Use constructor instead of SetParameter and remove needless file.
[PaGMO.git] / AstroToolbox / ZeroFinder.h
blob27499bc1832f4dbd82463e831dd6f159e19433ff
1 // ------------------------------------------------------------------------ //
2 // This source file is part of the 'ESA Advanced Concepts Team's //
3 // Space Mechanics Toolbox' software. //
4 // //
5 // The source files are for research use only, //
6 // and are distributed WITHOUT ANY WARRANTY. Use them on your own risk. //
7 // //
8 // Copyright (c) 2004-2007 European Space Agency //
9 // ------------------------------------------------------------------------ //
11 #ifndef ZEROFINDER_H
12 #define ZEROFINDER_H
14 #define NUMERATOR(dab,dcb,fa,fb,fc) fb*(dab*fc*(fc-fb)-fa*dcb*(fa-fb))
15 #define DENOMINATOR(fa,fb,fc) (fc-fb)*(fa-fb)*(fa-fc)
17 /// Namespace
18 namespace ZeroFinder
21 /// Class for one dimensional functions
22 // with some parameters
23 /** The ()-operator with one double argument
24 * has to be overloaded for a derived class
25 * The return value is the ordinate computed for
26 * the abscissa-argument.
28 class Function1D
30 public:
31 Function1D(const double &a, const double &b):p1(a),p2(b) {}
32 // parameters
33 const double p1,p2;
36 class Function1D_7param
38 public:
39 Function1D_7param(const double &a, const double &b, const double &c,
40 const double &d, const double &e, const double &f,
41 const double &g):p1(a),p2(b),p3(c),p4(d),p5(e),p6(f),p7(g) {}
42 // parameters
43 const double p1,p2,p3,p4,p5,p6,p7;
47 class FZero
49 private:
50 double a, c; // lower and upper bounds
52 public:
53 FZero(const double &a_, const double &c_):a(a_),c(c_) {} // constructor
54 // fzero procedure
55 template <class Functor>
56 double FindZero(const Functor &f);
59 //-------------------------------------------------------------------------------------//
60 // This part is an adaption of the 'Amsterdam method', which is an inversre quadratic //
61 // interpolation - bisection method //
62 // See http://mymathlib.webtrellis.net/roots/amsterdam.html //
64 //-------------------------------------------------------------------------------------//
65 template <class Functor>
66 double FZero::FindZero(const Functor &f)
68 static const int max_iterations = 500;
69 static const double tolerance = 1e-15;
71 double fa = f(a), b = 0.5 * ( a + c ), fc = f(c), fb = fa * fc;
72 double delta, dab, dcb;
73 int i;
75 // If the initial estimates do not bracket a root, set the err flag and //
76 // return. If an initial estimate is a root, then return the root. //
78 double err = 0;
79 if ( fb >= 0.0 )
80 if ( fb > 0.0 ) { err = -1; return 0.0; }
81 else return ( fa == 0.0 ) ? a : c;
83 // Insure that the initial estimate a < c. //
85 if ( a > c ) {
86 delta = a; a = c; c = delta; delta = fa; fa = fc; fc = delta;
89 // If the function at the left endpoint is positive, and the function //
90 // at the right endpoint is negative. Iterate reducing the length //
91 // of the interval by either bisection or quadratic inverse //
92 // interpolation. Note that the function at the left endpoint //
93 // remains nonnegative and the function at the right endpoint remains //
94 // nonpositive. //
96 if ( fa > 0.0 )
97 for ( i = 0; i < max_iterations; i++) {
99 // Are the two endpoints within the user specified tolerance ?
101 if ( ( c - a ) < tolerance ) return 0.5 * ( a + c);
103 // No, Continue iteration.
105 fb = f(b);
107 // Check that we are converging or that we have converged near //
108 // the left endpoint of the inverval. If it appears that the //
109 // interval is not decreasing fast enough, use bisection. //
110 if ( ( c - a ) < tolerance ) return 0.5 * ( a + c);
111 if ( ( b - a ) < tolerance )
112 if ( fb > 0 ) {
113 a = b; fa = fb; b = 0.5 * ( a + c ); continue;
115 else return b;
117 // Check that we are converging or that we have converged near //
118 // the right endpoint of the inverval. If it appears that the //
119 // interval is not decreasing fast enough, use bisection. //
121 if ( ( c - b ) < tolerance )
122 if ( fb < 0 ) {
123 c = b; fc = fb; b = 0.5 * ( a + c ); continue;
125 else return b;
127 // If quadratic inverse interpolation is feasible, try it. //
129 if ( ( fa > fb ) && ( fb > fc ) ) {
130 delta = DENOMINATOR(fa,fb,fc);
131 if ( delta != 0.0 ) {
132 dab = a - b;
133 dcb = c - b;
134 delta = NUMERATOR(dab,dcb,fa,fb,fc) / delta;
136 // Will the new estimate of the root be within the //
137 // interval? If yes, use it and update interval. //
138 // If no, use the bisection method. //
140 if ( delta > dab && delta < dcb ) {
141 if ( fb > 0.0 ) { a = b; fa = fb; }
142 else if ( fb < 0.0 ) { c = b; fc = fb; }
143 else return b;
144 b += delta;
145 continue;
150 // If not, use the bisection method. //
152 fb > 0 ? ( a = b, fa = fb ) : ( c = b, fc = fb );
153 b = 0.5 * ( a + c );
155 else
157 // If the function at the left endpoint is negative, and the function //
158 // at the right endpoint is positive. Iterate reducing the length //
159 // of the interval by either bisection or quadratic inverse //
160 // interpolation. Note that the function at the left endpoint //
161 // remains nonpositive and the function at the right endpoint remains //
162 // nonnegative. //
164 for ( i = 0; i < max_iterations; i++) {
165 if ( ( c - a ) < tolerance ) return 0.5 * ( a + c);
166 fb = f(b);
168 if ( ( b - a ) < tolerance )
169 if ( fb < 0 ) {
170 a = b; fa = fb; b = 0.5 * ( a + c ); continue;
172 else return b;
174 if ( ( c - b ) < tolerance )
175 if ( fb > 0 ) {
176 c = b; fc = fb; b = 0.5 * ( a + c ); continue;
178 else return b;
180 if ( ( fa < fb ) && ( fb < fc ) ) {
181 delta = DENOMINATOR(fa,fb,fc);
182 if ( delta != 0.0 ) {
183 dab = a - b;
184 dcb = c - b;
185 delta = NUMERATOR(dab,dcb,fa,fb,fc) / delta;
186 if ( delta > dab && delta < dcb ) {
187 if ( fb < 0.0 ) { a = b; fa = fb; }
188 else if ( fb > 0.0 ) { c = b; fc = fb; }
189 else return b;
190 b += delta;
191 continue;
195 fb < 0 ? ( a = b, fa = fb ) : ( c = b, fc = fb );
196 b = 0.5 * ( a + c );
198 err = -2;
199 return b;
203 #undef NUMERATOR
204 #undef DENOMINATOR
206 #endif