The Algebra of Continuous Functions

This theorem is the quiet workhorse behind an enormous share of elementary and advanced mathematics. In calculus itself, it is exactly what licenses the claim that every polynomial and every rational function is continuous on its natural domain, which in turn underlies the Intermediate Value Theorem applications used to locate roots and the justification that limits of rational expressions can be computed by direct substitution. In real analysis, it is the base case for inductive constructions: continuity of $\sin, \cos, e^x$, and their combinations (via power series or other definitions) reduces, after the hard work of defining those functions, to exactly this kind of algebraic bookkeeping. In numerical analysis and scientific computing, algorithms like Newton's method or bisection implicitly rely on continuity of the algebraic combinations of functions being solved, guaranteeing that small perturbations in input produce small, controlled perturbations in output. In control theory and engineering, transfer functions are rational functions of a complex variable, and the same quotient-continuity reasoning (adapted to $\mathbb{C}$) governs where poles — points of discontinuity — can occur, which is central to stability analysis. In statistics and probability, continuity of algebraic combinations of cumulative distribution functions and density functions is what allows sums and products of continuous random variables' functionals to remain well-behaved. In cryptography and computer science, continuity arguments appear more indirectly, but the same "build complex continuous objects from simple continuous pieces via algebra" philosophy recurs whenever one shows that an error function or loss function (as in optimization and machine learning, e.g., a sum of squared differences) is continuous, which is a prerequisite for many convergence guarantees of gradient-based methods.

Ask the tutor: Why does the quotient rule for continuity need $g(a) \neq 0$ specifically, and what exactly can go wrong at points where $g(a) = 0$?

Continuity at $a$ says: values of $f$ near $a$ cluster tightly around $f(a)$, and you can force the clustering as tight as you like ($\varepsilon$) by looking close enough ($\delta$). Now picture two continuous functions $f$ and $g$ near a point $a$, each trapped in its own thin horizontal band around $f(a)$ and $g(a)$ once $x$ is close enough to $a$. If both are trapped, their sum is trapped in a band around $f(a) + g(a)$ — you just add the widths. Their product is a little subtler (multiplying two small errors can interact with the sizes of $f(a)$ and $g(a)$ themselves), but it is still controllable by elementary bounding, exactly the way you tame products of limits. Division is the delicate one: dividing by something that could be zero or nearly zero is dangerous, so we need a firm buffer — the hypothesis $g(a) \neq 0$ — that keeps $g(x)$ bounded away from $0$ for $x$ near $a$. The picture below shows two continuous graphs $f$ and $g$ near $x = a$, each confined to a shrinking band as $x \to a$, and their sum landing in the correspondingly shrinking band around $f(a)+g(a)$.

Remark on non-circularity. Every step above used only: (i) the definition of continuity as a limit equality, and (ii) previously established limit laws (sum, difference, scalar multiple, product, quotient rules for limits), which are theorems about limits that do not themselves presuppose continuity or this theorem. No step assumed the conclusion or any special structure of $f, g$ beyond continuity at the single point $a$.

Example 1 — an asymmetric rational combination. Let

where $c_1$ is the real root of $x^3 + 2 = 0$ (namely $c_1 = -\sqrt[3]{2}$) and $c_2 = 1/2$. We show $h$ is continuous at $a = 2$, using only the theorem above plus the elementary facts that constants and the identity function $x \mapsto x$ are continuous everywhere (each checked directly from the $\varepsilon$-$\delta$ definition in one line). Building blocks. The identity function $\iota(x) = x$ is continuous at every point: given $\varepsilon > 0$, take $\delta = \varepsilon$; then $|x - a| < \delta$ immediately gives $|\iota(x) - \iota(a)| = |x-a| < \varepsilon$. Every constant function is continuous everywhere, as noted in the proof. By repeated use of parts (3) and (4) of the theorem (products and scalar multiples), $x \mapsto x^2 = \iota(x)\cdot \iota(x)$ is continuous everywhere, and $x \mapsto x^3 = \iota(x)\cdot \iota(x)\cdot \iota(x)$ is continuous everywhere (continuity of a product of two continuous functions is continuous, apply again with the third factor). By part (3), $x \mapsto 3x^2$ and $x \mapsto 7x$ and $x\mapsto 2x$ are continuous everywhere. By part (1)/(2), $x \mapsto 3x^2 - 5$, $x \mapsto x^3+2$, and $x \mapsto 2x - 1$ are continuous everywhere (subtracting/adding the constant functions $5, 2, 1$). First fraction. At $a = 2$: the denominator $D_1(x) = x^3+2$ satisfies $D_1(2) = 8+2 = 10 \neq 0$. Since numerator $N_1(x) = 3x^2-5$ and denominator $D_1$ are both continuous at $2$ and $D_1(2)\neq 0$, part (5) gives that $N_1/D_1$ is continuous at $2$, with value

Second fraction. The denominator $D_2(x) = 2x - 1$ satisfies $D_2(2) = 3 \neq 0$. Since the constant function $1$ and $D_2$ are continuous at $2$ with $D_2(2) \neq 0$, part (5) gives $1/D_2$ continuous at $2$, with value $1/3$. Assembling. $h = N_1/D_1 + 7\iota - 1/D_2$ is a sum and difference of three functions, each continuous at $a=2$ (the first fraction, $7x$, and the second fraction), so by parts (1) and (2), $h$ is continuous at $2$, and

Note that $2 \neq c_1 \approx -1.26$ and $2 \neq c_2 = 0.5$, so $2$ is genuinely in the domain of $h$, and the computation is valid. This whole argument required no epsilon-delta work at all beyond the one-line verification for the identity and constant functions — everything else was pure bookkeeping via the theorem.

Example 2 — a function built from a floor-type piece, testing the boundary of the machinery. Let

Is $k$ continuous at $a = 2$? Notice $k$ is not simply "$f/g$ with $f,g$ continuous and $g(2)\neq 0$" at $a=2$, since the denominator $g(x) = x-2$ vanishes exactly at $2$ — so part (5) of the theorem, as stated, does not directly apply at $a=2$ (it applies only away from $2$). This example is chosen to show precisely the boundary of the algebra-of-continuity theorem and how to handle it correctly, rather than misapplying the theorem. For $x \neq 2$,

valid for all $x\ne 2$ since we may cancel the nonzero factor $x - 2$. So as a function, $k(x) = x+2$ for $x \ne 2$, and $k(2) = 4$ by definition — which happens to equal $\lim_{x\to 2}(x+2) = 4$ as well. To check continuity at $2$ rigorously: the function $x \mapsto x+2$ is continuous at $2$ by parts (1) (sum of the continuous identity function and the constant function $2$), with value $2+2=4$. Since $k(x) = x+2$ for all $x$ in the punctured neighborhood $x\neq 2$ of $a=2$ (in fact for all $x \neq 2$), and $k(2) = 4 = \lim_{x\to 2}(x+2)$, we get $\lim_{x\to2} k(x) = 4 = k(2)$, so $k$ is continuous at $2$ — but this required the algebraic simplification (cancelling the common factor) before applying continuity facts, not a direct application of the quotient rule at the problematic point. Away from $2$, e.g. at $a=3$, part (5) applies directly to $f(x)=x^2-4$, $g(x) = x-2$ with $g(3) = 1 \neq 0$, confirming continuity there too, consistent with $k(x)=x+2$ being continuous everywhere. This example illustrates an important scope point: the quotient rule proves continuity of $f/g$ at points where $g(a)\neq 0$; at points where $g(a) = 0$, the quotient rule is silent, and continuity (if it holds at all, as a removable-singularity phenomenon) must be checked by other means, such as algebraic simplification as done here.

The rigorous $\varepsilon$-$\delta$ framework that makes this theorem precise is due to Augustin-Louis Cauchy and, in its final polished form, Karl Weierstrass, who insisted on banishing vague language like "infinitely small quantities" in favor of quantified inequalities.