StRoot  1
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Groups Pages
BoseEinstein.cc
1 // BoseEinstein.cc is a part of the PYTHIA event generator.
2 // Copyright (C) 2018 Torbjorn Sjostrand.
3 // PYTHIA is licenced under the GNU GPL v2 or later, see COPYING for details.
4 // Please respect the MCnet Guidelines, see GUIDELINES for details.
5 
6 // Function definitions (not found in the header) for the BoseEinsten class.
7 
8 #include "Pythia8/BoseEinstein.h"
9 
10 namespace Pythia8 {
11 
12 //==========================================================================
13 
14 // The BoseEinstein class.
15 
16 //--------------------------------------------------------------------------
17 
18 // Constants: could be changed here if desired, but normally should not.
19 // These are of technical nature, as described for each.
20 
21 // Enumeration of id codes and table for particle species considered.
22 const int BoseEinstein::IDHADRON[9] = { 211, -211, 111, 321, -321,
23  130, 310, 221, 331 };
24 const int BoseEinstein::ITABLE[9] = { 0, 0, 0, 1, 1, 1, 1, 2, 3 };
25 
26 // Distance between table entries, normalized to min( 2*mass, QRef).
27 const double BoseEinstein::STEPSIZE = 0.05;
28 
29 // Skip shift for two extremely close particles, to avoid instabilities.
30 const double BoseEinstein::Q2MIN = 1e-8;
31 
32 // Parameters of energy compensation procedure: maximally allowed
33 // relative energy error, iterative stepsize, and number of iterations.
34 const double BoseEinstein::COMPRELERR = 1e-10;
35 const double BoseEinstein::COMPFACMAX = 1000.;
36 const int BoseEinstein::NCOMPSTEP = 10;
37 
38 //--------------------------------------------------------------------------
39 
40 // Find settings. Precalculate table used to find momentum shifts.
41 
42 bool BoseEinstein::init(Info* infoPtrIn, Settings& settings,
43  ParticleData& particleData) {
44 
45  // Save pointer.
46  infoPtr = infoPtrIn;
47 
48  // Main flags.
49  doPion = settings.flag("BoseEinstein:Pion");
50  doKaon = settings.flag("BoseEinstein:Kaon");
51  doEta = settings.flag("BoseEinstein:Eta");
52 
53  // Shape of Bose-Einstein enhancement/suppression.
54  lambda = settings.parm("BoseEinstein:lambda");
55  QRef = settings.parm("BoseEinstein:QRef");
56 
57  // Multiples and inverses (= "radii") of distance parameters in Q-space.
58  QRef2 = 2. * QRef;
59  QRef3 = 3. * QRef;
60  R2Ref = 1. / (QRef * QRef);
61  R2Ref2 = 1. / (QRef2 * QRef2);
62  R2Ref3 = 1. / (QRef3 * QRef3);
63 
64  // Masses of particles with Bose-Einstein implemented.
65  for (int iSpecies = 0; iSpecies < 9; ++iSpecies)
66  mHadron[iSpecies] = particleData.m0( IDHADRON[iSpecies] );
67 
68  // Pair pi, K, eta and eta' masses for use in tables.
69  mPair[0] = 2. * mHadron[0];
70  mPair[1] = 2. * mHadron[3];
71  mPair[2] = 2. * mHadron[7];
72  mPair[3] = 2. * mHadron[8];
73 
74  // Loop over the four required tables. Local variables.
75  double Qnow, Q2now, centerCorr;
76  for (int iTab = 0; iTab < 4; ++iTab) {
77  m2Pair[iTab] = mPair[iTab] * mPair[iTab];
78 
79  // Step size and number of steps in normal table.
80  deltaQ[iTab] = STEPSIZE * min(mPair[iTab], QRef);
81  nStep[iTab] = min( 199, 1 + int(3. * QRef / deltaQ[iTab]) );
82  maxQ[iTab] = (nStep[iTab] - 0.1) * deltaQ[iTab];
83  centerCorr = deltaQ[iTab] * deltaQ[iTab] / 12.;
84 
85  // Construct normal table recursively in Q space.
86  shift[iTab][0] = 0.;
87  for (int i = 1; i <= nStep[iTab]; ++i) {
88  Qnow = deltaQ[iTab] * (i - 0.5);
89  Q2now = Qnow * Qnow;
90  shift[iTab][i] = shift[iTab][i - 1] + exp(-Q2now * R2Ref)
91  * deltaQ[iTab] * (Q2now + centerCorr) / sqrt(Q2now + m2Pair[iTab]);
92  }
93 
94  // Step size and number of steps in compensation table.
95  deltaQ3[iTab] = STEPSIZE * min(mPair[iTab], QRef3);
96  nStep3[iTab] = min( 199, 1 + int(9. * QRef / deltaQ3[iTab]) );
97  maxQ3[iTab] = (nStep3[iTab] - 0.1) * deltaQ3[iTab];
98  centerCorr = deltaQ3[iTab] * deltaQ3[iTab] / 12.;
99 
100  // Construct compensation table recursively in Q space.
101  shift3[iTab][0] = 0.;
102  for (int i = 1; i <= nStep3[iTab]; ++i) {
103  Qnow = deltaQ3[iTab] * (i - 0.5);
104  Q2now = Qnow * Qnow;
105  shift3[iTab][i] = shift3[iTab][i - 1] + exp(-Q2now * R2Ref3)
106  * deltaQ3[iTab] * (Q2now + centerCorr) / sqrt(Q2now + m2Pair[iTab]);
107  }
108 
109  }
110 
111  // Done.
112  return true;
113 
114 }
115 
116 //--------------------------------------------------------------------------
117 
118 // Perform Bose-Einstein corrections on an event.
119 
120 bool BoseEinstein::shiftEvent( Event& event) {
121 
122  // Reset list of identical particles.
123  hadronBE.resize(0);
124 
125  // Loop over all hadron species with BE effects.
126  nStored[0] = 0;
127  for (int iSpecies = 0; iSpecies < 9; ++iSpecies) {
128  nStored[iSpecies + 1] = nStored[iSpecies];
129  if (!doPion && iSpecies <= 2) continue;
130  if (!doKaon && iSpecies >= 3 && iSpecies <= 6) continue;
131  if (!doEta && iSpecies >= 7) continue;
132 
133  // Properties of current hadron species.
134  int idNow = IDHADRON[ iSpecies ];
135  int iTab = ITABLE[ iSpecies ];
136 
137  // Loop through event record to store copies of current species.
138  for (int i = 0; i < event.size(); ++i)
139  if ( event[i].id() == idNow && event[i].isFinal() )
140  hadronBE.push_back(
141  BoseEinsteinHadron( idNow, i, event[i].p(), event[i].m() ) );
142  nStored[iSpecies + 1] = hadronBE.size();
143 
144  // Loop through pairs of identical particles and find shifts.
145  for (int i1 = nStored[iSpecies]; i1 < nStored[iSpecies+1] - 1; ++i1)
146  for (int i2 = i1 + 1; i2 < nStored[iSpecies+1]; ++i2)
147  shiftPair( i1, i2, iTab);
148  }
149 
150  // Must have at least two pairs to carry out compensation.
151  if (nStored[9] < 2) return true;
152 
153  // Shift momenta and recalculate energies.
154  double eSumOriginal = 0.;
155  double eSumShifted = 0.;
156  double eDiffByComp = 0.;
157  for (int i = 0; i < nStored[9]; ++i) {
158  eSumOriginal += hadronBE[i].p.e();
159  hadronBE[i].p += hadronBE[i].pShift;
160  hadronBE[i].p.e( sqrt( hadronBE[i].p.pAbs2() + hadronBE[i].m2 ) );
161  eSumShifted += hadronBE[i].p.e();
162  eDiffByComp += dot3( hadronBE[i].pComp, hadronBE[i].p)
163  / hadronBE[i].p.e();
164  }
165 
166  // Iterate compensation shift until convergence.
167  int iStep = 0;
168  while ( abs(eSumShifted - eSumOriginal) > COMPRELERR * eSumOriginal
169  && abs(eSumShifted - eSumOriginal) < COMPFACMAX * abs(eDiffByComp)
170  && iStep < NCOMPSTEP ) {
171  ++iStep;
172  double compFac = (eSumOriginal - eSumShifted) / eDiffByComp;
173  eSumShifted = 0.;
174  eDiffByComp = 0.;
175  for (int i = 0; i < nStored[9]; ++i) {
176  hadronBE[i].p += compFac * hadronBE[i].pComp;
177  hadronBE[i].p.e( sqrt( hadronBE[i].p.pAbs2() + hadronBE[i].m2 ) );
178  eSumShifted += hadronBE[i].p.e();
179  eDiffByComp += dot3( hadronBE[i].pComp, hadronBE[i].p)
180  / hadronBE[i].p.e();
181  }
182  }
183 
184  // Error if no convergence, and then return without doing BE shift.
185  // However, not grave enough to kill event, so return true.
186  if ( abs(eSumShifted - eSumOriginal) > COMPRELERR * eSumOriginal ) {
187  infoPtr->errorMsg("Warning in BoseEinstein::shiftEvent: "
188  "no consistent BE shift topology found, so skip BE");
189  return true;
190  }
191 
192  // Store new particle copies with shifted momenta.
193  for (int i = 0; i < nStored[9]; ++i) {
194  int iNew = event.copy( hadronBE[i].iPos, 99);
195  event[ iNew ].p( hadronBE[i].p );
196  }
197 
198  // Done.
199  return true;
200 
201 }
202 
203 //--------------------------------------------------------------------------
204 
205 // Calculate shift and (unnormalized) compensation for pair.
206 
207 void BoseEinstein::shiftPair( int i1, int i2, int iTab) {
208 
209  // Calculate old relative momentum.
210  double Q2old = m2(hadronBE[i1].p, hadronBE[i2].p) - m2Pair[iTab];
211  if (Q2old < Q2MIN) return;
212  double Qold = sqrt(Q2old);
213  double psFac = sqrt(Q2old + m2Pair[iTab]) / Q2old;
214 
215  // Calculate new relative momentum for normal shift.
216  double Qmove = 0.;
217  if (Qold < deltaQ[iTab]) Qmove = Qold / 3.;
218  else if (Qold < maxQ[iTab]) {
219  double realQbin = Qold / deltaQ[iTab];
220  int intQbin = int( realQbin );
221  double inter = (pow3(realQbin) - pow3(intQbin))
222  / (3 * intQbin * (intQbin + 1) + 1);
223  Qmove = ( shift[iTab][intQbin] + inter * (shift[iTab][intQbin + 1]
224  - shift[iTab][intQbin]) ) * psFac;
225  }
226  else Qmove = shift[iTab][nStep[iTab]] * psFac;
227  double Q2new = Q2old * pow( Qold / (Qold + 3. * lambda * Qmove), 2. / 3.);
228 
229  // Calculate corresponding three-momentum shift.
230  double Q2Diff = Q2new - Q2old;
231  double p2DiffAbs = (hadronBE[i1].p - hadronBE[i2].p).pAbs2();
232  double p2AbsDiff = hadronBE[i1].p.pAbs2() - hadronBE[i2].p.pAbs2();
233  double eSum = hadronBE[i1].p.e() + hadronBE[i2].p.e();
234  double eDiff = hadronBE[i1].p.e() - hadronBE[i2].p.e();
235  double sumQ2E = Q2Diff + eSum * eSum;
236  double rootA = eSum * eDiff * p2AbsDiff - p2DiffAbs * sumQ2E;
237  double rootB = p2DiffAbs * sumQ2E - p2AbsDiff * p2AbsDiff;
238  double factor = 0.5 * ( rootA + sqrtpos(rootA * rootA
239  + Q2Diff * (sumQ2E - eDiff * eDiff) * rootB) ) / rootB;
240 
241  // Add shifts to sum. (Energy component dummy.)
242  Vec4 pDiff = factor * (hadronBE[i1].p - hadronBE[i2].p);
243  hadronBE[i1].pShift += pDiff;
244  hadronBE[i2].pShift -= pDiff;
245 
246  // Calculate new relative momentum for compensation shift.
247  double Qmove3 = 0.;
248  if (Qold < deltaQ3[iTab]) Qmove3 = Qold / 3.;
249  else if (Qold < maxQ3[iTab]) {
250  double realQbin = Qold / deltaQ3[iTab];
251  int intQbin = int( realQbin );
252  double inter = (pow3(realQbin) - pow3(intQbin))
253  / (3 * intQbin * (intQbin + 1) + 1);
254  Qmove3 = ( shift3[iTab][intQbin] + inter * (shift3[iTab][intQbin + 1]
255  - shift3[iTab][intQbin]) ) * psFac;
256  }
257  else Qmove3 = shift3[iTab][nStep3[iTab]] *psFac;
258  double Q2new3 = Q2old * pow( Qold / (Qold + 3. * lambda * Qmove3), 2. / 3.);
259 
260  // Calculate corresponding three-momentum shift.
261  Q2Diff = Q2new3 - Q2old;
262  sumQ2E = Q2Diff + eSum * eSum;
263  rootA = eSum * eDiff * p2AbsDiff - p2DiffAbs * sumQ2E;
264  rootB = p2DiffAbs * sumQ2E - p2AbsDiff * p2AbsDiff;
265  factor = 0.5 * ( rootA + sqrtpos(rootA * rootA
266  + Q2Diff * (sumQ2E - eDiff * eDiff) * rootB) ) / rootB;
267 
268  // Extra dampening factor to go from BE_3 to BE_32.
269  factor *= 1. - exp(-Q2old * R2Ref2);
270 
271  // Add shifts to sum. (Energy component dummy.)
272  pDiff = factor * (hadronBE[i1].p - hadronBE[i2].p);
273  hadronBE[i1].pComp += pDiff;
274  hadronBE[i2].pComp -= pDiff;
275 
276 }
277 
278 //==========================================================================
279 
280 } // end namespace Pythia8
Definition: AgUStep.h:26