Olympiad Maths Prep

Track / Stage 8 / 94 of 180 #1794 of 2000

Problem 1794

IMO Shortlist mid-range; USAMO P2/P5
Algebra Difficulty 8.3 Prove it Balkan Mathematical Olympiad Shortlisted Problems · Balkan Mathematical Olympiad

Find all functions f:RRf: \mathbb{R} \to \mathbb{R} such that
f(x+yf(x))+y=xy+f(x+y) f(x + y f(x)) + y = x y + f(x + y)
for all real numbers x,yx, y.

This one wants a proof. Work it on paper, read the official solution, then mark yourself honestly — the ladder only means something if the record is true.

Official solutions — 2

Solution 1

Let P(x,y)P(x, y) denote the given relation. If there is an aRa \in \mathbb{R} such that f(a)=0f(a) = 0, then P(a,y)P(a, y) gives that y=ay+f(a+y)y = a y + f(a + y), and so ff must be linear. Then we can easily check and get that the only linear solutions are f(x)=xf(x) = x and f(x)=2xf(x) = 2 - x (xRx \in \mathbb{R}).

Now suppose that f(x)0f(x) \neq 0 for all real numbers xx. From P(xy,y)P(x - y, y) we get that:
f(xy+yf(xy))=y2+y(x1)+f(x). f(x - y + y f(x - y)) = -y^2 + y(x - 1) + f(x).
Since f(t)0f(t) \neq 0 for all real numbers tt, it follows that y2+y(x1)+f(x)0-y^2 + y(x - 1) + f(x) \neq 0 for all real numbers x,yx, y, and so, its discriminant (as a polynomial in yy) must be negative. That is, (x1)2+4f(x)<0(x - 1)^2 + 4 f(x) < 0, which gives us
f(x)<(x1)240 f(x) < -\frac{(x-1)^2}{4} \leq 0
for all real numbers xx. Since (x+1)20(x + 1)^2 \geq 0 implies that (x1)24x-\frac{(x-1)^2}{4} \leq x, we see that
f(x)<(x1)24x f(x) < -\frac{(x-1)^2}{4} \leq x
for all real numbers xx. Now from P(x,y)P(x, y) for y>0y > 0 and xRx \in \mathbb{R}, we get that
xyy+f(x+y)=f(x+yf(x))<x+yf(x)<xy(x1)24 x y - y + f(x + y) = f(x + y f(x)) < x + y f(x) < x - y \frac{(x-1)^2}{4}
and so
f(x+y)<x+yy(x+(x1)24)=x+yy(x+1)24. f(x + y) < x + y - y(x + \frac{(x-1)^2}{4}) = x + y - y \frac{(x+1)^2}{4}.
Setting x=yx = -y above, we get that:
f(0)<y(y+1)24. f(0) < -y \frac{(-y + 1)^2}{4}.
for all positive real numbers yy. Letting y+y \to +\infty above, we reach a contradiction. Hence, the only solutions in this functional equation are f(x)=xf(x) = x and f(x)=2xf(x) = 2 - x.

Solution 2

Let P(x,y)P(x, y) denote the given relation. Similarly to the first solution, if a root exists (f(a)=0f(a) = 0 for any aa), we get that the function is linear and that the two solutions are f(x)=xf(x) = x and f(x)=2xf(x) = 2 - x. Assertion P(x,cx)P(x, c - x) gives us the following relation:
f(x+(cx)f(x))=(cx)(x1)+f(c)=x2+(c+1)x+(f(c)c) f(x + (c - x) f(x)) = (c - x)(x - 1) + f(c) = -x^2 + (c + 1)x + (f(c) - c)
The right hand side of the expression is a quadratic equation in xx with the discriminant Δ=Δ(c)=(c+1)2+4(f(c)c)=(c1)2+4f(c)\Delta = \Delta(c) = (c + 1)^2 + 4(f(c) - c) = (c - 1)^2 + 4 f(c). Therefore, if there exists a cc such that (c1)2+4f(c)0(c - 1)^2 + 4 f(c) \geq 0, the quadratic equation has a real solution which implies the existence of a root, in which case we are done.
If f(1)=0f(1) = 0, then we found a root and are done. If f(1)=1f(1) = 1, then by taking c=1c = 1 we obtain that Δ(1)=4\Delta(1) = 4, implying the existence of a root. We now check the case when f(1)=1f(1) = -1. From the assertion P(1x,x)P(1 - x, x), we obtain:
f(1x+xf(1x))=x21 f(1 - x + x f(1 - x)) = -x^2 - 1
Plugging in x=1x = 1, in the above assertion, we obtain that f(f(0))=2f(f(0)) = -2. Now plugging in x=1f(0)x = 1 - f(0) in the above assertion we get that f(f(0)+(1f(0))f(f(0)))=(1f(0))21f(f(0) + (1 - f(0)) f(f(0))) = -(1 - f(0))^2 - 1, simplifying and utilizing f(f(0))=2f(f(0)) = -2 we obtain f(3f(0)2)=f(0)2+2f(0)2f(3 f(0) - 2) = -f(0)^2 + 2 f(0) - 2. Note that if f(0)0f(0) \ge 0, we have that Δ(0)=1+4f(0)>0\Delta(0) = 1 + 4 f(0) > 0, implying the existence of a root, so assume that f(0)<0f(0) < 0. Now using c=3f(0)2c = 3 f(0) - 2 for our discriminant value, we obtain Δ(3f(0)2)=(3f(0)3)2+4f(3f(0)2)=9(f(0)1)2+4(f(0)2+2f(0)2)=5f(0)210f(0)+1>0\Delta(3 f(0) - 2) = (3 f(0) - 3)^2 + 4 f(3 f(0) - 2) = 9(f(0) - 1)^2 + 4(-f(0)^2 + 2 f(0) - 2) = 5 f(0)^2 - 10 f(0) + 1 > 0, implying the existence of a root, and resolving the case when f(1)=1f(1) = -1.
Now assume that f(1){0,1,1}f(1) \notin \{0, 1, -1\}. From P(1,y)P(1, y), we obtain the relation that f(1+yf(1))=f(1+y)f(1 + y f(1)) = f(1 + y). As f(1)0f(1) \ne 0, we can inductively show that f(1+yf(1)k)=f(1+y)f(1 + y f(1)^k) = f(1 + y) for all kZk \in \mathbb{Z}. Since f(1){1,1}f(1) \notin \{1, -1\}, there exists an unbounded sequence ana_n such that f(an)f(a_n) is constant. Namely, one can take an=1+f(1)2na_n = 1 + f(1)^{2n} if f(1)>1|f(1)| > 1, and an=1+f(1)2na_n = 1 + f(1)^{-2n} if f(1)<1|f(1)| < 1, both times it holds that f(an)=f(2)f(a_n) = f(2). The value of the discriminant along this sequence is Δ(an)=(an1)2+4f(an)=(an1)2+4f(2)\Delta(a_n) = (a_n - 1)^2 + 4 f(a_n) = (a_n - 1)^2 + 4 f(2), and since ana_n is unbounded this there exists nn where the value of the discriminant is positive, yielding our root. This finishes the problem.

Source: MathNet, licensed CC-BY-4.0. Statement and solution reproduced as published; topic, difficulty and ordering added by this site.