Mathematical proofs based on a technique called diagonalization can be relentlessly contrarian, but they help reveal the limits of algorithms. How hard is it to prove that problems are hard to solve?