Uniform Convergence of Functions

Convergence with one shared deadline for every input, which is what keeps continuity and integrals safe in the limit.

Uniform convergence squeezes every function inside one epsilon tube around the limit, so no point lags behind

Why Does This Exist?

Pointwise convergence lets each input converge on its own schedule, and that freedom breaks things: continuity can vanish and integrals can jump. Training needs the opposite, where a surrogate approaches its target evenly across the whole domain.

Uniform convergence is that shared schedule. One $N$ works for every $x$ at once. Under it, continuous approximants have a continuous limit, and integrals of the limit equal limits of the integrals.

This page defines the sup-norm test and the M-test that proves it. The weaker mode is in sequences and series of functions, and series practice is in Taylor series and approximation.

The one idea you need first

The supremum norm $\|g\|_\infty = \sup_x |g(x)|$ measures the worst gap. Uniform convergence means this worst gap goes to zero, not just each pointwise gap.

Think of It Like This

A marching band crossing a line

Each marcher is one input $x$ . Pointwise convergence lets marchers cross the finish line whenever they arrive. Uniform convergence ties them wrist to wrist: the whole line crosses together, because the slowest marcher sets the pace.

The tube in the diagram is the finish corridor of width $2\epsilon$ . The sup gap is the distance from the slowest marcher to the line. Where the analogy stops: functions need this for every $\epsilon$ , not one race; the band must tighten repeatedly.

How It Actually Works

$f_n \to f$ uniformly when for every $\epsilon > 0$ there is $N(\epsilon)$ , independent of $x$ , with $n > N$ implying $|f_n(x) - f(x)| < \epsilon$ for all $x$ . Equivalently $\|f_n - f\|_\infty \to 0$ .

Consequences: a uniform limit of continuous functions is continuous; on a bounded interval $\int \lim f_n = \lim \int f_n$ ; with uniformly convergent derivatives plus one pointwise value, $\frac{d}{dx}\lim f_n = \lim f_n'$ .

The Weierstrass M-test proves uniform convergence of $\sum g_n$ : find numbers $M_n$ with $|g_n(x)| \le M_n$ for all $x$ and $\sum M_n < \infty$ . Then the series converges uniformly.

Worked example

Take $f_n(x) = x/n$ on $[0,1]$ . Pointwise gaps are $|x|/n \le 1/n$ . So $\|f_n - 0\|_\infty = 1/n$ : values $1, 0.5, 0.1, 0.01$ for $n = 1, 2, 10, 100$ . Given $\epsilon = 0.05$ , $N = 20$ works for every $x$ at once. Uniform to $0$ . Contrast $x^n$ on $[0,1]$ , where $\|f_n - f\|_\infty = 1$ for all $n$ because near $x = 1$ the gap stays large: not uniform.

Watch Out For

Claiming uniformity from plots at few points

Symptom: checking $x = 0, 0.5, 1.0$ and declaring uniform convergence while a spike between samples persists. Fix: bound the sup over all $x$ analytically or on a dense grid plus a Lipschitz argument between grid points.

Differentiating a uniformly convergent series termwise

Symptom: uniform $f_n$ with wildly oscillating $f_n'$ gives a wrong derivative after swapping. Fix: uniform convergence of the derivatives is what licenses the swap, not of the functions alone.

The Quick Version

Uniform convergence uses one $N$ for all inputs, measured by the sup norm going to zero.
It preserves continuity and lets limits pass through integrals on bounded intervals.
The M-test proves it by dominating each term with a convergent number $M_n$ .
$x/n$ converges uniformly to $0$ with sup gap $1/n$ ; $x^n$ on $[0,1]$ does not.
Check the worst input, not a few convenient ones.