Analysis II Sheet 1

Baran Karakus

July 3, 2017

1. a) Fix  > 0. Let N1 be such that n N1 |vn v| /2 and let N2 be such
that n N2 |wn w| /2. Then n max(N1 , N2 ) |(vn + wn ) (v + w)|
|vn v| + |vn w|  by the Triangle Inequality. Since  was chosen arbitrarily, we deduce
(vn + wn ) v + w.

b) If = 0 then vn = 0 for all n and the result is trivial. So suppose is not 0. Fix
 > 0. Let N be such that n N |vn v| /||. Then n N |vn v| . Since
 was chosen arbitrarily, we deduce (vn ) v.

c) Fix  > 0. Let N be such that n N |vn v| . Pick M such that

k M nk N . Then k M |vnk v| . So (vnk ) v.

d) Suppose v and w are not equal. Then |v w| is positive. Let  = |v w|/3. Then
there exist N1 and N2 such that n N1 |vn v|  and n N2 |vn w| . But
then, by the Triangle Inequality, for n max(N1 , N2 ) we have |vw| |vn w|+|vn v| =
2 = 2|v w|/3; contradiction. So v = w.

Suppose f, g : V W are continuous, where V and W are normed vector spaces over
R. Take x V , let (xn ) be any sequence in V converging to x. Then f (xn ) f (x) and
similarly g(xn ) g(x). From (b), g(xn ) g(x). From (a), (f + g)(xn ) (f + g)(x).
So f + g is continuous.

2. Since |v w| = |||v w| R, we have ||v||X = || ||v||X R.

We have ||v||X = 0 maxwX |v w| = 0. Suppose v 6= 0. In general, for x, y Rn ,
x y = 0 x or y = 0 or x and y are perpendicular. If X spans Rn , the vectors
in X cannot all be zero or perpendicular to v (otherwise, we have a linear combination
1 x1 + ... + n xn = v, where all the xi are in X, and then, taking the dot product with v
on both sides, the RHS is non-zero and the LHS is zero; contradiction). Thus we have a
vector x in X such that |x v| > 0. So ||v||X 6= 0 if v 6= 0. So ||v||X = 0 v = 0.

1
 Now we prove the Triangle Inequality. Take x, y Rn . We wish to prove ||x||X +||y||X
||x+y||X . We have, by the Triangle Inequality for real numbers, |xw|+|yw| |(x+y)w)|.
It is then obvious that ||x||X + ||y||X = maxwX |x w| + maxwX |y w| maxwX (|x
w| + |y w|) maxwX |(x + y) w| = ||x + y||X .
So, || ||X is indeed a norm on Rn .

3. a) This set is closed. Take arbitrary (a, b) in the complement of this set. If b 6= 0
then let  = |b|. If b = 0, let  = |a| if a < 0, let  = a 1 if a > 1. Then the open ball
B (a, b) is contained in the complement. So the complement is open. So the set is closed.

b) This set is neither open nor closed. Take (1/2, 0) in the set. In any open ball
B (1/2, 0), there exists points (e.g. (1/2, /2)) not in the set. So the set is not open. It is
also not closed because the sequence ((1/n, 0)) has limit (0, 0) not in the set.

c) This set is open. Take arbitrary (a, b) in the set. Let  = |b|. Then every point in
the open ball B (a, b) has non-zero y co-ordinate, so the open ball is contained in the set.
Thus the set is open.

d) This set is neither open nor closed. Take (0, 0) in this set. For any  > 0 there exists
an irrational t with |t| < . Hence there is always a point not in the set, namely (t, t), in
an open ball around (0, 0). Thus the set is not open. Let s be an arbitrary irrational. We
can construct an increasing sequence of rationals with limit s inductively: take a1 to be
any rational less than s, take a2 to be a rational between (s + a1 )/2 and s, take a3 to be
a rational between (s + a2 )/2 and s, and so on. Then ((an , an )) is a sequence in the set
converging to (s, s), which is not in the set. So the set is not closed.

e) This set is closed. Let A = {(x, y) R2 : y 6= nx for any n N} be the complement.

Now let (a, b) A.
1
Consider the curve y = mx. Define D : R R by D(x) = ((ax)2 +(bmx)2 ) 2 . D(x)
is the distance between (a, b) and the point on the curve with x as its x co-ordinate. We
a+mb
compute the derivative of (D(x))2 w.r.t. x and set it to zero, which yields x = = 1+m 2.
|amb|
This is a minimum point of D. We have D() = 1 . This is the shortest distance
(1+m2 ) 2
between the curve and the point (a, b).
|anb|
Now define F : N R by F (n) = 1 . F (n) is the shortest distance between the
(1+n2 ) 2
curve y = nx and the point (a, b). Let n1 be the least natural number such that b > an1 ,
if such a natural number exists. Then b > a(n1 + 1). So
If a < 0 and b > 0, or a > 0 and b < 0, i.e. (a, b) is in the upper left quartile or
the lower right quartile, let  = min(|a|, |b|). Then every point in the open ball B (a, b)
is in the upper left quartile or the lower right quartile. So no such point has y = nx for

2
some n N, since the x and y co-ordinates have opposite sign. Hence the open balls are
contained in A.
Let n be the least natural number such that b > an, supposing it exists. Then a(n+1) >
b.

4. Let A = {f C[0, 1] : f (1/2) = 0}. Then AC = {f C[0, 1] : f (1/2) 6= 0}. Take

f AC . Let  = |f (1/2)|/2. Consider the open ball B (f ) = {g C[0, 1] : ||g f || < }.
For every g in this open ball and every x [0, 1], we have |f (x) g(x)| < . Thus
|g(1/2) f (1/2)| < |f (1/2)|/2 so g(1/2) 6= 0, so g AC . Hence this open ball is contained
in AC . Since f was arbitrary, AC is open. So A is closed.

However, A is not closed under the norm ||||1 . Take f in AC defined by f (x) = 1 for all
x. Take  > 0. Define h C[0, 1] such that on [ 21 4 , 12 ], h is the line joining ( 12 4 , 0) and
( 21 , 1), and on [ 12 , 12 + 4 ], h is the line joining ( 12 , 1) and ( 12 + 4 , 0). Let h be 0 everywhere
else. Then define g by g(x) = h(x) + f (x). We have ||f g||1 = ||h||1 /2, so g B (f ).
But g is not in AC since g(1/2) = 0. Since  was arbitrary, no open ball centred around f
is contained in AC . So AC is not open. So A is not closed.
R1
Now let B = {f C[0, 1] : 0 f (x)dx}. We prove that B C is open under the |||| norm.
R1
Take f B C . Then let  = | 0 f (x)dx| = 6 0. Now, for all g in the open ball B (f ), we have,
R1 R1 R1
for all x, f (x)  < g(x) < f (x) + , so 0 f (x)dx  < 0 g(x)dx < 0 f (x)dx + . Hence
R1 C C
0 g(x)dx 6= 0. So g B . So B (f ) is contained in B . Since f was chosen arbitrarily,
B C is open. So B is closed.

B is also closed under the || ||1 norm. Take a sequence (fn ) in C[0, 1] converging
toR some f C[0, R 1 1] under the || ||1 norm, where each of the fn are in B. We have
1 R1
| 0 f (x)dx| = | 0 (f (x) fn (x))dx| 0 |f (x) fn (x)|dx for all n. Given any  there
R1 R1
exists some N such that n N | 0 f (x)dx| 0 |f (x) fn (x)|dx. Hence any  > 0 is
R1 R1 R1
greater than | 0 f (x)dx|, so 0 f (x)dx = | 0 f (x)dx| = 0. Thus f B. So any sequence
(fn ) in B that is convergent in C[0, 1] has its limit in B. So B is closed.

5 Define the norm || ||1 by ||(xn )||1 =

P
k=1 |xk |. Define the norm || || by ||(xn )|| =
maxkN |xk |. It is very easy to check that these are both indeed norms. Now let b > 0. Let
(xn ) be such that exactly b + 1 elements are 1, the rest are zero. Then ||(xn )||1 = b + 1 >
b||(xn )|| = b. So for all b > 0 we can find a sequence in l0 such that ||(xn )||1 b||(xn )|| .
So these two norms are not Lipshitz equivalent. P
For all nonempty X P(N), define a norm || ||X on l0 by ||(xn )||X = kX |xk |. It
is straightforward to check that these are norms. We will show no two of these norms are

3
Lipshitz equivalent, and thus that we have uncountably many norms, none of which are
Lipshitz equivalent.
Take any non-empty, distinct X, Y that are subsets of N. Now let i be an element in
one of X, Y and not the other, wlog i X. Let b > 0. Define a sequence (xn ) by xi = 1,
xk = 0 for all other k. Then ||(xn )||X = 1 > b||(xn )||Y = 0. So for all b > 0 we can find a
sequence (xn ) such that ||(xn )||X > b||(xn )||Y . So the two norms are not equivalent. Hence
no two of the norms are equivalent.

6. Suppose S(L) is bounded. Let C be an upper bound. Fix  > 0, and take v, w V ,
v 6= w. We have ||A(vw)|| 
||vw|| C. Then ||v w|| < C ||Av Aw|| = ||A(v w)||
C||v w|| < . Hence L is continuous.

Now suppose L is continuous. Then it is continuous at 0. So there exists a > 0 such

that ||v|| < ||Lv|| < 1. Now let < . Then for any v with ||v|| = 1, ||v|| < , so
1
||L(v)|| < 1. Hence ||Lv|| < || for all v with ||v|| = 1. Now, given any w V , let be
||Lw|| ||L(w)|| 1 1
such that ||w|| = 1. Then ||w|| = ||w|| = ||L(w)|| < || . So S(L) is bounded by || .

We prove that supS(L) = sup{||Lx|| : ||x|| = 1}. Let = sup{||Lx|| : ||x|| = 1}.
Given, v, let be such that || ||v|| = 1. Then ||Lv|| ||L(v)||
||v|| = ||v|| = ||L(v)|| . Now
||Lv||
let < . There exists v with ||v|| = 1 such that ||v|| > . Then ||v|| > . Hence
= supS(L). We utilise this result in b) and c).

a) ||L|| = supS(L) = 0 ||Lv|| = 0 (v V /{0} L is the zero map.

||L|| = supS(L) = sup{ ||Lv|| ||Lv||

||v|| : v V /{0}} = ||sup{ ||v|| : v V /{0}} = ||||L||
R.

||L+K|| = sup{ ||(L+K)v||

||v|| : v V /{0}} sup{ ||Lv|| ||Kv|| ||Lv||
||v|| + ||v|| : v V /{0}} sup{ ||v|| :
v V /{0}} + sup{ ||Kv||
||v|| : v V /{0}} = supS(L) + supS(K) = ||L|| + ||K||.

Thus || || is a norm on B(V, W ).

b) Take any unit vector v in V1 such that ||L1 (v)|| = 6 0 (if there is no such unit
vector, then L1 is the zero map and the inequality is trivially true as the LHS becomes
0). Let be such that ||L1 (v)|| = 1. Then ||(L2 L1 )(v)|| = ||L2 (L1 (v))|| = ||
1
||L2 (L1 (v))|| ||L2 ||, so ||(L2 L1 )(v)|| ||L2 || || = ||L2 || ||L1 v|| ||L2 || ||L1 ||.

4
 c) Let A = (A1 , ..., An ) where Ai is the ith column of A. p As proven above, ||L|| =
sup{||Lx|| : |x| = 1}. Let x = (x1 , ..., xn ) where 1 = x21 + ... + x2n , i.e. x is a
unit vector. We have ||Lx|| = ||x1 A1 + ... + xn An || |x1 | ||A1 || + ... + |xn | ||An ||
max(||A1 ||, ..., ||An ||). Let i be such that ||Ai || = max(||A1 ||, ..., ||An ||). Then let xk = 1
for k = i, xk = 0 otherwise. Then ||Ax|| = ||Ai || = max(||A1 ||, ..., ||An ||). So ||L|| =
max(||A1 ||, ..., ||An ||).

7. a) Does not converge uniformly.

The sequence converges pointwise to f defined by f (x) = 0 for all x X. So if
it converged uniformly, it would converge to f . Take N N. Now pick x such that
1
log
N log x2 . Such an x exists since log 1 = 0 and log is continuous and strictly increasing on
(0, 1), so the RHS can be made arbitrarily large by making x small. Then we have xN 21 .
Since N was arbitrary, for every N N we deduce that we can find an x X such that
|fN (x) f (x)| = |fN (x)| = fN (x) 12 . So (fn ) does not converge uniformly on (0, 1).

b) Converges uniformly.
The sequence converges pointwise to the same f as before. Take  > 0. Now let N be
log 
such that N log n N 1 N  x X. Since 
1 . Then n N |fn (x) f (x)| = x x 2
2
was chosen arbitrarily, we deduce that (fn ) converges uniformly on (0, 21 ).

c) Converges uniformly.
We have (fn ) f pointwise where f (x) = 0 x [0, ). Fix  > 0. Let N N be such
that log (1/N ) 1 < log ; such an N exists since log is not bounded below and strictly
increasing on (0, ) and 1/N can be made arbitrarily small. We have, for all n N ,
log (1/n) 1 log (1/N ) 1 < log .
We have, for x (0, ), xenx <  log x nx < log . Consider the function g
on (0, ) defined by g(x) = log x nx. We have g 0 (x) = 1/x n = 0 x = 1/n. We
also have g 00 (1/n) = n2 < 0, so g has maximum value g(1/n) = log (1/n) 1 on (0, ).
Hence, n N log x nx log (1/n) 1 < log  for all x (0, ) 0 xenx <  for
all x [0, ).

d) Converges uniformly.
We again have (fn ) f pointwise where f is zero everywhere. Fix  > 0. We try to
2
find an N such that n N implies that |ex sin( nx )| <  for all real x. We note that the
2
LHS is a product of two things, |ex | and | sin( nx )|, both of which are always less than or
equal to 1.
2 2
Let x0 > 0 be such that ex0 < ; such an x0 exists since ex is strictly decreasing on
2 2
[0, ) with limit 0 as x . For all real x with |x| x0 , ex < , and so |ex sin( nx )| < .
2
Now we try to make N such that |ex sin( nx )| is also less than  on [x0 , x0 ].

5
 x0
Let N be such that N > sin1 ()
. Then for all x [x0 , x0 ], n N x
n [ xN0 , xN0 ]
2
[ sin1 (), sin1 ()], so and hence |ex sin( nx )| < .
x
| sin( n )| < 
Thus we have found an N such that n N |fn (x)| <  for all real x.

p
8. a) We have fn (0) 0. Take x (0, 1]. fn (x) 0 ennq x 0
p log n nq x (by continuity of the log and exp functions).
nq qnq1 qnq
We have, by LHopitals Rule, p log n p 1 p . Thus for every k > 0 there
n
exists N such that n N implies that nq > kp log n, so that nq x p log n > (kx 1)p log n.
Let k = x2 . We know that log n , so p log n , so nq x p log n . Hence
p log n nq x , which implies that fn (x) 0.

Thus (fn ) converges pointwise to f defined on [0, 1] by f (x) = 0 for all x. So if it

converges uniformly, it must do so to f .

b) and c)
q q
We have fn0 (x) = np en x np+q xen x = 0 x = n1q . We have fn00 (x) =
q q
2np+q en x + np+2q xen x , so fn00 ( n1q ) = 2np+q e1 < 0. For large enough n, n1q (0, 1).
p
We have fn (0) = 0, fn (1) = ennq , and fn ( n1q ) = npq e1 . A continuous function on a
closed bounded interval has its maximum value on the interval at stationary points or at
endpoints of the interval. For n large enough that n1q (0, 1), the maximum of fn on [0, 1]
p
is either fn (1) = ennq or fn ( n1q ) = npq e1 . (fn ) converges uniformly to f if and only if
p
both of these tend to 0 as n . We already know that ennq 0, from a). It is obvious
that npq e1 0 p < q.
So we have uniform convergence iff p < q.

nk
9. Claim: (1+) 0 as n for all k N, for all > 0.
n n!
Proof: By the Binomial Theorem, for n 2k, (1 + )n


k+1 = (nk1)!(k+1)! =
n(n1)...(nk) (n )k+1 nk (nk )(2k+1 )(k+1)! 2k+1 (k+1)! nk
(k+1)! 2
(k+1)! , so (1+) nk+1
= n 0, so (1+) 0.

a) For all real . We have n xn (1 x) 0 for x = 1 or 0, for all . So we need

only prove it for x (0, 1). Proving that n xn (1 x) 0 for x (0, 1) suffices, since
(1 x) is simply a constant for fixed x. Let k be a positive integer greater than . We
have 0 n nk for all n. Now let = x1 1 > 0. We have 1+ 1
= x. By the Claim,
nk
n xn nk xn = (1+)n 0. So n xn 0. Thus fn (x) 0 for all real .

b) For < 1. We require that  > 0 N n N fn (x) = n xn (1x) <  x [0, 1].

6
 Continuous functions on closed bounded intervals have their maximum point(s) at
either a stationary point or at endpoints of the interval. We have fn0 (x) = n+1 xn1
(n+1 + n )xn = 0 x = 0 or x = n+1 n
. We have fn ( n+1 n
) = n (1 + n1 )n n+1
1
, and
1 n 1
fn (0) = fn (1) = 0, so the maximum of fn on [0, 1] is n (1 + n ) n+1 = n 1 n
(1 + n ) 1+1 1 .
1
n
If we are to have uniform convergence, for all  > 0 there must exist an N such
that n N n (1 + n1 )n n+1
1
< ; we require n1 (1 + n1 )n 1+1 1 0. We know
n
1 n
that (1 + n) e (2, 3). So there exists N such that n N 2n1 1+1 1
n
1 n 1
n1 (1 + n) 1
1+ n
3n1 1+1 1 . For = 1, the LHS 2, for > 1 the LHS , for
n
< 1 the RHS 0. Hence n1 (1 + n1 )n 1+1 1 0 < 1.
n

R1 R1 R1
c) For < 2. We have ||fn ||1 = 0 n xn (1 x)dx = n ( 0 xn dx 0 xn+1 dx) =
n n2
(n+1)(n+2) = (1+ 1 )(1+ 2 ) 1 if = 2, if > 2, 0 if < 2.
n n

d) For < 0. We have fn0 (x) = n xn1 (n (n + 1)x). We require fn0 (x) 0 for all
x [0, 1].
Continuous functions on closed bounded intervals have their maximum point(s) at either
a stationary point or at endpoints of the interval. The functions fn0 are continuous. We
have fn00 (x) = xn2 (n+2 n+1 ) xn1 (n+2 + n+1 ) = 0 x = 0 or x = n+1
n1
.
0 0 0 n1 n1 n1
We have fn (0) = 0, fn (1) = n , fn ( n+1 ) = n ( n+1 ) < n . The maximum of fn0 on

[0, 1] is thus fn (1). We have fn (1) = n 0 < 0. So we have (fn0 ) 0 pointwise

0 0

< 0.

e) For < 0.
We cannot have uniform convergence to 0 for > 0 as we do not even have pointwise
convergence for > 0, and uniform convergence implies pointwise convergence.
log 
Now take < 0. Take any  > 0. Now let N > e . Then N < . For all n N ,
n N < . Since n is the maximum of fn0 (1) on [0, 1], as proven in (d), n N
0 fn0 (x) <  for all x [0, 1]. So we have uniform convergence.

P
12. Let x (1, 1). Since an converges, the an are bounded, i.e. there exists some
C > 0 such that C |an | for all n 0. Then |an xn | Cxn . Cxn converges because
P
it
an xn
P
is a geometric series with common ratio in (1, 1). So, by the Comparison Test,
converges absolutely, so converges.

Now define f on (1, 1] by f (x) = n

P
Pan x . This is an well-defined function since
n=0
f (x)
P is finiten forPall x. We havePf (x) = (sn sn1 )x (where
n=0P Pwe take s1 = 0)
= n=0 sn x
n=0 s n1 x n =
n=0 s n x nx
n=0 sn x n = (1 x) n
n=0 sn x .

7
 13. Since the function is periodic with period 2, we may as well prove the given
inequality for s, t [1, 3], with at least one of s, t, wlog s, in [1, 1], since both of
them being in [1, 3] is the same as both being in [1, 1]. Suppose s, t [1, 1]. Then
|(s) (t)| = ||s| |t|| |s t| follows from the Triangle Inequality. Now suppose
t [1, 3].
We first prove the inequality for s, t [1, 1]. |(s) (t)| = ||s| |t|| |s t| follows
from the Triangle Inequality. Now suppose t [2n + 1, 2n + 3] for some n N. There
exists some u [1, 1] such that (u) = (t) and u t = 2k for some k Z. We thus
have |(s) (t)| = |(s) (u)| |s u| = |s t 2k| |s t| + |2k| by the Triangle
Inequality, so |(s) (t)| |s t|.

14. a) Suppose there is some  > 0 such that there is no finite collection of open balls
of radius  that cover C. We shall construct recursively a sequence in C with no convergent
subsequence. Take v1 C. Take v2 C \ B (v1 ). Now take v3 C \ (B (v1 ) B (v2 )),
and so on. We are always able to pick the next element of the sequence because at no point
will C \ (B (v1 ) ... B (vk )) be empty. Thus (vn ) is a sequence where every element is
at least a distance  from the other elements. So it has no convergent subsequence. So C
is not sequentially compact; contradiction.
Thus, for every  > 0 there is a finite set of points {v1 , ..., vn } such that {B (v1 ), ..., B (vn )}
cover C.

b)