1 | /* ======================================================================== *\
|
---|
2 | !
|
---|
3 | ! *
|
---|
4 | ! * This file is part of MARS, the MAGIC Analysis and Reconstruction
|
---|
5 | ! * Software. It is distributed to you in the hope that it can be a useful
|
---|
6 | ! * and timesaving tool in analysing Data of imaging Cerenkov telescopes.
|
---|
7 | ! * It is distributed WITHOUT ANY WARRANTY.
|
---|
8 | ! *
|
---|
9 | ! * Permission to use, copy, modify and distribute this software and its
|
---|
10 | ! * documentation for any purpose is hereby granted without fee,
|
---|
11 | ! * provided that the above copyright notice appear in all copies and
|
---|
12 | ! * that both that copyright notice and this permission notice appear
|
---|
13 | ! * in supporting documentation. It is provided "as is" without express
|
---|
14 | ! * or implied warranty.
|
---|
15 | ! *
|
---|
16 | !
|
---|
17 | !
|
---|
18 | ! Author(s): Harald Kornmayer 1/2001 (harald@mppmu.mpg.de)
|
---|
19 | ! Author(s): Thomas Bretz 12/2000 (tbretz@uni-sw.gwdg.de)
|
---|
20 | !
|
---|
21 | ! Copyright: MAGIC Software Development, 2000-2001
|
---|
22 | !
|
---|
23 | !
|
---|
24 | \* ======================================================================== */
|
---|
25 |
|
---|
26 | //////////////////////////////////////////////////////////////////////////////
|
---|
27 | // //
|
---|
28 | // //
|
---|
29 | //////////////////////////////////////////////////////////////////////////////
|
---|
30 | #include "MElectron.h"
|
---|
31 |
|
---|
32 | #include <math.h> // for aqlphas
|
---|
33 | #include <iostream.h>
|
---|
34 |
|
---|
35 | #include <TF1.h>
|
---|
36 | #include <TH1.h>
|
---|
37 | #include <TMatrixD.h>
|
---|
38 | #include <TVectorD.h>
|
---|
39 |
|
---|
40 | #include <TRandom.h>
|
---|
41 |
|
---|
42 | #include <TPad.h>
|
---|
43 | #include <TCanvas.h>
|
---|
44 |
|
---|
45 | #include "MPhoton.h"
|
---|
46 |
|
---|
47 | ClassImp(MElectron);
|
---|
48 |
|
---|
49 | Double_t MElectron::Sigma_ge(Double_t *x, Double_t *k)
|
---|
50 | {
|
---|
51 | const Double_t E0 = 511e-6; // [GeV]
|
---|
52 | const Double_t c = 299792458; // [m/s]
|
---|
53 | const Double_t e = 1.602176462e-19; // [C]
|
---|
54 | const Double_t h = 1e-9/e*6.62606876e-34; // [GeVs]
|
---|
55 | const Double_t a = 1./137; // [1]
|
---|
56 |
|
---|
57 | const Double_t o = x[0];
|
---|
58 |
|
---|
59 | const Double_t s1 = TMath::Pi()*2;
|
---|
60 | const Double_t s2 = a*h*c/E0; // [m]
|
---|
61 |
|
---|
62 | if (o<1e-4)
|
---|
63 | return s2*s2/s1 * 4/3;
|
---|
64 |
|
---|
65 | const Double_t o1 = o+1;
|
---|
66 | const Double_t o21 = o*2+1;
|
---|
67 |
|
---|
68 | const Double_t s3 = o1/(o*o*o);
|
---|
69 | const Double_t s4 = o*2*o1/o21;
|
---|
70 | const Double_t s5 = log(o21);
|
---|
71 | const Double_t s6 = s5/o/2;
|
---|
72 | const Double_t s7 = (o*3+1)/(o21*o21);
|
---|
73 |
|
---|
74 | return s2*s2/s1*(s3*(s4-s5)+s6-s7); // [m^2]
|
---|
75 | }
|
---|
76 |
|
---|
77 | Double_t MElectron::Li(Double_t *x, Double_t *k)
|
---|
78 | {
|
---|
79 | const Double_t t = x[0];
|
---|
80 | return log(1.-t)/t;
|
---|
81 | }
|
---|
82 |
|
---|
83 | Double_t MElectron::DiSum(Double_t *x, Double_t *k)
|
---|
84 | {
|
---|
85 | Double_t t = x[0];
|
---|
86 |
|
---|
87 | const Double_t eps = fabs(t*1e-2);
|
---|
88 |
|
---|
89 | Double_t disum = t;
|
---|
90 | Double_t add = 0;
|
---|
91 |
|
---|
92 | Int_t n = 2;
|
---|
93 | Double_t pow = t*t; // t^2
|
---|
94 |
|
---|
95 | do
|
---|
96 | {
|
---|
97 | add = pow/n/n;
|
---|
98 |
|
---|
99 | pow *= t; // pow = t^n
|
---|
100 | n++;
|
---|
101 |
|
---|
102 | disum += add;
|
---|
103 |
|
---|
104 | } while (fabs(add)>eps);
|
---|
105 |
|
---|
106 | return disum;
|
---|
107 | }
|
---|
108 |
|
---|
109 | Double_t MElectron::Li2(Double_t *x, Double_t *k)
|
---|
110 | {
|
---|
111 | //
|
---|
112 | // Dilog, Li2
|
---|
113 | // ----------
|
---|
114 | //
|
---|
115 | // Integral(0, 1) = konst;
|
---|
116 | // Double_t konst = 1./6*TMath::Pi()*TMath::Pi();
|
---|
117 | //
|
---|
118 | // x[0]: z
|
---|
119 | //
|
---|
120 | const Double_t z = x[0];
|
---|
121 |
|
---|
122 | if (fabs(z)<1)
|
---|
123 | return DiSum(x);
|
---|
124 |
|
---|
125 | // TF1 IntLi("Li", Li, 0, z, 0);
|
---|
126 | static TF1 IntLi("Li", Li, 0, 0, 0);
|
---|
127 | const Double_t integ = IntLi.Integral(0, z, (Double_t*)NULL, 1e-2);
|
---|
128 | return -integ;
|
---|
129 | }
|
---|
130 |
|
---|
131 | Double_t MElectron::Flim(Double_t *x, Double_t *k) // F(omegap)-F(omegam) mit b-->1 (Maple)
|
---|
132 | {
|
---|
133 | const Double_t w = x[0];
|
---|
134 |
|
---|
135 | const Double_t w4 = w*4;
|
---|
136 | const Double_t wsqr = w*w;
|
---|
137 |
|
---|
138 | const Double_t u1 = (w*wsqr*16 + wsqr*40 + w*17 + 2)*log(w4 + 1);
|
---|
139 | const Double_t u2 = -w4*(wsqr*2 + w*9 + 2);
|
---|
140 | const Double_t d = w4*(w4 + 1);
|
---|
141 |
|
---|
142 | Double_t s = -w*2*(1+1); // -2*omega*(1+beta)
|
---|
143 | const Double_t li2 = Li2(&s);
|
---|
144 |
|
---|
145 | const Double_t res = (u1+u2)/d + li2;
|
---|
146 |
|
---|
147 | return res; //<1e-10? 0 : res;
|
---|
148 | }
|
---|
149 |
|
---|
150 | Double_t MElectron::Compton(Double_t *x, Double_t *k)
|
---|
151 | {
|
---|
152 | const Double_t E0 = 511e-6; //[GeV]
|
---|
153 |
|
---|
154 | Double_t epsilon = x[0];
|
---|
155 | Double_t z = k[1];
|
---|
156 |
|
---|
157 | const Double_t E = k[0];
|
---|
158 |
|
---|
159 | Double_t flim;
|
---|
160 | if (epsilon<1e-14)
|
---|
161 | {
|
---|
162 | const Double_t d = E/(E0*E0);
|
---|
163 |
|
---|
164 | Double_t omega1 = 1e-13*d;
|
---|
165 | Double_t omega2 = 1e-12*d;
|
---|
166 |
|
---|
167 | const Double_t f1 = Flim(&omega1);
|
---|
168 | const Double_t f2 = Flim(&omega2);
|
---|
169 |
|
---|
170 | const Double_t m = log10(f2/f1);
|
---|
171 | const Double_t t = pow(f2, 13)/pow(f1, 12);
|
---|
172 |
|
---|
173 | flim = pow(epsilon, m) * t;
|
---|
174 | }
|
---|
175 | else
|
---|
176 | {
|
---|
177 | Double_t omega = epsilon*E/(E0*E0);
|
---|
178 | flim = Flim(&omega);
|
---|
179 | }
|
---|
180 |
|
---|
181 | const Double_t n = MParticle::Planck(&epsilon, &z)/epsilon/epsilon; // [1]
|
---|
182 | return flim*n;
|
---|
183 | }
|
---|
184 |
|
---|
185 | Double_t MElectron::InteractionLength(Double_t *E, Double_t *k)
|
---|
186 | {
|
---|
187 | // E = electron energy, ~ TeV(?) 1e12
|
---|
188 | // e = photon energy, ~ meV(?) 1e-3
|
---|
189 | // mc^2 = electron rest mass energy ~.5keV(?) .5e3
|
---|
190 | //
|
---|
191 | // x^-1 = int( n(epsilon)/2beta * ((mc^2)^2/eE)^2 * int ( omega*sigma(omega), omega=o-..o+), epsilon=0..inf)
|
---|
192 | //
|
---|
193 | // o+/- = omage_0 (1 +- beta)
|
---|
194 | //
|
---|
195 | // omega_0 = eE/(mc^2)^2 ~1e12*1e-3/.25e6=4e3
|
---|
196 | //
|
---|
197 | // --> x^-1 = (alpha*hc)^2/4pibetaE^2 * int(n(epsilon)/epsilon^2 *( F(o+)-F(o-)), epsilon=0..inf)
|
---|
198 | //
|
---|
199 | // F(o) = -o/4 + (9/4 + 1/o + o/2) * ln(1+2o) + 1/8(1+2o) - 3/8 + Li2(-2o)
|
---|
200 | //
|
---|
201 | // Li2(x) = int(ln(1-t)/t, t=0..x)
|
---|
202 | //
|
---|
203 | // F(o+)~F(2o) = -o/2 + (9/4 + 1/2o + o) * ln(1+4o) + 1/8(1+4o) - 3/8 + Li2(-4o)
|
---|
204 | // F(o-)~F(0) = 14/8 = 1.75
|
---|
205 |
|
---|
206 | const Double_t E0 = 511e-6; // [GeV]
|
---|
207 | const Double_t E02 = E0*E0; // [GeV^2]
|
---|
208 | const Double_t c = 299792458; // [m/s]
|
---|
209 | const Double_t e = 1.602176462e-19; // [C]
|
---|
210 | const Double_t h = 1e-9/e*6.62606876e-34; // [GeVs]
|
---|
211 | const Double_t hc = h*c; // [GeVm]
|
---|
212 | const Double_t alpha = 1./137.; // [1]
|
---|
213 |
|
---|
214 | const Double_t z = k ? k[0] : 0;
|
---|
215 |
|
---|
216 | /* -------------- old ----------------
|
---|
217 | Double_t from = 1e-15;
|
---|
218 | Double_t to = 1e-11;
|
---|
219 | eps = [default];
|
---|
220 | -----------------------------------
|
---|
221 | */
|
---|
222 | static TF1 func("Compton", Compton, 0, 0, 2); // [0, inf]
|
---|
223 |
|
---|
224 | const Double_t from = 1e-17;
|
---|
225 | const Double_t to = 2e-11;
|
---|
226 |
|
---|
227 | Double_t val[3] = { E[0], z }; // E[GeV]
|
---|
228 |
|
---|
229 | Double_t integ = func.Integral(from, to, val, 1e-2); // [Gev] [0, inf]
|
---|
230 |
|
---|
231 | const Double_t aE = alpha/E[0]; // [1/GeV]
|
---|
232 |
|
---|
233 | const Double_t beta = 1;
|
---|
234 |
|
---|
235 | const Double_t konst = 2.*E02/hc/beta; // [1 / GeV m]
|
---|
236 | const Double_t ret = konst * (aE*aE) * integ; // [1 / m]
|
---|
237 |
|
---|
238 | const Double_t ly = 3600.*24.*365.*c; // [m/ly]
|
---|
239 | const Double_t pc = 1./3.258; // [pc/ly]
|
---|
240 |
|
---|
241 | return (1./ret)/ly*pc/1000; // [kpc]
|
---|
242 | }
|
---|
243 |
|
---|
244 | Double_t MElectron::GetInteractionLength(Double_t energy, Double_t z)
|
---|
245 | {
|
---|
246 | return InteractionLength(&energy, &z);
|
---|
247 | }
|
---|
248 |
|
---|
249 | Double_t MElectron::GetInteractionLength() const
|
---|
250 | {
|
---|
251 | return InteractionLength((Double_t*)&fEnergy, (Double_t*)&fZ);
|
---|
252 | }
|
---|
253 |
|
---|
254 | // --------------------------------------------------------------------------
|
---|
255 |
|
---|
256 | /*inline*/ Double_t MElectron::p_e(Double_t *x, Double_t *k)
|
---|
257 | {
|
---|
258 | Double_t e = pow(10, x[0]);
|
---|
259 | return Compton(&e, k);
|
---|
260 | /*
|
---|
261 | Double_t z = k[1];
|
---|
262 |
|
---|
263 | const Double_t E = k[0];
|
---|
264 |
|
---|
265 | const Double_t E0 = 511e-6; //[GeV]
|
---|
266 | const Double_t E02 = E0*E0;
|
---|
267 |
|
---|
268 | Double_t omega = e*E/E02;
|
---|
269 |
|
---|
270 | const Double_t n = Planck(&e, &z);
|
---|
271 |
|
---|
272 | const Double_t F = Flim(&omega)/omega/omega;
|
---|
273 |
|
---|
274 | return n*F*1e26;
|
---|
275 | */
|
---|
276 | }
|
---|
277 |
|
---|
278 | Double_t MElectron::G_q(Double_t *x, Double_t *k)
|
---|
279 | {
|
---|
280 | const Double_t q = x[0];
|
---|
281 | const Double_t Gamma = k[0];
|
---|
282 |
|
---|
283 | const Double_t Gq = Gamma*q;
|
---|
284 |
|
---|
285 | const Double_t s1 = 2.*q*log(q);
|
---|
286 | const Double_t s2 = (1.+2.*q);
|
---|
287 | const Double_t s3 = (Gq*Gq)/(1.+Gq)/2.;
|
---|
288 |
|
---|
289 | return s1+(s2+s3)*(1.-q);
|
---|
290 | }
|
---|
291 |
|
---|
292 |
|
---|
293 | Double_t MElectron::EnergyLoss(Double_t *x, Double_t *k, Double_t *ep)
|
---|
294 | {
|
---|
295 | const Double_t E = x[0];
|
---|
296 | const Double_t z = k ? k[0] : 0;
|
---|
297 |
|
---|
298 | const Double_t E0 = 511e-6; //[GeV]
|
---|
299 |
|
---|
300 | const Double_t lolim = -log10(E)/7*4-13.5;
|
---|
301 |
|
---|
302 | static TF1 fP("p_e", p_e, lolim, -10.6, 2);
|
---|
303 | static TF1 fQ("G", G_q, 0, 1., 1);
|
---|
304 |
|
---|
305 | fP.SetNpx(50);
|
---|
306 | fQ.SetNpx(50);
|
---|
307 |
|
---|
308 | fP.SetRange(lolim, -10.6);
|
---|
309 | fP.SetParameter(0, E);
|
---|
310 | fP.SetParameter(1, z);
|
---|
311 |
|
---|
312 | const Double_t e = pow(10, fP.GetRandom());
|
---|
313 |
|
---|
314 | if (ep)
|
---|
315 | *ep = e;
|
---|
316 |
|
---|
317 | const Double_t omega = (e/E0)*(E/E0);
|
---|
318 | const Double_t Gamma = 4.*omega;
|
---|
319 |
|
---|
320 | fQ.SetParameter(0, Gamma);
|
---|
321 |
|
---|
322 | const Double_t q = fQ.GetRandom();
|
---|
323 | const Double_t Gq = Gamma*q;
|
---|
324 |
|
---|
325 | const Double_t e1 = Gq*E/(1.+Gq);
|
---|
326 |
|
---|
327 | return e1;
|
---|
328 | }
|
---|
329 |
|
---|
330 | Double_t MElectron::GetEnergyLoss(Double_t E, Double_t z, Double_t *ep)
|
---|
331 | {
|
---|
332 | return EnergyLoss(&E, &z);
|
---|
333 | }
|
---|
334 |
|
---|
335 | Double_t MElectron::GetEnergyLoss(Double_t *ep) const
|
---|
336 | {
|
---|
337 | return EnergyLoss((Double_t*)&fEnergy, (Double_t*)&fZ, ep);
|
---|
338 | }
|
---|
339 |
|
---|
340 |
|
---|
341 | Double_t Omega_sigmae(Double_t *x, Double_t *k)
|
---|
342 | {
|
---|
343 | Double_t sbar = pow(10,x[0]);
|
---|
344 |
|
---|
345 | Double_t omega = (sbar-1)/2;
|
---|
346 |
|
---|
347 | Double_t sigma = MElectron::Sigma_ge(&omega);
|
---|
348 |
|
---|
349 | return (sbar-1)*sigma*1e28;
|
---|
350 | }
|
---|
351 |
|
---|
352 | Double_t RandomThetaE(Double_t Ee, Double_t Ep)
|
---|
353 | {
|
---|
354 | Double_t E0 = 511e-6; // [GeV]
|
---|
355 |
|
---|
356 | Double_t f = 2*Ee/E0*Ep/E0;
|
---|
357 |
|
---|
358 | static TF1 func("RndThetaE", Omega_sigmae, 0, 0, 0);
|
---|
359 |
|
---|
360 | Double_t beta = sqrt(1-E0/Ee*E0/Ee);
|
---|
361 |
|
---|
362 | //func.SetRange(0, log10(1+f*(1+beta)));
|
---|
363 | func.SetRange(log10(1+f*(1-beta)), log10(1+f*(1+beta)));
|
---|
364 | func.SetNpx(50);
|
---|
365 |
|
---|
366 | Double_t sbar = pow(10, func.GetRandom());
|
---|
367 |
|
---|
368 | Double_t bcost = 1 - (sbar-1)/f;
|
---|
369 | return bcost;
|
---|
370 |
|
---|
371 | /*
|
---|
372 | Double_t theta = acos(bcost/beta);
|
---|
373 | return theta;
|
---|
374 | */
|
---|
375 | }
|
---|
376 |
|
---|
377 | MPhoton *MElectron::DoInvCompton()
|
---|
378 | {
|
---|
379 | const Double_t E0 = 511e-6; //[GeV]
|
---|
380 |
|
---|
381 | Double_t epsilon;
|
---|
382 | const Double_t e = GetEnergyLoss(&epsilon);
|
---|
383 |
|
---|
384 | // epsilon: photon energy befor interaction, lab
|
---|
385 | // e: photon energy after interaction, lab
|
---|
386 |
|
---|
387 | const Double_t gamma = fEnergy/E0;
|
---|
388 | const Double_t gammabeta = sqrt(gamma*gamma-1);
|
---|
389 | const Double_t beta = gammabeta/gamma;
|
---|
390 |
|
---|
391 | const Double_t bcost = RandomThetaE(fEnergy, epsilon);
|
---|
392 | const Double_t t = acos(bcost/beta);
|
---|
393 |
|
---|
394 | const Double_t f = epsilon/fEnergy;
|
---|
395 | const Double_t r = gamma*(1-bcost);
|
---|
396 |
|
---|
397 | Double_t arg = (1 - 1/(gamma*r) - f)/(beta-f);
|
---|
398 |
|
---|
399 | if (arg<-1 || arg>1)
|
---|
400 | cout << "<" << (int)(t*180/TMath::Pi()) << "°>" << flush;
|
---|
401 |
|
---|
402 | //
|
---|
403 | // Due to numerical uncertanties arg can be something like:
|
---|
404 | // 1+1e-15 which we cannot allow
|
---|
405 | //
|
---|
406 | if (arg<-1)
|
---|
407 | arg = -1;
|
---|
408 | if (arg>1)
|
---|
409 | arg = 1;
|
---|
410 |
|
---|
411 | const Double_t theta1s = acos(arg);
|
---|
412 | const Double_t thetas = atan(-sin(t)/(beta*r)/*(1-bcost)/gammabeta*/);
|
---|
413 |
|
---|
414 | const Double_t thetastar = thetas-theta1s;
|
---|
415 |
|
---|
416 | // const Double_t theta1 = atan(sin(thetastar)/(gamma*cos(thetastar)+gammabeta));
|
---|
417 | const Double_t theta1 = atan(sin(thetastar)/(gamma*(cos(thetastar)+beta)));
|
---|
418 |
|
---|
419 | fEnergy -= e;
|
---|
420 |
|
---|
421 | const Double_t phi = gRandom->Uniform(TMath::Pi()*2);
|
---|
422 |
|
---|
423 | MPhoton &p = *new MPhoton(*this, e);
|
---|
424 | p.SetNewDirection(theta1, phi);
|
---|
425 |
|
---|
426 | /*
|
---|
427 | const Double_t beta2 = sqrt(1.-E0/fEnergy*E0/fEnergy);
|
---|
428 | const Double_t theta2 = asin((epsilon*sin(t)-e*sin(theta1))/fEnergy/beta2);
|
---|
429 | */
|
---|
430 |
|
---|
431 | const Double_t div = sqrt(fEnergy*fEnergy-E0*E0);
|
---|
432 | const Double_t theta2 = asin((epsilon*sin(t)-e*sin(theta1))/div);
|
---|
433 |
|
---|
434 | SetNewDirection(theta2, phi);
|
---|
435 |
|
---|
436 | return &p;
|
---|
437 | }
|
---|
438 |
|
---|
439 | void MElectron::DrawInteractionLength(Double_t z)
|
---|
440 | {
|
---|
441 | if (!gPad)
|
---|
442 | new TCanvas("IL_Electron", "Mean Interaction Length Electron");
|
---|
443 | else
|
---|
444 | gPad->GetVirtCanvas()->cd(3);
|
---|
445 |
|
---|
446 | TF1 f2("length", MElectron::InteractionLength, 1e3, 1e10, 0);
|
---|
447 | f2.SetParameter(0, z);
|
---|
448 |
|
---|
449 | gPad->SetLogx();
|
---|
450 | gPad->SetLogy();
|
---|
451 | gPad->SetGrid();
|
---|
452 | f2.SetLineWidth(1);
|
---|
453 |
|
---|
454 | TH1 &h=*f2.DrawCopy()->GetHistogram();
|
---|
455 |
|
---|
456 | h.SetTitle("Mean Interaction Length (Electron)");
|
---|
457 | h.SetXTitle("E [GeV]");
|
---|
458 | h.SetYTitle("x [kpc]");
|
---|
459 |
|
---|
460 | gPad->Modified();
|
---|
461 | gPad->Update();
|
---|
462 | }
|
---|
463 |
|
---|
464 | void MElectron::DrawInteractionLength() const
|
---|
465 | {
|
---|
466 | DrawInteractionLength(fZ);
|
---|
467 | }
|
---|
468 |
|
---|
469 | Bool_t MElectron::SetNewPositionB(Double_t B)
|
---|
470 | {
|
---|
471 | if (B==0)
|
---|
472 | return SetNewPosition();
|
---|
473 |
|
---|
474 | Double_t x = gRandom->Exp(GetInteractionLength());
|
---|
475 |
|
---|
476 | // -----------------------------
|
---|
477 |
|
---|
478 | const Double_t E0 = 511e-6; //[GeV]
|
---|
479 |
|
---|
480 | Double_t B_theta = gRandom->Uniform(TMath::Pi());
|
---|
481 | Double_t B_phi = gRandom->Uniform(TMath::Pi()*2);
|
---|
482 |
|
---|
483 | static TMatrixD M(3,3);
|
---|
484 |
|
---|
485 | M(0, 0) = sin(B_theta)*cos(B_phi);
|
---|
486 | M(1, 0) = cos(B_theta)*cos(B_phi);
|
---|
487 | M(2, 0) = -sin(B_phi);
|
---|
488 |
|
---|
489 | M(0, 1) = sin(B_theta)*sin(B_phi);
|
---|
490 | M(1, 1) = cos(B_theta)*sin(B_phi);
|
---|
491 | M(2, 1) = cos(B_phi);
|
---|
492 |
|
---|
493 | M(0, 2) = cos(B_theta);
|
---|
494 | M(1, 2) = -sin(B_theta);
|
---|
495 | M(2, 2) = 0;
|
---|
496 |
|
---|
497 | const Double_t beta = sqrt(1.-E0/fEnergy*E0/fEnergy);
|
---|
498 |
|
---|
499 | //
|
---|
500 | // rotate vector of velocity into a system in
|
---|
501 | // which the x-axis is identical with the B-Field
|
---|
502 | //
|
---|
503 | static TVectorD v(3);
|
---|
504 | v(0) = beta*sin(fTheta)*cos(fPsi);
|
---|
505 | v(1) = beta*sin(fTheta)*sin(fPsi);
|
---|
506 | v(2) = beta*cos(fTheta);
|
---|
507 | v *= M;
|
---|
508 |
|
---|
509 | //
|
---|
510 | // Now rotate the system so, that the velocity vector
|
---|
511 | // lays in the y-z-plain
|
---|
512 | //
|
---|
513 | Double_t chi = atan(-v(2)/v(1));
|
---|
514 |
|
---|
515 | // -----------------------------
|
---|
516 |
|
---|
517 | static TMatrixD N(3,3);
|
---|
518 | N(0, 0) = 1;
|
---|
519 | N(1, 0) = 0;
|
---|
520 | N(2, 0) = 0;
|
---|
521 |
|
---|
522 | N(0, 1) = 0;
|
---|
523 | N(1, 1) = -cos(chi);
|
---|
524 | N(2, 1) = -sin(chi);
|
---|
525 |
|
---|
526 | N(0, 2) = 0;
|
---|
527 | N(1, 2) = sin(chi);
|
---|
528 | N(2, 2) = -cos(chi);
|
---|
529 | v *= N;
|
---|
530 |
|
---|
531 | Double_t beta_p = v(0); // beta parallel
|
---|
532 | Double_t beta_o = v(1); // beta orthogonal
|
---|
533 | // v(2) = 0
|
---|
534 | // -----------------------------
|
---|
535 | static TVectorD p(3);
|
---|
536 |
|
---|
537 | Double_t rho = 0;
|
---|
538 | if (B>0)
|
---|
539 | {
|
---|
540 | const Double_t c = 299792458; // [m/s]
|
---|
541 | const Double_t ly = 3600.*24.*365.*c; // [m/ly]
|
---|
542 | const Double_t pc = 1./3.258; // [pc/ly]
|
---|
543 |
|
---|
544 | Double_t r = (fEnergy*1e9)*beta_o/(c*B)*(pc/ly/1e3); // [kpc]
|
---|
545 |
|
---|
546 | rho = beta_o/beta*x/r; // [2pi]
|
---|
547 |
|
---|
548 | // -----------------------------
|
---|
549 |
|
---|
550 | if (fabs(rho*3437)>5*60) // > 5*60min
|
---|
551 | {
|
---|
552 | cout << "r" << flush;
|
---|
553 | return kFALSE;
|
---|
554 | }
|
---|
555 |
|
---|
556 | p(0) = beta_p/beta*x;
|
---|
557 | p(1) = GetPType()==kEElectron?r*sin(rho):-r*sin(rho);
|
---|
558 | p(2) = r*(1.-cos(rho));
|
---|
559 | }
|
---|
560 | else
|
---|
561 | {
|
---|
562 | p(0) = beta_p/beta*x;
|
---|
563 | p(1) = beta_o/beta*x;
|
---|
564 | p(2) = 0;
|
---|
565 | cout << "------------- HEY! --------------" << endl;
|
---|
566 | }
|
---|
567 |
|
---|
568 | static TVectorD s(3);
|
---|
569 | s(0) = fR*cos(fPhi);
|
---|
570 | s(1) = fR*sin(fPhi);
|
---|
571 | s(2) = RofZ(&fZ);
|
---|
572 |
|
---|
573 | TMatrixD N2(TMatrixD::kTransposed, N);
|
---|
574 | TMatrixD M2(TMatrixD::kTransposed, M);
|
---|
575 | p *= N2;
|
---|
576 | p *= M2;
|
---|
577 |
|
---|
578 | if (p(2)<x) // happens sometimes in case B==0
|
---|
579 | {
|
---|
580 | cout << "----- HA: " << B << " " << x << " " << p(2) << " " << x-p(2) << endl;
|
---|
581 | p(2)=x;
|
---|
582 | }
|
---|
583 |
|
---|
584 | s -= p;
|
---|
585 |
|
---|
586 | fR = sqrt(s(0)*s(0)+s(1)*s(1));
|
---|
587 | fPhi = atan2(s(1), s(0));
|
---|
588 | fZ = ZofR(&s(2));
|
---|
589 | fX += x-p(2);
|
---|
590 |
|
---|
591 | // -----------------------------
|
---|
592 |
|
---|
593 | static TVectorD w(3);
|
---|
594 | w(0) = beta_p/beta;
|
---|
595 | w(1) = beta_o/beta*cos(rho);
|
---|
596 | w(2) = beta_o/beta*sin(rho);
|
---|
597 |
|
---|
598 | w *= N2;
|
---|
599 | w *= M2;
|
---|
600 |
|
---|
601 | fPsi = atan2(w(1), w(0));
|
---|
602 | fTheta = asin(sqrt(w(0)*w(0)+w(1)*w(1)));
|
---|
603 |
|
---|
604 | // -----------------------------
|
---|
605 |
|
---|
606 | return fZ<0 ? kFALSE : kTRUE;
|
---|
607 | }
|
---|