Critical solutions of nonlinear equations: local attraction for Newton-type methods

被引:17
作者
Izmailov, A. F. [1 ,2 ]
Kurennoy, A. S. [3 ]
Solodov, M. V. [4 ]
机构
[1] Lomonosov Moscow State Univ MSU, OR Dept, VMK Fac, Uchebniy Korpus 2, Moscow 119991, Russia
[2] Peoples Friendship Univ Russia, Miklukho Maklaya Str 6, Moscow 117198, Russia
[3] Derzhavin Tambov State Univ, TSU, Dept Math Phys & Comp Sci, Int Naya 33, Tambov 392000, Russia
[4] IMPA, Jardim Bot, Estr Dona Castorina 110, BR-22460320 Rio De Janeiro, RJ, Brazil
基金
俄罗斯基础研究基金会; 俄罗斯科学基金会;
关键词
Newton method; Critical solutions; 2-Regularity; Levenberg-Marquardt method; Linear-programming-Newton method; Stabilized sequential quadratic programming; LIPSCHITZIAN DERIVATIVES; NONISOLATED SOLUTIONS; CRITICAL MULTIPLIERS; ERROR-BOUNDS; SYSTEMS; CONVERGENCE; STABILITY; MAPPINGS;
D O I
10.1007/s10107-017-1128-5
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
We show that if the equation mapping is 2-regular at a solution in some nonzero direction in the null space of its Jacobian (in which case this solution is critical; in particular, the local Lipschitzian error bound does not hold), then this direction defines a star-like domain with nonempty interior from which the iterates generated by a certain class of Newton-type methods necessarily converge to the solution in question. This is despite the solution being degenerate, and possibly non-isolated (so that there are other solutions nearby). In this sense, Newtonian iterates are attracted to the specific (critical) solution. Those results are related to the ones due to A. Griewank for the basic Newton method but are also applicable, for example, to some methods developed specially for tackling the case of potentially non-isolated solutions, including the Levenberg-Marquardt and the LP-Newton methods for equations, and the stabilized sequential quadratic programming for optimization.
引用
收藏
页码:355 / 379
页数:25
相关论文
共 28 条
[1]  
[Anonymous], 2005, COMP MATH MATH PHYS+
[2]  
[Anonymous], SPRINGER SERIES OPER
[3]  
Arutyunov A.V., 2000, Optimality condition: abnormal and degenerate problems
[4]   EXTREMUM CONDITIONS FOR SMOOTH PROBLEMS WITH EQUALITY-TYPE CONSTRAINTS [J].
AVAKOV, ER .
USSR COMPUTATIONAL MATHEMATICS AND MATHEMATICAL PHYSICS, 1985, 25 (03) :24-32
[5]  
Clarke F., 1983, OPTIMIZATION NONSMOO
[6]   An LP-Newton method: nonsmooth equations, KKT systems, and nonisolated solutions [J].
Facchinei, Francisco ;
Fischer, Andreas ;
Herrich, Markus .
MATHEMATICAL PROGRAMMING, 2014, 146 (1-2) :1-36
[7]   A family of Newton methods for nonsmooth constrained systems with nonisolated solutions [J].
Facchinei, Francisco ;
Fischer, Andreas ;
Herrich, Markus .
MATHEMATICAL METHODS OF OPERATIONS RESEARCH, 2013, 77 (03) :433-443
[8]   Stabilized sequential quadratic programming for optimization and a stabilized Newton-type method for variational problems [J].
Fernandez, Damian ;
Solodov, Mikhail .
MATHEMATICAL PROGRAMMING, 2010, 125 (01) :47-73
[9]   Convergence conditions for Newton-type methods applied to complementarity systems with nonisolated solutions [J].
Fischer, Andreas ;
Herrich, Markus ;
Izmailov, Alexey F. ;
Solodov, Mikhail V. .
COMPUTATIONAL OPTIMIZATION AND APPLICATIONS, 2016, 63 (02) :425-459
[10]   On computation of limiting coderivatives of the normal-cone mapping to inequality systems and their applications [J].
Gfrerer, Helmut ;
Outrata, Jiri V. .
OPTIMIZATION, 2016, 65 (04) :671-700