Basic Inequalities for First-Order Optimization with Applications to Statistical Risk Analysis