25.08.2013 Views

Hamzi - Eurandom

Hamzi - Eurandom

Hamzi - Eurandom

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Reproducing Kernel Hilbert Spaces<br />

• Mercer Theorem: Let (X , µ) be a finite-measure space, and suppose<br />

k ∈ L∞(X 2 ) is a symmetric real-valued function such that the integral<br />

operator<br />

Tk : L2(X ) → L2(X )<br />

∫<br />

f ↦→ (Tkf)(x) =<br />

k(x, x<br />

X<br />

′ )f(x ′ )dµ(x ′ )<br />

is positive definite; that is, for all f ∈ L2(X ), we have<br />

∫<br />

X 2 k(x, x ′ )f(x)f(x ′ )dµ(x)dµ(x ′ ) ≥ 0.<br />

Let Ψj ∈ L2(X ) be the normalized orthogonal eigenfunctions of Tk<br />

associated with the eigenvalues λj > 0, sorted in non-increasing order.<br />

Then<br />

i. (λj)j ∈ ℓ1,<br />

ii. k(x, x ′ ) = ∑ NX<br />

j=1 λjΨj(x)Ψj(x ′ ) holds for almost all (x, x ′ ). Either<br />

NX ∈ N, or NX = ∞; in the latter case, the series converges<br />

absolutely and uniformly for almost all (x, x ′ ).<br />

. . . . . .<br />

Boumediene <strong>Hamzi</strong> (Imperial College) On Control and RDS in RKHS June 4th, 2012 22 / 55

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!