Maths Olympiad Prep

Track / Stage 8 / 166 of 180 #2346 of 2444

Problem 2346

IMO Shortlist mid-range; USAMO P2/P5
Algebra Difficulty 8.7 Prove it IMO Shortlist · IMO

Determine all functions f:(0,)Rf:(0, \infty) \rightarrow \mathbb{R} satisfying
(x+1x)f(y)=f(xy)+f(yx) \left(x+\frac{1}{x}\right) f(y)=f(x y)+f\left(\frac{y}{x}\right)
for all x,y>0x, y>0.

This one wants a proof. Work it on paper, then read the official solution and mark yourself. Be honest about it: the record is only any use to you if it is.

Next problem →

Official solution

Answer: f(x)=C1x+C2xf(x)=C_{1} x+\frac{C_{2}}{x} with arbitrary constants C1C_{1} and C2C_{2}.

Solution 1. Fix a real number a>1a>1, and take a new variable tt. For the values f(t),f(t2)f(t), f\left(t^{2}\right), f(at)f(a t) and f(a2t2)f\left(a^{2} t^{2}\right), the relation (1) provides a system of linear equations:
x=y=t:(t+1t)f(t)x=ta,y=at:(ta+at)f(at)=f(t2)+f(t2)+f(a2)x=a2t,y=t:(a2t+1a2t)f(t)=f(a2t2)+f(1a2)x=y=at:(at+1at)f(at)=f(a2t2)+f(1) \begin{array}{ll} x=y=t: & \left(t+\frac{1}{t}\right) f(t) \\ x=\frac{t}{a}, y=a t: & \left(\frac{t}{a}+\frac{a}{t}\right) f(a t)=f\left(t^{2}\right)+f\left(t^{2}\right)+f\left(a^{2}\right) \\ x=a^{2} t, y=t: & \left(a^{2} t+\frac{1}{a^{2} t}\right) f(t)=f\left(a^{2} t^{2}\right)+f\left(\frac{1}{a^{2}}\right) \\ x=y=a t: & \left(a t+\frac{1}{a t}\right) f(a t)=f\left(a^{2} t^{2}\right)+f(1) \end{array}
In order to eliminate f(t2)f\left(t^{2}\right), take the difference of (2a) and (2b); from (2c) and (2d) eliminate f(a2t2)f\left(a^{2} t^{2}\right); then by taking a linear combination, eliminate f(at)f(a t) as well:
(t+1t)f(t)(ta+at)f(at)=f(1)f(a2) and (a2t+1a2t)f(t)(at+1at)f(at)=f(1/a2)f(1), so ((at+1at)(t+1t)(ta+at)(a2t+1a2t))f(t)=(at+1at)(f(1)f(a2))(ta+at)(f(1/a2)f(1)) \begin{gathered} \left(t+\frac{1}{t}\right) f(t)-\left(\frac{t}{a}+\frac{a}{t}\right) f(a t)=f(1)-f\left(a^{2}\right) \quad \text{ and } \\ \left(a^{2} t+\frac{1}{a^{2} t}\right) f(t)-\left(a t+\frac{1}{a t}\right) f(a t)=f\left(1 / a^{2}\right)-f(1), \quad \text{ so } \\ \left(\left(a t+\frac{1}{a t}\right)\left(t+\frac{1}{t}\right)-\left(\frac{t}{a}+\frac{a}{t}\right)\left(a^{2} t+\frac{1}{a^{2} t}\right)\right) f(t) \\ =\left(a t+\frac{1}{a t}\right)\left(f(1)-f\left(a^{2}\right)\right)-\left(\frac{t}{a}+\frac{a}{t}\right)\left(f\left(1 / a^{2}\right)-f(1)\right) \end{gathered}
Notice that on the left-hand side, the coefficient of f(t)f(t) is nonzero and does not depend on tt :
(at+1at)(t+1t)(ta+at)(a2t+1a2t)=a+1a(a3+1a3)<0. \left(a t+\frac{1}{a t}\right)\left(t+\frac{1}{t}\right)-\left(\frac{t}{a}+\frac{a}{t}\right)\left(a^{2} t+\frac{1}{a^{2} t}\right)=a+\frac{1}{a}-\left(a^{3}+\frac{1}{a^{3}}\right)<0 .
After dividing by this fixed number, we get
f(t)=C1t+C2t \begin{equation*} f(t)=C_{1} t+\frac{C_{2}}{t} \end{equation*}
where the numbers C1C_{1} and C2C_{2} are expressed in terms of a,f(1),f(a2)a, f(1), f\left(a^{2}\right) and f(1/a2)f\left(1 / a^{2}\right), and they do not depend on tt.
The functions of the form (3) satisfy the equation:
(x+1x)f(y)=(x+1x)(C1y+C2y)=(C1xy+C2xy)+(C1yx+C2xy)=f(xy)+f(yx). \left(x+\frac{1}{x}\right) f(y)=\left(x+\frac{1}{x}\right)\left(C_{1} y+\frac{C_{2}}{y}\right)=\left(C_{1} x y+\frac{C_{2}}{x y}\right)+\left(C_{1} \frac{y}{x}+C_{2} \frac{x}{y}\right)=f(x y)+f\left(\frac{y}{x}\right) .

Solution 2. We start with an observation. If we substitute x=a1x=a \neq 1 and y=any=a^{n} in (1), we obtain
f(an+1)(a+1a)f(an)+f(an1)=0. f\left(a^{n+1}\right)-\left(a+\frac{1}{a}\right) f\left(a^{n}\right)+f\left(a^{n-1}\right)=0 .
For the sequence zn=anz_{n}=a^{n}, this is a homogeneous linear recurrence of the second order, and its characteristic polynomial is t2(a+1a)t+1=(ta)(t1a)t^{2}-\left(a+\frac{1}{a}\right) t+1=(t-a)\left(t-\frac{1}{a}\right) with two distinct nonzero roots, namely aa and 1/a1 / a. As is well-known, the general solution is zn=C1an+C2(1/a)nz_{n}=C_{1} a^{n}+C_{2}(1 / a)^{n} where the index nn can be as well positive as negative. Of course, the numbers C1C_{1} and C2C_{2} may depend of the choice of aa, so in fact we have two functions, C1C_{1} and C2C_{2}, such that
f(an)=C1(a)an+C2(a)an for every a1 and every integer n \begin{equation*} f\left(a^{n}\right)=C_{1}(a) \cdot a^{n}+\frac{C_{2}(a)}{a^{n}} \text{ for every } a \neq 1 \text{ and every integer } n \text{. } \end{equation*}
The relation (4) can be easily extended to rational values of nn, so we may conjecture that C1C_{1} and C2C_{2} are constants, and whence f(t)=C1t+C2tf(t)=C_{1} t+\frac{C_{2}}{t}. As it was seen in the previous solution, such functions indeed satisfy (1).
The equation (1) is linear in ff; so if some functions f1f_{1} and f2f_{2} satisfy (1) and c1,c2c_{1}, c_{2} are real numbers, then c1f1(x)+c2f2(x)c_{1} f_{1}(x)+c_{2} f_{2}(x) is also a solution of (1). In order to make our formulas simpler, define
f0(x)=f(x)f(1)x. f_{0}(x)=f(x)-f(1) \cdot x .
This function is another one satisfying (1) and the extra constraint f0(1)=0f_{0}(1)=0. Repeating the same argument on linear recurrences, we can write f0(a)=K(a)an+L(a)anf_{0}(a)=K(a) a^{n}+\frac{L(a)}{a^{n}} with some functions KK and LL. By substituting n=0n=0, we can see that K(a)+L(a)=f0(1)=0K(a)+L(a)=f_{0}(1)=0 for every aa. Hence,
f0(an)=K(a)(an1an). f_{0}\left(a^{n}\right)=K(a)\left(a^{n}-\frac{1}{a^{n}}\right) .
Now take two numbers a>b>1a>b>1 arbitrarily and substitute x=(a/b)nx=(a / b)^{n} and y=(ab)ny=(a b)^{n} in (1):
(anbn+bnan)f0((ab)n)=f0(a2n)+f0(b2n), so (anbn+bnan)K(ab)((ab)n1(ab)n)=K(a)(a2n1a2n)+K(b)(b2n1b2n), or equivalently K(ab)(a2n1a2n+b2n1b2n)=K(a)(a2n1a2n)+K(b)(b2n1b2n). \begin{align*} \left(\frac{a^{n}}{b^{n}}+\frac{b^{n}}{a^{n}}\right) f_{0}\left((a b)^{n}\right) & =f_{0}\left(a^{2 n}\right)+f_{0}\left(b^{2 n}\right), \quad \text{ so } \\ \left(\frac{a^{n}}{b^{n}}+\frac{b^{n}}{a^{n}}\right) K(a b)\left((a b)^{n}-\frac{1}{(a b)^{n}}\right) & =K(a)\left(a^{2 n}-\frac{1}{a^{2 n}}\right)+K(b)\left(b^{2 n}-\frac{1}{b^{2 n}}\right), \quad \text{ or equivalently } \\ K(a b)\left(a^{2 n}-\frac{1}{a^{2 n}}+b^{2 n}-\frac{1}{b^{2 n}}\right) & =K(a)\left(a^{2 n}-\frac{1}{a^{2 n}}\right)+K(b)\left(b^{2 n}-\frac{1}{b^{2 n}}\right) . \end{align*}
By dividing (5) by a2na^{2 n} and then taking limit with n+n \rightarrow+\infty we get K(ab)=K(a)K(a b)=K(a). Then (5) reduces to K(a)=K(b)K(a)=K(b). Hence, K(a)=K(b)K(a)=K(b) for all a>b>1a>b>1.
Fix a>1a>1. For every x>0x>0 there is some bb and an integer nn such that 1<b<a1<b<a and x=bnx=b^{n}. Then
f0(x)=f0(bn)=K(b)(bn1bn)=K(a)(x1x). f_{0}(x)=f_{0}\left(b^{n}\right)=K(b)\left(b^{n}-\frac{1}{b^{n}}\right)=K(a)\left(x-\frac{1}{x}\right) .
Hence, we have f(x)=f0(x)+f(1)x=C1x+C2xf(x)=f_{0}(x)+f(1) x=C_{1} x+\frac{C_{2}}{x} with C1=K(a)+f(1)C_{1}=K(a)+f(1) and C2=K(a)C_{2}=-K(a).

Source: MathNet, licensed CC-BY-4.0. Statement reproduced verbatim; metadata (topic, difficulty, ordering) added by this project.