24.02.2013 Views

Optimality

Optimality

Optimality

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

142 D. M. Dabrowska<br />

where λ =−1,<br />

ηθ(t) =<br />

� τ<br />

0<br />

Kθ(t, u)ρ − ˙ Γ (u, θ)EN(du),<br />

ρ−Γ ˙ (u, θ) = v(u, θ) ˙ Γθ(u)+ρ(u, θ) and Bθ is given by (2.4). For fixed θ, the kernel Kθ<br />

is symmetric, positive definite and square integrable with respect to Bθ. Therefore it<br />

can have only positive eigenvalues. For λ =−1, the equation has a unique solution<br />

given by<br />

(2.9) ψθ(t) = ηθ(u)−<br />

� τ<br />

0<br />

∆θ(t, u,−1)ηθ(u)Bθ(du),<br />

where ∆θ(t, u, λ) is the resolvent corresponding to the kernel Kθ. By definition, the<br />

resolvent satisfies a pair of integral equations<br />

Kθ(t, u) = ∆θ(t, u, λ)−λ<br />

= ∆θ(t, u, λ)−λ<br />

� τ<br />

0<br />

� τ<br />

0<br />

∆θ(t, w, λ)Bθ(dw)Kθ(w, u)<br />

Kθ(t, w)Bθ(dw)∆θ(w, u, λ),<br />

where integration is with respect to different variables in the two equations. For<br />

λ =−1 the solution to the equation is given by<br />

ψθ(t) =<br />

� τ<br />

0<br />

� τ<br />

−<br />

Kθ(t, u)ρ − ˙ Γ (u, θ)EN(du)<br />

0<br />

∆θ(t, w,−1)Bθ(dw)<br />

� τ<br />

0<br />

Kθ(w, u)ρ − ˙ Γ (u, θ)EN(du)<br />

and the resolvent equations imply that the right-hand side is equal to<br />

(2.10) ψθ(t) =<br />

� τ<br />

0<br />

∆θ(t, u,−1)ρ − ˙ Γ (u, θ)EN(du).<br />

For θ = θ0, substitution of this expression into the formula for the matrices<br />

Σ1,ϕ(θ0, τ) and Σ2,ϕ(θ0, τ) and application of the resolvent equations yields also<br />

Σ1,ϕ(θ0, τ) = Σ2,ϕ(θ0, τ)<br />

=<br />

� τ<br />

v − ˙ Γ (u, θ0)EN(du)<br />

0<br />

� τ � τ<br />

−<br />

0<br />

0<br />

∆θ0(t, u,−1)ρ − ˙ Γ (u, θ0)ρ − ˙ Γ (t, θ0) T EN(du)EN(dt).<br />

It remains to find the resolvent ∆θ. We shall consider first the case of θ = θ0.<br />

To simplify algebra, we multiply both sides of the equation (2.8) byPθ0(0, t) −1 =<br />

exp � t<br />

0 s′ (θ0,Γθ0(u), u)Cθ0(du). For this purpose set<br />

˜ψ(t) =Pθ0(0, t) −1 ψ(t),<br />

˙ G(t) =Pθ0(0, t) −1 ˙ Γθ0(t),<br />

˜v(t, θ0) = v(t, θ0)Pθ0(0, t) 2 , ˜ρ − ˙ G (t, θ0) =Pθ0(0, t)ρ − ˙ Γ (t, θ0),<br />

b(t) =<br />

� t<br />

0<br />

˜v(u, θ0)dEN(u), c(t) =<br />

Multiplication of (2.8) byPθ0(0, t) −1 yields<br />

(2.11)<br />

˜ ψ(t) +<br />

� τ<br />

0<br />

k(t, u) ˜ ψ(u)b(du) =<br />

� τ<br />

0<br />

� t<br />

0<br />

Pθ0(0, u) −2 dCθ0(u).<br />

k(t, u)˜ρ − ˙ G (u, θ0)EN(du),

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

Saved successfully!

Ooh no, something went wrong!