Deriving Uniform Control From Compactness

From Local Control to One Uniform Choice

Continuity gives control near each point, but the size of the neighborhood it provides may depend on that point. Compactness supplies the key step: if those neighborhoods cover the space, finitely many of them still cover it. This tutorial develops that finite-subcover method for uniform continuity and for uniform bounds on continuous functions.

For a metric space \((X,d_X)\), a map \(f:X\to Y\) into a metric space \((Y,d_Y)\) is continuous at \(x\in X\) if for every \(\varepsilon>0\) there is a \(\delta_x>0\) such that \(d_X(x,y)<\delta_x\) implies \(d_Y(f(x),f(y))<\varepsilon\). The radius \(\delta_x\) may depend on \(x\). The map is uniformly continuous if for every \(\varepsilon>0\) there is one \(\delta>0\) that works for every pair \(x,y\in X\): whenever \(d_X(x,y)<\delta\), we have \(d_Y(f(x),f(y))<\varepsilon\).

A topological space is compact if every cover by open sets has a finite subcover. The proof strategy here is to build an open neighborhood around each point where a desired estimate holds, use compactness to select finitely many such neighborhoods, and then take the most restrictive of the finitely many resulting controls.

Uniform Continuity on a Compact Metric Space

Proof. Suppose first that \(X\ne\varnothing\), and fix \(\varepsilon>0\). For each \(a\in X\), continuity at \(a\), applied with tolerance \(\varepsilon/3\), gives a radius \(r_a>0\) such that \(d_X(a,z)<r_a\) implies \(d_Y(f(a),f(z))<\varepsilon/3\). Consider the open balls \(B_{d_X}(a,r_a/2)\), one for each \(a\in X\). They cover \(X\), since \(a\) belongs to its own ball. Compactness gives a finite subcover with centers \(a_1,\ldots,a_m\). Set \(\delta=\min_{1\le i\le m} r_{a_i}/2\), which is positive because it is the minimum of finitely many positive numbers. Now let \(x,y\in X\) satisfy \(d_X(x,y)<\delta\). Choose an index \(i\) such that \(d_X(x,a_i)<r_{a_i}/2\). Then \(d_X(y,a_i)\le d_X(y,x)+d_X(x,a_i)<\delta+r_{a_i}/2\le r_{a_i}\). Both \(x\) and \(y\) are therefore within \(r_{a_i}\) of \(a_i\). The choices of the radii give \(d_Y(f(x),f(a_i))<\varepsilon/3\) and \(d_Y(f(y),f(a_i))<\varepsilon/3\). By the triangle inequality, \(d_Y(f(x),f(y))<2\varepsilon/3<\varepsilon\). This single \(\delta\) works for every pair \(x,y\), proving uniform continuity. \(\square\)

The cover uses balls with half the radii supplied by continuity. That margin matters: if \(x\) is in one of those smaller balls and \(y\) is sufficiently close to \(x\), then \(y\) is still within the full radius around the same center. Taking a minimum of the finitely many half-radii gives one distance control for the whole space.

This example has a direct algebraic estimate, so it does not need a finite cover to find \(\delta\). The theorem is more general: it applies even when no simple formula for the dependence of \(f(x)\) on \(x\) is available. Its finite-cover proof is precisely what converts the separate continuity radii into one uniform choice.

Finite Subcovers Also Produce Uniform Bounds

Proof. Assume \(X\ne\varnothing\). For every \(a\in X\), continuity makes \(U_a=f^{-1}((f(a)-1,f(a)+1))\) open, and \(a\in U_a\). These sets cover \(X\), so compactness gives a finite subcover \(U_{a_1},\ldots,U_{a_m}\). If \(x\in X\), it belongs to some \(U_{a_i}\), which means \(|f(x)-f(a_i)|<1\). The triangle inequality gives \(|f(x)|\le |f(a_i)|+|f(x)-f(a_i)|<|f(a_i)|+1\). Let \(M=1+\max_{1\le i\le m}|f(a_i)|\). This is a finite real number, and the displayed inequality gives \(|f(x)|<M\), hence \(|f(x)|\le M\), for every \(x\in X\). \(\square\)

The local allowance in this proof is \(1\), but the same method works with any fixed positive allowance. What matters is that finitely many centers suffice: their finitely many function values have a finite maximum, which then controls all values on the space.

A Uniform Positive Lower Bound

The same method controls a function away from zero. The function values need not be bounded above or below by a useful single local estimate in advance: positivity and continuity provide neighborhoods with positive local lower bounds, and compactness reduces these to finitely many.

Proof. For each \(a\in X\), continuity at \(a\) gives an open set \(U_a\) containing \(a\) on which \(|g(y)-g(a)|<g(a)/2\). The right-hand tolerance is positive because \(g(a)>0\). For \(y\in U_a\), this inequality implies \(g(y)>g(a)-g(a)/2=g(a)/2\). The \(U_a\) cover \(X\), so finitely many, say \(U_{a_1},\ldots,U_{a_m}\), cover \(X\). Set \(c=\min_{1\le i\le m}g(a_i)/2\). Each number in this finite minimum is positive, so \(c>0\). Every \(y\in X\) lies in some \(U_{a_i}\), and therefore \(g(y)>g(a_i)/2\ge c\). In particular, \(g(y)\ge c\) for every \(y\in X\). \(\square\)

Common Pitfall: Pointwise Choices Need Not Be Uniform

Continuity at every point does not by itself produce a single \(\delta\) that works everywhere. Compactness in the uniform-continuity theorem is a substantive hypothesis, not a formality. For a concrete failure, take \(X=[0,\infty)\) with its usual metric and \(f(x)=x^2\). This function is continuous at every \(x\), but it is not uniformly continuous.

To verify the failure, for each \(n\in\mathbb{N}\) set \(x_n=n\) and \(y_n=n+1/n\). Their distance is \(|x_n-y_n|=1/n\), which tends to \(0\). But their function values differ by \[ |f(y_n)-f(x_n)|=\left(n+\frac1n\right)^2-n^2 =n^2+2+\frac1{n^2}-n^2 =2+\frac1{n^2}>2. \] If uniform continuity held with output tolerance \(\varepsilon=1\), some \(\delta>0\) would work for every pair. Choose \(n\in\mathbb{N}\) large enough that \(1/n<\delta\). Then \(|x_n-y_n|<\delta\), but \(|f(y_n)-f(x_n)|>2>1\), a contradiction. The pointwise radii can shrink as the location moves farther out, and there is no finite subcover argument available for this noncompact domain.

A useful way to audit a proof is to ask whether the claimed radius or bound depends on the point. If it does, one still needs a reason that finitely many local choices suffice. Compactness supplies that reason for open covers; pointwise continuity alone does not.