Let $\{f_n\}$ be a decreasing sequence of continuous functions with $f_n:[0,1]\rightarrow\mathbb{R}$, with the property that there exists an $M\in(0,1)$ such that $|f_n|\leq M$. Furthermore, $f_n\rightarrow f$ pointwise, where $f$ is also continuous. Prove that the convergence is actually uniform.

This is what I tried: Let $\epsilon>0$ be given. Then as $f$ is continuous on $[0,1]$ it is uniformly continuous. Let $\delta$ be such that $\delta$-close points map to $\frac{\epsilon}{3}$-close points. Partition $[0,1]$ into intervals, all whose length are less than $\delta$. Let $I_1,...,I_k$ be such intervals. Let $y_i$ be the midpoint of $I_i$. Let $N_i$ be such that $|f_n(y_i)-f(y_i)|<\epsilon/3$. Let $N$ be the max of all $N_i$.

Let $x\in [0,1]$ be arbitrary. Then $x\in I_k$ for some $k$. Hence, if $n\geq N$, then: $$|f_n(x)-f(x)|\leq |f_n(x)-f_n(y_k)|+|f_n(y_k)+f(y_k)|+|f(y_k)+f(x)|$$By choice of $N$, we have that the middle term is less that $\epsilon/3$, by continuity of $f$, the last summand is also less than $\epsilon/3$, so I was wondering how can I finish it?



Actually, the assumption that $|f_n|\leq M$ is redundant, since it is given that each $f_n$ is continuous (hence bounded on $[0, 1]$), and the sequence is decreasing (i.e. I presume you mean that for each $m > n$ and each $x\in [0, 1]$, $f_m(x)\leq f_n(x)$. The assumption that the upper bound (that is, $M$) is in $(0, 1)$ is also not needed. We'll only use the fact that each $f_n$ is continuous, the interval $[0,1]$ is compact, the sequence $\{f_n\}$ is decreasing, and $f_n\rightarrow f$ pointwise on $[0, 1]$.

Observe that since $\{f_n\}$ is decreasing in the sense assumed above, for each $m > n$ and $x\in[0,1]$, we have $f(x)\leq f_m(x)\leq f_n(x)$. Given $\epsilon > 0$ and $n\in\mathbb{N}$, set

$$ O_n := \{x\in[0,1]: f_n(x) - f(x) < \epsilon\}. $$

By continuity of $f_n$ and $f$ (which is assumed in the original problem), for each $n$, $O_n$ is open. Let $C_n = O_n^c$. Then $C_n$ is compact and, moreover, by pointwise convergence of $\{f_n\}$ to $f$ and the fact that $\{f_n\}$ is decreasing, the sequence $\{C_n\}_{n\in\mathbb{N}}$ forms a nested sequence of compact sets; that is, $C_{n+1}\subset C_n$ for each $n\geq 1$. Now, if for each $n$, $C_n\neq \emptyset$, then

$$ C:= \bigcap_{n\in\mathbb{N}}C_n $$

is nonempty, and for each $x\in C$, $f_n(x)$ does not converge to $f$. This is contradiction of pointwise convergence, hence for some $n_0$, $C_{n_0}$ must be empty. In other words, for some $n_0\in\mathbb{N}$,

$$ f_{n_0}(x) - f(x)\leq \epsilon $$

for each $x\in[0,1]$. By monotonicity of the sequence $\{f_n\}$ (in the sense presumed above), for each $m\geq n_0$, we have also

$$ f_m(x) - f(x) \leq \epsilon $$

for each $x\in[0,1]$. Uniform convergence is thus proved.

  • $\begingroup$ Very nice... Thank you :D $\endgroup$ – Daniel Montealegre Mar 29 '12 at 8:00
  • $\begingroup$ Is there a mistake in the way $O_n$ and $C_n$ were defined? I think they should be the other way around! $\endgroup$ – user123276 Jun 1 '14 at 13:11
  • $\begingroup$ @user123276: You're right, thank you! The sign in the definition of $O_n$ was incorrectly written as $ > \epsilon$ instead of $ < \epsilon$. I think it is fine now. $\endgroup$ – user2093 Jun 2 '14 at 6:44

As mentioned by Salech Alhasov, this is Dini's Theorem.

Here is a proof closer in spirit to the OP's approach:

The result will follow if we can show that the sequence with terms $g_n=f_n-f$ converges uniformly to 0. Note that $(g_n)$ is a decreasing sequence of continuous functions converging pointwise to 0

Fix $\epsilon>0$.

For each $t\in I=[0,1]$, choose $g_{m_t}$ so that $0\le g_{m_t}(t)<{\epsilon\over2}$. Since $g_{m_t}$ is continuous at $t$, there is a $\delta(t)>0$ so that $$0\le g_{m_t}(x)<\textstyle{\epsilon }\quad\text{ for all}\quad |x-t|\le \delta(t).$$ Now, since $[0,1]$ is compact, there is a finite set $\{t_1,\ldots, t_k\}\subset I$ such that $\bigl\{ B_{t_1}(\delta(t_1)),\ldots, B_{t_k}(\delta(t_k))\bigr\}$ covers $[0,1]$.

Let $m>\max\limits_{1\le i\le k}\{m_{t_i}\}$ and $x\in I$. Then $x\in B_{t_i}(\delta(t_i))$ for some $i$ and thus $\phantom{\epsilon\over2}$ $$ 0\le g_m(x)\le g_{m_{t_i}}(x)<\epsilon. $$

Thus $(g_n)$ converges uniformly to $0$ on $[0,1]$.

  • $\begingroup$ Thank you. No wonder the problem also said "Do not just quote a theorem". I did not know which theorem were they referring to. $\endgroup$ – Daniel Montealegre Mar 29 '12 at 19:01

It is Dini's theorem.


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.