28.4. PROKHOROV AND LEVY THEOREMS 767

Lemma 28.4.3 If {µn}is a sequence of Borel probability measures defined on the Borelsets of Rpsuch that

limn→∞

φ µn(t) = ψ (t)

for all t, where ψ (0) = 1 and ψ is continuous at 0, then {µn}∞

n=1 is tight.

Proof: Let e j be the jth standard unit basis vector. Letting t= te j in the definition andu > 0 ∣∣∣∣1u

∫ u

−u

(1−φ µn

(te j))

dt∣∣∣∣= ∣∣∣∣1u

∫ u

−u

(1−

∫Rp

eitx j dµn (x))

dt∣∣∣∣ (28.8)

=

∣∣∣∣1u∫ u

−u

(∫Rp

(1− eitx j

)dµn (x)

)dt∣∣∣∣= ∣∣∣∣∫Rp

1u

∫ u

−u

(1− eitx j

)dtdµn (x)

∣∣∣∣=

∣∣∣∣2∫Rp

(1−

sin(ux j)

ux j

)dµn (x)

∣∣∣∣≥ 2∫[|x j|≥ 2

u ]

(1− 1∣∣ux j

∣∣)

dµn (x)

≥ 2∫[|x j|≥ 2

u ]

(1− 1

u(2/u)

)dµn (x) =

∫[|x j|≥ 2

u ]1dµn (x) = µn

([x :∣∣x j∣∣≥ 2

u

]).

If ε > 0 is given, there exists r > 0 such that if u ≤ r, 1u∫ u−u (1−ψ (te j))dt < ε/p for all

j = 1, · · · , p and so, by the dominated convergence theorem, the same is true with φ µnin place of ψ provided n is large enough, say n ≥ N (r). Thus, from 28.8, if n ≥ N (r),µn([x :∣∣x j∣∣> 2r

])< ε/p for all j ∈ {1, · · · , p}. It follows for n≥ N (r) ,

µn ([x : ∥x∥∞> 2r])< ε.

and so let Kε ≡ [−r,r]p. ■In the case of Rp, and µn,µ Borel probability measures, convergence of characteristic

functions yields something interesting for ψ ∈ G or S, the Schwartz class.

Lemma 28.4.4 If φ µn(t)→ φ µ (t) for all t, then whenever ψ ∈S, the Schwartz class,

µn (ψ)≡∫Rp

ψ (y)dµn (y)→∫Rp

ψ (y)dµ (y)≡ µ (ψ)

as n→ ∞.

Proof: By definition, φ µ (y) ≡∫Rp eiy·xdµ (x) . Also remember the inverse Fourier

transform. Letting ψ ∈S, the Schwartz class,

F−1 (µ)(ψ) ≡ µ(F−1

ψ)≡∫Rp

F−1ψdµ

=1

(2π)p/2

∫Rp

∫Rp

eiy·xψ (x)dxdµ (y)

=1

(2π)p/2

∫Rp

ψ (x)∫Rp

eiy·xdµ (y)dx =1

(2π)p/2

∫Rp

ψ (x)φ µ (x)dx

and so, considered as elements of S∗ or G ∗, F−1 (µ) = φ µ (·)(2π)−(p/2) ∈ L∞. By thedominated convergence theorem

(2π)p/2 F−1 (µn)(ψ) ≡∫Rp

φ µn(t)ψ (t)dt→

∫Rp

φ µ (t)ψ (t)dt

= (2π)p/2 F−1 (µ)(ψ)

28.4. PROKHOROV AND LEVY THEOREMS 767Lemma 28.4.3 /f {,, }is a sequence of Borel probability measures defined on the Borelsets of R? such thatlim 4, (t) = w(t)for all t, where yw (0) = 1 and y is continuous at 0, then {L,};-_, is tight.Proof: Let e; be the j'” standard unit basis vector. Letting t = re; in the definition and= a (1 4, (re;)) at) = EL (he ® dtFlr bile=P f. (1-52) ten] 22 fog (" 5 fin)22 fa (t- wiayuy) it 9 fjey Hal )=te([e:|ul2 fl).If € > 0 is given, there exists r > 0 such that if u < r,4 {“,(1—y/(te;)) dt < €/p for allj=1,--:,p and so, by the dominated convergence theorem, the same is true with @ Lt,in place of y provided n is large enough, say n > N(r). Thus, from 28.8, if n > N(r),Lt, ([x: |x| > 2r]) <e/p for all j € {1,--- , p}. It follows for n > N(r),(28.8)HM, ([a : lla]. > 2r]) <€and so let Ke = [—rr]”.In the case of R?, and p1,,, U Borel probability measures, convergence of characteristicfunctions yields something interesting for y € Y or G, the Schwartz class.Lemma 28.4.4 if u, (€) + 4 (t) for all t, then whenever y € 6, the Schwartz class,1W)= [van |) v(wano) =H)R?as Nn — ©,Proof: By definition, 9, (y) = Jp e'¥ du (x). Also remember the inverse Fouriertransform. Letting y € G, the Schwartz class,F'(u)(y) = “wm =[, F'wdune1 vap-ae 1Lo [yo ven te Lamanand so, considered as elements of G* or Y*, F~'(u) = Ou () (2m) (?/?) € L®. By thedominated convergence theoremny" F(u,)(v) = [ oy, wear [oy (t)w(eat= (20)? Fo! (uy (y)