You are on page 1of 6

Math 114 Problem Set 3

Roman Berens
with collaboration from Greg Parker and George Torres

September 23, 2014

1. Let En [0, 1] satisfy m(En ) > c > 0. Prove that lim sup En , the set of points belonging
to infinitely many En , has positive measure.
Since [0, 1] is compact and m(En ) [0, 1] for all n, we see that the sequence m(En ) has a convergent subsequence. Let Enk be this convergent subsequence. We also see that m (Enk ) > c
for all n, so lim m (Enk ) > 0. We see that
\

Enk lim sup Enk lim sup En .

By monotonicity of Lebesgue measure, we see that


!
\
m
Enk m (lim sup En ) .
k

But we know that m (

Enk ) = lim m (Enk ) > 0. Thus lim sup En has positive measure.

2. Let K be the middle-thirds Cantor set. Prove


T that K K contains an nonempty interval
(even though m(K) = 0). (Hint: Let K = Kn be the usual construction of K, where
Kn consists of 2n disjoint intervals of length 3n each. Prove by induction that Kn Kn = [1, 1].)
We see that K1 K1 consists of the union of four set differences:

 
 

2
1
1
, 1 0,
=
,1
3
3
3

 
 

1
2
1
0,
, 1 = 1,
3
3
3

 
 

1
1
1 1
0,
0,
= ,
3
3
3 3

 
 

2
2
1 1
,1 ,1 = ,
.
3
3
3 3
Thus we see that K1 K1 = [1, 1]. We now assume the
that
 inductive hypothesis,

m m+1
Kn1 Kn1 = [1, 1]. We consider an interval in Kn1 ,
,
. We see that
n1 3n1

3
1
1
the difference of any two points in this interval will lie in n1 , n1 . We then re3
3

m 3m + 1
move the middle third of this interval, dividing into two intervals,
,
and
3n1
3n


3m + 2 m + 1
, n1 . Considering either of these intervals separately, we see that the difference
3n
3


1 1
between any two points in the interval will lie in n , n . We see that the difference of any
3 3


1
1
two points with one in the first interval and the other in the second will lie in n1 , n or
3
3


1
1
,
. Thus we see that the total span of the differences of points in the two intervals
3n 3n1


1
1
in Kn is still n1 , n1 . Thus the removal of the middle third did not change the
3
3
span. Since this was true for an arbitrary interval, it must hold for every interval in Kn1 ,
2

so the span of the intervals of Kn1 does not change after all the middle thirds have been
removed. This means that the span of Kn is the same as that Kn1 , since it is precisely
by removing the middle thirds that we construct Kn from Kn1 . Since by our induction
hypothesis, Kn1 Kn1 = [1, 1], this must be true for Kn as well.
We then see that K =

Kn , so it follows that
\
\
\
K K =
Kn Kn =
(Kn Kn ) = [1, 1].
n

Thus K K contains a nonempty open interval.


3. Give an example of a continuous function h : R R and a measurable function
f : R R such that f h is not measurable.
Let f (x) be the Cantor function and let g(x) = f (x) + x. We see that g is bijective and thus
has a continuous inverse. It maps the Cantor set to a set of measure 1, i.e., m(g(C)) = 1.
Since g(C) has positive measure, it must have a nonmeasurable subset. Let this set be D.
Define E = g 1 (D). Since E C, m(E) = 0. Let E be the indicator function on E. E
is a measurable functions since E is a measurable set. We see that E g 1 = g(E) = D .
Letting h = g 1 , which is continuous, and f = E , which is measurable, we see that f h is
not measurable.
4. Prove that every open set U in R2 is a countable union of open rectangles, i.e., U =
where the Ii and Ji are open intervals in R.

Ii Ji ,

Consider an open set A R2 . We know that for any a A, there is some such that
B (a) A. We can choose coordinates so that a = (x, y) and consider the intervals (x, x+)
and (y , y + ), which we can think of as the projection of B (a) onto two axes. We then
pick four rational numbers qa1 , ra1 , qa2 , and ra2 , such that x qa1 1 (a) ra1 x + and
y qa2 2 (a) ra2 y + , where 1 (a) and 2 (a) are the projections of a onto the x
and y axes. We then define
Ra = (qa1 , ra1 ) (qa2 , ra2 ).
We see that Ra is a rectangle around a within B (a). We also see that we write A as
[
A=
Ra
aA

This is an uncountable union, but we can remove every rectangle in {Ra } that is repeated,
calling this new collection R. We thus can associate every rectangle in R to a unique set
of four rational numbers, i.e. we have an injection to Q4 . Thus R is countable. Since by
removing the repeated rectangles we have not lost any points in A, we see that
[
A=
Ra .
R

Since R is countable, we see that A is a countable union of open rectangles.

5. Prove that if h(x, y) is continuous and f (x) and g(x) are measurable functions, then
h(f (x), g(x)) is also measurable.
Define J : R R with J(x) = h(f (x), g(x)). Define H : R R2 with H(x) = (f (x), g(x)).
We see that J = h H, so J 1 = H 1 h1 . we want to show that J 1 (O) is measurable,
2
where O is an open set. Since h is continuous, we see that U = h1 (O) is an open set
S in R .
By our result from Problem S
4, U is a countableSunion of open rectangles,
i.e., U = Ii Ji .

1
1
1
1
We then see that H (U ) = H (Ii Ji ) =
f (Ii ) g (Ji ) . Since f and g are measurable functions, f 1 (Ii ) and g 1 (Ji ) are measurable sets. Thus J 1 (O) is a countable union
of a finite intersection of measurable sets, so it is measurable. Thus J is a measurable function.
6. Let f : R R be a Lipschitz function. This means that there exists a constant L
such that |f (x) f (y)| L|x y| for all x, y R.
(i) Show that if m(A) = 0 then m(f (A)) = 0.
Since a Lipschitz function is continuous, it must map open sets to open sets. Since A
is a set of measure zero, by Littlewoods first principle it can be covered by a collection of
open intervals {Ii } such that
X
`(Ii ) < .
i

We see that the maximum distance between any two points in an interval Ii is `(Ii ), so the
maximum
theSimages of any two points in the interval must be L`(Ii ). Since
S distance between
S
A Ii , f (A) f ( Ii ) = f (Ii ) because f is continuous. Thus by monotonicity and
subadditivity of Lebesgue measure and the definition of {Ii },
[
 X
X
m(f (A)) m
f (Ii )
m(f (Ii )) L
m(Ii ) < L.
i

Since  can be made arbitrarily small, we see that f (A) must have measure zero as well.
(ii) Show that if E is measurable then f (E) is measurable.
We first consider the case that E has finite measure. Given i = 1/i, we choose an compact
set Fi E such S
that m(E Fi ) = m(E) m(Fi ) < i . Thus m(Fi ) m(E) as i . We
then define F = Fi . Since m(E) < , we see that m(E F ) m(E Fi ) = m(E) m(Fi ).
Since the right hand side tends to 0 as i , it follows that E F has measure zero. We
then see that
f (E) = f (F ) f (E F ).
By the previous part, we see that f (E F ) is a set of measure zero. Thus,
[
f (E) = f (F ) =
f (Fi ).
i

Since the Fi are compact, they are measurable, and since a continuous function maps compact
sets to compact sets, the f (Fi ) are measurable. Since the measurable sets form a -algebra,
and f (E) is a countable intersection of measurable sets, f (E) is measurable.
Now we consider the case where m(E) = . We then express E as the disjoint union
of a countable collection {Ek } of measurable sets, each with finite measure. By the case
above, f (Ek ) is measurable, so f (E) = f (Ek ), so f (E) is measurable.
4

7. Let E [0, 1] be a set of positive measure.


(i) Construct a Lipschitz function f : [0, 1] [0, 1] such that f (E) = [0, 1] and f 0 (x) = 0
outside E. (Hint: first take a closed set F E with m(F ) > 0.)
We let F E be a closed set with m(F ) > 0. This set must always exist by Littlewoods
principles, which allow us to construct a closed subset of E with measure arbitrarily close to
m(F [0, x])
the measure of E. Let f (x) =
. We see that f (F ) = [0, 1], and it follows that
m(F )
f (E) = [0, 1]. Since F is closed, for all x
/ F there is some interval around x that is not in F
either. As x varies in this neighborhood the value of f (x) does not change. Thus f 0 (x) = 0
for x
/ F . Since F E, it follows a fortiori that f 0 (x) = 0 for all x
/ E.
We must now show that this function is Lipschitz. Without loss of generality, let us assume
that x > y. We see that f (x) f (y), as the interval [0, x] can only contain the same number
or more points of F than the interval [0, y], i.e., F [0, x] F [0, y]. However, even in the
maximal case where [0, x] and [0, y] are subsets of F , we see that |f (x) f (y)| = |x y|.
Thus it follows that |f (x) f (y)| |x y|, so f is Lipschitz with L = 1.
(ii) Prove that |E| = |R|.
We define an equivalence relation on E by x y if and only if f (x) = f (y). Let
E 0 = {x | x = c(E)}, where E is an equivalence class and c is a choice function. We
see that f (E 0 ) = [0, 1], since no values of f have been removed from the range. We also see
that f is injective on E 0 , since every element in E 0 maps to a different value in [0, 1]. Thus f
restricted to E 0 is a bijection. Thus |E 0 | = |[0, 1]|. Since E 0 E [0, 1], |E| = |[0, 1]| = |R|.
Use f to construct a bijection between E and [0, 1]. Use the fact that E [0, 1] and
the fact that f is surjective.
8. Let us say f : R R is a Borel function if f 1 (B) is a Borel set for every Borel
set B. Prove that the set A of all Borel functions is a vector space over R, and that it is
closed under composition (unlike the algebra of measurable functions).
We see that the zero map is a Borel function. If 0 A, the pre-image of A is R. If
0
/ A, the pre-image of A is . Both R and are Borel sets, so the zero map is a Borel
function.
We now show that the Borel functions are closed under scalar multiplication. Let f A,
R, and h = f . We only need to consider the action of h on sets of the form [a, ) and
(, b], as these generate the Borel sets. We see that
h1 ([a, )) = {x | f (x) a} = {x | f (x) a/} = f 1 ([a/, )).
Since f is a Borel function and [a/, ) is a Borel set, the pre-image of this set under f is a
Borel set. Thus h is a Borel function. (Note that if = 0, the h is the zero map, which we
already showed is a Borel function.)
We now shall show that the Borel functions are closed under addition. Letf , g A,
and h = f + g. We see that
h1 ([a, )) = {x | f (x) + g(x) a}.
5

We see that this set contains the set


{x | f (x) a + b and g(x) a b}
for any b R. If we take the union of sets of this form over all b R, the constraint
disappears and we recover the original set. Thus we can write this set up as
[
{x | f (x) + g(x) a} =
{x | f (x) a + b and g(x) a b}.
bR

We note that this is an uncountable union. However, we can restrict b to Q. Since Q is dense
in R, we can find a rational number less than b that is arbitrarily close to b, so this restriction
does not affect the set. We write it as
[
{x | f (x) + g(x) a} =
({x | f (x) a + b} {x | g(x) a b}
bQ

Since Q is countable, and f and g are Borel functions, this makes the set a countable union
of Borel sets, which is a Borel set. Thus f + g is a Borel function.
We now shall show that A is closed under composition. Let f , g A and h = f g.
Then we see that


h1 (B) = g 1 f 1 (B) = g 1 B 0 = B 00 ,
where B, B 0 , and B 00 , are Borel sets. Thus h is a Borel function.
Note that for all of the above, the construction for sets of the form (, b] proceeds
in the exact same way.

You might also like