| 1 | tim | 2440 | /********************************************************************** | 
| 2 |  |  | vector3.cpp - Handle 3D coordinates. | 
| 3 |  |  |  | 
| 4 |  |  | Copyright (C) 1998-2001 by OpenEye Scientific Software, Inc. | 
| 5 |  |  | Some portions Copyright (C) 2001-2005 by Geoffrey R. Hutchison | 
| 6 |  |  |  | 
| 7 |  |  | This file is part of the Open Babel project. | 
| 8 |  |  | For more information, see <http://openbabel.sourceforge.net/> | 
| 9 |  |  |  | 
| 10 |  |  | This program is free software; you can redistribute it and/or modify | 
| 11 |  |  | it under the terms of the GNU General Public License as published by | 
| 12 |  |  | the Free Software Foundation version 2 of the License. | 
| 13 |  |  |  | 
| 14 |  |  | This program is distributed in the hope that it will be useful, | 
| 15 |  |  | but WITHOUT ANY WARRANTY; without even the implied warranty of | 
| 16 |  |  | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the | 
| 17 |  |  | GNU General Public License for more details. | 
| 18 |  |  | ***********************************************************************/ | 
| 19 |  |  |  | 
| 20 |  |  | #include <math.h> | 
| 21 |  |  |  | 
| 22 |  |  | #include "mol.hpp" | 
| 23 |  |  | #include "vector3.hpp" | 
| 24 |  |  |  | 
| 25 |  |  | using namespace std; | 
| 26 |  |  |  | 
| 27 |  |  | namespace OpenBabel | 
| 28 |  |  | { | 
| 29 |  |  |  | 
| 30 |  |  | /*! \class vector3 | 
| 31 |  |  | \brief Represents a vector in the 3-dimensional real space. | 
| 32 |  |  |  | 
| 33 |  |  | The vector3 class was designed to simplify operations with doubleing | 
| 34 |  |  | point coordinates. To this end many of the common operations have been | 
| 35 |  |  | overloaded for simplicity. Vector addition, subtraction, scalar | 
| 36 |  |  | multiplication, dot product, cross product, magnitude and a number of | 
| 37 |  |  | other utility functions are built in to the vector class. For a full | 
| 38 |  |  | description of the class member functions please consult the header | 
| 39 |  |  | file vector3.h. The following code demonstrates several of the | 
| 40 |  |  | functions of the vector class: | 
| 41 |  |  | \code | 
| 42 |  |  | vector3 v1,v2,v3; | 
| 43 |  |  | v1 = VX; | 
| 44 |  |  | v2 = VY; | 
| 45 |  |  | v3 = cross(v1,v2); | 
| 46 |  |  | v3 *= 2.5; | 
| 47 |  |  | v3.normalize(); | 
| 48 |  |  | \endcode | 
| 49 |  |  | */ | 
| 50 |  |  |  | 
| 51 |  |  | /*! This (slow) method allows to access the elements of the | 
| 52 |  |  | vector as if it were an array of doubles. If the index is > 2, | 
| 53 |  |  | then a warning is printed, and the program is terminated via | 
| 54 |  |  | exit(-1). Otherwise, if i is 0, 1 or 2, then a reference to x, | 
| 55 |  |  | y or z is returned, respectively. | 
| 56 |  |  |  | 
| 57 |  |  | \warning This method is primarily designed to facilitate the | 
| 58 |  |  | integration ('Open Babelization') of code that uses arrays of | 
| 59 |  |  | doubles rather than the vector class. Due to the error checks | 
| 60 |  |  | the method is of course very slow and should therefore be | 
| 61 |  |  | avoided in production code. | 
| 62 |  |  | */ | 
| 63 |  |  | double& vector3::operator[] ( unsigned int i) | 
| 64 |  |  | { | 
| 65 |  |  | if (i > 2) | 
| 66 |  |  | { | 
| 67 |  |  | cerr << "ERROR in OpenBabel::vector3::operator[]" << endl | 
| 68 |  |  | << "The method has been called with an illegal index i=" << i << "." << endl | 
| 69 |  |  | << "Please contact the author of the offending program immediately." << endl; | 
| 70 |  |  | exit(-1); | 
| 71 |  |  | } | 
| 72 |  |  | if (i == 0) | 
| 73 |  |  | return _vx; | 
| 74 |  |  | if (i == 1) | 
| 75 |  |  | return _vy; | 
| 76 |  |  | return _vz; | 
| 77 |  |  | } | 
| 78 |  |  |  | 
| 79 |  |  | /*! replaces *this with a random unit vector, which is (supposed | 
| 80 |  |  | to be) uniformly distributed over the unit sphere. Uses the | 
| 81 |  |  | random number generator obRand, or uses the system number | 
| 82 |  |  | generator with a time seed if obRand == NULL. | 
| 83 |  |  |  | 
| 84 |  |  | @param obRandP random number generator to use, or 0L, if the | 
| 85 |  |  | system random number generator (with time seed) should be used | 
| 86 |  |  | */ | 
| 87 |  |  | void vector3::randomUnitVector(OBRandom *obRandP) | 
| 88 |  |  | { | 
| 89 |  |  | OBRandom *ptr; | 
| 90 |  |  | if (!obRandP) | 
| 91 |  |  | { | 
| 92 |  |  | ptr = new OBRandom(true); | 
| 93 |  |  | ptr->TimeSeed(); | 
| 94 |  |  | } | 
| 95 |  |  | else | 
| 96 |  |  | ptr = obRandP; | 
| 97 |  |  |  | 
| 98 |  |  | // obtain a random vector with 0.001 <= length^2 <= 1.0, normalize | 
| 99 |  |  | // the vector to obtain a random vector of length 1.0. | 
| 100 |  |  | double l; | 
| 101 |  |  | do | 
| 102 |  |  | { | 
| 103 |  |  | this->Set(ptr->NextFloat()-0.5, ptr->NextFloat()-0.5, ptr->NextFloat()-0.5); | 
| 104 |  |  | l = length_2(); | 
| 105 |  |  | } | 
| 106 |  |  | while ( (l > 1.0) || (l < 1e-4) ); | 
| 107 |  |  | this->normalize(); | 
| 108 |  |  |  | 
| 109 |  |  | if (!obRandP) | 
| 110 |  |  | delete ptr; | 
| 111 |  |  | } | 
| 112 |  |  |  | 
| 113 |  |  | OBAPI ostream& operator<< ( ostream& co, const vector3& v ) | 
| 114 |  |  | { | 
| 115 |  |  | co << "< " << v._vx << ", " << v._vy << ", " << v._vz << " >" ; | 
| 116 |  |  | return co ; | 
| 117 |  |  | } | 
| 118 |  |  |  | 
| 119 |  |  | OBAPI int operator== ( const vector3& v1, const vector3& v2 ) | 
| 120 |  |  | { | 
| 121 |  |  | if ( ( v1._vx == v2._vx ) && | 
| 122 |  |  | ( v1._vy == v2._vy ) && | 
| 123 |  |  | ( v1._vz == v2._vz ) ) | 
| 124 |  |  | return ( true ) ; | 
| 125 |  |  | else | 
| 126 |  |  | return ( false ) ; | 
| 127 |  |  | } | 
| 128 |  |  |  | 
| 129 |  |  | OBAPI int operator!= ( const vector3& v1, const vector3& v2 ) | 
| 130 |  |  | { | 
| 131 |  |  | if ( ( v1._vx != v2._vx ) || | 
| 132 |  |  | ( v1._vy != v2._vy ) || | 
| 133 |  |  | ( v1._vz != v2._vz ) ) | 
| 134 |  |  | return ( true ) ; | 
| 135 |  |  | else | 
| 136 |  |  | return ( false ) ; | 
| 137 |  |  | } | 
| 138 |  |  |  | 
| 139 |  |  | /*! This method checks if the current vector has length() == | 
| 140 |  |  | 0.0.  If so, *this remains unchanged. Otherwise, *this is | 
| 141 |  |  | scaled by 1.0/length(). | 
| 142 |  |  |  | 
| 143 |  |  | \warning If length() is very close to zero, but not == 0.0, | 
| 144 |  |  | this method may behave in unexpected ways and return almost | 
| 145 |  |  | random results; details may depend on your particular doubleing | 
| 146 |  |  | point implementation. The use of this method is therefore | 
| 147 |  |  | highly discouraged, unless you are certain that length() is in | 
| 148 |  |  | a reasonable range, away from 0.0 (Stefan Kebekus) | 
| 149 |  |  |  | 
| 150 |  |  | \deprecated This method will probably replaced by a safer | 
| 151 |  |  | algorithm in the future. | 
| 152 |  |  |  | 
| 153 |  |  | \todo Replace this method with a more fool-proof version. | 
| 154 |  |  |  | 
| 155 |  |  | @returns a reference to *this | 
| 156 |  |  | */ | 
| 157 |  |  | vector3& vector3 :: normalize () | 
| 158 |  |  | { | 
| 159 |  |  | double l = length (); | 
| 160 |  |  |  | 
| 161 |  |  | if (IsNearZero(l)) | 
| 162 |  |  | return(*this); | 
| 163 |  |  |  | 
| 164 |  |  | _vx = _vx / l ; | 
| 165 |  |  | _vy = _vy / l ; | 
| 166 |  |  | _vz = _vz / l ; | 
| 167 |  |  |  | 
| 168 |  |  | return(*this); | 
| 169 |  |  | } | 
| 170 |  |  |  | 
| 171 |  |  | OBAPI double dot ( const vector3& v1, const vector3& v2 ) | 
| 172 |  |  | { | 
| 173 |  |  | return v1._vx*v2._vx + v1._vy*v2._vy + v1._vz*v2._vz ; | 
| 174 |  |  | } | 
| 175 |  |  |  | 
| 176 |  |  | OBAPI vector3 cross ( const vector3& v1, const vector3& v2 ) | 
| 177 |  |  | { | 
| 178 |  |  | vector3 vv ; | 
| 179 |  |  |  | 
| 180 |  |  | vv._vx =   v1._vy*v2._vz - v1._vz*v2._vy ; | 
| 181 |  |  | vv._vy = - v1._vx*v2._vz + v1._vz*v2._vx ; | 
| 182 |  |  | vv._vz =   v1._vx*v2._vy - v1._vy*v2._vx ; | 
| 183 |  |  |  | 
| 184 |  |  | return ( vv ) ; | 
| 185 |  |  | } | 
| 186 |  |  |  | 
| 187 |  |  |  | 
| 188 |  |  | /*! This method calculates the angle between two vectors | 
| 189 |  |  |  | 
| 190 |  |  | \warning If length() of any of the two vectors is == 0.0, | 
| 191 |  |  | this method will divide by zero. If the product of the | 
| 192 |  |  | length() of the two vectors is very close to 0.0, but not == | 
| 193 |  |  | 0.0, this method may behave in unexpected ways and return | 
| 194 |  |  | almost random results; details may depend on your particular | 
| 195 |  |  | doubleing point implementation. The use of this method is | 
| 196 |  |  | therefore highly discouraged, unless you are certain that the | 
| 197 |  |  | length()es are in a reasonable range, away from 0.0 (Stefan | 
| 198 |  |  | Kebekus) | 
| 199 |  |  |  | 
| 200 |  |  | \deprecated This method will probably replaced by a safer | 
| 201 |  |  | algorithm in the future. | 
| 202 |  |  |  | 
| 203 |  |  | \todo Replace this method with a more fool-proof version. | 
| 204 |  |  |  | 
| 205 |  |  | @returns the angle in degrees (0-360) | 
| 206 |  |  | */ | 
| 207 |  |  | OBAPI double vectorAngle ( const vector3& v1, const vector3& v2 ) | 
| 208 |  |  | { | 
| 209 |  |  | double mag; | 
| 210 |  |  | double dp; | 
| 211 |  |  |  | 
| 212 |  |  | mag = v1.length() * v2.length(); | 
| 213 |  |  | dp = dot(v1,v2)/mag; | 
| 214 |  |  |  | 
| 215 |  |  | if (dp < -0.999999) | 
| 216 |  |  | dp = -0.9999999; | 
| 217 |  |  |  | 
| 218 |  |  | if (dp > 0.9999999) | 
| 219 |  |  | dp = 0.9999999; | 
| 220 |  |  |  | 
| 221 |  |  | if (dp > 1.0) | 
| 222 |  |  | dp = 1.0; | 
| 223 |  |  |  | 
| 224 |  |  | return((RAD_TO_DEG * acos(dp))); | 
| 225 |  |  | } | 
| 226 |  |  |  | 
| 227 |  |  | OBAPI double CalcTorsionAngle(const vector3 &a, const vector3 &b, | 
| 228 |  |  | const vector3 &c, const vector3 &d) | 
| 229 |  |  | { | 
| 230 |  |  | double torsion; | 
| 231 |  |  | vector3 b1,b2,b3,c1,c2,c3; | 
| 232 |  |  |  | 
| 233 |  |  | b1 = a - b; | 
| 234 |  |  | b2 = b - c; | 
| 235 |  |  | b3 = c - d; | 
| 236 |  |  |  | 
| 237 |  |  | c1 = cross(b1,b2); | 
| 238 |  |  | c2 = cross(b2,b3); | 
| 239 |  |  | c3 = cross(c1,c2); | 
| 240 |  |  |  | 
| 241 |  |  | if (c1.length() * c2.length() < 0.001) | 
| 242 |  |  | torsion = 0.0; | 
| 243 |  |  | else | 
| 244 |  |  | { | 
| 245 |  |  | torsion = vectorAngle(c1,c2); | 
| 246 |  |  | if (dot(b2,c3) > 0.0) | 
| 247 |  |  | torsion *= -1.0; | 
| 248 |  |  | } | 
| 249 |  |  |  | 
| 250 |  |  | return(torsion); | 
| 251 |  |  | } | 
| 252 |  |  |  | 
| 253 |  |  | /*! This method checks if the current vector *this is zero | 
| 254 |  |  | (i.e. if all entries == 0.0). If so, a warning message is | 
| 255 |  |  | printed, and the whole program is aborted with exit(0). | 
| 256 |  |  | Otherwise, a vector of length one is generated, which is | 
| 257 |  |  | orthogonal to *this, and stored in v. The resulting vector is | 
| 258 |  |  | not random. | 
| 259 |  |  |  | 
| 260 |  |  | \warning If the entries of the *this (in particular the | 
| 261 |  |  | z-component) are very close to zero, but not == 0.0, this | 
| 262 |  |  | method may behave in unexpected ways and return almost random | 
| 263 |  |  | results; details may depend on your particular floating point | 
| 264 |  |  | implementation. The use of this method is therefore highly | 
| 265 |  |  | discouraged, unless you are certain that all components of | 
| 266 |  |  | *this are in a reasonable range, away from 0.0 (Stefan | 
| 267 |  |  | Kebekus) | 
| 268 |  |  |  | 
| 269 |  |  | \deprecated This method will probably replaced by a safer | 
| 270 |  |  | algorithm in the future. | 
| 271 |  |  |  | 
| 272 |  |  | \todo Replace this method with a more fool-proof version that | 
| 273 |  |  | does not call exit() | 
| 274 |  |  |  | 
| 275 |  |  | @param res a reference to a vector where the result will be | 
| 276 |  |  | stored | 
| 277 |  |  | */ | 
| 278 |  |  | void vector3::createOrthoVector(vector3 &res) const | 
| 279 |  |  | { | 
| 280 |  |  | vector3 cO; | 
| 281 |  |  |  | 
| 282 |  |  | if ( ( IsNearZero(this->x())) && (IsNearZero(this->y())) ) | 
| 283 |  |  | { | 
| 284 |  |  | if ( IsNearZero(this->z()) ) | 
| 285 |  |  | { | 
| 286 |  |  | cerr << "makeorthovec zero vector" << endl; | 
| 287 |  |  | exit(0); | 
| 288 |  |  | } | 
| 289 |  |  | cO.SetX(1.0); | 
| 290 |  |  | } | 
| 291 |  |  | else | 
| 292 |  |  | { | 
| 293 |  |  | cO.SetZ(1.0); | 
| 294 |  |  | } | 
| 295 |  |  | res= cross(cO,*this); | 
| 296 |  |  | res.normalize(); | 
| 297 |  |  | } | 
| 298 |  |  |  | 
| 299 |  |  | const vector3 VZero ( 0.0, 0.0, 0.0 ) ; | 
| 300 |  |  | const vector3 VX    ( 1.0, 0.0, 0.0 ) ; | 
| 301 |  |  | const vector3 VY    ( 0.0, 1.0, 0.0 ) ; | 
| 302 |  |  | const vector3 VZ    ( 0.0, 0.0, 1.0 ) ; | 
| 303 |  |  |  | 
| 304 |  |  | /* Calculate the distance of point a to the plane determined by b,c,d */ | 
| 305 |  |  | double Point2Plane(vector3 a, vector3 b, vector3 c, vector3 d) | 
| 306 |  |  | { | 
| 307 |  |  | double angle =0; | 
| 308 |  |  | double dist_ab =0; | 
| 309 |  |  | vector3 v_ba = a-b; | 
| 310 |  |  | vector3 v_normal = cross(c-b, d-b).normalize(); | 
| 311 |  |  | angle = vectorAngle(v_normal, v_ba); | 
| 312 |  |  | dist_ab = v_ba.length(); | 
| 313 |  |  | return fabs(dist_ab * cos(DEG_TO_RAD * angle)); | 
| 314 |  |  | } | 
| 315 |  |  |  | 
| 316 |  |  | } // namespace OpenBabel | 
| 317 |  |  |  | 
| 318 |  |  | //! \file vector3.cpp | 
| 319 |  |  | //! \brief Handle 3D coordinates. |