Stationarity with Gradient-Dependent Third Derivatives
Descriptions may be stale. Consult the linked Lean sources.
Stationarity with Gradient-Dependent Third Derivatives
We consider a possibly nonconvex objective
The underlying optimization problem is
Our assumptions:
Assumption 1 (Third-derivative bound). The function
Assumption 2 (Initial bounds). The given point
Queries to either oracle may be made at any point of
After the allotted calls to each oracle, the output must satisfy
Results
On a finite-dimensional real inner-product space,
| Result | Oracle complexity | Novelty | Limitations |
|---|---|---|---|
| (1) EnergyHessianReuse
(Ours) |
|||
| (2) Parameter-free CRN (Fang et al. 2026), Cor. 1 | Includes backtracking trials; |
(1): |
(1) needs |
| (3) LazyCRN (Doikov et al. 2023), Thm. 3.1, Cor. 3.7 | Special case: |
(1): Scope extends
Hessian reuse to gradient-dependent third derivatives, including
unbounded Hessian variation. Matches known rate at |
(1) needs |
| (4) NALEN (Chen et al. 2026), Thm. 4.1 | Special case: globally |
(1): Scope beyond
globally Lipschitz Hessians. On the intersection, smaller accuracy term
if |
(1) has |
Abbr: Costs are