A sequence of functions fn: D → ℝ converges pointwise to f if, at every fixed x, the number sequence fn(x) → f(x). It converges uniformly if the worst-case gap over the whole domain vanishes:
Pointwise: ∀x ∀ε>0 ∃N(x, ε): n≥N ⟹ |fₙ(x)−f(x)| < ε
Uniform: ∀ε>0 ∃N(ε): n≥N ⟹ |fₙ(x)−f(x)| < ε for ALL x at once
Equivalently: uniform ⟺ sup_x |fₙ(x) − f(x)| → 0
The surface is a "waterfall" of the whole family: x runs across, n runs into the depth, height is fn(x), and colour is the error |fn(x) − f(x)| (blue = small, red = large). The translucent band hugs the limit function f(x) at height ±ε; the highlighted curve is the current slice n, with green/red instanced markers showing where that slice sits inside or outside the band.
- xⁿ on [0,1] — each fn is continuous, but the pointwise limit is 0 on [0,1) and jumps to 1 at x=1: discontinuous. Since a uniform limit of continuous functions must be continuous, this sequence cannot converge uniformly — and indeed sup|fn−f| stays exactly 1 for every n, no matter how large.
- sin(nx)/n on [0,2π] — |fn(x)| ≤ 1/n everywhere, so the sup-norm itself shrinks to 0. One N works for every x simultaneously: uniform convergence to f = 0.
- n·x·(1−x)ⁿ on [0,1] — a "travelling bump" that slides toward x = 0 as n grows. At every fixed x the value still → 0 (pointwise), and the limit function 0 is perfectly continuous — yet the bump's peak height settles near 1/e ≈ 0.368 forever, so the sup-norm never shrinks: not uniform, even though continuity alone didn't rule it out.
This is the central subtlety of real analysis courses: pointwise limits can lose properties (continuity, integrability, differentiability) that every term in the sequence had; uniform convergence is exactly the extra condition that preserves them.