First, note that the function is clearly strictly decreasing. Assume the supremum of the function is finite. In this case, for any ϵ>0, there exists δ such that for any x<δ, we have l−f(x)<2ϵ. Now, in the given inequality, let x,y tend to zero. We get f(f(f(l)))≤0, which is a contradiction. Next, set y=1 and let x tend to zero. The right side tends to infinity, so the expression inside the left side must tend to zero, which implies that the infimum of the function is zero.
We have so far obtained that the limit of the function as x→0 is infinity, and as x→∞ is zero. Now, fix x and let y tends to zero. We obtain that f(x)−f(x+y) must tend to zero, because otherwise the left side would be bounded while the right side would be unbounded. Therefore, the function is continuous. Now, set y=xk and let x tend to zero. We obtain
f3(k)=x→0limf(x)f(xk)
This implies that f3 is multiplicative, that is
f3(k′)f3(k)=x→0limf(xk′)f(xk)=g(k′k)
which implies
f3(kk′)f3(1)=f3(k)f3(k′)
Combined with the monotonicity and the limits of the function at zero and infinity, this implies f3(x)=xrc, for some real r.
Now, choose c such that f(c)<1 and set x=c. We will have
f(f(f(cy))+c2)<f(y)⟹f(f(cy))+c2>f(y)
which implies f(f(y))<Ay+B. On the other hand, if f(c)>1 and choose y sufficiently large, we will have
f(y)<f(f(cy))+c2⟹y<f(cy)+c2
which implies f(f(y))>A′y+B′. Therefore, f6(y)=xr2 falls between two linear functions, which implies f3(y)=xc. Now, let g(x)=f2(x) be a strictly increasing function such that g3(x)=x. Thus, we have f2(x)=x, which combined with f3(x)=xc implies f(x)=xc. It is easy to verify that f(x)=1/x is the only function that satisfies the statement of the problem. ■
Solution 2:
As in the first solution, we need the fact that f is strictly decreasing as well as limx→0+f(x)=+∞ and limx→+∞f(x)=0. Taking into account the monotonicity, the function would be differentiable almost everywhere, thus there exists a point z>0 such that f′(z) exists. Letting x=z and y→0+ yielding we get
0<f(z2/2)<x→z+limf(x2)=y→0+limyf(y)⋅yf(z)−f(z+y)=−f′(z)y→0+limyf(y),
which implies that limy→0+yf(y)=ℓ1 exists and is positive. Let's rewrite the hypothesis as
(f(f(xy))+x2)f(f(f(xy))+x2)=(xyf(f(xy))+yx)yf(y)(xf(x)−xf(x+y)).
Since limx→0+f(f(xy))+x2=0, letting x→0+ yielding
ℓ1=(x→0+limxyf(f(xy)))yf(y)ℓ1.
Since the inside limit is independent of y, we obtain at yf(y)=ℓ21, where ℓ2=limx→0+xf(f(x)) (which obviously must exist, precisely because of the above equation). Plugging f(y)=yℓ21 yields f(y)=y1.