A geometric property that guarantees convergence of gradient-based optimization without requiring convexity.