Continuity and Limits

Intuition: Continuity means that information pulls back nicely: if you "zoom in" on a region of the target $Y$ (an open set), the points of $X$ that land there form an open region too. There are no jumps — nearby points are sent to nearby points, where "nearby" is expressed through the open sets. This definition is powerful because it works in any topological space, not just metric spaces.

Intuition: Continuity has many equivalent faces. The open-set version is the cleanest, but sometimes it is more convenient to argue about closed sets, or to say that the image of the closure is contained in the closure of the image — which expresses "points accumulating in $A$ get sent to points accumulating in $f(A)$."

Intuition: Continuity at a point says that we can force $f$ to land in any prescribed neighbourhood of $f(x_0)$ by starting from a small enough neighbourhood of $x_0$. This is the topological generalization of "nearby inputs produce nearby outputs."

Intuition: This is the familiar definition from calculus: given any desired output tolerance $\varepsilon$, we can find an input tolerance $\delta$ that guarantees it. The quantifier order matters: $\varepsilon$ is chosen first, then $\delta$ depends on both $\varepsilon$ and $x_0$.

Intuition: Sequential continuity says: if inputs converge to $x_0$, then outputs converge to $f(x_0)$. This formulation is often the easiest to use in practice, since sequences are concrete objects.

Intuition: Continuity composes. Reading the proof, this is essentially just a restatement of the identity $(g \circ f)^{-1}(W) = f^{-1}(g^{-1}(W))$.

Intuition: Ordinary continuity allows $\delta$ to shrink as we move to different parts of the domain; uniform continuity forbids this. A uniformly continuous function never becomes "arbitrarily steep" over its domain. Any uniformly continuous function is continuous; the converse fails in general.

Intuition: On a compact set, there is no "room at infinity" for continuity to deteriorate. Because the domain is totally controlled, the pointwise $\delta$'s admit a uniform lower bound.

Intuition: Lipschitz continuity is a uniform "slope bound." The ratio of output change to input change can never exceed $K$. Differentiable functions on $\mathbb{R}$ with bounded derivative are Lipschitz; the Lipschitz constant equals the sup of $|f'|$.

Intuition: Holder continuity with exponent $\alpha < 1$ allows the modulus of continuity to blow up like $h^{\alpha - 1}$ near zero — steeper than Lipschitz but tamer than arbitrary uniform continuity. It is the natural setting for many fractal and regularity estimates in analysis. The canonical example is $f(x) = \sqrt{x}$ on $[0, 1]$: Holder-$1/2$ but not Lipschitz.

Hierarchy: Lipschitz $\Rightarrow$ Holder-$\alpha$ (for any $\alpha \in (0, 1]$, on bounded domains) $\Rightarrow$ uniformly continuous $\Rightarrow$ continuous. All implications are strict. Holder continuity is most often used with $\alpha \in (0, 1)$; the borderline $\alpha = 1$ is Lipschitz, while "Holder with $\alpha > 1$" on a connected domain would force $f$ to be constant (the difference quotient is bounded by $K |x - x'|^{\alpha - 1} \to 0$, so $f' \equiv 0$).

Intuition: A homeomorphism is an isomorphism of topological spaces: it matches up points AND open sets. Homeomorphic spaces have the same topological properties — connectedness, compactness, Hausdorffness, etc. A donut and a coffee cup are the classical example of homeomorphic spaces that differ geometrically but not topologically.

Intuition: This is the most general form: "pull back every target neighbourhood to a source neighbourhood, ignoring the point $a$ itself." The formulation makes no reference to distances — only to open sets. In Hausdorff spaces this limit, when it exists, is unique by essentially the same argument as for sequences.

Intuition: The limit of $f$ at $a$ is the value that $f(x)$ approaches as $x$ approaches $a$, but without touching $a$ itself. The requirement that $a$ be an accumulation point ensures we can actually approach $a$ through $A$. The value of $f$ at $a$ (if defined) is irrelevant.

Intuition: The naive "plug one limit into the other" is correct almost always, but fails in a subtle edge case: if $g(x)$ happens to equal $b$ for $x$ near $a$ but $b$ is not in the domain of $f$ with value $L$, the limit of $f \circ g$ can see the "punctured-out" value of $f$ at $b$. The two clauses rule out that pathology — either $f$ behaves at $b$ (it is continuous there) or $g$ never hits $b$ near $a$.

Intuition: One-sided limits only "approach from one side" of $a$. They let us describe jump discontinuities (where $f(a^-) \neq f(a^+)$) and are the natural concept at boundary points of intervals. The two-sided limit $\lim_{x \to a} f(x)$ exists iff both one-sided limits exist and agree.

Intuition: "$x$ large enough" plays the role of "$x$ close enough to $a$" — only the source has changed. Equivalently, $\lim_{x \to \infty} f(x) = L$ iff for every sequence $x_n \to \infty$ in $A$, $f(x_n) \to L$.

Intuition: "$f(x)$ close to $L$" has been replaced by "$f(x)$ bigger than any threshold $M$" — only the target has changed. These are not limits in the strict metric sense (since $\pm\infty \notin \mathbb{R}$); they are limits in the extended real line $[-\infty, +\infty]$, which is a compact topological space whose neighbourhoods of $\pm\infty$ are the sets $(M, \infty]$ and $[-\infty, M)$.

Intuition: The oscillation $\omega_f(x_0)$ measures how much $f$ jumps near $x_0$. It is zero exactly when $f$ is continuous at $x_0$, so it serves as a quantitative failure-of-continuity.

Intuition: Continuity is a "for every $\varepsilon$" condition; quantifying over countably many $\varepsilon = 1/n$ expresses $C_f$ as a countable intersection. Each individual condition cuts out an open set, so the intersection is $G_\delta$.

The converse holds: for every $G_\delta$ set $S \subseteq \mathbb{R}$ there does exist a function $f : \mathbb{R} \to \mathbb{R}$ with $C_f = S$. Sketch: write $E = S^c = \bigcup_n F_n$ with each $F_n$ closed and (WLOG) nested $F_1 \subseteq F_2 \subseteq \cdots$. Set $A_n = F_n \cap \mathbb{Q}$ and $f_n = \mathbf{1}_{F_n} - \mathbf{1}_{A_n}$. The series $f = \sum_{n=1}^\infty \frac{1}{n!} f_n$ converges uniformly on $\mathbb{R}$ to a bounded function whose continuity set is exactly $S$.