n -2(R TAmR
r . A z
Op (n 2) Op (n 1) (O
„ ⇒
O (n 1) n 2Л O
^ pv ' m J ∖
(A.30)
V 1
r,m
J
where the latter result follows from (15). Q.E.D.
Proof of Lemma 5: By Chebishev inequality:
P(λU ≤ n∣Kqκm) ≥ 1
E (λ1,m )
Kκ
у α, q-r, m
E [trace( Г 1, m)]
(A.31)
α, q-r, m
Moreover, it follows easily from (23), by first conditioning on the X*k’s, that
m
E( V,,m ) = ∑ γ 2Jr
k= 1
j-1
A-1
*T
m
= ∑γ I
к-1
m
= Σ γ2I
к- 1
∑YjEtrace (1/m)£ X^X^τ W? I,
j= 1
Ik=1
(A.32)
m
q~,∖∙^ 2τ
γ-Σt7 k1r ,
m k= 1
where the second equality follows from fact that the X*k’s are i.i.d., hence it follows from (22)
that
(
E [trace( V,+1,m )] = 1
∖
ɪɪɪ A
mJ
J
(A
£y2 trace(RrTD(1)D(1)TRr+J.
<k=1 √
(A.33)
Q.E.D.
Proof of Lemma 6: It follows from Fourier analysis that we can write without loss of generality:
Fk(x ) =
У2 cjkexp(2iπjx), where cjk = Jexp(2iπjx)Fk(x)dx.
(A.34)
-∞<z' <∞
43