Define three sequences $x_n, y_n, z_n$ for $n=1, 2, \dots, $ by $x_1 = 2$, $y_1 = 4$, $z_1 = \frac{6}{7}$ and the recursion $$ x_{n+1} = \frac{2x_n}{x_n^2-1}, \quad y_{n+1} = \frac{2y_n}{y_n^2-1}, \quad z_{n+1} = \frac{2z_n}{z_n^2-1} $$

Is it possible to have $x_n + y_n + z_n = 0$ for some $n$?

Looking at each sequence separately, they look quite messy and are not monotone and also don't seem to converge to a limit. I have tried an induction method, to start from $x_{n+1} + y_{n+1} + z_{n+1}$, expand out using the recursion, and try to show that it is non-zero provided that $x_n+y_n+z_n$ is non-zero, but I am not getting anywhere because it is too messy. Maybe it can be simplified to showing that $x_n + y_n > z_n$ for all $n$ or something along those lines which will be sufficient to prove the result.

Alternatively, I think it would be neater to find some invariant but I am really bad at finding invariants and I don't see any clue on how to start. The most obvious $x_n +y_n + z_n$ is not constant so that doesn't work.

Another thing maybe is to look at the sequence of numerators and denominators i.e. let $a_n/b_n = x_n$ where $a_n$ and $b_n$ are coprime and work from there but it also gets messy very quickly. I have $\frac{a_{n+1}}{b_{n+1}} = \frac{2a_nb_n}{a_n^2 - b_n^2} = \frac{\sqrt{a_n^2b_n^2}}{\frac{a_n^2-b_n^2}{2}}$ which doesn't seem to resemble anything.

How to approach this problem?

  • 1
    $\begingroup$ Can you share the source of the problem? $\endgroup$ Jan 4, 2020 at 8:24

1 Answer 1


Step 1: If $x_n = \tan(\phi)$ for some $\phi $ then $$ x_{n+1} = -\frac{2 \tan(\phi)}{1- \tan^2(\phi)} = - \tan(2 \phi) = \tan(-2\phi) \, . $$ Therefore, if we define $$ \alpha = \arctan(x_1), \, \beta = \arctan(y_1), \, \gamma = \arctan(z_1) \, , $$ then $$ x_n = \tan((-2)^{n-1} \alpha), \, y_n = \tan((-2)^{n-1} \beta), \, z_n = \tan((-2)^{n-1} \gamma) $$ for all $n$, as long as the sequences are defined, i.e. as long as no denominator in the sequence definitions becomes zero.

Step 2: Generally, $$ \tan (u) + \tan (v) + \tan (w) = \tan (u) \tan(v) \tan(w) $$ holds if and only if $u+v+w$ is an integral multiple of $\pi$, see for example Prove the identity if x+y+z=xyz on AoPS.

Step 3: The initial values $x_1 = 2$, $y_1 = 4$, $z_1 = 6/7$ satisfy $x_1 + y_1 + z_1 = x_1 y_1 z_1$. It follows that $\alpha + \beta + \gamma$ is an integral multiple of $\pi$. But then also $$ (-2)^{n-1} \alpha + (-2)^{n-1} \beta + (-2)^{n-1} \gamma = k \pi \text{ for some }k \in \Bbb Z $$ which in turn implies that $$ x_n + y_n + z_n = x_n y_n z_n $$ as long as the sequences are defined. It follows directly from the sequence definitions that the right-hand side is never zero, therefore $x_n + y_n + z_n$ can not be zero either.

Remark: One can also verify directly from the recursive definition that $$ x_{n+1} + y_{n+1} + z_{n+1} - x_{n+1} y_{n+1} z_{n+1} = \frac{2(1-x_n y_n - x_n z_n - y_n z_n)(x_n + y_n +z_n -x_ny_n z_n)}{(x_n^2-1)(y_n^2-1)(z_n^2-1)} $$ so that $$ x_n + y_n +z_n = x_ny_n z_n \implies x_{n+1} + y_{n+1} + z_{n+1} = x_{n+1} y_{n+1} z_{n+1} \, . $$

  • $\begingroup$ Ah okay so the idea here is to recognize the recurrence and use the trig substitution? Then the invariant $x_n + y_n + z_n - x_ny_nz_n = 0 \quad \forall n$ is less mysterious now. Thanks! $\endgroup$
    – eatfood
    Jan 4, 2020 at 11:14
  • $\begingroup$ $\alpha, \beta, \gamma$ are the angles of a triangle with sides $\sqrt{17}, \sqrt{20}, 3$. $\endgroup$ Jan 4, 2020 at 13:06

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.