A Globalization of L-BFGS and the Barzilai-Borwein Method for Nonconvex Unconstrained Optimization

被引:0
|
作者
Mannel, Florian [1 ]
机构
[1] Univ Lubeck, Inst Math & Image Comp, Maria Goeppert Str 3, D-23562 Lubeck, Germany
关键词
L-BFGS; Barzilai-Borwein methods; Cautious updates; Nonconvex optimization; Global convergence; Linear convergence; Hilbert space; QUASI-NEWTON METHODS; LIMITED-MEMORY; GLOBAL CONVERGENCE; LINEAR CONVERGENCE; ALGORITHMS; MINIMIZATION; CONSTRAINTS; MATRICES; FAMILY;
D O I
10.1007/s10957-024-02565-5
中图分类号
C93 [管理学]; O22 [运筹学];
学科分类号
070105 ; 12 ; 1201 ; 1202 ; 120202 ;
摘要
We present a modified limited memory BFGS (L-BFGS) method that converges globally and linearly for nonconvex objective functions. Its distinguishing feature is that it turns into L-BFGS if the iterates cluster at a point near which the objective is strongly convex with Lipschitz gradients, thereby inheriting the outstanding effectiveness of the classical method. These strong convergence guarantees are enabled by a novel form of cautious updating, where, among others, it is decided anew in each iteration which of the stored pairs are used for updating and which ones are skipped. In particular, this yields the first modification of cautious updating for which all cluster points are stationary while the spectrum of the L-BFGS operator is not permanently restricted, and this holds without Lipschitz continuity of the gradient. In fact, for Wolfe-Powell line searches we show that continuity of the gradient is sufficient for global convergence, which extends to other descent methods. Since we allow the memory size to be zero in the globalized L-BFGS method, we also obtain a new globalization of the Barzilai-Borwein spectral gradient (BB) method. The convergence analysis is developed in Hilbert space under comparably weak assumptions and covers Armijo and Wolfe-Powell line searches. We illustrate the theoretical findings with numerical experiments. The experiments indicate that if one of the parameters of the cautious updating is chosen sufficiently small, then the modified method agrees entirely with L-BFGS/BB. We also discuss this in the theoretical part. An implementation of the new method is available on arXiv.
引用
收藏
页数:34
相关论文
共 50 条
  • [1] The cyclic Barzilai-Borwein method for unconstrained optimization
    Dai, Yu-Hong
    Hager, William W.
    Schittkowski, Klaus
    Zhang, Hongchao
    IMA JOURNAL OF NUMERICAL ANALYSIS, 2006, 26 (03) : 604 - 627
  • [2] An Efficient Barzilai-Borwein Conjugate Gradient Method for Unconstrained Optimization
    Liu, Hongwei
    Liu, Zexian
    JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS, 2019, 180 (03) : 879 - 906
  • [3] An adaptive nonmonotone global Barzilai-Borwein gradient method for unconstrained optimization
    Nosratipour, Hadi
    Fard, Omid Solaymani
    Borzabadi, Akbar Hashemi
    OPTIMIZATION, 2017, 66 (04) : 641 - 655
  • [4] Stochastic Gradient Method with Barzilai-Borwein Step for Unconstrained Nonlinear Optimization
    Wang, L.
    Wu, H.
    Matveev, I. A.
    JOURNAL OF COMPUTER AND SYSTEMS SCIENCES INTERNATIONAL, 2021, 60 (01) : 75 - 86
  • [5] EXTENDED BARZILAI-BORWEIN METHOD FOR UNCONSTRAINED MINIMIZATION PROBLEMS
    Narushima, Yasushi
    Wakamatsu, Takahiko
    Yabe, Hiroshi
    PACIFIC JOURNAL OF OPTIMIZATION, 2010, 6 (03): : 591 - 613
  • [6] On the Barzilai-Borwein method
    Fletcher, R
    OPTIMIZATION AND CONTROL WITH APPLICATIONS, 2005, 96 : 235 - 256
  • [7] A New Nonmonotone Trust Region Barzilai-Borwein Method for Unconstrained Optimization Problems
    Xing Li
    Wen-li Dong
    Zheng Peng
    Acta Mathematicae Applicatae Sinica, English Series, 2021, 37 : 166 - 175
  • [8] A new nonmonotone filter Barzilai-Borwein method for solving unconstrained optimization problems
    Arzani, F.
    Peyghami, M. Reza
    INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS, 2016, 93 (03) : 596 - 608
  • [9] Nonmonotone globalization techniques for the Barzilai-Borwein gradient method
    Grippo, L
    Sciandrone, M
    COMPUTATIONAL OPTIMIZATION AND APPLICATIONS, 2002, 23 (02) : 143 - 169
  • [10] A New Nonmonotone Trust Region Barzilai-Borwein Method for Unconstrained Optimization Problems
    Xing LI
    Wen-li DONG
    Zheng PENG
    Acta Mathematicae Applicatae Sinica, 2021, 37 (01) : 166 - 175