## Analysis of Inexact Trust-Region SQP Algorithms (2000)

### Cached

### Download Links

Venue: | RICE UNIVERSITY, DEPARTMENT OF |

Citations: | 17 - 2 self |

### BibTeX

@TECHREPORT{Heinkenschloss00analysisof,

author = {Matthias Heinkenschloss and Luís N. Vicente},

title = {Analysis of Inexact Trust-Region SQP Algorithms},

institution = {RICE UNIVERSITY, DEPARTMENT OF},

year = {2000}

}

### OpenURL

### Abstract

In this paper we extend the design of a class of composite-step trust-region SQP methods and their global convergence analysis to allow inexact problem information. The inexact problem information can result from iterative linear systems solves within the trust-region SQP method or from approximations of first-order derivatives. Accuracy requirements in our trust-region SQP methods are adjusted based on feasibility and optimality of the iterates. Our accuracy requirements are stated in general terms, but we show how they can be enforced using information that is already available in matrix-free implementations of SQP methods. In the absence of inexactness our global convergence theory is equal to that of Dennis, El-Alem, Maciel (SIAM J. Optim., 7 (1997), pp. 177--207). If all iterates are feasible, i.e., if all iterates satisfy the equality constraints, then our results are related to the known convergence analyses for trust-region methods with inexact gradient information fo...

### Citations

1323 |
GMRES: A generalized minimal residual algorithm for solving nonsymmetric linear systems
- Saad, Schultz
- 1986
(Show Context)
Citation Context ...t may be more efficient to use methods that avoid the use . In this ����� case one can apply ���������� nonsymmetric Krylov subspace methods based on minimum residual approximations, such as GMRES( ) =-=[30]-=-. In the context of nonlinear system solving the use of such methods is described e.g. in [2]. In that context, trust–region subproblems of the type (3.3) also have to be solved and the solvers in [2]... |

361 |
Iterative Methods for Linear and Nonlinear Equations
- Kelley
- 1995
(Show Context)
Citation Context ...convergence theories and implementations. In the context of Newton methods for nonlinear equations and unconstrained optimization, the control of inexactness is relatively well understood. See, e.g., =-=[2, 7, 12, 13, 14, 25]-=-. Generalizations of the inexact Newton method concepts to the local convergence analysis of inexact SQP methods can be found, e.g., in [8, 9, 15, 22, 26]. In [23] global convergence of line-search re... |

172 | An interior trust region approach for nonlinear minimization subject to bounds - Coleman, Li - 1996 |

126 |
The conjugate gradient method and trust regions in large scale optimization
- Steihaug
- 1983
(Show Context)
Citation Context ...mponent of the tangential component. An approximate solution of (4.1) that satisfies (4.6) can be computed, e.g., using the conjugate gradient (cg) method with a modification as suggested by Steihaug =-=[33]-=- and Toint [34]. Here the cg method with starting value �is applied to the minimization of��. The conjugate gradient method is stopped if an approximate minimum of the If������ quadratic model��is rea... |

107 | Optimum Aerodynamic Design Using the Navier-Stokes Equations, Theoret
- Jameson, Martinelli, et al.
- 1998
(Show Context)
Citation Context ...mation. Our research is motivated by discretized optimal control problems [16, 18, 21], parameter identification problems and inverse problems [28, 31], and s.t.���� �� ��or�� ��� design optimization =-=[4, 24]-=-. In these applications represents the discretized state variables and represents the discretized controls, parameters, or design variables, respectively, and the nonlinear constraint���is the discret... |

101 | Trust-Region Methods - Conn, Gould, et al. - 2000 |

84 |
Globally convergent inexact Newton methods
- Eisenstat, Walker
- 1994
(Show Context)
Citation Context ...convergence theories and implementations. In the context of Newton methods for nonlinear equations and unconstrained optimization, the control of inexactness is relatively well understood. See, e.g., =-=[2, 7, 12, 13, 14, 25]-=-. Generalizations of the inexact Newton method concepts to the local convergence analysis of inexact SQP methods can be found, e.g., in [8, 9, 15, 22, 26]. In [23] global convergence of line-search re... |

78 |
Recent developments in algorithms and software for trust region m ethods
- Moré
- 1983
(Show Context)
Citation Context ...f (2.4) also gives the standard���global convergence result for the unconstrained problem (2.1). This may be seen using the proof in [27, Th. 4.10] and applying (4.4) in the estimate ��on page 278 of =-=[27]-=-. The approximate reduced Hessian has to satisfy �� �� ���������������������� (4.5) for some��independent of�. If���is evaluated exactly, then (4.5) is implied by assumption A.4. The approximate solut... |

54 |
U.: Quadratically constrained least squares and quadratic problems
- Golub, Matt
- 1991
(Show Context)
Citation Context ...sfies the fraction of Cauchy decrease condition ���� ���� ��������� ��� ����� (5.1) then a result due to Powell [29, Th. 4] (see also [5, 6.3], [27, L. 4.8]) shows that (3.4) is satisfied. The papers =-=[17]-=-, [32] describe two iterative methods based on Krylov subspaces for the computation of steps �satisfying where is the solution of (3.3). In particular these steps also satisfy (3.4). The iterative met... |

51 | Convergence properties of a class of minimization algorithms, Nonlinear programming - Powell - 1974 |

49 |
Towards an efficient sparsity exploiting Newton method for minimization. Sparse matrices and their uses
- Toint
- 1981
(Show Context)
Citation Context ...tangential component. An approximate solution of (4.1) that satisfies (4.6) can be computed, e.g., using the conjugate gradient (cg) method with a modification as suggested by Steihaug [33] and Toint =-=[34]-=-. Here the cg method with starting value �is applied to the minimization of��. The conjugate gradient method is stopped if an approximate minimum of the If������ quadratic model��is reached, if negati... |

48 | Trust-Region Interior-Point Algorithms for a Class of Nonlinear Programming Problems
- VICENTE
- 1996
(Show Context)
Citation Context ...crease faster. Since too large penalty parameters��can slow down the convergence of SQP methods this is another benefit of our accuracy requirement. Our initial and somewhat straight forward approach =-=[20, 36]-=- to deal with inaccuracy did not use the split���� ���� ��� ����� ���������� ��� ���� ������. Rather, the predicted decrease was defined by (3.10). After determination of�satisfying (3.4) we computed ... |

42 | A global convergence theory for general trust–region–based algorithms for equality constrained optimization
- DENNIS, EL-ALEM, et al.
- 1997
(Show Context)
Citation Context ...antities that are readily available in our algorithms. �� We give a global convergence analysis of a class of composite–step trust–region SQP algorithms for (1.1), which are reviewed in [5, 15.4] and =-=[10, 4]-=-. In the absence of inexactness our global convergence theory is that of [10]. If all iterates are feasible, i.e., if all iterates satisfy�����, then our results are related to the convergence analyse... |

35 | TRICE: Trust-region interior-point SQP algorithms for optimal control and engineering design problems. http://www.caam.rice.edu/~trice
- Dennis, Heinkenschloss, et al.
(Show Context)
Citation Context ...s is useful. Some extensions are rather straight forward, although tedious. For example, we believe our analysis can be generalized to the affine–scaling interior–point trust–region SQP algorithms in =-=[11]-=-, which tackle problems (1.1) with additional simple bounds on . In fact, the predecessor [20] of this paper contains many of the technical details of such an extension, although the assumptions on th... |

33 | A reduced Hessian method for large-scale constrained optimization
- Biegler, Nocedal, et al.
- 1995
(Show Context)
Citation Context ...in (3.7) is replaced by the reduced Hessian approximation��and the term��is approximated. The details of the latter approximation are not important �� in �� our global analysis and we refer to, e.g., =-=[1]-=- for more details. The tangential step does not need to solve (3.5) or (3.7) exactly. It is sufficient that the tangential component �satisfies a fraction of Cauchy decrease condition associated with ... |

28 |
On the global convergence of trust region algorithms using inexact gradient information
- Carter
- 1991
(Show Context)
Citation Context ...the absence of inexactness our global convergence theory is that of [10]. If all iterates are feasible, i.e., if all iterates satisfy�����, then our results are related to the convergence analyses in =-=[3, 5]-=- for trust–region methods with inexact function and gradient information for unconstrained optimization. This paper is organized as follows. In section 2 we will consider the reduced problem �obtained... |

27 |
Convergence theory of nonlinear Newton-Krylov algorithms
- Brown, Saad
- 1994
(Show Context)
Citation Context ...convergence theories and implementations. In the context of Newton methods for nonlinear equations and unconstrained optimization, the control of inexactness is relatively well understood. See, e.g., =-=[2, 7, 12, 13, 14, 25]-=-. Generalizations of the inexact Newton method concepts to the local convergence analysis of inexact SQP methods can be found, e.g., in [8, 9, 15, 22, 26]. In [23] global convergence of line-search re... |

23 |
Augmented Lagrangian–SQP methods for nonlinear optimal control problems of tracking type
- ITO, KUNISCH
- 1996
(Show Context)
Citation Context ...are at least twice continuously differentiable. Variants of the algorithms, however, require only first–order derivative information. Our research is motivated by discretized optimal control problems =-=[16, 18, 21]-=-, parameter identification problems and inverse problems [28, 31], and s.t.���� �� ��or�� ��� design optimization [4, 24]. In these applications represents the discretized state variables and represen... |

19 | Global inexact Newton methods for very large scale nonlinear problems
- DEUFLHARD
- 1991
(Show Context)
Citation Context |

19 |
Projected sequential quadratic programming methods
- HEINKENSCHLOSS
- 1996
(Show Context)
Citation Context ...ion of���is treated as a black–box in the optimization algorithms for (2.1). It can be shown that�� � � ���� ����� � ��������� �� ����� (2.2) where (2.3) and� solves� � �� � �. For details see, e.g., =-=[11, 19]-=-. Now, suppose that the nonlinear equations���� ���� �� � can not be solved exactly for �,but that an approximation��of�� �is computed by applying an iterative method to����. In this case the function... |

19 |
Optimal control and optimization of viscous, incompressible flows
- GUNZBURGER, HOU, et al.
- 1993
(Show Context)
Citation Context ...are at least twice continuously differentiable. Variants of the algorithms, however, require only first–order derivative information. Our research is motivated by discretized optimal control problems =-=[16, 18, 21]-=-, parameter identification problems and inverse problems [28, 31], and s.t.���� �� ��or�� ��� design optimization [4, 24]. In these applications represents the discretized state variables and represen... |

18 |
Minimization of a large scale quadratic function subject to an spherical constraint
- SORENSEN
- 1997
(Show Context)
Citation Context ...the fraction of Cauchy decrease condition ���� ���� ��������� ��� ����� (5.1) then a result due to Powell [29, Th. 4] (see also [5, 6.3], [27, L. 4.8]) shows that (3.4) is satisfied. The papers [17], =-=[32]-=- describe two iterative methods based on Krylov subspaces for the computation of steps �satisfying where is the solution of (3.3). In particular these steps also satisfy (3.4). The iterative method in... |

18 |
Optimal control of two– and three–dimensional Navier–Stokes flow
- GHATTAS, BARK
- 1997
(Show Context)
Citation Context ...are at least twice continuously differentiable. Variants of the algorithms, however, require only first–order derivative information. Our research is motivated by discretized optimal control problems =-=[16, 18, 21]-=-, parameter identification problems and inverse problems [28, 31], and s.t.���� �� ��or�� ��� design optimization [4, 24]. In these applications represents the discretized state variables and represen... |

12 |
Distributed Parameter Systems. Theory and Applications
- OMATU, SEINFELD
- 1989
(Show Context)
Citation Context ...ithms, however, require only first–order derivative information. Our research is motivated by discretized optimal control problems [16, 18, 21], parameter identification problems and inverse problems =-=[28, 31]-=-, and s.t.���� �� ��or�� ��� design optimization [4, 24]. In these applications represents the discretized state variables and represents the discretized controls, parameters, or design variables, res... |

11 | Analysis of inexact trust-region interior-point SQP algorithms
- HEINKENSCHLOSS, VICENTE
- 1995
(Show Context)
Citation Context ...crease faster. Since too large penalty parameters��can slow down the convergence of SQP methods this is another benefit of our accuracy requirement. Our initial and somewhat straight forward approach =-=[20, 36]-=- to deal with inaccuracy did not use the split���� ���� ��� ����� ���������� ��� ���� ������. Rather, the predicted decrease was defined by (3.10). After determination of�satisfying (3.4) we computed ... |

11 |
Inaccuracy in quasi-Newton methods: Local improvement theorems
- DENNIS, WALKER
- 1984
(Show Context)
Citation Context |

10 | Two-step algorithms for nonlinear optimization with structured applications
- Conn, Vicente, et al.
- 1999
(Show Context)
Citation Context ..., we then identify (4.13) as a condition that has already been imposed on steps of such types in the context of global convergence of trust–region algorithms for unconstrained optimization [5, 10.4], =-=[6]-=-. We note that the amount of positivity ����is determined by the reductions ����. Thus we can allow the more inaccuracy in the �computation, which typically translates into less expensive �computation... |

9 |
Inexact SQP interior point methods and large scale optimal control problems
- Leibfritz, Sachs
- 1999
(Show Context)
Citation Context ...is relatively well understood. See, e.g., [2, 7, 12, 13, 14, 25]. Generalizations of the inexact Newton method concepts to the local convergence analysis of inexact SQP methods can be found, e.g., in =-=[8, 9, 15, 22, 26]-=-. In [23] global convergence of line-search reduced SQP methods is studied. The influence of inexact problem information on the global convergence of trust–region SQP methods, however, is to our knowl... |

7 |
Inexact primal–dual interior–point iteration for linear programs in function spaces
- ITO, KELLEY, et al.
- 1995
(Show Context)
Citation Context ...is relatively well understood. See, e.g., [2, 7, 12, 13, 14, 25]. Generalizations of the inexact Newton method concepts to the local convergence analysis of inexact SQP methods can be found, e.g., in =-=[8, 9, 15, 22, 26]-=-. In [23] global convergence of line-search reduced SQP methods is studied. The influence of inexact problem information on the global convergence of trust–region SQP methods, however, is to our knowl... |

6 | Recent developments in algorithms and software for trust region methods - E - 1983 |

5 |
On inexact quasi–Newton methods for constrained optimization, in Numerical Optimization
- FONTECILLA
- 1984
(Show Context)
Citation Context ...is relatively well understood. See, e.g., [2, 7, 12, 13, 14, 25]. Generalizations of the inexact Newton method concepts to the local convergence analysis of inexact SQP methods can be found, e.g., in =-=[8, 9, 15, 22, 26]-=-. In [23] global convergence of line-search reduced SQP methods is studied. The influence of inexact problem information on the global convergence of trust–region SQP methods, however, is to our knowl... |

5 |
Airfoil design by an all–at–once method
- CLIFF, HEINKENSCHLOSS, et al.
- 1998
(Show Context)
Citation Context ...mation. Our research is motivated by discretized optimal control problems [16, 18, 21], parameter identification problems and inverse problems [28, 31], and s.t.���� �� ��or�� ��� design optimization =-=[4, 24]-=-. In these applications represents the discretized state variables and represents the discretized controls, parameters, or design variables, respectively, and the nonlinear constraint���is the discret... |

4 |
Global convergence of inexact reduced SQP-methods
- JÄGER, SACHS
- 1997
(Show Context)
Citation Context ...rstood. See, e.g., [2, 7, 12, 13, 14, 25]. Generalizations of the inexact Newton method concepts to the local convergence analysis of inexact SQP methods can be found, e.g., in [8, 9, 15, 22, 26]. In =-=[23]-=- global convergence of line-search reduced SQP methods is studied. The influence of inexact problem information on the global convergence of trust–region SQP methods, however, is to our knowledge not ... |

3 |
Multigrid optimization methods for stationary parameter identification problems in groundwater flow
- SCHULZ, WITTUM
- 1997
(Show Context)
Citation Context ...ithms, however, require only first–order derivative information. Our research is motivated by discretized optimal control problems [16, 18, 21], parameter identification problems and inverse problems =-=[28, 31]-=-, and s.t.���� �� ��or�� ��� design optimization [4, 24]. In these applications represents the discretized state variables and represents the discretized controls, parameters, or design variables, res... |

2 |
Local convergence analysis for successive inexact quadratic programming methods
- DEMBO, TULOWITZKI
- 1984
(Show Context)
Citation Context |

2 | Convergence theory of nonlinear Netwon–Krylov algorithms - BROWN, SAAD - 1994 |

2 |
convergence of a class of trust-region methods for nonconvex minimization in Hilbert space
- Global
- 1988
(Show Context)
Citation Context ... gradient. How does one need to control the inexactness in function values and gradients in trust–region methods for (2.1)? The influence of inexact gradient information is analyzed in [3], [5, 8.4], =-=[35]-=- ������ � ��� (for a detailed ������� ��� literature review see [5, p. 296]) and the influence of inexact function evaluations is studied in [5, 10.6]. We want to ensure that our inexactness assumptio... |

1 |
truncated quadratic programming, in Numerical Optimization
- Sequential
- 1984
(Show Context)
Citation Context |

1 | VICENTE,Trust--Region Interior--Point Algorithms for a Class of Nonlinear Programming Problems - N - 1996 |