24.02.2013 Views

Optimality

Optimality

Optimality

SHOW MORE
SHOW LESS

Create successful ePaper yourself

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

Optimal sampling strategies 285<br />

Since δi > 0∀i, for the concavity of δi it suffices to show that<br />

�<br />

�<br />

1−βzi<br />

(7.22)<br />

κi+2 − κi+1 ≤ 0.<br />

1−βzi+2<br />

Now<br />

(7.23)<br />

1−αzi<br />

1−αzi+2<br />

− 1−βzi<br />

1−βzi+2<br />

= (α−β)(zi+2− zi)<br />

(1−αzi+2)(1−βzi+2)<br />

≥ 0.<br />

Then (7.20) and (7.23) combined with the fact that κi≥ 0,∀i proves (7.22). �<br />

Proof of Theorem 3.1. We split the theorem into three claims.<br />

Claim (1): L ∗ :=∪kL (k) (x ∗ k )∈Lγ(n).<br />

From (3.10), (3.11), and (3.13) we obtain<br />

(7.24)<br />

µγ(n) + Pγ− 1<br />

var(Vγ)<br />

E(Vγ|Lγk)<br />

L∈Λγ(n)<br />

k=1<br />

−1<br />

= max<br />

Pγ �<br />

≤ max<br />

Pγ �<br />

X∈∆n(Nγ1,...,NγPγ )<br />

µγ,γk(xk).<br />

k=1<br />

Clearly L ∗ ∈ Λγ(n). We then have from (3.10) and (3.11) that<br />

(7.25)<br />

µγ(n) + Pγ− 1<br />

var(Vγ) ≥ E(Vγ|L ∗ ) −1 + Pγ− 1<br />

var(Vγ) =<br />

=<br />

Pγ �<br />

k=1<br />

Thus from (7.25) and (7.26) we have<br />

(7.26) µγ(n) =E(Vγ|L ∗ ) −1 = max<br />

which proves Claim (1).<br />

Pγ �<br />

k=1<br />

µγ,γk(x ∗ k) = max<br />

X∈∆n(Nγ1,...,NγPγ )<br />

k=1<br />

E(Vγ|L ∗ γk) −1<br />

Pγ �<br />

X∈∆n(Nγ1,...,NγPγ )<br />

µγ,γk(xk).<br />

k=1<br />

Pγ �<br />

µγ,γk(xk)− Pγ− 1<br />

var(Vγ) ,<br />

Claim (2): If L∈Lγk(n) then L∈Lγ,γk(n) and vice versa.<br />

Denote an arbitrary leaf node of the tree of γk as E. Then Vγ, Vγk, and E are<br />

related through<br />

(7.27) Vγk = ϱγkVγ + Wγk,<br />

and<br />

(7.28) E = ηVγk + F<br />

where η and ϱγk are scalars and Wγk, F and Vγ are independent random variables.<br />

We note that by definition var(Vγ) > 0∀γ (see Definition 2.5). From Lemma 7.1

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

Saved successfully!

Ooh no, something went wrong!