Continuity and Types of Discontinuities

A function $f$ is continuous at a point $a$ if $lim_{x \to a} f (x) = f (a)$ — the limit exists, the function is defined there, and the two agree. When this fails, the point is a discontinuity, classified as removable, jump, or essential (infinite/oscillatory). This is foundational calculus/analysis, traceable to Cauchy (1821) and Weierstrass (1860s). Surprising fact: a function can be continuous at exactly one point of the real line, or continuous at every irrational and discontinuous at every rational.

Continuity formalizes the intuitive idea that a function's graph can be drawn "without lifting the pen," but the rigorous $ε$ – $δ$ definition reveals subtleties the picture hides. A function $f : R \to R$ is continuous at $a$ when small changes in input produce small changes in output, made precise through limits. This matters because continuity is the minimal hypothesis behind nearly every major theorem of calculus — the Intermediate Value Theorem, Extreme Value Theorem, integrability of bounded functions, and the existence of solutions to differential equations. The classification of discontinuities (removable, jump, essential) gives a complete vocabulary for how and why continuity fails. A surprising fact: there exist functions, like Thomae's function, that are discontinuous at every rational number yet continuous at every irrational number — something no hand-drawn graph could ever capture.

The Core Insight

Continuity at a point is the equality of three things that are logically independent: (1) the function value $f (a)$ , (2) the limit $lim_{x \to a} f (x)$ , and (3) their being equal. Each can fail separately, and the type of discontinuity is determined precisely by which of the one-sided limits exist and whether they are finite. Removable means both one-sided limits exist, are finite, and equal each other (but not $f (a)$ ); jump means they exist and are finite but differ; essential means at least one fails to exist as a finite number.

Formal Statement

Continuity at a Point

Let $f$ be defined on an open interval containing $a$ . Then $f$ is continuous at $a$ if

x \to a lim f (x) = f (a) .

⟳This single equation silently demands three conditions —

f (a)

is defined, the limit exists, and they coincide.

Equivalently, in the $ε$ – $δ$ formulation:

\forall ε > 0 \exists δ > 0 such that ∣ x - a ∣ < δ ⟹ ∣ f (x) - f (a) ∣ < ε .

⟳Unfold the definition of

lim_{x \to a} f (x) = L

with

L = f (a)

; note

x = a

is now permitted since

∣ f (a) - f (a) ∣ = 0 < ε

automatically.

Here $ε$ measures output tolerance, $δ$ measures input tolerance, $a$ is the point, and $f (a)$ is the function value.

One-Sided Limits and Continuity

The left-hand and right-hand limits are

f (a^{-}) = x \to a^{-} lim f (x), f (a^{+}) = x \to a^{+} lim f (x) .

⟳Restrict the

ε

–

δ

statement to

x < a

(use

a - δ < x < a

) or

x > a

(

a < x < a + δ

) respectively.

A function is continuous at $a$ iff both one-sided limits exist, are finite, and

f (a^{-}) = f (a^{+}) = f (a) .

⟳A two-sided limit exists and equals

L

iff both one-sided limits exist and equal

L

; pair this with

[1]

Classification of Discontinuities

Suppose $f$ is discontinuous at $a$ . Then:

Removable discontinuity: $f (a^{-}) = f (a^{+}) = L$ (finite) but either $f (a)$ is undefined or $f (a) \neq = L$ .
Jump (first-kind) discontinuity: $f (a^{-})$ and $f (a^{+})$ both exist and are finite but $f (a^{-}) \neq = f (a^{+})$ . The jump is

J = f (a^{+}) - f (a^{-}) .

⟳Subtract the right-hand from the left-hand limiting value;

J = 0

collapses to removability.

Essential (second-kind) discontinuity: at least one of $f (a^{-}), f (a^{+})$ fails to exist as a finite number. Subtypes: infinite (a one-sided limit is $\pm \infty$ ) and oscillatory (a one-sided limit does not exist due to oscillation).

Algebra of Continuous Functions

If $f$ and $g$ are continuous at $a$ , then $f \pm g$ , $f g$ , and (if $g (a) \neq = 0$ ) $f / g$ are continuous at $a$ . If $g$ is continuous at $a$ and $f$ is continuous at $g (a)$ , then $f \circ g$ is continuous at $a$ .

Intermediate Value Theorem

If $f$ is continuous on $[a, b]$ and $y$ lies between $f (a)$ and $f (b)$ , then there exists $c \in [a, b]$ with $f (c) = y$ . This expresses that continuity forbids "gaps" in the range over an interval — the defining property a jump discontinuity violates.

Why It Holds / Derivation

Composition of Continuous Functions is Continuous

We prove $f \circ g$ is continuous at $a$ , given $g$ continuous at $a$ and $f$ continuous at $b := g (a)$ .

Fix $ε > 0$ . By continuity of $f$ at $b$ , there exists $η > 0$ such that $∣ y - b ∣ < η ⟹ ∣ f (y) - f (b) ∣ < ε$ .
Apply continuity of $g$ at $a$ with tolerance $η$ : there exists $δ > 0$ such that $∣ x - a ∣ < δ ⟹ ∣ g (x) - g (a) ∣ < η$ , i.e. $∣ g (x) - b ∣ < η$ .
Chain the implications: if $∣ x - a ∣ < δ$ , then $∣ g (x) - b ∣ < η$ (step 2), hence $∣ f (g (x)) - f (b) ∣ < ε$ (step 1).
Since $f (b) = f (g (a)) = (f \circ g) (a)$ , we have shown $∣ x - a ∣ < δ ⟹ ∣ (f \circ g) (x) - (f \circ g) (a) ∣ < ε$ .
As $ε$ was arbitrary, $f \circ g$ is continuous at $a$ . $■$

Sequential Characterization of Continuity

Claim:

f

is continuous at

a

iff for every sequence

x_{n} \to a

, we have

f (x_{n}) \to f (a)

( $\Rightarrow$ ) Suppose $f$ is continuous and $x_{n} \to a$ . Given $ε > 0$ , pick $δ$ from continuity. Since $x_{n} \to a$ , there is $N$ with $∣ x_{n} - a ∣ < δ$ for $n > N$ , hence $∣ f (x_{n}) - f (a) ∣ < ε$ . Thus $f (x_{n}) \to f (a)$ .
( $\Leftarrow$ ) Contrapositive: suppose $f$ is not continuous at $a$ . Then some $ε_{0} > 0$ defeats every $δ$ . Take $δ = 1/ n$ : there exists $x_{n}$ with $∣ x_{n} - a ∣ < 1/ n$ but $∣ f (x_{n}) - f (a) ∣ \geq ε_{0}$ .
Then $x_{n} \to a$ but $f (x_{n}) \neq \to f (a)$ , contradicting the hypothesis. Hence $f$ must be continuous. $■$

This equivalence is the workhorse for disproving continuity — exhibiting one bad sequence suffices.

Augustin-Louis Cauchy | 1789–1857

Cauchy gave the first systematic, limit-based definition of continuity in his Cours d'analyse (1821), replacing the vague Euler–Lagrange notion of a function "given by a single analytic expression." His phrasing — that an infinitesimal change in $x$ produces an infinitesimal change in $f (x)$ — was later sharpened into the static $ε$ – $δ$ form by Karl Weierstrass in his Berlin lectures (1860s). Cauchy's work also exposed a famous gap: he initially believed limits of continuous functions were continuous, an error corrected by uniform convergence (Seidel, Stokes, Weierstrass).

Building Intuition

Think of continuity as a promise about controllability: "Tell me how close you need the output, and I can tell you how close to keep the input." Equation $[2]$ is exactly this negotiation — you name $ε$ , I deliver $δ$ . A discontinuity is a broken promise: no matter how tightly you squeeze the input near $a$ , the output refuses to stay near $f (a)$ .

The three discontinuity types correspond to how the promise breaks:

Removable — the output does converge, but you wrote down the wrong value at the point (or forgot to write one). The fix is trivial: redefine $f (a)$ .
Jump — the output settles to one value from the left and a different value from the right; there is no single target.
Essential — the output never settles at all (it blows up or oscillates forever).

The three types of discontinuity at $x=a$

Worked Examples

Identifying a Removable Discontinuity

Given:

f (x) = \frac{x ^{2} - 9}{x - 3}

, with

f

undefined at

x = 3

. Steps:

Factor the numerator: $x^{2} - 9 = (x - 3) (x + 3)$ .
For $x \neq = 3$ : $f (x) = \frac{( x - 3 ) ( x + 3 )}{x - 3} = x + 3$ .
Compute one-sided limits: $f (3^{-}) = lim_{x \to 3^{-}} (x + 3) = 6$ and $f (3^{+}) = 6$ .
Both exist, are finite, and equal $6$ , but $f (3)$ is undefined.

Since the limits agree but $f (3)$ is missing, this is removable. Redefining $\tilde{f} (3) = 6$ makes $\tilde{f}$ continuous.

Removable discontinuity at x = 3, x \to 3 lim f (x) = 6

Computing a Jump and Classifying It

Given: The Heaviside-type function

f (x) = ⎩ ⎨ ⎧ 2 x + 1, 5, x^{2} + 1, x < 1, x = 1, x > 1.

Steps:

Left limit: $f (1^{-}) = lim_{x \to 1^{-}} (2 x + 1) = 2 (1) + 1 = 3$ .
Right limit: $f (1^{+}) = lim_{x \to 1^{+}} (x^{2} + 1) = 1 + 1 = 2$ .
Both finite, but $f (1^{-}) = 3 \neq = 2 = f (1^{+})$ .
Compute the jump using $[5]$ : $J = f (1^{+}) - f (1^{-}) = 2 - 3 = - 1$ .
Note that $f (1) = 5$ matches neither — irrelevant to classification once limits differ.

Jump discontinuity at x = 1, J = - 1

Common Misconceptions

"Continuous means you can draw it without lifting your pen." This is a heuristic, not a definition, and it fails badly:

Thomae's function is continuous at every irrational but you can never "draw" it.
The topologist's sine curve $sin (1/ x)$ is continuous on $(0, \infty)$ yet wiggles infinitely — no human pen can trace it near $0$ .
Continuity is pointwise/local; the "pen" intuition smuggles in connectedness of the domain, which is a separate concept.

"If $f (a^{-})$ and $f (a^{+})$ both exist, $f$ is continuous." False — they must also equal each other and $f (a)$ (equation

[4]

). Existence of one-sided limits alone permits a jump. "A removable discontinuity isn't a real discontinuity." It genuinely is one:

f

as given fails

[1]

. It is merely fixable by a single redefinition.

Continuity Cannot Be Tested by Limits Alone

Define $f (x) = sin (1/ x)$ for $x \neq = 0$ and $f (0) = 0$ . As $x \to 0^{+}$ , choose $x_{n} = \frac{1}{( 2 n + 1/2 ) π} \to 0$ giving $f (x_{n}) = sin ((2 n + 1/2) π) = 1$ , while $y_{n} = \frac{1}{( 2 n - 1/2 ) π} \to 0$ gives $f (y_{n}) = - 1$ . Two sequences approaching $0$ produce limits $1$ and $- 1$ , so $f (0^{+})$ does not exist. This is an oscillatory essential discontinuity. No choice of $f (0)$ removes it — contrast sharply with the removable case where redefinition cures everything.

Discontinuous Everywhere: The Dirichlet Function

D (x) = {1, 0, x \in Q, x \in / Q .

At every point

a

, choose rationals

q_{n} \to a

(giving

D (q_{n}) = 1

) and irrationals

r_{n} \to a

(giving

D (r_{n}) = 0

). By the sequential criterion, no limit exists anywhere, so

D

is discontinuous at every real number — every point is an essential (oscillatory) discontinuity. Yet Thomae's function

T (x) = 1/ q

for

x = p / q

in lowest terms and

T (x) = 0

for irrational

x

flips this: it is continuous at every irrational and discontinuous at every rational.

Connections

Continuity sits at the hub of analysis. It is the hypothesis that powers the Intermediate Value Theorem and Extreme Value Theorem, both of which fail instantly under a jump or infinite discontinuity. Differentiability strictly implies continuity (the converse fails — see the Weierstrass nowhere-differentiable function), connecting to Differentiability and the Derivative. The notion strengthens to Uniform Continuity, where the same $δ$ works across the whole domain, and to Lipschitz Continuity, which bounds the rate of change. In integration theory, the set of discontinuities controls Riemann integrability via the Lebesgue Criterion for Riemann Integrability (bounded $f$ is Riemann-integrable iff its discontinuity set has measure zero). Topologically, continuity generalizes to "preimages of open sets are open," linking to Topological Continuity and Open Sets.

Real-World Applications

Signal processing — jump detection. Edge detection in images and discontinuity detection in audio rely on locating jump discontinuities; the jump magnitude $J$ from $[5]$ quantifies edge strength (Canny edge detector).
Control systems. Bang-bang controllers and switching circuits produce intentional jump discontinuities; stability analysis must account for non-continuous control laws (Filippov solutions).
Thermodynamics — phase transitions. First-order phase transitions exhibit jump discontinuities in entropy/density (latent heat), while continuous (second-order) transitions show no jump — Ehrenfest's classification mirrors the discontinuity taxonomy exactly.
Financial modeling. Asset prices with "gaps" (overnight jumps) are modeled by jump-diffusion processes (Merton model); the jump component is a genuine discontinuity in the sample path.
Numerical analysis. Quadrature and root-finding (bisection relies on the IVT) require continuity; algorithms break or lose convergence order across discontinuities, motivating discontinuity-aware adaptive solvers.
Computer graphics. $C^{0}$ continuity (positional) vs. $C^{1}$ (tangent) vs. $C^{2}$ (curvature) continuity of splines and Bézier curves determines whether seams in surfaces look smooth — a direct engineering use of continuity hierarchy.
Fluid dynamics — shock waves. Solutions to hyperbolic conservation laws (Euler equations) develop genuine jump discontinuities (shocks) satisfying the Rankine–Hugoniot jump conditions.

Historical Development

c. 1665–1700 — Newton and Leibniz use continuity intuitively in the calculus, with no formal definition; functions are imagined as smoothly varying quantities.
1748 — Euler (Introductio in analysin infinitorum) defines functions via single analytic expressions, conflating continuity with the existence of one formula.
1817 — Bernard Bolzano proves an early form of the Intermediate Value Theorem and gives a purely analytic (limit-based) definition of continuity, decades ahead of its time but largely unread.
1821 — Cauchy's Cours d'analyse establishes the limit-based definition of continuity that becomes standard, founding rigorous analysis.
1829 — Dirichlet introduces his everywhere-discontinuous function, decisively separating "function" from "analytic formula."
1854–1872 — Riemann (integration) and Thomae (his function, 1875) explore functions with rich discontinuity sets, probing exactly which discontinuities permit integration.
1861–1872 — Weierstrass formalizes the static $ε$ – $δ$ definition in his Berlin lectures and constructs a continuous, nowhere-differentiable function (published 1872), shattering the belief that continuity implies smoothness.
1872 — Heine introduces uniform continuity and proves continuous functions on closed bounded intervals are uniformly continuous (Heine–Cantor theorem).
1902 — Lebesgue characterizes Riemann integrability through the measure of the discontinuity set, completing the analytic understanding of how discontinuities matter.

Subtleties & Limits

While continuity itself is fully understood, several deep questions surround the structure of discontinuity sets:

Which sets can be discontinuity sets? Theorem (classical): the set of discontinuities of any function $f : R \to R$ is an $F_{σ}$ set (countable union of closed sets). Conversely, every $F_{σ}$ set is the discontinuity set of some function. There is no function continuous exactly on the rationals, because $Q$ is not a $G_{δ}$ set (its complement-of-discontinuities would need to be $G_{δ}$ ). This is a striking, non-obvious constraint.
Baire classification. Pointwise limits of continuous functions (Baire class 1) have discontinuity sets that are meager, leading to the Baire hierarchy — a still-active area in descriptive set theory.
Generic differentiability. Continuous functions are "typically" nowhere differentiable (Banach, 1931): in the Baire-category sense, almost all continuous functions fail to be differentiable anywhere, inverting naive intuition about how special the Weierstrass function is.
Restriction problems. To what large subsets can an arbitrary function be restricted to become continuous? Blumberg's theorem (1922) guarantees a dense subset on which any $f : R \to R$ becomes continuous — a surprising rescue of order from chaos.