Potential-function proofs for gradient methods

Abstract

This note discusses proofs of convergence for gradient methods (also called “first-order methods”) based on simple potential-function arguments. We cover methods like gradient descent (for both smooth and non-smooth settings), mirror descent, and some accelerated variants. We hope the structure and presentation of these amortized-analysis proofs will be useful as a guiding principle in learning and using these proofs.