10.2. THE GAMMA FUNCTION 231

Now apply L’Hopital’s rule to conclude that the limit of this expression is 0 as t → ∞. Thusthe quotient e−t tα−1

e−(t/2) is less than some constant C.

∫ R

1e−ttα−1dt ≤

∫ R

1Ce−(t/2)dt ≤ 2Ce(−1/2)−2Ce(−R/2) ≤ 2Ce(−1/2)

Thus these integrals also converge as R → ∞ because they are increasing in R and boundedabove. Hence they converge to sup

{∫ R1 e−ttα−1dt : R > 1

}. It follows that Γ(α) makes

sense.The argument also implies the following proposition. Absolute convergence implies

convergence.

Proposition 10.2.3 If f ≥ 0, then∫

a f (t)dt exists if the partial integrals∫ R

a f (t)dtare bounded above independent of R. Also

∫∞

a f (t)dt exists if∫

a | f (t)|dt exists.

Proof: The first part is just like what was done with the gamma function. As to thesecond part, consider f+ (t) ≡ | f (t)|+ f (t)

2 , f− (t) ≡ | f (t)|− f (t)2 . These are both nonnegative

and if∫

a | f |dt exists, then∫ R

af+dt ≤

∫∞

a| f |dt,

∫ R

af−dt ≤

∫∞

a| f |dt

and so the first part implies limR→∞

∫ Ra f+dt and limR→∞

∫ Ra f−dt both exist. Hence∫ R

af dt =

∫ R

af+dt −

∫ R

af−dt

also must have a limit as R → ∞.This gamma function has some fundamental properties described in the following pro-

position. In case the improper integral exists, we can obviously compute it in the form

limδ→0+

∫ 1/δ

δ

f (t)dt

which is used in what follows. Thus also the usual algebraic properties of the Riemannintegral are inherited by the improper integral.

Proposition 10.2.4 For n a positive integer, n! = Γ(n+1). In general, the followingidentity holds. Γ(1) = 1,Γ(α +1) = αΓ(α)

Proof: First of all, Γ(1) = limδ→0∫

δ−1

δe−tdt = limδ→0

(e−δ − e−(δ

−1))= 1. Next,

for α > 0,

Γ(α +1) = limδ→0

∫δ−1

δ

e−ttα dt = limδ→0

[−e−ttα |δ

−1

δ+α

∫δ−1

δ

e−ttα−1dt

]

= limδ→0

(e−δ

δα − e−(δ

−1)δ−α +α

∫δ−1

δ

e−ttα−1dt

)= αΓ(α)

10.2. THE GAMMA FUNCTION 231Now apply L’Hopital’s rule to conclude that the limit of this expression is 0 as tf > oo. Thus. 10-1,the quotient ae is less than some constant C.ii et* dt < [ocean < 2Ce'-!/2) — 2Ce-F/2) < 2Ce-1/)1 1Thus these integrals also converge as R — co because they are increasing in R and boundedabove. Hence they converge to sup { f x e't?-ldt:R> i} . It follows that [(@) makessense. §fThe argument also implies the following proposition. Absolute convergence impliesconvergence.Proposition 10.2.3 [f f > 0, then [* f (t)dt exists if the partial integrals [* f (t) dtare bounded above independent of R. Also |’ f (t) dt exists if J” |f (t)| dt exists.Proof: The first part is just like what was done with the gamma function. As to thesecond part, consider f, (t) = Forse | f-@= Or ro | These are both nonnegativeand if {7° | f|dt exists, then[pars Pinan [ras Pinaand so the first part implies limr_,.. fk f,dt and limp_sco fk f_dt both exist. Hence[ora [ rar [ratalso must have a limit as Roo. JThis gamma function has some fundamental properties described in the following pro-position. In case the improper integral exists, we can obviously compute it in the form1/6li t)dtsim J, 10which is used in what follows. Thus also the usual algebraic properties of the Riemannintegral are inherited by the improper integral.Proposition 10.2.4 For n a positive integer, n! = (n+1). In general, the followingidentity holds. T (1) = 1,0 (@+1) = al (a)Proof: First of all, P(1) = limg_4o f° e~tdt = lims_4o (e8 -e (@)) = 1. Next,for a > 0,-15! _ 5T(a+1)= lim | e't%dt = lim en ! +a cua60/5 5-0 56-0- o= lim Gua sea | ue ta = al (a)5