Tutorial 9: L

p
-spaces, p [1, +] 1
9. L
p
-spaces, p [1, +]
In the following, (, F, ) is a measure space.
Exercise 1. Let f, g : (, F) [0, +] be non-negative and measurable maps.
Let p, q R
+
, such that 1/p + 1/q = 1.
1. Show that 1 < p < + and 1 < q < +.
2. For all ]0, +[, we dene

: [0, +] [0, +] by:

(x)

=
_
x

if x R
+
+ if x = +
Show that

is a continuous map.
3. Dene A = (
_
f
p
d)
1/p
, B = (
_
g
q
d)
1/q
and C =
_
fgd. Explain why
A, B and C are well dened elements of [0, +].
4. Show that if A = 0 or B = 0 then C AB.
5. Show that if A = + or B = + then C AB.
6. We assume from now on that 0 < A < + and 0 < B < +. Dene
F = f/A and G = g/B. Show that:
_

F
p
d =
_

G
p
d = 1
7. Let a, b ]0, +[. Show that:
ln(a) + ln(b) ln
_
1
p
a
p
+
1
q
b
q
_
and:
ab
1
p
a
p
+
1
q
b
q
Prove this last inequality for all a, b [0, +].
8. Show that for all , we have:
F()G()
1
p
(F())
p
+
1
q
(G())
q
9. Show that C AB.
Theorem 41 (Holder inequality) Let (, F, ) be a measure space and f, g :
(, F) [0, +] be two non-negative and measurable maps. Let p, q R
+
be
such that 1/p + 1/q = 1. Then:
_

fgd
__

f
p
d
_1
p
__

g
q
d
_1
q
www.probability.net
Tutorial 9: L
p
-spaces, p [1, +] 2
Theorem 42 (Cauchy-Schwarz inequality:rst)
Let (, F, ) be a measure space and f, g : (, F) [0, +] be two non-negative
and measurable maps. Then:
_

fgd
__

f
2
d
_1
2
__

g
2
d
_1
2
Exercise 2. Let f, g : (, F) [0, +] be two non-negative and measurable
maps. Let p ]1, +[. Dene A = (
_
f
p
d)
1/p
and B = (
_
g
p
d)
1/p
and
C = (
_
(f +g)
p
d)
1/p
.
1. Explain why A, B and C are well dened elements of [0, +].
2. Show that for all a, b ]0, +[, we have:
(a +b)
p
2
p1
(a
p
+b
p
)
Prove this inequality for all a, b [0, +].
3. Show that if A = + or B = + or C = 0 then C A +B.
4. Show that if A < + and B < + then C < +.
5. We assume from now that A, B [0, +[ and C ]0, +[. Show the
existence of some q R
+
such that 1/p + 1/q = 1.
6. Show that for all a, b [0, +], we have:
(a +b)
p
= (a +b).(a +b)
p1
7. Show that:
_

f.(f +g)
p1
d AC
p
q
_

g.(f +g)
p1
d BC
p
q
8. Show that: _

(f +g)
p
d C
p
q
(A +B)
9. Show that C A +B.
10. Show that the inequality still holds if we assume that p = 1.
Theorem 43 (Minkowski inequality) Let (, F, ) be a measure space and
f, g : (, F) [0, +] be two non-negative and measurable maps. Let p
[1, +[. Then:
__

(f +g)
p
d
_1
p

__

f
p
d
_1
p
+
__

g
p
d
_1
p
www.probability.net
Tutorial 9: L
p
-spaces, p [1, +] 3
Denition 73 The L
p
-spaces, p [1, +[, on (, F, ), are:
L
p
R
(, F, )

=
_
f : (, F)(R, B(R)) measurable,
_

|f|
p
d <+
_
L
p
C
(, F, )

=
_
f : (, F)(C, B(C)) measurable,
_

|f|
p
d <+
_
For all f L
p
C
(, F, ), we put:
f
p

=
__

|f|
p
d
_1
p
Exercise 3. Let p [1, +[. Let f, g L
p
C
(, F, ).
1. Show that L
p
R
(, F, ) = {f L
p
C
(, F, ) , f() R}.
2. Show that L
p
R
(, F, ) is closed under R-linear combinations.
3. Show that L
p
C
(, F, ) is closed under C-linear combinations.
4. Show that f +g
p
f
p
+g
p
.
5. Show that f
p
= 0 f = 0 -a.s.
6. Show that for all C, f
p
= ||.f
p
.
7. Explain why (f, g) f g
p
is not a metric on L
p
C
(, F, )
Denition 74 For all f : (, F) (C, B(C)) measurable, Let:
f

= inf{M R
+
, |f| M -a.s.}
The L

-spaces on a measure space (, F, ) are:

L

R
(, F, )

={f : (, F) (R, B(R)) measurable, f

< +}
L

C
(, F, )

={f : (, F) (C, B(C)) measurable, f

< +}
Exercise 4. Let f, g L

C
(, F, ).
1. Show that L

R
(, F, ) = {f L

C
(, F, ) , f() R}.
2. Show that |f| f

-a.s.
3. Show that f +g

+g

.
4. Show that L

R
(, F, ) is closed under R-linear combinations.
5. Show that L

C
(, F, ) is closed under C-linear combinations.
6. Show that f

= 0 f = 0 -a.s..
www.probability.net
Tutorial 9: L
p
-spaces, p [1, +] 4
7. Show that for all C, f

= ||.f

.
8. Explain why (f, g) f g

is not a metric on L

C
(, F, )
Denition 75 Let p [1, +]. Let K = R or C. For all > 0 and f
L
p
K
(, F, ), we dene the so-called open ball in L
p
K
(, F, ):
B(f, )

= {g : g L
p
K
(, F, ), f g
p
< }
We call usual topology in L
p
K
(, F, ), the set T dened by:
T

= {U : U L
p
K
(, F, ), f U, > 0, B(f, ) U}
Note that if (f, g) f g
p
was a metric, the usual topology in L
p
K
(, F, ),
would be nothing but the metric topology.
Exercise 5. Let p [1, +]. Suppose there exists N F with (N) = 0 and
N = . Put f = 1
N
and g = 0
1. Show that f, g L
p
C
(, F, ) and f = g.
2. Show that any open set containing f also contains g.
3. Show that L
p
C
(, F, ) and L
p
R
(, F, ) are not Hausdor.
Exercise 6. Show that the usual topology on L
p
R
(, F, ) is induced by the
usual topology on L
p
C
(, F, ), where p [1, +].
Denition 76 Let (E, T ) be a topological space. A sequence (x
n
)
n1
in E is
said to converge to x E, and we write x
n
T
x, if and only if, for all U T
such that x U, there exists n
0
1 such that:
n n
0
x
n
U
When E = L
p
C
(, F, ) or E = L
p
R
(, F, ), we write x
n
L
p
x.
Exercise 7. Let (E, T ) be a topological space and E

E. Let T

= T
|E

be the induced topology on E

. Show that if (x
n
)
n1
is a sequence in E

and
x E

, then x
n
T
x is equivalent to x
n
T

x.
Exercise 8. Let f, g, (f
n
)
n1
be in L
p
C
(, F, ) and p [1, +].
1. Recall what the notation f
n
f means.
2. Show that f
n
L
p
f is equivalent to f
n
f
p
0.
3. Show that if f
n
L
p
f and f
n
L
p
g then f = g -a.s.
Exercise 9. Let p [1, +]. Suppose there exists some N F such that
(N) = 0 and N = . Find a sequence (f
n
)
n1
in L
p
C
(, F, ) and f, g in
L
p
C
(, F, ), f = g such that f
n
L
p
f and f
n
L
p
g.
www.probability.net
Tutorial 9: L
p
-spaces, p [1, +] 5
Denition 77 Let (f
n
)
n1
be a sequence in L
p
C
(, F, ), where (, F, ) is a
measure space and p [1, +]. We say that (f
n
)
n1
is a Cauchy sequence,
if and only if, for all > 0, there exists n
0
1 such that:
n, m n
0
f
n
f
m

p

Exercise 10. Let f, (f
n
)
n1
be in L
p
C
(, F, ) and p [1, +]. Show that if
f
n
L
p
f, then (f
n
)
n1
is a Cauchy sequence.
Exercise 11. Let (f
n
)
n1
be Cauchy in L
p
C
(, F, ), p [1, +].
1. Show the existence of n
1
1 such that:
n n
1
f
n
f
n1

p

1
2
2. Suppose we have found n
1
< n
2
< . . . < n
k
, k 1, such that:
j {1, . . . , k} , n n
j
f
n
f
nj

p

1
2
j
Show the existence of n
k+1
, n
k
< n
k+1
such that:
n n
k+1
f
n
f
n
k+1

p

1
2
k+1
3. Show that there exists a subsequence (f
n
k
)
k1
of (f
n
)
n1
with:
+

k=1
f
n
k+1
f
n
k

p
< +
Exercise 12. Let p [1, +], and (f
n
)
n1
be in L
p
C
(, F, ), with:
+

n=1
f
n+1
f
n

p
< +
We dene:
g

=
+

n=1
|f
n+1
f
n
|
1. Show that g : (, F) [0, +] is non-negative and measurable.
2. If p = +, show that g

+
n=1
f
n+1
f
n

-a.s.
3. If p [1, +[, show that for all N 1, we have:
_
_
_
_
_
N

n=1
|f
n+1
f
n
|
_
_
_
_
_
p

n=1
f
n+1
f
n

p
www.probability.net
Tutorial 9: L
p
-spaces, p [1, +] 6
4. If p [1, +[, show that:
__

g
p
d
_1
p

n=1
f
n+1
f
n

p
5. Show that for p [1, +], we have g < + -a.s.
6. Dene A = {g < +}. Show that for all A, (f
n
())
n1
is a Cauchy
sequence in C. We denote z() its limit.
7. Dene f : (, F) (C, B(C)), by:
f()

=
_
z() if A
0 if A
Show that f is measurable and f
n
f -a.s.
8. if p = +, show that for all n 1, |f
n
| |f
1
| + g and conclude that
f L

C
(, F, ).
9. If p [1, +[, show the existence of n
0
1, such that:
n n
0

_

|f
n
f
n0
|
p
d 1
Deduce from Fatou lemma that f f
n0
L
p
C
(, F, ).
10. Show that for p [1, +], f L
p
C
(, F, ).
11. Suppose that f
n
L
p
R
(, F, ), for all n 1. Show the existence of
f L
p
R
(, F, ), such that f
n
f -a.s.
Exercise 13. Let p [1, +], and (f
n
)
n1
be in L
p
C
(, F, ), with:
+

n=1
f
n+1
f
n

p
< +
1. Does there exist f L
p
C
(, F, ) such that f
n
f -a.s.
2. Suppose p = +. Show that for all n < m, we have:
|f
m+1
f
n
|
m

k=n
f
k+1
f
k

-a.s.
3. Suppose p = +. Show that for all n 1, we have:
f f
n

k=n
f
k+1
f
k

www.probability.net
Tutorial 9: L
p
-spaces, p [1, +] 7
4. Suppose p [1, +[. Show that for all n < m, we have:
__

|f
m+1
f
n
|
p
d
_1
p

k=n
f
k+1
f
k

p
5. Suppose p [1, +[. Show that for all n 1, we have:
f f
n

p

+

k=n
f
k+1
f
k

p
6. Show that for p [1, +], we also have f
n
L
p
f.
7. Suppose conversely that g L
p
C
(, F, ) is such that f
n
L
p
g. Show that
f = g -a.s.. Conclude that f
n
g -a.s..
Theorem 44 Let (, F, ) be a measure space. Let p [1, +], and (f
n
)
n1
be a sequence in L
p
C
(, F, ) such that:
+

n=1
f
n+1
f
n

p
< +
Then, there exists f L
p
C
(, F, ) such that f
n
f -a.s. Moreover, for all
g L
p
C
(, F, ), the convergence f
n
g -a.s. and f
n
L
p
g are equivalent.
Exercise 14. Let f, (f
n
)
n1
be in L
p
C
(, F, ) such that f
n
L
p
f, where p
[1, +].
1. Show that there exists a sub-sequence (f
n
k
)
k1
of (f
n
)
n1
, with:
+

k=1
f
n
k+1
f
n
k

p
< +
2. Show that there exists g L
p
C
(, F, ) such that f
n
k
g -a.s.
3. Show that f
n
k
L
p
g and g = f -a.s.
4. Conclude with the following:
Theorem 45 Let (f
n
)
n1
be in L
p
C
(, F, ) and f L
p
C
(, F, ) such that
f
n
L
p
f, where p [1, +]. Then, we can extract a sub-sequence (f
n
k
)
k1
of
(f
n
)
n1
such that f
n
k
f -a.s.
Exercise 15. Prove the last theorem for L
p
R
(, F, ).
Exercise 16. Let (f
n
)
n1
be Cauchy in L
p
C
(, F, ), p [1, +].
www.probability.net
Tutorial 9: L
p
-spaces, p [1, +] 8
1. Show that there exists a subsequence (f
n
k
)
k1
of (f
n
)
n1
and f belonging
to L
p
C
(, F, ), such that f
n
k
L
p
f.
2. Using the fact that (f
n
)
n1
is Cauchy, show that f
n
L
p
f.
Theorem 46 Let p [1, +]. Let (f
n
)
n1
be a Cauchy sequence in L
p
C
(, F, ).
Then, there exists f L
p
C
(, F, ) such that f
n
L
p
f.
Exercise 17. Prove the last theorem for L
p
R
(, F, ).
www.probability.net
Solutions to Exercises 9
Solutions to Exercises
Exercise 1.
1. Since p, q R
+
, we have p < + and q < +. From the inequality
1/p 1/p+1/q = 1, we obtain p 1. If p = 1, then 1/q = 0, contradicting
q < +. So p > 1, and similarly q > 1. We have proved that 1 < p < +
and 1 < q < +.
2. Let ]0, +[ and =

. We want to prove that is continuous. For

all a R
+
, it is clear that lim
xa
(x) = (a). So is continuous at
x = a. Furthermore, lim
x+
(x) = (+). So is also continuous
at +. For many of us, this is sucient proof of the fact that is a
continuous map. However, for those who want to apply denition (27),
the following can be said: let V be open in [0, +]. We want to show
that
1
(V ) is open in [0, +]. Let a
1
(V ). Then (a) V . Since
is continuous at x = a, there exists U
a
open in [0, +], containing a,
such that (U
a
) V . So a U
a

1
(V ). It follows that
1
(V ) can
be written as
1
(V ) =
a
1
(V )
U
a
, and
1
(V ) is therefore open in
[0, +]. From denition (27), we conclude that : [0, +] [0, +] is
a continuous map.
3. f
p
can be viewed as f
p
=
p
f, where
p
is dened as in 2. We proved that

p
is a continuous map. It is therefore measurable with respect to the Borel
-algebra B([0, +]) on [0, +]. It follows that f
p
: (, F) [0, +] is a
measurable map, which is also non-negative. Hence, the integral
_
f
p
d is
a well-dened element of [0, +], and A = (
_
f
p
d)
1/p
is also well-dened,
being understood that (+)
1/p
= +. Similarly, B = (
_
f
q
d)
1/q
is a
well-dened element of [0, +]. Finally, the map fg : (, F) [0, +]
is non-negative and measurable, and C =
_
fgd is a well-dened element
of [0 +].
4. Suppose A = 0. Then
_
f
p
d = 0, and since f
p
is non-negative, we see
that f
p
= 0 -a.s., and consequently f = 0 -a.s. So fg = 0 -a.s., and
nally C =
_
fgd = 0. So C AB. Similarly, B = 0 implies C = 0, and
therefore C AB.
5. Suppose A = +. Then, either B = 0 or B > 0. If B = 0, then C AB
is true from 4. If B > 0, then AB = +, and consequently C AB. In
any case, we see that C AB. Similarly, B = + implies C AB.
6. Suppose A, B ]0, +[. Let F = f/A and G = g/B. We have:
_
F
p
d =
_
(f/A)
p
d =
1
A
p
_
f
p
d = 1
and similarly,
_
G
p
d = 1.
www.probability.net
Solutions to Exercises 10
7. Let a, b ]0, +[. The map x ln(x) being convex on ]0, +[, since
1/p + 1/q = 1, we have:
ln(
1
p
a
p
+
1
q
b
q
)
1
p
ln(a
p
)
1
q
ln(b
q
) = ln(ab)
and consequently ln(ab) ln(a
p
/p + b
q
/q). The map x e
x
being non-
decreasing, we conclude that:
ab
1
p
a
p
+
1
q
b
q
(1)
It is easy to check that inequality (1) is in fact true for all a, b [0, +].
8. For all , F() and G() are elements of [0, +]. From 7.:
F()G()
1
p
F()
p
+
1
q
G()
q
9. Integrating on both side of 8., we obtain:
_
FGd
1
p
_
F
p
d +
1
q
_
G
q
d = 1
where we have used the fact that
_
F
p
d =
_
G
q
d = 1. Since
_
FGd =
(
_
fgd)/AB = C/AB, we conclude that C AB.
Exercise 1
Exercise 2.
1. f
p
, g
p
and (f +g)
p
are all non-negative and measurable. All three integrals
_
f
p
d,
_
g
p
d and
_
(f +g)
p
d are therefore well-dened. It follows that
A, B and C are well-dened elements of [0, +].
2. Since p > 1, the map x x
p
is convex on ]0, +[. In particular, for all
a, b ]0, +[, we have ((a +b)/2)
p
(a
p
+b
p
)/2. We conclude that:
(a +b)
p
2
p1
(a
p
+b
p
) (2)
In fact, it is easy to check that (2) holds for all a, b [0, +].
3. If A = + or B = +, then A + B = +, and C A + B. If C = 0,
then clearly C A +B.
4. Using 2., for all , we have:
(f() +g())
p
2
p1
(f()
p
+g()
p
)
Integrating on both side of the inequality, we obtain:
_
(f +g)
p
d 2
p1
__
f
p
d +
_
g
p
d
_
(3)
If A < + and B < +, then both integrals
_
f
p
d and
_
g
p
d are
nite, and we see from (3) that
_
(f +g)
p
d is itself nite. So C < +.
www.probability.net
Solutions to Exercises 11
5. Take q = p/(p 1). Since p ]1, +[, q is a well-dened element of R
+
,
and 1/p + 1/q = 1.
6. Let a, b [0, +]. If a, b R
+
, then:
(a +b)
p
= (a +b).(a +b)
p1
(4)
If a = + or b = +, then a + b = + and both sides of (4) are equal
to +. So (4) is true for all a, b [0, +].
7. Using holder inequality (41), since q(p 1) = p, we have:
_
f.(f +g)
p1
d
__
f
p
d
_1
p
__
(f +g)
q(p1)
d
_1
q
= AC
p
q
and:
_
g.(f +g)
p1
d
__
g
p
d
_1
p
__
(f +g)
q(p1)
d
_1
q
= BC
p
q
8. From 6., we have:
_
(f +g)
p
d =
_
f.(f +g)
p1
d +
_
g.(f +g)
p1
d
and using 7., we obtain:
_
(f +g)
p
d C
p
q
(A +B)
9. From 8., we have C
p
C
p
q
(A+B). Having assumed in 5. that C ]0, +[,
we can divide both side of this inequality by C
p
q
, to obtain C
p
p
q
A+B.
Since p p/q = 1, we conclude that C A +B.
10. If p = 1, then C = A +B is equivalent to:
_
(f + g)d =
_
fd +
_
gd
which is true by linearity. In particular, C A +B. The purpose of this
exercise is to prove minkowski inequality (43).
Exercise 2
Exercise 3.
1. Let f : (, F) (C, B(C)) be a map. Then, if f has values in R,
i.e. f() R, then the measurability of f with respect to (C, B(C)) is
equivalent to its measurability with respect to (R, B(R)). Hence:
L
p
R
(, F, ) = {f L
p
C
(, F, ) , f() R}
The equivalence of measurability with respect to B(C) and B(R) may
be taken for granted by many. It is easily proved from the fact that
B(R) = B(C)
|R
, i.e. the Borel -algebra on R is the trace on R, of the
www.probability.net
Solutions to Exercises 12
Borel -algebra on C. This fact can be seen from the trace theorem (10),
and the fact that the usual topology on R is induced on R, by the usual
topology on C.
2. Let f, g L
p
R
(, F, ) and R. The map f + g is R-valued and
measurable. Moreover, we have |f +g| |f|+||.|g|. Since p 1, (and in
particular p 0), the map x x
p
is non-decreasing on R
+
, so |f +g|
p

(|f| + ||.|g|)
p
. Hence, we see that
_
|f + g|
p
d
_
(|f| + ||.|g|)
p
d.
However, using minkowski inequality (43), we have:
__
(|f| +||.|g|)
p
d
_1
p

__
|f|
p
d
_1
p
+||.
__
|g|
p
d
_1
p
We conclude that
_
|f +g|
p
d < +. So f +g L
p
R
(, F, ), and we
have proved that L
p
R
(, F, ) is closed under R-linear combinations.
3. The fact that L
p
C
(, F, ) is closed under C-linear combinations, is proved
identically to 2., replacing R by C.
4. Using |f +g|
p
(|f| +|g|)
p
and minkowski inequality (43):
__
(|f| +|g|)
p
d
_1
p

__
|f|
p
d
_1
p
+
__
|g|
p
d
_1
p
we see that f +g
p
f
p
+g
p
.
5. Suppose f
p
= 0. Then
_
|f|
p
d = 0. Since |f|
p
is non-negative, |f|
p
= 0
-a.s., and consequently f = 0 -a.s. Conversely, if f = 0 -a.s., then
|f|
p
= 0 -a.s., so
_
|f|
p
d = 0 and nally f
p
= 0.
6. Let C. We have:
f
p
=
__
|f|
p
_1
p
= ||.
__
|f|
p
_1
p
= ||.f
p
7. f g
p
= 0 only implies f = g -.a.s, and not necessarily f = g. So
(f, g) f g
p
, may not be a metric on L
p
C
(, F, ).
Exercise 3
Exercise 4.
1. For all f : (, F) (C, B(C)) with values in R, the measurability of
f with respect to B(C) is equivalent to its measurability with respect to
B(R). Hence:
L

R
(, F, ) = {f L

C
(, F, ) , f() R}
2. Since f

< +, for all n 1, we have f

< f

+ 1/n. f

being the greatest lower bound of all -almost sure upper bounds of |f|,
f

+1/n cannot be such lower bound. There exists M R

+
, such that
www.probability.net
Solutions to Exercises 13
|f| M -a.s., and M < f

+1/n. In particular, |f| < f

+1/n -
a.s. Let A
n
be the set dened by A
n
= {f

+1/n |f|}. Then A

n
F
and (A
n
) = 0. Moreover, A
n
A
n+1
and
+
n=1
A
n
= {f

< |f|}. It
follows that A
n
{f

< |f|}, and from theorem (7), we see that:

({f

< |f|}) = lim

n+
(A
n
) = 0
We conclude that |f| f

-a.s.
3. Since |f +g| |f| +|g|, using 2., we have:
|f +g| f

+g

-a.s.
Hence, f

+g

is a -almost sure upper bound of |f +g|. f +g

being a lower bound of all such upper bounds, we have f + g

+g

.
4. Let f, g L

R
(, F, ) and R. Then f +g is R-valued and measur-
able. Furthermore, using 2., we have:
|f +g| |f| +||.|g| f

+||.g

-a.s.
It follows that f + g

+ ||.g

< +. We conclude that

f + g L

R
(, F, ), and we have proved that L

R
(, F, ) is closed
under R-linear combinations.
5. The fact that L

C
(, F, ) is closed under C-linear combinations can be
proved identically, replacing R by C.
6. Suppose f

= 0. Then |f| 0 -a.s., and consequently f = 0 -a.s.

Conversely, if f = 0 -a.s., then |f| 0 -a.s., and 0 is therefore a -
almost sure upper bound of |f|. So f

0. Since f

is an inmum
of a subset of R
+
, it is either + (when such subset is empty), or lies in
R
+
. So f

0 and nally f

= 0.
7. We have |f| ||.f

-a.s., and hence f

||.f

. if = 0,
we have:
f

=
1

.(f)

1
||
f

It follows that f

= ||.f

, (also true if = 0).

8. f g

= 0 implies f = g -a.s., but not f = g. It follows that

(f, g) f g

may not be a metric on L

C
(, F, ).
Exercise 4
Exercise 5.
1. Since N = , 1
N
= 0, so f = g. Since N F, the map f = 1
N
is
measurable, and being R-valued, it is also C-valued. Moreover, since
(N) = 0, f
p
= 0 < + (whether p = + or lies in [1, +[), and
we see that f L
p
C
(, F, ). Since g = 0, it is C-valued, measurable and
g
p
= 0 < +, so g L
p
C
(, F, ).
www.probability.net
Solutions to Exercises 14
2. Let U be open in L
p
C
(, F, ), such that f U. By denition (75), there
exists > 0, such that B(f, ) U. However, f g
p
= f
p
= 0
(p = + or p [1, +[). So in particular f g
p
< . So g B(f, )
and nally g U.
3. If L
p
C
(, F, ) was Hausdor, since f = g, there would exist U, V open
sets in L
p
C
(, F, ) such that f U, g V and U V = . However from
2., this is impossible, as g would always be an element of U as well as V .
We conclude similarly that L
p
R
(, F, ) is not Hausdor.
Exercise 5
Exercise 6. Let L
p
R
and L
p
C
denote L
p
R
(, F, ) and L
p
C
(, F, ) respectively.
Let T be the usual topology on L
p
C
and T

be the usual topology on L

p
R
. We
want to prove that T

= T
|L
p
R
, i.e. that T

is the topology on L
p
R
induced by
T . Given f L
p
R
and > 0, let B(f, ) denote the open ball in L
p
C
and B

(f, )
denote the open ball the L
p
R
. Then B

(f, ) = B(f, ) L
p
R
. It is a simple
exercise to show that any open ball in L
p
R
or L
p
C
, is in fact open with respect
to their usual topology. Let U

. For all f U

, there exists
f
> 0 such
that f B

(f,
f
) U

. It follows that:
U

=
fU
B

(f,
f
) = (
fU
B(f,
f
)) L
p
R
and we see that U

T
|L
p
R
. So T

T
|L
p
R
. Conversely, let U

T
|L
p
R
. There
exists U T such that U

= U L
p
R
. Let f U

. Then f U. There exists

> 0 such that B(f, ) U. It follows that B

(f, ) = B(f, ) L
p
R
U

. So U

is open with respect to the usual topology in L

p
R
, i.e. U

T

. We have proved
that T
|L
p
R
T

, and nally T

= T
|L
p
R
.
Exercise 6
Exercise 7. let (E, T ) be a topological space and E

E. Let T

= T
|E
be the
induced topology on E

. We assume that (x
n
)
n1
is a sequence in E

, and that
x E

. Suppose that x
n
T
x. Let U

T

be such that x U

. There exists
U T such that U

= U E

. Since x U and x
n
T
x, there exists n
0
1
such that x
n
U for all n n
0
. But x
n
E

for all n 1. So x
n
U E

= U

for all n n
0
, and we see that x
n
T

x. Conversely, suppose that x

n
T

x. Let
U T be such that x U. Then U E

and x U E

. There exists
n
0
1, such that x
n
U E

for all n n
0
. In particular, x
n
U for all
n n
0
, and we see that x
n
T
x. We have proved that x
n
T

x and x
n
T
x are
equivalent.
Exercise 7
Exercise 8.
1. The notation f
n
f has been used throughout these tutorials, to refer
to a simple convergence, i.e. f
n
() f() as n +, for all .
www.probability.net
Solutions to Exercises 15
2. Suppose f
n
L
p
f. Let > 0. The open ball B(f, ) being open with
respect to the usual topology in L
p
C
(, F, ), there exists n
0
1, such
that f
n
B(f, ) for all n n
0
, i.e.:
n n
0
f
n
f
p
<
So f
n
f
p
0. Conversely, suppose f
n
f
p
0. Let U be open
in L
p
C
(, F, ), such that f U. From denition (75), there exists > 0
such that B(f, ) U. By assumption, there exists n
0
0, such that
f
n
f
p
< for all n n
0
. So f
n
B(f, ) for all n n
0
. Hence, we
see that f
n
U for all n n
0
, and we have proved that f
n
L
p
f. We
conclude that f
n
L
p
f and f
n
f
p
0 are equivalent.
3. Suppose f
n
L
p
f and f
n
L
p
g. From 2., we have f
n
f
p
0 and
f
n
g
p
0. Using the triangle inequality (ex. (3) for p [1, +[ and
ex. (4) for p = +):
f g
p
f
n
f
p
+f
n
g
p
for all n 1. It follows that we have f g
p
< for all > 0, and
consequently f g
p
= 0. From ex. (3) and ex. (4) we conclude that
f = g -a.s.
Exercise 8
Exercise 9. Take f
n
= 1
N
= f for all n 1. Take g = 0. Then f
n
, f and g
are all elements of L
p
C
(, F, ), and f = g. Moreover, for all n 1, we have
f
n
f
p
= f
n
g
p
= 0. So f
n
L
p
f and f
n
L
p
g. The purpose of this exercise
is to show that a limit in L
p
may not be unique (f = g). However, it is unique,
up to -almost sure equality (See exercise (8)).
Exercise 9
Exercise 10. Suppose f
n
L
p
f. Let > 0. There exists n
0
1, with:
n n
0
f
n
f
p
/2
From the triangle inequality, for all n, m 1:
f
n
f
m

p
f
n
f
p
+f
m
f
p
It follows that we have:
n, m n
0
f
n
f
m

p

We conclude that (f
n
)
n1
is a Cauchy sequence in L
p
C
(, F, ).
Exercise 10
Exercise 11.
1. Take = 1/2. There exists n
1
1, such that:
n, m n
1
f
n
f
m

p

1
2
www.probability.net
Solutions to Exercises 16
In particular, we have:
n n
1
f
n
f
n1

p

1
2
2. Let k 1. We have n
1
< . . . < n
k
, such that for all j = 1, . . . , k:
n n
j
f
n
f
nj

p

1
2
j
Take = 1/2
k+1
. There exists n

k+1
1, such that:
n, m n

k+1
f
n
f
m

p

1
2
k+1
Take n
k+1
= max(n
k
+ 1, n

k+1
). Then n
k
< n
k+1
, and:
n n
k+1
f
n
f
n
k+1

p

1
2
k+1
3. By induction from 2., we can construct a strictly increasing sequence of
integers (n
k
)
k1
, such that for all k 1:
n n
k
f
n
f
n
k

p

1
2
k
In particular, f
n
k+1
f
n
k

p
1/2
k
for all k 1. It follows that we have
found a subsequence (f
n
k
)
k1
of (f
n
)
n1
, such that:
+

k=1
f
n
k+1
f
n
k

p
< +
Exercise 11
Exercise 12.
1. Each nite sum g
N
=

N
n=1
|f
n+1
f
n
| is well-dened and measurable.
It follows that g = sup
N1
g
N
is itself measurable. It is obviously non-
negative.
2. Suppose p = +. From exercise (4), for all n 1, we have:
|f
n+1
f
n
| f
n+1
f
n

, -a.s.
The set N
n
= {|f
n+1
f
n
| > f
n+1
f
n

} which lies in F, is such that

(N
n
) = 0. It follows that if N =
n1
N
n
, then (N) = 0. However, for
all N
c
, we have:
g() =
+

n=1
|f
n+1
() f
n
()|
+

n=1
f
n+1
f
n

We conclude that g

n=1
f
n+1
f
n

-a.s.
www.probability.net
Solutions to Exercises 17
3. Let p [1, +[ and N 1. By the triangle inequality (ex. (3)):
_
_
_
_
_
N

n=1
|f
n+1
f
n
|
_
_
_
_
_
p

n=1
f
n+1
f
n

p

+

n=1
f
n+1
f
n

p
4. Let p [1, +[. Given N 1, let g
N
=

N
n=1
|f
n+1
f
n
|. Then g
N
g
as N +. The map x x
p
being continuous on [0, +], we have
g
p
N
g
p
, and in particular g
p
= liminf g
p
N
as N +. Using Fatou
lemma (20), we see that:
_
g
p
d liminf
N1
_
g
p
N
d (5)
However, from 3., we have g
N

+
n=1
f
n+1
f
n

p
, for all N 1.
Since p 0, the map x x
p
is non-decreasing on [0, +], and therefore:
_
g
p
N
d
_
+

n=1
f
n+1
f
n

p
_
p
(6)
From inequalities (5) and (6), we conclude that:
_
g
p
d
_
+

n=1
f
n+1
f
n

p
_
p
and nally:
__
g
p
d
_1
p

n=1
f
n+1
f
n

p
5. Let p [1, +]. If p = +, from 2. we have:
g
+

n=1
f
n+1
f
n

p
, -a.s. (7)
By assumption, the series in (7) is nite. So g < + -a.s.
If p [1, +[, from 4. we have:
__
g
p
d
_1
p

n=1
f
n+1
f
n

p
So
_
g
p
d < +. Since (+)({g
p
= +})
_
g
p
d, we see that
({g
p
= +}) = 0 and nally g < + -a.s.
6. Let A = {g < +}. Let A. Then g() < +. The series

+
n=1
|f
n+1
() f
n
()| is therefore nite. Let > 0. There exists n
0
1,
such that:
n n
0

+

k=n
|f
k+1
() f
k
()|
www.probability.net
Solutions to Exercises 18
Given m > n n
0
, we have:
|f
m
() f
n
()|
m1

k=n
|f
k+1
() f
k
()|
We conclude that the sequence (f
n
())
n1
is Cauchy in C. It therefore
has a limit
1
, denoted z().
7. From 6., f
n
() z() = f() for all A. Since by denition, f() = 0
for all A
c
, we see that f
n
()1
A
() f() for all . Hence, we
have f
n
1
A
f, and since f
n
1
A
is measurable for all n 1, we see from
theorem (17) that f = limf
n
1
A
is itself measurable. Since (A
c
) = 0 and
f
n
() f() on A, we have f
n
f -a.s.
8. Suppose p = +. For all n 1, we have:
|f
n
f
1
|
n1

k=1
|f
k+1
f
k
| g
So |f
n
| |f
1
| +g. Taking the limit as n +, we obtain |f| |f
1
| +g
-a.s. Let M =

+
n=1
f
n+1
f
n

. Then by assumption, M < + and

from 2. we have g M -a.s. Moreover, since f
1
L

C
(, F, ), using
exercise (4), we have |f
1
| f
1

-a.s. with f
1

< +. Hence, we
see that |f| f
1

+ M -a.s., and consequently:

f

f
1

+
+

n=1
f
n+1
f
n

< +
f is therefore C-valued, measurable and with f

< +. We have
proved that f L

C
(, F, ).
9. Let p [1, +[. The series

+
n=1
f
n+1
f
n

p
being nite, there exists
n
0
1, such that:
n n
0

+

k=n
f
k+1
f
k

p
1
Let n n
0
. By the triangle inequality:
f
n
f
n0

p

n1

k=n0
f
k+1
f
k

p
1
Hence, we see that:
n n
0

_
|f
n
f
n0
|
p
d 1
p
= 1 (8)
1
The completeness of C is proved in the next Tutorial.
www.probability.net
Solutions to Exercises 19
From 6., f
n
() f() as n +, for all A, where (A
c
) = 0. In
particular:
1
A
|f f
n0
|
p
= liminf
n1
1
A
|f
n
f
n0
|
p
Using inequality (8) and Fatou lemma (20), we obtain:
2
_
|f f
n0
|
p
d liminf
n1
_
|f
n
f
n0
|
p
d 1
In particular,
_
|f f
n0
|
p
d < +. Since f f
n0
is C-valued and mea-
surable, we conclude that f f
n0
L
p
C
(, F, ).
10. Let p [1, +]. If p = +, then f L

C
(, F, ) was proved in 8. If
p [1, +[, we saw in 9. that f f
n0
L
p
C
(, F, ) for some n
0
1.
Since f
n0
is itself an element of L
p
C
(, F, ), we conclude from exercise (3)
that f = (f f
n0
) +f
n0
is also an element of L
p
C
(, F, ).
11. The purpose of this exercise is to prove that given a sequence (f
n
)
n1
in L
p
C
(, F, ) such that

+
n=1
f
n+1
f
n

p
< +, there exists f
L
p
C
(, F, ), such that f
n
f -a.s. We now assume that all f
n
s are
in fact R-valued, i.e. f
n
L
p
R
(, F, ). There exists f

L
p
C
(, F, )
such that f
n
f

-a.s. However, f

() may not be R-valued for all

. Yet, if N F is such that (N) = 0 and f
n
() f

()
for all N
c
, then f

is R-valued on N
c
(as a limit of an R-valued
sequence). If we dene f = f

1
N
c, then f is R-valued and measurable,
with f
p
= f

p
< +. So f L
p
R
(, F, ) and furthermore since
f = f

-a.s., f
n
f -a.s.
Exercise 12
Exercise 13.
1. Yes, there does exist f L
p
C
(, F, ) such that f
n
f -a.s. This was
precisely the object of the previous exercise.
2. Suppose p = +, and let n < m. From exercise (4), we have |f
m+1

f
n
| f
m+1
f
n

-a.s. Furthermore, from the triangle inequality,

f
m+1
f
n

m
k=n
f
k+1
f
k

. It follows that:
|f
m+1
f
n
|
m

k=n
f
k+1
f
k

, -a.s. (9)
3. Suppose p = + and let n 1. For all m > n, let N
m
F be such that
(N
m
) = 0, and inequality (9) holds for all N
c
m
. Furthermore, since
f
m+1
f -a.s., let M F be such that (M) = 0, and f
m+1
() f()
for all M
c
. Then, if N = M (
m>n
N
m
), we have N F, (N) = 0
and for all N
c
, f
m+1
() f(), together with, for all m > n:
|f
m+1
() f
n
()|
m

k=n
f
k+1
f
k

2
Note that n n
0
un 1 is enough to ensure liminf
n1
un 1.
www.probability.net
Solutions to Exercises 20
Taking the limit as m +, we obtain:
|f() f
n
()|
+

k=n
f
k+1
f
k

This being true for all N

c
, we have proved that:
|f f
n
|
+

k=n
f
k+1
f
k

, -a.s.
From denition (74), we conclude that:
f f
n

k=n
f
k+1
f
k

4. Let p [1, +[ and n < m. From exercise (3), we have:

__
|f
m+1
f
n
|
p
d
_1
p
= f
m+1
f
n

p

m

k=n
f
k+1
f
k

p
5. Let p [1, +[ and n 1. Let N F be such that (N) = 0, and
f
m+1
() f() for all N
c
. Then, we have:
|f f
n
|
p
1
N
c = liminf
m>n
|f
m+1
f
n
|
p
1
N
c
Using Fatou lemma (20), we obtain:
_
|f f
n
|
p
d liminf
m>n
_
|f
m+1
f
n
|
p
d
Hence, from 4. we see that:
_
|f f
n
|
p
d
_
+

k=n
f
k+1
f
k

p
_
p
and consequently:
f f
n

p

+

k=n
f
k+1
f
k

p
6. Let p [1, +]. whether p = + or p [1, +[, from 3. and 5., for all
n 1, we have f f
n

p


+
k=n
f
k+1
f
k

p
. Since by assumption,
the series

+
k=1
f
k+1
f
k

p
is nite, we conclude that f f
n

p
0,
as n +. It follows that not only f
n
f -a.s., but also f
n
L
p
f.
7. Suppose g L
p
C
(, F, ) is such that f
n
L
p
g. Then f
n
L
p
f together with
f
n
L
p
g. From ex. (8), f = g -a.s. Furthermore, since f
n
f -a.s., we
see that f
n
g -a.s. The purpose of this exercise (and the previous) is
to prove theorem (44).
www.probability.net
Solutions to Exercises 21
Exercise 13
Exercise 14.
1. Since f
n
L
p
f, from exercise (10), (f
n
)
n1
is a Cauchy sequence in L
p
C
(, F, ).
Using exercise (11), there exists a sub-sequence (f
n
k
)
k1
of (f
n
)
n1
, such
that

+
k=1
f
n
k+1
f
n
k

p
< +.
2. Applying theorem (44) to the sequence (f
n
k
)
k1
, there exists g L
p
C
(, F, ),
such that f
n
k
g -a.s.
3. Also from theorem (44), the convergence f
n
k
g -a.s. and f
n
k
L
p
g are
equivalent. Hence, we also have f
n
k
L
p
g. However, since by assumption
f
n
L
p
f, we see that f
n
k
L
p
f, and consequently from exercise (8), f = g
-a.s.
4. From 2., f
n
k
g -a.s., and from 3., f = g -a.s. It follows that f
n
k
f
-a.s. Given a sequence (f
n
)
n1
and f in L
p
C
(, F, ), such that f
n
L
p
f,
we have been able to extract a sub-sequence (f
n
k
)
k1
such that f
n
k
f
-a.s. This proves theorem (45).
Exercise 14
Exercise 15. Suppose (f
n
)
n1
is a sequence in L
p
R
(, F, ), and f L
p
R
(, F, )
such that f
n
L
p
f. Then in particular, all f
n
s and f are elements of L
p
C
(, F, )
with f f
n

p
0 as n +. From theorem (45), we can extract a sub-
sequence (f
n
k
)
k1
of (f
n
)
n1
, such that f
n
k
f -a.s. This proves theo-
rem (45), where L
p
C
(, F, ) is replaced by L
p
R
(, F, ). Anyone who feels
there was very little to prove in this exercise, could make a very good point.
Exercise 15
Exercise 16.
1. Since (f
n
)
n1
is Cauchy in L
p
C
(, F, ), from exercise (11), we can extract
a sub-sequence (f
n
k
)
k1
of (f
n
)
n1
, such that:
+

k=1
f
n
k+1
f
n
k

p
< +
From theorem (44), there exists f L
p
C
(, F, ), such that f
n
k
f
-a.s., as well as f
n
k
L
p
f.
2. Let > 0. (f
n
)
n1
being Cauchy, there exists n
0
1, such that:
n, m n
0
f
m
f
n

p


2
Furthermore, since f
n
k
L
p
f, there exists k
0
1, such that:
k k
0
f f
n
k

p


2
www.probability.net
Solutions to Exercises 22
However, n
k
+ as k +. There exists k

0
1, such that k k

0

n
k
n
0
. Choose an arbitrary k max(k
0
, k

0
). Then f f
n
k

p
/2
together with n
k
n
0
. Hence, for all n n
0
, we have:
f f
n

p
f f
n
k

p
+f
n
k
f
n

p

We have found n
0
1 such that:
n n
0
f f
n

p

This shows that f
n
L
p
f. The purpose of this exercise, is to prove theo-
rem (46). It is customary to say in light of this theorem, that L
p
C
(, F, )
is complete, even though as dened in these tutorials, L
p
C
(, F, ) is not
strictly speaking a metric space.
Exercise 16
Exercise 17. Let (f
n
)
n1
be a Cauchy sequence in L
p
R
(, F, ). Then in
particular, it is a Cauchy sequence in L
p
C
(, F, ). From theorem (46), there
exists f

L
p
C
(, F, ) such that f
n
L
p
f

. Furthermore, from theorem (45),

there exists a sub-sequence (f
n
k
)
k1
of (f
n
)
n1
, such that f
n
k
f

-a.s.
It follows that f

is in fact R-valued -almost surely. There exists N F,

(N) = 0, such that f

() R for all N
c
. Take f = f

1
N
c. Then
f is R-valued, measurable and f
p
= f

p
< +. So f L
p
R
(, F, ).
Furthermore, f f
n

p
= f

f
n

p
0, which shows that f
n
L
p
f. This
proves theorem (46), where L
p
C
(, F, ) is replaced by L
p
R
(, F, ).
Exercise 17
www.probability.net