Friday, November 20, 2026

On IMO 1988 P6

 The famous "hardest" problem of IMO: If positive integers $a,b$ are such that

$$\frac{a^2 + b^2}{1 + ab} = k$$

and $k$ is a positive integer, then $k$ must be a perfect square.

In this post we will allow $a,b$ to be integers (while keeping $k > 0$), and also try to characterize the solution set when $k$ is a perfect square.

.

.

.

.

.

First we start of with a useful observation.

$\textbf{Observation:}$ Any solution $(a,b)$ satisfies $ab \geq 0$.

$\textbf{Proof:}$ We must have $1 + ab = \frac{a^2 + b^2}{k} > 0$ and thus $ab \geq 0$. 

Now because the quadratic in a 

$$a^2 - (kb) a + (b^2 - k) = 0$$

has sum of roots $kb$, if $(a_0, b)$ is a solution, then so is $(kb - a_0, b)$. Similary treating it as a quadratic in $b$, if $(a, b_0)$ is a solution and thus so is $(a, ka - b_0)$.

Thus if $(a,b)$ is a solution, then so are $(a, ka -b) = (a,x)$ and $(kb - a, b) = (y,b)$. Call these the companion solutions of $(a,b)$

Also observe that from both $(a,x)$ and $(y,b)$ we get back $(a,b)$ as one of the companion solutions.

Now $bx = (ka - b)b = a^2 -k$ and $ay = (kb-a)a = b^2 -k$.

Thus $axby = (a^2 - k)(b^2 - k)$. If $ab > 0$, then we must have $axby < a^2b^2$.

Thus at-least one of $ax$ or $by$ is a non-negative integer $< ab$

Thus if we have a solution $(a,b)$ with $ab > 0$, one of its companion solution has a lower value of the product $ab$.

Since $ab \geq 0$, we must ultimately have that $xy = 0$ (say $y = 0$) for some solution $(x,y)$, and thus $k = \frac{x^2 + 0}{1 + x\times0} = x^2$ and so $k$ must be a perfect square (and this completes the proof for the IMO problem).

$\textbf{Characterization of the solutions}:$

Say $k = m^2$.

Now observe that, what we showed that, if $(a,b)$ is any solution with $ab > 0$, then there is a sequence of companions leading to the $(0, m)$ solution. We also noted that this is a reversible step, and thus any solution can be obtained by starting with $(0, m)$ and taking the companions.

Thus we could characterize solutions as a sequence of companions.

$$(a_{n+1}, b_{n+1})  = (k a_n - b_n, a_n)$$

Starting with  $(a_0, b_0) = (m, 0)$

This can be written as a linear recurrence (with characteristic equation $x^2 -kx + 1 = 0$) or we can use matrix multiplication (applying diagonalization) etc and we can easily give a closed form formula, which we will omit the details.

The negative integer solutions are easily dealt with because we always have $ab \geq 0$ and thus we can negate the solutions in which both $a,b$ are non-negative.

Sunday, November 1, 2026

When the third derivative is the function itself

 Suppose $f: \mathbb{R} \to \mathbb{R}$ is an infinite differentiable function such that the third derivative of $f$ is $f$ itself. i.e.

$$ f'''(x) = f(x) \quad \forall x \in \mathbb{R}$$

It is easy to see that $f(x) = Ae^x$ satisfies this. Can you find others?

Scroll down for a solution.

.

.

.

.

Solution:

If the equation was $f'''(x) = 8f(x)$ then we could easily see that $Ae^{2x}$ would be a solution.

What if there was a solution of the form $Ae^{kx}$? Then we must have that $k^3 = 1$. $k = 1$ already corresponds to the $Ae^x$ case. But the others are complex, so is that the end of this approach?

No!

We can still get our solution out of this. Consider $$f(x) = Re(e^{wx})$$

i.e. we take the real part of the solution that allows complex numbers. It must satisfy the conditions of the problem!

In this case

$$f(x) = \cos \left(\frac{\sqrt{3}}{2}x \right) e^{-\frac{x}{2}}$$

works! Try it.

There are other approaches, like power series. But the matrix differential equation based approach gives us the complete set of solutions and comes out to be any linear combination of $$e^x, \cos \left(\frac{\sqrt{3}}{2}x \right) e^{-\frac{x}{2}}, \sin \left(\frac{\sqrt{3}}{2}x \right) e^{-\frac{x}{2}}$$