Existence of Continuous Functions
Existence of Continuous Functions
S. Kumaresan
School of Math. and Stat.
University of Hyderabad
Hyderabad 500046
kumaresa@gmail.com
The crucial fact here is the simple observation: If (X, d) is a metric space and x ∈ X, then
the function fx (y) := d(x, y) is continuous on X. For, by triangle inequality we have
Lemma 1. Let A be any nonempty subset of a metric space (X, d). Define d(x, A) ≡ dA (x) :=
inf{d(x, a) : a ∈ A}. Then |dA (x) − dA (y)| ≤ d(x, y) and hence dA is uniformly continuous
on X.
Proof. Observe from the triangle inequality d(x, a) ≤ d(x, y) + d(y, a), we obtain
so that dA (x) ≤ d(x, y) + dA (y). Thus, da (x) − dA (y) ≤ d(x, y). Interchanging x and y yields
the result.
Ex. 2. dA (x) = 0 iff x is a limit point of A. Hence if A is a closed set then d(x, A) = 0 iff
x ∈ A.
1
Lemma 3 (Urysohn’s Lemma for Metric Spaces). Let A and B be nonempty disjoint closed
subsets of a metric space X. Then there exists an f ∈ C(X, R) such that 0 ≤ f (x) ≤ 1 for
x ∈ X and f = 0 on A and f = 1 on B.
Proof. Note that for any x ∈ X, d(x, A) + d(x, B) 6= 0. For, if it were so, then d(x, A) = 0 =
d(x, B). Since A and B are closed x ∈ A and x ∈ B by the last exercise. This contradicts
our hypothesis that A ∩ B = ∅.
d(x,A)
The function f (x) := d(x,A)+d(x,B) meets our requirements.
Theorem 4 (Tietze extension theorem for metric spaces). Let Y be a closed subspace of a
metric space (X, d). Let f : Y → R be a bounded continuous function. Then there exists a
continuous function g : X → R such that g(y) = f (y) for all y ∈ Y and
Proof. Assume that f ≥ 0. Consider the function Mx (r) := sup{f (y) : y ∈ Y ∩ B(x, r)}.
Then, for each x ∈ X, Mx is real valued, bounded and monotonic increasing in r. Hence it is
Riemann integrable as a function of r over any finite interval. Let δ(x) := d(x, Y ). Note that
δ(x) > 0 iff x ∈
/ Y . We define g by g(x) = f (x) if x ∈ Y and if x ∈
/ Y,
Z 2δ(x)
1
g(x) := Mx (r) dr.
δ(x) δ(x)
Since 3δ → 0 as x → y (with x ∈ / Y ), it follows that, for any ε > 0, |g(x) − f (y)| < ε if x ∈
/Y
and δ(x) < δ0 for δ0 sufficiently small. On the other hand, |g(x) − f (y)| = |f (x) − f (y)| < ε
if x ∈ Y and d(x, y) < δ1 by continuity of f on Y . Thus g is continuous at y.
Consider next the continuity of g at z ∈ / Y . Let x be any point in X with d(x, z) <
d(z, Y )/3. Let α := d(x, z). Then 2α < d(x, Y ). For, otherwise, d(x, Y ) ≤ 2α so that
d(z, Y ) ≤ d(z, x) + d(x, Y ) < 3α. Hence α > d(z, X)/3, contradicting our assumption on x.
Since |δ(x) − δ(z)| ≤ d(x, z) = α (by Lemma 1), Mz (r) ≥ Mx (r − α) as B(x, r − α) ⊂ B(z, r)
2
and Mz (r) ≥ 0, we have
Z 2δ(x) Z 2δ(z)
1 1
g(x) − g(z) = Mx (r) dr − Mz (r) dr
δ(x) δ(x) δ(z) δ(z)
Z 2δ(x) Z 2δ(x)−2α
1 1
≤ Mx (r) dr − Mx (r − α) dr
δ(x) δ(x) δ(x) + α δ(x)+α
Z 2δ(x)−3α Z 2δ(x)
1 1
= Mx (r) dr + Mx (r) dr
δ(x) δ(x) δ(x) 2δ(x)−3α
Z 2δ(x)−3α
1
− Mx (s) ds, using a change of variable
δ(x) + α δ(x)
Z 2δ(x)−3α Z 2δ(x)
α 1
= Mx (r) dr + Mx (r) dr
δ(x)[δ(x) + α] δ(x) δ(x) 2δ(x)−3α
4M α
≤ ,
δ(x)
where M = supY f . A similar inequality holds with x and z interchanged. Hence g(x) → g(z)
as x → z. One easily checks that g is as desired.
To treat the general case, let m := inf Y f . Consider F := f − m. Apply the first case to
F to get a continuous extension G. Then g := G + c is as required.
We shall use Weierstrass approximation theorem to give a proof of Tietze theorem for Rn .
Proof. (of Tietze theorem for Rn .) Let us prove the result when the closed set is compact.
So, we assume that f : K → R is a continuous function on a compact subset of Rn . By
Weierstrass approximation theorem, for each k ∈ Z+ , there exists a polynomial pk such
P that
− p(x)| < 2−k−2 for all x ∈ K. We let q0 = p0 and qk := pk − pk−1 . Then pk = ki=1 qi
|f (x)P
and qk converges uniformly to f on K.
Let M := sup{|f (x)| : x ∈ K}. Then |p0 (x)| ≤ 2−2 + M for x ∈ K. Also, |qk (x)| < 2−k
for k ≥ 1 and x ∈ K. We let
h0 := max{−2−2 − M, min{q0 , 2−2 + M }},
hk := max{−2−k , min{qk , 2−k }}, for k ≥ 1.
n −k n
P for x ∈ K, hk is continuous non R and |hk (x)| ≤ 2 for x ∈ R and
Then hk (x) = qk (x)
for all k. Hence hk converges uniformly on R to a continuous function h. Then h is
continuous and h(x) = f (x) for x ∈ K.
We now extend to result if the subset K is any arbitrary closed subset. If K is bounded
the result follows from the previous paragraph. So, we assume that K is not bounded. Let
k ∈ N be such that B[0, k] ∩ K is nonempty. Let fk be the restriction of f to this nonempty
compact set. Then there exists a continuous function hk on Rn which extends fk . Define
(
hk (x), if x ∈ B[0, k]
gk (x) :=
f (x), if x ∈ K ∩ B[0, k + 1].
3
Then gk is continuous on the compact set B[0, k] ∪ (K ∩ B[0, k + 1]). There is an extension
hk+1 on Rn . Let (
hk+1 (x), if x ∈ B[0, k + 1]
gk+1 (x) :=
f (x), if x ∈ K ∩ B[0, k + 2].
Continuing in this way, we obtain a sequence (gm ) whose domains are increasing to Rn . Define
g(x) := gm (x) if x ∈ B[0, m]. Then g is an extension of f .
2 Normal Spaces
Lemma 5. A space X is a normal space iff for each closed set F and an open set V containing
F there exists an open set U such that F ⊂ U ⊂ U ⊂ V .
Proof. Let X be normal and F , V as above. Then F and X \ V are disjoint closed sets.
By normality of X there exist open sets U and W such that F ⊂ U and X \ V ⊂ W and
U ∩ W = ∅. Since U ⊂ X \ W and X \ W is closed, we see that U ⊂ X \ W ⊂ V . Thus U is
as required. The converse is left as an exercise.
Ex. 6. Recall that a dyadic rational is a rational number of the form p/2n , where p and n
are integers. Show that the set of dyadic rationals is dense in R.
Lemma 7. urys2 Let X be a normal space. If A and B are closed subsets of X, for each
dyadic rational r = k2−n ∈ (0, 1], there is an open set Ur with the following properties: (i)
A ⊂ Ur ⊂ X \ B, (ii) U r ⊂ Us for r < s.
Proof. Let U1 := X \ B. By the last lemma, there exist disjoint open sets V and W such that
A ⊂ V and B ⊂ W . Let U1/2 = V . Then, since X \ W is closed, we have
A ⊂ U1/2 ⊂ U 1/2 ⊂ X \ W ⊂ X \ B = U1 .
Applying the same lemma once again to the open set U1/2 containing A and to the open set
U1 containing U 1/2 , we get open sets U1/4 and U3/4 such that
Continuing this manner, we construct, for each dyadic rational r ∈ (0, 1), an open set Ur with
the following properties:
(i) U r ⊂ Us , 0 < r < s ≤ 1.
(ii) A ⊂ Ur , 0 < r ≤ 1. (iii) Ur ⊂ U1 , 0 < r ≤ 1.
More formally, we proceed as follows. We select Ur for r = k2−n by induction on n.
Assume that we have chosen Ur for r = k2−n , 0 < k < 2n , 1 ≤ n ≤ N − 1. To find Ur for
r = (2j + 1)2−N , 0 ≤ j < 2N −1 , observe that U j21−N and X \ U(j+1)21−N are disjoint closed
sets. So once again appealing to the last lemma, we can choose an open set Ur such that
U j21−N ⊂ Ur ⊂ U r ⊂ U(j+1)21−N .
4
Theorem 8. Urysohn’s Lemma. A space X is a normal space iff the following is true: For
any two disjoint closed subsets A and B of X there exists a continuous function f : X → [0, 1]
such that f = 0 on A and f = 1 on B.
Proof. Let Ur ’s be as in the last lemma. We define the function f so that the sets ∂Ur are
the level sets of f for the value r. We achieve this by defining
(
0, x ∈ Ur for all r
f (x) =
sup{r : x ∈
/ Ur }, otherwise.
Lemma 9. Let X and Y be Banach spaces. Let T : X → Y be a bounded linear map. Assume
that for y0 ∈ Y there exist constants M and r ∈ (0, 1) such that there exists x ∈ X such that
kxk ≤ M ky0 k and ky0 − T xk ≤ r ky k. Then there exists z ∈ X such that T z = y0 with
kz k ≤ M/(1 − r).
Proof. Let y ∈ Y be given. We may assume without loss of generality that ky k = 1. Given
y ∈ Y let z1 = x as given in the lemma. For y0 = y − T z1 , we can find a z2 ∈ X such
that kz2 k ≤ M ky − T z1 k ≤ M r and ky − T z1 − T z2 k ≤ r ky − T z1 k ≤ r2 . Proceeding
Pn by
n−1
P a sequence (zn ) in X such that (i) kzn k ≤ mr
induction, we get and (ii) ky − i=1 T zi k ≤
rn . The series ∞ z
n=1 n converges to an element z ∈ X. We have T z = y0 .
Theorem 10 (Tietze Extension Theorem). Let X be a normal space and Y a closed subset
of X. Let f ∈ Y := Cb (Y, R). Then there exists a g ∈ X := Cb (X, R) such that g(y) = f (y)
for all y ∈ Y and sup{g(x) : x ∈ X} = sup{f (y) : y ∈ Y }.
Proof. Let T : X → Y denote the restriction map g 7→ g|Y . We show that T satisfies the
hypothesis of the previous lemma. Without loss of generality, assume that |f (y)| ≤ 1 for all
y ∈ Y . Let A := f −1 ([−1, −1/3]) and B := f −1 ([1/3, 1]). Then A and B are closed in Y and
hence in X. By Urysohn’s lemma, there exists a g ∈ X such that |g(x)| ≤ 1/3 for x ∈ X
and g = −1/3 on A and g = 1/3 on B. One easily checks that kT g − f kX ≤ 1/3. If we take
M = 1/3 and r = 2/3, then T satisfies the previous lemma. Note that the assertion about
the equality of the norms is also obtained.
Ex. 11. Let X be a normal space and F a closed subset. Assume that f : F → (−R, R) be
a continuous function. Then f can be extended to a continuous function from X to (−R, R).
Hint: You may need Urysohn’s lemma.
Ex. 12. Let X be a normal space and F a closed subset. Assume that f : F → R be a
continuous function. Then f can be extended to a continuous function from X to R. Hint:
R is homeomorphic to (−1, 1).
5
Ex. 13. Assuming Tietze extension theorem, prove Urysohn’s lemma.
Ex. 14. Let A be a closed subset of a normal space X. Let f : A → S n be continuous. Show
that there exists an open set U ⊃ A (U depends on f ) and an extension g of f to U .
Ex. 15. Show that with the notation of Exer. 14 that f may not extend to all of X. Hint:
What happens (i) if n = 0 and X is connected or (ii) if X := B[0, 1] ⊂ Rn+1 , A := S n and f
is the identity?