Problems
Problems
(with solutions)
Dr Nikolai Chernov
Contents
1 Lebesgue measure
2 Measurable functions
17
20
31
33
9 Lp spaces: general
34
42
11 `p spaces
46
Lebesgue measure
m(int(A)) = 0.
m(A) = m(A)
Now
A = int(A) (A \ int(A)),
1
A \ int(A) A.
Hence A is the union of an open set, int(A), and a subset of the null set A. Since
the latter is always measurable, we conclude that A is a measurable set.
JPE, Sept 2010. Is the following true of false?
Let E be a subset of Rn , and int(E) the set of all interior points of E. Then
int(E) = if and only if (E) = 0. (Here denotes the outer measure.)
If (E) = 0, then m(E) = 0, so int(E) is indeed empty. But the converse is
not true. The set of points with irrational coordinates has infinite measure and
empty interior.
JPE, May 2005. Show that if A [0, 1] and m(A) > 0, then there are x and y
in A such that |x y| is an irrational number.
If |x y| Q for any x, y A, then A x + Q for any point x A, hence A
would be a countable set and we would have m(A) = 0.
JPE, Sept 2004 and Jan 1989. Is the following true or false?
There is a subset A of R which is not measurable, but such that B = {x
A : x is irrational} is measurable.
False. The set A \ B Q is countable, hence measurable. So if B was measurable, then A = B (A \ B) would be measurable, too.
JPE, May 2001. Does there exist a non-measurable subset of R whose complement in R has outer measure zero?
No. If the outer measure of a set is zero, then its inner measure is also zero, so
the set is measurable. Then its complement is measurable, too.
JPE, May 2000. Do there exist two non-measurable sets whose union is measurable?
Yes. If A is any non-measurable set, then its complement Ac is also nonmeasurable, but their union is the whole space (a measurable set).
JPE, May 2000. Is the following true of false?
If the boundary of Rk has outer measure zero, then is measurable.
True. Since the outer measure of is zero, its inner measure is zero, too,
hence its Lebesgue measure is zero. Then any subset of is a null set, and
therefore it is measurable, too. Now is the union of two sets:
= int() ( \ int()).
Note that int() is an open set, hence it is measurable. And ( \ int()) is
a subset of a null set, hence it is also measurable. Therefore is measurable.
n=1 (rn
1
, rn
n2
1
)
n2
m (rn
n=1
1
, rn
n2
1
)
n2
2
n2
< ,
n=1
Measurable functions
en|1sin x| >
|1 sin x| <
1
1
ln
n
f (x + n1 ) f (x)
1
n
f is measurable?
No. Indeed, f = ( f )2 is a composition of a measurable function f and a continuous (and thus Borel) function, x2 , thus f is measurable.
JPE, May 1994 Let {fn } be a sequence of measurable functions on a measurable
space (X, M). Define the set
E = {x X : lim fn (x) exists}
n
[0,1]
, where the infimum is taken over all simple measurable functions with f .
True. Since f is bounded, let M = sup f be its upper bound. Now for every
simple function f there is a simple function such that
R M andR f
. Indeed, it is enough to take = min{, M }. Note that [0,1] dm [0,1] dm,
5
hence
inf
dm =
[0,1]
inf
f M
dm,
[0,1]
0sg
[0,1]
[0,1]
where the infimum is taken over all simple functions s such that 0 s g.
For each such s we have = M s a simple function satisfying f M .
Conversely, for every simple function satisfying f M we have s = M
a simple function satisfying 0 s g. Thus the above identity gives
Z
Z
Z
() dm = inf
dm.
f dm = sup
f M
[0,1]
f M
[0,1]
[0,1]
where the infimum is taken over all simple measurable functions with f ?
False. This would be true if the measure of the whole space was finite (like in
the previous problem). But here m(R) = , in which case the claim is false.
A counterexample is any bounded function f (x) L1m (R) such that f (x) > 0
2
1
for all x R. For example, f (x) = ex or f (x) = 1+x
2 , which you might
remember from Calculus or Probability Theory. If you do not remember any, you
can construct f as follows:
f=
2|n| [n,n+1)
n=
2|n| = 3 <
n=
Now since f (x) > 0 for all x R, then any simple function f must also be
positive everywhere, so that in the representation
=
n
X
i=1
i Ai
all the values i are positive: i > 0 for all i = 1, . . . , n. At the same time at least
one subset Ai0 must have infinite measure, i.e., m(Ai0 ) = . Therefore
Z
n
X
dm =
i m(Ai ) i0 m(Ai0 ) =
i=1
dm = , while
R
R
R
f dm < .
JPE, May 2011. Prove that a measurable function f (x) belongs to L1 (0, 1) if
and only if
X
2n m{x [0, 1] : |f (x)| 2n } < .
n=1
JPE, May 2006. Let (X) < . Prove that a non-negative measurable function
f (x) belongs to L1 (X, ) if and only if
2n {x X : f (x) 2n } < .
n=1
The above two problems are almost identical, we only solve the first one. Let
us partition [0, 1] into subsets
E0 = {x : |f (x)| < 2}
E1 = {x : 2 |f (x)| < 22 }
...
En = {x : 2n |f (x)| < 2n+1 }
...
g(x) = 1
g(x) = 2
g(x) = 2n
and a new function g by g(x) = 2n for all x En , as shown above. Note that
Z
Z
Z
g(x) 1 |f (x)| 2g(x)
g 1 |f | 2 g,
thus f L1 (0, 1) if and only if g L1 (0, 1).
Denote
X
S: =
2n m{x [0, 1] : |f (x)| 2n }
n=1
m{x : |f (x)| 2 } =
X
k=n
m(Ek ).
X
X
n
g dm =
2 m(En ) 2m(E0 ) +
2n m{x : |f (x)| 2n } 2 + S.
[0,1]
n=0
n=1
X
n=1
m(Ek ) =
k=n
m(Ek )
n=1
k=1
thus
k
X
k+1
Z
m(Ek ) 2
k=1
g dm
[0,1]
Z
g dm S + 2.
[0,1]
and
fk = k 1 + (1)k [0,k1 ]
R
Then R fk dm alternatively takes values 2 (for all even k) and 0 (for all odd k), so
it does not have a limit.
R
We note, however, that whenever limk R fk dm does exist, the claimed indeed immediately follows from Fatous Lemma.
JPE, May 2010. If f L1 (0, 1), find
Z 1
|f (x)|2
lim
k ln 1 +
dx.
k 0
k2
The limit is zero. The integrand can be written (and bounded above) as
1
|f (x)|2 k2
1
1
2
ln 1 +
ln e|f (x)| = |f (x)|2 ,
2
k
k
k
k
thus it converges to zero pointwise a.e. (more precisely, for every x [0, 1] such
that |f (x)| < ). However, the Lebesgue Dominated Convergence
apply
R 1 does not
2
1
2
(yet), because |f | is not necessarily in L (0, 1), i.e., we may have 0 |f (x)| dx =
R1
ln(1 + t) 2 t
1 + t e2
4t
= 1 + 2 t + +
2
therefore
fn
f.
therefore
Rn
Rn
|f |
Rn
JPE, May 2009. Let f L1 ([0, 1]) be real-valued. Prove the following statements:
(a) xk f (x) R L1 ([0, 1]) for all k N.
1
(b) limk 0 xk f (x) dx = 0.
(c) If limx1 f (x) = a for some real number a, then
Z
lim k
xk f (x) dx = a.
k+1
1
f (x) = 1k+1 f (1) 0k+1 f (0) = a,
0
R 1
0
k1
10
(iii) Note that a < f (x) < a + on the interval [1 , 1], thus
Z 1
Z 1
Z 1
k
k
xk dx.
x f (x) dx (a + )k
x dx k
(a )k
1
1 (1 )k+1
k+1
xk dx =
gives
(a )k
1 (1 )k+1 k
k+1
xk f (x) dx
(a + )k
1 (1 )k+1
k+1
Taking the limit k shows that the middle integral will be eventually squeezed
between a and a + . Since > 0 is arbitrary, Part (c) follows.
JPE, May 2009. Suppose that fn is a sequence of non-negative Lebesgue measurable functions
on (0, 10) such
Rx
R x that fn (x) f (x) for almost all x (0, 10). Let
F (x) = 0 f dm and Fn (x) = 0 fn dm. Prove that
10
Z
(f + F ) dm lim inf
n
10
(fn + Fn ) dm.
0
Second,
Z
10
10
Z
(f + F ) dm
(f + lim inf Fn ) dm
0
10
Z
=
lim inf(fn + Fn ) dm
Z 10
lim inf
(fn + Fn ) dm.
0
JPE, Sept 2008. Let f L1 (0, ). Prove that there is a sequence xn such
that limn xn f (xn ) = 0.
Denote c = lim inf x x|f (x)|. If c = 0, then a sequence as above exists. If
c > 0, then there exists A > 0 such that x|f (x)| > c/2 for all x > A. Then
Z
Z
Z
c
|f | dm
|f | dm
dx = ,
(0,)
(A,)
A 2x
11
(0,)
[0,1]
12
Suppose that
R
[0,1]
f n dm
See also a much harder version of this problem in Section on the Lp spaces.
We solve the 2004 problem. The 1995 and 1993 problems will follow as a side
result.
Part (a): Let us partition [0, 1] into four subsets:
G = {x [0, 1] :
E = {x [0, 1] :
L = {x [0, 1] :
Z = {x [0, 1] :
Note that
f (x) > 1}
f (x) = 1}
0 < f (x) < 1}
f (x) = 0}.
Z
f dm = m(E)
and
f n dm = 0
for all x G
for all x L.
On the subset L [0, 1], we have f n (x) < 1 for all n and all x L, hence we can
apply the Lebesgue Dominated Convergence:
Z
Z
n
lim
f dm =
0 dm = 0.
n
f n dm = m(G) + m(E).
[0,1]
13
if m(G) > 0
0 if m(G) = 0
f n dm = C <
[0,1]
[0,1]
As a result, f = E a.e.
JPE, May 2001. Let f and fn , n = 1, 2, . . ., be non-negative measurable functions on [0, 1] such that fn converges pointwise
each of the following
R 1 to f . RUnder
1
additional assumptions, either prove that 0 f dm 0 f dm or show that this is
not generally true.
(a) fn f and fn L1 ([0, 1]) for all n.
(b) fn fn+1 for all n.
(c) fn f for all n.
(Question (b) was also given in JPE, Sept 1993 and Jan 1989.)
(Question (c) was also given in JPE, Sept 1995.)
(a) False: fn = n(0,1/n) and f = 0.
(b) False: fn = 1/(nx) and f = 0. (Note that
R1
0
(c) True. First note that we cannot apply Lebesgue Dominated Convergence,
because f may not be integrable. We proceed as follows. On the one hand,
Z 1
Z 1
Z 1
Z 1
fn f
fn dm
f dm lim sup
fn dm
f dm
0
thus lim
R1
0
R1
0
14
f dm.
fn dm,
En
exists.
R
The above limit equals E f dm, where E = n1 En . Indeed, we have
Z
Z
Z
Z
f dm
=
f
dm
f
dm
|f | dm.
E
En
E\En
E\En
R
R
We know that (A) = A |f | dm is a measure on R, and (R) = R |f | dm < .
Thus by the continuity
Z
|f | dm = (E \ En ) (E \
n=1 En ) = () = 0.
E\En
JPE, Sept 1995, May 1995 and May 1989. Let > 2 be a real number.
Define
o
n
p
2
E = x [0, 1] : x < q for infinitely many p, q N .
q
Prove that m(E) = 0.
In the May 1995 and May 1989 versions, was set to 3.
Denote
n
Ep,q = x [0, 1] : x
1o
p
< .
q
q
m(Ep,q )
p,q
X
2(q + 1)
q=1
X
4
< ,
1
q
q=1
[0,1]
R
R
Note that it is not necessarily true that [0,1] fn dm [0,1] f dm under the given
conditions. For example, let f 0 and fn = n(0, 1 ) . Then we have fn (x) f (x)
n
R
R
for all x [0, 1] but [0,1] fn dm = 1 6= 0 = [0,1] f dm.
15
R
R
It is then really amazing that [0,1] fn efn dm [0,1] f ef dm.
The reason is that fn (x)efn (x) f (x)ef (x) a.e. (quite obviously), and the
integrands here are bounded by one integrable function:
0 fn (x)efn (x) g(x) = 1.
In fact, the bound can be tightened: the function tet for t 0 reaches its maximum at t = 1 and its maximum value is e1 < 1.
Now the Lebesgue Dominated Convergence applies and completes the solution.
JPE, Sept 1994. Let (X, M, ) be a measure space with a positive measure .
Let f be a non-negative function on X such that for any n = 1, 2, . . . there are
two measurable
hn on X such that gn (x) f (x) hn (x) for all
R functions gn and
1
x X, and X (hn gn ) d < n . Prove that f is measurable and
Z
Z
f d = lim
gn d.
n
R
Note that hn gn 0, hence X (hn gn ) d = khn gn k1 , so we have khn
gn k1 < n1 , i.e., khn gn k1 0. By a corollary proved in class, there is a subsequence
nk such that hnk gnk 0 a.e.
A side note: we can arrive at the same
R conclusion without using the corollary
or the 1-norm. We can first prove that X (hn gn ) d 0 implies hn gn 0 in
measure, and then refer to a homework exercise were we proved that convergence
in measure implies the existence of a subsequence converging a.e.
Now since hnk (x) gnk (x) 0 a.e. and gnk (x) f (x) hnk (x), then gnk (x)
f (x) a.e. In other words, there is a full measure set E X such that gnk (x) f (x)
for all x E. Thus the restriction f |E of f to E is a limit of measurable functions,
hence f |E is measurable. The values of f on the null set X \ E do not affect the
measurability of f , hence f is measurable
on the whole space X.
R
Next we need to prove that X gn d exists in a certain
sense. The function gn
R
is not necessarily non-negative, so the existence of X gn d is not a trivial issue.
Since hn (x) f (x) 0, we have
0 gn (x) = max{0, gn (x)} max{0, hn (x) gn (x)} = hn (x) gn (x),
hence
Z
X
gn
Z
d
(hn gn ) d < .
X
R
This implies that X gn d exists, at least in the extended sense (defined in class);
though note thatR its value may
R be +. Also note that f 0 and f gn 0, so
the existence of X f d and X (f gn ) d need no justification. Now we have
Z
Z
Z
f = gn + (f gn )
f d =
gn d + (f gn ) d.
X
16
Note that
1
.
n
X
X
Thus taking the limit n proves the last claim of the problem.
0
(f gn ) d
(hn gn ) d
Note that for each x R, the sequence n2n+x2 monotonically increases and
2
converges to 1, as n . Thus the integrand n2n+x2 f (x) monotonically increases
(in n) and converges to f (x) as n . By the Lebesgue Monotone Convergence
Z
Z
n2
f (x) dm
f (x) dm = kf k1 .
2
2
R
R n +x
JPE, Sept 1989. Let fn be a sequence of continuous functions Lebesgue integrable on [0, ) which converges uniformly to a function f Lebesgue integrable on
[0, ). Is it true that
Z
lim
|f (x) fn (x)| dx = 0.
n
False. Example: we set f (x)R 0 and define fn by fn (x) = n1 nx2 for x (0, n)
Note: in all the problems of this section the function f must be real-valued, even
though this assumption is NOT made explicitly in any of them, for some reason.
JPE, May 2011. Let f L1 (, ). Let
P{an } be a sequence of strictly positive
numbers, i.e., an > 0 for any n, such that
n=1 an = 1.R Prove that there
R exists a
17
R
(,x)
F (x + ) F (x) =
f dm
[x,x+)
|f | dm
[x,x+)
R
We know that (E)
=
|f | dm is a measure in R, and it is a finite measure
E
R
because (R) = R |f | dm < . Hence by the continuity we have
Z
Z
|f | dm = [x, x + ) ({x}) =
|f | dm = 0,
0
[x,x+)
{x}
because {x} is a singleton whose Lebesgue measure is zero. (Note: the continuity
of F (x) was a separate problem in JPE, Sept 1995 and May 1992.)
Next, we have the following limits:
Z
lim F (x) = 0,
lim F (x) = I : =
f dm.
x
Indeed,
Z
|f | dm
|F (x)|
(,x)
Z
|f | dm =
|f | dm = 0,
(,x)
Similarly,
Z
|I F (x)|
|f | dm
[x,)
Z
|f | dm =
|f | dm = 0.
[x,)
Now suppose first that I 6= 0. By the intermediate value theorem there are points
< x1 < x2 < < such that
F (xn ) = (a1 + + an )I
for all n 1.
R
In the case I = 0, i.e., R f dm = 0, we can just choose E2 , E3 , . . . to be arbitrary
disjoint null sets, and Rdefine E1 = R \
n=2 En .
Note: in the case R f dm = 0 we can also choose sets En differently, so that
each of them has positive measure. To this end we need to apply the previous
argument to f + and f separately.
JPE, May
R 2003. LetRf be integrable on [0, 1]. Prove that there exists c [0, 1]
such that [0,c] f dm = [c,1] f dm.
R
R
Similarly to the previous problem, let F (c) = [0,c] f dm [c,1] f dm. It is a
continuous function on [0, 1] and F (0) = F (1). If F (0) 6= 0, the existence of c
such that F (c) = 0 follows from the intermediate value theorem. If F (0) = 0, we
can choose c = 0 (or c = 1).
Note: in the case F (0) = 0 the choices c R= 0 and c = 1 may be the only
1
possible. RExample: f (x) =Rsin(x). We have 0 f (x) = 0, but for any c (0, 1)
we have [0,c] f (x) > 0 and [c,1] f (x) < 0.
JPE, Sept 2007 and May 2001. Let f L1 (R2 ). Prove that there exists a
subset E R2 such that
Z
Z
f dm =
f dm.
R2 \E
f dm,
R2 \Dr
Dr
where Rr,r+ = {r2 x2 + y 2 < (r + )2 } is the ring with the inner radius r and
the outer radius r + . By the triangle inequality
Z
|F (r + ) F (r)| 2
|f | dm
Rr,r+
2
We know that (E)
R = E |f | dm is a measure in R , and it is a finite measure
2
because (R ) = R2 |f | dm < . Hence by the continuity we have
Z
Z
lim
|f | dm =
|f | dm = 0,
0
Rr,r+
Rr,r
because Rr,r = {x2 +y 2 = r2 } is a circle (a ring with zero width) whose Lebesgue
measure is zero. Next, similarly to the previous problems,
Z
Z
F (0) =
f dm and lim F (r) =
f dm = F (0).
R2
19
R2
If F (0) 6= 0, the existence of r such that F (r) = 0 follows from the intermediate
value theorem, and we can choose E = Dr . If F (0) = 0, we can choose E to be
any null set (or make RR2 \ E a null set).
Note: in the case R2 f dm = 0 we can choose a more interesting set E so
that m(E) > 0 and m(E c ) > 0. To this end we need to apply the above argument
to f + and f separately.
Preliminary Note. In many problems one has to use the following upper bound:
x n
ex
(1)
1+
n
which holds for all
n1
and
x > n
(2)
This bound needs to be proved! You cannot just refer to Calculus or Advanced
Calculus where this fact may have been mentioned.
To prove it, take logarithm of both sides in (1) and convert it to
ln
Since t =
x+n
n
x
x+n
x+n
=
1
n
n
n
t > 0
(3)
dn
n
x+n
Since an > 0, we have
dan
dn
0 if and only if
ln
or equivalently
ln
x+n
x
n
x+n
n
x
n
=
1
x+n
x+n
x+n
20
Since t =
n
x+n
due to (2), we again reduce our inequality to the standard fact (3).
lim
1+
[0,n]
x n x
e
dx
n
fn (x) =
x 0.
1 + x ln(n+2011)
JPE, May 2006. Find
Z
lim
1
n
1 + x ln n+2006
dx.
The above two problems are almost identical, we only solve the first one.
Note that
n
lim
= ,
n ln n + 2011
hence we have pointwise convergence fn (x) f (x), where
n
2
ln n + 2011
for all n n0 for some n0 1, hence
fn (x)
1
1 + x2
21
dx
k .
1 + xk k x
lim
k
k
k 1 + x
e
1
x
k
k
pointwise. Due to Preliminary Note, 1 + xk is a monotonically increasing sequence (converging to ex ). Thus for all k 2 the integrand is bounded by
lim
1
1
2
k
1 + x2
1 + xk k x
which is an integrable function (it is inR L1 (0, )). Thus by Lebesgue Dominated
0 n
/2
n sin nx dx.
The pointwise limit of the integrand is x, which is also its upper bound.
R /2
Hence by Lebesgue Dominated Convergence the limit is 0
x dx = 23 ( 2 )3/2 .
22
sin x n
lim
[0,n]
dm.
You can use the fact that | sin x| < x for all x > 0.
JPE, Sept 2006. Find
Z
sin x n
dx.
lim
n 0
x
For all x > 0 we have sinx x < 1 hence
sin x n
x
0
n
pointwise.
Next we find a good upper bound. First, the integrand is bounded by g1 (x) = 1.
Second, for all n 2 the integrand is bounded by g2 (x) = 1/x2 . Thus we have
sin x n
min{g1 (x), g2 (x)} = min{1, 1/x2 }
x
and this upper bound is integrable (it belongs to L1 (0, )). Thus the Lebesgue
Dominated Convergence applies. Therefore the above limit is zero.
JPE, Sept 2005. Find the limit and justify your answer
Z
ln(nx)
lim
dx.
n 1
x + x2 ln n
The integrand converges to f (x) =
1
x2
ln n + ln x
1
ln x
2+ 2 ,
2
x + x ln n
x
x
which is an integrable function (it is in LR1 (1, )). Thus the Lebesgue Dominated
23
dx
n 1 + x2 0
n 0 (1 + x2 )2
Z
1
1 2x cos(nx)
=
dx.
n n 0 (1 + x2 )2
For the last integral we have
Z
Z
Z
|2x
cos(nx)|
2xdx
2x
cos(nx)
dx
dx
= 1,
(1 + x2 )2
(1 + x2 )2
(1 + x2 )2
0
0
0
Hence our original integral converges to zero.
JPE, May 2003 and Sept 1999 Let E = [0, ). Prove that limn
exists and find its value.
x
E 1+xn
dx
x
We have limn 1+x
n = f (x) pointwise, where f (x) = x for x [0, 1), f (x) = 0
for x (1, ), and f (1) = 0.5. For all n 3 our integrand is bounded by the
x
function g such that g(x) = x for x [0, 1] and g(x) = 1+x
3 for x (1, ). We
1
can easily see that g L (E),
thus Lebesgue Dominated Convergence applies, and
R
our integral converges to E f dm = 0.5.
JPE, Sept 2002 and May 1994. Find the limit and justify your steps
Z 1
(nx)2
dx.
lim
n 0 (1 + x2 )n
This problem is unusual, because the limit of the integral here is NOT equal
to the integral of the pointwise limit function. In fact, the integrand converges to
zero pointwise while the integral converges to infinity.
A quick inspection shows that the integrand takes high values for x 1/ n.
So let us estimate the integral from below by
Z 1
Z 1/n
(nx)2
(nx)2
dx
dx
2 n
(1 + x2 )n
0 (1 + x )
0
Now for all x [0, 1n ] we have
(nx)2
(nx)2
(nx)2
(1 + x2 )n
(1 + 1/n)n
e
because (1 + 1/n)n e. Therefore
Z 1
Z
(nx)2
1 1/ n
n
2
dx
(nx) dx =
.
2
n
e 0
3e
0 (1 + x )
24
Z k
x k x/3
e dx.
k
x k x/2
e dx.
k
Z k
n
(a) Due to Preliminary Note, 1 nx is a monotonically increasing sequence converging to ex . Thus the integrand monotonically converges to f (x) =Re2x/3 point
wise. By Lebesgue Monotone Convergence, our integral converges to 0 f dx = 32 .
(b) The solution is almost identical to Part (a). The answer is 2.
JPE, May 2001 and May 1991. Compute
Z 1 n1
nx
dx.
lim
n 0 2 + x
(in JPE, May 1991, the value of the limit (= 31 ) was given.)
This problem is unusual, because the limit of the integral here is NOT equal
to the integral of the pointwise limit function. In fact, the integrand converges to
zero pointwise while the integral converges to a positive number.
We can integrate by parts first:
1 Z 1
Z 1 n1
Z 1
nx
dxn
xn
xn
dx =
=
+
dx
2
2 + x 0
0 2+x
0 2+x
0 (2 + x)
The first term on the right hand side is 13 . The last integral converges to zero,
because the integrand converges to zero pointwise for all x [0, 1) and is bounded
xn
1
by an integrable function (in fact, it is bounded by a constant function: (2+x)
2 4 ).
Alternatively, we can change variable y = xn , which gives us
Z 1
dy
lim
.
n 0 2 + y 1/n
Now the integrand is bounded by
converges to 13 .
1
2
cos(xn ) dx.
lim
[0,1]
25
cos(xn ) cos 0 = 1,
and for x = 1 we have cos(xn ) = cos 1. Note that | cos(xn )| 1, hence the
Lebesgue Dominated Convergence applies and gives
Z
Z
n
cos(x ) dm =
1 dm = 1.
lim
n
[0,1]
[0,1]
nx < 1 + nx 2nx,
hence
ln n + ln x
ln(1 + nx)
ln n + ln x + ln 2
.
1 + x2 ln n
1 + x2 ln n
1 + x2 ln n
By the squeeze theorem, for each x 1
lim
ln(1 + nx)
1
= 2.
2
1 + x ln n
x
+
+ 2
1 + x2 ln n
x2
x2
x
which is an integrable function on (1, ). Thus the Lebesgue Dominated Convergence applies and gives
Z
Z
ln(1 + nx)
1
lim
dx = 1.
dx =
2
n 1
1 + x ln n
x2
1
JPE, May 1998 and Sept 1995. Find the limit
Z
n sin(x/n)
lim
dx.
n 0
x(1 + x2 )
We have
lim
sin(x/n)
=1
(x/n)
lim
26
n sin(x/n)
1
=
2
x(1 + x )
1 + x2
for all x > 0, pointwise. Also note that sin for all 0, thus
n sin(x/n)
1
,
x(1 + x2 )
1 + x2
which is an integrable function on (0, ). Thus the Lebesgue Dominated Convergence applies and gives
Z
Z
n sin(x/n)
1
1
lim
dx. =
dx = tan x = /2.
n 0
x(1 + x2 )
1 + x2
0
0
JPE, Sept 1997. Show that the hyperelliptic integral
Z
x dx
p
(x2 2 )(x2 1)(x 2)
2
converges to the elliptic integral
Z
dx
p
(x2
1)(x 2)
as tends to zero.
For each x > 2, we have pointwise convergence, as 0,
x
p
x
x2 (x2 1)(x 2)
=p
(x2 1)(x 2)
(x2
2 )(x2
1)(x 2)
(x2
1)(x2
1)(x 2)
(x2
x
p
.
1) (x 2)
X
(1)m n
(1)m
=
n
n + nm2 + 1 m=1 m2 + 1
m=1
lim
Let X = N and be the counting measure. Then the above claim may be
stated in terms of Lebesgue integrals:
Z
Z
(1)x n
(1)x
lim
d
=
d.
2
n X n + nx2 + 1
X x +1
27
n + nx2 + 1
x2 + 1
as n .
n + nx2 + 1
x2 + 1
P
1
which is an integrable function on X, because
m=1 m2 +1 < . Thus the Lebesgue
Dominated Convergence applies and gives the result.
lim
ex /n
dx
1 + x2
lim
ex /n
dx =
1 + x2
dx
1
= .
=
tan
x
1 + x2
t
JPE, Oct 1991. (1) Prove that f (t) = 1+t
2 is a bounded function on (, ).
(2) Let
nx
fn (x) =
(x [0, 1], n = 1, 2, . . .)
1 + n2 x2
Then prove that fRn (x) does not converge uniformly on [0, 1].
1
(3) Find limn 0 fn (x) dx, if it exists.
t
(1) This a calculus exercise. Let us just denote the bound by C = max 1+t
2.
(2) For x = 1/n we have fn (x) = 1/2, thus there is no uniform convergence.
(3) We have pointwise convergence fn (x) 0 for all x [0, 1] and a common
upper bound fn (x) C. Thus Lebesgue Dominated Convergence applies.
JPE, Oct 1991. Let
fn = n [
1
,1]
n+1 n
then
Z
dm =
[0,1]
n [
1
,1]
n+1 n
n=1
1
[ n+1
, n1 ]
n=1
X
X
n1
1
=
= .
n + 1 n=1 n + 1
n=1
n1
0.
n+1
JPE, Oct 1991. (i) Let f be a step function on a bounded interval [a, b]. Then
prove that
Z b
f (x) sin(nx) dx.
()
lim
n
(ii) Let f be a bounded measurable function on [a, b]. Then prove (*).
(iii) Let f be an integrable function on (, ). Then prove (*).
This problem is almost a research project... We only sketch the solution. First,
let f (x) = (c,d) be a constant function on an interval (c, d). Then
Z d
d
f (x) sin(nx) dx = n cos(nx) = n (cos(nd) cos(nc)),
c
hence
f (x) sin(nx) dx 2||
0.
n
c
P
Next, if f is a step function, then f = N
i=1 i (ci ,di ) , and the result follows by
additivity.
(ii) If f is a bounded measurable function on [a, b], then f L1 ([a, b]). We can
approximate f by step functions (in the L1 norm) and apply Part (i).
(iii) The same strategy: approximate f by step functions (in the L1 norm) and
apply Part (i).
JPE, May 1991. Let
2
n1/4 ex n
fn (x) =
.
1 + x2
(a) Prove that fnR L1 (0, ).
(b) Find limn (0,) fn dm.
29
(a) Since ex
Z
2n
1, we have
Z
fn dm
(0,)
(0,)
n1/4
1/4
1
dm = n tan x = n1/4 /2,
2
1+x
0
hence fn L1 (0, ).
(b) We will need a common upper bound for all fn s. To this end we combine n1/4
2
4
with ex n as follows. The function g(t) = tet is bounded on (0, ).
Indeed,
by a standard Calculus-I argument its maximum is achieved at t = 1/ 2 and its
maximum value is C = 12 e1/4 . Therefore
2
2
x1/2 n1/4 ex n C
n1/4 ex n C/ x.
So we have
Z
fn dm
(0,)
(0,)
C/ x
dm < .
1 + x2
The last integral is finite, because the integrand has tail O(x5/2 ), as x
, which
is integrable. And it grows to infinity near x = 0 at a rate O(1/ x), which
is also integrable. Thus fn L1 (0, ) and fn are bounded by one integrable
function. It is easy to see that limn fn (x) = 0 for each x > 0, thus by the
Lebesgue Dominated Convergence
Z
Z
lim
fn dm =
0 dm = 0.
n
(0,)
(0,)
nxenx .
3
3
n1/2 x3/2 enx C
nxenx C/ x.
So we have
Z
fn dm
[0,1]
[0,1]
C
dm = 2C.
x
It is easy to see that limn fn (x) = 0 for each x > 0, thus by the Lebesgue
Dominated Convergence
Z
Z
lim
fn dm =
0 dm = 0.
n
[0,1]
[0,1]
30
X
n=1
n=1
X
n=1
X
n=1
R k
Z
lim inf
k
fn,k dm
R
1
< ,
n2
thus f L1 (R).
JPE, Sept 2008 and Sept 2007. Let fn , n =R 1, 2, . . . , be a sequence of
non-negative continuous functions on R such that R fn dm < n13 . Let f (x) =
P
X
X
1
< .
f dm =
fn dm <
3
n
R
R
n=1
n=1
Note: the continuity of fn s is not necessary, it is enough to assume their measur1
ability. The bound n13 can be relaxed to n12 or n1+a
for any a > 0.
JPE, May 2006. Let fn , n = 1, 2, .R. . , be a sequence of non-negative
Lebesgue
P
f
(x).
Show
measurable functions on R such that fn dm < 21n . Let f (x) =
n=1 n
1
that f L (0, ).
For non-negative functions, the summation and integration commute, hence
Z
Z
X
X
1
f dm =
fn dm <
< .
2n
(0,)
n=1 (0,)
n=1
Note: strangely, the domain of integration is not specified in the first formula of
the problem. One may naturally assume that it is R. But then it is not clear why
31
(1 et )n et dt.
1/2
Since the integrands are non-negative functions, the summation and integration
commute, hence
Z
X
n=0
t n t2
(1 e ) e
Z
dt =
1/2
(1 et )n et dt
1/2 n=0
X
t2
Z
=
1/2
n=0
t2
=
1/2
Z
=
(1 et )n dt
1
dt
1 (1 et )
ett dt
1/2
The last integral is computed by completing the square and changing variable:
Z
Z
Z
2
tt2
1/4
(t1/2)2
1/4
e
dt = e
e
dt = e
es ds = e1/4 /2.
1/2
1/2
Note:R apparently it is assumed here that the students know the value of the inte2
/2
(1
Since the integrands are non-negative functions, the summation and integration
32
commute, hence
Z
X
n=0
/2
(1
sin x) cos x dx =
/2 X
(1
cos x
0
cos x
0
=
0
(1
sin x)n dx
n=0
/2
=
Z
n=0
/2
=
Z
/2
dx
1 (1 sin x)
cos x
dx
sin x
/2
cos x
dx =
sin x
Z
0
/2
/2
d sin x
= 2 sin x = 2.
0
sin x
JPE, Sept 2011. Is it true that the characteristic function of the Cantor set is
Lebesgue integrable in [0, 1] but not Riemann integrable?
False. The characteristic function of the Cantor set is continuous on the complement to the Cantor set (that complement consists of open intervals on which
the function is identically zero). Thus the set of discontinuity points is exactly the
Cantor set, which measure zero. This implies Riemann integrability.
JPE, Sept 2004. Let f : [0, 1] R be defined by
x if x is irrational
f (x) =
0
otherwise
(i) Show that f is measurable.
(ii) Is f Lebesgue integrable? If yes, find its Lebesgue integral.
(iii) Is f Riemann integrable? If yes, find its Riemann integral.
(i) just like in some homework exercises.
(ii) yes, because f is measurable and bounded. Changing f on a set of measure
zero will not affect its Lebesgue integral, so we can replace f with g(x) = x for
all x [0, 1]. Now
Z
Z
Z 1
2
f dm =
g dm =
x dx = .
3
[0,1]
[0,1]
0
33
(iii) no, because f is discontinuous at every x > 0, hence the set of the discontinuity points has positive Lebesgue measure.
JPE, Sept 2001. Let f : [0, 1] R be defined by
x if x is rational
f (x) =
0
otherwise
(i) Show that f is measurable.
(ii) Prove or disprove that f is of bounded variation on [0, 1].
(iii) Is f Lebesgue integrable but not Riemann integrable?
(i) just like in some homework exercises.
(ii) see some other section...
(iii) yes. It is Lebesgue integrable because it is measurable and bounded. It is not
Riemann integrable because f is discontinuous at every x > 0, hence the set of the
discontinuity points has positive Lebesgue measure.
Lp spaces: general
JPE, May 2012. Let p [1, ). Suppose that fn Lp ([0, 1]) converges a.e. to
f Lp ([0, 1]). Show that fn converges to f in Lp ([0, 1]) if and only if kfn kp kf kp .
Note: this is the most popular JPE problem. Its variants were given in the
following exams: JPE, May 2005, Sept 1998, and Jan 1989 (for p = 2),
and JPE, May 2003, Sept 2002, Sept 2001, Sept 1999, May 1995, Sept
1994, and Sept 1993 (for p = 1). Sometimes [0, 1] is replaced with R or with an
unspecified measurable subset E R. See also a clever extended version of this
problem given in JPE, May 1994, in this section.
The only if part is simple: just use the triangle inequality
kfn f kp kfn kp kf kp ,
hence kfn f kp 0 implies kfn kp kf kp .
We now turn to the difficult if part. Consider first the simpler case p = 1.
There is a relatively fast solution that goes as follows. By the triangle inequality
|fn | |fn f | |f |.
Since f L1 ([0, 1]), the Lebesgue Dominated Convergence gives
Z
Z
Z
Z
|fn | |fn f | =
|fn | |fn f | |f |
n
34
R
therefore |fn f | 0, i.e., fn f in L1 . However, this solution does not seem
to generalize to the p > 1 case. Hence we present another solution below.
By the triangle inequality |fn f | |fn | + |f |, hence
|fn | + |f | |fn f | 0.
Now by Fatous Lemma (as in the proof of Lebesgue Dominated Convergence)
Z
Z
2 |f | = lim inf(|fn | + |f | |fn f |)
Z
lim inf (|fn | + |f | |fn f |)
Z
Z
Z
= lim inf |fn | + |f | |fn f |
Z
Z
= 2 |f | lim sup |fn f |.
R
R
Therefore lim sup |fn f | 0, which implies |fn f | 0, i.e., fn f in L1 .
In the case p > 1 we use the convexity of the function (x) = |x|p on the entire
real line < x < and Jensens inequality to get
f f p f + (f ) p |f |p + | f |p
|fn |p + |f |p
n
n
n
=
=
2
2
2
2
hence
2p1 |fn |p + 2p1 |f |p |fn f |p 0;
after that the previous solution for p = 1 is repeated almost verbatim.
JPE, May 2012 and Sept 2004. Are the following true of false?
(a) If f Lp ([0, 1]) for all p (1, ), then f L ([0, 1]).
(b) If 1 p < q < , then Lq ([1, )) Lp ([1, )).
(a) False. Counterexample: f = ln x.
(b) False. Counterexample: f = xr for any
1
q
< r < p1 .
JPE, Sept 2010 and Sept 1988. Is the following true of false?
Suppose fn L1 (0, 1) for all n N, that fn (x) g(x) for almost every x [0, 1]
and that fn f in L1 (0, 1). Then f (x) = g(x) for almost every x [0, 1].
True. By a corollary proved in class: If fn f in the Lp norm (p [1, )),
then there is a subsequence {fnk } such that fnk f a.e. So the subsequence fnk
converges a.e. to f and to g. This implies f = g a.e.
JPE, Sept 2009. Is the following true of false?
If {fk } Lp (Rn ) Lr (Rn ) for some p, r [1, ), fk g in Lp (Rn ) and fk h in
Lr (Rn ). Then g(x) = h(x) a.e. in Rn .
35
1
n
36
where q 0 = q/(q p) is just the exponent conjugate to q/p. Now raising both sides
to the power 1/p proves that g(p) g(q).
(b) For any A > 0 we have mA : = m{x : |f (x)| > A} > 0, hence
Z
1/p
g(p)
|f (x)| dx
1/p
{x : |f (x)|>A}
Taking the limit p gives lim inf p g(p) A. Since A is arbitrary, we have
limp g(p) = .
JPE, May 2009. Let (X, M, ) be a measure space with a positive, finite measure
. Consider a function f L () such that kf k > 0.
(a) Show that, for every positive , the set {x : |f (x)| > kf k } has positive
measure.
(b) Show that
kf kn+1
n+1
lim kf kn = lim
= kf k .
n
n kf kn
n
(The convergence kf kn kf k was also given, as a separate problem, in JPE,
May 1998, Sept 1995, May 1994, and May 1990.)
(a) Trivial. Let us just denote m : = {x : |f (x)| > kf k } > 0.
(b) First,
1/n
kf kn kf kn (X)
= kf k (X)1/n .
Taking the limit n gives lim sup kf kn kf k . On the other hand, for any
> 0 we have
1/n
kf kn (kf k )n m
= (kf k )m1/n
.
Taking the limit n gives lim inf kf kn kf k . Since > 0 is arbitrary,
we have lim inf kf kn kf k . Thus lim kf kn = kf k .
Next, we have
Z
Z
n+1
n+1
kf kn+1 =
|f |
d
kf k |f |n d kf k kf knn ,
X
thus
lim sup
kf kn+1
n+1
kf k .
kf knn
1n+1 d
= kf knn+1 (X) n+1
X
37
hence
kf kn+1
kf kn+1
n+1
1
n
kf kn
(X) n+1
n+1
kf kn+1
kf kn+1
= kf k
lim
1
kf knn
(X) n+1
1
kf kn+1
n+1
kf kn
n
= kf k .
JPE, May 2001. Show by example that there exist two functions f L1 (R) and
g L2 (R) such that f + g is neither in L1 (R) nor in L2 (R).
f (x) =
1/ x if x (0, 1]
1/x
if x (1, )
38
JPE, May 1995. Let M be a subspace of L2 ([0, 1]) with the following property:
there is a constant C so that if f M, then |f (x)| Ckf k2 for a.e. x [0, 1].
Let f1 , . . . , fn be
set in M.
Pnan orthonormal
2
2
(a) Show that j=1 |fj (x)| C for a.e. x [0, 1].
(b) Show that the dimension of M is bounded above by C 2 .
R
(a) An orthonormal set f1 , . . . , fn satisfies [0,1] fi fj dm = ij , which is the Kronecker delta symbol (the bar here means complex conjugate). Now let c1 , . . . , cn
be any complex numbers. Then c1 f1 + + cn fn M, hence
1/2
|c1 f1 (x) + + cn fn (x)| Ckc1 f1 + + cn fn k2 = C |c1 |2 + + |cn |2
for a.e. x M. Setting c1 = f1 (x), . . . , cn = fn (x) and squaring gives
n
X
|fj (x)|2
2
C2
j=1
n
X
|fj (x)|2 ,
j=1
therefore f L1 (R).
Next, by the triangle inequality
|fn | |fn f | |f |.
Since f L1 (R), the Lebesgue Dominated Convergence gives
Z
Z
Z
Z
|fn | |fn f | =
|fn | |fn f | |f |
n
therefore
Z
kfn f k1 =
|fn f | A kf k1 .
f d
X
This looks very much like Jensens inequality, but a crucial assumption is missing: (X) = 1. Without this assumption, the statement is false.
= x2 ,
R Example:
X = [0, 10], the Lebesgue measure, and f 1. Then X f d = 102 = 100
R
and X ( f ) d = 10.
JPE, May 1993. Let f L2 (R). Prove that
Z n+1
lim
f dm = 0.
n
|f | dm.
n+1
n+1
2
1 dm
1/2
Z
n+1
2
|f | dm
1/2
.
40
[0,1]
[0,1]
[0,1]
[0,1]
f dm
i1/2
f 2 dm.
[0,1]
[0,1]
Thus we have an equality in the Schwarz inequality, which is only possible if the
two involved functions (in this case f and f 2 ) are proportional to each other. From
the relation af = bf 2 a.e. with some a, b R (at least one of which differs from
zero) we can easily show that f = f 2 a.e.
Note that f = f 2 a.e. implies that f only takes two values, 0 and 1 (a.e.), hence
f = A a.e. for a subset A [0, 1].
JPE, Sept 1989. Let p > 1 and f Lp [1, 1] , i.e.
Z
|f |p dm < .
[1,1]
Z
|f | dm = 0.
In
Z
|f | 1 dm
[1,1]
p
p1
1/q
|f | dm
1 dm
[1,1]
1/q
=2
where q =
1/p Z
[1,1]
kf kp < ,
1
q
= .
|f | 1 dm
In
1/p Z
|f | dm
In
1/q
1 dm
In
Thus
n
Z
|f | dm 2
|f | dm
In
In
41
2 Z
n
In
1/p
.
|f | dm
1/p
.
|f | dm = (In )
(
n=1 In )
|f |p dm = 0.
= ({0}) =
{0}
In
10
[0,1]
(b) Describe the class of functions for which the equality takes place.
p
(a) Denote g = 1 f 2 . Observe that f 2 + g 2 = 1 a.e. Now by the Schwarz
inequality, applied twice,
Z
Z
Z
hZ i 2 h Z i 2 Z
2
2
2
2
f +
g f + g = (f + g ) = 1 = 1.
(b) For strictly convex functions, equality in Jensens inequality occurs if and only
if the integrand is constant a.e., hence f must be constant a.e.
JPE, Sept 2011. Given that
Z
R
0
ex 3 + 2 sin x dx 2.
R
1
First of all, in the given formula 0 ex sin x dx
the integral is a Riemann
R = 2 x
integral. The corresponding Lebesgue integral (0,) e sin x dm has the same
value, because the function is absolutely integrable:
Z
Z
x
e sin x dx
ex dx = 1.
0
e
(0,)
Z
3 + 2 sin x dm
1/2 Z
dm
(0,)
e (3 + 2 sin x) dm
(0,)
11/2 (3 + 2 12 )1/2 = 2.
42
1/2
1+x
3
dx 6.
2
x
3
+1
22/3 = 6.
=
2
(ii) The Holder inequality does not turn into an equality here, because the func1
are obviously not proportional to one another.
tions 1+x
and x3/2
x3
JPE, Sept 2008. Find all functions g(x) L3 (0, 1) satisfying the equation
3
Z
xg(x) dx
[0,1]
4
=
25
3/2
g 3 (x) dx.
[0,1]
2/3 Z
[0,1]
2 2/3 Z
5
1/3
g (x) dx
dx
[0,1]
[0,1]
1/3
g (x) dx
[0,1]
3
3/2
Since the Holder inequality turns
into an equality, the functions g and x must
be proportional, i.e., g(x) = c x, where c C is an arbitrary constant.
x3
dx
(1 x)1/5
5
16
.
81
Z
0
x3
dx
(1 x)1/5
3/5 Z
Z
1
(x ) dx
dx
1/5 ]5/2
0 [(1 x)
0
Z 1
2/5
3/5 Z 1
1
5
=
dx
x dx
1/2
0
0 (1 x)
1 3/5 2 2/5 1 1/5
=
=
6
1
54
3 5/3
2/5
1 1/5
0.4503
54
0.723,
0.3916,
1/5
4 3 14 19
0 (1 x)
which can be easily found by change of variable x = 1y. One may wonder why we
care about applying Holder inequality to get a rough upper bound for something
that can be computed precisely by any Calculus-II student. Go figure...
JPE, Sept 2004. Prove that
Z 1
x4 + 4x2 + 3 dx
x2 + 3
2
10.
3
x2 + 1 dx
Z
1
2
1/2 Z
(x + 3) dx
r
r
10
4
2
=
10.
=
3
3
3
0
1
2
1/2
(x + 1) dx
0
JPE, May 2001. (a) Let a1 , . . . , an be positive numbers. Prove that their harmonic mean is bounded by their arithmetic mean, i.e.
X
1
n
n
1
1
1X
ak .
n k=1 ak
n k=1
(b) Characterize the vectors for which equality holds in (a).
44
k
2
f
ak
X
X
X
X f
k=1
k=1
(b) Equality holds whenever f is proportional to 1/f , i.e., ak = c/ak for some
constant c and all k = 1, . . . , n. This implies a1 = = an .
JPE, May 1999. (i) Prove that
/2
x sin x dx .
2 2
1/2 Z
/2
sin x dx
x dx
0
1/2
/2
1/2
1 2
=
11/2 = .
2 2
2 2
(ii) The Schwarz inequality does not turn into an equality here, because the functions x and sin x are obviously not proportional to one another.
JPE, Sept 1997. Show that
Z
x1/2
dx
(1 x)1/3
3
8
.
5
1 + x3
dx
x4
45
7
.
10
5 2
10
JPE, Sept 1993. Prove that
Z
ex 1 + sin x dx
3
.
2
x
x
x
e (1 + sin x) dx
e
e dx
1 + sin x dx
0
0
0
r
3
1/2
1 1/2
1 (1 + 2 ) =
.
2
R
Note that we needed to compute here the integral 0 ex sin x dx = 12 . Even
though this is a routine Calculus-II exercise, it certainly takes valuable time during the exam. At a later JPE (Sept 2011) the value of this elementary integral was
provided, to save time.
11
`p spaces
JPE, May 2000. Do there exist two sequences (an ) `1 and (bn ) `2 such that
(an + bn ) is neither in `1 nor in `2 ?
No. Any sequence in (an ) `1 is also in `2 . Indeed, there are only finitely many
2
an s such that |an | P
> 1, and for all the other terms we
P have2 |an | |an |. Hence
the convergence of
|an | implies the convergence of
|an | . Thus (an + bn ) will
2
be in ` .
JPE, May 1998. Does there exist a sequence (xn ) which is in `1+ for all > 0
but which is not in `1 ?
Yes, xn = 1/n.
JPE, May 1997. Suppose 1 p q . Prove or disprove: `p `q .
The inclusion is true. If p = , then q = and the inclusion is trivial. If
p < and (an ) `p , then there are only finitely many an s such that |an | > 1. This
46
47
1
3
< b 12 .)