Three Faces of Continuity

Continuity has three faces: the ε\varepsilon-δ\delta argument, the sequential test, and preimages of open sets. Placing f(x)=x2f(x) = x^2 next to a jump function shows the three faces saying the same thing.

∀ε>0 ∃δ>0: ∣x−x0∣<δ⇒∣f(x)−f(x0)∣<ε\forall\varepsilon>0\ \exists\delta>0:\ \lvert x-x_0\rvert<\delta \Rightarrow \lvert f(x)-f(x_0)\rvert<\varepsilon
xn→x0⇒f(xn)→f(x0)x_n \to x_0 \Rightarrow f(x_n) \to f(x_0)
f continuous  ⟺  f−1(V) open for every open Vf \text{ continuous} \iff f^{-1}(V) \text{ open for every open } V

Examples & Exercises

Prove completely that ε\varepsilon-δ\delta continuity of ff at x0x_0 implies sequential continuity: f(xn)→f(x0)f(x_n) \to f(x_0) for every sequence xn→x0x_n \to x_0.

∀ε>0 ∃δ>0: ∣x−x0∣<δ⇒∣f(x)−f(x0)∣<ε\forall\varepsilon>0\ \exists\delta>0:\ \lvert x-x_0\rvert<\delta \Rightarrow \lvert f(x)-f(x_0)\rvert<\varepsilon
xn→x0 ⟹ f(xn)→f(x0)x_n \to x_0 \ \Longrightarrow\ f(x_n) \to f(x_0)
f(x)=x2f(x) = x^2, ε=0.3\varepsilon = 0.3: δmax⁡≈0.54\delta_{\max} \approx 0.54 (=0.3=\sqrt{0.3}) keeps every xx inside the band — this δ\delta is the heart of the proof below.
Show proof
  1. Assume ff is ε\varepsilon-δ\delta continuous at x0x_0: for every ε>0\varepsilon > 0 there is δ>0\delta > 0 with ∣x−x0∣<δ⇒∣f(x)−f(x0)∣<ε|x - x_0| < \delta \Rightarrow |f(x) - f(x_0)| < \varepsilon.
  2. Take a sequence xn→x0x_n \to x_0 and ε>0\varepsilon > 0. Fetch the δ\delta for this ε\varepsilon from above; since xn→x0x_n \to x_0, there is NN with n>N⇒∣xn−x0∣<δn > N \Rightarrow |x_n - x_0| < \delta.
  3. For all n>Nn > N, ∣f(xn)−f(x0)∣<ε|f(x_n) - f(x_0)| < \varepsilon, so f(xn)→f(x0)f(x_n) \to f(x_0). Since ε>0\varepsilon > 0 was arbitrary, sequential continuity holds.

Prove by contrapositive: if ff is not ε\varepsilon-δ\delta continuous at x0x_0, one can build a sequence (xn)(x_n) with xn→x0x_n \to x_0 but f(xn)↛f(x0)f(x_n) \not\to f(x_0).

∃ε0>0 ∀δ>0 ∃x: ∣x−x0∣<δ and ∣f(x)−f(x0)∣≥ε0\exists\varepsilon_0>0\ \forall\delta>0\ \exists x:\ \lvert x-x_0\rvert<\delta \text{ and } \lvert f(x)-f(x_0)\rvert \ge \varepsilon_0
∣xn−x0∣<1n,∣f(xn)−f(x0)∣≥ε0\lvert x_n - x_0\rvert < \tfrac{1}{n}, \qquad \lvert f(x_n) - f(x_0)\rvert \ge \varepsilon_0
The jump function, ε=0.3\varepsilon = 0.3: xn=(−1)n/n→0x_n = (-1)^n/n \to 0, yet f(xn)f(x_n) bounces near 00 and 11 and never settles at f(0)=1f(0) = 1 — the proof below builds this sequence in general form.
Show proof
  1. Assume ff is not ε\varepsilon-δ\delta continuous at x0x_0. Negating the definition: some ε0>0\varepsilon_0 > 0 exists such that for every δ>0\delta > 0 there is xx with ∣x−x0∣<δ|x - x_0| < \delta and ∣f(x)−f(x0)∣≥ε0|f(x) - f(x_0)| \ge \varepsilon_0.
  2. For each natural nn, plug in δ=1/n\delta = 1/n and pick xnx_n: ∣xn−x0∣<1/n|x_n - x_0| < 1/n and ∣f(xn)−f(x0)∣≥ε0|f(x_n) - f(x_0)| \ge \varepsilon_0.
  3. By the squeeze theorem, ∣xn−x0∣<1/n→0|x_n - x_0| < 1/n \to 0, so xn→x0x_n \to x_0. But ∣f(xn)−f(x0)∣≥ε0|f(x_n) - f(x_0)| \ge \varepsilon_0 for every nn, so f(xn)↛f(x0)f(x_n) \not\to f(x_0) — if f(xn)→f(x0)f(x_n) \to f(x_0) there would have to be nn with ∣f(xn)−f(x0)∣<ε0/2|f(x_n) - f(x_0)| < \varepsilon_0/2, a contradiction. Hence sequential continuity fails.

Prove completely in both directions that ff is continuous on its whole domain if and only if the preimage f−1(V)f^{-1}(V) of every open set is open.

f continuous  ⟺  f−1(V) open for every open Vf \text{ continuous} \iff f^{-1}(V) \text{ open for every open } V
V=(f(x0)−ε, f(x0)+ε),U=f−1(V)∋x0 openV = (f(x_0)-\varepsilon,\ f(x_0)+\varepsilon), \qquad U = f^{-1}(V) \ni x_0 \text{ open}
The jump function, ε=0.3\varepsilon = 0.3: the preimage of V=(0.7,1.3)V = (0.7, 1.3) contains no interval around 00 (on the left only ∣x∣≈0.84|x| \approx 0.84 and beyond qualifies) — the preimage of an open set is not open.
Show proof
  1. (⇒\Rightarrow) Let ff be continuous and VV open. Take x0∈f−1(V)x_0 \in f^{-1}(V); then f(x0)∈Vf(x_0) \in V, and since VV is open there is ε>0\varepsilon > 0 with (f(x0)−ε,f(x0)+ε)⊂V(f(x_0)-\varepsilon, f(x_0)+\varepsilon) \subset V.
  2. Continuity at x0x_0 gives δ>0\delta > 0 with ∣x−x0∣<δ⇒∣f(x)−f(x0)∣<ε|x - x_0| < \delta \Rightarrow |f(x) - f(x_0)| < \varepsilon. Then every xx in (x0−δ,x0+δ)(x_0-\delta, x_0+\delta) satisfies f(x)∈(f(x0)−ε,f(x0)+ε)⊂Vf(x) \in (f(x_0)-\varepsilon, f(x_0)+\varepsilon) \subset V, so x∈f−1(V)x \in f^{-1}(V). Thus x0x_0 is an interior point of f−1(V)f^{-1}(V); x0x_0 was arbitrary, so f−1(V)f^{-1}(V) is open.
  3. (⇐\Leftarrow) Assume the preimage of every open set is open. For x0x_0 and ε>0\varepsilon > 0, V=(f(x0)−ε,f(x0)+ε)V = (f(x_0)-\varepsilon, f(x_0)+\varepsilon) is open, so U=f−1(V)U = f^{-1}(V) is open with x0∈Ux_0 \in U. Hence some δ>0\delta > 0 has (x0−δ,x0+δ)⊂U(x_0-\delta, x_0+\delta) \subset U, and ∣x−x0∣<δ|x - x_0| < \delta gives f(x)∈Vf(x) \in V, i.e. ∣f(x)−f(x0)∣<ε|f(x) - f(x_0)| < \varepsilon. Since x0x_0 was arbitrary, ff is continuous.

3. Continuity