Maths Olympiad Prep

Library / /39 of 42

Algebra Difficulty 7.3 National olympiad, round 2 Prove it Ireland

A function f:Q+Q+f : \mathbb{Q}_+ \to \mathbb{Q}_+, where Q+\mathbb{Q}_+ denotes the (strictly) positive rational numbers, satisfies
f(f(x)+f(y))=xyx+y f(f(x) + f(y)) = \frac{xy}{x+y}
for any x,yQ+x, y \in \mathbb{Q}_+. Given f(1)=2023f(1) = 2023, find f(2023)f(2023).

Solution

We first show that ff is injective. Suppose that f(y)=f(z)f(y) = f(z). Then we have:
xyx+y=f(f(x)+f(y))=f(f(x)+f(z))=xzx+z \frac{xy}{x+y} = f(f(x) + f(y)) = f(f(x) + f(z)) = \frac{xz}{x+z}
Taking reciprocals of the first and last expression implies y=zy = z, as required for injectivity.

Note that the right hand side of the functional equation can be written as
xyx+y=11x+1y \frac{xy}{x+y} = \frac{1}{\frac{1}{x} + \frac{1}{y}}
If x=knax = \frac{k}{n-a} and y=kn+ay = \frac{k}{n+a} with a±na \ne \pm n, the expression 1x+1y=2nk\frac{1}{x} + \frac{1}{y} = \frac{2n}{k} does not depend on aa. Hence, for such xx and yy, the right hand side of the functional equation has a value that does not depend on aa. In particular, with n2n \ge 2, a=0a = 0 and a=1a = 1, the functional equation implies
f(f(kn)+f(kn))=f(f(kn1)+f(kn+1)). f\left(f\left(\frac{k}{n}\right) + f\left(\frac{k}{n}\right)\right) = f\left(f\left(\frac{k}{n-1}\right) + f\left(\frac{k}{n+1}\right)\right).
As ff is injective, we can equate the arguments of the two outer ff's to deduce:
f(kn+1)f(kn)=f(kn)f(kn1).(20) f\left(\frac{k}{n+1}\right) - f\left(\frac{k}{n}\right) = f\left(\frac{k}{n}\right) - f\left(\frac{k}{n-1}\right). \quad (20)

Next, we are going to prove that there are rational numbers a,ba, b such that
f(x)=a+bxfor all xQ+.(21) f(x) = a + \frac{b}{x} \quad \text{for all } x \in \mathbb{Q}_+. \qquad (21)

Our first proof of (21) starts with the claim that for each kQ+k \in \mathbb{Q}_+ and any integer n1n \ge 1 there exist rational numbers a(k)a(k) and b(k)b(k) such that
f(kn)=a(k)+b(k)nk.(22) f\left(\frac{k}{n}\right) = a(k) + b(k)\frac{n}{k}. \qquad (22)
Indeed, from (20) we know that
b(k):=k(f(kn+1)f(kn)) b(k) := k \left( f\left(\frac{k}{n+1}\right) - f\left(\frac{k}{n}\right) \right)
does not depend on nn. In particular, we have b(k)k=f(k2)f(k)\frac{b(k)}{k} = f(\frac{k}{2}) - f(k). If we let a(k)=2f(k)f(k2)a(k) = 2f(k) - f(\frac{k}{2}) we obtain f(k)=a(k)+b(k)kf(k) = a(k) + \frac{b(k)}{k}. This is the case n=1n = 1 of (20) which gets our proof by induction going. For the inductive step, we use the definition of b(k)b(k) and the inductive hypothesis:
f(kn+1)=f(kn)+b(k)k=a(k)+b(k)nk+b(k)k=a(k)+b(k)n+1k. f\left(\frac{k}{n+1}\right) = f\left(\frac{k}{n}\right) + \frac{b(k)}{k} = a(k) + b(k)\frac{n}{k} + \frac{b(k)}{k} = a(k) + b(k)\frac{n+1}{k}.
We next show that a(k)a(k) and b(k)b(k) do not depend on kk. For any kQ+k \in \mathbb{Q}_+ and nZ+n \in \mathbb{Z}_+ we have
f(kn)=f(k1)andf(k2n)=f(k2)wherek=kn f\left(\frac{k}{n}\right) = f\left(\frac{k'}{1}\right) \quad \text{and} \quad f\left(\frac{k}{2n}\right) = f\left(\frac{k'}{2}\right) \quad \text{where} \quad k' = \frac{k}{n}
and (22) gives us
a(k)+b(k)nk=a(k)+b(k)nkanda(k)+b(k)2nk=a(k)+b(k)2nk. a(k) + b(k) \frac{n}{k} = a(k') + b(k') \frac{n}{k} \quad \text{and} \quad a(k) + b(k) \frac{2n}{k} = a(k') + b(k') \frac{2n}{k}.
Subtracting the second from twice the first equation, and the first from the second, gives
a(k)=a(k)andb(k)nk=b(k)nkhence a(k) = a(k') \quad \text{and} \quad b(k) \frac{n}{k} = b(k') \frac{n}{k'} \quad \text{hence}
a(k)=a(kn)andb(k)=b(kn)for all kQ+,nZ+. a(k) = a\left(\frac{k}{n}\right) \quad \text{and} \quad b(k) = b\left(\frac{k}{n}\right) \quad \text{for all } k \in \mathbb{Q}_+, n \in \mathbb{Z}_+.
Letting k=1k = 1 we now obtain a(1)=a(1n)a(1) = a(\frac{1}{n}) for all nZ+n \in \mathbb{Z}_+. Using k=n/mk = n/m with arbitrary m,nZ+m, n \in \mathbb{Z}_+ we obtain a(nm)=a(1m)a(\frac{n}{m}) = a(\frac{1}{m}). Together these show that a(k)=a(1)a(k) = a(1) for all kQ+k \in \mathbb{Q}_+. Similarly it follows that b(k)=b(1)b(k) = b(1) for all kQ+k \in \mathbb{Q}_+. This establishes (21) with a=a(1)a = a(1) and b=b(1)b = b(1).

For our second proof of (21) we define
g(k):=f(kn+1)f(kn) g(k) := f\left(\frac{k}{n+1}\right) - f\left(\frac{k}{n}\right)
For all kQ+k \in \mathbb{Q}_+ and integers n1n \ge 1. From (20) we know that the right hand side does not depend on nn. A straightforward induction yields
f(kn+i)=f(kn)+ig(k)for all kQ+,n1,i0.(23) f\left(\frac{k}{n+i}\right) = f\left(\frac{k}{n}\right) + ig(k) \quad \text{for all } k \in \mathbb{Q}_+, n \ge 1, i \ge 0. \quad (23)
Next we show that vg(1)=ug(u/v)vg(1) = ug(u/v) for all positive integers u,vu, v. To see this, we first substitute k=1k = 1 and i=n=vi = n = v in equation (23) to obtain
f(12v)=f(1v)+vg(1). f\left(\frac{1}{2v}\right) = f\left(\frac{1}{v}\right) + vg(1).
Next we substitute k=u/vk = u/v and i=n=ui = n = u in equation (23) and obtain
f(12v)=f(1v)+ug(uv). f\left(\frac{1}{2v}\right) = f\left(\frac{1}{v}\right) + ug\left(\frac{u}{v}\right).
Comparing these two equations we get the desired equality, which means that kg(k)=g(1)kg(k) = g(1) for all kQ+k \in \mathbb{Q}_+.
Note that substituting n=1n = 1 and i=m10i = m - 1 \ge 0 in (23) gives
f(km)=f(k)+(m1)g(k).(24) f\left(\frac{k}{m}\right) = f(k) + (m-1)g(k). \quad (24)
For kQ+k \in \mathbb{Q}_+ we now define a(k)=f(k)g(k)a(k) = f(k) - g(k). Setting k=1k = 1 in (24) gives
f(1m)=f(1)+(m1)g(1)=a(1)+mg(1). f\left(\frac{1}{m}\right) = f(1) + (m-1)g(1) = a(1) + mg(1).
With k=n/mk = n/m and mm replaced by nn in (24) we obtain
f(1m)=f(nmn)=f(nm)+(n1)g(nm)=a(nm)+ng(nm)=a(nm)+mg(1). \begin{aligned} f\left(\frac{1}{m}\right) &= f\left(\frac{\frac{n}{m}}{n}\right) = f\left(\frac{n}{m}\right) + (n-1)g\left(\frac{n}{m}\right) \\ &= a\left(\frac{n}{m}\right) + ng\left(\frac{n}{m}\right) = a\left(\frac{n}{m}\right) + mg(1). \end{aligned}
Comparing these two equations, we obtain a(n/m)=a(1)a(n/m) = a(1) for all positive integers m,nm, n. If we now let a=a(1)a = a(1) and b=g(1)b = g(1) we finally obtain
f(k)=a(k)+g(k)=a(1)+g(1)k=a+bk f(k) = a(k) + g(k) = a(1) + \frac{g(1)}{k} = a + \frac{b}{k}
for all kQ+k \in \mathbb{Q}_+. This finishes the second proof of (21).

Our penultimate step is to show that a=0a = 0 in (21). To show this, we start by substituting (21) into the original functional equation with y=xy = x:
x2=f(f(x)+f(x))=f(2f(x))=f(2a+2bx)=a+bx2ax+2b. \frac{x}{2} = f(f(x) + f(x)) = f(2f(x)) = f\left(2a + \frac{2b}{x}\right) = a + \frac{bx}{2ax + 2b}.
This implies x(ax+b)=2a(ax+b)+bxx(ax+b) = 2a(ax+b)+bx, i.e. a(2axx2+2b)=0a(2ax-x^2+2b) = 0. So, either a=0a=0 or 2b=x22ax2b = x^2-2ax for all xQ+x \in \mathbb{Q}_+. But the polynomial x22axx^2-2ax is not constant, so we must have a=0a=0. Thus, f(x)=bxf(x) = \frac{b}{x} with bQ+b \in \mathbb{Q}_+. An easy check reveals that all such functions satisfy the functional equation. As f(1)=2023f(1) = 2023, we must have b=2023b = 2023 so that f(2023)=1f(2023) = 1.

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.