Geant4 11.1.1
Toolkit for the simulation of the passage of particles through matter
Loading...
Searching...
No Matches
RandBreitWigner.cc
Go to the documentation of this file.
1// -*- C++ -*-
2//
3// -----------------------------------------------------------------------
4// HEP Random
5// --- RandBreitWigner ---
6// class implementation file
7// -----------------------------------------------------------------------
8// This file is part of Geant4 (simulation toolkit for HEP).
9
10// =======================================================================
11// Gabriele Cosmo - Created: 5th September 1995
12// - Added methods to shoot arrays: 28th July 1997
13// J.Marraffino - Added default arguments as attributes and
14// operator() with arguments: 16th Feb 1998
15// M Fischler - put and get to/from streams 12/10/04
16// M Fischler - put/get to/from streams uses pairs of ulongs when
17// + storing doubles avoid problems with precision
18// 4/14/05
19// =======================================================================
20
24#include <algorithm> // for min() and max()
25#include <cmath>
26#include <iostream>
27#include <string>
28#include <vector>
29
30namespace CLHEP {
31
32std::string RandBreitWigner::name() const {return "RandBreitWigner";}
33HepRandomEngine & RandBreitWigner::engine() {return *localEngine;}
34
36}
37
39 return fire( defaultA, defaultB );
40}
41
42double RandBreitWigner::operator()( double a, double b ) {
43 return fire( a, b );
44}
45
46double RandBreitWigner::operator()( double a, double b, double c ) {
47 return fire( a, b, c );
48}
49
50double RandBreitWigner::shoot(double mean, double gamma)
51{
52 double rval, displ;
53
54 rval = 2.0*HepRandom::getTheEngine()->flat()-1.0;
55 displ = 0.5*gamma*std::tan(rval*CLHEP::halfpi);
56
57 return mean + displ;
58}
59
60double RandBreitWigner::shoot(double mean, double gamma, double cut)
61{
62 double val, rval, displ;
63
64 if ( gamma == 0.0 ) return mean;
65 val = std::atan(2.0*cut/gamma);
66 rval = 2.0*HepRandom::getTheEngine()->flat()-1.0;
67 displ = 0.5*gamma*std::tan(rval*val);
68
69 return mean + displ;
70}
71
72double RandBreitWigner::shootM2(double mean, double gamma )
73{
74 double val, rval, displ;
75
76 if ( gamma == 0.0 ) return mean;
77 val = std::atan(-mean/gamma);
78 rval = RandFlat::shoot(val, CLHEP::halfpi);
79 displ = gamma*std::tan(rval);
80
81 return std::sqrt(mean*mean + mean*displ);
82}
83
84double RandBreitWigner::shootM2(double mean, double gamma, double cut )
85{
86 double rval, displ;
87 double lower, upper, tmp;
88
89 if ( gamma == 0.0 ) return mean;
90 tmp = std::max(0.0,(mean-cut));
91 lower = std::atan( (tmp*tmp-mean*mean)/(mean*gamma) );
92 upper = std::atan( ((mean+cut)*(mean+cut)-mean*mean)/(mean*gamma) );
93 rval = RandFlat::shoot(lower, upper);
94 displ = gamma*std::tan(rval);
95
96 return std::sqrt(std::max(0.0, mean*mean + mean*displ));
97}
98
99void RandBreitWigner::shootArray ( const int size, double* vect )
100{
101 for( double* v = vect; v != vect + size; ++v )
102 *v = shoot( 1.0, 0.2 );
103}
104
105void RandBreitWigner::shootArray ( const int size, double* vect,
106 double a, double b )
107{
108 for( double* v = vect; v != vect + size; ++v )
109 *v = shoot( a, b );
110}
111
112void RandBreitWigner::shootArray ( const int size, double* vect,
113 double a, double b,
114 double c )
115{
116 for( double* v = vect; v != vect + size; ++v )
117 *v = shoot( a, b, c );
118}
119
120//----------------
121
123 double mean, double gamma)
124{
125 double rval, displ;
126
127 rval = 2.0*anEngine->flat()-1.0;
128 displ = 0.5*gamma*std::tan(rval*CLHEP::halfpi);
129
130 return mean + displ;
131}
132
134 double mean, double gamma, double cut )
135{
136 double val, rval, displ;
137
138 if ( gamma == 0.0 ) return mean;
139 val = std::atan(2.0*cut/gamma);
140 rval = 2.0*anEngine->flat()-1.0;
141 displ = 0.5*gamma*std::tan(rval*val);
142
143 return mean + displ;
144}
145
147 double mean, double gamma )
148{
149 double val, rval, displ;
150
151 if ( gamma == 0.0 ) return mean;
152 val = std::atan(-mean/gamma);
153 rval = RandFlat::shoot(anEngine,val, CLHEP::halfpi);
154 displ = gamma*std::tan(rval);
155
156 return std::sqrt(mean*mean + mean*displ);
157}
158
160 double mean, double gamma, double cut )
161{
162 double rval, displ;
163 double lower, upper, tmp;
164
165 if ( gamma == 0.0 ) return mean;
166 tmp = std::max(0.0,(mean-cut));
167 lower = std::atan( (tmp*tmp-mean*mean)/(mean*gamma) );
168 upper = std::atan( ((mean+cut)*(mean+cut)-mean*mean)/(mean*gamma) );
169 rval = RandFlat::shoot(anEngine, lower, upper);
170 displ = gamma*std::tan(rval);
171
172 return std::sqrt( std::max(0.0, mean*mean+mean*displ) );
173}
174
176 const int size, double* vect )
177{
178 for( double* v = vect; v != vect + size; ++v )
179 *v = shoot( anEngine, 1.0, 0.2 );
180}
181
183 const int size, double* vect,
184 double a, double b )
185{
186 for( double* v = vect; v != vect + size; ++v )
187 *v = shoot( anEngine, a, b );
188}
189
191 const int size, double* vect,
192 double a, double b, double c )
193{
194 for( double* v = vect; v != vect + size; ++v )
195 *v = shoot( anEngine, a, b, c );
196}
197
198//----------------
199
201{
202 return fire( defaultA, defaultB );
203}
204
205double RandBreitWigner::fire(double mean, double gamma)
206{
207 double rval, displ;
208
209 rval = 2.0*localEngine->flat()-1.0;
210 displ = 0.5*gamma*std::tan(rval*CLHEP::halfpi);
211
212 return mean + displ;
213}
214
215double RandBreitWigner::fire(double mean, double gamma, double cut)
216{
217 double val, rval, displ;
218
219 if ( gamma == 0.0 ) return mean;
220 val = std::atan(2.0*cut/gamma);
221 rval = 2.0*localEngine->flat()-1.0;
222 displ = 0.5*gamma*std::tan(rval*val);
223
224 return mean + displ;
225}
226
228{
229 return fireM2( defaultA, defaultB );
230}
231
232double RandBreitWigner::fireM2(double mean, double gamma )
233{
234 double val, rval, displ;
235
236 if ( gamma == 0.0 ) return mean;
237 val = std::atan(-mean/gamma);
238 rval = RandFlat::shoot(localEngine.get(),val, CLHEP::halfpi);
239 displ = gamma*std::tan(rval);
240
241 return std::sqrt(mean*mean + mean*displ);
242}
243
244double RandBreitWigner::fireM2(double mean, double gamma, double cut )
245{
246 double rval, displ;
247 double lower, upper, tmp;
248
249 if ( gamma == 0.0 ) return mean;
250 tmp = std::max(0.0,(mean-cut));
251 lower = std::atan( (tmp*tmp-mean*mean)/(mean*gamma) );
252 upper = std::atan( ((mean+cut)*(mean+cut)-mean*mean)/(mean*gamma) );
253 rval = RandFlat::shoot(localEngine.get(),lower, upper);
254 displ = gamma*std::tan(rval);
255
256 return std::sqrt(std::max(0.0, mean*mean + mean*displ));
257}
258
259void RandBreitWigner::fireArray ( const int size, double* vect)
260{
261 for( double* v = vect; v != vect + size; ++v )
262 *v = fire(defaultA, defaultB );
263}
264
265void RandBreitWigner::fireArray ( const int size, double* vect,
266 double a, double b )
267{
268 for( double* v = vect; v != vect + size; ++v )
269 *v = fire( a, b );
270}
271
272void RandBreitWigner::fireArray ( const int size, double* vect,
273 double a, double b, double c )
274{
275 for( double* v = vect; v != vect + size; ++v )
276 *v = fire( a, b, c );
277}
278
279
280std::ostream & RandBreitWigner::put ( std::ostream & os ) const {
281 int pr=os.precision(20);
282 std::vector<unsigned long> t(2);
283 os << " " << name() << "\n";
284 os << "Uvec" << "\n";
285 t = DoubConv::dto2longs(defaultA);
286 os << defaultA << " " << t[0] << " " << t[1] << "\n";
287 t = DoubConv::dto2longs(defaultB);
288 os << defaultB << " " << t[0] << " " << t[1] << "\n";
289 os.precision(pr);
290 return os;
291}
292
293std::istream & RandBreitWigner::get ( std::istream & is ) {
294 std::string inName;
295 is >> inName;
296 if (inName != name()) {
297 is.clear(std::ios::badbit | is.rdstate());
298 std::cerr << "Mismatch when expecting to read state of a "
299 << name() << " distribution\n"
300 << "Name found was " << inName
301 << "\nistream is left in the badbit state\n";
302 return is;
303 }
304 if (possibleKeywordInput(is, "Uvec", defaultA)) {
305 std::vector<unsigned long> t(2);
306 is >> defaultA >> t[0] >> t[1]; defaultA = DoubConv::longs2double(t);
307 is >> defaultB >> t[0] >> t[1]; defaultB = DoubConv::longs2double(t);
308 return is;
309 }
310 // is >> defaultA encompassed by possibleKeywordInput
311 is >> defaultB;
312 return is;
313}
314
315
316} // namespace CLHEP
317
static double longs2double(const std::vector< unsigned long > &v)
Definition: DoubConv.cc:110
static std::vector< unsigned long > dto2longs(double d)
Definition: DoubConv.cc:94
virtual double flat()=0
static HepRandomEngine * getTheEngine()
Definition: Random.cc:268
HepRandomEngine & engine()
static double shootM2(double a=1.0, double b=0.2)
static void shootArray(const int size, double *vect)
std::istream & get(std::istream &is)
static double shoot(double a=1.0, double b=0.2)
std::ostream & put(std::ostream &os) const
std::string name() const
void fireArray(const int size, double *vect)
static double shoot()
Definition: RandFlat.cc:61
Definition: DoubConv.h:17
bool possibleKeywordInput(IS &is, const std::string &key, T &t)
Definition: RandomEngine.h:166