You are on page 1of 6

MAT1000 ASSIGNMENT 1

VITALY KUZNETSOV

Question 1 (Exercise 2 on page 37). Tne Cantor set C can also be described in terms of
ternary expansions.
(a) Every number in [0, 1] has a ternary expansion


X
x= ak 3k
k=1

where
Pak = 0, k1 or 2. We also note that this decomposition is not unique since, for example
1/3 = k=1 2/3 . We show that x C if and only if x has a representation as above where
every ak is 0 or 2.
Proof. Recall that be definition the Cantor set C = k=0 Ck , where C0 = [0, 1] and Ck+1
is obtained from Ck be removing the open middle-third subinterval of each of the disjoint
intervals that constitute Ck .
First we use induction to show that for all non-negative integers k, if a P is such that
[a, b] is one of the disjoint intervals constituting Ck , then a can be written as n=1 an 3
n
,
where P an {0, 2} for 0 < n k and an = 0 for n > k. If k = 0 then C0 = [0, 1] and
0= n=1 0/3k
as needed. Now we assume that the statement holds for k N and we show
that it is also true for k + 1. Assume that s is such that [s, t] is one of the disjoint intervals
constituting
P Ck+1 . If s is also an endpoint of a disjoint interval in Ck then we know that
s = n=1 an 3n , where an {0, 2} for 0 < n k and an = 0 for n > k. Hence, an {0, 2}
for 0 < n k + 1 and an = 0 for n > k + 1 as well. Now if s is not an endpoint of the
interval in Ck then by construction there is an interval [a, t] in Ck such that [s, t] [a, t].
Furthermore, s a = 2(t a)/3 by construction of C. Since length of [a, t] is 3k (on each
step we reduce the length of disjoint intervals by a factor of 3), we have that s = a + 2/3k+1 .
Using induction hypothesis, s = kn=1 an 3n + 2/3k+1 . That is, s = n
P P
n=1 an 3 such that
an {0, 2} for 0 < n k + 1 and an = 0 for n > k + 1 as claimed.
Now we make an observation that Ck consists of exactly P 2k disjoint intervals, which is
n
exactly the number of points in [0, 1] of the form s = n=1 an 3 , where an {0, 2}
for 0 < n k and an = 0 for n > k. Therefore, this observation combined with our
previous result implies that each such number is an endpoint of one of the disjoint intervals
constituting Ck .PWe also note that if s is P an endpoint of an interval [s, t] in Ck then t =
s + 1/3k = s + n=k+1 2/3 n
. That is, t =
n=1 an 3
n
such that the first k of an s are the
same as for s andPthe rest are equal to 2.

Now if x = n=1 an 3n such that an {0, 2} for all n, then we have that sk =

+
Pk n
P n
Pk n
P n
n=1 an 3 x = n=1 an 3 n=1 an 3 n=k+1 2/3 = tk for all k. That, is
x [sk , tk ] Ck for all k, which shows that x C.
Date: September 29, 2011.
1
2 VITALY KUZNETSOV

Conversely, assume that x C. Then for each k, there is an interval [xk , yk ] in Ck containing
x. First observe that the sequence xk converges to x since |xxk | < 2/3k for each k N. Next
we note that the previous results imply that for i < j, we can write xj = xi + jn=i+1 an 3n
P

with an {0, 2}. Thus, x = limk xk = limk kn=1 an 3n = n


P P
n=1 an 3 with an
{0, 2} for all n. 
(b) The Cantor-Lebesgue function is defined on C by


X bk
F (x) =
k=1
2k
where x = k
with ak {0, 2} and bk = a2k . We show that F is well-defined and
P
k=1 ak 3
continuous on C, and moreover F (0) = 0 as well as F (1) = 1.
Proof.
P Tok show that F is well-defined it suffices to show that a ternary
P representationP of0 x k=
k
a
k=1 k 3 with a k {0, 2} is unique. Assume x C and x = k=1 ka 3 = a
k=1 k 3
with ak , a0k {0, 2}. For a contradiction, let an 6= a0n for some n (and by well-ordering
principle of natural numbers we can choose the smallest suchPn). Then by our result from
n1
part (a), we see that x lies in an interval [s, t] Cn with s = k=1 ak 3k . At the same time
n
it must also lie in the interval [a, b] Cn with a = s + 2/3 . We know that such intervals
are disjoint which leads to contradiction. Thus, ak = a0k for all k and F is well-defined.
Next we show that F is continuous. It is enough to show that if xn is a sequence of
points in C which converges to x, then F (xn ) converges to F (x). Fix  > 0. Then there
is n0 N such that 1/2n0 < . Since xn converges to x there is n1 > n0 such that for all
n > n1 , |x xn | < 1/3n1 . Then x and xn must lie in the same interval in Cn for all n > n1 .
Therefore, ternary expansion inPterms of 0 and 2 of x and xn must agree for the first n1
digits. Thus, |F (x) F (xn )| < k=n1 +1 2
k
= 2n1 < 2n0 <  for all n > n1 , which implies
that F (xn ) converges to F (x). This proves that F is continuous.
P P
k k
Finally,
Pusing geometric series we have that 0 = k=1 0/3 and 1 = k=1 2/3 , thus

F (0) = k=1 0/2k = 0 and F (1) = k=1 1/2k = 1.
P

(c) We prove that F : C [0, 1] is surjective, that is, for every y [0, 1] there is x C
such that F (x) = y.
= k
P
Proof. Let y [0, 1]. Then y has a binary expansion, that P is y k k=1 bk 2 such that
bk {0, 1}. We let ak = 2bk
P{0, 2}. kBy part (a), x = k=1 ak 3 C. Now we observe
that by construction, F (x) = k=1 kb 2 = y, which shows that F is indeed surjective. 
(d) One can also extend F to be a continuous function on [0, 1] as follows. Note that if
(a, b) is an open interval of the compliment of C, then F (a) = F (b). Hence we may define
F to have the constant value F (a) in that interval.
Proof. First we show that if (a, b) is an open interval of the compliment of C, then F (a) =
F (b). If (a, b) is a subinterval of the compliment of C with endpoints in C, then this interval
was removed from [0, 1] during construction of C on stage n + 1, for some n P0. In fact
it was removed from some interval [s, t] Cn . By partP(a), we know P that s = nk=1 ak 3k .
Since the length of [s, t] is 1/3n , a = s + 1/3n+1 = n
k=1 ak 3
k
+ k
k=n+2 2/3 and b =
Pn k n+1
k=1 ak 3 + 2/3 .
MAT1000 ASSIGNMENT 1 3

We compute that F (a) = nk=1 bk 2k + k


= nk=1 bk 2k + 2n1 = F (b) as
P P P
k=n+2 2
required.
Next we show that this extension of F is in fact continuous. It is immediate that F is
continuous at x 6 C, since there is a neighbourhood of x on which F is constant. Now
assume x C and we wish to show that F is continuous at x. As before, it enough to show
that if xn is a sequence that converges to x, then F (xn ) converges to F (x). If xn C, then
let yn = xn . If xn 6 C, then it lies in (sn , tn ) C c . In that case, let yn = sn if x < xn and
yn = tn if xn < x. By construction, yn C and |yn x| |xn x| for all n. Therefore, yn is
a sequence of points in C that converge to x. Hence, by continuity of F on C we must have
that F (yn ) converges to F (x). We note that by construction we have that F (xn ) = F (yn ) for
all n, which shows that F (xn ) converges to F (x) as required. Thus F is indeed continuous
on [0, 1].

Question 2 (Exercise 4 on page 38). We construct a closed set C so that starting with [0, 1]
at the k th stage of the construction one removes 2k1 centrally situated open intervals each
of length lk with l1 + 2l2 + . . . + 2k1 lk < 1.

(a) We show that if
P
2k1 lk < 1 then m(C) = 1 P 2k1 lk > 0.
k=1 k=1

Proof. We consider a compliment of C in [0, 1], that is the set S = [0, 1] C. We know that
the set S is a union of the collection {I }A of open intervals removed from [0, 1] during
the construction of C. Each interval in this collection is disjoint from all the other intervals
in the collection since each time we remove centrally situated intervals. Furthermore, this
collection of open intervals is countable since on each stage of the construction we remove 2k1
intervals. Therefore, S is measurable as countable union of measurable sets and by countable
interval removed on the k th stage of the
P
additivity we have m(S) = m(I ). Since each P
construction is of length lk this sum is equal to k=1 2
k1
lk . Now we use the fact that

[0, 1] is a disjoint union of C and S and countable additivity to conclude that 1 = m[0, 1] =

m(C)+m(S) P 2k1 lk (note that we implicitly use the fact that C is measurable
= m(C)+ k=1
as a compliment of measurable set). Finally, we write m(C) = 1 P 2k1 lk > 0 and the
k=1
proof is complete. 
(b) We show that if x C then there exists a sequence of points {xn }
n=1 such that xn 6 C,
yet xn x and x In , where In is a sub-interval in the compliment of C with |In | 0.
Proof. Let x C and define Cn to be the set containing all points of [0, 1] which has not
been removed during the first n stages of the construction of C (this set is analogous to the
set Cn in the construction of the Cantor set). By construction, Cn is a disjoint union of 2n
closed intervals. Note that there are exactly 2n closed intervals in Cn since each interval in
Cn1 gives a rise to two new intervals in Cn , that is on each step the number of intervals is
twice the number of intervals on the previous step. Let [a, b] be one of these intervals in Cn
such that x [a, b]. On the next stage of the construction an open interval (s, t) of length
ln is removed from [a, b]. We take xn to be the midpoint of (s, t) and we let In = (s, t).
Note that the distance from x to xn is less than the length of the interval [a, b] since both
of these points lie in it. On the other hand, by construction each of the intervals in Cn
has equal length and using the fact that there are 2n such intervals, we conclude that the
4 VITALY KUZNETSOV

length of [a, b] is bounded above by 2n . Therefore, we obtain a sequence {xn }


n=1 such that
n
|xn x| < 2 , xn 6 C and xn In [0, 1] Cn for all n. It follows that this sequence
P n1to x and it remains to verify that |In | vanishes as n . We
converges
n1
know that the series
n=1 2 ln converge and hence it must be the case that limn 2 ln = 0. We observe
that 0 ln 2n1 ln and by Squeeze theorem we conclude that |In | = ln 0 as n as
desired. 
Remark: The sequence of intervals In constructed in the proof above has the following
property. For all n, if J is an open interval which contains xn and is a subset of the
then it must be the case J In = (s, t). This is the case, since we always
compliment of C,
remove centrally situated open intervals when constructing C and this implies that endpoints
s and t will never be removed and will remain in C. This observation will be used in the
next proof.
(c) We show that as consequence of (b), C is perfect and it contains no open interval.
Proof. We have seen earlier that the compliment S of C in [0, 1] is a union of open intervals
and hence it is open. Therefore, C is closed in [0, 1] and it implies that it is also closed in
R. Therefore, to show that C is perfect we only need to verify that it has no isolated points.
For a contradiction, assume that x is an isolated point of C. Then there exist r > 0 such

that (x r, x + r) C = {x}. Now let {xn }n=1 be a sequence of points in S converging to
x satisfying properties in (b). Since xn x as n , there exists n0 N such that for
all n n0 , |x xn | < r. That is, for all n n0 , xn lies either in (x r, x) or (x, x + r).
Hence, by the remark we have made earlier, for all n n0 , In contains either of these two
intervals which implies that the length of each In with n n0 is at least r > 0. However,
this contradicts the fact that |In | 0 as n . Therefore, we conclude that C has no
isolated points and is perfect.
Now we prove that C contains no open interval. We assume otherwise and derive a
contradiction. Let (a, b) C and consider the midpoint x of (a, b). Since x C, by

part (b) we know that there exist a sequence {xn }n=1 such that xn 6 C and xn x. Then
there exists n N such that |x xn | < ba which implies that xn (a, b) C. This is
2
a contradiction since xn 6 C by assumption. Thus, we conclude that C contains no open
interval. 
(d) We show that C is uncountable.1
Proof. To show that C is uncountable we will construct a surjection f : C [0, 1] and since
[0, 1] is uncountable it will complete the proof.
Let x C and [a, b] be one of the disjoint intervals in Cn1 , n > 1 such that x [a, b].
On the next stage of the construction of C an open interval (s, t) is removed from [a, b],
resulting in disjoint intervals [a, s] and [s, t] lying in Cn . We let In0 = [a, s] and In1 = [s, t].
Since x C Cn it must lie in one of theseP intervals. We define an = 0 if x In0 and
n
a n = 1 otherwise.
PFinally, we set f (x) = n=1 an 2 (Note that this series converge since
P n n
n=1 an 2 n=1 2 = 1 and f is well-defined).
1Note that my original argument was only valid in the case when C has positive measure and was based
on a simple fact that any set of positive measure is uncountable. I would like to thank Mustazee Rahman
for this correction.
MAT1000 ASSIGNMENT 1 5

Now we P show that f is surjective. Let y [0, 1] be arbitrary. Then y has a binary
expansion n=1 an 2
n
where an {0, 1}. Note that the sequence {an }n=1 induces a nested
an
sequence of intervals In constructed as above starting with I0 = [0, 1]. By Nested Intervals
Theorem, an an
n=1 In is non-empty. Furthermore, since each In is one of the disjoint intervals
in Cn , we know from part (b) that |Inan | < 2n and hence |Inan | 0 as n . Therefore,
Nested Intervals Theorem tells us that in this case an
n=1 In = {x} for some unique real
number x. Note that x C since x Cn for all n. Moreover, by definition of f we have that
f (x) = y and this proves that f is surjective. 
Question 3 (Exercise 5 on page 39). Suppose E is a given set and On is the open set

On = {x : d(x, E) < 1/n}


(a) If E is compact, then m(E) = limn m(On ).
1
Proof. We start with an observation that On+1 On since if x On+1 then d(x, E) < n+1 <
1
n
and hence x On . Also, each On is measurable since each On is an open set.TThus, On s
form a decreasing sequence of measurable sets. Therefore, if we show that E = n=1 On and
m(O1 ) < then the desired T conclusion will follow from downward measure continuity.
We first show that E =
T
n=1 On . To simplify the notation we will write O = n=1 On .
If x E then d(x, E) = inf yE |x y| = 0 < 1/n for all n N. Thus, x On for all
n N and hence x O, which shows that E O. On the other hand, if x On then
d(x, E) = inf yE |x y| < 1/n and hence there exists yn E such that |x yn | < 1/n.
Therefore, if x O then for all n N there exists yn E such that |x yn | < 1/n. Note
that the sequence yn converges to x by construction and this implies that x E since E is
compact (and hence it is closed). Thus we conclude that O E and it follows that O = E.
It remains to show that m(O1 ) < . We will do so by showing that O1 is bounded and
the conclusion will follow since every bounded measurable set has a finite measure. Since E
is compact it is bounded and thus it is contained in some ball Br where r is the radius of the
ball. Consider now Br+2 which is a ball of radius r + 2 centred at the same point as Br . We
claim that O1 Br+2 , that is no point of O1 lies outside of Br+2 . For a contradiction, assume
c c
x Br+2 O1 . In this case we must have that d(x, Br ) 2 since d(Br , Br+2 ) = inf |z y| = 2
c
and infinum is taken over all y Br , x Br+2 . However, since E Br and x O1 we have
that 1 > d(x, E) d(x, Br ) 2 which is a contradiction. Therefore, O1 is bounded and the
proof is complete.

(b) We give an example when conclusion in (a) fails for E which is closed and unbounded
or for E which is open and bounded.
Example 1. We first give an example when conclusion in (a) fails when E is closed
and unbounded. First consider E = N. Note that E is closed in R since its compliment
(, 0)
S
k=0 (k, k + 1) which is open (as a union of open sets). Furthermore, the set E
is unbounded, since any ball (interval) in R contains finitely many natural numbers and
hence can not contain the whole E. We notice thatSfor an arbitrary positive  we can cover
E with cubes [k 2k+1  
, k + 2k+1 ]. That is, E k=0 [k P

2k+1

, k + 2k+1 ] for any  > 0.
Therefore, by monotonicity and subaddivity we have m(E) k=0 m([k 
2k+1

, k + 2k+1 ])
P k
k=0 2 = 2 and since  was arbitrary we conclude that m(E) = 0. On the other
S
hand, we have On = {x : d(x, E) < 1/n} = k=0 (k 1/n, k + 1/n). For all n > 1 this
6 VITALY KUZNETSOV

union is disjoint and we can use additivity to write m(On ) =


P
k=0 2/n > . Therefore,
m(E) = 0 6= limn m(On ) as desired.
Example 2. Now we give an example when conclusion in (a) fails with E being open
and bounded. S Since Q 1is countable let q1 , q2 , q3 , . . . be an enumeration of Q [0, 1]. We
1
define E = k=1 (qk 2k+2 , qk + 2k+2 ). We note that E is open as a union of open P intervals
and it is bounded P since E [1, 2]. Furthermore, by subadditivity m(E) k=1 m(qk
1 1 k1
2k+2
, qk + 2k+2
) = k=1 2 = 1/2. On the other hand, since Q [0, 1] is dense in [0, 1],
we know that for any x [0, 1] there is q Q [0, 1] such that |x q| < 1/n. In other
words, d(x, E) < 1/n for any x [0, 1] which implies that [0, 1] On for all n. Therefore,
by monotonicity 1 = m([0, 1]) m(On ) for all n, which implies that 1 limn m(On ).
Thus, m(E) 1/2 < 1 limn m(On ) as desired.

You might also like