Creating a proof-of-stake algorithm that doesn't degenerate to proof-of-work and that doesn't require a central authority (checkpoints, etc) is still an open problem.