Tietze extension theorem: Difference between revisions

From Topospaces
Line 19: Line 19:
'''Proof''': We write <math>f_0 = f</math>.
'''Proof''': We write <math>f_0 = f</math>.


# Let <math>C_1 = f^{-1}([-1,-1/3])</math> and <math>D_1 = f^{-1}([1/3,1])</math>. These are both closed subsets of <math>X</math> since they are both the inverse image of a closed set under a continuous map. Moreover, they are disjoint by definition. So, by fact (1), there exists a continuous function <math>g_1: X \to [-1/3,1/3]</math> such that <math>g_1(C_1) = -1/3</math> and <math>g_1(D_1) = 1/3</math>.
# Let <math>\! C_1 = f^{-1}([-1,-1/3])</math> and <math>\! D_1 = f^{-1}([1/3,1])</math>. These are both closed subsets of <math>A</math> since they are both the inverse image of a closed set under a continuous map. Moreover, they are disjoint by definition. Further, since <math>A</math> is closed in <math>X</math>, they are both closed in <math>X</math>. So, by fact (1), there exists a continuous function <math>\! g_1: X \to [-1/3,1/3]</math> such that <math>\! g_1(C_1) = -1/3</math> and <math>\! g_1(D_1) = 1/3</math>.
# Define <math>f_1 = f_0 - g_1</math> as a function on <math>A</math>. Note that <math>f_1</math> is a function on <math>A</math> taking values in <math>[-2/3,2/3]</math>. Iteratively, we proceed as follows:
# Define <math>\! f_1 = f_0 - g_1</math> as a function on <math>A</math>. Note that <math>\! f_1</math> is a function on <math>\! A</math> taking values in <math>\! [-2/3,2/3]</math>. Iteratively, we proceed as follows:
## From the previous stage, we have a continuous function <math>f_i</math> on the closed subset <math>A</math> taking values in <math>[-(2/3)^i,(2/3)^i]</math>.
## From the previous stage, we have a continuous function <math>f_i</math> on the closed subset <math>A</math> taking values in <math>\! [-(2/3)^i,(2/3)^i]</math>.
## Let <math>C_i = f_i^{-1}[-(2/3)^i,-(1/3)(2/3)^i]</math> and <math>D_i = f_i^{-1}[(1/3)(2/3)^i,(2/3)^i]</math>. Note that both <math>C_i</math> and <math>D_i</math> are closed subsets of <math>A</math> (since <math>f_i</math> is continuous) and hence in <math>X</math>, since <math>A</math> is closed in <mah>X</math>.
## Let <math>\! C_{i+1} = f_i^{-1}[-(2/3)^i,-(1/3)(2/3)^i]</math> and <math>\! D_{i+1} = f_i^{-1}[(1/3)(2/3)^i,(2/3)^i]</math>. Note that both <math>C_{i+1}</math> and <math>D_{i+1}</math> are closed subsets of <math>A</math> (since <math>f_i</math> is continuous) and hence in <math>X</math>, since <math>A</math> is closed in <math>X</math>.
## By fact (1), find a function <math>g_{i+1}:X \to [-(1/3)(2/3)^i,(1(3/(2/3)^i]</math> such that <math>g_{i+1}(C_i) = (-1/3)(2/3)^i</math> and <math>g_{i+1}(D_i) = (1/3)(2/3)^i</math>.
## By fact (1), find a function <math>g_{i+1}:X \to [-(1/3)(2/3)^i,(1(3/(2/3)^i]</math> such that <math>\! g_{i+1}(C_{i+1}) = (-1/3)(2/3)^i</math> and <math>\! g_{i+1}(D_{i+1}) = (1/3)(2/3)^i</math>.
## Define <math>f_{i+1} = f_i - g_{i+1}</math>. We see that <math>f_{i+1}</math> is a function on <math>A</math> taking values in <math>[-(2/3)^{i+1},(2/3)^{i+1}</math>.
## Define <math>\! f_{i+1} = f_i - g_{i+1}</math>. We see that <math>f_{i+1}</math> is a function on <math>A</math> taking values in <math>\! [-(2/3)^{i+1},(2/3)^{i+1}]</math>.
# Define <math>g</math> as <math>\sum g_i</math>. This sum is well-defined at each point and takes values in <math>[-1,1]</math>: Note that the absolute value of <math>g_i</math> is bounded by the geometric progression <math>(1/3) + (1/3)(2/3) + (1/3)(2/3)^2 + \dots = 1</math>. Similarly, the lower bound is <math>-1</math>. Further, since <math>g_n</math> are bounded by the geometric progression in absolute value, the series <math>g_n</math> coverges. So the sum is well-defined and takes values in <math>[-1,1]</math>.
# Define <math>g = \sum_{i=1}^\infty g_i</math>. This sum is well-defined at each point and takes values in <math>[-1,1]</math>: Note that the absolute value of <math>g_i</math> is bounded by the geometric progression <math>(1/3) + (1/3)(2/3) + (1/3)(2/3)^2 + \dots = 1</math>. Similarly, the lower bound is <math>-1</math>. Further, since <math>g_n</math> are bounded by the geometric progression in absolute value, the series <math>g_n</math> converges. So the sum is well-defined and takes values in <math>[-1,1]</math>.
# The function <math>g</math> is continuous: This follows from the fact that each <math>g_i</math> is continuous and the co-domain of the <math>g_i</math>s approaches zero. {{fillin}}
# The function <math>g</math> is continuous: This follows from the fact that each <math>g_i</math> is continuous and the co-domain of the <math>g_i</math>s approaches zero. {{fillin}}
# <math>g|_A = f</math>: Let <math>x \in A</math>. Then, <math>f_1(x) = f(x) - g_1(x)</math>. Inductively, <math>f_n(x) = f(x) - \sum_{i=1}^n g_i(x)</math>. Since the upper and lower bound on <math>f_n</math> tend to zero as <math>n \to \infty</math>, <math>f_n(x) \to 0</math> as <math>n \to \infty</math>. Thus, <math>f(x) = \sum_{i=1}^\infty g_i(x) = g(x)</math> for <math>x \in A</math>.
# <math>\! g|_A = f</math>: Let <math>x \in A</math>. Then, <math>\! f_1(x) = f(x) - g_1(x)</math>. Inductively, <math>f_n(x) = f(x) - \sum_{i=1}^n g_i(x)</math>. Since the upper and lower bound on <math>f_n</math> tend to zero as <math>n \to \infty</math>, <math>f_n(x) \to 0</math> as <math>n \to \infty</math>. Thus, <math>f(x) = \sum_{i=1}^\infty g_i(x) = g(x)</math> for <math>x \in A</math>.

Revision as of 04:46, 17 July 2009

This article gives the statement, and possibly proof, of a basic fact in topology.

Statement

Suppose X is a normal space (i.e., a topological space that is T1 and where any two disjoint closed subsets can be separated by disjoint open subsets). Suppose A is a closed subset of X, and f:A[0,1] is a continuous map. Then, there exists a continuous map g:X[0,1] such that the restriction of g to A is f.

Facts used

  1. Urysohn's lemma: This states that for, given two closed subsets A,B of a normal space X, there is a continuous function h:X[0,1] such that h(A)=0 and h(B)=1.

Proof

Note that since [0,1] is homeomorphic to [1,1], it suffices to prove the result replacing [0,1] with [1,1]. We will also freely use that any closed interval is homeomorphic to [0,1], so Urysohn's lemma can be stated replacing [0,1] by any closed interval.

Given: A normal space X. A closed subset A of X. A continuous function f:A[1,1].

To prove: There exists a continuous function g:X[1,1] such that the restriction of g to A is f.

Proof: We write f0=f.

  1. Let C1=f1([1,1/3]) and D1=f1([1/3,1]). These are both closed subsets of A since they are both the inverse image of a closed set under a continuous map. Moreover, they are disjoint by definition. Further, since A is closed in X, they are both closed in X. So, by fact (1), there exists a continuous function g1:X[1/3,1/3] such that g1(C1)=1/3 and g1(D1)=1/3.
  2. Define f1=f0g1 as a function on A. Note that f1 is a function on A taking values in [2/3,2/3]. Iteratively, we proceed as follows:
    1. From the previous stage, we have a continuous function fi on the closed subset A taking values in [(2/3)i,(2/3)i].
    2. Let Ci+1=fi1[(2/3)i,(1/3)(2/3)i] and Di+1=fi1[(1/3)(2/3)i,(2/3)i]. Note that both Ci+1 and Di+1 are closed subsets of A (since fi is continuous) and hence in X, since A is closed in X.
    3. By fact (1), find a function gi+1:X[(1/3)(2/3)i,(1(3/(2/3)i] such that gi+1(Ci+1)=(1/3)(2/3)i and gi+1(Di+1)=(1/3)(2/3)i.
    4. Define fi+1=figi+1. We see that fi+1 is a function on A taking values in [(2/3)i+1,(2/3)i+1].
  3. Define g=i=1gi. This sum is well-defined at each point and takes values in [1,1]: Note that the absolute value of gi is bounded by the geometric progression (1/3)+(1/3)(2/3)+(1/3)(2/3)2+=1. Similarly, the lower bound is 1. Further, since gn are bounded by the geometric progression in absolute value, the series gn converges. So the sum is well-defined and takes values in [1,1].
  4. The function g is continuous: This follows from the fact that each gi is continuous and the co-domain of the gis approaches zero. Fill this in later
  5. g|A=f: Let xA. Then, f1(x)=f(x)g1(x). Inductively, fn(x)=f(x)i=1ngi(x). Since the upper and lower bound on fn tend to zero as n, fn(x)0 as n. Thus, f(x)=i=1gi(x)=g(x) for xA.