OOPSE/libmdtools/NLModel.hpp

#ifndef _NLMODEL_H_
#define _NLMODEL_H_

#include <vector>
#include <utility>
#include <math.h>

#include "SymMatrix.hpp"
#include "Functor.hpp"
#include "ConstraintList.hpp"

using namespace std;


typedef enum  {backward, forward, central} FDType;

// special property of nonlinear object function
typedef enum {linear, quadratic, general} NLOFProp;

//abstract class of nonlinear optimization model
class NLModel{
  public:    
    NLModel(int dim, ConstraintList* cons) { ndim = dim, constraints = cons;}
    virtual ~NLModel() {  if (constraints != NULL) delete constraints;}

    virtual void setX(const vector<double>& x)= 0;
    virtual vector<double> getX() = 0;

    virtual void setF(double f) = 0;
    virtual double getF() = 0;

    virtual int  getDim() {return ndim;}

    bool hasConstraints() {  return constraints == NULL ? false : true;}
    int getConsType() {  return constraints->getConsType();}

    virtual double calcF()  = 0;
    virtual double calcF(vector<double>& x)  = 0;
    virtual vector<double> calcGrad()  = 0;
    virtual vector<double> calcGrad(vector<double>& x)  = 0;
    virtual SymMatrix calcHessian()  = 0;
    virtual SymMatrix calcHessian(vector<double>& x)  = 0;

#ifdef IS_MPI
    //void setMPIINITFunctor(MPIINITFunctor* func);
    //int getLocalDim() {return localDim;}
    
    //virtual void update();  //a hook function to load balancing
#endif

  protected:
    NLModel() {}
    ConstraintList* constraints;       //constraints of nonlinear optimization model
    int numOfFunEval;                  //number of function evaluation
    int ndim;

#ifdef IS_MPI
    bool mpiInitFlag;
    int myRank;                           //rank of current node
    int numOfProc;                      // number of processors
    //MPIINITFunctor * mpiInitFunc;
    
    int localDim;
    vector<int> procMappingArray; 
    int beginGlobalIndex;   
#endif
};

//abstract class of nonlinear optimization model without derivatives
class NLModel0 : public NLModel{
  public:

    NLModel0(int dim,  ConstraintList* cons) : NLModel(dim, cons) { currentX.resize(dim);}
    ~NLModel0() {}

    virtual void setX(const vector<double>& x) {currentX = x;}
    vector<double>  getX() {return currentX;}

    void setF(double f) {currentF = f;}
    double getF() {return currentF;}

    //Using finite difference methods to approximate the gradient
    //It is inappropriate to apply these methods in large scale problem
    
    vector<double> BackwardGrad(const vector<double>& x, double& fx, vector<double>& grad, const vector<double>& h);
    vector<double> ForwardGrad(const vector<double>& x, double& fx, vector<double>& grad, const vector<double>& h);
    vector<double> CentralGrad(const vector<double>& x, double& fx, vector<double>& grad, const vector<double>& h);

    //Using finite difference methods to approximate the hessian
    //It is inappropriate to apply this method in large scale problem
    //virtual SymMatrix FiniteHessian(vector<double>& x, double fx, vector<double>& h);
    SymMatrix FiniteHessian(vector<double>& x, double fx, vector<double>& h);
  protected:
    NLModel0() {}

    FDType fdType;
    vector<double> currentX;
    double currentF;
};

//concrete class of nonlinear optimization model without derivatives

class ConcreteNLModel0 : public NLModel0{

  public:

    ConcreteNLModel0(int dim, ObjFunctor0* func ,  ConstraintList* cons = NULL) : NLModel0(dim, cons){objfunc = func;}
   

    virtual double calcF();
    virtual double calcF(vector<double>& x);
    virtual vector<double> calcGrad();
    virtual vector<double> calcGrad(vector<double>& x);
    virtual SymMatrix calcHessian() ;
    virtual SymMatrix calcHessian(vector<double>& x) ; 
    
  protected:

    ObjFunctor0* objfunc;

};

//abstract class of nonlinear optimization model with first derivatives
class NLModel1 : public NLModel0{

  public:
    NLModel1(int dim,  ConstraintList* cons ) : NLModel0(dim, cons){currentGrad.resize(dim);}
    //Using finite difference methods to approximate the hessian
    //It is inappropriate to apply this method in large scale problem    
    virtual SymMatrix FiniteHessian(vector<double>& x, vector<double>& h);

    void setGrad(vector<double>& grad) {currentGrad = grad;}
    vector<double> getGrad() {return currentGrad;}
  protected:

    vector<double> currentGrad;
};

//concrete class of nonlinear optimization model with first derivatives
class ConcreteNLModel1 : public NLModel1{

  public:

    ConcreteNLModel1(int dim, ObjFunctor1* func ,  ConstraintList* cons = NULL);
    
    
    virtual double calcF();
    virtual double calcF(vector<double>& x);
    virtual vector<double> calcGrad();
    virtual vector<double> calcGrad( vector<double>& x);
    virtual SymMatrix calcHessian() ;
    virtual SymMatrix calcHessian(vector<double>& x) ; 

  protected:

    ObjFunctor1* objfunc;
};

/*
//abstract class of nonlinear optimization model with second derivatives
class NLModel2 : public NLModel1{
  public:
    
  protected:    
    SymMatrix currentHessian;

};

//concrete class of nonlinear optimization model with second derivatives
class ConcreteNLModel2 : public NLModel2{
  public:

    ConcreteNLModel2(int dim, ObjFunctor2* func ,  ConstraintList* cons = NULL);
    ConcreteNLModel2(int dim, ConstraintList* cons = NULL);
    
    virtual double calcF();
    virtual double calcF(vector<double>& x);
    virtual vector<double> calcGrad();
    virtual vector<double> calcGrad(vector<double>& x);
    virtual SymMatrix calcHessian() ;
    virtual SymMatrix calcHessian(vector<double>& x) ;
    
  protected:
    
    ObjFunctor2* objFunc;
};
*/
#endif
Revision:	1031
Committed:	Fri Feb 6 18:58:06 2004 UTC (21 years, 8 months ago) by tim
File size:	5841 byte(s)
Log Message:	Add some lines into global.cpp to make it work with energy minimization
#	Content
1	#ifndef _NLMODEL_H_
2	#define _NLMODEL_H_
3
4	#include <vector>
5	#include <utility>
6	#include <math.h>
7
8	#include "SymMatrix.hpp"
9	#include "Functor.hpp"
10	#include "ConstraintList.hpp"
11
12	using namespace std;
13
14
15	typedef enum {backward, forward, central} FDType;
16
17	// special property of nonlinear object function
18	typedef enum {linear, quadratic, general} NLOFProp;
19
20	//abstract class of nonlinear optimization model
21	class NLModel{
22	public:
23	NLModel(int dim, ConstraintList* cons) { ndim = dim, constraints = cons;}
24	virtual ~NLModel() { if (constraints != NULL) delete constraints;}
25
26	virtual void setX(const vector<double>& x)= 0;
27	virtual vector<double> getX() = 0;
28
29	virtual void setF(double f) = 0;
30	virtual double getF() = 0;
31
32	virtual int getDim() {return ndim;}
33
34	bool hasConstraints() { return constraints == NULL ? false : true;}
35	int getConsType() { return constraints->getConsType();}
36
37	virtual double calcF() = 0;
38	virtual double calcF(vector<double>& x) = 0;
39	virtual vector<double> calcGrad() = 0;
40	virtual vector<double> calcGrad(vector<double>& x) = 0;
41	virtual SymMatrix calcHessian() = 0;
42	virtual SymMatrix calcHessian(vector<double>& x) = 0;
43
44	#ifdef IS_MPI
45	//void setMPIINITFunctor(MPIINITFunctor* func);
46	//int getLocalDim() {return localDim;}
47
48	//virtual void update(); //a hook function to load balancing
49	#endif
50
51	protected:
52	NLModel() {}
53	ConstraintList* constraints; //constraints of nonlinear optimization model
54	int numOfFunEval; //number of function evaluation
55	int ndim;
56
57	#ifdef IS_MPI
58	bool mpiInitFlag;
59	int myRank; //rank of current node
60	int numOfProc; // number of processors
61	//MPIINITFunctor * mpiInitFunc;
62
63	int localDim;
64	vector<int> procMappingArray;
65	int beginGlobalIndex;
66	#endif
67	};
68
69	//abstract class of nonlinear optimization model without derivatives
70	class NLModel0 : public NLModel{
71	public:
72
73	NLModel0(int dim, ConstraintList* cons) : NLModel(dim, cons) { currentX.resize(dim);}
74	~NLModel0() {}
75
76	virtual void setX(const vector<double>& x) {currentX = x;}
77	vector<double> getX() {return currentX;}
78
79	void setF(double f) {currentF = f;}
80	double getF() {return currentF;}
81
82	//Using finite difference methods to approximate the gradient
83	//It is inappropriate to apply these methods in large scale problem
84
85	vector<double> BackwardGrad(const vector<double>& x, double& fx, vector<double>& grad, const vector<double>& h);
86	vector<double> ForwardGrad(const vector<double>& x, double& fx, vector<double>& grad, const vector<double>& h);
87	vector<double> CentralGrad(const vector<double>& x, double& fx, vector<double>& grad, const vector<double>& h);
88
89	//Using finite difference methods to approximate the hessian
90	//It is inappropriate to apply this method in large scale problem
91	//virtual SymMatrix FiniteHessian(vector<double>& x, double fx, vector<double>& h);
92	SymMatrix FiniteHessian(vector<double>& x, double fx, vector<double>& h);
93	protected:
94	NLModel0() {}
95
96	FDType fdType;
97	vector<double> currentX;
98	double currentF;
99	};
100
101	//concrete class of nonlinear optimization model without derivatives
102
103	class ConcreteNLModel0 : public NLModel0{
104
105	public:
106
107	ConcreteNLModel0(int dim, ObjFunctor0* func , ConstraintList* cons = NULL) : NLModel0(dim, cons){objfunc = func;}
108
109
110	virtual double calcF();
111	virtual double calcF(vector<double>& x);
112	virtual vector<double> calcGrad();
113	virtual vector<double> calcGrad(vector<double>& x);
114	virtual SymMatrix calcHessian() ;
115	virtual SymMatrix calcHessian(vector<double>& x) ;
116
117	protected:
118
119	ObjFunctor0* objfunc;
120
121	};
122
123	//abstract class of nonlinear optimization model with first derivatives
124	class NLModel1 : public NLModel0{
125
126	public:
127	NLModel1(int dim, ConstraintList* cons ) : NLModel0(dim, cons){currentGrad.resize(dim);}
128	//Using finite difference methods to approximate the hessian
129	//It is inappropriate to apply this method in large scale problem
130	virtual SymMatrix FiniteHessian(vector<double>& x, vector<double>& h);
131
132	void setGrad(vector<double>& grad) {currentGrad = grad;}
133	vector<double> getGrad() {return currentGrad;}
134	protected:
135
136	vector<double> currentGrad;
137	};
138
139	//concrete class of nonlinear optimization model with first derivatives
140	class ConcreteNLModel1 : public NLModel1{
141
142	public:
143
144	ConcreteNLModel1(int dim, ObjFunctor1* func , ConstraintList* cons = NULL);
145
146
147	virtual double calcF();
148	virtual double calcF(vector<double>& x);
149	virtual vector<double> calcGrad();
150	virtual vector<double> calcGrad( vector<double>& x);
151	virtual SymMatrix calcHessian() ;
152	virtual SymMatrix calcHessian(vector<double>& x) ;
153
154	protected:
155
156	ObjFunctor1* objfunc;
157	};
158
159	/*
160	//abstract class of nonlinear optimization model with second derivatives
161	class NLModel2 : public NLModel1{
162	public:
163
164	protected:
165	SymMatrix currentHessian;
166
167	};
168
169	//concrete class of nonlinear optimization model with second derivatives
170	class ConcreteNLModel2 : public NLModel2{
171	public:
172
173	ConcreteNLModel2(int dim, ObjFunctor2* func , ConstraintList* cons = NULL);
174	ConcreteNLModel2(int dim, ConstraintList* cons = NULL);
175
176	virtual double calcF();
177	virtual double calcF(vector<double>& x);
178	virtual vector<double> calcGrad();
179	virtual vector<double> calcGrad(vector<double>& x);
180	virtual SymMatrix calcHessian() ;
181	virtual SymMatrix calcHessian(vector<double>& x) ;
182
183	protected:
184
185	ObjFunctor2* objFunc;
186	};
187	*/
188	#endif