1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
|
// Copyright (C) 2004, 2006 International Business Machines and others.
// All Rights Reserved.
// This code is published under the Eclipse Public License.
//
// $Id: IpPDSystemSolver.hpp 1861 2010-12-21 21:34:47Z andreasw $
//
// Authors: Carl Laird, Andreas Waechter IBM 2004-08-13
#ifndef __IPPDSYSTEMSOLVER_HPP__
#define __IPPDSYSTEMSOLVER_HPP__
#include "IpUtils.hpp"
#include "IpSymMatrix.hpp"
#include "IpAlgStrategy.hpp"
#include "IpIteratesVector.hpp"
namespace Ipopt
{
/** Pure Primal Dual System Solver Base Class.
* This is the base class for all derived Primal-Dual System Solver Types.
*
* Here, we understand the primal-dual system as the following linear
* system:
*
* \f$
* \left[\begin{array}{cccccccc}
* W & 0 & J_c^T & J_d^T & -P^x_L & P^x_U & 0 & 0 \\
* 0 & 0 & 0 & -I & 0 & 0 & -P_L^d & P_U^d \\
* J_c & 0 & 0 & 0 & 0 & 0 & 0 & 0\\
* J_d & -I & 0 & 0 & 0 & 0 & 0 & 0\\
* Z_L(P_L^x)^T & 0 & 0 & 0 & Sl^x_L & 0 & 0 & 0\\
* -Z_U(P_U^x)^T & 0 & 0 & 0 & 0 & Sl^x_U & 0 & 0\\
* 0 & V_L(P_L^d)^T & 0 & 0 & 0 & 0 & Sl^s_L & 0 \\
* 0 & -V_U(P_U^d)^T & 0 & 0 & 0 & 0 & 0 & Sl^s_U \\
* \end{array}\right]
* \left(\begin{array}{c}
* sol_x\\ sol_s\\ sol_c\\ sol_d\\ sol^z_L\\ sol^z_U\\ sol^v_L\\
* sol^v_U
* \end{array}\right) =
* \left(\begin{array}{c}
* rhs_x\\ rhs_s\\ rhs_c\\ rhs_d\\ rhs^z_L\\ rhs^z_U\\ rhs^v_L\\
* rhs^v_U
* \end{array}\right)
* \f$
*
* Here, \f$Sl^x_L = (P^x_L)^T x - x_L\f$,
* \f$Sl^x_U = x_U - (P^x_U)^T x\f$, \f$Sl^d_L = (P^d_L)^T d(x) - d_L\f$,
* \f$Sl^d_U = d_U - (P^d_U)^T d(x)\f$. The results returned to the
* caller is \f$res = \alpha * sol + \beta * res\f$.
*
* The solution of this linear system (in order to compute the search
* direction of the algorthim) usually requires a considerable amount of
* computation time. Therefore, it is important to tailor the solution
* of this system to the characteristics of the problem. The purpose of
* this base class is to provide a generic interface to the algorithm
* that it can use whenever it requires a solution of the above system.
* Particular implementation can then be written to provide the methods
* defined here.
*
* It is implicitly assumed here, that the upper left 2 by 2 block
* is possibly modified (implicitly or explicitly) so that its
* projection onto the null space of the overall constraint
* Jacobian \f$\left[\begin{array}{cc}J_c & 0\\J_d &
* -I\end{array}\right]\f$ is positive definite. This is necessary
* to guarantee certain descent properties of the resulting search
* direction. For example, in the full space implementation, a
* multiple of the identity might be added to the upper left 2 by 2
* block.
*
* Note that the Solve method might be called several times for different
* right hand sides, but with identical data. Therefore, if possible,
* an implemetation of PDSystem should check whether the incoming data has
* changed, and not redo factorization etc. unless necessary.
*/
class PDSystemSolver: public AlgorithmStrategyObject
{
public:
/** @name /Destructor */
//@{
/** Default Constructor */
PDSystemSolver()
{}
/** Default destructor */
virtual ~PDSystemSolver()
{}
//@}
/** overloaded from AlgorithmStrategyObject */
virtual bool InitializeImpl(const OptionsList& options,
const std::string& prefix) = 0;
/** Solve the primal dual system, given one right hand side. If
* the flag allow_inexact is set to true, it is not necessary to
* solve the system to best accuracy; for example, we don't want
* iterative refinement during the computation of the second
* order correction. On the other hand, if improve_solution is
* true, the solution given in res should be improved (here beta
* has to be zero, and res is assume to be the solution for the
* system using rhs, without the factor alpha...). THe return
* value is false, if a solution could not be computed (for
* example, when the Hessian regularization parameter becomes too
* large.)
*/
virtual bool Solve(Number alpha,
Number beta,
const IteratesVector& rhs,
IteratesVector& res,
bool allow_inexact=false,
bool improve_solution=false) =0;
private:
/**@name Default Compiler Generated Methods
* (Hidden to avoid implicit creation/calling).
* These methods are not implemented and
* we do not want the compiler to implement
* them for us, so we declare them private
* and do not define them. This ensures that
* they will not be implicitly created/called. */
//@{
/** Overloaded Equals Operator */
PDSystemSolver& operator=(const PDSystemSolver&);
//@}
};
} // namespace Ipopt
#endif
|