6

If sequence $\{a_n\}$ is such that $a_1=2$ and $a_{n+1}=a_n+\frac{a_n}{a_n^2-1}$, show that $a_{2024}<64$.

First, I tried $a_{n+1}<a_n+\frac{a_n+1}{a_n^2-1}=a_n+\frac1{a_n-1}$. If $b_n=a_n-1$, then $b_{n+1}<b_n+\frac1{b_n}$. After squaring and summation, we get $$b_{n+1}^2<2n+b_1^2+\sum_{i=1}^n\frac1{b_i^2}.$$ We need to find a lower bound for $b_n$ ($a_n$). In fact, $a_{n+1}>a_n+\frac1{a_n}$, so squaring and summation gives $a_n\ge\sqrt{2n+2}$. Thus, $$b_{n+1}^2<2n+1+\sum_{i=1}^n\frac1{\left(\sqrt{2i+2}-1\right)^2}.$$ I checked that this bound is not enough.

Second, I tried $a_n=a_1+\sum\limits_{i=1}^{n-1}\frac{a_i}{a_i^2-1}$. Since $\frac x{x^2-1}$ is a decreasing function, we apply the above lower bound of $a_n$ again, but this is not enough as well.

youthdoo
  • 3,663
  • @Semiclassical: I get $a_{2024} \approx 63.7247$ and $a_{2025} \approx 63.7404$ with PARI/GP, but perhaps I made an off-by-one index error. – Martin R Mar 15 '25 at 18:40
  • @MartinR No, I'm the one with the off-by-one-error. – Semiclassical Mar 15 '25 at 19:38
  • 3
    (As an abstract duplicate. IE Michael's solution is similar to Did's) This question is similar to: Closed form for the sequence defined by $a_0=1$ and $a_{n+1} = a_n + a_n^{-1}$. If you believe it’s different, please [edit] the question, make it clear how it’s different and/or how the answers on that question are not helpful for your problem. – Calvin Lin Mar 15 '25 at 21:16
  • 2
    @CalvinLin: If the sequence $(b_n)$ is defined by $b_1=1$, $b_{n+1}=b_n+1/b_n$ then $a_n \le b_n+1$, which (numerically) gives $a_{2024} < 64.651629$, i.e. that is not good enough for this problem. The problems and solutions are similar, but I am reluctant to close it as a duplicate. – Martin R Mar 15 '25 at 21:40
  • 1
    @MartinR I don't consider it as a duplicate (and I agree the loose bound of $b_{n+1} < b_n + 1/b_n$ isn't sufficient), but it is potentially an abstract duplicate (which is very loosely defined). Partly I've been hammered by abstract duplicates too many times, so I now stay away from such. $\quad$ The approach shown there has a clear generalization, and those steps taken are what these 3 current solutions do (just with varying tightness). – Calvin Lin Mar 15 '25 at 21:44

4 Answers4

6

By your work $$a_{2024}^2=a_1^2+\sum_{k=1}^{2023}\left(\frac{2a_k^2}{a_k^2-1}+\frac{a_k^2}{(a_k^2-1)^2}\right)=4+\sum_{k=1}^{2023}\frac{a_k^2(2a_k^2-1)}{(a_k^2-1)^2}=$$ $$=4050+\sum_{k=1}^{2023}\left(\frac{a_k^2(2a_k^2-1)}{(a_k^2-1)^2}-2\right)=4050+\sum_{k=1}^{2023}\frac{3a_k^2-2}{(a_k^2-1)^2}\leq$$ $$\leq4050+\sum_{k=1}^{2023}\frac{6k+4}{(2k+1)^2}<4050+\int\limits_0^{2023}\frac{6x+4}{(2x+1)^2}dx<4096.$$

  • 1
    @ John Omielan It was typo. I fixed. Thank you! – Michael Rozenberg Mar 15 '25 at 19:38
  • 1
    You're welcome. Also, I agree with Martin that everything appears to now be correct. Note that, if my calculations are correct, your integral is $\approx 12.958$, so $a_n^2 \lt 4063$, i.e., $a_n$ is less than $\approx 63.74$, which is very close to the result in Martin's comment of $\approx 63.7247$. – John Omielan Mar 15 '25 at 19:41
  • 2
    In the step from the second line to the third line you use $a_n^2 \ge 2n+2$, which OP already proved. But for the convenience of future readers of this answer I would suggest to mention that explicitly here again. – Martin R Mar 15 '25 at 19:45
  • Just one more thing. Note that I wasn't pinged with your comment to me due to the space between the @ and my first name. Instead, I was still looking at your answer at the time, which is how I knew you had made the change, and had commented to me. – John Omielan Mar 15 '25 at 19:51
  • @Martin R It's trivial: $a_{n+1}=a_n+\frac{1}{a_n-\frac{1}{a_n}}>a_n+\frac{1}{a_n}.$ Thus, by telescoping summation again $a_n^2>a_1^2+2(n-1)=2n+2.$ – Michael Rozenberg Mar 15 '25 at 19:52
  • 1
    I know that, and OP proved it already. That was just a suggestion to make your answer better understandable for future readers. You are free to ignore it. – Martin R Mar 15 '25 at 19:55
  • How was the bound ($<16$) on the value of the final integral deduced? – Semiclassical Mar 15 '25 at 19:56
  • @Semiclassical $\int\frac{6x+4}{(2x+1)^2}dx=\int\frac{3(2x+1)+1}{(2x+1)^2}dx...$ Can you end it now? – Michael Rozenberg Mar 15 '25 at 19:58
4

$b_n = a_n^2$ satisfies $b_1 = 4$ and $$ b_{n+1} = b_n + \frac{b_n(2b_n-1)}{(b_n-1)^2} = b_n + 2 + \frac{3}{b_n-1} + \frac{1}{(b_n-1)^2} \, . $$ You already proved that $b_n \ge 2n+2 > 2n+1$, therefore is $$ b_{n+1} < b_n + 2 + \frac{3}{2n} + \frac{1}{4 n^2} \, . $$ It follows that $$ \begin{align} b_n &< 2n+2 + \frac 32 \sum_{k=1}^{n-1} \frac 1k + \frac 14 \sum_{k=1}^{n-1}\frac{1}{k^2} \\ &< 2n+2 + \frac 32 \bigl(1 + \ln(n)\bigr) + \frac 14 \frac{\pi^2}{6} \\ &< 2n + \ln(n) + 4 \, . \end{align} $$ In particular is $$ b_{2024} < 4052 + \frac 32 \ln(2024) \approx 4063.41925 < 4096 $$ and therefore $a_{2024} < 64$.

Martin R
  • 128,226
0

$b_n:=a_n^2$ satisfies $b_1=4,\quad b_{n+1}=\frac{b_n^3}{(b_n-1)^2}$ hence $c_n:=b_n-1$ satisfies$$c_1=3,\quad c_{n+1}=\frac{(c_n+1)^3}{c_n^2}-1=c_n+2+\frac{3c_n+1}{c_n^2}>c_n+2.$$ Therefore,$$c_n>3+2(n-1)=2n+1,$$ hence $$\begin{align}c_{2024}&<4049+\sum_{k=1}^{2023}\frac{6k+4}{(2k+1)^2} \\&<4095\\&=64^2-1.\end{align}$$

Anne Bauval
  • 49,005
0

Consider the function $f(x) = x+ \frac{x}{x^2-1}$. Our sequence is $$a_n = f^n(2)$$ for $n\ge 1$. We have $f'(x) = \frac{x^2(x^2-3)}{(x^2-1)^2}$, so $f$ is strictly increasing on $[\sqrt{3}, \infty)$.

Consider also the function $g(t) = \sqrt{2 t + \sqrt{t}}$. We want to show that $f^n(2)< g(n)$ for $n$ large enough. It is enough to show that

  1. $f^{N}(2) < g(N)$ for some low $N$
  2. $f(g(t))< g(t+1)$

The first $N$ for which the above holds is $N=112$, since $f^{112}(2) =15.315\ldots$, while $g(112)=15.316\ldots$.

Now the inequality $2$ holds for $t\ge 8$, as one can check.

Now by induction one sees that $f^{n}(2) < g(n)$ for $n\ge 112$. In particular $f^{2024}(2) < g(2024)=63.97\ldots < 64$.

Note: We have $a_n \simeq \sqrt{2n}$. Indeed, for the sequence $b_n = \frac{1}{a_n}$ we have the recurrence $b_{n+1} = b_n(1-b_n^2)$, very similar to $b_{n+1} = \sin b_n$, and these kind of recurrences appear elsewhere on the site.

orangeskid
  • 56,630