Continuity and Uniform Continuity — Formal Treatment

Continuity means small input changes produce small output changes near a fixed point; uniform continuity demands a single rate of control that works everywhere on a set at once. Key result: every continuous function on a compact set is automatically uniformly continuous (Heine–Cantor). Applies throughout real analysis, metric space theory, and numerical analysis. Formalized by Cauchy, Weierstrass, Bolzano, and Heine in the 19th century.

A function $f$ is continuous at a point $x_{0}$ if for every tolerance $ε > 0$ there is a neighborhood width $δ > 0$ (allowed to depend on both $ε$ and $x_{0}$ ) within which outputs stay $ε$ -close to $f (x_{0})$ . Uniform continuity strengthens this by requiring one $δ$ that depends on $ε$ alone, valid simultaneously across the entire domain. The distinction matters because uniform continuity guarantees the preservation of Cauchy sequences, the existence of continuous extensions, and the convergence of numerical schemes—properties ordinary continuity cannot ensure. The surprising fact: continuity and uniform continuity coincide on compact domains, so the gap between them only opens up on unbounded or "open-ended" sets like $(0, 1)$ or $R$ .

The Core Insight

Continuity is a local, pointwise property: the order of quantifiers is $\forall x_{0} \forall ε \exists δ$ . Uniform continuity is a global property: the order becomes $\forall ε \exists δ \forall x$ . Swapping the quantifier $\exists δ$ before $\forall x$ is the entire content of the distinction—the $δ$ must be chosen before you know which point you are looking at.

Formal Statement

Continuity at a Point

Let $f : D \to R$ with $D \subseteq R$ and $x_{0} \in D$ . Then $f$ is continuous at $x_{0}$ if

\forall ε > 0 \exists δ > 0 \forall x \in D : ∣ x - x_{0} ∣ < δ ⟹ ∣ f (x) - f (x_{0}) ∣ < ε .

⟳This is the

ε

–

δ

unpacking of

lim_{x \to x_{0}} f (x) = f (x_{0})

; the

δ

may legitimately depend on the pair

(ε, x_{0})

Here $ε$ is the output tolerance, $δ$ the input radius, $x_{0}$ the base point, and $D$ the domain. The function is continuous on $D$ if it is continuous at every $x_{0} \in D$ .

Uniform Continuity

A function $f : D \to R$ is uniformly continuous on $D$ if

\forall ε > 0 \exists δ > 0 \forall x, y \in D : ∣ x - y ∣ < δ ⟹ ∣ f (x) - f (y) ∣ < ε .

⟳Move the

\exists δ

to the left of all point quantifiers; now a single

δ (ε)

must control every pair

(x, y)

of nearby points.

The crucial structural difference from [1] is that $δ = δ (ε)$ has no dependence on the location of $x$ or $y$ .

Modulus of Continuity

The modulus of continuity of $f$ on $D$ is the function

ω_{f} (t) = sup {∣ f (x) - f (y) ∣ : x, y \in D, ∣ x - y ∣ \leq t}, t \geq 0.

⟳Take the supremum of output gaps over all input pairs separated by at most

t

Then $f$ is uniformly continuous on $D$ iff $ω_{f} (t) \to 0$ as $t \to 0^{+}$ . This recasts a logical condition into a single limit.

Heine–Cantor Theorem

If $K \subseteq R$ is compact (closed and bounded) and $f : K \to R$ is continuous, then $f$ is uniformly continuous on $K$ .

Lipschitz ⟹ Uniform Continuity

If there exists $L \geq 0$ such that

∣ f (x) - f (y) ∣ \leq L ∣ x - y ∣ \forall x, y \in D,

⟳Bound the output gap linearly by the input gap; this is the Lipschitz condition.

then $f$ is uniformly continuous on $D$ : given $ε$ , choose $δ = ε / L$ (for $L > 0$ ). The converse fails—uniform continuity does not imply Lipschitz (see $x$ below).

Why It Holds / Derivation

Heine–Cantor via Contradiction

Claim. Continuous

f

on compact

K

is uniformly continuous. Step 1. Suppose not. Then the negation of [2] holds: there exists a bad

ε_{0} > 0

such that for every

δ > 0

some pair fails. Take

δ = \frac{1}{n}

for each

n \in N

. Step 2. For each

n

obtain points

x_{n}, y_{n} \in K

with

∣ x_{n} - y_{n} ∣ < \frac{1}{n} yet ∣ f (x_{n}) - f (y_{n}) ∣ \geq ε_{0} .

⟳This is the literal failure of uniform continuity at scale

δ = 1/ n

Step 3. Since

K

is compact, the Bolzano–Weierstrass theorem gives a convergent subsequence

x_{n_{k}} \to p \in K

. Step 4. Because

∣ x_{n_{k}} - y_{n_{k}} ∣ < \frac{1}{n _{k}} \to 0

, the companion subsequence also satisfies

y_{n_{k}} \to p

. Step 5. By continuity of

f

p

f (x_{n_{k}}) \to f (p) and f (y_{n_{k}}) \to f (p),

⟳Sequential characterization of continuity applied to both sequences converging to the same limit

p

so $∣ f (x_{n_{k}}) - f (y_{n_{k}}) ∣ \to 0$ . This contradicts $∣ f (x_{n_{k}}) - f (y_{n_{k}}) ∣ \geq ε_{0} > 0$ from Step 2. Hence $f$ is uniformly continuous. $■$

Sequential Criterion for Failure of Uniform Continuity

f

is not uniformly continuous on

D

iff there exist sequences

(x_{n}), (y_{n})

D

with

∣ x_{n} - y_{n} ∣ \to 0 but ∣ f (x_{n}) - f (y_{n}) ∣ \neq \to 0.

⟳Directly negate [2]; "fails for some

ε_{0}

at every

δ

" produces witness sequences with vanishing input gap and non-vanishing output gap.

This is the standard tool for disproving uniform continuity in concrete cases.

Eduard Heine | 1821–1881

Heine, working in Halle, gave in 1872 the first rigorous published proof that continuity on a closed bounded interval implies what we now call uniform continuity, building on Weierstrass's lectures and Bolzano's earlier unpublished insights. The compactness-based argument was later abstracted by Cantor, and the result bears both names. Heine's broader program—pinning down the exact hypotheses under which limit operations may be exchanged—was central to the 19th-century "arithmetization of analysis" that replaced geometric intuition with $ε$ – $δ$ rigor.

Building Intuition

Picture an $ε$ -wide horizontal band and a $δ$ -wide vertical strip sliding along the graph. Continuity says: at each fixed point you can find some strip width keeping the curve inside the band there—but the required width may shrink to zero as you move into a steep region. Uniform continuity says: one strip width works everywhere at once; you can carry the same $δ$ -box along the whole graph and never have the curve escape the $ε$ -band. On a finite closed interval the function "can't get away" because compactness forbids the slope from blowing up unboundedly while staying continuous; on an open or infinite domain the function can sneak off to steeper and steeper behavior near a missing endpoint or out at infinity.

The $\delta$-box test: $f(x)=1/x$ on $(0,1)$ needs a smaller $\delta$ near $0$

Worked Examples

$f(x)=x^2$ on $\mathbb{R}$ is NOT uniformly continuous

Given.

f (x) = x^{2}

D = R

. Test using the sequential criterion [7]. Step 1. Choose

x_{n} = n + \frac{1}{n}

and

y_{n} = n

. Then

∣ x_{n} - y_{n} ∣ = \frac{1}{n} \to 0.

⟳Pick two sequences whose separation vanishes but that march off to infinity together.

Step 2. Compute the output gap:

∣ f (x_{n}) - f (y_{n}) ∣ = (n + \frac{1}{n})^{2} - n^{2} = 2 + \frac{1}{n ^{2}} \geq 2.

⟳Expand

(n + 1/ n)^{2} = n^{2} + 2 + 1/ n^{2}

and subtract

n^{2}

Step 3. The output gap stays

\geq 2

while the input gap

\to 0

. By [7],

f

is not uniformly continuous.

x^{2} is continuous on R but NOT uniformly continuous; ε_{0} = 2 works as the bad tolerance.

$f(x)=\sqrt{x}$ on $[0,\infty)$ IS uniformly continuous

Given.

f (x) = x

D = [0, \infty)

. Find

δ (ε)

. Step 1. Use the algebraic inequality (valid for all

x, y \geq 0

∣ x - y ∣ \leq ∣ x - y ∣ .

⟳From

(x - y)^{2} = x + y - 2 x y \leq ∣ x - y ∣

since

x + y - 2 x y \leq ∣ x - y ∣

holds for

x, y \geq 0

; take square roots.

Step 2. Given

ε > 0

, choose

δ = ε^{2}

. Then

∣ x - y ∣ < δ

gives

∣ x - y ∣ \leq ∣ x - y ∣ < δ = ε .

⟳Substitute

δ = ε^{2}

into [10] and simplify the square root.

Step 3. This

δ

depends on

ε

alone, so [2] holds globally.

x is uniformly continuous on [0, \infty) with δ = ε^{2}, yet it is n o t Lipschitz (slope \to \infty at 0) .

Common Misconceptions

Uniform continuity is a property of a function on a set, not of a function alone. The very same formula

f (x) = 1/ x

is uniformly continuous on

[1, \infty)

but not on

(0, 1)

. Always state the domain. Likewise, "

f

is continuous on a bounded set" does not imply uniform continuity—the set must also be closed (compact). And differentiability with bounded derivative gives Lipschitz hence uniform continuity, but unbounded derivative says nothing by itself:

x

has unbounded derivative yet is uniformly continuous.

Continuous but not uniformly continuous on $(0,1)$

Take $f (x) = sin (1/ x)$ on $(0, 1)$ . It is continuous everywhere on the open interval, and bounded by $1$ . Let $x_{n} = \frac{1}{2 π n}$ and $y_{n} = \frac{1}{2 π n + π /2}$ . Then $∣ x_{n} - y_{n} ∣ \to 0$ , but $f (x_{n}) = sin (2 π n) = 0$ while $f (y_{n}) = sin (2 π n + π /2) = 1$ , so the output gap is constantly $1$ . By [7] it is not uniformly continuous—the oscillation crowds infinitely near $0$ , defeating any single $δ$ .

Connections

Uniform continuity sits between mere Continuity at a Point and the stronger metric notions of Lipschitz Continuity and Hölder Continuity: Lipschitz ⟹ Hölder ⟹ uniformly continuous ⟹ continuous, with each implication strict. The Heine–Cantor theorem is a manifestation of Compactness in Metric Spaces, and the modulus-of-continuity formulation connects to Equicontinuity in the Arzelà–Ascoli theorem. Uniform continuity is precisely the regularity needed to extend a function continuously from a dense subset (e.g. $Q$ ) to its completion, linking it to Completion of Metric Spaces and to the construction of the Riemann Integral, where uniform continuity guarantees integrability of continuous functions on $[a, b]$ .

Real-World Applications

Numerical analysis (quadrature & ODE solvers): Uniform continuity bounds the error of Riemann-sum and Euler-method approximations uniformly over a grid, since one $δ$ controls all subintervals.
Function extension in signal processing: Continuously extending a uniformly continuous signal sampled on a dense set to its closure underlies reconstruction guarantees.
Computer-aided geometric design (CAD): Lipschitz/uniform-continuity bounds on parametric curves give provable bounds on rendering and tessellation error.
Control theory & stability: Uniform continuity of feedback maps ensures bounded-input bounded-output behavior independent of operating point.
Machine learning generalization: Lipschitz constants (a strong form of uniform continuity) bound how much a neural network's output can change, central to robustness and generalization certificates.
Physics — thermodynamic limits: Uniform continuity of state functions over compact phase regions justifies interchanging limits in equilibrium calculations.
Financial mathematics: Uniformly continuous payoff functions ensure stable discretization of option-pricing PDEs and Monte-Carlo schemes.

Historical Development

1817 — Bernard Bolzano, in Rein analytischer Beweis, gives an early rigorous (if unpublished-in-influence) $ε$ -style definition of continuity, anticipating uniform continuity implicitly.
1821 — Augustin-Louis Cauchy's Cours d'analyse defines continuity via infinitesimals, conflating pointwise and uniform notions—a famous source of later confusion.
1853 — Cauchy "proves" that a convergent series of continuous functions has a continuous sum, an error traceable to missing uniformity assumptions.
1854 — Bernhard Riemann's integration theory implicitly relies on uniform continuity of continuous functions on intervals.
1861–1870 — Karl Weierstrass, in Berlin lectures, sharply distinguishes pointwise from uniform behavior, introducing modern $ε$ – $δ$ language and uniform continuity explicitly.
1870 — Heinrich Eduard Heine articulates uniform continuity in print and connects it to closed bounded intervals.
1872 — Heine publishes the proof now called the Heine–Cantor theorem; Georg Cantor's work on closed sets supplies the compactness backbone.
1895–1906 — Émile Borel and Henri Lebesgue formalize the covering (compactness) principle, recasting Heine–Cantor as a corollary of the Heine–Borel theorem.
1906 — Maurice Fréchet's thesis abstracts these ideas to general metric spaces, giving uniform continuity its modern setting.

Subtleties & Limits

On general topological spaces uniform continuity is not even definable—it requires a uniform structure (Weil, 1937) or a metric, since it speaks of "comparable closeness at different points." This raises subtle questions: which topological spaces admit a metric making a given continuous map uniformly continuous? In infinite dimensions, the Heine–Cantor theorem fails for non-compact closed bounded sets (closed unit balls are not compact), so uniform continuity becomes genuinely restrictive; characterizing the functions and spaces where continuity still upgrades to uniform continuity, and quantifying moduli of continuity sharply for solution operators of nonlinear PDEs, remains an active research theme in analysis and optimization.