CMS 3D CMS Logo

EMShower.cc
Go to the documentation of this file.
1 //FAMOS Headers
6 
9 
10 #include <cmath>
11 
12 //#include "FastSimulation/Utilities/interface/Histos.h"
13 
14 using std::vector;
15 
18  EMECALShowerParametrization* const myParam,
19  vector<const RawParticle*>* const myPart,
20  EcalHitMaker* const myGrid,
21  PreshowerHitMaker* const myPresh,
22  bool bFixedLength)
23 
24  : theParam(myParam),
25  thePart(myPart),
26  theGrid(myGrid),
27  thePreshower(myPresh),
28  random(engine),
29  myGammaGenerator(gamma),
30  bFixedLength_(bFixedLength) {
31  // Get the Famos Histos pointer
32  // myHistos = Histos::instance();
33  // myGammaGenerator = GammaFunctionGenerator::instance();
34  stepsCalculated = false;
35  hasPreshower = myPresh != nullptr;
36  theECAL = myParam->ecalProperties();
37  theHCAL = myParam->hcalProperties();
38  theLayer1 = myParam->layer1Properties();
39  theLayer2 = myParam->layer2Properties();
40 
42 
43  nPart = thePart->size();
44  totalEnergy = 0.;
45  globalMaximum = 0.;
46  double meanDepth = 0.;
47  // Initialize the shower parameters for each particle
48 
49  for (unsigned int i = 0; i < nPart; ++i) {
50  // std::cout << " AAA " << *(*thePart)[i] << std::endl;
51  // The particle and the shower energy
52  Etot.push_back(0.);
53  E.push_back(((*thePart)[i])->e());
54  totalEnergy += E[i];
55 
56  double lny = std::log(E[i] / theECAL->criticalEnergy());
57 
58  // Average and Sigma for T and alpha
59  double theMeanT = myParam->meanT(lny);
60  double theMeanAlpha = myParam->meanAlpha(lny);
61  double theMeanLnT = myParam->meanLnT(lny);
62  double theMeanLnAlpha = myParam->meanLnAlpha(lny);
63  double theSigmaLnT = myParam->sigmaLnT(lny);
64  double theSigmaLnAlpha = myParam->sigmaLnAlpha(lny);
65 
66  // The correlation matrix
67  double theCorrelation = myParam->correlationAlphaT(lny);
68  double rhop = std::sqrt((1. + theCorrelation) / 2.);
69  double rhom = std::sqrt((1. - theCorrelation) / 2.);
70 
71  // The number of spots in ECAL / HCAL
72  theNumberOfSpots.push_back(myParam->nSpots(E[i]));
73  // theNumberOfSpots.push_back(myParam->nSpots(E[i])*spotFraction);
74  //theNumberOfSpots = random->poissonShoot(myParam->nSpots(myPart->e()));
75 
76  // Photo-statistics
77  photos.push_back(E[i] * fotos);
78 
79  // The longitudinal shower development parameters
80  // Fluctuations of alpha, T and beta
81  double z1 = 0.;
82  double z2 = 0.;
83  double aa = 0.;
84 
85  // Protect against too large fluctuations (a < 1) for small energies
86  while (aa <= 1.) {
87  z1 = random->gaussShoot(0., 1.);
88  z2 = random->gaussShoot(0., 1.);
89  aa = std::exp(theMeanLnAlpha + theSigmaLnAlpha * (z1 * rhop - z2 * rhom));
90  }
91 
92  a.push_back(aa);
93  T.push_back(std::exp(theMeanLnT + theSigmaLnT * (z1 * rhop + z2 * rhom)));
94  b.push_back((a[i] - 1.) / T[i]);
95  maximumOfShower.push_back((a[i] - 1.) / b[i]);
97  meanDepth += a[i] / b[i] * E[i];
98  // std::cout << " Adding max " << maximumOfShower[i] << " " << E[i] << " " <<maximumOfShower[i]*E[i] << std::endl;
99  // std::cout << std::setw(8) << std::setprecision(5) << " a /b " << a[i] << " " << b[i] << std::endl;
100  Ti.push_back(a[i] / b[i] * (std::exp(theMeanLnAlpha) - 1.) / std::exp(theMeanLnAlpha));
101 
102  // The parameters for the number of energy spots
103  TSpot.push_back(theParam->meanTSpot(theMeanT));
104  aSpot.push_back(theParam->meanAlphaSpot(theMeanAlpha));
105  bSpot.push_back((aSpot[i] - 1.) / TSpot[i]);
106  // myHistos->fill("h7000",a[i]);
107  // myHistos->fill("h7002",E[i],a[i]);
108  }
109  // std::cout << " PS1 : " << myGrid->ps1TotalX0()
110  // << " PS2 : " << myGrid->ps2TotalX0()
111  // << " ECAL : " << myGrid->ecalTotalX0()
112  // << " HCAL : " << myGrid->hcalTotalX0()
113  // << " Offset : " << myGrid->x0DepthOffset()
114  // << std::endl;
115 
118  // std::cout << " Total Energy " << totalEnergy << " Global max " << globalMaximum << std::endl;
119 }
120 
122  // TimeMe theT("EMShower::compute");
123 
124  // Determine the longitudinal intervals
125  // std::cout << " EMShower compute" << std::endl;
126  double dt;
127  double radlen;
128  int stps;
129  int first_Ecal_step = 0;
130  int last_Ecal_step = 0;
131 
132  // The maximum is in principe 8 (with 5X0 steps in the ECAL)
133  steps.reserve(24);
134 
135  /*
136  std::cout << " PS1 : " << theGrid->ps1TotalX0()
137  << " PS2 : " << theGrid->ps2TotalX0()
138  << " PS2 and ECAL : " << theGrid->ps2eeTotalX0()
139  << " ECAL : " << theGrid->ecalTotalX0()
140  << " HCAL : " << theGrid->hcalTotalX0()
141  << " Offset : " << theGrid->x0DepthOffset()
142  << std::endl;
143  */
144 
145  radlen = -theGrid->x0DepthOffset();
146 
147  // Preshower Layer 1
148  radlen += theGrid->ps1TotalX0();
149  if (radlen > 0.) {
150  steps.push_back(Step(0, radlen));
151  radlen = 0.;
152  }
153 
154  // Preshower Layer 2
155  radlen += theGrid->ps2TotalX0();
156  if (radlen > 0.) {
157  steps.push_back(Step(1, radlen));
158  radlen = 0.;
159  }
160 
161  // add a step between preshower and ee
162  radlen += theGrid->ps2eeTotalX0();
163  if (radlen > 0.) {
164  steps.push_back(Step(5, radlen));
165  radlen = 0.;
166  }
167 
168  // ECAL
169  radlen += theGrid->ecalTotalX0();
170 
171  // std::cout << "theGrid->ecalTotalX0() = " << theGrid->ecalTotalX0() << std::endl;
172 
173  if (radlen > 0.) {
174  if (!bFixedLength_) {
175  stps = (int)((radlen + 2.5) / 5.);
176  // stps=(int)((radlen+.5)/1.);
177  if (stps == 0)
178  stps = 1;
179  dt = radlen / (double)stps;
180  Step step(2, dt);
181  first_Ecal_step = steps.size();
182  for (int ist = 0; ist < stps; ++ist)
183  steps.push_back(step);
184  last_Ecal_step = steps.size() - 1;
185  radlen = 0.;
186  } else {
187  dt = 1.0;
188  stps = static_cast<int>(radlen);
189  if (stps == 0)
190  stps = 1;
191  Step step(2, dt);
192  first_Ecal_step = steps.size();
193  for (int ist = 0; ist < stps; ++ist)
194  steps.push_back(step);
195  dt = radlen - stps;
196  if (dt > 0) {
197  Step stepLast(2, dt);
198  steps.push_back(stepLast);
199  }
200  last_Ecal_step = steps.size() - 1;
201  // std::cout << "radlen = " << radlen << " stps = " << stps << " dt = " << dt << std::endl;
202  radlen = 0.;
203  }
204  }
205 
206  // I should had a gap here !
207 
208  // HCAL
209  radlen += theGrid->hcalTotalX0();
210  if (radlen > 0.) {
211  double dtFrontHcal = theGrid->totalX0() - theGrid->hcalTotalX0();
212  // One single step for the full HCAL
213  if (dtFrontHcal < 30.) {
214  dt = 30. - dtFrontHcal;
215  Step step(3, dt);
216  steps.push_back(step);
217  }
218  }
219 
220  nSteps = steps.size();
221  if (nSteps == 0)
222  return;
223  double ESliceTot = 0.;
224  double MeanDepth = 0.;
225  depositedEnergy.resize(nSteps);
226  meanDepth.resize(nSteps);
227  double t = 0.;
228 
229  int offset = 0;
230  for (unsigned iStep = 0; iStep < nSteps; ++iStep) {
231  ESliceTot = 0.;
232  MeanDepth = 0.;
233  double realTotalEnergy = 0;
234  dt = steps[iStep].second;
235  t += dt;
236  for (unsigned int i = 0; i < nPart; ++i) {
237  depositedEnergy[iStep].push_back(deposit(t, a[i], b[i], dt));
238  ESliceTot += depositedEnergy[iStep][i];
239  MeanDepth += deposit(t, a[i] + 1., b[i], dt) / b[i] * a[i];
240  realTotalEnergy += depositedEnergy[iStep][i] * E[i];
241  }
242 
243  if (ESliceTot > 0.) // can happen for the shower tails; this depth will be skipped anyway
244  MeanDepth /= ESliceTot;
245  else
246  MeanDepth = t - dt;
247 
248  meanDepth[iStep] = MeanDepth;
249  if (realTotalEnergy < 0.001) {
250  offset -= 1;
251  }
252  }
253 
254  innerDepth = meanDepth[first_Ecal_step];
255  if (last_Ecal_step + offset >= 0)
256  outerDepth = meanDepth[last_Ecal_step + offset];
257  else
259 
260  stepsCalculated = true;
261 }
262 
264  double t = 0.;
265  double dt = 0.;
266  if (!stepsCalculated)
267  prepareSteps();
268 
269  // Prepare the grids in EcalHitMaker
270  // theGrid->setInnerAndOuterDepth(innerDepth,outerDepth);
271  float pstot = 0.;
272  float ps2tot = 0.;
273  float ps1tot = 0.;
274  bool status = false;
275  // double E1 = 0.; // Energy layer 1
276  // double E2 = 0.; // Energy layer 2
277  // double n1 = 0.; // #mips layer 1
278  // double n2 = 0.; // #mips layer 2
279  // double E9 = 0.; // Energy ECAL
280 
281  // Loop over all segments for the longitudinal development
282  double totECalc = 0;
283 
284  for (unsigned iStep = 0; iStep < nSteps; ++iStep) {
285  // The length of the shower in this segment
286  dt = steps[iStep].second;
287 
288  // std::cout << " Detector " << steps[iStep].first << " t " << t << " " << dt << std::endl;
289 
290  // The elapsed length
291  t += dt;
292 
293  // In what detector are we ?
294  unsigned detector = steps[iStep].first;
295 
296  bool presh1 = detector == 0;
297  bool presh2 = detector == 1;
298  bool ecal = detector == 2;
299  bool hcal = detector == 3;
300  bool vfcal = detector == 4;
301  bool gap = detector == 5;
302 
303  // Temporary. Will be removed
304  if (theHCAL == nullptr)
305  hcal = false;
306 
307  // Keep only ECAL for now
308  if (vfcal)
309  continue;
310 
311  // Nothing to do in the gap
312  if (gap)
313  continue;
314 
315  // cout << " t = " << t << endl;
316  // Build the grid of crystals at this ECAL depth
317  // Actually, it might be useful to check if this grid is empty or not.
318  // If it is empty (because no crystal at this depth), it is of no use
319  // (and time consuming) to generate the spots
320 
321  // middle of the step
322  double tt = t - 0.5 * dt;
323 
324  double realTotalEnergy = 0.;
325  for (unsigned int i = 0; i < nPart; ++i) {
326  realTotalEnergy += depositedEnergy[iStep][i] * E[i];
327  }
328 
329  // std::cout << " Step " << tt << std::endl;
330  // std::cout << "ecal " << ecal << " hcal " << hcal <<std::endl;
331 
332  // If the amount of energy is greater than 1 MeV, make a new grid
333  // otherwise put in the previous one.
334  bool usePreviousGrid = (realTotalEnergy < 0.001);
335 
336  // If the amount of energy is greater than 1 MeV, make a new grid
337  // otherwise put in the previous one.
338 
339  // If less than 1 kEV. Just skip
340  if (iStep > 2 && realTotalEnergy < 0.000001)
341  continue;
342 
343  if (ecal && !usePreviousGrid) {
344  status = theGrid->getPads(meanDepth[iStep]);
345  }
346  if (hcal) {
348  }
349  if ((ecal || hcal) && !status)
350  continue;
351 
352  bool detailedShowerTail = false;
353  // check if a detailed treatment of the rear leakage should be applied
354  if (ecal && !usePreviousGrid) {
355  detailedShowerTail = (t - dt > theGrid->getX0back());
356  }
357 
358  // The particles of the shower are processed in parallel
359  for (unsigned int i = 0; i < nPart; ++i) {
360  // double Edepo=deposit(t,a[i],b[i],dt);
361 
362  // integration of the shower profile between t-dt and t
363  double dE = (!hcal) ? depositedEnergy[iStep][i] : 1. - deposit(a[i], b[i], t - dt);
364 
365  // no need to do the full machinery if there is ~nothing to distribute)
366  if (dE * E[i] < 0.000001)
367  continue;
368 
369  if (ecal && !theECAL->isHom()) {
370  double mean = dE * E[i];
371  double sigma = theECAL->resE() * sqrt(mean);
372 
373  /*
374  double meanLn = log(mean);
375  double kLn = sigma/mean+1;
376  double sigmaLn = log(kLn);
377  */
378 
379  double dE0 = dE;
380 
381  // std::cout << "dE before shoot = " << dE << std::endl;
382  dE = random->gaussShoot(mean, sigma) / E[i];
383 
384  // myGammaGenerator->setParameters(aSam,bSam,0);
385  // dE = myGammaGenerator->shoot()/E[i];
386  // std::cout << "dE shooted = " << dE << " E[i] = " << E[i] << std::endl;
387  if (dE * E[i] < 0.000001)
388  continue;
389  photos[i] = photos[i] * dE / dE0;
390  }
391 
392  /*
393  if (ecal && !theParam->ecalProperties()->isHom()){
394 
395  double cSquare = TMath::Power(theParam->ecalProperties()->resE(),2);
396  double aSam = dE/cSquare;
397  double bSam = 1./cSquare;
398 
399  // dE = dE*gam(bSam*dE, aSam)/tgamma(aSam);
400  }
401  */
402 
403  totECalc += dE;
404 
405  // The number of energy spots (or mips)
406  double nS = 0;
407 
408  // ECAL case : Account for photostatistics and long'al non-uniformity
409  if (ecal) {
410  // double aSam = E[i]*dE*one_over_resoSquare;
411  // double bSam = one_over_resoSquare;
412 
413  dE = random->poissonShoot(dE * photos[i]) / photos[i];
414  double z0 = random->gaussShoot(0., 1.);
415  dE *= 1. + z0 * theECAL->lightCollectionUniformity();
416 
417  // Expected spot number
418  nS = (theNumberOfSpots[i] * gam(bSpot[i] * tt, aSpot[i]) * bSpot[i] * dt / tgamma(aSpot[i]));
419 
420  // Preshower : Expected number of mips + fluctuation
421  } else if (hcal) {
422  nS = (theNumberOfSpots[i] * gam(bSpot[i] * tt, aSpot[i]) * bSpot[i] * dt / tgamma(aSpot[i])) *
424  double nSo = nS;
425  nS = random->poissonShoot(nS);
426  // 'Quick and dirty' fix (but this line should be better removed):
427  if (nSo > 0. && nS / nSo < 10.)
428  dE *= nS / nSo;
429 
430  // if(true)
431  // {
432  // std::cout << " theHCAL->spotFraction = " <<theHCAL->spotFraction() <<std::endl;
433  // std::cout << " nSpot Ecal : " << nSo/theHCAL->spotFraction() << " Final " << nS << std::endl;
434  // }
435  } else if (presh1) {
436  nS = random->poissonShoot(dE * E[i] * theLayer1->mipsPerGeV());
437  // std::cout << " dE *E[i] (1)" << dE*E[i] << " " << dE*E[i]*theLayer1->mipsPerGeV() << " "<< nS << std::endl;
438  pstot += dE * E[i];
439  ps1tot += dE * E[i];
440  dE = nS / (E[i] * theLayer1->mipsPerGeV());
441 
442  // E1 += dE*E[i];
443  // n1 += nS;
444  // if (presh2) { E2 += SpotEnergy; ++n2; }
445 
446  } else if (presh2) {
447  nS = random->poissonShoot(dE * E[i] * theLayer2->mipsPerGeV());
448  // std::cout << " dE *E[i] (2) " << dE*E[i] << " " << dE*E[i]*theLayer2->mipsPerGeV() << " "<< nS << std::endl;
449  pstot += dE * E[i];
450  ps2tot += dE * E[i];
451  dE = nS / (E[i] * theLayer2->mipsPerGeV());
452 
453  // E2 += dE*E[i];
454  // n2 += nS;
455  }
456 
457  if (detailedShowerTail)
458  myGammaGenerator->setParameters(floor(a[i] + 0.5), b[i], t - dt);
459 
460  // myHistos->fill("h100",t,dE);
461 
462  // The lateral development parameters
463 
464  // Energy of the spots
465  double eSpot = (nS > 0.) ? dE / nS : 0.;
466  double SpotEnergy = eSpot * E[i];
467 
468  if (hasPreshower && (presh1 || presh2))
469  thePreshower->setSpotEnergy(0.00009);
470  if (hcal) {
471  SpotEnergy *= theHCAL->hOverPi();
472  theHcalHitMaker->setSpotEnergy(SpotEnergy);
473  }
474  // Poissonian fluctuations for the number of spots
475  // int nSpot = random->poissonShoot(nS);
476  int nSpot = (int)(nS + 0.5);
477 
478  // Fig. 11 (right) *** Does not match.
479  // myHistos->fill("h101",t,(double)nSpot/theNumberOfSpots);
480 
481  //double taui = t/T;
482  double taui = tt / Ti[i];
483  double proba = theParam->p(taui, E[i]);
484  double theRC = theParam->rC(taui, E[i]);
485  double theRT = theParam->rT(taui, E[i]);
486 
487  // Fig. 10
488  // myHistos->fill("h300",taui,theRC);
489  // myHistos->fill("h301",taui,theRT);
490  // myHistos->fill("h302",taui,proba);
491 
492  double dSpotsCore = random->gaussShoot(proba * nSpot, std::sqrt(proba * (1. - proba) * nSpot));
493 
494  if (dSpotsCore < 0)
495  dSpotsCore = 0;
496 
497  unsigned nSpots_core = (unsigned)(dSpotsCore + 0.5);
498  unsigned nSpots_tail = ((unsigned)nSpot > nSpots_core) ? nSpot - nSpots_core : 0;
499 
500  for (unsigned icomp = 0; icomp < 2; ++icomp) {
501  double theR = (icomp == 0) ? theRC : theRT;
502  unsigned ncompspots = (icomp == 0) ? nSpots_core : nSpots_tail;
503 
504  RadialInterval radInterval(theR, ncompspots, SpotEnergy, random);
505  if (ecal) {
506  if (icomp == 0) {
507  setIntervals(icomp, radInterval);
508  } else {
509  setIntervals(icomp, radInterval);
510  }
511  } else {
512  radInterval.addInterval(100., 1.); // 100% of the spots
513  }
514 
515  radInterval.compute();
516  // irad = 0 : central circle; irad=1 : outside
517 
518  unsigned nrad = radInterval.nIntervals();
519 
520  for (unsigned irad = 0; irad < nrad; ++irad) {
521  double spote = radInterval.getSpotEnergy(irad);
522  if (ecal)
523  theGrid->setSpotEnergy(spote);
524  if (hcal)
526  unsigned nradspots = radInterval.getNumberOfSpots(irad);
527  double umin = radInterval.getUmin(irad);
528  double umax = radInterval.getUmax(irad);
529  // Go for the lateral development
530  // std::cout << "Couche " << iStep << " irad = " << irad << " Ene = " << E[i] << " eSpot = " << eSpot << " spote = " << spote << " nSpot = " << nS << std::endl;
531 
532  for (unsigned ispot = 0; ispot < nradspots; ++ispot) {
533  double z3 = random->flatShoot(umin, umax);
534  double ri = theR * std::sqrt(z3 / (1. - z3));
535 
536  // Generate phi
537  double phi = 2. * M_PI * random->flatShoot();
538 
539  // Add the hit in the crystal
540  // if( ecal ) theGrid->addHit(ri*theECAL->moliereRadius(),phi);
541  // Now the *moliereRadius is done in EcalHitMaker
542  if (ecal) {
543  if (detailedShowerTail) {
544  // std::cout << "About to call addHitDepth " << std::endl;
545  double depth;
546  do {
548  } while (depth > t);
549  theGrid->addHitDepth(ri, phi, depth);
550  // std::cout << " Done " << std::endl;
551  } else
552  theGrid->addHit(ri, phi);
553  } else if (hasPreshower && presh1)
554  thePreshower->addHit(ri, phi, 1);
555  else if (hasPreshower && presh2)
556  thePreshower->addHit(ri, phi, 2);
557  else if (hcal) {
558  // std::cout << " About to add a spot in the HCAL" << status << std::endl;
559  theHcalHitMaker->addHit(ri, phi);
560  // std::cout << " Added a spot in the HCAL" << status << std::endl;
561  }
562  // if (ecal) E9 += SpotEnergy;
563  // if (presh1) { E1 += SpotEnergy; ++n1; }
564  // if (presh2) { E2 += SpotEnergy; ++n2; }
565 
566  Etot[i] += spote;
567  }
568  }
569  }
570  // std::cout << " Done with the step " << std::endl;
571  // The shower!
572  //myHistos->fill("h500",theSpot.z(),theSpot.perp());
573  }
574  // std::cout << " nPart " << nPart << std::endl;
575  }
576  // std::cout << " Finshed the step loop " << std::endl;
577  // myHistos->fill("h500",E1+0.7*E2,E9);
578  // myHistos->fill("h501",n1+0.7*n2,E9);
579  // myHistos->fill("h400",n1);
580  // myHistos->fill("h401",n2);
581  // myHistos->fill("h402",E9+E1+0.7*E2);
582  // if(!standalone)theGrid->printGrid();
583  double Etotal = 0.;
584  for (unsigned i = 0; i < nPart; ++i) {
585  // myHistos->fill("h10",Etot[i]);
586  Etotal += Etot[i];
587  }
588 
589  // std::cout << "Etotal = " << Etotal << " nPart = "<< nPart << std::endl;
590  // std::cout << "totECalc = " << totECalc << std::endl;
591 
592  // myHistos->fill("h20",Etotal);
593  // if(thePreshower)
594  // std::cout << " PS " << thePreshower->layer1Calibrated() << " " << thePreshower->layer2Calibrated() << " " << thePreshower->totalCalibrated() << " " << ps1tot << " " <<ps2tot << " " << pstot << std::endl;
595 }
596 
597 double EMShower::gam(double x, double a) const {
598  // A stupid gamma function
599  return std::pow(x, a - 1.) * std::exp(-x);
600 }
601 
602 //double
603 //EMShower::deposit(double t, double a, double b, double dt) {
604 //
605 // // The number of integration steps (about 1 / X0)
606 // int numberOfSteps = (int)dt+1;
607 //
608 // // The size if the integration step
609 // double integrationstep = dt/(double)numberOfSteps;
610 //
611 // // Half this size
612 // double halfis = 0.5*integrationstep;
613 //
614 // double dE = 0.;
615 //
616 // for(double tt=t-dt+halfis;tt<t;tt+=integrationstep) {
617 //
618 // // Simpson integration over each of these steps
619 // dE += gam(b*(tt-halfis),a)
620 // + 4.*gam(b* tt ,a)
621 // + gam(b*(tt+halfis),a);
622 //
623 // }
624 //
625 // // Normalization
626 // dE *= b*integrationstep/tgamma(a)/6.;
627 //
628 // // There we go.
629 // return dE;
630 //}
631 
632 double EMShower::deposit(double t, double a, double b, double dt) {
633  myIncompleteGamma.a().setValue(a);
634  double b1 = b * (t - dt);
635  double b2 = b * t;
636  double result = 0.;
637  double rb1 = (b1 != 0.) ? myIncompleteGamma(b1) : 0.;
638  double rb2 = (b2 != 0.) ? myIncompleteGamma(b2) : 0.;
639  result = (rb2 - rb1);
640  return result;
641 }
642 
643 void EMShower::setIntervals(unsigned icomp, RadialInterval& rad) {
644  // std::cout << " Got the pointer " << std::endl;
645  const std::vector<double>& myValues((icomp) ? theParam->getTailIntervals() : theParam->getCoreIntervals());
646  // std::cout << " Got the vector " << myValues.size () << std::endl;
647  unsigned nvals = myValues.size() / 2;
648  for (unsigned iv = 0; iv < nvals; ++iv) {
649  // std::cout << myValues[2*iv] << " " << myValues[2*iv+1] <<std::endl;
650  rad.addInterval(myValues[2 * iv], myValues[2 * iv + 1]);
651  }
652 }
653 
655  if (myPresh != nullptr) {
656  thePreshower = myPresh;
657  hasPreshower = true;
658  }
659 }
660 
661 void EMShower::setHcal(HcalHitMaker* const myHcal) { theHcalHitMaker = myHcal; }
662 
663 double EMShower::deposit(double a, double b, double t) {
664  // std::cout << " Deposit " << std::endl;
665  myIncompleteGamma.a().setValue(a);
666  double b2 = b * t;
667  double result = 0.;
668  if (fabs(b2) < 1.e-9)
669  b2 = 1.e-9;
671  // std::cout << " deposit t = " << t << " " << result <<std::endl;
672  return result;
673 }
EcalHitMaker::getX0back
double getX0back() const
Definition: EcalHitMaker.h:104
RadialInterval::addInterval
void addInterval(double, double)
Definition: RadialInterval.cc:15
ECALProperties::lightCollectionUniformity
virtual double lightCollectionUniformity() const =0
Light Collection uniformity.
EMECALShowerParametrization::rT
double rT(double tau, double E) const
Definition: EMECALShowerParametrization.h:141
PreshowerHitMaker.h
PreshowerHitMaker::setSpotEnergy
void setSpotEnergy(double e) override
Definition: PreshowerHitMaker.h:27
EMShower::nPart
unsigned int nPart
Definition: EMShower.h:90
mps_fire.i
i
Definition: mps_fire.py:428
SiStripPI::mean
Definition: SiStripPayloadInspectorHelper.h:169
EMShower::a
std::vector< double > a
Definition: EMShower.h:98
EMShower::Etot
std::vector< double > Etot
Definition: EMShower.h:94
groupFilesInBlocks.tt
int tt
Definition: groupFilesInBlocks.py:144
step
step
Definition: StallMonitor.cc:94
EMECALShowerParametrization::p
double p(double tau, double E) const
Definition: EMECALShowerParametrization.h:147
mps_update.status
status
Definition: mps_update.py:69
hcal
Definition: ConfigurationDatabase.cc:13
EMShower::theLayer1
const PreshowerLayer1Properties * theLayer1
Definition: EMShower.h:85
EMShower::globalMaximum
double globalMaximum
Definition: EMShower.h:113
gpuVertexFinder::iv
int32_t *__restrict__ iv
Definition: gpuClusterTracksDBSCAN.h:42
GammaFunctionGenerator::setParameters
void setParameters(double a, double b, double xm)
The parameters must be set before shooting.
Definition: GammaFunctionGenerator.cc:93
EMShower::theGrid
EcalHitMaker * theGrid
Definition: EMShower.h:123
EMECALShowerParametrization::sigmaLnAlpha
double sigmaLnAlpha(double lny) const
Definition: EMECALShowerParametrization.h:91
EMShower::photos
std::vector< double > photos
Definition: EMShower.h:96
EMShower::theECAL
const ECALProperties * theECAL
Definition: EMShower.h:83
CustomPhysics_cfi.gamma
gamma
Definition: CustomPhysics_cfi.py:17
EMECALShowerParametrization::nSpots
double nSpots(double E) const
Definition: EMECALShowerParametrization.h:183
EMShower::aSpot
std::vector< double > aSpot
Definition: EMShower.h:102
EMShower::E
std::vector< double > E
Definition: EMShower.h:95
EcalHitMaker::ps2TotalX0
double ps2TotalX0() const
total number of X0 in the PS (Layer2).
Definition: EcalHitMaker.h:61
DDAxes::x
EMShower::thePreshower
PreshowerHitMaker * thePreshower
Definition: EMShower.h:126
EMShower::thePart
std::vector< const RawParticle * > *const thePart
Definition: EMShower.h:89
ECALProperties::criticalEnergy
double criticalEnergy() const override
Critical energy in GeV (2.66E-3*(x0*Z/A)^1.1): 8.74E-3 for Standard ECAL.
Definition: ECALProperties.h:46
b2
static constexpr float b2
Definition: L1EGammaCrystalsEmulatorProducer.cc:83
EMShower::bFixedLength_
bool bFixedLength_
Definition: EMShower.h:144
GammaFunctionGenerator
Definition: GammaFunctionGenerator.h:21
EMECALShowerParametrization::layer1Properties
const PreshowerLayer1Properties * layer1Properties() const
Definition: EMECALShowerParametrization.h:223
ecal
Definition: ElectronicsMappingGPU.h:13
GammaFunctionGenerator::shoot
double shoot(RandomEngineAndDistribution const *) const
Definition: GammaFunctionGenerator.cc:30
EMShower::totalEnergy
double totalEnergy
Definition: EMShower.h:115
EMECALShowerParametrization::meanLnT
double meanLnT(double lny) const
Definition: EMECALShowerParametrization.h:73
EcalHitMaker::getPads
bool getPads(double depth, bool inCm=false)
Definition: EcalHitMaker.cc:820
EcalHitMaker.h
HcalHitMaker.h
RandomEngineAndDistribution.h
EMShower::setHcal
void setHcal(HcalHitMaker *const myHcal)
set the HCAL address
Definition: EMShower.cc:661
EMShower::nSteps
unsigned nSteps
Definition: EMShower.h:119
EMShower::setIntervals
void setIntervals(unsigned icomp, RadialInterval &rad)
Definition: EMShower.cc:643
GammaFunctionGenerator.h
hltEgammaHLTExtra_cfi.hcal
hcal
Definition: hltEgammaHLTExtra_cfi.py:42
EcalHitMaker::addHit
bool addHit(double r, double phi, unsigned layer=0) override
Definition: EcalHitMaker.cc:184
EMShower::prepareSteps
void prepareSteps()
Computes the steps before the real compute.
Definition: EMShower.cc:121
testProducerWithPsetDescEmpty_cfi.z2
z2
Definition: testProducerWithPsetDescEmpty_cfi.py:41
EMShower::theHCAL
const HCALProperties * theHCAL
Definition: EMShower.h:84
dt
float dt
Definition: AMPTWrapper.h:136
RadialInterval::getUmax
double getUmax(unsigned i) const
Upper limit of the argument in the radius generator.
Definition: RadialInterval.h:42
b1
static constexpr float b1
Definition: L1EGammaCrystalsEmulatorProducer.cc:83
ECALProperties::resE
double resE() const
a rough estimate of ECAL resolution sigma/E = resE/sqrt(E)
Definition: ECALProperties.h:58
RadialInterval::nIntervals
unsigned nIntervals() const
Number of intervals.
Definition: RadialInterval.h:25
ECALProperties::lightCollectionEfficiency
virtual double lightCollectionEfficiency() const =0
Light Collection efficiency.
EcalHitMaker::ps2eeTotalX0
double ps2eeTotalX0() const
Definition: EcalHitMaker.h:64
EcalHitMaker::x0DepthOffset
double x0DepthOffset() const
get the offset (e.g the number of X0 after which the shower starts)
Definition: EcalHitMaker.h:55
mathSSE::sqrt
T sqrt(T t)
Definition: SSEVec.h:19
EMShower::innerDepth
double innerDepth
Definition: EMShower.h:111
RadialInterval::getUmin
double getUmin(unsigned i) const
Lower limit of the argument in the radius generator.
Definition: RadialInterval.h:37
HLTMuonOfflineAnalyzer_cfi.z0
z0
Definition: HLTMuonOfflineAnalyzer_cfi.py:98
PreshowerHitMaker::addHit
bool addHit(double r, double phi, unsigned layer=0) override
Definition: PreshowerHitMaker.cc:88
EMECALShowerParametrization::meanT
double meanT(double lny) const
Definition: EMECALShowerParametrization.h:44
EMECALShowerParametrization::ecalProperties
const ECALProperties * ecalProperties() const
Definition: EMECALShowerParametrization.h:219
HLT_FULL_cff.gap
gap
Definition: HLT_FULL_cff.py:8498
b
double b
Definition: hdecay.h:118
EMECALShowerParametrization::rC
double rC(double tau, double E) const
Definition: EMECALShowerParametrization.h:135
RadialInterval::compute
void compute()
Definition: RadialInterval.cc:66
LEDCalibrationChannels.depth
depth
Definition: LEDCalibrationChannels.py:65
EMShower::compute
void compute()
Compute the shower longitudinal and lateral development.
Definition: EMShower.cc:263
EMShower::meanDepth
std::vector< double > meanDepth
Definition: EMShower.h:110
RadialInterval
Definition: RadialInterval.h:12
RadialInterval::getNumberOfSpots
unsigned getNumberOfSpots(unsigned i) const
Number of spots in a given interval.
Definition: RadialInterval.h:32
EMShower::theLayer2
const PreshowerLayer2Properties * theLayer2
Definition: EMShower.h:86
a
double a
Definition: hdecay.h:119
jetAnalyzer_cfi.rhom
rhom
Definition: jetAnalyzer_cfi.py:251
EMShower::gam
double gam(double x, double a) const
Definition: EMShower.cc:597
EMShower::depositedEnergy
std::vector< std::vector< double > > depositedEnergy
Definition: EMShower.h:109
EMShower::random
const RandomEngineAndDistribution * random
Definition: EMShower.h:139
PreshowerHitMaker
Definition: PreshowerHitMaker.h:11
EcalHitMaker::setSpotEnergy
void setSpotEnergy(double e) override
Definition: EcalHitMaker.h:112
EMECALShowerParametrization::meanAlpha
double meanAlpha(double lny) const
Definition: EMECALShowerParametrization.h:50
EMShower::myIncompleteGamma
Genfun::IncompleteGamma myIncompleteGamma
Definition: EMShower.h:136
EMShower::TSpot
std::vector< double > TSpot
Definition: EMShower.h:101
createfilelist.int
int
Definition: createfilelist.py:10
EcalHitMaker::totalX0
double totalX0() const
Definition: EcalHitMaker.h:49
EMECALShowerParametrization::meanAlphaSpot
double meanAlphaSpot(double alpha) const
Definition: EMECALShowerParametrization.h:195
HCALProperties::hOverPi
double hOverPi() const
Definition: HCALProperties.h:60
M_PI
#define M_PI
Definition: BXVectorInputProducer.cc:49
EMECALShowerParametrization::getTailIntervals
const std::vector< double > & getTailIntervals() const
Definition: EMECALShowerParametrization.h:229
trackerHitRTTI::vector
Definition: trackerHitRTTI.h:21
EMECALShowerParametrization::layer2Properties
const PreshowerLayer2Properties * layer2Properties() const
Definition: EMECALShowerParametrization.h:225
EMShower::setPreshower
void setPreshower(PreshowerHitMaker *const myPresh)
set the preshower address
Definition: EMShower.cc:654
HcalHitMaker
Definition: HcalHitMaker.h:16
EcalHitMaker::ecalTotalX0
double ecalTotalX0() const
in the ECAL
Definition: EcalHitMaker.h:67
EMShower::bSpot
std::vector< double > bSpot
Definition: EMShower.h:103
EMShower::outerDepth
double outerDepth
Definition: EMShower.h:111
ECALProperties::photoStatistics
virtual double photoStatistics() const =0
Photostatistics (photons/GeV) in the homegeneous material.
EMECALShowerParametrization::hcalProperties
const HCALProperties * hcalProperties() const
Definition: EMECALShowerParametrization.h:221
EMShower::theHcalHitMaker
HcalHitMaker * theHcalHitMaker
Definition: EMShower.h:129
EMShower::b
std::vector< double > b
Definition: EMShower.h:99
RadialInterval::getSpotEnergy
double getSpotEnergy(unsigned i) const
Spot energy in a given interval.
Definition: RadialInterval.h:27
RandomEngineAndDistribution::flatShoot
double flatShoot(double xmin=0.0, double xmax=1.0) const
Definition: RandomEngineAndDistribution.h:27
EMECALShowerParametrization::meanTSpot
double meanTSpot(double T) const
Definition: EMECALShowerParametrization.h:189
EMShower.h
EcalHitMaker::ps1TotalX0
double ps1TotalX0() const
Definition: EcalHitMaker.h:58
HcalHitMaker::setDepth
bool setDepth(double, bool inCm=false)
set the depth in X0 or Lambda0 units depending on showerType
Definition: HcalHitMaker.cc:114
EMECALShowerParametrization
Definition: EMECALShowerParametrization.h:19
EcalHitMaker::addHitDepth
bool addHitDepth(double r, double phi, double depth=-1)
Definition: EcalHitMaker.cc:133
Calorimetry_cff.bFixedLength
bFixedLength
Definition: Calorimetry_cff.py:24
ECALProperties::isHom
bool isHom() const
a rough estimate of ECAL resolution sigma/E = resE/sqrt(E)
Definition: ECALProperties.h:67
DDAxes::phi
EMShower::Ti
std::vector< double > Ti
Definition: EMShower.h:100
EMShower::EMShower
EMShower(RandomEngineAndDistribution const *engine, GammaFunctionGenerator *gamma, EMECALShowerParametrization *const myParam, std::vector< const RawParticle * > *const myPart, EcalHitMaker *const myGrid=nullptr, PreshowerHitMaker *const myPreshower=nullptr, bool bFixedLength=false)
Definition: EMShower.cc:16
HCALProperties::spotFraction
double spotFraction() const
Spot fraction wrt ECAL.
Definition: HCALProperties.h:63
EMShower::myGammaGenerator
GammaFunctionGenerator * myGammaGenerator
Definition: EMShower.h:142
EMShower::deposit
double deposit(double t, double a, double b, double dt)
Definition: EMShower.cc:632
T
long double T
Definition: Basic3DVectorLD.h:48
EMECALShowerParametrization::getCoreIntervals
const std::vector< double > & getCoreIntervals() const
Definition: EMECALShowerParametrization.h:227
genVertex_cff.x
x
Definition: genVertex_cff.py:12
EcalHitMaker
Definition: EcalHitMaker.h:24
EcalHitMaker::hcalTotalX0
double hcalTotalX0() const
in the HCAL
Definition: EcalHitMaker.h:73
EMShower::stepsCalculated
bool stepsCalculated
Definition: EMShower.h:120
EMECALShowerParametrization::sigmaLnT
double sigmaLnT(double lny) const
Definition: EMECALShowerParametrization.h:79
dqm-mbProfile.log
log
Definition: dqm-mbProfile.py:17
EMShower::hasPreshower
bool hasPreshower
Definition: EMShower.h:132
hgcalTestNeighbor_cfi.detector
detector
Definition: hgcalTestNeighbor_cfi.py:6
funct::pow
Power< A, B >::type pow(const A &a, const B &b)
Definition: Power.h:29
mps_fire.result
result
Definition: mps_fire.py:311
HcalHitMaker::addHit
bool addHit(double r, double phi, unsigned layer=0) override
add the hit in the HCAL in local coordinates
Definition: HcalHitMaker.cc:30
EMShower::theNumberOfSpots
std::vector< double > theNumberOfSpots
Definition: EMShower.h:93
EMShower::steps
Steps steps
Definition: EMShower.h:118
JetChargeProducer_cfi.exp
exp
Definition: JetChargeProducer_cfi.py:6
hltrates_dqm_sourceclient-live_cfg.offset
offset
Definition: hltrates_dqm_sourceclient-live_cfg.py:82
EMShower::maximumOfShower
std::vector< double > maximumOfShower
Definition: EMShower.h:108
PreshowerLayer1Properties::mipsPerGeV
double mipsPerGeV() const override
Number of Mips/GeV [Default : 41.7 Mips/GeV or 24 MeV/Mips].
Definition: PreshowerLayer1Properties.h:29
HcalHitMaker::setSpotEnergy
void setSpotEnergy(double e) override
Set the spot energy.
Definition: HcalHitMaker.h:26
submitPVValidationJobs.t
string t
Definition: submitPVValidationJobs.py:644
EMShower::theParam
EMECALShowerParametrization *const theParam
Definition: EMShower.h:80
PreshowerLayer2Properties::mipsPerGeV
double mipsPerGeV() const override
Number of Mips/GeV [Default : 59.5 Mips/GeV or 0.7*24 MeV/Mips].
Definition: PreshowerLayer2Properties.h:29
EMShower::Step
std::pair< unsigned int, double > Step
Definition: EMShower.h:30
RandomEngineAndDistribution::gaussShoot
double gaussShoot(double mean=0.0, double sigma=1.0) const
Definition: RandomEngineAndDistribution.h:29
RandomEngineAndDistribution::poissonShoot
unsigned int poissonShoot(double mean) const
Definition: RandomEngineAndDistribution.h:33
EMECALShowerParametrization::meanLnAlpha
double meanLnAlpha(double lny) const
Definition: EMECALShowerParametrization.h:85
EMECALShowerParametrization::correlationAlphaT
double correlationAlphaT(double lny) const
Definition: EMECALShowerParametrization.h:97
MillePedeFileConverter_cfg.e
e
Definition: MillePedeFileConverter_cfg.py:37
RandomEngineAndDistribution
Definition: RandomEngineAndDistribution.h:18