IMO 2017 · Problem 2

Let \(\mathbb{R}\) be the set of real numbers. Determine all functions \(f:\mathbb{R}\to\mathbb{R}\) such that, for all real numbers \(x\) and \(y\), \[ f(f(x)f(y))+f(x+y)=f(xy). \]

Official text of the problem. Official problem paper (PDF)

Answer

The solutions are \[ \boxed{f(x)\equiv 0,\qquad f(x)=x-1,\qquad f(x)=1-x.} \]

Solution

Denote the given equation by \(P(x,y)\), and put \(c=f(0)\). From \(P(x,0)\), \[ f(cf(x))=c-f(x). \tag{1} \] If \(c=0\), this immediately gives \(f(x)=0\) for every \(x\). Henceforth assume \(c\ne0\).

First, \(f\) has no nonzero period. Indeed, if \(p\ne0\) and \(f(t+p)=f(t)\) for all \(t\), then \(f(p)=c\). Comparing \(P(p,y)\) with \(P(0,y)\) gives \[ f(py)=c \] for every real \(y\). Thus \(f\) is constant, and the original equation forces that constant to be zero, contradicting \(c\ne0\).

We next prove that \(c^2=1\). Set \(a=c^2>0\). Equation (1) at \(x=0\) gives \(f(a)=0\), and \(P(a,1)\) gives \(f(a+1)=-c\). Applying (1) successively at \(x=a+1\) and \(x=-a\), we obtain \[ f(-a)=2c,\qquad f(2a)=-c. \] Consequently, \(P(a,a)\) gives \(f(a^2)=0\).

Since \(a\) and \(a^2\) are zeros of \(f\), the original equation yields \[ f(at)=f(t+a)+c,\qquad f(a^2t)=f(t+a^2)+c. \tag{2} \] Applying the first identity twice also gives \[ f(a^2t)=f(at+a)+c=f(t+a+1)+2c. \] Therefore, with \(d=a^2-a-1\), \[ f(t+d)=f(t)+c \qquad(t\in\mathbb R). \tag{3} \] In particular, \(d\ne0\). Combining (2) and (3), \[ f(a(t+d)) =f(t+d+a)+c =f(t+a)+2c =f(at)+c. \] Because \(a\ne0\), this says \(f(t+ad)=f(t)+c\) for every real \(t\). Comparing with (3), we find that \((a-1)d\) is a period of \(f\). There are no nonzero periods, and \(d\ne0\), so \(a=1\). Thus \(c=\pm1\).

Replacing a solution by its negative again gives a solution. We may therefore assume \(f(0)=1\), restoring the sign at the end. We now have \(f(1)=0\), and \(P(x,0)\), \(P(x,1)\) give \[ f(f(x))=1-f(x),\qquad f(x+1)=f(x)-1. \tag{4} \] Applying the first identity at \(f(x)\) gives \[ f(1-f(x))=f(x). \] The second identity then implies \(f(-f(x))=f(x)+1\). Since \(f(2)=-1\), equation \(P(x,2)\) yields \[ f(2x)=f(-f(x))+f(x+2)=2f(x)-1. \]

Define \[ h(x)=1-f(x). \] The identities just established imply \[ \begin{gathered} h(0)=0,\qquad h(1)=1,\qquad h(x+1)=h(x)+1,\\ h(2x)=2h(x),\qquad h(h(x))=h(x). \end{gathered} \tag{5} \] Substituting \(f=1-h\) into the original equation and using the translation identity in (5), we obtain \[ h\bigl(h(x)h(y)-h(x)-h(y)\bigr)+h(x+y)=h(xy). \tag{6} \]

We will show that \(h\) is additive. For fixed \(x,y\), write \[ A=h(x)h(y),\qquad B=h(x)+h(y). \] Applying (6) at \((h(x),h(y))\), and using \(h(h(t))=h(t)\), gives \[ h(A-B)+h(B)=h(A). \] Applying (6) at \((x+1,y+1)\), and using the translation identity, gives \[ h(xy+x+y)=h(A)+h(x+y). \] Together with (6), these equations imply \[ h(xy+x+y)-h(xy)=h(h(x)+h(y)). \tag{7} \]

Apply (7) at \((2x,2y)\) and use \(h(2t)=2h(t)\). After dividing by \(2\), we obtain \[ h(2xy+x+y)-2h(xy)=h(h(x)+h(y)). \] Comparing this with (7), \[ h(2xy+x+y)=h(xy+x+y)+h(xy). \tag{8} \]

Let \(t\in\mathbb R\) and \(v\le0\). There exist real \(x,y\) satisfying \[ xy=v,\qquad x+y=t-v, \] because the quadratic with these roots has discriminant \[ (t-v)^2-4v\ge0. \] Equation (8) consequently gives \[ h(t+v)=h(t)+h(v)\qquad(v\le0). \tag{9} \] For \(v>0\), applying (9) with increment \(-v\) first gives \(h(-v)=-h(v)\), and then \[ h(t)=h(t+v)+h(-v). \] Thus (9) holds for positive \(v\) as well, proving that \(h\) is additive.

Using additivity and \(h(h(x))=h(x)\), equation (6) simplifies to \[ h(xy)=h(h(x)h(y)). \tag{10} \] If \(h(u)=0\) for some \(u\ne0\), then (10) with \(x=u\), \(y=1/u\) gives \[ 1=h(1)=h(0)=0, \] a contradiction. Hence the only zero of \(h\) is \(0\). But additivity and (5) give \[ h(x-h(x))=h(x)-h(h(x))=0. \] Therefore \(h(x)=x\) for every real \(x\), so the normalized solution is \(f(x)=1-x\). Restoring the possible sign gives \(f(x)=x-1\) as well.

Finally, the zero function plainly satisfies the equation. For \(\varepsilon\in\{-1,1\}\), the function \(f(x)=\varepsilon(x-1)\) satisfies \[ \begin{aligned} f(f(x)f(y))+f(x+y) &=\varepsilon\bigl((x-1)(y-1)-1+x+y-1\bigr)\\ &=\varepsilon(xy-1)=f(xy). \end{aligned} \] Thus all three listed functions are solutions.

The solution was checked against the official IMO shortlist solutions.