Zach Conn

Math 425 HW 1
Due 31 August 2011
#1. Assume f
n
(k) 0 for k, n 1. Dene S
n
=

k=1
f
n
(k). Prove that
liminf
n
S
n

k=1
(liminf
n
f
n
(k)).
Solution. {S
n
} is a monotonically increasing sequence. By denition,
liminf
n
f
n
(k) = lim
n
(inf{f
N
(k) : N n}).
Dene g
n
(k) = inf
Nn
f
n
(k) and notice that g
n
(k) f
N
(k) for N n. Now
suppose N < n and m < . Then
S
N
=

k=1
f
N
(k)

k=1
g
N
(k)
m

k=1
g
n
(k).
Therefore
liminf
N
S
N

m

k=1
liminf
N
g
n
(k) =
m

k=1
liminf
n
f
n
(k)
Finally, taking the limit as m in this inequality gives the result.
#2. Assume f
n+1
(k) f
n
(k) 0 for k, n 1. Use Problem 1 to prove
that lim
n
S
n
=

k=1
lim
n
f
n
(k). Prove that if 0 f
n+1
(k) f
n
(k) for
k, n 1 and if S
n
< for some n then this result remains true.
Solution. According to Problem 1,

k=1
f(k) liminf
n

k=1
f
n
(k)
where f = lim
n
f
n
. For each index n we have f
n
f for all k, so

k=1
f
n
(k)

k=1
f(k).
Consequently
limsup
n

k=1
f
n
(k)

k=1
f(k).
Hence

k=1
f(k) = lim
n

k=1
f
n
(k).
In the alternative case that 0 f
n+1
(k) f
n
(k) for k, n 1 and S
q
< ,
dene the functions g
n
by g
n
(k) = f
q
(k) f
n
(k). Now g
n+1
(k) g
n
(k) 0
for suciently large n, so the result just proved implies for g = lim
n
that

k=1
g(k) = lim
n

k=1
g
n
(k).
(Strictly speaking, the entire sequence {g
n
} is not monotonically increasing,
but since a tail end of it is this does not aect the result.) This is the same
as

k=1
(f
q
(k) f(k)) = lim
n

k=1
(f
q
(k) f
n
(k))
or
S
q

k=1
f(k) = S
q
lim
n

k=1
f
n
(k).
The S
q
s cancel, giving the desired result.
#3. Assume g(k) 0 and f
n
(k) R satisfy

k=1
g(k) < ,
|f
n
(k)| g(k), k, n 1,
lim
n
f
n
(k) = f(k) exists, k 1.
Use Problem 1 to prove that
lim
n
S
n
=

k=1
_
lim
n
f
n
(k)
_
.
Show also that
lim
n

k=1
|f
n
(k) f(k)| = 0.
Solution. Observe that |f f
n
| < 2g. Apply Problem 1 to the nonnegative
sequence given by 2g |f f
n
|. This yields
limsup
n

k=1
|f(k) f
n
(k)|

k=1
limsup
n
|f(k) f
n
(k)| = 0,
so
lim
n

k=1
|f
n
(k) f(k)| = 0.
This implies lim
n

k=1
f
n
(k) =

k=1
f(k) (i.e, lim
n
S
n
=

k=1
(lim
n
f
n
(k)))
since

k=1
f(k)

k=1
f
n
(k)

k=1
(f(k) f
n
(k))

k=1
|f(k) f
n
(k)|.
#4. Let {a
k
} and {b
k
} be sequences of real numbers dened for k r.
Dene
A
s
=
_

s
k=r
a
k
, s r
0, s < r
and prove that
q

k=p
a
k
b
k
=
q1

k=p
A
k
(b
k
b
k+1
) + A
q
b
q
A
p1
b
p
, r p < q < .
Solution. Note that a
k
= A
k
A
k1
for k r. Consequently
q

k=p
a
k
b
k
= A
p
b
p
+
q

k=p+1
(A
k
A
k1
)b
k
= A
p
b
p
+
q

k=p+1
A
k
b
k

q1

k=p
A
k
b
k+1
=
q

k=p
A
k
b
k

q

k=p
A
k
b
k+1
+ A
q
b
q+1
= A
q
b
q+1

q

k=p
A
k
(b
k+1
b
k
)
= A
q
b
q+1
A
q
(b
q+1
b
q
) +
q1

k=p
A
k
(b
k
b
k+1
)
= A
q
b
q
+
q1

k=p
A
k
(b
k
b
k+1
)
#5. Given a sequence {b
k
}, k 0, such that b
k
b
k+1
for k 0, set up the
sequence
S
n
=
n

k=0
(1)
k
b
k
, n 1.
Prove that
_
0 S
n
S
m
b
m+1
, 0 m n, m odd
0 S
n
S
m
b
m+1
, 0 m n, m even
.
Conclude that the sequence {S
n
} is Cauchy, hence convergent, if and only if
lim
k
b
k
= 0.
Solution. If m is odd, then
S
n
S
m
=
n

k=m+1
(1)
k
b
k
= b
m+1
b
m+2
+ b
m+3
b
m+4
+ + b
n
= b
m+1
(b
m+2
b
m+3
) (b
m+4
b
m+5
) b
n
b
m+1
On the other hand, its nonnegative because b
m+1
b
m+2
, b
m+3
b
m+4
, and
so on are nonnegative.
If m is even, then
S
n
S
m
=
n

k=m+1
(1)
k
b
k
= b
m+1
+ b
m+2
b
m+3
+ b
m+4
+ + (1)
n
b
n
= b
m+1
(b
m+3
b
m+2
) (b
m+5
b
m+4
) b
n
b
m+1
.
On the other hand, its nonpositive because b
m+2
b
m+1
, b
m+4
b
m+3
, and
so on are nonpositive.
Now we move on to the convergence condition. If lim
k
b
k
= 0, then
the two estimates established above imply that {S
n
} is Cauchy (by directly
going back to the denition of Cauchy sequent) and so convergent (R is
complete). Conversely, it is a well-known fact that if the series is convergent
then lim
k
(1)
k
b
k
= 0. By taking norms this implies lim
k
|b
k
| = 0,
which implies lim
k
b
k
= 0, nishing the proof.
#6. Dene polynomials P
n
(x) by
P
n
(x) =
_
1 +
x
n
_
n
and express P
n
(x) by
P
n
(x) =
n

k=0
_
n
k
_
_
x
n
_
k
=
n

k=0
f
n
(k)x
k
.
Prove that 0 f
n
(k) f
n+1
(k), n 1, k 0. Conclude from Problem 2
that
lim
n
_
1 +
x
n
_
n
=

k=0
x
k
k!
= e
x
, x 0.
Show that
lim
n
P
n
(x)P
n
(x) = 1, x 0.
Solution. The functions f
n
(k) look like
f
n
(k) =
_
n
k
_
1
n
k
=
n!
k!(n k)!n
k
.
The successive ratios are
f
n+1
(k)
f
n
(k)
=
(n + 1)!k!(n k)!n
k
n!k!(n + 1 k)!(n + 1)
k
= (n + 1)(n + 1 k)
_
n
n + 1
_
k
1
for 0 k n. Applying Problem 2 with the sequence g
n
(k) = f
n
(k)x
k
for
a xed nonnegative value of x (so that the ordering g
n+1
g
n
is preserved),
we nd that
lim
n
_
1 +
x
n
_
n
=

k=0
x
k
k!
since lim
n
f
n
(k) = 1/k!. Now observe that
1
x
2
n

_
1
_
x
n
_
2
_
n
1,
so by the squeeze theorem
lim
n
_
1
_
x
n
_
2
_
n
= 1.
But
_
1 +
x
n
_
n
_
1
x
n
_
n
=
__
1 +
x
n
__
1
x
n
__
n
=
_
1
x
2
n
2
_
n
,
so
lim
n
P
n
(x)P
n
(x) = 1.
#7. Let = be a nite set. Let f : (0, ) be a positive real-valued
function on . Dene v
f
on [0, ) by
v
f
() = Cardinality(f
1
((, ])), 0.
Let be real-valued and continuously dierentiable on [0, ). Establish

f
1
([0,R])
(f()) = (0)v
f
(0) (R)v
f
(R) +
_
R
0

()v
f
() d.
Conclude that

(f()) = (0)v
f
(0) +
_

0

()v
f
() d.
Solution. v
f
() is a monotonically decreasing step function. Consequently
there exist a partition 0 = p
0
< p
1
< < p
n
= R of the interval [0, R] and
corresponding constants c
0
, c
1
, . . . , c
n1
{0, 1, 2, . . . } such that
_
R
0

()v
f
() d =
n1

i=0
_
c
i
_
p
i+1
p
i

() d
_
=
n1

i=0
c
i
((p
i+1
) (p
i
)) .
The points {p
i
} of the partition are precisely the points where v
f
jumps in
value and consequently exhaust the set f
1
([0, R]). Moreover, the constants
{c
i
} are determined by c
i
= v
f
(p
i
). So the sum becomes
n1

i=0
c
i
((p
i+1
) (p
i
)) = v
f
(p
0
)((p
1
) (p
0
)) + + v
f
(p
n1
)((p
n
) (p
n1
))
= v
f
(0)(0) + v
f
(p
n1
)(p
n
) +
n1

i=1
((p
i
)(v
f
(p
i1
) v
f
(p
i
)))
But v
f
(p
i1
) v
f
(p
i
) = 1, v
f
(p
n1
) = v
f
(R), and (p
n
) = (R), so we nd
_
R
0

()v
f
() d = v
f
(0)(0) + v
f
(R)(R) +

f
1
([0,R])
(f()),
which was to be proved. To obtain the nal formula, we just pass to the
limit as R . The term v
f
(R)(R) 0 since (R) 0 (because is a
nite set), and we are left with
_
R
0

()v
f
() d = v
f
(0)(0) +

(f()).
#8. Let {a
n
}
n1
be a sequence of real numbers that is subadditive, i.e.,
assume a
n+m
a
n
+ a
m
for m, n 1. Prove that for any xed n 1
limsup
N
a
N
N

a
n
n
and use this to conclude that
limsup
N
a
N
N
inf
n1
a
n
n
liminf
n
a
n
n
so that
lim
N
a
N
N
= inf
n1
a
n
n
.
Since this is true for all n,
limsup
N
a
N
N
inf
n1
a
n
n
.
But inf
n1
an
n
liminf
n
an
n
, so we can conclude
limsup
N
a
N
N
inf
n1
a
n
n
liminf
n
a
n
n
.
Solution. For all i, j, k 1 we have a
jk+i
ja
k
+a
i
by subadditivity. Hence,
xing i, k 1 we nd
limsup
j
a
jk+i
jk + i
limsup
j
ja
k
+ a
i
jk + i
= limsup
j
_
a
k
j

jk
jk + i
+
a
i
jk + i
_
=
a
k
k
.
This holds for each i 1, so for each k 1
limsup
n
a
n
n
= inf
i=1,...,k
_
liminf
j
a
jk+i
jk + i
_

a
k
k
.
#9. Prove that if a -algebra of subsets of R contains intervals of the form
(a, ), then it contains all intervals.
Solution. The algebra also contains the intervals (, a] since it is closed
under complementation. Since it is closed under intersections and unions as
well, it contains (a, b) since (a, b) = (a, ) (b, ); it contains [a, b] since
[a, b] = (, a] (, b]; it contains [a, b) since [a, b) = (a, b) [a, a]; and it
contains (a, b] since (a, b] = (a, b) [b, b].
#10. Let E be a nonempty set of real numbers such that EE = E. Prove
that if there exists > 0 such that
E (, ) = {0}
then there exists u 0 with
E = Zu.
Prove that if otherwise then E is dense in R.
Solution. If E = {0}, then it is of the form Z0, so suppose from now on that
E contains an element distinct from zero. Then E must contain a positive
element, since if it contains a negative element x then 0 x E is positive.
If such an exists, then E contains a smallest positive element u. For
any other nonzero element x E write x = uv + r where v Z and r R
satises 0 < r u. Then r = x uv E, forcing |r| = u since u is the
minimal positive element. Consequently x Zu. Since x was arbitrary, we
conclude that E = Zu.
Otherwise E contains elements arbitrarily close to zero. More precisely,
for any there exists x E with |x| < . Since E is additive it contains Zx.
Since is arbitrary, E must be dense.
Zach Conn
Math 425 HW 2
Due 9 September 2011
1. Prove that U E is relatively open if and only if C = E U is relatively
closed. (E U = E

U, where

U denotes the complement of U in R.)
Solution. Suppose U is relatively open. This means there exists an
open set V R such that U = E V . We want to nd a closed set
D R such that E U = E D. Consider D =

V , the complement
of V in R. This is closed in R since V is open in R.
If x E U, then x E but x / U. Since U = E V , this means
x / E V as well. The only way to have x E and x / E V is
to have x E V . This means x E and x / V . In particular,
x E

V . So we get the inclusion E U E D.
Now suppose x E D. This means x E but x / V . Since
U = E V V , we nd x / U. Therefore x E U (since x is in
E). So we get the reverse inclusion E D E U. This proves that
if U is relatively open then E U is relatively closed.
Now suppose E U is relatively closed. This means there exists a
closed set D R such that E U = E D. We want to nd an open
set V R such that U = E V . Consider V =

D, the complement of
D in R. This is open in R since D is closed in R.
If x U, then x E (since U E) but x / E U. Therefore
x / E D (since E U = E D). If x E but x / E D, then
x / D, i.e., x

D = V . Hence x E V . So we get the inclusion
U E V .
Now suppose x EV . Then x E and x / D. This means x / ED
since E D D. Therefore x / E U since E D = E U. But
x E, so we nd x U. So we get the reverse inclusion E V U.
This proves that if E U is relatively closed then U is relatively open.
2. Prove and E are clopen in E.
Solution. = E. Since is open in R, it follows that is relatively
open in E by the denition of relatively open. Similarly, since is
closed in R, it follows that is relatively closed in E by the denition
of relatively closed.
E = ER. Since R is open in R, it follows that E is relatively open in
E by the denition of relatively open. Similarly, since R is closed in R,
it follows that E is relatively closed in E by the denition of relatively
closed.
3. Prove for any E and x that if x / E, then E (x, ) and E (, x)
are clopen in E.
Solution. They are relatively open in E directly by the denition of rela-
tively open since the intervals (x, ) and (, x) are open in R
To see that E (x, ) is relatively closed in E, we need to nd a closed
set D R such that E (x, ) = E D. The set E (x, ) is bounded
below by x but does not contain x. Therefore, since E does not contain x,
we do not add any points to E (x, ) by expanding it to E [x, ).
But [x, ) is closed in R since its complement is the open interval (, x).
This implies that E (x, ) is relatively closed in E.
Similarly, to see that E(, x) is relatively closed in E, we can consider
instead E (, x], which does not dier from E (, x) since x / E.
(, x] is closed in R since its complement is the open interval (x, ). This
implies that E (, x) is relatively closed in E.
Altogether we conclude that E (x, ) and E (, x) are each both
relatively open and relatively closed in E, i.e., clopen in E.
4. Prove that if E R is connected, then either E = or else E =
(inf E, sup E) H, where H is a subset of the (nite!) set {inf E, sup E}R.
Solution. Suppose E is nonempty. I will show that E must then have the
desired form.
By problem 3, the sets U(x) = E (x, ) and L(x) = E (, x)
are clopen in E. Since E is connected, its only clopen subsets are E and .
Therefore U(x), L(x) {E, } for all x R E.
Now we show that E contains (inf E, sup E) via an argument by contra-
diction. Suppose (inf E, sup E) is not contained in E. Then there is at least
one point p (inf E, sup E) with p / E. Because p > inf E, there exists
x E satisfying inf E x < p. Because p < sup E, there exists y E
satisfying p < y sup E.
Because p R E and E contains an element less than p, namely x,
L(p) is nonempty, so L(p) = E. Similarly, because p R E and E
contains an element greater than p, namely y, U(p) is nonempty, so U(p) = E.
Therefore L(p) = U(p), i.e., E (, p) = E (p, ). But then E
(, p) (p, ) = , a contradiction since E is nonempty by supposition.
Therefore we can conclude that (inf E, sup E) E.
Whether E contains its inmum and supremum cannot be decided in
the general case, but its certainly possible. However, by denition it cant
contain anything smaller than inf E nor can it contain anything larger than
sup E. Therefore, in the general case E is just the interval (inf E, sup E)
with neither, one, or both of the numbers inf E, sup E. Since certainly these
numbers are not in E if they are innite, we can conclude with the nal
result that
E = (inf E, sup E) H
where H is some subset of {inf E, sup E} R.
5. Let f be a real-valued continuous function on a connected set E. Prove
that the image of E under f, i.e. the set f(E) = {f(x) : x E}, is also
connected. Why does this imply Proposition 19, Pg. 48?
Solution. By Problem 41, Pg. 49, the preimage of an open set is relatively
open in E. Similarly, the preimage of a closed set is relatively closed in E.
To deduce this from Problem 41, suppose D R is closed and consider its
complement V =

D. V is open since D is closed. Therefore the preimage
f
1
(V ) of V is relatively open in E, so there exists an open set O R such
that f
1
(V ) = E O. The complement C =

O is a closed set in R. Now
f
1
(D) = f
1
(

V ) = {x E : f(x) / V } = E f
1
(V ) = E (E O) =
E O = E

O = E C, meaning the preimage f
1
(D) is relatively closed
in E, as desired.
These results are for preimages of sets which are closed or open in R, but
the analogues hold for sets which are relatively closed or relatively open in
f(E). If H f(E) is relatively open, then H = f(E) V for some open set
V in R. Now f
1
(H) = f
1
(f(E) V ) = {x E : f(x) E and f(x)
V } = f
1
(f(E)) f
1
(V ) = E f
1
(V ). Now E is relatively open in E
(proved in Problem 2) and f
1
(V ) is relatively open in V , as proved in the
previous paragraph. We now appeal to the fact that the relatively open sets
in E assemble to form their own topology on the set E. Therefore, as in any
topology, the union of two open sets is open, so Ef
1
(V ) must be relatively
open in E.
1
A totally analogous argument shows that the preimage of a set
relatively closed in f(E) is also relatively closed in E.
Therefore the preimage of a clopen set in f(E) is relatively clopen in E.
But since E is connected the only relatively clopen sets it contains are and
E. So suppose H is a relatively clopen set strictly contained in f(E). Then
the preimage f
1
(H) is relatively clopen in E and hence equal to either or
E. However, if it were equal to E, then H would have to be equal to f(E)
by denition of f(E), so the only possibility is that f
1
(H) = . Therefore
H = . We conclude that f(E) can contain no relatively clopen subsets
besides and f(E), so f(E) must be connected.
In conjunction with Problem 4 this implies Proposition 19, Pg. 48 (the
intermediate value theorem). The statement of that proposition is as follows:
Let f be a continuous real-valued function on [a, b] and suppose f(a)
f(b) [or f(b) f(a)]; then there is a point c [a, b] such that
f(c) = . To prove this, rst notice that [a, b] is connected.
2
Therefore
f([a, b]) is connected by the foregoing. Assume without loss of generality that
f(a) f(b) (otherwise just swap a, b in what follows). f([a, b]) contains f(a)
and f(b), so it must contain [f(a), f(b)] by Problem 4 (since that problem
1
Direct proof: if A, B are relatively open in E, then A = E C and B = E D for
open sets C, D in R. Then AB = (EC) (ED) = E(C D). We take for granted
that the union of two open sets in R is again open in R, so C D is an open set in R.
This means that A B is relatively open in E.
2
Proof: The function f : R (0, 1) dened by f(x) = 1/((arctanx + /2)) is
surjective and continuous, so by Problem 5 the open interval (0, 1) is connected (since
R is connected by Remark 1 in the problem set). Now the closure [0, 1] of (0, 1) is also
connected, and then any closed interval [a, b] is a continuous image of [0, 1] via the map
t [0, 1] (1 t)a +tb.
implies that f([a, b]) is an interval, though its left endpoint may be less than
f(a) and its right endpoint may be greater than f(b)). Therefore, since
[f(a), f(b)] f([a, b]), for any [f(a), f(b)] there must be a preimage
c [a, b] with f(c) = .
6. Let E R be any compact set, i.e., a set which enjoys the Heine-Borel
property that every open cover has a nite subcover. Prove that if f is a
continuous real-valued function on E, then f(E) is also compact.
Solution. Suppose that V is an open cover of f(E). Dene U = {f
1
(V ) :
V V}.
First, U must be an open cover of E. Its elements are open sets by
Problem 41, Pg. 48. To see that its a cover of E, take any x E. Its image
f(x) must be in some set V V since f(x) f(E) and V is an open cover
of f(E). Then f
1
(V ) U contains x. So every point of E is contained in
an open set in U.
Second, since E is compact, U must have a nite subcover, say {U
1
, . . . , U
N
}.
Thus, for any point x E there is an index i(x) such that x U
i(x)
. There-
fore
N
_
i=1
f(U
i
)
contains f(E) since f(x) f(U
i(x)
). But f(U
i
) V by the denition of U.
This implies {f(U
1
), . . . , f(U
N
)} is a nite subcover of V.
7. Use Problem 6 to prove that a continuous real-valued function on a
nonempty compact set E attains its maximum and its minimum value. Use
this and Problem 4 to prove that if f is a continuous function on the closed
interval [a, b], < a < b < , then f([a, b]) = [c, d] for some d = max f
and c = min f. Think about this fact in connection with Rolles theorem,
but do not supply a proof of Rolles theorem.
Solution. Suppose E is a nonempty compact set, and suppose f : E R
is a continuous function. By Problem 6 the image f(E) is compact. The
lemma below shows that f(E) is closed and bounded. Since its bounded,
the construction of R guarantees that it has a least upper bound u and a
greatest lower bound l. By the denition of supremum and inmum, u and
l are limit points of f(E). By the denition of closed, f(E) must in fact
contain all its limit points, so it must contain u and l. u is the maximum
value of f, l the minimum value.
Now Problem 4 implies that if E = [a, b], then f(E) is an interval, but
the question of whether it contains either of its endpoints is open. In fact,
the endpoints are precisely the minimum and maximum values, so it does
contain them in this case. Therefore f([a, b]) = [c, d] where c = min f and
d = max f.
In particular, if c < 0 and d > 0, then f([a, b]) contains 0, meaning there
is some point x E where f(x) = 0. If f is the derivative of another function
on the interval [a, b], this means the original function has a horizontal tangent
somewhere in [a, b]. This is roughly Rolles theorem.
Now for the lemma:
Lemma If E R is compact, then it is closed and bounded.
Proof. First, I will show that its closed. Suppose E is not closed.
Then there is a limit point p of E that is not in E. Consider the
collection C = {N(x) : x E} where N(x) is an open neighborhood
of x that does not intersect some neighborhood V (x) of p. C is an
open cover of E. However, it cannot contain a nite subcover. If
it did, then let W denote the intersection of all sets V (x). This
is a neighborhood of p since it is the intersection of nitely many
open sets, and since p is a limit point W must contain a point q
of E. q is not covered by any set, however, a contradiction. This
in turn means that E is not compact, i.e., we have shown that if
E is not closed, then it is not compact. Therefore, by taking the
contrapositive, compactness implies closedness.
Pick any point p R and consider the family C of all open intervals
centered at p. This is an open cover of any subset of R; in par-
ticular, its an open cover of E. If E is compact, then it must be
covered by some nite subcover of C. This subcover contains some
largest interval since it contains only nitely many intervals, and
this furnishes both upper and lower bounds on E (since E must be
contained entirely within this largest interval).
8. Let f be a continuous real-valued function on an interval [a, b], < a <
b < . Dene a set U (a, b) by U = {x : a < x < b and there exists y, x <
y < b, such that f(y) > f(x)}. Prove that U is an open set.
Solution. Suppose x U. It suces to show that there exists some > 0
so that (x , x + ) U.
Suppose there is no such . Then any interval (x , x + ) contains a
point x

/ U. This means there does not exist y

satisfying x

< y

< b and
f(y

) > f(x

). In other words, f(y

) f(x

) for all y

satisfying x

< y

< b.
But as decreases, x

must get closer to x. By the continuity of f, it

follows that f(x

) f(x) as x

x. This implies that sending in the limit

f(y

) f(x) for all y

satisfying x < y

< b. This is a contradiction since

x U, meaning there must be a y with x < y < b and f(y) > f(x).
9(i) Let [a, b], f and U be as in Problem 8. By Proposition 8, Pg. 42, U
can be represented as a countable union of pairwise disjoint open intervals:
U =

n
(a
n
, b
n
). For each n either a
n
= a or a
n
/ U, for the reason that any
open interval I (such as one of the intervals (a
k
, b
k
)) that contains a
n
must
satisfy I (a
n
, b
n
) = . Analogously, for each n either b
n
= b or b
n
/ U, for
the reason that any open interval I (such as one of the interval (a
k
, b
k
)) that
contains b
n
must satisfy I (a
n
, b
n
) = .
9(ii-iv) (ii) Use Part (i) and the denitions to prove for each n such that
a
n
= a it is true that f(a
n
) f(b
n
).
(iii) Use Part (i) and the continuity of f to prove that for any n and x that
if a
n
< x < b
n
, then f(x) f(b
n
).
(iv) Apply (ii), (iii) and the continuity of f to conclude that for all n,
f(a
n
) f(b
n
), and if a
n
= a, then f(a
n
) = f(b
n
).
Solution. (ii) If a
n
= b
n
, then this is trivially true, so suppose b
n
> a
n
. By
Part (i), if a
n
= a, then a
n
/ U. By the denition of U, this means
that f(y) f(a
n
) for all y satisfying a
n
< y < b. In particular, we can
consider the choice y = b
n
, since b
n
> a
n
. This yields f(a
n
) f(b
n
) in
this case as well, as desired.
(iii) Suppose by way of contradiction that there was a
n
< x < b
n
with
f(b
n
) < f(x). Let A = {s [x, b] : f(s) f(x)}. A is a closed set
3
that contains x but not b. Set x

= sup A, so x

[x, b) [a
n
, b
n
] U
(x

= b because, being the supremum, it is a limit point of A and A,

being closed, must contain all its limit points, but A does not contain
b). Hence there exists y such that x

< y b and f(y) > f(x

). Since
f(x

) f(x) > f(b

n
) and f(b
n
) f(z) for all z satisfying b
n
z b,
we see that y cannot exceed b
n
and hence lies in A. But this contradicts
the fact that x

is the supremum of A.
(iv) Since f is continuous, sending x a
n
(under the conditions of Part
(iii)) must yield f(a
n
) f(b
n
). This is for all n since Part (iii) is true
for all n. However, if n is such that a
n
= a, then Part (ii) gives the
reverse inequality f(a
n
) f(b
n
), which forces f(a
n
) = f(b
n
).
Zach Conn
Math 425 HW 1
Due 19 September 2011
1. Let
1
,
2
be partitions of [a, b]. Prove there exists

such that (a)

j

,
j = 1, 2, and (b) if
j

, j = 1, 2, then

also.
Solution. (a) Take

=
1

2
. By denition of the join of two partitions,
each point in
j
is in
1

2
for j = 1, 2.
(b) With

=
1

2
, if
j

, then
1

2

, so

.
2. Prove that L

(f) = U

(f). Conclude that either inequality in (3)

implies the other.
Solution. This follows from the relations sup
x
i1
xx
i
(f) = inf
x
i1
xx
i
(f).
Given the inequality L

(f) L

(f), we obtain U

(f) U

(f)
or U

(f) U

(f). This in turn implies U

(f) U

(f). These steps are

reversible, so the reverse implication follows as well.
3
Proof: By Problem 7, f achieves a maximum on [x, b], say M, so A = [x, b]
f
1
([f(x), M]). Now [x, b] is closed since its a closed interval and the preimage is closed
as proved during the course of Problem 5. Since the intersection of two closed sets is
closed, A must be closed.
3. Assume f R[a, b]. Given > 0, prove that there exists a partition
= {x
0
, . . . , x
n
} with the following property: If

is the totality of intervals

[x
i1
, x
i
] having M
i
m , then

has total length at most . That is,

the sum of those dierences x
i
x
i1
with M
i
m
i
is at most .
Solution. By Proposition 3 there exists a partition such that
U

(f) L

(f) <
2
.
Hence

2
>

(M
i
m
i
)(x
i
x
i1
).
Now we eliminate those terms for which M
i
m
i
< (which we can do
because M
i
m
i
0 for all i), and we nd

2
>

(x
i
x
i1
)
or
>

(x
i
x
i1
),
as desired.
4. Given f, g R[a, b] and > 0, prove there exists with U

(f)L

(f) < ,
U

(g) L

(f) < . (That is, one working for both.)

Solution. By Riemann integrability there exist partitions
f
,
g
so that
U

f
(f) L

f
(f) < and U
g
(g) L
g
(g) < .
Consider the join =
f

g
, which is subordinate to both
f
and
g
.
By applying Proposition 1, we nd that the dierences U

(f) L

(f) and
U

(g) L

(g) are at most the dierences above, hence at most .

5. Evaluate the limits
(i)
lim
n

k=1
n k
n
2
cos 2
k
n
,
(ii)
lim
n

k=1
k
2
(k n)
n
4
.
Solution. (i) Each term splits into
1
n
cos 2
k
n

k
n
2
cos 2
k
n
.
This is a Riemann sum for cos(2x)xcos(2x) on [0, 1]. The integral,
hence the limit, is 0.
(ii) This is a Riemann sum for x
3
x
2
on [0, 1]. The integral, hence the
limit, is 1/12.
6. Assume f, g have continuous derivatives. Prove that
I
b
a
(fg

) = (f(b)g(b) f(a)g(a)) I
b
a
(f

g).
Solution. By Proposition 5,
I
b
a
(fg

) + I
b
a
(f

g) = I
b
a
(fg

+ f

g) = I
b
a
((fg)

).
Now fg is certainly an antiderivative of (fg)

. By the existence theorem,

then, we can calculate I
b
a
((fg)

) = (fg)(b) (fg)(a) = f(b)g(b) f(a)g(a).

This gives the desired equality.
7. Let f, g be continuous with g continuously dierentiable. Assume g is
monotone (i.e., g

0 or g

0 always). Prove there exists , a b,

such that
I
b
a
(fg

) = f()(g(b) g(a)).
Solution. Assume g

0. Otherwise replace f, g by f, g. Let m =

inf f, M = sup f. Since g

0, we nd
m(g(b) g(a)) I
b
a
(fg

) M(g(b) g(a)).
Since f(g(b) g(a)) is continuous, there exists so that f()(g(b) g(a)) =
I
b
a
(fg

). This concludes the proof.

8. Let f and g be continuously dierentiable with f monotone. Prove that
there exists , a < < b, such that
I
b
a
(fg

) = f(a)(g() g(a)) + f(b)(g(b) g()).

Solution. By problem 6,
I
b
a
(fg

) = (f(b)g(b) f(a)g(a)) I
b
a
(f

g).
Now apply problem 7 to the integral I
b
a
(gf

) to nd so that
I
b
a
(f

g) = g()(f(b) f(a)).
This yields
I
b
a
(fg

) = (f(b)g(b)f(a)g(a))g()(f(b)f(a)) = f(a)(g()g(a))+f(b)(g(b)g()),
as desired.
9. Let f be twice continuously dierentiable, and assume f

is monotone.
Assume in addition there exists > 0 such that either f

or f

everywhere. Prove
|I
b
a
(cos(f(x)))|
2

and |I
b
a
(sin(f(x)))|
2

.
Solution. Assume without loss of generality that f everywhere. We
make the change of variables t = f(x), and the integral becomes
J(a, b) :=
_
f(b)
f(a)
cos(t)
f

(f
1
(t))
dt.
The integrand is positive on (n 1/2) < t < (n + 1/2) if n is even and
negative if n is odd. Let
J
n
=
_
(n+1/2)
(n1/2)
cos(t)
f

(f
1
(t))
dt.
So J
n
alternates in sign and decreases in absolute value. Therefore
|J(a, b)| |J
n
|
1

_
(n+1/2)
(n1/2)
cos(t) dt =
2

,
where f(a) > (n 1/2).
10. Prove
(a) |I
e
20
e
10
(cos(xlog x))| 2/11;
(b) lim
T
I
T
1
(cos(xlog x)) exists;
(c) set sin 0/0 = 1 and prove lim
T
I
T
0
sin x/x exists. The limit is known
to be /2.
Solution. (a) Since log x + 1 is monotonically increasing in x, by problem
9 its enough to show that (xlog x)

= log x+1 11 on [e
10
, e
20
]. Since
log e
10
+ 1 = 11 and log x + 1 is increasing in x, this is true.
(b) We partition [1, ) into the intervals [1, e
10
] and [e
10
k
, e
10
k+1
] for k 1.
On [e
10
k
, e
10
k+1
] the integral is bounded above by 2/(log e
10
k
+ 1) =
2/(10
k
+ 1). By the ratio test for series, the series, hence the limit of
the integral, converges to a nite number.
(c) Dene
a
k
=
_
(k+1)
k
sin x
x
dx,
First observe that
|a
k
| =
_
(k+1)
k
| sin x|
x
dx
since sin does not change sign on [k, (k+1)]. Now | sin((k+1)+t)| =
| sin(k + t)| for t (0, ), so
| sin((k + 1) + t)|
(k + 1) + t
<
| sin(k + t)|
k + t
.
Hence
|a
k+1
| =
_
(k+2)
(k+1)
| sin x|
x
dx =
_

0
| sin((k + 1) + t)|
(k + 1) + t
dt
<
_

0
| sin(k + t)|
k + t
dt =
_
(k+1)
k
| sinx|
x
dx = |a
k
|.
Thus |a
k+1
| < |a
k
|.
But the a
k
alternate in sign (positive for even values of k, negative
for odd values). It follows by the alternating series test that

k=1
a
k
converges, so lim
T
I
T
0
sin x/x exists and is nite.
11. Dene f
t
(x) by f
t
(x) = t sin(2x) +2x, f
t
(x +1) = f
t
(x) +2, so that
cos(f
t
(x)) has period one. Observe that
f

t
(x) = 2t cos(2x) + 2 2(1 t) , 0 < t < 1/2, x R
f
t+s
(x) = f
t
(x) + s sin(2x).
Use the mean value theorem to prove that

cos(f
t+s
(x)) cos(f
t
(x))
s
+ sin(2x) sin(f
t
(x))

s, s > 0, t 0, x R
and
| sin(2x) sin(f
t
(x)) sin
2
(2x)| t, t 0, x R.
With these notations, dene F(t) = I
1
0
(cos(f
t
(x))) and prove F

(t) = I
1
0
((sin(2x)) sin(f
t
(x))),
F

() is continuous, and F

(0) = 1/2.
Conclude that there exists > 0 such that F(t) < 0, 0 < t < .
Finally, prove that if [T], T > 0, is the greatest integer T, then
|I
T
0
(cos(f
t
(x))) [T]F(t)| 1, T > 0
and
lim
T
I
T
0
(cos(f
t
(x))) = , 0 < t < .
Contrast this with Problems 9 and 10 on the Riemann integral notes. What
is missing?
Solution. By calculating the derivative we nd f

t
(x) = 2t cos(2x)+
2 = 2(1 + t cos(2x)). But cos(2x) 1, so f

t
(x) 2(1 t). If
0 < t < 1/2, then in turn we nd 2(1 t) since 2(1 1/2) =
2(1/2) = .
Now f
t+s
(x) = (t +s) sin(2x) +2x = t sin(2x) +2x+s sin(2x) =
f
t
(x) + s sin(2x).
Consider the function u cos(f
t+u
(x)). The mean value theorem
yields
cos(f
t+s
(x)) cos(f
t
(x)) = sin(f
t+c
(x)) sin(2x)s
for some 0 < c s. Then

cos(f
t+s
(x)) cos(f
t
(x))
s
+ sin(2x) sin(f
t
(x))

= | sin(f
t+c
(x)) sin(f
t
(x))|
= | cos(f
t+z
(x)) sin(2x)c| c s
by another application of the mean value theorem, where z c. This
establishes the rst inequality.
Consider the function g(u) = sin(2x) sin(f
u
(x)). The mean value
theorem furnishes 0 c t so that
|g(t) g(0)| = |g

(c)t| = | sin
2
(2x) cos(f
c
(x))t| t.
But g(t) g(0) = sin(2x) sin(f
t
(x)) sin
2
(2x). This establishes the
second inequality.
If we know that we can dierentiate under the integral sign, then, since
the derivative of cos(f
t
(x)) with respect to t is sin(2x) sin(f
t
(x)),
we nd
F

(t) = I
1
0
(sin(2x) sin(f
t
(x))).
The following lemma guarantees that this interchange of operations is
permissible.
Lemma Suppose f(t, x) is dierentiable in t with continuous partial
derivatives. Then

t
I
1
0
(f(t, x)) = I
1
0
_

t
f(t, x)
_
.
Proof. Dene F(t) := I
1
0
(f(t, x)), where the integral is with respect
to x. We rst note that
F(t + t) F(t) = I
1
0
(f(t + t, x) f(t, x)),
where we have used the additivity of the integral, proved in the notes.
Since f(t, x) is continuous on a compact set, it is also uniformly
continuous, so for any > 0 there exists such that
|f(t + t, x) f(t, x)| <
for all values of x [a, b] and t < .
Combining these facts, we nd that for any > 0 there exists such
that
|F(t + t) F(t)|
for t < . In particular, F(t) is continuous.
Similarly, since /tf(t, x) exists and is continuous, for any > 0
there exists such that if t < then

f(t + t, x) f(t, x)
t

f
t

<
for x [0, 1]. Consequently
F(t + t, x) F(t, x)
t
= I
1
0
_
f(t + t, x) f(t, x)
t
_
= I
1
0
_
f(t, x)
t
_
+ R
where |R| < I
1
0
() = .
As t 0, so does . So we obtain
lim
t0
F(t + t, x) F(t, x)
t
= I
1
0
_

t
f(t, x)
_
,
which is what was to be proved since the left side is F/t.
During the proof of the previous lemma it was shown that I
1
0
(f(x, t))
is continuous. Apply that proof with f(t, x) = sin(2x) sin(f
t
(x)).
F

(0) = I
1
0
(sin(2x) sin(f
0
(x))) = I
1
0
(sin
2
(2x)). To calculate this,
we use the identity sin
2
(2x) = (1 cos(2x))/2. The result must be
1/2 since cos(2x) is symmetric about the x-axis on [0, 1].
For u > 0 we can write F(u) = F(0) +I
u
0
F

(t). Now I
u
0
F

(t) is a dier-
entiable function of u (Proposition 6 in the notes), so it is in particular
continuous. Moreover, since F

(0) < 0 and F

(t) is continuous, it fol-

lows that there exists so that F

(t) < 0 for 0 < t < . Since I

u
0
F

(t)
is a limit of Riemann sums, for 0 < u < we must have I
u
0
F

(t) < 0.
Therefore F(u) < 0 for 0 < u < .
Because f
t
(x +1) = f
t
(x) +2, the function cos(f
t
(x)) has period one.
Therefore, if T is a positive integer, then the additivity of the Riemann
integral yields I
T
0
(cos(f
t
(x))) = T I
1
0
(cos(f
t
(x))).
Similarly, if T = n + for n a positive integer and 0 < < 1, then
I
T
0
(cos(f
t
(x))) = n I
1
0
(cos(f
t
(x))) + I

0
(cos(f
t
(x))).
Now |I
1
0
(cos(f
t
(x)))| 1 since | cos(f
t
(x))| 1. Therefore
|I
T
0
(cos(f
t
(x))) n I
1
0
(cos(f
t
(x)))| 1.
But n I
1
0
(cos(f
t
(x))) = [T]F(t). This establishes the rst of the nal
inequality.
If 0 < t < , then we have shown that F(t) < 0, i.e., I
1
0
(cos(f
t
(x))) < 0.
Now by the previous inequality I
T
0
(cos(f
t
(x))) is always within one unit
of [T]F(t), which is negative. Since F(t) is a constant with respect
to the limit sending T , we conclude that [T]F(t) and hence
I
T
0
(cos(f
t
(x))) diverge to as T . This establishes the nal
result.
The dierence is that f

t
(x) is not monotone. It includes the factor
cos(2x), so there can be no > 0 as in Problems 9 and 10.