Sharp First-Order Lower Bounds for Higher-Order Smooth Nonconvex Optimization
For optimization theorists, this resolves the open problem of tight lower bounds for higher-order smooth nonconvex optimization, confirming optimality of existing algorithms.
The paper proves matching lower bounds for first-order optimization of higher-order smooth nonconvex functions, achieving Ω(ε^{-7/4}) for Lipschitz Hessians and Ω(ε^{-5/3}) for Lipschitz third derivatives, closing the gap with known upper bounds.
We study the deterministic first-order oracle complexity of finding \(ε\)-stationary points in smooth nonconvex optimization when the objective satisfies higher-order smoothness assumptions. While the classical \(ε^{-2}\) rate is optimal under only Lipschitz gradients, higher-order smoothness leads to accelerated first-order upper bounds, most notably the \(ε^{-7/4}\) rate under Lipschitz Hessians and the \(ε^{-5/3}\) rate under Lipschitz third derivatives. The matching lower bounds, however, have remained open. We resolve this gap by proving a new dimension-free first-order lower bound for higher-order smooth nonconvex functions, valid for every finite smoothness order. In particular, our construction gives a matching \(Ω(ε^{-7/4})\) lower bound in the Hessian-Lipschitz case and a matching \(Ω(ε^{-5/3})\) lower bound in the third-order-smooth regime. The hard instance is based on a \emph{block-chain} mechanism that enforces blockwise oracle revelation while preserving the smoothness structure needed for the scalar hard instance. The lower-bound construction was discovered with the assistance of ChatGPT 5.5 Pro and subsequently verified by the authors.