Practice final exam solutions

1
We write d = dSN CF for the French Railroad metric on R2 . In this problem
we will often use the easily checked fact that if (xn ) is a Cauchy sequence in
any metric space, then xn x if and only if xnk x for some subsequence
(xnk ).
So let (xn ) R2 be a Cauchy sequence with respect to d. We will show
that (xn ) is convergent, and hence that d is a complete metric. If xn 0
then obviously we are done, so assume (xn ) does not converge to 0. Note
that since 0 lies on every line through the origin by denition, we have that
d(xn , 0) = |xn |.
I claim that there exists E > 0 such that |xn | > E for all n N. Indeed, if
not then there exists a subsequence (xnk ) with |xnk | 0 as k , which
means that the subsequence (xnk ) converges to 0 with respect to d. Then
since (xn ) is Cauchy, xn 0, contradiction.
Now let N N be suciently large that d(xn , xm ) < E for n, m > N .
Then xn and xm must lie on the same line. If they didnt, then
E > d(xn , xm ) = |xn | + |xm | > E + E
Contradiction.
In other words, there exists a line l R2 with xn l for n > N . For
any n, m > N , we then have d(xn , xm ) = |xn xm |. Thus, (xn ) is a Cauchy
sequence with respect to the standard Euclidean metric on R2 , since d will
agree that metric for suciently large n. Since | | is complete, there exists
1

x R2 with limn |xxn | = 0. Lines are closed subsets of R2 with respect

to | |, so x l. Then d(xn , x) = |xn x|. Thus xn x with respect to d,
and so d is complete.

2
Let E = Q [0, 1]. E is an innite subset of a countably innite set,
hence is countably innite. In other words, there exists a bijective function
f : N E. Dene the sequence (xn ) via xn = f (n). Note that E = [0, 1]
Let F be the set of all subsequential limits of E. I claim that F = [0, 1].
Suppose that x F . Take a subsequence (xnk ) converging to x. Then every
neighbourhood of x contains all but nitely many (xnk ), and in particular
intersects E. So x E = [0, 1].
Conversely, suppose x [0, 1]. We will construct a subsequence xnk x
inductively. Let n1 = 1. Suppose we have dented n1 , n2 , . . . nk . Consider
the subset Ak+1 N, dened by
Ak+1 = {n N||x f (n)| <

1
}
k

Recall that xn = f (n). Since x is a limit point of E, B1/k+1 (x) contains

innitely many points of E; since f is surjective, this implies that Ak+1 is
innite. Thus we can pick a nk+1 Ak+1 with nk+1 > nk .
Thus we have constructed a subsequence (xnk ) with |x xnk | < 1/k, which
means that xnk x, so x F . Thus F = [0, 1] and we are done.

3
We have a continuous function f : [0, 1] [0, 1] R. For a xed x [0, 1],
consider the function hx : [0, 1] R dened by hx (y) = f (x, y). Then hx
is continuous; indeed, for any y [0, 1] and E > 0, take a > 0 that works
for f and E at (x, y). Since hx is a continuous function on a compact set,
it attains a nite maximum; in other words, for some y0 [0, 1], we have
f (x, y0 ) = hx (y0 ) hx (y) = f (x, y) for all y [0, 1]. Then
g(x) = sup {f (x, y)} = f (x, y0 )
y[0,1]

Is well dened. We need to show that g : [0, 1] R is continuous. Note that

we have not only proved that g is well dened, but have also shown that for
any x [0, 1], there exists y [0, 1] with g(x) = f (x, y).
Let x [0, 1]. We need to show that limzx g(z) = g(x). So suppose
this is false. Then there exists E > 0 and a sequence (xn ) with xn x but
|g(xn ) g(x)| > E.
For each xn , pick yn [0, 1] such that g(xn ) = f (xn , yn ). Now consider
the sequence ((xn , yn ))nN . This a sequence in the compact set [0, 1] [0, 1],
hence has a convergent subsequence. In other words there exists (x/ , y / )
[0, 1] [0, 1] with (xnk , ynk ) (x/ , y / ). This implies that xnk x/ , but since
this a subsequence of a convergent sequence, it must also converge to x, and
so x/ = x.
f is continuous, and so
f (x, y / ) = lim f (xnk , ynk ) = lim g(xnk )
k

Hence, we must have |g(x)f (x, y / )| E. Pick y [0, 1] with f (x, y) = g(x).
Then f (x, y) f (x, y / ), by the denition of g, and so
f (x, y) f (x, y / ) E
On the other hand, f is uniformally continuous, since it is a continuous
function on a compact set. Pick a > 0 such that
d((z, w), (z / , w/ )) < = d(f (z, w), f (z / , w/ )) < E/3

and k suciently large that |x xnk |, |y / ynk | < / 2. Then |f (x, y)

f (xnk , y)| < E/3, and so
f (xnk , y) > f (x, y) E/3
Similarly, |f (xnk , ynk ) f (x, y / )| < E/3, and so
f (x, y / ) + E/3 > f (xnk , ynk )

Putting these together, we have

f (xnk , y) > f (x, y) E/3 > f (x, y / ) + E/3 > f (xnk , ynk )
This is a contradiction, since f (xnk , ynk ) = g(xnk ) = supy {f (xnk , y)}.

4
We will show that g / (x0 ) = f // (x0 )/2 by directly evaluating the limit of
dierence quotients. We have

g(x) g(x0 )
lim
= lim
xx0
xx0
x x0

f (x)f (x0 )
xx0

f / (x0 )

x x0

= lim

xx0

f (x) f (x0 ) (x x0 )f / (x)

(x x0 )2

Note that both the numerator and the denominator of the above expression
converge to 0 as x x0 . Since f is twice dierentiable at x0 , it must
be once dierentiable in some neighbourhood of x0 , otherwise the second
derivative would not even make sense. Thus we can apply LHopitals rule;
the derivative of the numerator is f / (x) f / (x0 ), while the derivative of the
denominator is 2(x x0 ). In other words, we have
lim

xx0

g(x) g(x0 )
f / (x) f / (x0 )
f // (0)
= lim
=
xx0
x x0
2(x x0 )
2

In particular, the limit exists, i.e. g is dierentiable at x0 .

5
f is integrable, with integral 0. Note that any closed interval [x, y] with
x < y contains a point z with f (z) = 0. Since f is non-negative, this implies
that for any partition P , we have L(f, P ) = 0.
Let E > 0. We will nd a partition P with the upper Riemann sum U (f, P ) <
2E, which will prove the result. Consider the function g : [E, 1] R, which
is equal tof restricted to the interval [E, 1]. g has only nitely many points
of discontinuity, namely, the nitely many points of the form 1/n > E for

n N. Hence by Rudin Theorem 6.10, g is integrable. Since all lower

Riemann Sums of g are zero, we must have
1

g(x) = 0
1

In particular, there exists a partition P of [E, 1] with U (g, P ) < E.

Now consider the partition of [0, 1] dened by P / = P {0}. Then all
but the rst term of U (f, P / ) is contained in U (g, P ). More precisely, we
have
U (f, P / ) = ( sup f (x))(E 0) + U (g, P ) < E + E = 2E.
x[0,1]

Which proves the result.

6
Since f : [0, 1] R is integrable, it is bounded, i.e. |f (x)| < M for all
x [0, 1]. We may assume M > 1. Let E > 0. Let < E/(2M ).
Note that 0 < 1 < 1, and so by Rudin Theorem 3.20 limn (1 )n = 0.
Let N be sucienly large that n > N = (1 )n < . Then for any
0 x 1 and any n > N , we have xn < . Hence for n > N , we have
1

f (x)xn dx
0

|f (x)|xn dx <
0

M dx < M <
0

E
2

On the other hand xn 1 for x [0, 1], and so

1

f (x)xn dx <
1

|f (x)|xn dx <

M dx = M <
1

E
2

Putting these together, we have

1

f (x)xn dx
0

f (x)xn dx <

f (x)xn dx +
1

E
E
+ = E.
2 2

Thus limn

1
0

f (x)xn dx = 0.

MIT OpenCourseWare
http://ocw.mit.edu

18.100C Real Analysis

Fall 2012

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.