]> git.uio.no Git - u/mrichter/AliRoot.git/blame - MUON/AliMUONTrackParam.cxx
Connect only MUON branches when reading the event to speed up digitisation.
[u/mrichter/AliRoot.git] / MUON / AliMUONTrackParam.cxx
CommitLineData
a9e2aefa 1/**************************************************************************
2 * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
3 * *
4 * Author: The ALICE Off-line Project. *
5 * Contributors are mentioned in the code where appropriate. *
6 * *
7 * Permission to use, copy, modify and distribute this software and its *
8 * documentation strictly for non-commercial purposes is hereby granted *
9 * without fee, provided that the above copyright notice appears in all *
10 * copies and that both the copyright notice and this permission notice *
11 * appear in the supporting documentation. The authors make no claims *
12 * about the suitability of this software for any purpose. It is *
13 * provided "as is" without express or implied warranty. *
14 **************************************************************************/
15
16/*
17$Log$
43af2cb6 18Revision 1.10 2001/04/25 14:50:42 gosset
19Corrections to violations of coding conventions
20
3831f268 21Revision 1.9 2000/10/16 15:30:40 gosset
22TotalMomentumEnergyLoss:
23correction for change in the absorber composition (JP Cussonneau)
24
6372a28d 25Revision 1.8 2000/10/02 21:28:09 fca
26Removal of useless dependecies via forward declarations
27
94de3818 28Revision 1.7 2000/10/02 16:58:29 egangler
29Cleaning of the code :
30-> coding conventions
31-> void Streamers
32-> some useless includes removed or replaced by "class" statement
33
ecfa008b 34Revision 1.6 2000/09/19 09:49:50 gosset
35AliMUONEventReconstructor package
36* track extrapolation independent from reco_muon.F, use of AliMagF...
37* possibility to use new magnetic field (automatic from generated root file)
38
a6f03ddb 39Revision 1.5 2000/07/18 16:04:06 gosset
40AliMUONEventReconstructor package:
41* a few minor modifications and more comments
42* a few corrections
43 * right sign for Z of raw clusters
44 * right loop over chambers inside station
45 * symmetrized covariance matrix for measurements (TrackChi2MCS)
46 * right sign of charge in extrapolation (ExtrapToZ)
47 * right zEndAbsorber for Branson correction below 3 degrees
48* use of TVirtualFitter instead of TMinuit for AliMUONTrack::Fit
49* no parameter for AliMUONTrack::Fit() but more fit parameters in Track object
50
956019b6 51Revision 1.4 2000/07/03 07:53:31 morsch
52Double declaration problem on HP solved.
53
88962f0c 54Revision 1.3 2000/06/30 10:15:48 gosset
55Changes to EventReconstructor...:
56precision fit with multiple Coulomb scattering;
57extrapolation to vertex with Branson correction in absorber (JPC)
58
04b5ea16 59Revision 1.2 2000/06/15 07:58:49 morsch
60Code from MUON-dev joined
61
a9e2aefa 62Revision 1.1.2.3 2000/06/09 21:03:09 morsch
63Make includes consistent with new file structure.
64
65Revision 1.1.2.2 2000/06/09 12:58:05 gosset
66Removed comment beginnings in Log sections of .cxx files
67Suppressed most violations of coding rules
68
69Revision 1.1.2.1 2000/06/07 14:44:53 gosset
70Addition of files for track reconstruction in C++
71*/
72
3831f268 73///////////////////////////////////////////////////
74//
75// Track parameters
76// in
77// ALICE
78// dimuon
79// spectrometer
a9e2aefa 80//
3831f268 81///////////////////////////////////////////////////
a9e2aefa 82
83#include <iostream.h>
84
85#include "AliCallf77.h"
3831f268 86#include "AliMUON.h"
a9e2aefa 87#include "AliMUONTrackParam.h"
3831f268 88#include "AliMUONChamber.h"
a9e2aefa 89#include "AliRun.h"
94de3818 90#include "AliMagF.h"
a9e2aefa 91
92ClassImp(AliMUONTrackParam) // Class implementation in ROOT context
93
a6f03ddb 94 // A few calls in Fortran or from Fortran (extrap.F).
95 // Needed, instead of calls to Geant subroutines,
96 // because double precision is necessary for track fit converging with Minuit.
97 // The "extrap" functions should be translated into C++ ????
a9e2aefa 98#ifndef WIN32
a6f03ddb 99# define extrap_onestep_helix extrap_onestep_helix_
100# define extrap_onestep_helix3 extrap_onestep_helix3_
101# define extrap_onestep_rungekutta extrap_onestep_rungekutta_
102# define gufld_double gufld_double_
a9e2aefa 103#else
a6f03ddb 104# define extrap_onestep_helix EXTRAP_ONESTEP_HELIX
105# define extrap_onestep_helix3 EXTRAP_ONESTEP_HELIX3
106# define extrap_onestep_rungekutta EXTRAP_ONESTEP_RUNGEKUTTA
107# define gufld_double GUFLD_DOUBLE
a9e2aefa 108#endif
109
a6f03ddb 110extern "C" {
111 void type_of_call extrap_onestep_helix
112 (Double_t &Charge, Double_t &StepLength, Double_t *VGeant3, Double_t *VGeant3New);
113
114 void type_of_call extrap_onestep_helix3
115 (Double_t &Field, Double_t &StepLength, Double_t *VGeant3, Double_t *VGeant3New);
116
117 void type_of_call extrap_onestep_rungekutta
118 (Double_t &Charge, Double_t &StepLength, Double_t *VGeant3, Double_t *VGeant3New);
119
120 void type_of_call gufld_double(Double_t *Position, Double_t *Field) {
121 // interface to "gAlice->Field()->Field" for arguments in double precision
122 Float_t x[3], b[3];
123 x[0] = Position[0]; x[1] = Position[1]; x[2] = Position[2];
124 gAlice->Field()->Field(x, b);
125 Field[0] = b[0]; Field[1] = b[1]; Field[2] = b[2];
126 }
a9e2aefa 127}
128
a9e2aefa 129 //__________________________________________________________________________
130void AliMUONTrackParam::ExtrapToZ(Double_t Z)
131{
132 // Track parameter extrapolation to the plane at "Z".
133 // On return, the track parameters resulting from the extrapolation
134 // replace the current track parameters.
a9e2aefa 135 if (this->fZ == Z) return; // nothing to be done if same Z
136 Double_t forwardBackward; // +1 if forward, -1 if backward
137 if (Z > this->fZ) forwardBackward = 1.0;
138 else forwardBackward = -1.0;
a6f03ddb 139 Double_t vGeant3[7], vGeant3New[7]; // 7 in parameter ????
a9e2aefa 140 Int_t iGeant3, stepNumber;
141 Int_t maxStepNumber = 5000; // in parameter ????
142 // For safety: return kTRUE or kFALSE ????
a6f03ddb 143 // Parameter vector for calling EXTRAP_ONESTEP
a9e2aefa 144 SetGeant3Parameters(vGeant3, forwardBackward);
956019b6 145 // sign of charge (sign of fInverseBendingMomentum if forward motion)
a6f03ddb 146 // must be changed if backward extrapolation
956019b6 147 Double_t chargeExtrap = forwardBackward *
148 TMath::Sign(Double_t(1.0), this->fInverseBendingMomentum);
a9e2aefa 149 Double_t stepLength = 6.0; // in parameter ????
150 // Extrapolation loop
151 stepNumber = 0;
152 while (((forwardBackward * (vGeant3[2] - Z)) <= 0.0) &&
153 (stepNumber < maxStepNumber)) {
154 stepNumber++;
a6f03ddb 155 // Option for switching between helix and Runge-Kutta ????
156 // extrap_onestep_rungekutta(chargeExtrap, stepLength, vGeant3, vGeant3New);
157 extrap_onestep_helix(chargeExtrap, stepLength, vGeant3, vGeant3New);
a9e2aefa 158 if ((forwardBackward * (vGeant3New[2] - Z)) > 0.0) break; // one is beyond Z
159 // better use TArray ????
160 for (iGeant3 = 0; iGeant3 < 7; iGeant3++)
161 {vGeant3[iGeant3] = vGeant3New[iGeant3];}
162 }
163 // check maxStepNumber ????
a9e2aefa 164 // Interpolation back to exact Z (2nd order)
165 // should be in function ???? using TArray ????
166 Double_t dZ12 = vGeant3New[2] - vGeant3[2]; // 1->2
167 Double_t dZ1i = Z - vGeant3[2]; // 1-i
168 Double_t dZi2 = vGeant3New[2] - Z; // i->2
169 Double_t xPrime = (vGeant3New[0] - vGeant3[0]) / dZ12;
170 Double_t xSecond =
171 ((vGeant3New[3] / vGeant3New[5]) - (vGeant3[3] / vGeant3[5])) / dZ12;
172 Double_t yPrime = (vGeant3New[1] - vGeant3[1]) / dZ12;
173 Double_t ySecond =
174 ((vGeant3New[4] / vGeant3New[5]) - (vGeant3[4] / vGeant3[5])) / dZ12;
175 vGeant3[0] = vGeant3[0] + xPrime * dZ1i - 0.5 * xSecond * dZ1i * dZi2; // X
176 vGeant3[1] = vGeant3[1] + yPrime * dZ1i - 0.5 * ySecond * dZ1i * dZi2; // Y
177 vGeant3[2] = Z; // Z
178 Double_t xPrimeI = xPrime - 0.5 * xSecond * (dZi2 - dZ1i);
179 Double_t yPrimeI = yPrime - 0.5 * ySecond * (dZi2 - dZ1i);
956019b6 180 // (PX, PY, PZ)/PTOT assuming forward motion
a9e2aefa 181 vGeant3[5] =
182 1.0 / TMath::Sqrt(1.0 + xPrimeI * xPrimeI + yPrimeI * yPrimeI); // PZ/PTOT
183 vGeant3[3] = xPrimeI * vGeant3[5]; // PX/PTOT
184 vGeant3[4] = yPrimeI * vGeant3[5]; // PY/PTOT
956019b6 185 // Track parameters from Geant3 parameters,
186 // with charge back for forward motion
187 GetFromGeant3Parameters(vGeant3, chargeExtrap * forwardBackward);
a9e2aefa 188}
189
190 //__________________________________________________________________________
191void AliMUONTrackParam::SetGeant3Parameters(Double_t *VGeant3, Double_t ForwardBackward)
192{
193 // Set vector of Geant3 parameters pointed to by "VGeant3"
194 // from track parameters in current AliMUONTrackParam.
195 // Since AliMUONTrackParam is only geometry, one uses "ForwardBackward"
196 // to know whether the particle is going forward (+1) or backward (-1).
197 VGeant3[0] = this->fNonBendingCoor; // X
198 VGeant3[1] = this->fBendingCoor; // Y
199 VGeant3[2] = this->fZ; // Z
200 Double_t pYZ = TMath::Abs(1.0 / this->fInverseBendingMomentum);
201 Double_t pZ =
202 pYZ / TMath::Sqrt(1.0 + this->fBendingSlope * this->fBendingSlope);
203 VGeant3[6] =
204 TMath::Sqrt(pYZ * pYZ +
205 pZ * pZ * this->fNonBendingSlope * this->fNonBendingSlope); // PTOT
206 VGeant3[5] = ForwardBackward * pZ / VGeant3[6]; // PZ/PTOT
207 VGeant3[3] = this->fNonBendingSlope * VGeant3[5]; // PX/PTOT
208 VGeant3[4] = this->fBendingSlope * VGeant3[5]; // PY/PTOT
209}
210
211 //__________________________________________________________________________
212void AliMUONTrackParam::GetFromGeant3Parameters(Double_t *VGeant3, Double_t Charge)
213{
214 // Get track parameters in current AliMUONTrackParam
956019b6 215 // from Geant3 parameters pointed to by "VGeant3",
216 // assumed to be calculated for forward motion in Z.
a9e2aefa 217 // "InverseBendingMomentum" is signed with "Charge".
218 this->fNonBendingCoor = VGeant3[0]; // X
219 this->fBendingCoor = VGeant3[1]; // Y
220 this->fZ = VGeant3[2]; // Z
221 Double_t pYZ = VGeant3[6] * TMath::Sqrt(1.0 - VGeant3[3] * VGeant3[3]);
222 this->fInverseBendingMomentum = Charge / pYZ;
223 this->fBendingSlope = VGeant3[4] / VGeant3[5];
224 this->fNonBendingSlope = VGeant3[3] / VGeant3[5];
225}
226
227 //__________________________________________________________________________
228void AliMUONTrackParam::ExtrapToStation(Int_t Station, AliMUONTrackParam *TrackParam)
229{
230 // Track parameters extrapolated from current track parameters ("this")
231 // to both chambers of the station(0..) "Station"
232 // are returned in the array (dimension 2) of track parameters
233 // pointed to by "TrackParam" (index 0 and 1 for first and second chambers).
234 Double_t extZ[2], z1, z2;
ecfa008b 235 Int_t i1 = -1, i2 = -1; // = -1 to avoid compilation warnings
a9e2aefa 236 AliMUON *pMUON = (AliMUON*) gAlice->GetModule("MUON"); // necessary ????
237 // range of Station to be checked ????
238 z1 = (&(pMUON->Chamber(2 * Station)))->Z(); // Z of first chamber
239 z2 = (&(pMUON->Chamber(2 * Station + 1)))->Z(); // Z of second chamber
240 // First and second Z to extrapolate at
241 if ((z1 > this->fZ) && (z2 > this->fZ)) {i1 = 0; i2 = 1;}
242 else if ((z1 < this->fZ) && (z2 < this->fZ)) {i1 = 1; i2 = 0;}
243 else {
244 cout << "ERROR in AliMUONTrackParam::CreateExtrapSegmentInStation" << endl;
245 cout << "Starting Z (" << this->fZ << ") in between z1 (" << z1 <<
246 ") and z2 (" << z2 << ") of station(0..) " << Station << endl;
247 }
248 extZ[i1] = z1;
249 extZ[i2] = z2;
250 // copy of track parameters
251 TrackParam[i1] = *this;
252 // first extrapolation
253 (&(TrackParam[i1]))->ExtrapToZ(extZ[0]);
254 TrackParam[i2] = TrackParam[i1];
255 // second extrapolation
256 (&(TrackParam[i2]))->ExtrapToZ(extZ[1]);
257 return;
258}
259
04b5ea16 260 //__________________________________________________________________________
261void AliMUONTrackParam::ExtrapToVertex()
262{
263 // Extrapolation to the vertex.
264 // Returns the track parameters resulting from the extrapolation,
265 // in the current TrackParam.
956019b6 266 // Changes parameters according to Branson correction through the absorber
04b5ea16 267
268 Double_t zAbsorber = 503.0; // to be coherent with the Geant absorber geometry !!!!
269 // Extrapolates track parameters upstream to the "Z" end of the front absorber
270 ExtrapToZ(zAbsorber);
271 // Makes Branson correction (multiple scattering + energy loss)
272 BransonCorrection();
273}
274
43af2cb6 275
276// Keep this version for future developments
04b5ea16 277 //__________________________________________________________________________
43af2cb6 278// void AliMUONTrackParam::BransonCorrection()
279// {
280// // Branson correction of track parameters
281// // the entry parameters have to be calculated at the end of the absorber
282// Double_t zEndAbsorber, zBP, xBP, yBP;
283// Double_t pYZ, pX, pY, pZ, pTotal, xEndAbsorber, yEndAbsorber, radiusEndAbsorber2, pT, theta;
284// Int_t sign;
285// // Would it be possible to calculate all that from Geant configuration ????
286// // and to get the Branson parameters from a function in ABSO module ????
287// // with an eventual contribution from other detectors like START ????
288// // Radiation lengths outer part theta > 3 degres
289// static Double_t x01[9] = { 18.8, // C (cm)
290// 10.397, // Concrete (cm)
291// 0.56, // Plomb (cm)
292// 47.26, // Polyethylene (cm)
293// 0.56, // Plomb (cm)
294// 47.26, // Polyethylene (cm)
295// 0.56, // Plomb (cm)
296// 47.26, // Polyethylene (cm)
297// 0.56 }; // Plomb (cm)
298// // inner part theta < 3 degres
299// static Double_t x02[3] = { 18.8, // C (cm)
300// 10.397, // Concrete (cm)
301// 0.35 }; // W (cm)
302// // z positions of the materials inside the absober outer part theta > 3 degres
303// static Double_t z1[10] = { 90, 315, 467, 472, 477, 482, 487, 492, 497, 502 };
304// // inner part theta < 3 degres
305// static Double_t z2[4] = { 90, 315, 467, 503 };
306// static Bool_t first = kTRUE;
307// static Double_t zBP1, zBP2, rLimit;
308// // Calculates z positions of the Branson's planes: zBP1 for outer part and zBP2 for inner part (only at the first call)
309// if (first) {
310// first = kFALSE;
311// Double_t aNBP = 0.0;
312// Double_t aDBP = 0.0;
313// Int_t iBound;
314
315// for (iBound = 0; iBound < 9; iBound++) {
316// aNBP = aNBP +
317// (z1[iBound+1] * z1[iBound+1] * z1[iBound+1] -
318// z1[iBound] * z1[iBound] * z1[iBound] ) / x01[iBound];
319// aDBP = aDBP +
320// (z1[iBound+1] * z1[iBound+1] - z1[iBound] * z1[iBound] ) / x01[iBound];
321// }
322// zBP1 = (2.0 * aNBP) / (3.0 * aDBP);
323// aNBP = 0.0;
324// aDBP = 0.0;
325// for (iBound = 0; iBound < 3; iBound++) {
326// aNBP = aNBP +
327// (z2[iBound+1] * z2[iBound+1] * z2[iBound+1] -
328// z2[iBound] * z2[iBound ] * z2[iBound] ) / x02[iBound];
329// aDBP = aDBP +
330// (z2[iBound+1] * z2[iBound+1] - z2[iBound] * z2[iBound]) / x02[iBound];
331// }
332// zBP2 = (2.0 * aNBP) / (3.0 * aDBP);
333// rLimit = z2[3] * TMath::Tan(3.0 * (TMath::Pi()) / 180.);
334// }
335
336// pYZ = TMath::Abs(1.0 / fInverseBendingMomentum);
337// sign = 1;
338// if (fInverseBendingMomentum < 0) sign = -1;
339// pZ = pYZ / (TMath::Sqrt(1.0 + fBendingSlope * fBendingSlope));
340// pX = pZ * fNonBendingSlope;
341// pY = pZ * fBendingSlope;
342// pTotal = TMath::Sqrt(pYZ *pYZ + pX * pX);
343// xEndAbsorber = fNonBendingCoor;
344// yEndAbsorber = fBendingCoor;
345// radiusEndAbsorber2 = xEndAbsorber * xEndAbsorber + yEndAbsorber * yEndAbsorber;
346
347// if (radiusEndAbsorber2 > rLimit*rLimit) {
348// zEndAbsorber = z1[9];
349// zBP = zBP1;
350// } else {
351// zEndAbsorber = z2[3];
352// zBP = zBP2;
353// }
354
355// xBP = xEndAbsorber - (pX / pZ) * (zEndAbsorber - zBP);
356// yBP = yEndAbsorber - (pY / pZ) * (zEndAbsorber - zBP);
357
358// // new parameters after Branson and energy loss corrections
359// pZ = pTotal * zBP / TMath::Sqrt(xBP * xBP + yBP * yBP + zBP * zBP);
360// pX = pZ * xBP / zBP;
361// pY = pZ * yBP / zBP;
362// fBendingSlope = pY / pZ;
363// fNonBendingSlope = pX / pZ;
364
365// pT = TMath::Sqrt(pX * pX + pY * pY);
366// theta = TMath::ATan2(pT, pZ);
367// pTotal =
368// TotalMomentumEnergyLoss(rLimit, pTotal, theta, xEndAbsorber, yEndAbsorber);
369
370// fInverseBendingMomentum = (sign / pTotal) *
371// TMath::Sqrt(1.0 +
372// fBendingSlope * fBendingSlope +
373// fNonBendingSlope * fNonBendingSlope) /
374// TMath::Sqrt(1.0 + fBendingSlope * fBendingSlope);
375
376// // vertex position at (0,0,0)
377// // should be taken from vertex measurement ???
378// fBendingCoor = 0.0;
379// fNonBendingCoor = 0;
380// fZ= 0;
381// }
382
04b5ea16 383void AliMUONTrackParam::BransonCorrection()
384{
385 // Branson correction of track parameters
386 // the entry parameters have to be calculated at the end of the absorber
43af2cb6 387 // simplified version: the z positions of Branson's planes are no longer calculated
388 // but are given as inputs. One can use the macros MUONTestAbso.C and DrawTestAbso.C
389 // to test this correction.
04b5ea16 390 // Would it be possible to calculate all that from Geant configuration ????
956019b6 391 // and to get the Branson parameters from a function in ABSO module ????
392 // with an eventual contribution from other detectors like START ????
43af2cb6 393 Double_t zBP, xBP, yBP;
394 Double_t pYZ, pX, pY, pZ, pTotal, xEndAbsorber, yEndAbsorber, radiusEndAbsorber2, pT, theta;
395 Int_t sign;
04b5ea16 396 static Bool_t first = kTRUE;
43af2cb6 397 static Double_t zBP1, zBP2, rLimit, zEndAbsorber;
398 // zBP1 for outer part and zBP2 for inner part (only at the first call)
04b5ea16 399 if (first) {
400 first = kFALSE;
43af2cb6 401
402 zEndAbsorber = 503;
403 rLimit = zEndAbsorber * TMath::Tan(3.0 * (TMath::Pi()) / 180.);
404 zBP1 = 450;
405 zBP2 = 480;
04b5ea16 406 }
407
408 pYZ = TMath::Abs(1.0 / fInverseBendingMomentum);
409 sign = 1;
410 if (fInverseBendingMomentum < 0) sign = -1;
411 pZ = pYZ / (TMath::Sqrt(1.0 + fBendingSlope * fBendingSlope));
412 pX = pZ * fNonBendingSlope;
413 pY = pZ * fBendingSlope;
414 pTotal = TMath::Sqrt(pYZ *pYZ + pX * pX);
415 xEndAbsorber = fNonBendingCoor;
416 yEndAbsorber = fBendingCoor;
417 radiusEndAbsorber2 = xEndAbsorber * xEndAbsorber + yEndAbsorber * yEndAbsorber;
418
419 if (radiusEndAbsorber2 > rLimit*rLimit) {
04b5ea16 420 zBP = zBP1;
421 } else {
04b5ea16 422 zBP = zBP2;
423 }
424
425 xBP = xEndAbsorber - (pX / pZ) * (zEndAbsorber - zBP);
426 yBP = yEndAbsorber - (pY / pZ) * (zEndAbsorber - zBP);
427
428 // new parameters after Branson and energy loss corrections
429 pZ = pTotal * zBP / TMath::Sqrt(xBP * xBP + yBP * yBP + zBP * zBP);
430 pX = pZ * xBP / zBP;
431 pY = pZ * yBP / zBP;
432 fBendingSlope = pY / pZ;
433 fNonBendingSlope = pX / pZ;
434
435 pT = TMath::Sqrt(pX * pX + pY * pY);
436 theta = TMath::ATan2(pT, pZ);
437 pTotal =
438 TotalMomentumEnergyLoss(rLimit, pTotal, theta, xEndAbsorber, yEndAbsorber);
439
440 fInverseBendingMomentum = (sign / pTotal) *
441 TMath::Sqrt(1.0 +
442 fBendingSlope * fBendingSlope +
443 fNonBendingSlope * fNonBendingSlope) /
444 TMath::Sqrt(1.0 + fBendingSlope * fBendingSlope);
445
446 // vertex position at (0,0,0)
447 // should be taken from vertex measurement ???
448 fBendingCoor = 0.0;
449 fNonBendingCoor = 0;
450 fZ= 0;
451}
04b5ea16 452 //__________________________________________________________________________
453Double_t AliMUONTrackParam::TotalMomentumEnergyLoss(Double_t rLimit, Double_t pTotal, Double_t theta, Double_t xEndAbsorber, Double_t yEndAbsorber)
454{
455 // Returns the total momentum corrected from energy loss in the front absorber
43af2cb6 456 // One can use the macros MUONTestAbso.C and DrawTestAbso.C
457 // to test this correction.
04b5ea16 458 Double_t deltaP, pTotalCorrected;
459
460 Double_t radiusEndAbsorber2 =
461 xEndAbsorber *xEndAbsorber + yEndAbsorber * yEndAbsorber;
462 // Parametrization to be redone according to change of absorber material ????
956019b6 463 // See remark in function BransonCorrection !!!!
04b5ea16 464 // The name is not so good, and there are many arguments !!!!
465 if (radiusEndAbsorber2 < rLimit * rLimit) {
466 if (pTotal < 15) {
467 deltaP = 2.737 + 0.0494 * pTotal - 0.001123 * pTotal * pTotal;
468 } else {
469 deltaP = 3.0643 + 0.01346 *pTotal;
470 }
43af2cb6 471 deltaP = 0.63 * deltaP; // !!!! changes in the absorber composition ????
04b5ea16 472 } else {
473 if (pTotal < 15) {
474 deltaP = 2.1380 + 0.0351 * pTotal - 0.000853 * pTotal * pTotal;
475 } else {
476 deltaP = 2.407 + 0.00702 * pTotal;
477 }
43af2cb6 478 deltaP = 0.67 * deltaP; // !!!! changes in the absorber composition ????
04b5ea16 479 }
480 pTotalCorrected = pTotal + deltaP / TMath::Cos(theta);
481 return pTotalCorrected;
482}
483