## Numerical Optimal Control Of Parabolic PDEs Using DASOPT (1997)

Citations: | 11 - 6 self |

### BibTeX

@MISC{Petzold97numericaloptimal,

author = {Linda Petzold and J. Ben Rosen and PHILIP E. GILL and Laurent O. Jay and KIHONG PARK},

title = {Numerical Optimal Control Of Parabolic PDEs Using DASOPT},

year = {1997}

}

### OpenURL

### Abstract

. This paper gives a preliminary description of DASOPT, a software system for the optimal control of processes described by time-dependent partial differential equations (PDEs). DASOPT combines the use of efficient numerical methods for solving differential-algebraic equations (DAEs) with a package for large-scale optimization based on sequential quadratic programming (SQP). DASOPT is intended for the computation of the optimal control of time-dependent nonlinear systems of PDEs in two (and eventually three) spatial dimensions, including possible inequality constraints on the state variables. By the use of either finite-difference or finite-element approximations to the spatial derivatives, the PDEs are converted into a large system of ODEs or DAEs. Special techniques are needed in order to solve this very large optimal control problem. The use of DASOPT is illustrated by its application to a nonlinear parabolic PDE boundary control problem in two spatial dimensions. Computational resu...

### Citations

1341 |
Practical Optimization
- Gill, Murray, et al.
- 1981
(Show Context)
Citation Context ...bproblem is itself an iterative procedure, with the minor iterations of an SQP method being the iterations of the QP method. (For an overview of SQP methods, see, for example, Gill, Murray and Wright =-=[17].) Each -=-QP subproblem minimizes a quadratic model of the modified Lagrangian L(p; p k ; �� k ) = `(p) \Gamma �� T k d L (p; p k ); (3.1) which is defined in terms of the constraint linearization, r L ... |

327 | SNOPT: an SQP algorithm for large-scale constrained optimization, Numerical Analysis
- Gill, Murray, et al.
- 1997
(Show Context)
Citation Context ... ! + ff k / b p k \Gamma p k b �� k \Gamma �� k ! ; where ff k is found from a line search that enforces a sufficient decrease in an augmented Lagrangian merit function (see Gill, Murray and S=-=aunders [15]-=-). In this SQP formulation, the objective and constraint derivatives r` and J are required once each major iteration. They are needed to define the objective and constraints of the QP subproblem. The ... |

253 |
Namerical Solution of Boundary Value Problems for Ordinanj Diflerential Equations. SIAM
- Ascher, Mattheij, et al.
- 1995
(Show Context)
Citation Context ...E software, and when it converges, it can be very efficient. However, it is well-known that single shooting can suffer from a lack of robustness and stability (see, e.g., Ascher, Mattheij and Russell =-=[1]-=-). For some nonlinear problems it can generate intermediate iterates that are nonphysical and/or not computable. For some well-conditioned boundary-value problems, it can generate unstable initial-val... |

214 |
Updating quasi-newton matrices with limited storage
- Nocedal
- 1980
(Show Context)
Citation Context ...ssigned fixed number (say `) of these updates are stored and products H k u are computed using O(`) inner-products. For a discussion of limited-memory methods see, e.g., Gill and Murray [14], Nocedal =-=[26]-=-), Buckley and LeNir [8], and Gilbert and Lemar'echal [13]. Currently, SNOPT uses a simple limited-memory implementation of the BFGS quasi-Newton method. As the iterations proceed, the two vectors (q ... |

149 | Generating derivative codes from Fortran programs
- Bischof, Carle, et al.
- 1991
(Show Context)
Citation Context ...can specify directly the residual of the sensitivity system at the same time as the residual of the original system. Eventually, we intend to incorporate the automatic differentiation software ADIFOR =-=[5]-=- for this purpose. Alternatively, we can approximate the right-hand side of the sensitivity equations using a directional derivative finite-difference approximation. As an example, define s i = dv=dp ... |

83 |
Direct Trajectory optimization Using Nonlinear Programming and Collocation
- Hargraves, Paris
- 1987
(Show Context)
Citation Context ...val is divided into subintervals in a multiple-shooting type approach that provides a source of parallelism. (For other approaches, see, e.g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris =-=[19]-=-, Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], ... |

76 |
Minos 5.4 user's guide
- Murtagh, Saunders
- 1993
(Show Context)
Citation Context ...give an approximate minimizer with a reduced-gradient norm less than 10 \Gamma5 and a maximum nonlinear constraint violation less than 10 \Gamma6 (for further details of the termination criteria, see =-=[25]-=-). The default maximum number of limited memory updates is 20. The last four columns in Table 6.1 give the results of the computation. The minimum value of the objective function OE, scaled by 10 5 , ... |

60 |
Some numerical experiments with variable-storage quasi-Newton algorithms
- Gilbert, Lemaréchal
- 1989
(Show Context)
Citation Context ...nd products H k u are computed using O(`) inner-products. For a discussion of limited-memory methods see, e.g., Gill and Murray [14], Nocedal [26]), Buckley and LeNir [8], and Gilbert and Lemar'echal =-=[13]-=-. Currently, SNOPT uses a simple limited-memory implementation of the BFGS quasi-Newton method. As the iterations proceed, the two vectors (q k ; y k ) defining the current update are added to an expa... |

56 |
Direct and indirect methods for trajectory optimisation
- Stryk, Bulirsch
- 1992
(Show Context)
Citation Context ...ovides a source of parallelism. (For other approaches, see, e.g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch =-=[35]-=-, Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis [27], and Gritsis, Pantelides and Sargent... |

32 | A reduced Hessian method for large–scale constrained optimization
- BIEGLER, NOCEDAL, et al.
- 1995
(Show Context)
Citation Context ... problems. Much of this work has focused on reducedHessian methods, which maintain a dense quasi-Newton approximation to a smaller dimensional reduced Hessian (see, e.g., Biegler, Nocedal and Schmidt =-=[4]-=-, Eldersveld [12], Tjoa and Biegler [33], and Schultz [29]). Our preference for approximating the full Hessian is motivated by substantial improvements in reliability and efficiency compared to earlie... |

31 | Josef: Real-Time Computation of Feedback Controls for Constrained Optimal Control Problems. Part 2: A Correction Method Based on Multiple Shooting
- Pesch
- 1989
(Show Context)
Citation Context ...ed into subintervals in a multiple-shooting type approach that provides a source of parallelism. (For other approaches, see, e.g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris [19], Pesch =-=[28]-=-, Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, ... |

30 |
Maintaining LU factors of a general sparse matrix
- Gill, Murray, et al.
- 1986
(Show Context)
Citation Context ...systems formed from rows and columns of the structured derivatives. In the current version of SNOPT, these sparse systems are solved using the general-purpose sparse LU package LUSOL (see Gill et al. =-=[16]-=-). Current research is directed towards other factorization methods that more fully exploit the block-diagonal structure of the derivatives (see, e.g., Steinbach [31]). SQP methods are most robust whe... |

22 | Numerical methods and software for sensitivity analysis of differential-algebraic systems
- Maly, Petzold
- 1996
(Show Context)
Citation Context ...ly with the original system, yielding a nonlinear system at each time step. We will outline the algorithms here; further details on the algorithms, codes, theory and numerical results can be found in =-=[24]-=-. The new codes are easy to use, highly efficient, and well-suited for large-scale problems. First, we briefly give some background on the algorithms in DASSL and DASPK. Further details can be found i... |

19 | M.: Exploiting invariants in the numerical solution of multipoint boundary value problems for DAEs
- Schulz, Bock, et al.
- 1998
(Show Context)
Citation Context ...aft [20], Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach =-=[30]-=-, Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis [27], and Gritsis, Pantelides and Sargent [18].) The associated finite-dimensional optimization problem is characterized by: (a) many ... |

17 |
On converting optimal control problems into nonlinear programming problems
- Kraft
- 1985
(Show Context)
Citation Context ...m. The original time interval is divided into subintervals in a multiple-shooting type approach that provides a source of parallelism. (For other approaches, see, e.g., Dickmanns and Well [11], Kraft =-=[20]-=-, Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Ta... |

15 |
QN-like variable storage conjugate gradients
- Buckley, Lenir
- 1983
(Show Context)
Citation Context ... `) of these updates are stored and products H k u are computed using O(`) inner-products. For a discussion of limited-memory methods see, e.g., Gill and Murray [14], Nocedal [26]), Buckley and LeNir =-=[8]-=-, and Gilbert and Lemar'echal [13]. Currently, SNOPT uses a simple limited-memory implementation of the BFGS quasi-Newton method. As the iterations proceed, the two vectors (q k ; y k ) defining the c... |

15 |
Large-Scale Sequential Quadratic Programming Algorithms
- Eldersveld
- 1991
(Show Context)
Citation Context ...of this work has focused on reducedHessian methods, which maintain a dense quasi-Newton approximation to a smaller dimensional reduced Hessian (see, e.g., Biegler, Nocedal and Schmidt [4], Eldersveld =-=[12]-=-, Tjoa and Biegler [33], and Schultz [29]). Our preference for approximating the full Hessian is motivated by substantial improvements in reliability and efficiency compared to earlier versions of SNO... |

15 |
Conjugate-gradient methods for largescale nonlinear optimization, report sol 79-15, department of operations research
- Gill, Murray
- 1979
(Show Context)
Citation Context ...proach, a preassigned fixed number (say `) of these updates are stored and products H k u are computed using O(`) inner-products. For a discussion of limited-memory methods see, e.g., Gill and Murray =-=[14]-=-, Nocedal [26]), Buckley and LeNir [8], and Gilbert and Lemar'echal [13]. Currently, SNOPT uses a simple limited-memory implementation of the BFGS quasi-Newton method. As the iterations proceed, the t... |

15 | Structured interior point SQP methods in optimal control, Zeitschrift für Angewandte Mathematik und Mechanik
- Steinbach
- 1996
(Show Context)
Citation Context ...LU package LUSOL (see Gill et al. [16]). Current research is directed towards other factorization methods that more fully exploit the block-diagonal structure of the derivatives (see, e.g., Steinbach =-=[31]-=-). SQP methods are most robust when the derivatives of the objective and constraint functions are computed exactly. As described in x4, the function and derivative computations involve computing the s... |

14 |
Sensitivity analysis of initial value problems with mixed ODES and algebraic equations, Computers and Chemical Engineering
- Caracotsios, Stewart
- 1985
(Show Context)
Citation Context ... C C C C C C A ; (4.4) with J = 1 h @F @v 0 + @F @v and J i = @J @v s i + @J @p i : A number of codes for ODEs and DAEs solve the sensitivity system (4.2), or its special case for ODEs, directly (see =-=[10]-=-). If the partial derivative matrices are not available analytically, they are approximated by finite differences. The nonlinear system is usually solved by a so-called staggered scheme, in which the ... |

13 |
Application of sparse nonlinear programming to trajectory optimization
- Huffman
- 1992
(Show Context)
Citation Context ...ooting type approach that provides a source of parallelism. (For other approaches, see, e.g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman =-=[3]-=-, von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis [27], and Gr... |

12 |
Stryk, Combining direct and indirect methods in nonlinear optimal control: Range maximization of a hang glider
- Bulirsch, Nerz, et al.
- 1993
(Show Context)
Citation Context ...allelism. (For other approaches, see, e.g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. =-=[9]-=-, von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis [27], and Gritsis, Pantelides and Sargent [18].) The associate... |

8 |
Numerical solution of parabolic state constrained control problems using SQP-- 46
- Leibfritz, Sachs
- 1994
(Show Context)
Citation Context ...ed a boundary control heating problem in two spatial dimensions. This model problem is described by a nonlinear parabolic PDE. It is a two-dimensional generalization of the model problem described in =-=[22]-=-. A rectangular domain in space is heated by controlling the temperature on its boundaries. It is desired that the transient temperature in a specified interior subdomain follow a prescribed temperatu... |

8 |
Stable decomposition for dynamic optimization
- Tanartkit, Biegler
- 1995
(Show Context)
Citation Context ...s [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler =-=[32]-=-, Pantelides, Sargent and Vassiliadis [27], and Gritsis, Pantelides and Sargent [18].) The associated finite-dimensional optimization problem is characterized by: (a) many variables and constraints; (... |

8 |
Stryk, “Numerical solution of optimal control problems by direct collocation
- von
- 1993
(Show Context)
Citation Context ...other approaches, see, e.g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk =-=[34]-=-, Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis [27], and Gritsis, Pantelides and Sargent [18].) The associated finite-dimensi... |

5 |
Approximate Solution of Optimal Control Problems Using Third-Order Hermite Polynomial Functions
- Dickmanns, Well
- 1975
(Show Context)
Citation Context ...ntrol problem. The original time interval is divided into subintervals in a multiple-shooting type approach that provides a source of parallelism. (For other approaches, see, e.g., Dickmanns and Well =-=[11]-=-, Kraft [20], Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinb... |

4 |
Differential-algebraic equations issues in the direct transcription of path constrained optimal control problems
- Brenan
- 1994
(Show Context)
Citation Context ....g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan =-=[6]-=-, Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis [27], and Gritsis, Pantelides and Sargent [18].) The associated finite-dimensional optimization probl... |

4 |
Optimal control of multistage systems described by high-index differential-algebraic equations
- Pantelides, Sargent, et al.
- 1994
(Show Context)
Citation Context ... Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis =-=[27]-=-, and Gritsis, Pantelides and Sargent [18].) The associated finite-dimensional optimization problem is characterized by: (a) many variables and constraints; (b) sparse constraint and objective derivat... |

3 |
Direct transcription of optimal control problems with high order state constraints: Theory vs. practice
- Betts, Campbell, et al.
(Show Context)
Citation Context ...ches, see, e.g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris [19], Pesch [28], Lamour [21], Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts =-=[2]-=-, Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis [27], and Gritsis, Pantelides and Sargent [18].) The associated finite-dimensional optimi... |

3 |
Optimal control of systems described by index two differential-algebraic equations
- Gritsis, Pantelides, et al.
- 1995
(Show Context)
Citation Context ... Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and Vassiliadis [27], and Gritsis, Pantelides and Sargent =-=[18]-=-.) The associated finite-dimensional optimization problem is characterized by: (a) many variables and constraints; (b) sparse constraint and objective derivatives; and (c) many constraints active at t... |

3 |
The Calculus of Variations and Optimal
- Leitmann
- 1981
(Show Context)
Citation Context ...iscussed, for example, in Brenan, Campbell and Petzold [7]. We also assume that the control u(t) satisfies some standard conditions needed for the existence of an optimal control (see, e.g., Leitmann =-=[23]-=-). For simplicity of presentation, we assume that v 0 is given and fixed. However, there is no difficulty in treating v 0 as a vector of parameters to be determined by the optimization. Note also that... |

3 |
Reduced SQP methods for large-scale optimal control problems in DAE with application to path planning problems for satellite mounted robots
- Schultz
- 1996
(Show Context)
Citation Context ...n methods, which maintain a dense quasi-Newton approximation to a smaller dimensional reduced Hessian (see, e.g., Biegler, Nocedal and Schmidt [4], Eldersveld [12], Tjoa and Biegler [33], and Schultz =-=[29]-=-). Our preference for approximating the full Hessian is motivated by substantial improvements in reliability and efficiency compared to earlier versions of SNOPT based on the reduced-Hessian approach.... |

2 |
A well posed shooting method for transferable DAEs
- Lamour
- 1991
(Show Context)
Citation Context ...tervals in a multiple-shooting type approach that provides a source of parallelism. (For other approaches, see, e.g., Dickmanns and Well [11], Kraft [20], Hargraves and Paris [19], Pesch [28], Lamour =-=[21]-=-, Betts and Huffman [3], von Stryk and Bulirsch [35], Bulirsch et al. [9], von Stryk [34], Betts [2], Brenan [6], Schulz, Bock and Steinbach [30], Tanartkit and Biegler [32], Pantelides, Sargent and V... |

2 |
A reduced SQP strategy for errors-in-variables estimation
- Tjoa, Biegler
- 1992
(Show Context)
Citation Context ...d on reducedHessian methods, which maintain a dense quasi-Newton approximation to a smaller dimensional reduced Hessian (see, e.g., Biegler, Nocedal and Schmidt [4], Eldersveld [12], Tjoa and Biegler =-=[33]-=-, and Schultz [29]). Our preference for approximating the full Hessian is motivated by substantial improvements in reliability and efficiency compared to earlier versions of SNOPT based on the reduced... |