Ma · Mathematics

When Does a Uniformly Continuous Map Extend to the Closure?

Published By the 9to5Science editorial team

At a glance

  1. 1. Check completeness

    The codomain Y must be complete for the general extension theorem to hold.

  2. 2. Approach a closure point

    Choose a sequence in A converging to the point in the closure of A.

  3. 3. Take the image limit

    Uniform continuity makes the image sequence Cauchy, so completeness gives its limit in Y.

  4. 4. Verify the extension

    Show the limit does not depend on the chosen sequence, then prove uniform continuity and uniqueness.

Advertisement

A uniformly continuous function cannot make nearby points in its domain jump far apart. This controls the values of the function along sequences approaching points outside the domain. If the codomain is complete, those values converge inside the codomain, giving an extension to the closure.

The completeness assumption is essential. The statement is not true for an arbitrary metric space Y, so it should be included in the theorem.

The extension theorem

Let X and Y be metric spaces, let A be a subset of X, and suppose that Y is complete. If f:A→Yf:A\to Y is uniformly continuous, then there is a unique uniformly continuous function

f‾:A‾→Y\overline{f}:\overline{A}\to Y

such that f‾(a)=f(a)\overline{f}(a)=f(a) for every a∈Aa\in A. Here A‾\overline{A} denotes the closure of A in X. No completeness assumption on X is needed.

The idea is to define the value at a point x∈A‾x\in\overline{A} by taking a sequence from A that approaches x and then taking the limit of its image under f.

Constructing the extension

Fix x∈A‾x\in\overline{A}. Since x is in the closure of A, there is a sequence (an)(a_n) in A such that an→xa_n\to x. Because ff is uniformly continuous, the sequence (f(an))(f(a_n)) is Cauchy: for any ε>0\varepsilon>0, choose δ>0\delta>0 so that points of A less than δ\delta apart have images less than ε\varepsilon apart. Since (an)(a_n) converges, its terms are eventually within δ\delta of one another, so their images are eventually within ε\varepsilon.

Completeness of Y now guarantees that (f(an))(f(a_n)) converges to a point of Y. Define

f‾(x)=lim⁡n→∞f(an).\overline{f}(x)=\lim_{n\to\infty}f(a_n).

This definition must not depend on which sequence approaching x is chosen. Suppose also that bn∈Ab_n\in A and bn→xb_n\to x. Then

dX(an,bn)≤dX(an,x)+dX(x,bn)⟶0.d_X(a_n,b_n)\le d_X(a_n,x)+d_X(x,b_n)\longrightarrow 0.

Uniform continuity implies dY(f(an),f(bn))→0d_Y(f(a_n),f(b_n))\to 0. Since both image sequences converge, their limits are equal. Thus f‾(x)\overline{f}(x) is well-defined.

If x∈Ax\in A, you can use the constant sequence an=xa_n=x. The definition then gives f‾(x)=f(x)\overline{f}(x)=f(x), so this really is an extension.

Why the extension is uniformly continuous

Let ε>0\varepsilon>0. Uniform continuity of f gives a δ>0\delta>0 such that, for all a,b∈Aa,b\in A,

dX(a,b)<δ⟹dY(f(a),f(b))<ε/2.d_X(a,b)<\delta\quad\Longrightarrow\quad d_Y(f(a),f(b))<\varepsilon/2.

Take x,z∈A‾x,z\in\overline{A} with dX(x,z)<δ/2d_X(x,z)<\delta/2. Choose sequences an,bn∈Aa_n,b_n\in A with an→xa_n\to x and bn→zb_n\to z. The distances dX(an,bn)d_X(a_n,b_n) converge to dX(x,z)d_X(x,z), so they are eventually less than δ\delta. Therefore the corresponding image distances are eventually less than ε/2\varepsilon/2. Taking limits gives

dY(f‾(x),f‾(z))≤ε/2<ε.d_Y(\overline{f}(x),\overline{f}(z))\le \varepsilon/2<\varepsilon.

Thus f‾\overline{f} is uniformly continuous. The use of ε/2\varepsilon/2 allows for a non-strict inequality when passing to the limit.

Why the extension is unique

Suppose g:A‾→Yg:\overline{A}\to Y is another continuous extension of f. For any x∈A‾x\in\overline{A}, choose an∈Aa_n\in A with an→xa_n\to x. Continuity of g gives

g(x)=lim⁡n→∞g(an)=lim⁡n→∞f(an)=f‾(x).g(x)=\lim_{n\to\infty}g(a_n)=\lim_{n\to\infty}f(a_n)=\overline{f}(x).

So g and f‾\overline{f} agree at every point. In fact, continuity is enough for uniqueness; uniform continuity of the competing extension is not required.

A concrete example

Let A=(0,1)A=(0,1), viewed as a subset of X=RX=\mathbb{R}, and define f(x)=sin⁡xf(x)=\sin x with codomain Y=RY=\mathbb{R}. The function is uniformly continuous on A: by the mean value theorem, ∣sin⁡x−sin⁡y∣≤∣x−y∣|\sin x-\sin y|\le |x-y|. Since R\mathbb{R} is complete, the theorem gives a unique uniformly continuous extension to [0,1][0,1]. It is

f‾(x)=sin⁡x(x∈[0,1]).\overline{f}(x)=\sin x\qquad (x\in[0,1]).

For example, at the newly added endpoint 0, take an=1/na_n=1/n. Then f(an)=sin⁡(1/n)→0f(a_n)=\sin(1/n)\to 0, so the extension must assign the value 0 there. At the endpoint 1, any sequence in (0,1)(0,1) converging to 1 gives the value sin⁡1\sin 1.

What can fail without completeness

Take X=RX=\mathbb{R}, A=Q∩(0,1)A=\mathbb{Q}\cap(0,1), and Y=QY=\mathbb{Q} with the usual metric. The identity map f(q)=qf(q)=q is uniformly continuous. But its extension to A‾=[0,1]\overline{A}=[0,1] would have to send an irrational point such as 2/2\sqrt{2}/2 to that same real number, which is not in Y. No continuous Q-valued extension exists. The failure occurs because Q is not complete.

Common mistakes

  • Leaving out completeness: The theorem requires a complete codomain, or some separate reason that the image limits lie in Y.
  • Assuming the sequence choice is obvious: The extension is defined through sequences, so independence from the chosen approximating sequence must be proved.
  • Confusing continuity with uniform continuity: A continuous function on A need not extend continuously to its closure. Uniform continuity supplies the needed Cauchy control.
  • Using completeness of the wrong space: It is Y, the codomain, that must be complete. X itself may be incomplete.

If A is empty, its closure is empty and the extension statement is vacuous. Otherwise the construction above applies to each point of A‾\overline{A}.

References

  1. Principles of Mathematical Analysis, Walter Rudin
  2. Topology, James R. Munkres

Spotted an error? Send a correction. Read how we write and check articles in our editorial policy.