Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

How can I find all functions $f:\mathbb{R}\to\mathbb{R}$ such that $f(1)=1$ and $$f\left(f(x)y+\frac{x}{y}\right)=xyf\left(x^2+y^2\right)$$ for all real numbers $x$ and $y$ with $y\neq0$?

PS. This is from USA MOSP (Mathematical Olympiad Summer Program) 2007, and is one of the most difficult problems I've seen ever. This problem has been posted a couple of times on MathLinks fora, but I couldn't find a complete solution for it... And the only solution is $f(x)=\frac{1}{x}$. Please try it, thanks!

share|cite|improve this question
If the only solution is $f(x)=\frac{1}{x}$ then the function should be defined on $\Bbb{R}^*$. – Beni Bogosel Jun 14 '11 at 9:42
Showing $f(0)=0$ is hard, but it's true. Now I think it's better. – Amir Hossein Jun 14 '11 at 10:24
Showing $f(0)=0$ is actually easy. Substitute $x=0$. Then the relation becomes $f(y f(0))=0$. If $f(0)$ were nonzero, then the argument $yf(0)$ could be any nonzero number and it would be true that $f(z)=0$ for all nonzero $z$ which contradicts $f(1)=1$. – Luboš Motl Jun 14 '11 at 11:13
Is there anything to be gained by changing to polar coordinates? That would simplify the LHS. – Simon Jun 16 '11 at 16:38
up vote 10 down vote accepted

Here is my approach. The equation $f(x)y+\frac{x}{y}=x^2+y^2$ for $x \neq 0$ has at least one real root, since is equivalent to a cubic equation. Denote that one root with $y_0$.

Substituting $y_0$ in the functional equation we get $f(x^2+y^2)=xy_0f(x^2+y^2)$. Notice that $x^2+y_0^2\neq 0$. If we can prove that $f(c)=0$ if and only if $c=0$ we are done, since then the relation above imples $y_0=\frac{1}{x}$ and therefore, from the relation $f(x)y_0+\frac{x}{y_0}=x^2+y_0^2$ we get $f(x)=\frac{1}{x}$.

Therefore, to prove that the only solution of the functional equation is the one mentioned above, we just need to prove the following:

$$f(c)=0 \Leftrightarrow c=0$$

I've made some small steps in proving this but the proof is not complete.

The relation $f(f(x)+x)=xf(x^2+1)$ allows us to prove that if $x \neq 0$ then $f(x)=0\Rightarrow f(x^2+1)=0$. This implies that there exists a sequence $K_n \to \infty$ such that $f(K_n)=0$.

Assume that $f$ is continuous.

Take now $x,y$ with $x^2+y^2=K_n$. The initial relation implies that $f(f(x)y+\frac{x}{y})=0,\ \forall x^2+y^2=K_n$. For $y \to 0+$ and $x>0$ we get that $f(x)y+\frac{x}{y} \to \infty$ (since $x \to \sqrt{K_n})$ This means that $f$ takes zero values in a neighborhood of $\infty$. Since $f(y+\frac{1}{y})=yf(1+y^2)$ implies $f(x)=-f(-x),\forall x$ with $|x| \geq 2$, we have the same thing in a neighborhood of $-\infty$.

Notice that for $y >1 $ we have $y^2+1>y+\frac{1}{y}$, and we can translate the "zeros" in the neighborhood of $\infty$ until we reach $2$, i.e. $f(x)=0$ on $(-\infty,-2]\cup [2,\infty)$.

I don't know where to go from here. This is not an answer, and I added an extra hypothesis towards the end, but I think this might lead to a result.

Maybe there are some more general solutions such as

$$ f(x)=\begin{cases} \frac{1}{x}, &x \in A \\ 0, & x \notin A \end{cases}$$, where $A$ is a set with some given properties.

share|cite|improve this answer
Very nice, Beni! Thanks for this nice approach. I'll try to show that $f(c)=0 \iff c=0$. – Amir Hossein Jun 15 '11 at 3:25
Nice, thinking :) – user9413 Jun 16 '11 at 15:35
Thank you. Too bad I cannot find the end of the proof... – Beni Bogosel Jun 16 '11 at 15:39

This is a partial answer but too long for comment.

First plug in $x = 1$ to get $$f(y + {1 \over y}) = yf(1 + y^2)$$ and $y = 1$ to get $$f(f(x) + x) = xf(x^2 + 1)$$ Observe that the RHS of both expressions has the same form and therefore $$f(y + {1 \over y}) = f(f(y) + y)$$

Assuming $f$ is injective we immediately obtain $f(y) = {1 \over y}$. I am not sure how hard it is to obtain injectivity of $f$. Perhaps as hard as solving the problem altogether, so this answer of mine might not help much.

share|cite|improve this answer
You can get the same result by substituting $x=\frac{1}{y}$, yielding $f(f(\frac{1}{y})y + \frac{1}{y^2}) = f(\frac{1}{y^2} + y^2)$. This yields that $f(\frac{1}{y})y = y^2$ if $f$ is 1-1. – Thomas Andrews Jun 14 '11 at 16:24
@Thomas: ah, that is indeed neater. – Marek Jun 14 '11 at 17:43

Let $K = \{x\in \mathbb R : f(x) = 0, x \ne 0 \}$ and suppose $x\in K$.
Beni Bogosel showed that under those conditions $x^2 + 1 \in K$ and there is a sequence $(x_n)_{n\in \mathbb N}\subseteq K$ such that $\lim_n x_n = +\infty$.
$f(x + 1/x) = xf(1 + x^2)$ implies that $\frac {x^2 + 1} x \in K$ too.
From the invariance of $xyf(x^2 + y^2)$ with respect to the transformation $x \to y, y\to x$ we deduce $$f(f(x)y + x/y) = f(f(y)x + y/x)$$ If $k_1, k_2\in K$, the previous equality becomes $$f(k_1/k_2) = f(k_2/k_1)$$
Applying the above equality to $x^2 + 1$ and $\frac {x^2 + 1} x$ we obtain $$f(1/x) = f(x) = 0$$ even $1/x$ belongs to $K$.
Now, we can conclude there exists a sequence $(y_n)_{n\in \mathbb N}\subseteq K$ such that $\lim_n y_n = 0$.
If $k\in K$ and $0 < k < 1$ we have $k^2 < k$ and $$f\left(k/\sqrt{k - k^2}\right) = k\sqrt{k - k^2}f(k) = 0$$ so $\sqrt{\frac k {1 - k}}\in K$ and the same is for $$\left(\sqrt{\frac k {1-k}}\right)^2 + 1 = \frac 1 {1 - k}$$ Applying the previous result to the sequence $(y_n)_{n\in \mathbb N}$ we can construct a third sequence $(z_n)_{n\in \mathbb N}\subseteq K$ such that $\lim_n z_n = 1$. Assuming $f$ is continuous we can write: $$0 = \lim_n f(z_n) = f\left(\lim_n z_n\right) = f(1) = 1$$ Therefore we must conclude $K = \emptyset$.
For $x, y \ne 0$ we have $$f(f(y)x + y/x) = yxf(y^2 + x^2) = xy f(x^2 + y^2) = f(f(x)y + x/y)$$ if $f(x) = f(y) = 0$ the previous equality becomes $$f(y/x) = f(x/y)$$

share|cite|improve this answer
+1 for the work. I would like to see a more detailed proof that if $k_1,k_2 \in K$ then $f(k_1/k_2)=f(k_2/k_1)$. If you could prove this more clearly, then the problem is finalized, with your arguments. – Beni Bogosel Jun 16 '11 at 15:42
I had reached the same conclusion (if f is continuous at 1, then there's only one solution). The case where we don't assume continuity is still avoiding me, though. You can construct a lot of points $x_0$ where $f(x_0) = 0$. I have been able to prove that if $f(x_0) = 0$, $x_0 \neq 0$, then $f(2x_0^2) = \frac{1}{x_0^2}$, so one way to reach a contradiction would be to show that $2K^2$ and $K$ intersect. Oh and by the way, if there is $x_0 \neq 0$ such that $f(x_0) = 0$, then $f(x) = -f(-x)$ for all $x \in \mathbb{R}$. – J. J. Jun 17 '11 at 17:32

If anyone is interested, here is a proof that $f$ vanishes only at $0$, copied from my post on AoPS.

Let $S=\{x\ne0: f(x)=0\}$, and suppose $S$ is nonempty for the sake of contradiction. (We assume $f(0)=0$ and the reduction posted earlier.)

For nonzero $x,y$, let $P(x,y)$ denote the assertion $f(f(x)y+x/y) = xyf(x^2+y^2) = f(f(y)x+y/x)$.

(1) $P(x,1)\implies f(f(x)+x) = xf(x^2+1) = f(x+1/x)$, so $s\in S\implies s^2+1,s+s^{-1}\in S$.

(2) For $s\in S$, $$ P(s,y)\implies f(s/y) = sy f(s^2+y^2) = f(f(y)s+y/s). $$ In particular, $P(s,-y)$ reveals that $f$ is odd. Also, if $t\in S$, $$ P(s,t)\implies f(s/t) = stf(s^2+t^2) = f(t/s), $$ so $P(s+s^{-1},s^2+1) \implies f(s^{-1}) = f(s) = 0\implies s^{-1}\in S$ (according to (1)) and $P(s,s)\implies s^2f(2s^2) = f(1) = 1\implies 2s^2\notin S$.

(3) If $s,t\in S$ with $t>s^2$, then $$ P(s,\sqrt{t-s^2}) \implies \frac{s}{\sqrt{t-s^2}}\in S\implies \frac{\sqrt{t-s^2}}{s}\in S. $$ Combined with (1), this gives $$ \left(\frac{\sqrt{t-s^2}}{s}\right)^2+1 = \frac{t}{s^2}\in S \implies \frac{t-s^2}{s} = \frac{s}{(s/\sqrt{t-s^2})^2}\in S $$ as well. Now WLOG fix $s\in(0,1)$ (if necessary, we can replace $s$ with one of $\pm s,\pm s^{-1}$, since $\pm1\notin S$). Then $t=s>s^2$ yields $\sqrt{s/(1-s)},\sqrt{(1-s)/s}\in S$ and $1-s = (s-s^2)/s \in S$ (from the first and second implications, respectively). But then (1) forces $(1-s)^2+1\in S$, so if we take $t=(1-s)^2+1>1>s^2$ instead, we get $$ 2\frac{1-s}{s} = \frac{(1-s)^2+1-s^2}{s} \in S $$ from the second implication. Finally, (twice) using the fact that $uv^2\in S$ whenever $u,v\in S$ and $uv^2\ne0,\pm1$ (*) twice, we first obtain $$ 2 = \left(2\frac{1-s}{s}\right)\left(\sqrt{\frac{s}{1-s}}\right)^2 \in S, $$ and then $2(s^{-1})^2>1$ must lie in $S$, contradicting the end of (2).

Edit (thanks to tenniskidperson3 for pointing out). (*) was originally incorrectly stated and missing proof. Here's a proof: By the second implication in (3), we know that if $u,v\in S$ and $uv^2>1$, then since $1/v\in S$, $u>(1/v)^2$ gives $u/(1/(1/v))^2 = uv^2$ in $S$. If $u,v\in S$ and $0<uv^2<1$ instead, then since $1/u,1/v\in S$, we have $(1/u)(1/v)^2>1$ so $(1/u)(1/v)^2\in S$ by the previous sentence, and then the inverse $uv^2$ must lie in $S$. If instead $u,v\in S$ and $uv^2<0$ but $uv^2\ne-1$, then negating $u$ shows from the previous two sentences that $uv^2\in S$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.