 , Thm 4.23 ] .

Theorem 6.3 Reproducing identity

Lean: rkhs_eval_eq_inner

\(f(x) = \langle k_x,\, f \rangle \) for every \(f \in \mathcal{H}\) and \(x \in X\) [ 1 ] .

Theorem 6.4 Continuity of RKHS evaluations

Lean: continuous_rkhs_apply

If the feature map \(x \mapsto k_x\) is continuous, then every \(f \in \mathcal{H}\) is continuous on \(X\) [ 21 , Lemma 4.29 ] .

Theorem 6.5 RKHS functions are integrable under a finite measure

Lean: rkhs_fun_integrable

Under IsFiniteMeasure \(P\) and a continuous feature map, every \(f \in \mathcal{H}\) is \(P\)-integrable (via \(L^\infty \subset L^1\) on finite measures).