Math 424 B / 574 B Due Wednesday, Nov 25

Autumn 2015

Solutions for Homework 7

Solution for Problem 6, page 99. I’ll follow the hint from the assignment and work with
the function F : E → R2 defined by
F (x) = (x, f (x)).
(⇒) Suppose f is continuous. Then, since x 7→ x is continuous, F is continuous by part
(a) of Theorem 4.10 (because its components x 7→ x and f are continuous).
So the graph
Γ(f ) = {(x, f (x)) : x ∈ E}
of f is the image of E through the continuous function F . Since continuous functions preserve
compactness (Theorem 4.14), the graph is indeed compact.
(⇐) Conversely, assume that the graph Γ(f ) is compact. We want to show that f is
continuous, or equivalently (by the corollary to Theorem 4.8) that the inverse image f −1 (C)
of a closed subset C ⊆ R is closed.
f −1 (C) consists of those points x ∈ E such that f (x) ∈ C, or equivalently, such that the
second component of F (x) = (x, f (x)) belongs to C. In other words, f −1 (C) consists of the
first components of those points (x, f (x)) ∈ Γ(f ) which also belong to R × C. If I denote the
two coordinate functions on R2 by φ1 and φ2 as in Example 4.11, so that
φ1 (x1 , x2 ) = x1 and φ2 (x1 , x2 ) = x2 ,
then we’ve just proven the following formula:
f −1 (C) = φ1 Γ(f ) ∩ φ−1


2 (C) . (1)

C ⊆ R is closed, so φ−1 2
2 (C) ⊆ R is closed because the coordinate functions φi are continuous
(Example 4.11).
We are assuming that Γ(f ) is compact, so Γ(f ) ∩ φ−1
2 (C) is again compact, being a closed
subset of a compact metric space (Theorem 2.35).
Finally, φ1 : R2 → R is continuous, so it maps compact sets onto compact sets by Theorem
4.14. All in all, (1) now ensures that f −1 (C) ⊂ E ⊂ R is compact and hence closed by
Theorem 2.34. 

Solution for Problem 8, page 99. By uniform continuity there is a δ > 0 such that
|f (x) − f (y)| < 1 provided x, y ∈ E and |x − y| < δ. (2)
1
2

Since E ⊂ R is bounded, it can be covered by finitely many open intervals I1 up to In of

length δ. We then get
 
[n [n
f (E) = f  (E ∩ Ij ) ⊆ f (E ∩ Ij ).
j=1 j=1

(2) means that each f (E ∩ Ij ) is bounded (because it has diameter ≤ 1), so that f (E) is
contained in a finite union of bounded sets. Finite unions of bounded sets are bounded, so
we are done.
To see that the boundedness of E is essential, take say E = R and define f by f (x) = x for
all x ∈ E. It is immediate that f is uniformly bounded, but the range of f is all of R which
is unbounded. 

Solution for Problem 11, page 99. I will give the alternate proof for Problem 13 sepa-
rately, after this one. Here, I’m just showing that (f (xn ))n is Cauchy if f : X → Y is
uniformly continuous and (xn )n is Cauchy.
Fix ε > 0. We want to show that there is some positive integer N such that
n, m ≥ N ⇒ d(f (xn ), f (xm )) < ε. (3)
Now, by uniform continuity, we know that there is some δ > 0 such that
d(xn , xm ) < δ ⇒ d(f (xn ), f (xm )) < ε. (4)
On the other hand, because (xn )n is Cauchy, there is a positive integer N such that
n, m ≥ N ⇒ d(xn , xm ) < δ. (5)
Equations (4) and (5) together give (3), which completes the proof. 

Alternate solution for Problem 13, page 99. Let x ∈ X \E be an arbitrary point. First
off, we want to define the value g(x) of an extension g : X → R of f (this is what it means to
extend f from E to all of X; see Problem 5 from the textbook for the definition of ’extension’).
Since E is dense, there is a sequence (xn )n in E that converges to x. Since xn → x in
X, the sequence (xn )n is Cauchy. By Problem 11 solved above the absolute continuity of f
implies that (f (xn ))n is also Cauchy. Since by part (c) of Theorem 3.11 the real line R is
complete, the Cauchy sequence (f (xn ))n converges to some y ∈ R; set g(x) = y. We now
have to check a few things.
Step 1: The value g(x) = y from above is well defined. Here’s what this means:
Recall that in order to set g(x) = y I chose some sequence (xn )n ⊂ E converging to x. What I
want to prove now is that the real number y from the construction above does not depend on
the specific sequence xn that converges to x, only on x itself (otherwise the formula g(x) = y
is not justified).
Suppose you have sequences xn → x and x0n → x with xn , x0n ∈ E and fix an arbitrarily
small ε > 0. Since f is uniformly continuous, there is some δ > 0 such that
d(p, q) < δ ⇒ d(f (p), f (q)) < ε. (6)
 3

For sufficiently large n we have

δ δ
d(xn , x) < , d(x0n , x) < ,
2 2
which implies
δ δ
d(xn , x0n ) ≤ d(xn , x) + d(x0n , x) <
+ = δ.
2 2
This together with (6) implies d(f (xn ), f (x0n )) < ε for sufficiently large n, so the distance
between the limits of the Cauchy sequences (f (xn ))n and (f (x0n ))n is at most ε (this last step
follows for example from Problem 6 from Homework 4). This is true for any ε > 0, so the two
sequences (f (xn ))n and (f (x0n ))n must have the same limit. So indeed y = limn f (xn ) does
not depend on (xn )n itself so long as this sequence converges to x.
This concludes the proof of Step 1. Since I now have a well defined value for g at each
point in X \ E. Setting g = f on E gives me a function g : X → R which is an extension of
f in the sense of Problem 5. We want a continuous extension, so there’s one more thing to
check:
Step 2: The function g : X → R defined above is continuous. We will actually show
that g is uniformly continuous, which is even better.
Note first that the exact same procedure as above actually works for x ∈ E as well, so
we can define g uniformly for all points of X: pick your point x ∈ X arbitrarily, take some
sequence E 3 xn → x converging to it, and set g(x) to be the limit of f (xn ). This observation
will allow us to treat all points of X on an equal footing, without having to split into cases
depending on whether or not they are in E.
Fix ε > 0. We want to find some δ > 0 such that
d(p, x) < δ ⇒ d(g(p), g(x)) < ε.
Choose δ > 0 so small that
ε
p0 , x0 ∈ E, d(p0 , x0 ) < 3δ ⇒ d(f (p0 ), f (x0 )) < . (7)
3
(this is possible, since f : E → R is uniformly continuous).
Now, by the way we defined g, we can find x0 ∈ E such that d(x0 , x) < δ and d(f (x0 ), g(x)) <
ε
3.
Similarly, for any p ∈ X with d(p, x) < δ we can find p0 ∈ E with d(p0 , p) < δ and
d(f (p0 ), g(p)) < 3ε .
We then have
d(p0 , x0 ) ≤ d(p0 , p) + d(p, x) + d(x0 , x) < δ + δ + δ = 3δ,
which by (7) implies d(f (p0 ), f (x0 )) < 3ε . But then we get
ε ε ε
d(g(p), g(x)) ≤ d(f (p0 ), g(p)) + d(f (p0 ), f (x0 )) + d(f (x0 ), g(x)) < + + = ε,
3 3 3
which is what we wanted. 
Remark 1. Problem 13 also asks whether the result remains true if instead of real-valued
functions f we consider functions taking values in any Euclidean space Rk , or any compact
or complete or arbitrary metric space. I purposely phrased the proof so that it goes through
whenever f : E → Y lands inside a complete metric space Y .
You can’t drop completeness though: the identity function id : Q → Q cannot be extended
to a continuous function R → Q. Indeed, if it could, the image of this continuous function
4

would contain both 0 and 1 and hence would have to contain the entire interval [0, 1] by
Theorem 4.23. This means that its image can’t be contained in Q.
In conclusion, the answer to all of the questions at the end of Problem 13 is ’yes’ except
for the last one, which has a negative answer. 

E1. Give an example of a continuous function f : (0, 1) → R and a Cauchy sequence (xn )n
in (0, 1) such that (f (xn ))n is not Cauchy in R.
Solution. Take say f (x) = x1 . This is continuous by, say, Theorem 4.9 (dividing a continuous
function by a non-zero continuous functions yields something continuous).
Now have a look at the sequence n1 in (0, 1). It is Cauchy because it is convergent in the
larger metric space [0, 1), but applying f to it turns it into the sequence n, which is unbounded
and hence not Cauchy. 

E2. Show that a metric space (X, d) is disconnected if and only if there exists a continuous
function f : X → R whose range f (X) is the two-element set {0, 1}.
Solution. According to Problem 1 from Homework 4, a metric space is disconnected if and
only if it can be written as the disjoint union of two non-empty open subsets (or equivalently,
two non-empty closed subsets); see also Definition 3 in that homework assignment.
(⇒) Suppose first that I can write X A ∪ B, with A ∩ B = ∅ and both A and B are open
and non-empty. Define a function f : X → R by
(
0 if x ∈ A
f (x) =
1 if x ∈ B
Clearly, f (X) = {0, 1}. To see that f is continuous, use the criterion that the preimage of
every relatively open subset of {0, 1} ⊂ R is open. {0, 1} has four subsets, namely
∅, {0}, {1} and {0, 1}.
Their preimages through f are
∅, A, B and respectively X,
all of which are indeed open.
(⇐) If there is a function f : X → R as in the statement, then the inverse images A =
f (0) and B = f −1 (1) of the relatively open subsets
−1

{0} and {1} ⊂ {0, 1}

are open and non-empty in X, and moreover
X = f −1 ({0, 1}) = f −1 (0) ∪ f −1 (1) = A ∪ B,
so we can write X as a disjoint union of two non-empty open subsets. According to the
observation made at the beginning of the proof, this means that X is disconnected.