## Choosing the Forcing Terms in an Inexact Newton Method (1994)

Venue: | SIAM J. Sci. Comput |

Citations: | 109 - 2 self |

### BibTeX

@ARTICLE{Eisenstat94choosingthe,

author = {Stanley C. Eisenstat and Homer F. Walker and Stanley C. Eisenstatt and Homer and F. Walker},

title = {Choosing the Forcing Terms in an Inexact Newton Method},

journal = {SIAM J. Sci. Comput},

year = {1994},

volume = {17},

pages = {16--32}

}

### Years of Citing Articles

### OpenURL

### Abstract

An inexact Newton method is a generalization of Newton's method for solving F(x) = 0, F:/ -/, in which, at the kth iteration, the step sk from the current approximate solution xk is required to satisfy a condition ]lF(x) + F'(x)s]l _< /]lF(xk)]l for a "forcing term" / [0,1). In typical applications, the choice of the forcing terms is critical to the efficiency of the method and can affect robustness as well. Promising choices of the forcing terms arc given, their local convergence properties are analyzed, and their practical performance is shown on a representative set of test problems.

### Citations

691 | Introduction to Numerical Analysis - Stoer, Bulirsch - 1980 |

360 |
Numerical Methods for Unconstrained Optimization and Nonlinear Equations
- SCHNABEL
- 1983
(Show Context)
Citation Context ...the scale of F. 3.1. The algorithm. A globazed inexact Newton algorithm was necessary because initial appromate solutions were not alwws near a solution. We used Algorithm INB of Eisenstat and Walker =-=[7, 6]-=-. This is an inexact Newton method globazed by backtracking, which we write here as foows: Algorithm INB: Inexact Newton Backtracking Method [7] LrT xo, ?max 6 [0, 1), t 6 (0, 1), AD 0 nsFOR k = 0 STE... |

257 |
Inexact Newton methods
- Dembo, Eisenstat, et al.
- 1982
(Show Context)
Citation Context ...continuous at x, with constant A, i.e., (1.1) IIF'(x)- F'(x,)[ I _ for x near x,, where 11' 11 denotes some norm on and the induced norm on x. An inexact Newton method (Dembo, Eisenstat, and Steihaug =-=[4]) is an ex-=-tension of classical Newton's method for approximating x, formulated as follows: Algorithm IN: Inexact Newton Method [4] LET x0 BE GIVEN. FOl k = 0 STEP 1 UNTIL "CoNvERGENCE" DO: FIND some r... |

171 | Hybrid Krylov methods for nonlinear systems of equations - Brown, Saad - 1990 |

99 |
Globally convergent inexact Newton methods
- EISENSTAT, WALKER
- 1994
(Show Context)
Citation Context ...the scale of F. 3.1. The algorithm. A globazed inexact Newton algorithm was necessary because initial appromate solutions were not alwws near a solution. We used Algorithm INB of Eisenstat and Walker =-=[7, 6]-=-. This is an inexact Newton method globazed by backtracking, which we write here as foows: Algorithm INB: Inexact Newton Backtracking Method [7] LrT xo, ?max 6 [0, 1), t 6 (0, 1), AD 0 nsFOR k = 0 STE... |

72 |
GMRES: A generalized minimal residual method for solving nonsymmetric linear systems
- Saad, Schultz
- 1986
(Show Context)
Citation Context ...rst chose each initial ? (with 70 = 1/2 for Choices i and 2) and then determined an initial s by appromately solving the Newton equation using GMRES(m), the restatted GMRES method of Saad and Schultz =-=[12]-=-, with restart value m = 20. Products of Ft(x) with vectors were evaluated analyticay in some cases and appromated by finite differences of F-values in others; see 3.2. When finite-difference ppromtio... |

28 |
Continuation-conjugate gradient methods for the least squares solution of nonlinear boundary value problems
- Glowinski, Keller, et al.
- 1985
(Show Context)
Citation Context ...em is harder. 3.2.2. The (modified) Bratu problem. The problem is 0u Au + nXl + Ae ' = 0 in f, u = 0 on Of. The actual Bratu (or Gelfend) problem has n = 0; see, e.g., G1owinski, Keller, and Reinhart =-=[8]-=- or the description by Glowinski and Keller in the collection of nonlinear model problems assembled by Mor6 [11, pp. 733-737]. As n and ), grow, solving the Newton equations for the discretized proble... |

20 | Newton-Krylov-Schwarz methods - CAI, GROPP, et al. - 1994 |

18 |
Truncated Newton algorithms for large-scale unconstrained optimization
- Dembo, Steihaug
- 1983
(Show Context)
Citation Context ...d dlows relatively inaccurate ppromtions of Newton steps for sm k, whensmy not be near ,; however, it incorporates no information bout F. 4. The choices= min(1/(k + 2),llF(x)ll) of Dembo and Steihaug =-=[5]-=-. This choice results in q-quadratic local convergence and also mw aow relatively inaccurate appromations of Newton steps for sma k. It incorporates some information about F; however, it does not refl... |

18 |
Ecient high accuracy solutions with GMRES(m
- TURNER, WALKER
- 1992
(Show Context)
Citation Context ...were used within the cycle. This selective second-order differencing gave essentiay the same accuracy as if central differences had been used throughout, but at much lower cost (see Turner and Walker =-=[16]-=-). The parameters used were ?x = .9, t = 10-4, 0n = 1/10, and 0x = 1/2. The om w he Eucde om I1' I1. I he whie4oop, ech 0 w choe o miimiz ov [, Smx] tn qatic p(S) o which p(0) p(1) = (1), wn ()sIIF( +... |

16 |
Algorithm 541: Efficient FORTRAN subprograms for the solution of separable elliptic partial differential equations
- Swartztrauber, Sweet
- 1979
(Show Context)
Citation Context ...he usual centered differences on a 100 x 100 uniform grid, so that n = 10 4. The discretized problem was preconditioned on the right using a fast Poisson solver from FISHPACK (Swartztrauber and Sweet =-=[15]-=-). Products of Fswith vectors were evaluated analytically. The initial approximate solution was a discretization of no(x)s;Xl(1- Xl)X2(1- x2), which should lead to the everywherepositive solution for ... |

11 |
Fast numerical solution of the biharmonic Dirichlet problem on rectangles
- Bjørstad
- 1983
(Show Context)
Citation Context ..., r-qudmticy. However, this possibity ws not s successful in our experiments s the other choices proposed here, nd we do not consider it further. Our second choice is the foowing: Choice 2: Given 7 6 =-=[0, 1]-=-, a 6 (1, 2], and 70 6 [0, 1), choose (llF(xll h = 1,. (')s=skllF(X_l)llJ ' .... The choice (2.6) does not directly reflect the agreement between F and its local near model, as does Choice 1. However,... |

4 |
Private communication
- MCKENNA
- 1992
(Show Context)
Citation Context ...y value problems posed on f = [0, 1] x [0, 1] C_ 2. 3.2.1. A PDE problem. The problem is Au+u 3=0in f, u=0on0ff. This problem has multiple solutions, but only one that is positive everywhere (McKenna =-=[10]-=-, Schaaf [13]). These properties appear to be shared by the discretized problem, and finding the everywhere-positive solution can be difficult without a good initial approximate solution. Discretizati... |

4 |
Large time behaviour of solutions of the porous medium equation with convection
- DUIJN, GRAAF
- 1990
(Show Context)
Citation Context ...O in l2, with u = 1 on the bottom and left sides of fi and u = 0 on the top and right sides. This is more or less a steady-state special case of a general problem considered by van Duijn and de Graaf =-=[17]-=-. Discretization was by the usual centered differences on a 64 x 64 uniform grid, so that n = 4096. The discretized problem was preconditioned on the right using the tridiagonal part of the JacobJan. ... |

4 | A collection of nonlinear model problems, in Computational Solutions of Nonlinear Systems of Equations - MOR - 1990 |

3 | Fast Numerical Solution of the Biharmonic Dirichlet Problem on Rectangles - Bj��rstad - 1983 |

2 |
A pointwise quasi-Newton method for integral equations
- KELLEY, NORTHRUP
- 1988
(Show Context)
Citation Context ...e more backtracking for negative d and to cause more oversolving for positive d. Two test cases were considered: d = 50 and d = -50. 3.2.5. An integral equation. The problem, from Kelley and Northrup =-=[9]-=-, is if01 1 cu(x)2- cos(yu(x))u(y)dy+sinl-c=O, x E [0,1]. Clearly, u(x)s1 is always a solution, and there exist other solutions for at least some values of c. The discretized problem was determined by... |

2 | de Graaf, Large time behaviour of solutions of the porous medium equation with convection - Duijn, M - 1990 |

1 | A collection of nonlinear model problems, in Computational Solution of Nonlinear Systems of Equations - MORI - 1990 |

1 |
Private communication
- SCHAAF
- 1994
(Show Context)
Citation Context ...ems posed on f = [0, 1] x [0, 1] C_ 2. 3.2.1. A PDE problem. The problem is Au+u 3=0in f, u=0on0ff. This problem has multiple solutions, but only one that is positive everywhere (McKenna [10], Schaaf =-=[13]-=-). These properties appear to be shared by the discretized problem, and finding the everywhere-positive solution can be difficult without a good initial approximate solution. Discretization was by the... |