๐Ÿงฎ Brain Teaser

Dini's Theorem: When Pointwise Becomes Uniform


Problem

We know that pointwise convergence of continuous functions does not in general imply uniform convergence. (Classic counterexample: fn(x)=xnf_n(x) = x^n on [0,1][0,1].)

But now suppose the convergence is monotone. Prove:

Let fn:[0,1]โ†’Rf_n : [0,1] \to \mathbb{R} be continuous, and suppose fnโ†˜ff_n \searrow f pointwise (i.e. f1โ‰ฅf2โ‰ฅโ‹ฏf_1 \geq f_2 \geq \cdots and fn(x)โ†’f(x)f_n(x) \to f(x) for each xx). If ff is also continuous, then fnโ†’ff_n \to f uniformly.


Why It's Interesting

The counterexample xnโ†’0x^n \to 0 shows that continuity + pointwise convergence is not enough. Dini's theorem says that one extra assumption โ€” monotonicity โ€” tips the balance. The proof is a slick application of compactness that makes you appreciate why compactness is so powerful. It is also a cautionary reminder: drop any one of the three conditions (compact domain / monotone / continuous limit) and the conclusion fails.


Answer: view

Uniform convergencePointwise convergenceCompactnessDini

Answer: Dini's Theorem

fnโ†’fย uniformlyย onย [0,1]\boxed{f_n \to f \text{ uniformly on } [0,1]}


Intuition First

Define gn=fnโˆ’fโ‰ฅ0g_n = f_n - f \geq 0. We need to show supโกxโˆˆ[0,1]gn(x)โ†’0\sup_{x \in [0,1]} g_n(x) \to 0.

The key picture: fix any ฮต>0\varepsilon > 0. For each point xx, there is some N(x)N(x) such that gN(x)(x)<ฮตg_{N(x)}(x) < \varepsilon (by pointwise convergence). So the open set Un={x:gn(x)<ฮต}U_n = \{x : g_n(x) < \varepsilon\} expands as nn grows (monotonicity means gnโ‰ฅgn+1g_n \geq g_{n+1}, so UnโŠ‚Un+1U_n \subset U_{n+1}), and every point eventually falls inside one of these sets.

The sets {Un}\{U_n\} form an open cover of the compact set [0,1][0,1]. By compactness, finitely many suffice โ€” so one single UNU_N already covers everything. That means gN(x)<ฮตg_N(x) < \varepsilon for all xx simultaneously. Done.

Compactness is doing exactly one thing here: turning "every point is eventually fine" into "there is a single NN that works for all points at once."


Why Each Hypothesis Is Necessary

Before the proof, here's why you cannot drop any condition:

| Drop this | Counterexample | |-----------|---------------| | Compact domain | fn(x)=1nx+1f_n(x) = \frac{1}{nx+1} on (0,1)(0,1) โ€” monotone, fnโ†’0f_n \to 0, but not uniformly | | Monotone | fn(x)=xn(1โˆ’xn)f_n(x) = x^n(1-x^n) on [0,1][0,1] โ€” pointwise โ†’0\to 0, not monotone, not uniform | | Continuous limit | fn(x)=xnf_n(x) = x^n on [0,1][0,1] โ€” limit is 1{1}\mathbf{1}_{\{1\}}, discontinuous, not uniform |


Formal Proof

Let gn=fnโˆ’fg_n = f_n - f. Each gng_n is continuous (difference of continuous functions), gnโ‰ฅ0g_n \geq 0, and gnโ†˜0g_n \searrow 0 pointwise on [0,1][0,1].

Goal: โˆฅgnโˆฅโˆžโ†’0\|g_n\|_\infty \to 0.

Fix ฮต>0\varepsilon > 0. Define: Un={xโˆˆ[0,1]:gn(x)<ฮต}U_n = \{x \in [0,1] : g_n(x) < \varepsilon\}

Each UnU_n is open in [0,1][0,1] (preimage of (โˆ’โˆž,ฮต)(-\infty, \varepsilon) under the continuous function gng_n).

Since gnโ†˜0g_n \searrow 0 pointwise, for each xx there exists N(x)N(x) with gN(x)(x)<ฮตg_{N(x)}(x) < \varepsilon, so xโˆˆUN(x)x \in U_{N(x)}. Thus: [0,1]=โ‹ƒn=1โˆžUn[0,1] = \bigcup_{n=1}^{\infty} U_n

Since gng_n is decreasing, UnโŠ‚Un+1U_n \subset U_{n+1} (if gn(x)<ฮตg_n(x) < \varepsilon then gn+1(x)โ‰คgn(x)<ฮตg_{n+1}(x) \leq g_n(x) < \varepsilon).

By compactness of [0,1][0,1], the open cover {Un}\{U_n\} has a finite subcover. Since the UnU_n are nested, the finite subcover is just {UN}\{U_N\} for the largest index NN appearing. So: [0,1]=UN={x:gN(x)<ฮต}[0,1] = U_N = \{x : g_N(x) < \varepsilon\}

For all nโ‰ฅNn \geq N and all xโˆˆ[0,1]x \in [0,1]: 0โ‰คgn(x)โ‰คgN(x)<ฮต0 \leq g_n(x) \leq g_N(x) < \varepsilon

Therefore โˆฅgnโˆฅโˆž<ฮต\|g_n\|_\infty < \varepsilon for all nโ‰ฅNn \geq N, which is exactly uniform convergence. โ– \blacksquare