-
本文考虑下面的无约束优化问题:
这里f:
${\mathbb{R}^n} \to \mathbb{R}$ 是一个光滑函数,$\mathbb{R}^n$ 表示n维欧氏空间,函数f的梯度记为g(x):=▽f(x). 在实际求解大规模无约束优化问题中,建立一些具有低存储的快速数值算法是十分重要的. 共轭梯度法是求解大型无约束优化问题(1)的一类十分有效的方法[1],其通常的迭代形式如下这里xk∈
$\mathbb{R}^n$ 是一个解的第k次逼近,αk > 0是由一种合适的线搜索所确定的步长,dk∈$\mathbb{R}^n$ 是搜索方向,其定义如下其中:gk表示梯度g(xk),βk∈
$\mathbb{R}$ 为共轭参数. 步长αk由标准的Wolfe线搜索或者强Wolfe线搜索
确定,其中0 < δ < σ < 1. 不同的共轭梯度法对应不同的共轭参数βk. 众所周知的共轭梯度法包括Hestenes-Stiefel(HS)[2],Polak-Ribière-Polyak(PRP)[3-4],Dai-Yuan(DY)[5],Liu-Storey(LS)[6],Fletcher-Reeves(FR)[7]和Conjugate-Descent(CD)[8],它们所对应的共轭参数如下
其中yk-1:=gk-gk-1且||·||表示欧氏范数. 目标函数是一个严格凸二次函数并且步长由精确线搜索得到,且在一般情形下,它们的理论性质和数值表现不尽相同. 众所周知,DY法、CD法和FR法具有良好的收敛性质,但它们的数值计算效果一般;相反,PRP法、HS法和LS法具备出色的数值表现,但它们可能不收敛. 为了获得收敛性数值效果的算法,不少学者对共轭参数自身作了一些改进[9-10]和一些混合[11-12]. 本文主要考虑共轭梯度法的混合形式. 文献[13]通过对PRP法和DY法进行凸组合,得到一种新的共轭梯度法,其共轭参数如下
其中参数θk∈[0, 1]由共轭条件
$\boldsymbol{y}_{k-1}^{\mathrm{T}} \boldsymbol{d}_{k}=0$ 确定. 最近,文献[14]对PRP法和FR法进行凸组合,引进了一种混合共轭梯度法:这里参数θk∈[0, 1]根据求解优化问题
$\mathop {\min }\limits_{_{{\theta _k}}}\left\{\left\|\boldsymbol{d}_{k}^{H C G}-\boldsymbol{d}_{k}^{Z Z L}\right\|^{2}: k \geqslant 1\right\}$ 得到. 在标准的Wolfe线搜索条件下,该算法具有全局收敛性和良好的计算表现.受文献[14]的启发,本文构造了一种新的混合共轭梯度法
这里
其中
为优化问题
$\mathop {\min }\limits_{_{{\theta _k}}}\left\{\left\|\boldsymbol{d}_{k}^{\mathrm{NH}+}-\boldsymbol{d}_{k}^{\mathrm{NYF}}\right\|^{2}: k \geqslant 1\right\}$ 的最优解. 基于共轭参数βkNH+所得相应的共轭梯度法称作NH+算法或NH+法. 因为文献[15]提出的ZZL法是文献[16]提出的NYF法特殊情况. 因此,文献[14]提出的方法是NH+法的特殊情况(当pk=gk即为文献[14]中的方法).
A New Hybrid Conjugate Gradient Method with Sufficient Descent Property
-
摘要: 对PRP法和FR法进行凸组合, 提出了一种求解无约束优化问题的新共轭梯度法.该方法总是能生成一个充分下降方向, 且它的凸组合参数为Babaie-Kafaki和Ghanbari的推广形式.在Wolfe线搜索条件下, 新算法的全局收敛性得以建立, 数值结果也说明提出的算法是有效的.Abstract: In this paper, a new hybrid conjugate gradient algorithm has been developed for solving unconstrained optimization problems. The proposed method can be viewed as a convex combination of Polak-Ribière-Polyak method and Fletcher-Reeves method. In this method, a sufficient descent direction can always be generated, with its convex combination parameter as an extension of the parameter proposed by Babaie-Kafaki and Ghanbari. Under the Wolfe line search, the global convergence has been established. Some numerical results show that proposed method is effective.
-
Key words:
- unconstrained optimization /
- conjugate gradient method /
- sufficient descent /
- global convergence .
-
表 1 部分数值实验结果
函数名 维度/维 算法 迭代次数/次 迭代时间/s 梯度值 LIARWHD 20 NH+ 138 2 137.390 9.37E-07 20 PRP+ 94 1 303.897 9.93E-07 20 NYF 345 11 117.170 9.02E-06 20 HCG+ 340 3 605.104 1.10E-05 DIAGONAL 50 NH+ 179 1 957.943 8.47E-07 50 PRP+ 132 1 292.290 7.51E-07 50 NYF 708 2 381.197 8.95E-07 50 HCG+ 740 3 615.365 1.04E-05 QUADRATICQF1 100 NH+ 181 2 089.250 0.00E+00 100 PRP+ 136 1 315.878 0.00E+00 100 NYF 231 4 130.011 1.15E-02 100 HCG+ 282 3 613.380 9.89E-03 QUARTC 400 NH+ 17 1 859.335 9.91E-07 400 PRP+ 48 3 633.389 4.55E-03 400 NYF 50 3 643.121 3.70E-03 400 HCG+ 17 1 956.870 9.91E-07 -
[1] 唐天国. 一种求解无约束优化问题的新混合共轭梯度法[J]. 西南师范大学学报(自然科学版), 2019, 44(9): 34-39. doi: http://xbgjxt.swu.edu.cn/article/doi/10.13718/j.cnki.xsxb.2019.09.006 [2] HESTENES M R, STIEFEL E. Methods of Conjugate Gradients for Solving Linear Systems[J]. Journal of Research of the National Bureau of Standards, 1952, 49(6): 409-436. doi: 10.6028/jres.049.044 [3] POLAK E, RIBIERE G. Note Sur La Convergence de Méthodes de Directions Conjuguées[J]. Revue Française d'Informatique et De Recherche Opérationnelle Série Rouge, 1969, 3(16): 35-43. [4] POLYAK B T. The Conjugate Gradient Method in Extremal Problems[J]. USSR Computational Mathematics and Mathematical Physics, 1969, 9(4): 94-112. doi: 10.1016/0041-5553(69)90035-4 [5] DAI Y H, YUAN Y. A Nonlinear Conjugate Gradient Method with a Strong Global Convergence Property[J]. SIAM Journal on Optimization, 1999, 10(1): 177-182. doi: 10.1137/S1052623497318992 [6] LIU Y, STOREY C. Efficient Generalized Conjugate Gradient Algorithms, Part 1: Theory[J]. Journal of Optimization Theory and Applications, 1991, 69(1): 129-137. doi: 10.1007/BF00940464 [7] FLETCHER R, REEVES C M. Function Minimization by Conjugate Gradients[J]. The Computer Journal, 1964, 7(2): 149-154. doi: 10.1093/comjnl/7.2.149 [8] FLETCHER R. Practical Method of Optimization, Unconstrained Optimization[M]. New York: John Wiley and Sons, 1987. [9] SHENGWEI Y, WEI Z X, HUANG H. A Note about WYL's Conjugate Gradient Method and Its Applications[J]. Applied Mathematics and Computation, 2007, 191(2): 381-388. doi: 10.1016/j.amc.2007.02.094 [10] DAI Z F, WEN F H. Another Improved Wei-Yao-Liu Nonlinear Conjugate Gradient Method with Sufficient Descent Property[J]. Applied Mathematics and Computation, 2012, 218(14): 7421-7430. doi: 10.1016/j.amc.2011.12.091 [11] 韩信, 张俊容, 王森森. 一种新的混合共轭梯度算法[J]. 西南大学学报(自然科学版), 2017, 39(5): 132-138. doi: https://www.cnki.com.cn/Article/CJFDTOTAL-XNND201705020.htm [12] LIU J K, DU X L. Global Convergence of an Efficient Hybrid Conjugate Gradient Method for Unconstrained Optimization [J]. Bulletin of the Korean Mathematical Society, 2013, 50(1): 73-81. doi: 10.4134/BKMS.2013.50.1.073 [13] ANDREI N. Hybrid Conjugate Gradient Algorithm for Unconstrained Optimization[J]. Journal of Optimization Theory and Applications, 2009, 141(2): 249-264. doi: 10.1007/s10957-008-9505-0 [14] BABAIE-KAFAKI S, GHANBARI R. A Hybridization of the Polak-Ribière-Polyak and Fletcher-Reeves Conjugate Gradient Methods[J]. Numerical Algorithms, 2015, 68(3): 481-495. doi: 10.1007/s11075-014-9856-6 [15] ZHANG L, ZHOU W J, LI D H. A Descent Modified Polak-Ribière-Polyak Conjugate Gradient Method and Its Global Convergence[J]. IMA Journal of Numerical Analysis, 2006, 26(4): 629-640. doi: 10.1093/imanum/drl016 [16] NARUSHIMA Y, YABE H, FORD J A. A Three-Term Conjugate Gradient Method with Sufficient Descent Property for Unconstrained Optimization[J]. SIAM Journal on Optimization, 2011, 21(1): 212-230. doi: 10.1137/080743573 [17] ZOUTENDIJK G. Nonlinear Programming, Computational Methods[J]. Integer and Nonlinear Programming, 1970, 143: 37-86. [18] GILBERT J C, NOCEDAL J. Global Convergence Properties of Conjugate Gradient Methods for Optimization[J]. SIAM Journal on Optimization, 1992, 2(1): 21-42. doi: 10.1137/0802003 [19] ANDREI N. An Unconstrained Optimization Test Functions Collection[J]. Environmental Science and Technology, 2008, 10(1): 6552-6558. [20] DOLAN E D, MORÉ J J. Benchmarking Optimization Software with Performance Profiles[J]. Mathematical Programming, 2002, 91(2): 201-213. doi: 10.1007/s101070100263