Let $E ⊂ \mathbb{R}$ be a compact (i.e., closed bounded) set of real numbers. Suppose $\{f_n\}$ is a sequence of real-valued continuous functions which converges pointwise on $E$ to a function $f$ that is also continuous on $E.$ Suppose further that the sequence $\{f_n\}$ is monotonic: $f_{n+1}(x) ≤ f_n(x)$ for all $x ∈ E$ and all $n = 1,2,....$

(a) Prove that $f_n(x) → f(x)$ uniformly on $E$ as $n → ∞.$

(b) Show by example that the hypothesis of compactness is essential.

I need to show that $||f_n - f||_{\infty}$ tends to $0$ as $n$ tends to $\infty$ but I am not really sure how to do this. Advice would be awesome. Thanks.


This is known as Dini's theorem. Since $f_n-f$ is also continuous, we may assume $f_n\searrow 0$. Given $\varepsilon >0$, consider the sets $O_n=f_n^{-1}(-\infty,\varepsilon)$. Since $f_n$ is continuous, each one is open. Prove that

$(\rm i)$ $E=\bigcup O_n$

$(\rm ii)$ $O_n\subseteq O_{n+1}$

Since $E$ is compact, you will find $N$ such that $E\subseteq O_N$. But if $x\in O_N$, $|f_n(x)|<\varepsilon$ for $n\geqslant N$ (why?), which shows that $f_n\to 0$ uniformly.

The hypotheses that $E$ is compact, the $f_n$ are monotone and the limit is cotinuous are all necessary and independent. Can you find counterexamples in each case, that is, a sequences satisfying two of the three premises but not the third, such that the theorem fails?

Solutions $(1)$ Take $E=[0,1]$, $f_n(x)=x^n$. Then we have monotonicity, compactness, but the limit is not continuous and the theorem fails. $(2)$ Take $\chi_{[n,n+1])}(x-n)+\chi_{(n,\infty)}$ over $E=\Bbb R$. Then we have monotonicty, continuity, but $\Bbb R$ is not compact. $(3)$ Take $E=[0,1]$, $f_n(x)=(1-|nx-1|)\chi_{[0,2n^{-1}]}$. Then each $f_n$ is continuous, the limit is continuous, $E$ is comapct but the sequence is not monotone, and the theorem fails.


Let $g_n(x)=f_n(x)-f(x)$ and note that $0\le g_{n+1}(x)\le g_n(x)$ for all $x\in E$. Let $x_n\in E$ be such that $$g_n(x_n)=\|g_n\|_\infty,$$

As $E$ is compact, there is a subsequence $x_{n_k}$ such that $x_{n_k}\to x$ in $E$. Take $\epsilon>0$ and choose $N$ such that $|g_n(x)|<\epsilon$ for $n>N$. If $n_k>n$, we have that $$g_{n_k}(x_{n_k})<g_n(x_{n_k}),$$

therefore, by taking $n>N$ and making $k\to \infty$, we conclude that $$\lim_{k\to \infty} g_{n_k}(x_{n_k})<\lim_{k\to \infty}g_n(x_{n_k})=g_n(x)<\epsilon.$$

In particular, $g_{n_k}(x_{n_k})\to 0$.

To conclude, choose $K>0$ such that $g_{n_k}(x_{n_k})<\epsilon$ for $k>K$. Once $g_n$ is monotone, we have that $g_n(x_n)\le g_{n_K}(x_{n_K})$ for all $n>K$, therefore, $g_n(x_n)<\epsilon$ for $n>K$.

  • $\begingroup$ This is a lovely solution. $\endgroup$
    – Pedro Tamaroff
    Jul 15 '14 at 21:09
  • $\begingroup$ (It might be good to prove why we can assume that $x_n$ converges to something, and not that a subsequence converges to something, i.e. why can we be done by working with a subsequence of the $g_n$. This is not immediately evident, in my opinion.) $\endgroup$
    – Pedro Tamaroff
    Jul 15 '14 at 21:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.