Accelerating Trust-Region Methods: An Attempt to Balance Global and Local Efficiency
Yuntian Jiang, Chuwen Zhang, Bo Jiang, Yinyu Ye · 2025
Read Accelerating Trust-Region Methods: An Attempt to Balance Global and Local Efficiency by Yuntian Jiang, Chuwen Zhang, Bo Jiang, Yinyu Ye (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-00680.txt.
Description
Yuntian Jiang, Chuwen Zhang, Bo Jiang, and Yinyu Ye attack a standing trade-off in second-order optimization for convex problems. Newton's method converges quadratically fast near a solution but can diverge from far away. Accelerated methods guarantee global progress yet crawl locally. The question is how much global acceleration is possible before the fast local finish is lost.
Their accelerated trust-region method uses Lagrange multiplier information, already present in the primal-dual structure, to detect when the search enters a local region. It achieves strong global complexity while keeping quadratic local convergence. Pushing further, an extragradient variant reaches still better global complexity but loses the quadratic finish, revealing a phase-transition-like boundary. Numerical experiments track the theory, mapping exactly where the trade-off turns.
Catalog record
| CREATOR | Yuntian Jiang, Chuwen Zhang, Bo Jiang, Yinyu Ye |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 92 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-00680