14.14. EXERCISES 353

and so, there is a Borel set of measure zero N such that if x /∈ N,

limn→∞

1m(B(0,rn))

∫B(x,rn)

| f (y)− f (x)|dm(y) = 0

and solimr→∞

1m(B(0,r))

∫B(x,r)

| f (y)− f (x)|dm(y) = 0

which implies for x /∈ N,

limr→0

1m(B(0,r))

∫B(x,r)

f (y)dm(y) = f (x)

27. If you only know that f X[a,b] ∈ L1 (R,m) for all finite intervals [a,b] , show exactlythe same conclusion holds.

28. Use the above result to show that if f X[a,b] ∈ L1 (R,m) for all finite intervals [a,b] ,then for a.e. x,

limh→0

1h

∫ x+h

0f (t)dm = f (x)

which can also be termed the Lebesgue fundamental theorem of calculus.

29. This problem outlines an approach to Stirling’s formula following [24] and [9]. Aneasier approach comes from the traditional Stirling’s formula and Problem 16 onPage 247 From the above problems, Γ(n+1) = n! for n ≥ 0. Consider more gen-erally Γ(x+1) for x > 0. Actually, we will always assume x > 1 since it is thelimit as x→ ∞ which is of interest. Γ(x+1) =

∫∞

0 e−ttxdt. Change variables let-ting t = x(1+u) to obtain Γ(x+1) = xx+1e−x ∫ ∞

−1 ((1+u)e−u)x du Next let h(u) be

such that h(0) = 1 and (1+u)e−u = exp(− u2

2 h(u))

Show the thing which works

is h(u) = 2u2 (u− ln(1+u)). Use L’Hospital’s rule to verify that the limit of h(u)

as u→ 0 is 1. The graph of h is illustrated in the following picture. Verify that itsgraph is like this, with an asymptote at u = −1 decreasing and equal to 1 at 0 andconverging to 0 as u→ ∞.

−1

1

Next change the variables again letting u = s√

2x . This yields, from the original

description of h, Γ(x+1) = xxe−x√

2x∫

−√

x/2exp(−s2h

(s√

2x

))ds. For s < 1,

h

(s

√2x

)> 2−2ln2 = 0.61371

so the above integrand is dominated by e−(2−2ln2)s2. Consider the integrand in the

above for s ≥ 1. The exponent part is −(√

2√

xs− x ln(

1+ s√

2x

)). Now for

14.14. EXERCISES 35327.28.29.and so, there is a Borel set of measure zero N such that if x ¢ N,; Ilim TBO) Ihnen Mf 0 ~F l4m(o) =0and so ' ;im 7 (B(0.r) bcs lf (y) — f (x)|dm(y) = 0which implies for x ¢ N,. 1lim aon) (B07) Irvun f (y)dm(y) = f (x)If you only know that f 2 jap) € L' (R,m) for all finite intervals [a,b], show exactlythe same conclusion holds.Use the above result to show that if f 2j,,) € L' (R,m) for all finite intervals [a,b] ,then for a.e. x,1 xthlim — t)dm=tim 5 [fam = F(a)which can also be termed the Lebesgue fundamental theorem of calculus.This problem outlines an approach to Stirling’s formula following [24] and [9]. Aneasier approach comes from the traditional Stirling’s formula and Problem 16 onPage 247 From the above problems, (n+ 1) =n! for n > 0. Consider more gen-erally [(x+1) for x > 0. Actually, we will always assume x > 1 since it is thelimit as x — oe which is of interest. T'(x+1) = fo e ‘t*dt. Change variables let-ting f= x(1+u) to obtain (x+1) =x le [™ ((1+u)e“)* du Next let h(u) besuch that #(0) = 1 and (1+u)e"" =exp (-$ (u)) Show the thing which worksis h(u) = 5 (u—In(1+u)). Use L’Hospital’s rule to verify that the limit of A (u)as u—> 0 is 1. The graph of h is illustrated in the following picture. Verify that itsgraph is like this, with an asymptote at u = —1 decreasing and equal to | at 0 andconverging to 0 as u > ©,-1 |Next change the variables again letting u = s\/2. This yields, from the originaldescription of h, T(x+1) = we SV 2x [™ payexp (<n (s\/2)) ds. Fors < 1,h (+\/2) >2—2In2 = 0.61371x. : . 2so the above integrand is dominated by e~(2~2!n2)sabove for s > 1. The exponent part is — (v2vie—in (1 +s?) . Now for. Consider the integrand in the