The Extremal Principle

Some interview puzzles ask you to prove that something exists — a strategy, a player, a subset — without telling you which one. Trying to construct it directly can stall, because you have no way to point at the right object. The extremal principle sidesteps this: instead of hunting for the object, pick out whichever candidate is the biggest, smallest, first, or otherwise most extreme by some measure, and show that this one must have the property you need.

Worked example: the tournament king

In a round-robin tournament, every pair of players plays exactly one game and there are no ties. Prove that there is always a player — call them the "king" — who either beat every other player directly, or beat every other player who wasn't beaten directly, through one player they beat.

Pick the extreme. Let $K$ be a player with the most wins in the whole tournament (if several players tie for most wins, pick any one of them).

Suppose $K$ isn't a king. Then some player $P$ is not beaten by $K$ directly, and is not beaten by anyone $K$ beat either. Since $K$ and $P$ played each other and $K$ didn't beat $P$, $P$ beat $K$. And since nobody $K$ beat managed to beat $P$, $P$ beat every single player that $K$ beat too.

The contradiction. So $P$ beat $K$, and everyone $K$ beat. That means $P$ has strictly more wins than $K$ — at least one more, from beating $K$ itself. But $K$ was chosen to have the most wins in the entire tournament. Contradiction. So no such $P$ exists, and the player with the most wins is always a king.

The same move proves a finite set of positive integers has a smallest element (trivial, but it's the base case of every induction), and that any non-empty set of positive reals bounded above has candidates worth checking only at its supremum. Whenever a puzzle says "prove there exists..." and direct construction feels blocked, ask: what happens at the extreme case, and can I derive a contradiction if it lacks the property?

Quant research uses the same reflex when hunting for what breaks a model: rather than testing every parameter combination, look at the parameter that maximises some exposure or minimises some diversification — that boundary case is usually where the assumption fails first.