Chapter 28

The Normal DistributionThis particular distribution is likely the most important one in statistics and it will be essen-tial to understand in developing the Wiener process later. To begin with, 1√

∫R e−

12 u2

du =

1 as is easily shown as done earlier by the standard calculus trick of

I =∫R

e−12 u2

du, I2 =∫R

∫R

e−12 (u2+v2)dudv

and then changing to polar coordinates to obtain I2 = 2π . I will use this identity wheneverconvenient. Also useful is the following lemma.

Lemma 28.0.1 1√2π

∫R e−

12 (u−it)2

du = 1.

Proof: e−12 (u−it)2

= e−12 (u2−2itu−t2) = e−

12 u2

e12 t2

(cos(tu)+ isin(tu)) and so, the inte-gral equals

1√2π

∫R

e−12 u2

e12 t2

cos(tu)du

Now let f (t)≡ 1√2π

∫R e−

12 u2

e12 t2

cos(tu)du. Using the dominated convergence theorem,

f ′ (t) =1√2π

∫R

ddt

(e−

12 u2

e12 t2

cos(tu))

du

=1√2π

∫R

(e−

12 u2(

te12 t2

cos(tu)− e12 t2

usin(tu)))

du

Now f (0) is known to be 1. Assume then that t ̸= 0.

− 1√2π

e12 t2∫R

e−12 u2

usin(tu)du =1√2π

e12 t2∫R

e−12 u2

t cos(tu)du

and this shows that f ′ (t) = 0 so f (t) is the constant 1. ■

28.1 The Multivariate Normal DistributionThe multivariate normal distribution is very important in statistics and it will be shown inthis chapter why this is the case.

Definition 28.1.1 A random vectorX,with values inRp has a multivariate normaldistribution written asX ∼ Np (m,Σ) if for all Borel E ⊆ Rp,

λX (E) =∫Rp

XE (x)1

(2π)p/2 det(Σ)1/2 e−12 (x−m)∗Σ−1(x−m)dx

for µ a given vector and Σ a given positive definite symmetric matrix, called the covariancematrix. In case p = 1, this is called the variance.

Theorem 28.1.2 ForX ∼ Np (m,Σ) ,m= E (X) and

Σ = E((X−m)(X−m)∗

).

759

Chapter 28The Normal DistributionThis particular distribution is likely the most important one in statistics and it will be essen-: : . . : : 12tial to understand in developing the Wiener process later. To begin with, Ta Spe 2" du=1 as is easily shown as done earlier by the standard calculus trick ofT= [ en?" du, P. = [ [ eo 20°") dudyR JRIRand then changing to polar coordinates to obtain /* = 27. I will use this identity wheneverconvenient. Also useful is the following lemma.Lemma 28.0.1 <= Jz eit) dy = 1,Proof: = 2("-i#)” = e~3(w—2itu-#) — pp? p30? (cos (tu) +isin(tu)) and so, the inte-gral equals1 12 12—= | e 2" e2" cos(tu) duaah1,2 1,2 . :Now let f (t) = Jpe 2“ e2" cos (tu) du. Using the dominated convergence theorem,ala1 od 12 1,2/ —4w Att) = — L(e% e2 cos (tu) ) dufo ha (ru)1 12 12 1,2= —— e 2" (te* cos (tu) — e2" usin (tu )) du= I (tu) (tu)Now f (0) is known to be 1. Assume then that t 4 0.1 ap 12... 1 ip _12e2” | e 2 usin(tu)du= e2! fe 2“ tcos (tu) duRV2n R V2nand this shows that f’ (t) =0 so f(t) is the constant 1. Hl28.1 The Multivariate Normal DistributionThe multivariate normal distribution is very important in statistics and it will be shown inthis chapter why this is the case.Definition 28.1.1 4 random vector X , with values in R? has a multivariate normaldistribution written as X ~ N,(m,~) if for all Borel E C R?,1 =1 kylAx (E)= By Pn (em) "EV (@—m)x(E) IRP e(#) (2m)? det (Z) 2" *for La given vector and Xa given positive definite symmetric matrix, called the covariancematrix. In case p = 1, this is called the variance.Theorem 28.1.2 for X ~ N,(m,2),m = E(X) andL=E((X—m)(X—m)").759