I have tried to do this exercise. Do you think my solution is ok? Is it possible to get more information about the convergence? Is there a better way to do it?

Let $ f:[0,1]\mapsto[0,1] $ be a continous function such that $ f(x)<x \ \forall x \in (0,1] $ and the right limit of the derivative in 0 is 1/2. Fixed a point $a_0 \in [0,1]$ consider the succession $a_{n+1}:=f(a_n)$. Study the series: $\sum_{n=0}^\infty a_n$.

Thank you everyone for the kind help!


PS I'm doing some exercises for a very important test which is about a lot of different stuff I might not even have seen in a course yet. So every kind comment, suggestion about somehow similar exercises or theorems/results/whatever that you think might help me out is more than welcome.


As requested by Samprity, this is my answer. This is a little bit long as I try to spell out as many details as possible.

Claim 1: $\quad f(0) = 0$.

  • Assume the contrary. If $f(0) > 0$, the the fact $f$ is continuous implies for some $\epsilon > 0$, $f(x) > \frac{f(0)}{2}$ for $x \in [0,\epsilon)$. If one pick a $y$ such that $0 < y < \min(\epsilon,\frac{f(0)}{2})$, we will find $f(y) > \frac{f(0)}{2} > y$. This contradicts with the given condition $f(x) < x$ for $x > 0$.

Claim 2: $\quad a_n \to 0$ as $n \to \infty.$

  • By Claim 1, $f(0) = 0$. Combine these with $f(x) < x$ over $(0,1]$, we find $f(x) \le x$ over $[0,1]$. This in turn implies the sequence $a_n$ determined by the recurrence relation $a_{n+1} = f(a_n)$ is monotonic decreasing.

  • Since $a_n$ is also bounded below by $0$, the sequence $a_n$ converges. Let us call the limit $a$ and it is clear $a \ge 0$.

  • Since $f(x)$ is a continuous function, we have $$f(a) = f(\lim_{n\to\infty}a_n) = \lim_{n\to\infty} f(a_n) = \lim_{n\to\infty} a_{n+1} = a$$
  • Since $f(x) < x$ for $x > 0$, $a$ cannot be positive and hence it must be $0$.

Please note that in above claims, we have not used the fact the right derivative of $f$ at $0$ exists and equal to $\frac12$. By definition of right derivative, for any $\epsilon > 0$, there exists a $\delta(\epsilon) > 0$ such that:

$$\left|\frac{f(x) - f(0)}{x - 0} - \frac12\right| < \epsilon \quad\text{ whenever }\quad x \in (0,\delta(\epsilon))$$

Take $\epsilon$ as $\frac14$ and notice $f(0) = 0$, this implies:

$$f(x) \le \frac34 x \quad\text{ whenever }\quad x \in [ 0, \delta(\frac14) )$$

By Claim 2, $a_n \to 0$ as $n \to \infty$. This means there is an integer $N$ such that

$$a_n < \delta(\frac14) \quad\text{ whenever }\quad n \ge N$$

For such $n$, we have $$ a_{n+1} = f(a_n) \le \frac34 a_n \quad\implies\quad a_{n} \le \left(\frac34\right)^{n-N} a_N$$

As a consequence, for any integer $M > N$, the partial sums of the original series we care is bounded from above by a number independent of $M$.

$$ \sum_{k=0}^{M} a_k = \sum_{k=0}^{N-1} a_k + \sum_{k=N}^{M} a_k \le \sum_{k=0}^{N-1} a_k + a_N \sum_{k=N}^{M} \left(\frac34\right)^{k-N} \le \sum_{k=0}^{N-1} a_k + 4a_N $$

Being a series of non-negative numbers, this implies $\displaystyle \sum_{k=0}^{\infty} a_k$ converges.


For right continuity (we have the right derivative) we have that $f(0)=0$. Thus the case $a_0=0$ trivially converges. Otherwise we have that the succession is strictly monotone (thus it has a limit).

There exists a neighbourhood of $0$, $V=(0,a)$ such that, $\forall h \in V$ we have $\vert \frac{f(h)-f(0)}{h} \vert =\vert \frac{f(h)}{h} \vert \leq \frac{1}{2}+\epsilon = M$.

Now let $ a_0=h\in (0,1]$. If there is a V, as above, such that $h \in V$ and $M < \frac{1}{h}$ the series converges using the comparison test with the geometric series, because: $\vert a_1 \vert = \vert f(h) \vert \leq M h \Rightarrow \vert a_n \vert \leq (M h)^n$ (with $\vert M h \vert < 1$).

If no such V exists then $M\geq h$ and the series doesn't converge because $\vert a_n \vert \geq (M h)^n \geq 1$

  • $\begingroup$ Since the right limit of the derivative at $0$ is $1/2$, there is some neighborhood $[0,\epsilon)$ in which $f(x) < (2/3)x$ (the $2/3$ is arbitrary; anything larger than $1/2$ can be used). That estimate should allow you to prove that the series will converge. $\endgroup$ – Carl Mummert Sep 9 '13 at 13:40

I am getting the followings for your given problem without using the right hand differentiability at $x = 0$, at not much required for getting convergence of the series $\sum a_n$.

As $f :[0,1] \rightarrow [0,1]$ is a continuous function all the value of $a_n$ will be in $[0,1]$. So $\{a_n\}$ is a bounded positive term sequence.

Now $a_{n+1} = f(a_n) < a_n$ as $f(x) < x$. So we can conclude that $\{a_n\}$ is a monotone sequence. A monotone bounded sequence is always convergent. You can prove it and so $\{a_n\}$ is convergent, though it is not your question which is about the convergence of $\sum a_n$.

As the sequence $\{a_n\}$ is monotone we shall say $a_n < a_0$ and as all the terms are positive we shall say $[a_n]^\frac{1}{n} < [a_0]^\frac{1}{n}$. Let $a_0 < 1$ then $\lim_{n \rightarrow \infty} a_n < 1$ because of monotonicity of $\{a_n\}$. Then by root text we can say $\sum a_n$ is convergent.

  • $\begingroup$ In this case, root test doesn't work because $\lim_{n\to\infty} a_0^{\frac{1}{n}} = 1$. Another thing is by this argument, the series corresponds to $a_n = \frac{1}{n+2}$ ( generated by $f(x) = x/(x+1)$ with $a_0 = \frac12$ ) converges. This is clearly not true. $\endgroup$ – achille hui Sep 9 '13 at 14:17
  • $\begingroup$ @achillehui Thank you for instructions. What will be correct argument? $\endgroup$ – Dutta Sep 10 '13 at 2:50
  • $\begingroup$ The correct argument consists of two parts. 1) show $a_n$ converges to 0 ($a_n$ is strictly monotonic decreasing + bounded from below + $f$ continuous is enough to prove this) 2) Since $a_n \to 0$, aside from finitely many terms, you can apply the argument in Carl Mummert's comment to bound $a_n$ from above by a geometric series. By comparison test, you are done. $\endgroup$ – achille hui Sep 10 '13 at 4:54
  • $\begingroup$ I have understood. Please write it down as an answer. $\endgroup$ – Dutta Sep 10 '13 at 6:33

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.