PRACTICE PROBLEMS

Complete any six problems in 3 hours. Please do not work in groups or refer to your notes. After
the time limit has passed, try and solve the other problems as well. These problems will not be
graded.
Problem 1. Let us define the function Γ : R+ → R by the integral
Z ∞
Γ(t) = xt−1 e−x dx.
0
This function is usually called the gamma function.
(1) Show that the integral for Γ(t) is well-defined for each 0 < t < ∞ and that it is infinitely
differentiable there.
(2) Show that for each positive integer n, Γ(n + 1) = n! = n(n − 1)(n − 2) · · · 2 · 1.
Solution. The argument to the integral is non-negative so it suffices to show that the integral is
finite. Splitting the integral, we have
Z 1 Z 1
t−1 −x
x e dx ≤ xt−1 dx = t−1 < ∞
0 0
If t ≤ 1, then we have Z ∞ Z ∞
t−1 −x
x e dx ≤ e−x dx = e−1 < ∞.
1 1
Otherwise, we integrate by parts btc times to get
Z ∞ Z ∞
t−1 −x
x e dx = const. + (t − 1) · · · (t − btc) xt−btc−1 e−x dx < ∞
1 1
by the previous case.
We now claim that Z ∞
Γ (n)
(t) = (log x)n xt−1 e−x dx,
0
which is absolutely integrable by an analogous argument. We prove this by induction on n, noting
that n = 0 is trivial. Then we have
Z ∞
(n)
Γ (t) = lim (log x)n−1 xt−1 e−x ∆(h log x) log x dx,
h→0 0
where
ey − 1
∆(y) =.
y
∆ is a nice function: it is continuous (in fact, analytic) on the whole real line, ∆(0) = 1, and is
monotone increasing. This last fact can be verified by computing its derivative, which is
yey − (ey − 1)
∆0 (y) = .
y2
We want to show that this quantity is always non-negative, or equivalently that
yey − ey − 1 ≥ 0
Replacing y with −y and rearranging it suffices to show that
1 + y ≤ ey
which follows since ey is convex and 1 + y is the tangent line to ey at y = 0.
Now we can complete the proof. It suffices to consider the integral over (0, 1) and (1, ∞), since
if the limit exists for each integral than it exists for the sum. We can also consider positive and
negative hs separately. For either case, we have a monotone sequence of functions which are either
2 PRACTICE PROBLEMS

non-negative or non-positive, which converges pointwise to the desired limit, so the result follows
from the monotone convergence theorem.
Next we compute Γ(n + 1). We have
Z ∞ Z −1
n −t
Γ(n + 1) = t e dt = lim tn e−t dt
0 →0 
by the monotone convergence theorem. The latter integral can be evaluated by parts, so that
Z −1 Z −1
n −t n − −n −−1
t e dt = (− e +  e )+n tn−1 e−t dt.
 
If we send  → 0, then the first term converges to zero (as n is fixed) while the second term converges
to nΓ(n), again by the monotone convergence theorem. The result now follows by induction and the
base case Γ(1) = 1, which can be verified directly.
Problem 2. Fix some δ ∈ R and let f : [0, ∞) → R be given by the equation
sin(x2 ) δx
f (x) = + .
x 1+x
Show that Z a
lim f (nx) dx = aδ
n→∞ 0
for each a > 0.
Solution. Consider the sequence of functions fn (x) = f (nx). For each fixed 0 ≤ x ≤ a, fn (x) =
f (nx) → δ as n → ∞. Furthermore, the sequence is uniformly bounded by some M > 0. Since we
are working on a finite interval [0, a] we have
Z a Z a Z a
lim f (nx) dx = lim fn (x) dx = lim fn (x) dx = δa
n→∞ 0 n→∞ 0 0 n→∞

by the dominated convergence theorem, with dominating functions g(x) = M 10≤x≤a .

Problem 3. Show that
n Z n Z ∞
 X 1 x n
lim log n − = lim 1− log x dx = e−x log x dx.
n→∞
k=1
k n→∞ 0 n 0

Solution. Let fn (x) = (1 − xn−1 )n 10≤x≤n . Then 0 ≤ fn (x) and fn (x) ≤ e−x by the convexity of e−x .
By a theorem of Euler we have fn (x) → e−x for each x, so since
Z ∞
e−x log dx < ∞
0
(say, by bounding the integral on (0, 1) by the integral of log and the integral on (1, ∞) by integrating
by parts) we have by the dominated convergence theorem
Z ∞ Z ∞
lim fn (x) log x dx = e−x log x dx.
n→∞ 0 0
Now we evaluate the sequence of integrals. We have, using the substitution u = 1 − xn−1 ,
Z n Z 1
−1 n
(1 − xn ) log x dx = n un log(n(1 − u)) du
0 0
Of course log(n(1 − u)) = log n + log(1 − u) so the above integral is equal to
Z 1 Z 1 Z 1
n n n
n u log n du + n u log(1 − u) = log n + n un log(1 − u).
0 0 n+1 0
Let us study the latter integral. The integrand is bounded by log(1−u) which is integrable. log(1−u)
has the power series expansion
∞
X uk
log(1 − u) = −
k=1
k
 PRACTICE PROBLEMS 3

which converges uniformly in compact subsets of (−1, 1). Furthermore, the partial sums of the series
are monotone decreasing in k, so by the monotone convergence theorem
Z 1 ∞ Z ∞
n
X uk+n X 1
u log(1 − u) = − du = − .
0 k=1
k k=1
k(n + k + 1)
We observe that  
1 1 1 1
= −
k(n + k + 1) n+1 k n+k+1
so that the series is telescoping and
Z 1 n+1
n 1 X1
u log(1 − u) = − .
0 n + 1 k=1 k
We have therefore shown that
Z n n+1
x n n n X1
1− log x dx = log n +
0 n n+1 n + 1 k=1 n

The right hand side differs from the sequence log n − nk=1 k −1 by a quantity bounded by a constant
P
times n−1 log n, so the result follows.
Problem 4. Let φ ∈ L1 ([0, 1]). Define the function f : R → R by the integral
Z 1
f (t) = |φ(x) − t| dx.
0

(1) Show that f is a continuous function.

(2) Show that if m({φ = t}) = 0 for all t in an interval then f is continuously differentiable on
that interval.
(3) Show that if f is continuously differentiable on an interval then m({φ = t}) = 0 for all t in
the interval.
Solution. (1) We shall in fact show that f is 1-Lipschitz. For, for any s and t ∈ R,
Z 1 Z 1
|f (t) − f (s)| ≤ ||φ(x) − t| − |φ(x) − s|| dx ≤ |t − s| dx ≤ |t − s|
0 0

by the triangle inequality, and the result follows.

(2) Fix t in the interior of the interval. We claim that if Et = {x | φ(x) > t} then f 0 (t) =
2m(Et ) − 1; Note that the above is a continuous function by the monotone convergence
theorem for measures and the hypothesis that m(φ = t) = 0.
Let δ > 0 and split the integral for f for s near t, so that
Z Z Z
f (s) = + + |φ(x) − s| dx.
φ≤t−δ t−δ<φ≤t+δ t+δ<φ

Now let us consider difference quotients for f with h < δ. We have

f (t + h) − f (t)
=A+B+C
h
where
A = −m(φ ≤ t − δ), C = m(φ > t + δ),
and
|φ(x) − t − h| − |φ(x) − t|
Z
B= dx.
t−δ<φ≤t+δ h
The latter integrand is bounded by 1 by the triangle inequality, so the integral is bounded
by m(t − δ < φ ≤ t + δ). If we send δ → 0, then A converges to −1 + m(Et ), C converges to
m(Et ), and B converges to zero, all by the monotone convergence theorem for measures.
4 PRACTICE PROBLEMS

(3) We now consider the three limits without hypothesis on f . A and C converge as before
(without hypothesis) by the monotone convergence theorem. For B, we further split the
integral
|φ(x) − t − h| − |φ(x) − t| |φ(x) − t − h| − |φ(x) − t| |φ(x) − t − h| − |φ(
Z Z Z
B= dx+ dx+
t−δ<φ<t h φ=t h t<φ<t+δ h
By the dominated convergence theorem, the first and third of these integrals converge to zero
as δ → 0. Thus we see that the limit f 0 (t) exists if and only if
|φ(x) − t − h| − |φ(x) − t|
Z
lim dx
h→0 φ=t h
exists. However, the integrand is identically equal to |h|h−1 = sign h, so the limit exists if
and only if m(φ = t) = 0.
Problem 5. Suppose f : X → (0, ∞) is a measurable function so that
Z Z
1
f (x) dµ(x) < ∞ and dµ(x) < ∞.
X X f (x)
Show that µ(X) < ∞.
p
Solution. We have 1 = f (x)f (x)−1 so
Z Z p sZ sZ
1 1
µ(X) = dµ(x) ≤ f (x) p dµ(x) ≤ f (x) dµ(x) dµ(x) < ∞,
X X f (x) X X f (x)

by Cauchy-Schwarz.
Problem 6. (1) Suppose µ(X) < ∞. Show that if f ∈ Lq (X, µ) for some 1 ≤ q < ∞ then
r
f ∈ L (X, µ) for 1 ≤ r ≤ q.
(2) Give an example of a measure space (X, µ) with µ(X) = ∞ and a function f ∈ L2 (X, µ)
where f ∈/ L1 (X, µ).
R
Solution. (1) Let f : X → R with X |f (x)|q dx < ∞. Let 1 ≤ r ≤ q so that q/r ≥ 1. Let p be
the dual exponent to q/r. Then by Hölder’s inequality
Z Z r/q Z 1/p
r q p
|f (x)| dµ(x) ≤ |f (x)| 1 dµ(x) < ∞,
X X X
as required.
(2) RLet X = [1, ∞) with Lebesgue measureRand f : X → R the function f (x) = x−1 . Then
∞
f (x)2 dx = 1 x−2 dx = 1 < ∞ while X f (x) dx = ∞.
R
X
p→∞
Problem 7. Suppose µ(X) < ∞. Show that if f ∈ L∞ (X, µ) then f ∈ Lp (X, µ) and kf kp −−−→
kf k∞ .
Solution. We note that we can assume kf k∞ > 0, as otherwise f = 0 a.e. and the result is trivial.
We first show that lim supp→∞ kf kp ≤ kf k∞ . Let E = {x ∈ X | |f (x)| > kf k∞ }. Then µ(E) = 0
and so Z Z Z
p p p
|f (x)| dµ(x) = |f (x)| dµ(x) ≤ kf k∞ dµ(x) = kf kp∞ µ(X).
X Ec Ec
Taking pth roots we have
kf kp ≤ kf k∞ µ(X)1/p
so that
lim supkf kp ≤ kf k∞ .
p→∞
Next we establish the lower bound. Let  > 0 be arbitrary and set F = {x ∈ X | |f (x)| ≥ kf k∞ − }.
If  < kf k∞ then µ(F ) > 0 and
Z Z
|f (x)| dµ(x) ≥ (kf k∞ − )p dµ(x) = (kf k∞ − )p µ(F ).
p
X F
 PRACTICE PROBLEMS 5

Taking pth roots,

kf kp ≥ (kf k∞ )µ(F )1/p
and so
lim inf kf kp ≥ (kf k∞ − )
p→∞

since µ(F )1/p → 1 as µ(F ) > 0. But since  > 0 was arbitrary the result follows.
Problem 8. Show that the infinitely differentiable functions are dense in Lp (R, dx) for 1 ≤ p < ∞.
Solution. By the monotone convergence theorem, we see that bounded and compactly supported
functions are dense in Lp (apply monotone convergence to |1|f |<M 1|x|<M f − f | as M → ∞). By the
definition of integrability we can approximate this function with a simple function. Since each simple
function is the sum of a finite number of indicator functions, it suffices to approximate each indicator
function by a smooth function.
We can approximate an indicator function 1E with a step function as follows. Let E ⊆ U with
m(U \ E) <  and U open. Now
Z Z
|1U − 1E | = |1U − 1E |p < .

U is the union of an at-most countable number of open intervals, call them Uk . Then
∞
[
U= Uk
k=1

so by monotone convergence we must have N large so that

∞
!
[
m <
k=N +1

and so
Z X N Z XN
| 1Uk − 1E | = | 1Uk − 1E |p < 2.
k=1 k=1
It therefore suffices to approximate each 1Uk by a smooth function. Consider
( 2
e−1/(x −1) , |x| ≤ 1
f (x) =
0, otherwise
This function is clearly smooth away from ±1. It suffices to show that the derivatives of this function
converge to zero as x → ±1 from inside the interval. In fact, we have
dn −1/(x2 −1) 2

n
e = Rn (x)e−1/(x −1)
dx
where Rn is an (explicit) rational function in x. But this converges to zero at ±1 so the function is
smooth on the R whole real line.
Let C = R
f (y) dy (which is finite as f is smooth and compactly supported) and define g(x) =
−1 x
R
C 0
f (y) dy. Clearly g = 0 for x < −1 and g = 1 for x > 1, and g is a smooth function by the
fundamental theorem of calculus. We can then approximate the indicator function of an interval (say
[0, 1]) arbitrarily closely by
h (x) = g(−1 x) − g(−1 (x − 1)).
which is a smooth function such that
Z
|1[0,1] (x) − h (x)| dx ≤ 2.
R

Problem 9. Let f : [0, 1] → R be Riemann integrable. Show that f is also Lebesgue integrable and
that f is continuous a.e.
6 PRACTICE PROBLEMS

Solution. Let Pk be a sequence of nested partitions of [0, 1] such that

Z 1 Z 1
lim U (f, Pk , x) dx = lim L(f, Pk , x) dx.
k→∞ 0 k→∞ 0
Each U (f, Pk , ·) and L(f, Pk , ·) as a sequence of functions in k is strictly decreasing (resp. increas-
ing). Let U (f, x) = inf k≥1 U (f, Pk , x) and L(f, x) = supk≥1 L(f, Pk , x). These functions are clearly
measurable and
Z 1 Z 1
U (f, x) − L(f, x) dx ≤ lim inf U (f, Pk , x) − L(f, Pk , x) dx = 0.
0 k→∞ 0
Since U (f, ·) − L(f, ·) ≥ 0 the difference must be zero a.e..
Let E be the set of points where the difference is zero and not contained in any of the partitions
Pk . Clearly m(E) = 1 as there are a countable number of points in any Pk . On E we have
L(f, Pk , x) ≤ f ≤ U (f, Pk , x) for all k so that L(f, x) ≤ f ≤ U (fx ); since L(f, x) = U (f, x), we
conclude that f = U , say, on E, hence f is measurable. It is integrable since it is bounded above
and below by integrable functions, so it suffices to show that it is continuous on E.
Since x is not contained on any of the points in any Pk , we have a sequence of intervals ak < x < bk
such that ak → x and bk → x. Since x ∈ E, we must have for k sufficiently large,
U (f, Pk , x) −  ≤ L(f, Pk , x) = inf f (y) ≤ f (x) ≤ sup f (y) = U (f, Pk , x)
ak <y<bk ak <y<bk

so |f (x) − f (y)| <  for all other y in the interval. As  was arbitrary, f must be continuous at x,
hence all of E as required.