Maths Olympiad Prep

Library / /87 of 106

Algebra Difficulty 8.8 Shortlist Prove it IMO

Let R>0\mathbb{R}_{>0} be the set of positive real numbers. Find all functions f:R>0R>0f: \mathbb{R}_{>0} \rightarrow \mathbb{R}_{>0} such that, for every xR>0x \in \mathbb{R}_{>0}, there exists a unique yR>0y \in \mathbb{R}_{>0} satisfying
xf(y)+yf(x)2. x f(y)+y f(x) \leqslant 2 .

Solutions — 4

Solution 1

First we prove that the function f(x)=1/xf(x)=1 / x satisfies the condition of the problem statement. The AM-GM inequality gives
xy+yx2 \frac{x}{y}+\frac{y}{x} \geqslant 2
for every x,y>0x, y>0, with equality if and only if x=yx=y. This means that, for every x>0x>0, there exists a unique y>0y>0 such that
xy+yx2 \frac{x}{y}+\frac{y}{x} \leqslant 2
namely y=xy=x.

Let now f:R>0R>0f: \mathbb{R}_{>0} \rightarrow \mathbb{R}_{>0} be a function that satisfies the condition of the problem statement. We say that a pair of positive real numbers (x,y)(x, y) is good if xf(y)+yf(x)2x f(y)+y f(x) \leqslant 2. Observe that if (x,y)(x, y) is good, then so is (y,x)(y, x).

Lemma 1.0. If (x,y)(x, y) is good, then x=yx=y.

Proof. Assume that there exist positive real numbers xyx \neq y such that (x,y)(x, y) is good. The uniqueness assumption says that yy is the unique positive real number such that (x,y)(x, y) is good. In particular, (x,x)(x, x) is not a good pair. This means that
xf(x)+xf(x)>2 x f(x)+x f(x)>2
and thus xf(x)>1x f(x)>1. Similarly, (y,x)(y, x) is a good pair, so (y,y)(y, y) is not a good pair, which implies yf(y)>1y f(y)>1. We apply the AM-GM inequality to obtain
xf(y)+yf(x)2xf(y)yf(x)=2xf(x)yf(y)>2. x f(y)+y f(x) \geqslant 2 \sqrt{x f(y) \cdot y f(x)}=2 \sqrt{x f(x) \cdot y f(y)}>2 .
This is a contradiction, since (x,y)(x, y) is a good pair.

By assumption, for any x>0x>0, there always exists a good pair containing xx, however Lemma 1 implies that the only good pair that can contain xx is (x,x)(x, x), so
xf(x)1f(x)1x, x f(x) \leqslant 1 \quad \Longleftrightarrow \quad f(x) \leqslant \frac{1}{x},
for every x>0x>0.

Solution 1.1. We give an alternative way to prove that f(x)=1/xf(x)=1 / x assuming f(x)1/xf(x) \leqslant 1 / x for every x>0x>0.
Indeed, if f(x)<1/xf(x)<1 / x then for every a>0a>0 with f(x)<1/a<1/xf(x)<1 / a<1 / x (and there are at least two of them), we have
af(x)+xf(a)<1+xa<2. a f(x)+x f(a)<1+\frac{x}{a}<2 .
Hence (x,a)(x, a) is a good pair for every such aa, a contradiction. We conclude that f(x)=1/xf(x)=1 / x.

Solution 1.2. We can also conclude from Lemma 1 and f(x)1/xf(x) \leqslant 1 / x as follows.
Lemma 2. The function ff is decreasing.
Proof. Let y>x>0y>x>0. Lemma 1 says that (x,y)(x, y) is not a good pair, but (y,y)(y, y) is. Hence
xf(y)+yf(x)>22yf(y)>yf(y)+xf(y), x f(y)+y f(x)>2 \geqslant 2 y f(y)>y f(y)+x f(y),
where we used y>xy>x (and f(y)>0f(y)>0 ) in the last inequality. This implies that f(x)>f(y)f(x)>f(y), showing that ff is decreasing.

We now prove that f(x)=1/xf(x)=1 / x for all xx. Fix a value of xx and note that for y>xy>x we must have xf(x)+yf(x)>xf(y)+yf(x)>2x f(x)+y f(x)>x f(y)+y f(x)>2 (using that ff is decreasing for the first step), hence f(x)>2x+yf(x)>\frac{2}{x+y}. The last inequality is true for every y>x>0y>x>0. If we fix xx and look for the supremum of the expression 2x+y\frac{2}{x+y} over all y>xy>x, we get
f(x)2x+x=1x. f(x) \geqslant \frac{2}{x+x}=\frac{1}{x} .
Since we already know that f(x)1/xf(x) \leqslant 1 / x, we conclude that f(x)=1/xf(x)=1 / x.

Solution 2

As in the first solution, we note that f(x)=1/xf(x)=1 / x is a solution, and we set out to prove that it is the only one. We write g(x)g(x) for the unique positive real number such that (x,g(x))(x, g(x)) is a good pair. In this solution, we prove Lemma 2 without assuming Lemma 1.

Lemma 2. The function ff is decreasing.
Proof. Consider x<yx<y. It holds that yf(g(y))+g(y)f(y)2y f(g(y))+g(y) f(y) \leqslant 2. Moreover, because yy is the only positive real number such that (g(y),y)(g(y), y) is a good pair and xyx \neq y, we have xf(g(y))+g(y)f(x)>2x f(g(y))+g(y) f(x)>2. Combining these two inequalities yields
xf(g(y))+g(y)f(x)>2yf(g(y))+g(y)f(y), x f(g(y))+g(y) f(x)>2 \geqslant y f(g(y))+g(y) f(y),
or f(g(y))(xy)>g(y)(f(y)f(x))f(g(y))(x-y)>g(y)(f(y)-f(x)). Because g(y)g(y) and f(g(y))f(g(y)) are both positive while xyx-y is negative, it follows that f(y)<f(x)f(y)<f(x), showing that ff is decreasing.

We now prove Lemma 1 using Lemma 2. Suppose that xyx \neq y but xf(y)+yf(x)2x f(y)+y f(x) \leqslant 2. As in the first solution, we get xf(x)+xf(x)>2x f(x)+x f(x)>2 and yf(y)+yf(y)>2y f(y)+y f(y)>2, which implies xf(x)+yf(y)>2x f(x)+y f(y)>2. Now
xf(x)+yf(y)>2xf(y)+yf(x) x f(x)+y f(y)>2 \geqslant x f(y)+y f(x)
implies (xy)(f(x)f(y))>0(x-y)(f(x)-f(y))>0, which contradicts the fact that ff is decreasing. So y=xy=x is the unique yy such that (x,y)(x, y) is a good pair, and in particular we have f(x)1/xf(x) \leqslant 1 / x.

We can now conclude the proof as in any of the Solutions 1.x.

Solution 3

As in the other solutions we verify that the function f(x)=1/xf(x)=1 / x is a solution. We first want to prove the following lemma:

Lemma 3. For all xR>0x \in \mathbb{R}_{>0} we actually have xf(g(x))+g(x)f(x)=2x f(g(x))+g(x) f(x)=2 (that is: the inequality is actually an equality).

Proof. We proceed by contradiction: Assume there exists some number x>0x>0 such that for y=g(x)y=g(x) we have xf(y)+yf(x)<2x f(y)+y f(x)<2. Then for any 0<ϵ<2xf(y)yf(x)2f(x)0<\epsilon<\frac{2-x f(y)-y f(x)}{2 f(x)} we have, by uniqueness of yy, that xf(y+ϵ)+(y+ϵ)f(x)>2x f(y+\epsilon)+(y+\epsilon) f(x)>2. Therefore
f(y+ϵ)>2(y+ϵ)f(x)x=2yf(x)ϵf(x)x>2yf(x)2xf(y)yf(x)2x=2xf(y)yf(x)2x+f(y)>f(y) \begin{align*} f(y+\epsilon) & >\frac{2-(y+\epsilon) f(x)}{x}=\frac{2-y f(x)-\epsilon f(x)}{x} \\ & >\frac{2-y f(x)-\frac{2-x f(y)-y f(x)}{2}}{x} \\ & =\frac{2-x f(y)-y f(x)}{2 x}+f(y)>f(y) \tag{1} \end{align*}
Furthermore, for every such ϵ\epsilon we have g(y+ϵ)f(y+ϵ)+(y+ϵ)f(g(y+ϵ))2g(y+\epsilon) f(y+\epsilon)+(y+\epsilon) f(g(y+\epsilon)) \leqslant 2 and g(y+ϵ)f(y)+yf(g(y+ϵ))>2g(y+\epsilon) f(y)+y f(g(y+\epsilon))>2 (since yy+ϵ=g(g(y+ϵ)))y \neq y+\epsilon=g(g(y+\epsilon))). This gives us the two inequalities
f(g(y+ϵ))2g(y+ϵ)f(y+ϵ)y+ϵ and f(g(y+ϵ))>2g(y+ϵ)f(y)y. f(g(y+\epsilon)) \leqslant \frac{2-g(y+\epsilon) f(y+\epsilon)}{y+\epsilon} \quad \text{ and } \quad f(g(y+\epsilon))>\frac{2-g(y+\epsilon) f(y)}{y} .
Combining these two inequalities and rearranging the terms leads to the inequality
2ϵ<g(y+ϵ)[(y+ϵ)f(y)yf(y+ϵ)]. 2 \epsilon<g(y+\epsilon)[(y+\epsilon) f(y)-y f(y+\epsilon)] .
Moreover combining with the inequality (1) we obtain
2ϵ<g(y+ϵ)[(y+ϵ)f(y)y2xf(y)yf(x)2x+f(y)]=g(y+ϵ)[ϵf(y)y2xf(y)yf(x)2x]2 \epsilon<g(y+\epsilon)\left[(y+\epsilon) f(y)-y \frac{2-x f(y)-y f(x)}{2 x}+f(y)\right]=g(y+\epsilon)\left[\epsilon f(y)-y \frac{2-x f(y)-y f(x)}{2 x}\right].
We now reach the desired contradiction, since for ϵ\epsilon sufficiently small we have that the left hand side is positive while the right hand side is negative.

With this lemma it then follows that for all x,yR>0x, y \in \mathbb{R}_{>0} we have
xf(y)+yf(x)2 x f(y)+y f(x) \geqslant 2
since for y=g(x)y=g(x) we have equality and by uniqueness for yg(x)y \neq g(x) the inequality is strict.

In particular for every xR>0x \in \mathbb{R}_{>0} and for y=xy=x we have 2xf(x)22 x f(x) \geqslant 2, or equivalently f(x)1/xf(x) \geqslant 1 / x for all xR>0x \in \mathbb{R}_{>0}. With this inequality we obtain for all xR>0x \in \mathbb{R}_{>0}
2xf(g(x))+g(x)f(x)xg(x)+g(x)x2 2 \geqslant x f(g(x))+g(x) f(x) \geqslant \frac{x}{g(x)}+\frac{g(x)}{x} \geqslant 2
where the first inequality comes from the problem statement. Consequently each of these inequalities must actually be an equality, and in particular we obtain f(x)=1/xf(x)=1 / x for all xR>0x \in \mathbb{R}_{>0}.

Solution 4

Again, let us prove that f(x)=1/xf(x)=1 / x is the only solution. Let again g(x)g(x) be the unique positive real number such that (x,g(x))(x, g(x)) is a good pair.

Lemma 4. The function ff is strictly convex.
Proof. Consider the function qs(x)=f(x)+sxq_{s}(x)=f(x)+s x for some real number ss. If ff is not strictly convex, then there exist u<vu<v and t(0,1)t \in(0,1) such that
f(tu+(1t)v)tf(u)+(1t)f(v). f(t u+(1-t) v) \geqslant t f(u)+(1-t) f(v) .
Hence
qs(tu+(1t)v)tf(u)+(1t)f(v)+s(tu+(1t)v)=tqs(u)+(1t)qs(v) \begin{aligned} q_{s}(t u+(1-t) v) & \geqslant t f(u)+(1-t) f(v)+s(t u+(1-t) v) \\ & =t q_{s}(u)+(1-t) q_{s}(v) \end{aligned}
Let w=tu+(1t)vw=t u+(1-t) v and consider the case s=f(g(w))/g(w)s=f(g(w)) / g(w). For that particular choice of ss, the function qs(x)q_{s}(x) has a unique minimum at x=wx=w. However, since qs(w)tqs(u)+(1t)qs(v)q_{s}(w) \geqslant t q_{s}(u)+(1-t) q_{s}(v), it must hold qs(u)qs(w)q_{s}(u) \leqslant q_{s}(w) or qs(v)qs(w)q_{s}(v) \leqslant q_{s}(w), a contradiction.

Lemma 5. The function ff is continuous.
Proof. Since ff is strictly convex and defined on an open interval, it is also continuous.

As in Solution 1, we can prove that f(x)1/xf(x) \leqslant 1 / x. If f(x)<1/xf(x)<1 / x, then we consider the function h(y)=xf(y)+yf(x)h(y)=x f(y)+y f(x) which is continuous. Since h(x)<2h(x)<2, there exist at least two distinct zxz \neq x such that h(z)<2h(z)<2 giving that (x,z)(x, z) is good pair for both values of zz, a contradiction. We conclude that f(x)=1/xf(x)=1 / x as desired.

Want a route through all this instead of an archive? The track puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

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