2

I would require some help with a series problem my math teacher gave me. I am first year of university. The problem is as follows:

Let there be $x_1 \gt 0$ and the following recurrent relation :

$$x_{n+1} = x_n + \frac{n^2}{x_1+x_2+...+x_n}$$

Study the nature of convergence for $\frac{x_n}{n}$.

Thanks in advance!

Jack D'Aurizio
  • 353,855
Nicklarca
  • 21
  • 1

1 Answers1

3

The sequence $\{x_n\}_{n\geq 1}$ is clearly positive and increasing. It follows that

$$ x_{n+1}=x_n+\frac{n^2}{x_1+\ldots+x_n}\geq x_n+\frac{n}{x_n} \tag{A}$$ $$ x_{n+1}^2 \geq x_n^2 + 2n,\qquad x_n\geq (n-1) \tag{B} $$ $$ x_{n+1}\leq x_n +\frac{n^2}{0+1+\ldots+(n-1)},\qquad x_n\leq x_0+2n\tag{C} $$ so $\frac{x_n}{n}$ is bounded. Assuming $\left\{\frac{x_n}{n}\right\}_{n\geq 1}$ is convergent, by the Cesàro-Stolz theorem we have

$$ L=\lim_{n\to +\infty}\frac{x_n}{n}=\lim_{n\to +\infty}\frac{n^2}{x_1+\ldots+x_n}=\lim_{n\to +\infty}\frac{2n+1}{x_{n+1}}=\frac{2}{L}\tag{D}$$ hence $L=\color{red}{\sqrt{2}}$.

Now, let us prove that the limit exists. By $(B)$ and $(C)$ $x_n$ is bounded between $n+O(1)$ and $2n+O(1)$, so $a_n = \frac{x_n}{n}$ is bounded between $1-\varepsilon$ and $2+\varepsilon$. We have

$$ n(a_{n+1}-a_n) = \frac{2n}{(n+1)}\cdot\frac{n(n+1)/2}{a_1+ 2a_2+\ldots + na_n}-a_{n+1} \tag{E}$$ where $\frac{n(n+1)/2}{a_1+ 2a_2+\ldots + na_n}$ is the reciprocal of a weighted mean of $a_1,\ldots,a_n$. It follows that $\{a_n\}_{n\geq 1}$ is "almost constant", in the sense that $a_{n+1}-a_n = O\left(\frac{1}{n}\right)$. Let us assume that $\{a_n\}_{n\geq 1}$ has a subsequence $\{a_{n_k}\}_{k\geq 1}$ convergent to $c\neq \sqrt{2}$. By $(E)$ $$ a_{n_k+1}-a_{n_k} = \frac{1}{n_k}\left(\frac{2}{c}-c+o(1)\right)\tag{F} $$ so the weighted mean of $a_1,a_2,\ldots,a_{n_k},a_{n_k+1}$ is also close to $c$ and $(F)$ continues to hold by replacing $n_k$ with $n_k+1$. On the other hand, by summing multiple instances of $(F)$ we violate the boundedness of $\{a_n\}_{n\geq 1}$, since the harmonic series is divergent. It follows that $\sqrt{2}$ is the only limit point of $\{a_n\}_{n\geq 1}$.

Jack D'Aurizio
  • 353,855
  • Thank you for the fast reply! I'll take until tomorrow to study your proof a little more in depth, if there's anything unclear I'll make sure to ask – Nicklarca Oct 30 '17 at 19:01
  • It might help the OP to show how the conclusions of A, B and C are derived. For example: $x_{n+1} \ge x_n \implies x_1+...+x_n \le nx_n$ and $x_{n+1}^2 \geq x_n^2 + 2n \implies x_{n+1}^2 - x_n^2 \ge 2n \implies \sum (x_{n+1}^2 - x_n^2) \ge \sum 2n \implies ...$. – marty cohen Oct 30 '17 at 20:14
  • @martycohen: on the other hand, I think that to leave something to complete, on purpose, in not a bad thing. This is probably some form of "advanced homework" and I don't want to spoil all the fun ;) – Jack D'Aurizio Oct 30 '17 at 20:16
  • I would have just a couple of questions :
    1. How do you go from $x^2_n+_1$ $\ge$ $x_n+\frac{n}{x_n}$ to $x^2_n+_1$ = $x^2_n+ 2n$;

    2)Could you also show me the steps to reach $x_n≥(n−1)$ from the formula above. I assume that for point (C) it is the same procedure for $x_n$ $\le$ $x_0+2n$.

    3)And one more thing, why when calculating the limit, $lim$ $\frac{x_n}{n}$ = $lim$ $\frac{n^2}{x_1+x_2+...+x_n}$.

    – Nicklarca Oct 31 '17 at 15:24
  • @Nicklarca: $x_{n+1}\geq x_{n}+\frac{n}{x_n}$ leads to $x_{n+1}^2 \geq x_n^2 +2n$ by simply squaring both sides, then the lower bound for $x_n$ follows by setting $y_n=x_n^2$ and solving $y_{n+1}=y_n+2n$. The multiple equality among those limits follows from the Stolz-Cesàro theorem, as mentioned. – Jack D'Aurizio Oct 31 '17 at 15:27
  • When squaring each side, wouldn't you have a $\frac{n^2}{x_n^2}$ left over on the right side? – Nicklarca Oct 31 '17 at 15:34
  • That is $\geq 0$, so? – Jack D'Aurizio Oct 31 '17 at 15:35
  • Is it like a negligible quantity or why exactly do we leave it out instead of writing : $x_{n+1}^2 \geq x_n^2 +2n + \frac{n^2}{x_n^2}$ – Nicklarca Oct 31 '17 at 15:40
  • @Nicklarca: we neglect it because $y_{n+1}=y_n+2n$ is way easier to solve than $y_{n+1}=y_n+2n+\frac{n^2}{y_n}$. – Jack D'Aurizio Oct 31 '17 at 15:41
  • 1
    As in the current question https://math.stackexchange.com/q/4488666/42969, proving the existence of the limit of $(x_n/n)$ is the crucial part, and that seems to be missing here. – Martin R Jul 08 '22 at 16:44