Problem Set 4 Solutions, 18.

100C, Fall 2012

October 11, 2012

1
Let X be a complete metric space with metric d, and let f : X → X be a con­
traction, meaning that there exists λ < 1 such that d(f (x), f (y)) ≤ λd(x, y)
for all x, y ∈ X. Then there is a unique point x0 ∈ X such that f (x0 ) = x0 .

Proof:

Existence: Let x1 ∈ X be arbitary and inductively let xn+1 = f (xn ) for

n ∈ N. We will prove that (xn ) is a Cauchy sequence. Suppose inductively
that
d(xr+1 , xr ) ≤ λr−1 d(x2 , x1 ).
Then

d(xr+2 , xr+1 ) = d(f (xr+1 ), f (xr )) ≤ λd(xr+1 , xr ) ≤ λr d(x2 , x1 )

so that the above equation holds for all r ∈ N. For m > n, by repeated use
of the triangle inequality

d(xm , xn ) ≤ d(xm , xm−1 ) + d(xm−1 , xm−2 ) + . . . + d(xn+1 , xn ).

Hence,
λn−1 (1 − λm−n ) λn−1
d(xm , xn ) ≤ (λm−2 +. . . λn−1 )d(x2 , x1 ) = d(x2 , x1 ) ≤ d(x2 , x1 ).
1−λ 1−λ
Let E > 0. By theorem 3.20(e) and 3.3(b), there exists an N ∈ N such that

n ≥ N =⇒ λn−1 d(x2 , x1 ) < E(1 − λ)

and so
m, n ≥ N =⇒ d(xm , xn ) < E,

1
which shows (xn ) is Cauchy. Since X is complete (xn ) converges to some
x0 ∈ X. Given E > 0 there exists N ∈ N such that
E
n ≥ N =⇒ d(x0 , xn ) <
2
and so

d(x0 , f (x0 )) ≤ d(x0 , xN +1 )+d(f (xN ), f (x0 )) ≤ d(x0 , xN +1 )+λd(xN , x0 ) < E.

Since E was arbitary, d(x0 , f (x0 )) = 0 giving x0 = f (x0 ), as required.

Uniqueness: If f (x0 ) = x0 and f (y0 ) = y0 then

d(x0 , y0 ) = d(f (x0 ), f (y0 )) ≤ λd(x0 , y0 ) =⇒ (1 − λ)d(x0 , y0 ) ≤

0 =⇒ d(x0 , y0 ) ≤ 0.
Thus d(x0 , y0 ) = 0 giving x0 = y0 .

2
We have a convergent sequence xk → x, and a bijective function g : N → N,
with an inverse function g −1 , and we wish to show that x'k also converges to
x. Pick E > 0; we will find N ' such that k ' > N ' =⇒ d(x, xk' ) < E

Since xk → x, there exists N ∈ N such that k > N =⇒ d(x, xk ) < E. Now,

pick an N ' ∈ N greater than max{g −1 (1), g −1 (2) . . . g −1 (N )}, which is al­
ways possible since this is a finite set. Let k ' > N ' , and consider x'k = xg(k' ) .
Let k = g(k ' ); then we must have k > N . If not, then k ≤ N and by the
definition of N ' we have k ' = g −1 (k) < N ' < k ' , a contradiction. So we have
d(x, x'k ) = d(x, xg(k) ) < E. So this N ' works, and we are done.

This statement is no longer true if g is not one-to-one. As a counter-example,

consider the sequence of real numbers xk = 1/k, and the function g : N → N
give by g(n) = 1 if n is odd, g(n) = 2 if n is even. Then xk → 0, but x'k
simply alternates between 1 and 1/2, and hence is not a Cauchy sequence
and cannot converge.

2
3
To set notation, we use boldface for vectors in Rn , i.e. x ∈ Rn , and super­
scripts with the same letter non-boldfaced for components of that vector,
i.e. xj is the j’th component of x, 1 ≤ j ≤ n. We use lower subscripts for
sequences; {xi } will be a sequence in Rn , and xji is the j’th component of
the i’th vector in the sequence.

With that set, we can proceed with the problem. Suppose first that xi → x;
we need to show that xij → xj as i → ∞. Note first that, for any two vectors
n
y, z ∈ Rn , we have (y j − z j )2 ≤ l=1 (y l − z l )2 . Taking the positive square
root of both sides, we see that |y − z j | ≤ ||y − z||. Now, take any E > 0,
j

and N sufficiently large that for k > N , ||x − xk || < E. Then by what we
just showed |xj − xjk | < E, so this N also works for E and the sequence {xjk },
so we have xjk → xj as desired.

For the other direction, suppose that xjk converges to some real number
xj for each 1 ≤ j ≤ n. Then take the vector x whose j’th component is
xj . We will show that xk → x. Fix E > 0. Since xkj → xj , we can choose
√
for each j a natural number N j such that for k > N j , |xj − xjk | < E/ n;
recall that n here is the dimension Rn . Now take N bigger than any of the
N 1 , N 2 , . . . N n ; we claim that for k > N , ||x − xk || < E, so that this N works
for this choice of E, and we have shown that xk → x. We compute

� � √ √
||x−xk || = (x1 − xk1 )2 + (x2 − xk2 )2 + · · · + (xn − xkn )2 < (E/ n)2 + · · · + (E/ n)2

= nE2 /n = E

Where we used the fact that the square root function is increasing in the
second step. This completes the proof.

4
Recall that the p-adic metric is defined as follows: if a, b ∈ Z, let n be the
largest power of p that divides a − b, i.e. pn |(a − b), but pn+1 t (a − b). Then
d(a, b) = 1/pn .

3
 k−1 i
Pk−1
Now we wish to show that the sequnce xk = i=0 p is Cauchy. Note
that p does not divide any xk ; indeed, xk − 1 is divisible by p, and no conse­
quetive numbers are divisible by p. Consider any pair xn , xm , where without
loss of generality n > m. Then

n−1
n m
n −1 n
n −1 n−
nm−1
i i i m
xn − xm = p −( p)= p =p ( pi ) = pm xn−m
i=0 i=0 i=m i=0

Since p does not divide xn−m , this means that the largest power of p divid­
ing xn − xm is m. In other words, d(xn , xm ) = 1/pm . This formula shows
that {xk } is Cauchy. Indeed, let E > 0. By Rudin 3.20 (e), we can find
N ∈ N sufficiently large that 1/pN < E. Then if n > m > N , we have
d(xn , xm ) = 1/pm < 1/pN < E, and so the sequence is Cauchy.

Now consider the case p = 2. By the formula for the sum of a geomet­
ric series we have
n −1
n 2n − 1
xn = 2i = = 2n − 1
2−1
i=0

So xn − (−1) = 2n , and has n the highest power of 2 dividing it. This says
that d(xn , −1) = 1/2n . But since the numbers 1/2n → 0 as n → ∞, this
shows that xn → 1, so the sequence converges.

4
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.