Geant4 Cross Reference |
1 // 1 // 2 // ******************************************* 2 // ******************************************************************** 3 // * License and Disclaimer 3 // * License and Disclaimer * 4 // * 4 // * * 5 // * The Geant4 software is copyright of th 5 // * The Geant4 software is copyright of the Copyright Holders of * 6 // * the Geant4 Collaboration. It is provided 6 // * the Geant4 Collaboration. It is provided under the terms and * 7 // * conditions of the Geant4 Software License 7 // * conditions of the Geant4 Software License, included in the file * 8 // * LICENSE and available at http://cern.ch/ 8 // * LICENSE and available at http://cern.ch/geant4/license . These * 9 // * include a list of copyright holders. 9 // * include a list of copyright holders. * 10 // * 10 // * * 11 // * Neither the authors of this software syst 11 // * Neither the authors of this software system, nor their employing * 12 // * institutes,nor the agencies providing fin 12 // * institutes,nor the agencies providing financial support for this * 13 // * work make any representation or warran 13 // * work make any representation or warranty, express or implied, * 14 // * regarding this software system or assum 14 // * regarding this software system or assume any liability for its * 15 // * use. Please see the license in the file 15 // * use. Please see the license in the file LICENSE and URL above * 16 // * for the full disclaimer and the limitatio 16 // * for the full disclaimer and the limitation of liability. * 17 // * 17 // * * 18 // * This code implementation is the result 18 // * This code implementation is the result of the scientific and * 19 // * technical work of the GEANT4 collaboratio 19 // * technical work of the GEANT4 collaboration. * 20 // * By using, copying, modifying or distri 20 // * By using, copying, modifying or distributing the software (or * 21 // * any work based on the software) you ag 21 // * any work based on the software) you agree to acknowledge its * 22 // * use in resulting scientific publicati 22 // * use in resulting scientific publications, and indicate your * 23 // * acceptance of all terms of the Geant4 Sof 23 // * acceptance of all terms of the Geant4 Software license. * 24 // ******************************************* 24 // ******************************************************************** 25 // 25 // 26 // 26 // >> 27 // $Id: G4ChipsHyperonElasticXS.cc 70680 2013-06-04 07:51:03Z gcosmo $ 27 // 28 // 28 // 29 // 29 // G4 Physics class: G4ChipsHyperonElasticXS f 30 // G4 Physics class: G4ChipsHyperonElasticXS for pA elastic cross sections 30 // Created: M.V. Kossov, CERN/ITEP(Moscow), 5- 31 // Created: M.V. Kossov, CERN/ITEP(Moscow), 5-Feb-2010 31 // The last update: M.V. Kossov, CERN/ITEP (Mo 32 // The last update: M.V. Kossov, CERN/ITEP (Moscow) 5-Feb-2010 32 // 33 // 33 // ------------------------------------------- 34 // ------------------------------------------------------------------------------- 34 // Short description: Interaction cross-sectio 35 // Short description: Interaction cross-sections for the elastic process. 35 // Class extracted from CHIPS and integrated i 36 // Class extracted from CHIPS and integrated in Geant4 by W.Pokorski 36 // ------------------------------------------- 37 // ------------------------------------------------------------------------------- 37 // 38 // 38 39 39 #include "G4ChipsHyperonElasticXS.hh" 40 #include "G4ChipsHyperonElasticXS.hh" 40 #include "G4SystemOfUnits.hh" 41 #include "G4SystemOfUnits.hh" 41 #include "G4DynamicParticle.hh" 42 #include "G4DynamicParticle.hh" 42 #include "G4ParticleDefinition.hh" 43 #include "G4ParticleDefinition.hh" 43 #include "G4Lambda.hh" 44 #include "G4Lambda.hh" 44 #include "G4SigmaPlus.hh" 45 #include "G4SigmaPlus.hh" 45 #include "G4SigmaMinus.hh" 46 #include "G4SigmaMinus.hh" 46 #include "G4SigmaZero.hh" 47 #include "G4SigmaZero.hh" 47 #include "G4XiMinus.hh" 48 #include "G4XiMinus.hh" 48 #include "G4XiZero.hh" 49 #include "G4XiZero.hh" 49 #include "G4OmegaMinus.hh" 50 #include "G4OmegaMinus.hh" 50 #include "G4Nucleus.hh" 51 #include "G4Nucleus.hh" 51 #include "G4ParticleTable.hh" 52 #include "G4ParticleTable.hh" 52 #include "G4NucleiProperties.hh" 53 #include "G4NucleiProperties.hh" 53 #include "G4IonTable.hh" 54 #include "G4IonTable.hh" 54 #include "G4Exp.hh" << 55 #include "G4Log.hh" << 56 #include "G4Pow.hh" << 57 55 58 // factory 56 // factory 59 #include "G4CrossSectionFactory.hh" 57 #include "G4CrossSectionFactory.hh" 60 // 58 // 61 G4_DECLARE_XS_FACTORY(G4ChipsHyperonElasticXS) 59 G4_DECLARE_XS_FACTORY(G4ChipsHyperonElasticXS); 62 60 63 G4ChipsHyperonElasticXS::G4ChipsHyperonElastic 61 G4ChipsHyperonElasticXS::G4ChipsHyperonElasticXS():G4VCrossSectionDataSet(Default_Name()), nPoints(128), nLast(nPoints-1) 64 { 62 { 65 lPMin=-8.; //Min tabulatedLogarithmMomentum 63 lPMin=-8.; //Min tabulatedLogarithmMomentum(D) 66 lPMax= 8.; //Max tabulatedLogarithmMomentum 64 lPMax= 8.; //Max tabulatedLogarithmMomentum(D) 67 dlnP=(lPMax-lPMin)/nLast;// LogStep inTable 65 dlnP=(lPMax-lPMin)/nLast;// LogStep inTable (D) 68 onlyCS=true;//Flag toCalculOnlyCS(not Si/Bi) 66 onlyCS=true;//Flag toCalculOnlyCS(not Si/Bi)(L) 69 lastSIG=0.; //Last calculated cross section 67 lastSIG=0.; //Last calculated cross section (L) 70 lastLP=-10.;//LastLog(mom_of IncidentHadron) 68 lastLP=-10.;//LastLog(mom_of IncidentHadron)(L) 71 lastTM=0.; //Last t_maximum 69 lastTM=0.; //Last t_maximum (L) 72 theSS=0.; //TheLastSqSlope of 1st difr.Max 70 theSS=0.; //TheLastSqSlope of 1st difr.Max(L) 73 theS1=0.; //TheLastMantissa of 1st difrMax 71 theS1=0.; //TheLastMantissa of 1st difrMax(L) 74 theB1=0.; //TheLastSlope of 1st difructMax 72 theB1=0.; //TheLastSlope of 1st difructMax(L) 75 theS2=0.; //TheLastMantissa of 2nd difrMax 73 theS2=0.; //TheLastMantissa of 2nd difrMax(L) 76 theB2=0.; //TheLastSlope of 2nd difructMax 74 theB2=0.; //TheLastSlope of 2nd difructMax(L) 77 theS3=0.; //TheLastMantissa of 3d difr.Max 75 theS3=0.; //TheLastMantissa of 3d difr.Max(L) 78 theB3=0.; //TheLastSlope of 3d difruct.Max 76 theB3=0.; //TheLastSlope of 3d difruct.Max(L) 79 theS4=0.; //TheLastMantissa of 4th difrMax 77 theS4=0.; //TheLastMantissa of 4th difrMax(L) 80 theB4=0.; //TheLastSlope of 4th difructMax 78 theB4=0.; //TheLastSlope of 4th difructMax(L) 81 lastTZ=0; // Last atomic number of the tar 79 lastTZ=0; // Last atomic number of the target 82 lastTN=0; // Last # of neutrons in the tar 80 lastTN=0; // Last # of neutrons in the target 83 lastPIN=0.; // Last initialized max momentum 81 lastPIN=0.; // Last initialized max momentum 84 lastCST=0; // Elastic cross-section table 82 lastCST=0; // Elastic cross-section table 85 lastPAR=0; // ParametersForFunctionCalculat 83 lastPAR=0; // ParametersForFunctionCalculation 86 lastSST=0; // E-dep ofSqardSlope of 1st dif 84 lastSST=0; // E-dep ofSqardSlope of 1st difMax 87 lastS1T=0; // E-dep of mantissa of 1st dif. 85 lastS1T=0; // E-dep of mantissa of 1st dif.Max 88 lastB1T=0; // E-dep of the slope of 1st dif 86 lastB1T=0; // E-dep of the slope of 1st difMax 89 lastS2T=0; // E-dep of mantissa of 2nd difr 87 lastS2T=0; // E-dep of mantissa of 2nd difrMax 90 lastB2T=0; // E-dep of the slope of 2nd dif 88 lastB2T=0; // E-dep of the slope of 2nd difMax 91 lastS3T=0; // E-dep of mantissa of 3d difr. 89 lastS3T=0; // E-dep of mantissa of 3d difr.Max 92 lastB3T=0; // E-dep of the slope of 3d difr 90 lastB3T=0; // E-dep of the slope of 3d difrMax 93 lastS4T=0; // E-dep of mantissa of 4th difr 91 lastS4T=0; // E-dep of mantissa of 4th difrMax 94 lastB4T=0; // E-dep of the slope of 4th dif 92 lastB4T=0; // E-dep of the slope of 4th difMax 95 lastN=0; // The last N of calculated nucl 93 lastN=0; // The last N of calculated nucleus 96 lastZ=0; // The last Z of calculated nucl 94 lastZ=0; // The last Z of calculated nucleus 97 lastP=0.; // LastUsed inCrossSection Momen 95 lastP=0.; // LastUsed inCrossSection Momentum 98 lastTH=0.; // Last threshold momentum 96 lastTH=0.; // Last threshold momentum 99 lastCS=0.; // Last value of the Cross Secti 97 lastCS=0.; // Last value of the Cross Section 100 lastI=0; // The last position in the DAMD 98 lastI=0; // The last position in the DAMDB 101 } 99 } 102 100 103 G4ChipsHyperonElasticXS::~G4ChipsHyperonElasti 101 G4ChipsHyperonElasticXS::~G4ChipsHyperonElasticXS() 104 { 102 { 105 std::vector<G4double*>::iterator pos; 103 std::vector<G4double*>::iterator pos; 106 for (pos=CST.begin(); pos<CST.end(); pos++) 104 for (pos=CST.begin(); pos<CST.end(); pos++) 107 { delete [] *pos; } 105 { delete [] *pos; } 108 CST.clear(); 106 CST.clear(); 109 for (pos=PAR.begin(); pos<PAR.end(); pos++) 107 for (pos=PAR.begin(); pos<PAR.end(); pos++) 110 { delete [] *pos; } 108 { delete [] *pos; } 111 PAR.clear(); 109 PAR.clear(); 112 for (pos=SST.begin(); pos<SST.end(); pos++) 110 for (pos=SST.begin(); pos<SST.end(); pos++) 113 { delete [] *pos; } 111 { delete [] *pos; } 114 SST.clear(); 112 SST.clear(); 115 for (pos=S1T.begin(); pos<S1T.end(); pos++) 113 for (pos=S1T.begin(); pos<S1T.end(); pos++) 116 { delete [] *pos; } 114 { delete [] *pos; } 117 S1T.clear(); 115 S1T.clear(); 118 for (pos=B1T.begin(); pos<B1T.end(); pos++) 116 for (pos=B1T.begin(); pos<B1T.end(); pos++) 119 { delete [] *pos; } 117 { delete [] *pos; } 120 B1T.clear(); 118 B1T.clear(); 121 for (pos=S2T.begin(); pos<S2T.end(); pos++) 119 for (pos=S2T.begin(); pos<S2T.end(); pos++) 122 { delete [] *pos; } 120 { delete [] *pos; } 123 S2T.clear(); 121 S2T.clear(); 124 for (pos=B2T.begin(); pos<B2T.end(); pos++) 122 for (pos=B2T.begin(); pos<B2T.end(); pos++) 125 { delete [] *pos; } 123 { delete [] *pos; } 126 B2T.clear(); 124 B2T.clear(); 127 for (pos=S3T.begin(); pos<S3T.end(); pos++) 125 for (pos=S3T.begin(); pos<S3T.end(); pos++) 128 { delete [] *pos; } 126 { delete [] *pos; } 129 S3T.clear(); 127 S3T.clear(); 130 for (pos=B3T.begin(); pos<B3T.end(); pos++) 128 for (pos=B3T.begin(); pos<B3T.end(); pos++) 131 { delete [] *pos; } 129 { delete [] *pos; } 132 B3T.clear(); 130 B3T.clear(); 133 for (pos=S4T.begin(); pos<S4T.end(); pos++) 131 for (pos=S4T.begin(); pos<S4T.end(); pos++) 134 { delete [] *pos; } 132 { delete [] *pos; } 135 S4T.clear(); 133 S4T.clear(); 136 for (pos=B4T.begin(); pos<B4T.end(); pos++) 134 for (pos=B4T.begin(); pos<B4T.end(); pos++) 137 { delete [] *pos; } 135 { delete [] *pos; } 138 B4T.clear(); 136 B4T.clear(); 139 } 137 } 140 138 141 void << 139 G4bool G4ChipsHyperonElasticXS::IsIsoApplicable(const G4DynamicParticle* Pt, G4int, G4int, 142 G4ChipsHyperonElasticXS::CrossSectionDescripti << 143 { << 144 outFile << "G4ChipsHyperonElasticXS provid << 145 << "section for hyperon nucleus sc << 146 << "momentum. The cross section is << 147 << "CHIPS parameterization of cros << 148 } << 149 << 150 G4bool G4ChipsHyperonElasticXS::IsIsoApplicabl << 151 const G4Element*, 140 const G4Element*, 152 const G4Material*) 141 const G4Material*) 153 { 142 { 154 return true; << 143 G4ParticleDefinition* particle = Pt->GetDefinition(); >> 144 if (particle == G4Lambda::Lambda()) >> 145 { >> 146 return true; >> 147 } >> 148 else if(particle == G4SigmaPlus::SigmaPlus()) >> 149 { >> 150 return true; >> 151 } >> 152 else if(particle == G4SigmaMinus::SigmaMinus()) >> 153 { >> 154 return true; >> 155 } >> 156 else if(particle == G4SigmaZero::SigmaZero()) >> 157 { >> 158 return true; >> 159 } >> 160 else if(particle == G4XiMinus::XiMinus()) >> 161 { >> 162 return true; >> 163 } >> 164 else if(particle == G4XiZero::XiZero()) >> 165 { >> 166 return true; >> 167 } >> 168 else if(particle == G4OmegaMinus::OmegaMinus()) >> 169 { >> 170 return true; >> 171 } >> 172 return false; 155 } 173 } 156 174 157 // The main member function giving the collisi 175 // The main member function giving the collision cross section (P is in IU, CS is in mb) 158 // Make pMom in independent units ! (Now it is 176 // Make pMom in independent units ! (Now it is MeV) 159 G4double G4ChipsHyperonElasticXS::GetIsoCrossS 177 G4double G4ChipsHyperonElasticXS::GetIsoCrossSection(const G4DynamicParticle* Pt, G4int tgZ, G4int A, 160 const G4Isotope*, 178 const G4Isotope*, 161 const G4Element*, 179 const G4Element*, 162 const G4Material*) 180 const G4Material*) 163 { 181 { 164 G4double pMom=Pt->GetTotalMomentum(); 182 G4double pMom=Pt->GetTotalMomentum(); 165 G4int tgN = A - tgZ; 183 G4int tgN = A - tgZ; 166 G4int pdg = Pt->GetDefinition()->GetPDGEncod 184 G4int pdg = Pt->GetDefinition()->GetPDGEncoding(); 167 185 168 return GetChipsCrossSection(pMom, tgZ, tgN, 186 return GetChipsCrossSection(pMom, tgZ, tgN, pdg); 169 } 187 } 170 188 171 G4double G4ChipsHyperonElasticXS::GetChipsCros 189 G4double G4ChipsHyperonElasticXS::GetChipsCrossSection(G4double pMom, G4int tgZ, G4int tgN, G4int pPDG) 172 { 190 { >> 191 static G4ThreadLocal std::vector <G4int> *colN_G4MT_TLS_ = 0 ; if (!colN_G4MT_TLS_) colN_G4MT_TLS_ = new std::vector <G4int> ; std::vector <G4int> &colN = *colN_G4MT_TLS_; // Vector of N for calculated nuclei (isotops) >> 192 static G4ThreadLocal std::vector <G4int> *colZ_G4MT_TLS_ = 0 ; if (!colZ_G4MT_TLS_) colZ_G4MT_TLS_ = new std::vector <G4int> ; std::vector <G4int> &colZ = *colZ_G4MT_TLS_; // Vector of Z for calculated nuclei (isotops) >> 193 static G4ThreadLocal std::vector <G4double> *colP_G4MT_TLS_ = 0 ; if (!colP_G4MT_TLS_) colP_G4MT_TLS_ = new std::vector <G4double> ; std::vector <G4double> &colP = *colP_G4MT_TLS_; // Vector of last momenta for the reaction >> 194 static G4ThreadLocal std::vector <G4double> *colTH_G4MT_TLS_ = 0 ; if (!colTH_G4MT_TLS_) colTH_G4MT_TLS_ = new std::vector <G4double> ; std::vector <G4double> &colTH = *colTH_G4MT_TLS_; // Vector of energy thresholds for the reaction >> 195 static G4ThreadLocal std::vector <G4double> *colCS_G4MT_TLS_ = 0 ; if (!colCS_G4MT_TLS_) colCS_G4MT_TLS_ = new std::vector <G4double> ; std::vector <G4double> &colCS = *colCS_G4MT_TLS_; // Vector of last cross sections for the reaction >> 196 // ***---*** End of the mandatory Static Definitions of the Associative Memory ***---*** 173 197 174 G4bool fCS = false; 198 G4bool fCS = false; 175 G4double pEn=pMom; 199 G4double pEn=pMom; 176 200 177 onlyCS=fCS; 201 onlyCS=fCS; 178 202 179 G4bool in=false; // By def 203 G4bool in=false; // By default the isotope must be found in the AMDB 180 lastP = 0.; // New mo 204 lastP = 0.; // New momentum history (nothing to compare with) 181 lastN = tgN; // The la 205 lastN = tgN; // The last N of the calculated nucleus 182 lastZ = tgZ; // The la 206 lastZ = tgZ; // The last Z of the calculated nucleus 183 lastI = (G4int)colN.size(); // Size o << 207 lastI = colN.size(); // Size of the Associative Memory DB in the heap 184 if(lastI) for(G4int i=0; i<lastI; ++i) // Lo << 208 if(lastI) for(G4int i=0; i<lastI; i++) // Loop over proj/tgZ/tgN lines of DB 185 { // The nu 209 { // The nucleus with projPDG is found in AMDB 186 if(colN[i]==tgN && colZ[i]==tgZ) // Isotop 210 if(colN[i]==tgN && colZ[i]==tgZ) // Isotope is foind in AMDB 187 { 211 { 188 lastI=i; 212 lastI=i; 189 lastTH =colTH[i]; // Last T 213 lastTH =colTH[i]; // Last THreshold (A-dependent) 190 if(pEn<=lastTH) 214 if(pEn<=lastTH) 191 { 215 { 192 return 0.; // Energy 216 return 0.; // Energy is below the Threshold value 193 } 217 } 194 lastP =colP [i]; // Last M 218 lastP =colP [i]; // Last Momentum (A-dependent) 195 lastCS =colCS[i]; // Last C 219 lastCS =colCS[i]; // Last CrossSect (A-dependent) 196 // if(std::fabs(lastP/pMom-1.)<toleranc 220 // if(std::fabs(lastP/pMom-1.)<tolerance) //VI (do not use tolerance) 197 if(lastP == pMom) // Do not 221 if(lastP == pMom) // Do not recalculate 198 { 222 { 199 CalculateCrossSection(fCS,-1,i,pPDG,la 223 CalculateCrossSection(fCS,-1,i,pPDG,lastZ,lastN,pMom); // Update param's only 200 return lastCS*millibarn; // Use th 224 return lastCS*millibarn; // Use theLastCS 201 } 225 } 202 in = true; // This 226 in = true; // This is the case when the isotop is found in DB 203 // Momentum pMom is in IU ! @@ Units 227 // Momentum pMom is in IU ! @@ Units 204 lastCS=CalculateCrossSection(fCS,-1,i,pP 228 lastCS=CalculateCrossSection(fCS,-1,i,pPDG,lastZ,lastN,pMom); // read & update 205 if(lastCS<=0. && pEn>lastTH) // Corre 229 if(lastCS<=0. && pEn>lastTH) // Correct the threshold 206 { 230 { 207 lastTH=pEn; 231 lastTH=pEn; 208 } 232 } 209 break; // Go o 233 break; // Go out of the LOOP with found lastI 210 } 234 } 211 } // End of attampt to find the nucleus in D 235 } // End of attampt to find the nucleus in DB 212 if(!in) // This n 236 if(!in) // This nucleus has not been calculated previously 213 { 237 { 214 //!!The slave functions must provide cross 238 //!!The slave functions must provide cross-sections in millibarns (mb) !! (not in IU) 215 lastCS=CalculateCrossSection(fCS,0,lastI,p 239 lastCS=CalculateCrossSection(fCS,0,lastI,pPDG,lastZ,lastN,pMom);//calculate&create 216 if(lastCS<=0.) 240 if(lastCS<=0.) 217 { 241 { 218 lastTH = 0; //ThresholdEnergy(tgZ, tgN); 242 lastTH = 0; //ThresholdEnergy(tgZ, tgN); // The Threshold Energy which is now the last 219 if(pEn>lastTH) 243 if(pEn>lastTH) 220 { 244 { 221 lastTH=pEn; 245 lastTH=pEn; 222 } 246 } 223 } 247 } 224 colN.push_back(tgN); 248 colN.push_back(tgN); 225 colZ.push_back(tgZ); 249 colZ.push_back(tgZ); 226 colP.push_back(pMom); 250 colP.push_back(pMom); 227 colTH.push_back(lastTH); 251 colTH.push_back(lastTH); 228 colCS.push_back(lastCS); 252 colCS.push_back(lastCS); 229 return lastCS*millibarn; 253 return lastCS*millibarn; 230 } // End of creation of the new set of param 254 } // End of creation of the new set of parameters 231 else 255 else 232 { 256 { 233 colP[lastI]=pMom; 257 colP[lastI]=pMom; 234 colCS[lastI]=lastCS; 258 colCS[lastI]=lastCS; 235 } 259 } 236 return lastCS*millibarn; 260 return lastCS*millibarn; 237 } 261 } 238 262 239 // Calculation of total elastic cross section 263 // Calculation of total elastic cross section (p in IU, CS in mb) @@ Units (?) 240 // F=0 - create AMDB, F=-1 - read&update AMDB, 264 // F=0 - create AMDB, F=-1 - read&update AMDB, F=1 - update AMDB (sinchro with higher AMDB) 241 G4double G4ChipsHyperonElasticXS::CalculateCro 265 G4double G4ChipsHyperonElasticXS::CalculateCrossSection(G4bool CS,G4int F,G4int I, 242 G 266 G4int PDG, G4int tgZ, G4int tgN, G4double pIU) 243 { 267 { >> 268 // *** Begin of Associative Memory DB for acceleration of the cross section calculations >> 269 static G4ThreadLocal std::vector <G4double> *PIN_G4MT_TLS_ = 0 ; if (!PIN_G4MT_TLS_) PIN_G4MT_TLS_ = new std::vector <G4double> ; std::vector <G4double> &PIN = *PIN_G4MT_TLS_; // Vector of max initialized log(P) in the table >> 270 // *** End of Static Definitions (Associative Memory Data Base) *** 244 G4double pMom=pIU/GeV; // All 271 G4double pMom=pIU/GeV; // All calculations are in GeV 245 onlyCS=CS; // Fla 272 onlyCS=CS; // Flag to calculate only CS (not Si/Bi) 246 lastLP=G4Log(pMom); // Make a << 273 lastLP=std::log(pMom); // Make a logarithm of the momentum for calculation 247 if(F) // Thi 274 if(F) // This isotope was found in AMDB =>RETRIEVE/UPDATE 248 { 275 { 249 if(F<0) // the 276 if(F<0) // the AMDB must be loded 250 { 277 { 251 lastPIN = PIN[I]; // Max 278 lastPIN = PIN[I]; // Max log(P) initialised for this table set 252 lastPAR = PAR[I]; // Poi 279 lastPAR = PAR[I]; // Pointer to the parameter set 253 280 254 lastCST = CST[I]; // Poi 281 lastCST = CST[I]; // Pointer to the total sross-section table 255 lastSST = SST[I]; // Poi 282 lastSST = SST[I]; // Pointer to the first squared slope 256 lastS1T = S1T[I]; // Poi 283 lastS1T = S1T[I]; // Pointer to the first mantissa 257 lastB1T = B1T[I]; // Poi 284 lastB1T = B1T[I]; // Pointer to the first slope 258 lastS2T = S2T[I]; // Poi 285 lastS2T = S2T[I]; // Pointer to the second mantissa 259 lastB2T = B2T[I]; // Poi 286 lastB2T = B2T[I]; // Pointer to the second slope 260 lastS3T = S3T[I]; // Poi 287 lastS3T = S3T[I]; // Pointer to the third mantissa 261 lastB3T = B3T[I]; // Poi 288 lastB3T = B3T[I]; // Pointer to the rhird slope 262 lastS4T = S4T[I]; // Poi 289 lastS4T = S4T[I]; // Pointer to the 4-th mantissa 263 lastB4T = B4T[I]; // Poi 290 lastB4T = B4T[I]; // Pointer to the 4-th slope 264 } 291 } 265 if(lastLP>lastPIN && lastLP<lPMax) 292 if(lastLP>lastPIN && lastLP<lPMax) 266 { 293 { 267 lastPIN=GetPTables(lastLP,lastPIN,PDG,tg 294 lastPIN=GetPTables(lastLP,lastPIN,PDG,tgZ,tgN);// Can update upper logP-Limit in tabs 268 PIN[I]=lastPIN; // Rem 295 PIN[I]=lastPIN; // Remember the new P-Limit of the tables 269 } 296 } 270 } 297 } 271 else // Thi 298 else // This isotope wasn't initialized => CREATE 272 { 299 { 273 lastPAR = new G4double[nPoints]; // All 300 lastPAR = new G4double[nPoints]; // Allocate memory for parameters of CS function 274 lastPAR[nLast]=0; // Ini 301 lastPAR[nLast]=0; // Initialization for VALGRIND 275 lastCST = new G4double[nPoints]; // All 302 lastCST = new G4double[nPoints]; // Allocate memory for Tabulated CS function 276 lastSST = new G4double[nPoints]; // All 303 lastSST = new G4double[nPoints]; // Allocate memory for Tabulated first sqaredSlope 277 lastS1T = new G4double[nPoints]; // All 304 lastS1T = new G4double[nPoints]; // Allocate memory for Tabulated first mantissa 278 lastB1T = new G4double[nPoints]; // All 305 lastB1T = new G4double[nPoints]; // Allocate memory for Tabulated first slope 279 lastS2T = new G4double[nPoints]; // All 306 lastS2T = new G4double[nPoints]; // Allocate memory for Tabulated second mantissa 280 lastB2T = new G4double[nPoints]; // All 307 lastB2T = new G4double[nPoints]; // Allocate memory for Tabulated second slope 281 lastS3T = new G4double[nPoints]; // All 308 lastS3T = new G4double[nPoints]; // Allocate memory for Tabulated third mantissa 282 lastB3T = new G4double[nPoints]; // All 309 lastB3T = new G4double[nPoints]; // Allocate memory for Tabulated third slope 283 lastS4T = new G4double[nPoints]; // All 310 lastS4T = new G4double[nPoints]; // Allocate memory for Tabulated 4-th mantissa 284 lastB4T = new G4double[nPoints]; // All 311 lastB4T = new G4double[nPoints]; // Allocate memory for Tabulated 4-th slope 285 lastPIN = GetPTables(lastLP,lPMin,PDG,tgZ, 312 lastPIN = GetPTables(lastLP,lPMin,PDG,tgZ,tgN); // Returns the new P-limit for tables 286 PIN.push_back(lastPIN); // Fil 313 PIN.push_back(lastPIN); // Fill parameters of CS function to AMDB 287 PAR.push_back(lastPAR); // Fil 314 PAR.push_back(lastPAR); // Fill parameters of CS function to AMDB 288 CST.push_back(lastCST); // Fil 315 CST.push_back(lastCST); // Fill Tabulated CS function to AMDB 289 SST.push_back(lastSST); // Fil 316 SST.push_back(lastSST); // Fill Tabulated first sq.slope to AMDB 290 S1T.push_back(lastS1T); // Fil 317 S1T.push_back(lastS1T); // Fill Tabulated first mantissa to AMDB 291 B1T.push_back(lastB1T); // Fil 318 B1T.push_back(lastB1T); // Fill Tabulated first slope to AMDB 292 S2T.push_back(lastS2T); // Fil 319 S2T.push_back(lastS2T); // Fill Tabulated second mantissa to AMDB 293 B2T.push_back(lastB2T); // Fil 320 B2T.push_back(lastB2T); // Fill Tabulated second slope to AMDB 294 S3T.push_back(lastS3T); // Fil 321 S3T.push_back(lastS3T); // Fill Tabulated third mantissa to AMDB 295 B3T.push_back(lastB3T); // Fil 322 B3T.push_back(lastB3T); // Fill Tabulated third slope to AMDB 296 S4T.push_back(lastS4T); // Fil 323 S4T.push_back(lastS4T); // Fill Tabulated 4-th mantissa to AMDB 297 B4T.push_back(lastB4T); // Fil 324 B4T.push_back(lastB4T); // Fill Tabulated 4-th slope to AMDB 298 } // End of creation/update of the new set o 325 } // End of creation/update of the new set of parameters and tables 299 // =-----------= NOW Update (if necessary) a 326 // =-----------= NOW Update (if necessary) and Calculate the Cross Section =-----------= 300 if(lastLP>lastPIN && lastLP<lPMax) 327 if(lastLP>lastPIN && lastLP<lPMax) 301 { 328 { 302 lastPIN = GetPTables(lastLP,lastPIN,PDG,tg 329 lastPIN = GetPTables(lastLP,lastPIN,PDG,tgZ,tgN); 303 } 330 } 304 if(!onlyCS) lastTM=GetQ2max(PDG, tgZ, tgN, p 331 if(!onlyCS) lastTM=GetQ2max(PDG, tgZ, tgN, pMom); // Calculate (-t)_max=Q2_max (GeV2) 305 if(lastLP>lPMin && lastLP<=lastPIN) // Lin 332 if(lastLP>lPMin && lastLP<=lastPIN) // Linear fit is made using precalculated tables 306 { 333 { 307 if(lastLP==lastPIN) 334 if(lastLP==lastPIN) 308 { 335 { 309 G4double shift=(lastLP-lPMin)/dlnP+.0000 336 G4double shift=(lastLP-lPMin)/dlnP+.000001; // Log distance from lPMin 310 G4int blast=static_cast<int>(shift); 337 G4int blast=static_cast<int>(shift); // this is a bin number of the lower edge (0) 311 if(blast<0 || blast>=nLast)G4cout<<"G4QH 338 if(blast<0 || blast>=nLast)G4cout<<"G4QHyperElCS::CCS:b="<<blast<<","<<nLast<<G4endl; 312 lastSIG = lastCST[blast]; 339 lastSIG = lastCST[blast]; 313 if(!onlyCS) // Ski 340 if(!onlyCS) // Skip the differential cross-section parameters 314 { 341 { 315 theSS = lastSST[blast]; 342 theSS = lastSST[blast]; 316 theS1 = lastS1T[blast]; 343 theS1 = lastS1T[blast]; 317 theB1 = lastB1T[blast]; 344 theB1 = lastB1T[blast]; 318 theS2 = lastS2T[blast]; 345 theS2 = lastS2T[blast]; 319 theB2 = lastB2T[blast]; 346 theB2 = lastB2T[blast]; 320 theS3 = lastS3T[blast]; 347 theS3 = lastS3T[blast]; 321 theB3 = lastB3T[blast]; 348 theB3 = lastB3T[blast]; 322 theS4 = lastS4T[blast]; 349 theS4 = lastS4T[blast]; 323 theB4 = lastB4T[blast]; 350 theB4 = lastB4T[blast]; 324 } 351 } 325 } 352 } 326 else 353 else 327 { 354 { 328 G4double shift=(lastLP-lPMin)/dlnP; 355 G4double shift=(lastLP-lPMin)/dlnP; // a shift from the beginning of the table 329 G4int blast=static_cast<int>(shift); 356 G4int blast=static_cast<int>(shift); // the lower bin number 330 if(blast<0) blast=0; 357 if(blast<0) blast=0; 331 if(blast>=nLast) blast=nLast-1; 358 if(blast>=nLast) blast=nLast-1; // low edge of the last bin 332 shift-=blast; 359 shift-=blast; // step inside the unit bin 333 G4int lastL=blast+1; 360 G4int lastL=blast+1; // the upper bin number 334 G4double SIGL=lastCST[blast]; 361 G4double SIGL=lastCST[blast]; // the basic value of the cross-section 335 lastSIG= SIGL+shift*(lastCST[lastL]-SIGL 362 lastSIG= SIGL+shift*(lastCST[lastL]-SIGL); // calculated total elastic cross-section 336 if(!onlyCS) // Ski 363 if(!onlyCS) // Skip the differential cross-section parameters 337 { 364 { 338 G4double SSTL=lastSST[blast]; 365 G4double SSTL=lastSST[blast]; // the low bin of the first squared slope 339 theSS=SSTL+shift*(lastSST[lastL]-SSTL) 366 theSS=SSTL+shift*(lastSST[lastL]-SSTL); // the basic value of the first sq.slope 340 G4double S1TL=lastS1T[blast]; 367 G4double S1TL=lastS1T[blast]; // the low bin of the first mantissa 341 theS1=S1TL+shift*(lastS1T[lastL]-S1TL) 368 theS1=S1TL+shift*(lastS1T[lastL]-S1TL); // the basic value of the first mantissa 342 G4double B1TL=lastB1T[blast]; 369 G4double B1TL=lastB1T[blast]; // the low bin of the first slope 343 theB1=B1TL+shift*(lastB1T[lastL]-B1TL) 370 theB1=B1TL+shift*(lastB1T[lastL]-B1TL); // the basic value of the first slope 344 G4double S2TL=lastS2T[blast]; 371 G4double S2TL=lastS2T[blast]; // the low bin of the second mantissa 345 theS2=S2TL+shift*(lastS2T[lastL]-S2TL) 372 theS2=S2TL+shift*(lastS2T[lastL]-S2TL); // the basic value of the second mantissa 346 G4double B2TL=lastB2T[blast]; 373 G4double B2TL=lastB2T[blast]; // the low bin of the second slope 347 theB2=B2TL+shift*(lastB2T[lastL]-B2TL) 374 theB2=B2TL+shift*(lastB2T[lastL]-B2TL); // the basic value of the second slope 348 G4double S3TL=lastS3T[blast]; 375 G4double S3TL=lastS3T[blast]; // the low bin of the third mantissa 349 theS3=S3TL+shift*(lastS3T[lastL]-S3TL) 376 theS3=S3TL+shift*(lastS3T[lastL]-S3TL); // the basic value of the third mantissa 350 G4double B3TL=lastB3T[blast]; 377 G4double B3TL=lastB3T[blast]; // the low bin of the third slope 351 theB3=B3TL+shift*(lastB3T[lastL]-B3TL) 378 theB3=B3TL+shift*(lastB3T[lastL]-B3TL); // the basic value of the third slope 352 G4double S4TL=lastS4T[blast]; 379 G4double S4TL=lastS4T[blast]; // the low bin of the 4-th mantissa 353 theS4=S4TL+shift*(lastS4T[lastL]-S4TL) 380 theS4=S4TL+shift*(lastS4T[lastL]-S4TL); // the basic value of the 4-th mantissa 354 G4double B4TL=lastB4T[blast]; 381 G4double B4TL=lastB4T[blast]; // the low bin of the 4-th slope 355 theB4=B4TL+shift*(lastB4T[lastL]-B4TL) 382 theB4=B4TL+shift*(lastB4T[lastL]-B4TL); // the basic value of the 4-th slope 356 } 383 } 357 } 384 } 358 } 385 } 359 else lastSIG=GetTabValues(lastLP, PDG, tgZ, 386 else lastSIG=GetTabValues(lastLP, PDG, tgZ, tgN); // Direct calculation beyond the table 360 if(lastSIG<0.) lastSIG = 0.; 387 if(lastSIG<0.) lastSIG = 0.; // @@ a Warning print can be added 361 return lastSIG; 388 return lastSIG; 362 } 389 } 363 390 364 // It has parameter sets for all tZ/tN/PDG, us 391 // It has parameter sets for all tZ/tN/PDG, using them the tables can be created/updated 365 G4double G4ChipsHyperonElasticXS::GetPTables(G 392 G4double G4ChipsHyperonElasticXS::GetPTables(G4double LP, G4double ILP, G4int PDG, 366 393 G4int tgZ, G4int tgN) 367 { 394 { 368 // @@ At present all nA==pA ---------> Each 395 // @@ At present all nA==pA ---------> Each neucleus can have not more than 51 parameters 369 static const G4double pwd=2727; 396 static const G4double pwd=2727; 370 const G4int n_hypel=33; // #o 397 const G4int n_hypel=33; // #of parameters for pp-elastic (<nPoints=128) 371 // -0- -1- -2- -3- 398 // -0- -1- -2- -3- -4- -5- -6--7--8--9--10--11--12-13--14- 372 G4double hyp_el[n_hypel]={1.,.002,.12,.0557, 399 G4double hyp_el[n_hypel]={1.,.002,.12,.0557,3.5,6.72,99.,2.,3.,5.,74.,3.,3.4,.2,.17, 373 .001,8.,.055,3.64, 400 .001,8.,.055,3.64,5.e-5,4000.,1500.,.46,1.2e6,3.5e6,5.e-5, 374 1.e10,8.5e8,1.e10, 401 1.e10,8.5e8,1.e10,1.1,3.4e6,6.8e6,0.}; 375 // -15--16- -17- -18- 402 // -15--16- -17- -18- -19- -20- -21- -22- -23- -24- -25- 376 // -26- -27- -28- 403 // -26- -27- -28- -29- -30- -31- -32- 377 //AR-04Jun2014 if(PDG!=3222 && PDG>3000 && << 404 if(PDG!=3222 && PDG>3000 && PDG<3335) 378 if(PDG>3000 && PDG<3335) << 379 { 405 { 380 // -- Total pp elastic cross section cs & 406 // -- Total pp elastic cross section cs & s1/b1 (main), s2/b2 (tail1), s3/b3 (tail2) -- 381 //p2=p*p;p3=p2*p;sp=sqrt(p);p2s=p2*sp;lp=l 407 //p2=p*p;p3=p2*p;sp=sqrt(p);p2s=p2*sp;lp=log(p);dl1=lp-(3.=par(3));p4=p2*p2; p=|3-mom| 382 //CS=2.865/p2s/(1+.0022/p2s)+(18.9+.6461*d 408 //CS=2.865/p2s/(1+.0022/p2s)+(18.9+.6461*dl1*dl1+9./p)/(1.+.425*lp)/(1.+.4276/p4); 383 // par(0) par(7) par(1) par(2) 409 // par(0) par(7) par(1) par(2) par(4) par(5) par(6) 384 //dl2=lp-5., s1=(74.+3.*dl2*dl2)/(1+3.4/p4 410 //dl2=lp-5., s1=(74.+3.*dl2*dl2)/(1+3.4/p4/p)+(.2/p2+17.*p)/(p4+.001*sp), 385 // par(8) par(9) par(10) par(11 411 // par(8) par(9) par(10) par(11) par(12)par(13) par(14) 386 // b1=8.*p**.055/(1.+3.64/p3); s2=5.e-5+40 412 // b1=8.*p**.055/(1.+3.64/p3); s2=5.e-5+4000./(p4+1500.*p); b2=.46+1.2e6/(p4+3.5e6/sp); 387 // par(15) par(16) par(17) par(18) pa 413 // par(15) par(16) par(17) par(18) par(19) par(20) par(21) par(22) par(23) 388 // s3=5.e-5+1.e10/(p4*p4+8.5e8*p2+1.e10); 414 // s3=5.e-5+1.e10/(p4*p4+8.5e8*p2+1.e10); b3=1.1+3.4e6/(p4+6.8e6); ss=0. 389 // par(24) par(25) par(26) par(27) p 415 // par(24) par(25) par(26) par(27) par(28) par(29) par(30) par(31) 390 // 416 // 391 if(lastPAR[nLast]!=pwd) // A unique flag t 417 if(lastPAR[nLast]!=pwd) // A unique flag to avoid the repeatable definition 392 { 418 { 393 if ( tgZ == 1 && tgN == 0 ) 419 if ( tgZ == 1 && tgN == 0 ) 394 { 420 { 395 for (G4int ip=0; ip<n_hypel; ip++) las 421 for (G4int ip=0; ip<n_hypel; ip++) lastPAR[ip]=hyp_el[ip]; // Hyperon+P 396 } 422 } 397 else 423 else 398 { 424 { 399 G4double a=tgZ+tgN; 425 G4double a=tgZ+tgN; 400 G4double sa=std::sqrt(a); 426 G4double sa=std::sqrt(a); 401 G4double ssa=std::sqrt(sa); 427 G4double ssa=std::sqrt(sa); 402 G4double asa=a*sa; 428 G4double asa=a*sa; 403 G4double a2=a*a; 429 G4double a2=a*a; 404 G4double a3=a2*a; 430 G4double a3=a2*a; 405 G4double a4=a3*a; 431 G4double a4=a3*a; 406 G4double a5=a4*a; 432 G4double a5=a4*a; 407 G4double a6=a4*a2; 433 G4double a6=a4*a2; 408 G4double a7=a6*a; 434 G4double a7=a6*a; 409 G4double a8=a7*a; 435 G4double a8=a7*a; 410 G4double a9=a8*a; 436 G4double a9=a8*a; 411 G4double a10=a5*a5; 437 G4double a10=a5*a5; 412 G4double a12=a6*a6; 438 G4double a12=a6*a6; 413 G4double a14=a7*a7; 439 G4double a14=a7*a7; 414 G4double a16=a8*a8; 440 G4double a16=a8*a8; 415 G4double a17=a16*a; 441 G4double a17=a16*a; 416 //G4double a20=a16*a4; 442 //G4double a20=a16*a4; 417 G4double a32=a16*a16; 443 G4double a32=a16*a16; 418 // Reaction cross-section parameters ( 444 // Reaction cross-section parameters (pel=peh_fit.f) 419 lastPAR[0]=4./(1.+22/asa); 445 lastPAR[0]=4./(1.+22/asa); // p1 420 lastPAR[1]=2.36*asa/(1.+a*.055/ssa); 446 lastPAR[1]=2.36*asa/(1.+a*.055/ssa); // p2 421 lastPAR[2]=(1.+.00007*a3/ssa)/(1.+.002 447 lastPAR[2]=(1.+.00007*a3/ssa)/(1.+.0026*a2); // p3 422 lastPAR[3]=1.76*a/ssa+.00003*a3; 448 lastPAR[3]=1.76*a/ssa+.00003*a3; // p4 423 lastPAR[4]=(.03+200./a3)/(1.+1.E5/a3/s 449 lastPAR[4]=(.03+200./a3)/(1.+1.E5/a3/sa); // p5 424 lastPAR[5]=5.; 450 lastPAR[5]=5.; // p6 425 lastPAR[6]=0.; 451 lastPAR[6]=0.; // p7 not used 426 lastPAR[7]=0.; 452 lastPAR[7]=0.; // p8 not used 427 lastPAR[8]=0.; 453 lastPAR[8]=0.; // p9 not used 428 // @@ the differential cross-section i 454 // @@ the differential cross-section is parameterized separately for A>6 & A<7 429 if(a<6.5) 455 if(a<6.5) 430 { 456 { 431 G4double a28=a16*a12; 457 G4double a28=a16*a12; 432 // The main pre-exponent (pel_s 458 // The main pre-exponent (pel_sg) 433 lastPAR[ 9]=4000*a; 459 lastPAR[ 9]=4000*a; // p1 434 lastPAR[10]=1.2e7*a8+380*a17; 460 lastPAR[10]=1.2e7*a8+380*a17; // p2 435 lastPAR[11]=.7/(1.+4.e-12*a16); 461 lastPAR[11]=.7/(1.+4.e-12*a16); // p3 436 lastPAR[12]=2.5/a8/(a4+1.e-16*a32); 462 lastPAR[12]=2.5/a8/(a4+1.e-16*a32); // p4 437 lastPAR[13]=.28*a; 463 lastPAR[13]=.28*a; // p5 438 lastPAR[14]=1.2*a2+2.3; 464 lastPAR[14]=1.2*a2+2.3; // p6 439 lastPAR[15]=3.8/a; 465 lastPAR[15]=3.8/a; // p7 440 // The main slope (pel_s 466 // The main slope (pel_sl) 441 lastPAR[16]=.01/(1.+.0024*a5); 467 lastPAR[16]=.01/(1.+.0024*a5); // p1 442 lastPAR[17]=.2*a; 468 lastPAR[17]=.2*a; // p2 443 lastPAR[18]=9.e-7/(1.+.035*a5); 469 lastPAR[18]=9.e-7/(1.+.035*a5); // p3 444 lastPAR[19]=(42.+2.7e-11*a16)/(1.+.1 470 lastPAR[19]=(42.+2.7e-11*a16)/(1.+.14*a); // p4 445 // The main quadratic (pel_s 471 // The main quadratic (pel_sh) 446 lastPAR[20]=2.25*a3; 472 lastPAR[20]=2.25*a3; // p1 447 lastPAR[21]=18.; 473 lastPAR[21]=18.; // p2 448 lastPAR[22]=2.4e-3*a8/(1.+2.6e-4*a7) 474 lastPAR[22]=2.4e-3*a8/(1.+2.6e-4*a7); // p3 449 lastPAR[23]=3.5e-36*a32*a8/(1.+5.e-1 475 lastPAR[23]=3.5e-36*a32*a8/(1.+5.e-15*a32/a); // p4 450 // The 1st max pre-exponent (pel_q 476 // The 1st max pre-exponent (pel_qq) 451 lastPAR[24]=1.e5/(a8+2.5e12/a16); 477 lastPAR[24]=1.e5/(a8+2.5e12/a16); // p1 452 lastPAR[25]=8.e7/(a12+1.e-27*a28*a28 478 lastPAR[25]=8.e7/(a12+1.e-27*a28*a28); // p2 453 lastPAR[26]=.0006*a3; 479 lastPAR[26]=.0006*a3; // p3 454 // The 1st max slope (pel_q 480 // The 1st max slope (pel_qs) 455 lastPAR[27]=10.+4.e-8*a12*a; 481 lastPAR[27]=10.+4.e-8*a12*a; // p1 456 lastPAR[28]=.114; 482 lastPAR[28]=.114; // p2 457 lastPAR[29]=.003; 483 lastPAR[29]=.003; // p3 458 lastPAR[30]=2.e-23; 484 lastPAR[30]=2.e-23; // p4 459 // The effective pre-exponent (pel_s 485 // The effective pre-exponent (pel_ss) 460 lastPAR[31]=1./(1.+.0001*a8); 486 lastPAR[31]=1./(1.+.0001*a8); // p1 461 lastPAR[32]=1.5e-4/(1.+5.e-6*a12); 487 lastPAR[32]=1.5e-4/(1.+5.e-6*a12); // p2 462 lastPAR[33]=.03; 488 lastPAR[33]=.03; // p3 463 // The effective slope (pel_s 489 // The effective slope (pel_sb) 464 lastPAR[34]=a/2; 490 lastPAR[34]=a/2; // p1 465 lastPAR[35]=2.e-7*a4; 491 lastPAR[35]=2.e-7*a4; // p2 466 lastPAR[36]=4.; 492 lastPAR[36]=4.; // p3 467 lastPAR[37]=64./a3; 493 lastPAR[37]=64./a3; // p4 468 // The gloria pre-exponent (pel_u 494 // The gloria pre-exponent (pel_us) 469 lastPAR[38]=1.e8*G4Exp(.32*asa); << 495 lastPAR[38]=1.e8*std::exp(.32*asa); // p1 470 lastPAR[39]=20.*G4Exp(.45*asa); << 496 lastPAR[39]=20.*std::exp(.45*asa); // p2 471 lastPAR[40]=7.e3+2.4e6/a5; 497 lastPAR[40]=7.e3+2.4e6/a5; // p3 472 lastPAR[41]=2.5e5*G4Exp(.085*a3); << 498 lastPAR[41]=2.5e5*std::exp(.085*a3); // p4 473 lastPAR[42]=2.5*a; 499 lastPAR[42]=2.5*a; // p5 474 // The gloria slope (pel_u 500 // The gloria slope (pel_ub) 475 lastPAR[43]=920.+.03*a8*a3; 501 lastPAR[43]=920.+.03*a8*a3; // p1 476 lastPAR[44]=93.+.0023*a12; 502 lastPAR[44]=93.+.0023*a12; // p2 477 } 503 } 478 else 504 else 479 { 505 { 480 G4double p1a10=2.2e-28*a10; 506 G4double p1a10=2.2e-28*a10; 481 G4double r4a16=6.e14/a16; 507 G4double r4a16=6.e14/a16; 482 G4double s4a16=r4a16*r4a16; 508 G4double s4a16=r4a16*r4a16; 483 // a24 509 // a24 484 // a36 510 // a36 485 // The main pre-exponent (peh_s 511 // The main pre-exponent (peh_sg) 486 lastPAR[ 9]=4.5*G4Pow::GetInstance() << 512 lastPAR[ 9]=4.5*std::pow(a,1.15); // p1 487 lastPAR[10]=.06*G4Pow::GetInstance() << 513 lastPAR[10]=.06*std::pow(a,.6); // p2 488 lastPAR[11]=.6*a/(1.+2.e15/a16); 514 lastPAR[11]=.6*a/(1.+2.e15/a16); // p3 489 lastPAR[12]=.17/(a+9.e5/a3+1.5e33/a3 515 lastPAR[12]=.17/(a+9.e5/a3+1.5e33/a32); // p4 490 lastPAR[13]=(.001+7.e-11*a5)/(1.+4.4 516 lastPAR[13]=(.001+7.e-11*a5)/(1.+4.4e-11*a5); // p5 491 lastPAR[14]=(p1a10*p1a10+2.e-29)/(1. 517 lastPAR[14]=(p1a10*p1a10+2.e-29)/(1.+2.e-22*a12); // p6 492 // The main slope (peh_s 518 // The main slope (peh_sl) 493 lastPAR[15]=400./a12+2.e-22*a9; 519 lastPAR[15]=400./a12+2.e-22*a9; // p1 494 lastPAR[16]=1.e-32*a12/(1.+5.e22/a14 520 lastPAR[16]=1.e-32*a12/(1.+5.e22/a14); // p2 495 lastPAR[17]=1000./a2+9.5*sa*ssa; 521 lastPAR[17]=1000./a2+9.5*sa*ssa; // p3 496 lastPAR[18]=4.e-6*a*asa+1.e11/a16; 522 lastPAR[18]=4.e-6*a*asa+1.e11/a16; // p4 497 lastPAR[19]=(120./a+.002*a2)/(1.+2.e 523 lastPAR[19]=(120./a+.002*a2)/(1.+2.e14/a16); // p5 498 lastPAR[20]=9.+100./a; 524 lastPAR[20]=9.+100./a; // p6 499 // The main quadratic (peh_s 525 // The main quadratic (peh_sh) 500 lastPAR[21]=.002*a3+3.e7/a6; 526 lastPAR[21]=.002*a3+3.e7/a6; // p1 501 lastPAR[22]=7.e-15*a4*asa; 527 lastPAR[22]=7.e-15*a4*asa; // p2 502 lastPAR[23]=9000./a4; 528 lastPAR[23]=9000./a4; // p3 503 // The 1st max pre-exponent (peh_q 529 // The 1st max pre-exponent (peh_qq) 504 lastPAR[24]=.0011*asa/(1.+3.e34/a32/ 530 lastPAR[24]=.0011*asa/(1.+3.e34/a32/a4); // p1 505 lastPAR[25]=1.e-5*a2+2.e14/a16; 531 lastPAR[25]=1.e-5*a2+2.e14/a16; // p2 506 lastPAR[26]=1.2e-11*a2/(1.+1.5e19/a1 532 lastPAR[26]=1.2e-11*a2/(1.+1.5e19/a12); // p3 507 lastPAR[27]=.016*asa/(1.+5.e16/a16); 533 lastPAR[27]=.016*asa/(1.+5.e16/a16); // p4 508 // The 1st max slope (peh_q 534 // The 1st max slope (peh_qs) 509 lastPAR[28]=.002*a4/(1.+7.e7/G4Pow:: << 535 lastPAR[28]=.002*a4/(1.+7.e7/std::pow(a-6.83,14)); // p1 510 lastPAR[29]=2.e6/a6+7.2/G4Pow::GetIn << 536 lastPAR[29]=2.e6/a6+7.2/std::pow(a,.11); // p2 511 lastPAR[30]=11.*a3/(1.+7.e23/a16/a8) 537 lastPAR[30]=11.*a3/(1.+7.e23/a16/a8); // p3 512 lastPAR[31]=100./asa; 538 lastPAR[31]=100./asa; // p4 513 // The 2nd max pre-exponent (peh_s 539 // The 2nd max pre-exponent (peh_ss) 514 lastPAR[32]=(.1+4.4e-5*a2)/(1.+5.e5/ 540 lastPAR[32]=(.1+4.4e-5*a2)/(1.+5.e5/a4); // p1 515 lastPAR[33]=3.5e-4*a2/(1.+1.e8/a8); 541 lastPAR[33]=3.5e-4*a2/(1.+1.e8/a8); // p2 516 lastPAR[34]=1.3+3.e5/a4; 542 lastPAR[34]=1.3+3.e5/a4; // p3 517 lastPAR[35]=500./(a2+50.)+3; 543 lastPAR[35]=500./(a2+50.)+3; // p4 518 lastPAR[36]=1.e-9/a+s4a16*s4a16; 544 lastPAR[36]=1.e-9/a+s4a16*s4a16; // p5 519 // The 2nd max slope (peh_s 545 // The 2nd max slope (peh_sb) 520 lastPAR[37]=.4*asa+3.e-9*a6; 546 lastPAR[37]=.4*asa+3.e-9*a6; // p1 521 lastPAR[38]=.0005*a5; 547 lastPAR[38]=.0005*a5; // p2 522 lastPAR[39]=.002*a5; 548 lastPAR[39]=.002*a5; // p3 523 lastPAR[40]=10.; 549 lastPAR[40]=10.; // p4 524 // The effective pre-exponent (peh_u 550 // The effective pre-exponent (peh_us) 525 lastPAR[41]=.05+.005*a; 551 lastPAR[41]=.05+.005*a; // p1 526 lastPAR[42]=7.e-8/sa; 552 lastPAR[42]=7.e-8/sa; // p2 527 lastPAR[43]=.8*sa; 553 lastPAR[43]=.8*sa; // p3 528 lastPAR[44]=.02*sa; 554 lastPAR[44]=.02*sa; // p4 529 lastPAR[45]=1.e8/a3; 555 lastPAR[45]=1.e8/a3; // p5 530 lastPAR[46]=3.e32/(a32+1.e32); 556 lastPAR[46]=3.e32/(a32+1.e32); // p6 531 // The effective slope (peh_u 557 // The effective slope (peh_ub) 532 lastPAR[47]=24.; 558 lastPAR[47]=24.; // p1 533 lastPAR[48]=20./sa; 559 lastPAR[48]=20./sa; // p2 534 lastPAR[49]=7.e3*a/(sa+1.); 560 lastPAR[49]=7.e3*a/(sa+1.); // p3 535 lastPAR[50]=900.*sa/(1.+500./a3); 561 lastPAR[50]=900.*sa/(1.+500./a3); // p4 536 } 562 } 537 // Parameter for lowEnergyNeutrons 563 // Parameter for lowEnergyNeutrons 538 lastPAR[51]=1.e15+2.e27/a4/(1.+2.e-18* 564 lastPAR[51]=1.e15+2.e27/a4/(1.+2.e-18*a16); 539 } 565 } 540 lastPAR[nLast]=pwd; 566 lastPAR[nLast]=pwd; 541 // and initialize the zero element of th 567 // and initialize the zero element of the table 542 G4double lp=lPMin; 568 G4double lp=lPMin; // ln(momentum) 543 G4bool memCS=onlyCS; 569 G4bool memCS=onlyCS; // ?? 544 onlyCS=false; 570 onlyCS=false; 545 lastCST[0]=GetTabValues(lp, PDG, tgZ, tg 571 lastCST[0]=GetTabValues(lp, PDG, tgZ, tgN); // Calculate AMDB tables 546 onlyCS=memCS; 572 onlyCS=memCS; 547 lastSST[0]=theSS; 573 lastSST[0]=theSS; 548 lastS1T[0]=theS1; 574 lastS1T[0]=theS1; 549 lastB1T[0]=theB1; 575 lastB1T[0]=theB1; 550 lastS2T[0]=theS2; 576 lastS2T[0]=theS2; 551 lastB2T[0]=theB2; 577 lastB2T[0]=theB2; 552 lastS3T[0]=theS3; 578 lastS3T[0]=theS3; 553 lastB3T[0]=theB3; 579 lastB3T[0]=theB3; 554 lastS4T[0]=theS4; 580 lastS4T[0]=theS4; 555 lastB4T[0]=theB4; 581 lastB4T[0]=theB4; 556 } 582 } 557 if(LP>ILP) 583 if(LP>ILP) 558 { 584 { 559 G4int ini = static_cast<int>((ILP-lPMin+ 585 G4int ini = static_cast<int>((ILP-lPMin+.000001)/dlnP)+1; // already inited till this 560 if(ini<0) ini=0; 586 if(ini<0) ini=0; 561 if(ini<nPoints) 587 if(ini<nPoints) 562 { 588 { 563 G4int fin = static_cast<int>((LP-lPMin 589 G4int fin = static_cast<int>((LP-lPMin)/dlnP)+1; // final bin of initialization 564 if(fin>=nPoints) fin=nLast; 590 if(fin>=nPoints) fin=nLast; // Limit of the tabular initialization 565 if(fin>=ini) 591 if(fin>=ini) 566 { 592 { 567 G4double lp=0.; 593 G4double lp=0.; 568 for(G4int ip=ini; ip<=fin; ip++) 594 for(G4int ip=ini; ip<=fin; ip++) // Calculate tabular CS,S1,B1,S2,B2,S3,B3 569 { 595 { 570 lp=lPMin+ip*dlnP; 596 lp=lPMin+ip*dlnP; // ln(momentum) 571 G4bool memCS=onlyCS; 597 G4bool memCS=onlyCS; 572 onlyCS=false; 598 onlyCS=false; 573 lastCST[ip]=GetTabValues(lp, PDG, 599 lastCST[ip]=GetTabValues(lp, PDG, tgZ, tgN); // Calculate AMDB tables (ret CS) 574 onlyCS=memCS; 600 onlyCS=memCS; 575 lastSST[ip]=theSS; 601 lastSST[ip]=theSS; 576 lastS1T[ip]=theS1; 602 lastS1T[ip]=theS1; 577 lastB1T[ip]=theB1; 603 lastB1T[ip]=theB1; 578 lastS2T[ip]=theS2; 604 lastS2T[ip]=theS2; 579 lastB2T[ip]=theB2; 605 lastB2T[ip]=theB2; 580 lastS3T[ip]=theS3; 606 lastS3T[ip]=theS3; 581 lastB3T[ip]=theB3; 607 lastB3T[ip]=theB3; 582 lastS4T[ip]=theS4; 608 lastS4T[ip]=theS4; 583 lastB4T[ip]=theB4; 609 lastB4T[ip]=theB4; 584 } 610 } 585 return lp; 611 return lp; 586 } 612 } 587 else G4cout<<"*Warning*G4ChipsHyperonE 613 else G4cout<<"*Warning*G4ChipsHyperonElasticXS::GetPTables: PDG="<<PDG 588 <<", Z="<<tgZ<<", N="<<tgN< 614 <<", Z="<<tgZ<<", N="<<tgN<<", i="<<ini<<" > fin="<<fin<<", LP="<<LP 589 <<" > ILP="<<ILP<<" nothing 615 <<" > ILP="<<ILP<<" nothing is done!"<<G4endl; 590 } 616 } 591 else G4cout<<"*Warning*G4ChipsHyperonEla 617 else G4cout<<"*Warning*G4ChipsHyperonElasticXS::GetPTables: PDG="<<PDG 592 <<", Z="<<tgZ<<", N="<<tgN<<" 618 <<", Z="<<tgZ<<", N="<<tgN<<", i="<<ini<<">= max="<<nPoints<<", LP="<<LP 593 <<" > ILP="<<ILP<<", lPMax="< 619 <<" > ILP="<<ILP<<", lPMax="<<lPMax<<" nothing is done!"<<G4endl; 594 } 620 } 595 } else { 621 } else { 596 // G4cout<<"*Error*G4ChipsHyperonElasticXS 622 // G4cout<<"*Error*G4ChipsHyperonElasticXS::GetPTables: PDG="<<PDG<<", Z="<<tgZ 597 // <<", N="<<tgN<<", while it is def 623 // <<", N="<<tgN<<", while it is defined only for Hyperons"<<G4endl; 598 // throw G4QException("G4ChipsHyperonElast 624 // throw G4QException("G4ChipsHyperonElasticXS::GetPTables:onlyaBA implemented"); 599 G4ExceptionDescription ed; 625 G4ExceptionDescription ed; 600 ed << "PDG = " << PDG << ", Z = " << tgZ < 626 ed << "PDG = " << PDG << ", Z = " << tgZ << ", N = " << tgN 601 << ", while it is defined only for Hype 627 << ", while it is defined only for Hyperons" << G4endl; 602 G4Exception("G4ChipsHyperonElasticXS::GetP 628 G4Exception("G4ChipsHyperonElasticXS::GetPTables()", "HAD_CHPS_0000", 603 FatalException, ed); 629 FatalException, ed); 604 } 630 } 605 return ILP; 631 return ILP; 606 } 632 } 607 633 608 // Returns Q2=-t in independent units (MeV^2) 634 // Returns Q2=-t in independent units (MeV^2) (all internal calculations are in GeV) 609 G4double G4ChipsHyperonElasticXS::GetExchangeT 635 G4double G4ChipsHyperonElasticXS::GetExchangeT(G4int tgZ, G4int tgN, G4int PDG) 610 { 636 { 611 static const G4double GeVSQ=gigaelectronvolt 637 static const G4double GeVSQ=gigaelectronvolt*gigaelectronvolt; 612 static const G4double third=1./3.; 638 static const G4double third=1./3.; 613 static const G4double fifth=1./5.; 639 static const G4double fifth=1./5.; 614 static const G4double sevth=1./7.; << 640 static const G4double sevth=1./7.; 615 //AR-04Jun2014 if(PDG==3222 || PDG<3000 || << 641 if(PDG==3222 || PDG<3000 || PDG>3334)G4cout<<"*Warning*G4QHyElCS::GET:PDG="<<PDG<<G4endl; 616 if(PDG<3000 || PDG>3334)G4cout<<"*Warning*G4 << 617 if(onlyCS)G4cout<<"*Warning*G4ChipsHyperonEl 642 if(onlyCS)G4cout<<"*Warning*G4ChipsHyperonElasticXS::GetExchanT: onlyCS=1"<<G4endl; 618 if(lastLP<-4.3) return lastTM*GeVSQ*G4Unifor 643 if(lastLP<-4.3) return lastTM*GeVSQ*G4UniformRand();// S-wave for p<14 MeV/c (kinE<.1MeV) 619 G4double q2=0.; 644 G4double q2=0.; 620 if(tgZ==1 && tgN==0) // ===> 645 if(tgZ==1 && tgN==0) // ===> p+p=p+p 621 { 646 { 622 G4double E1=lastTM*theB1; 647 G4double E1=lastTM*theB1; 623 G4double R1=(1.-G4Exp(-E1)); << 648 G4double R1=(1.-std::exp(-E1)); 624 G4double E2=lastTM*theB2; 649 G4double E2=lastTM*theB2; 625 G4double R2=(1.-G4Exp(-E2*E2*E2)); << 650 G4double R2=(1.-std::exp(-E2*E2*E2)); 626 G4double E3=lastTM*theB3; 651 G4double E3=lastTM*theB3; 627 G4double R3=(1.-G4Exp(-E3)); << 652 G4double R3=(1.-std::exp(-E3)); 628 G4double I1=R1*theS1/theB1; 653 G4double I1=R1*theS1/theB1; 629 G4double I2=R2*theS2; 654 G4double I2=R2*theS2; 630 G4double I3=R3*theS3; 655 G4double I3=R3*theS3; 631 G4double I12=I1+I2; 656 G4double I12=I1+I2; 632 G4double rand=(I12+I3)*G4UniformRand(); 657 G4double rand=(I12+I3)*G4UniformRand(); 633 if (rand<I1 ) 658 if (rand<I1 ) 634 { 659 { 635 G4double ran=R1*G4UniformRand(); 660 G4double ran=R1*G4UniformRand(); 636 if(ran>1.) ran=1.; 661 if(ran>1.) ran=1.; 637 q2=-G4Log(1.-ran)/theB1; << 662 q2=-std::log(1.-ran)/theB1; 638 } 663 } 639 else if(rand<I12) 664 else if(rand<I12) 640 { 665 { 641 G4double ran=R2*G4UniformRand(); 666 G4double ran=R2*G4UniformRand(); 642 if(ran>1.) ran=1.; 667 if(ran>1.) ran=1.; 643 q2=-G4Log(1.-ran); << 668 q2=-std::log(1.-ran); 644 if(q2<0.) q2=0.; 669 if(q2<0.) q2=0.; 645 q2=G4Pow::GetInstance()->powA(q2,third)/ << 670 q2=std::pow(q2,third)/theB2; 646 } 671 } 647 else 672 else 648 { 673 { 649 G4double ran=R3*G4UniformRand(); 674 G4double ran=R3*G4UniformRand(); 650 if(ran>1.) ran=1.; 675 if(ran>1.) ran=1.; 651 q2=-G4Log(1.-ran)/theB3; << 676 q2=-std::log(1.-ran)/theB3; 652 } 677 } 653 } 678 } 654 else 679 else 655 { 680 { 656 G4double a=tgZ+tgN; 681 G4double a=tgZ+tgN; 657 G4double E1=lastTM*(theB1+lastTM*theSS); 682 G4double E1=lastTM*(theB1+lastTM*theSS); 658 G4double R1=(1.-G4Exp(-E1)); << 683 G4double R1=(1.-std::exp(-E1)); 659 G4double tss=theSS+theSS; // for future so 684 G4double tss=theSS+theSS; // for future solution of quadratic equation (imediate check) 660 G4double tm2=lastTM*lastTM; 685 G4double tm2=lastTM*lastTM; 661 G4double E2=lastTM*tm2*theB2; 686 G4double E2=lastTM*tm2*theB2; // power 3 for lowA, 5 for HighA (1st) 662 if(a>6.5)E2*=tm2; 687 if(a>6.5)E2*=tm2; // for heavy nuclei 663 G4double R2=(1.-G4Exp(-E2)); << 688 G4double R2=(1.-std::exp(-E2)); 664 G4double E3=lastTM*theB3; 689 G4double E3=lastTM*theB3; 665 if(a>6.5)E3*=tm2*tm2*tm2; 690 if(a>6.5)E3*=tm2*tm2*tm2; // power 1 for lowA, 7 (2nd) for HighA 666 G4double R3=(1.-G4Exp(-E3)); << 691 G4double R3=(1.-std::exp(-E3)); 667 G4double E4=lastTM*theB4; 692 G4double E4=lastTM*theB4; 668 G4double R4=(1.-G4Exp(-E4)); << 693 G4double R4=(1.-std::exp(-E4)); 669 G4double I1=R1*theS1; 694 G4double I1=R1*theS1; 670 G4double I2=R2*theS2; 695 G4double I2=R2*theS2; 671 G4double I3=R3*theS3; 696 G4double I3=R3*theS3; 672 G4double I4=R4*theS4; 697 G4double I4=R4*theS4; 673 G4double I12=I1+I2; 698 G4double I12=I1+I2; 674 G4double I13=I12+I3; 699 G4double I13=I12+I3; 675 G4double rand=(I13+I4)*G4UniformRand(); 700 G4double rand=(I13+I4)*G4UniformRand(); 676 if(rand<I1) 701 if(rand<I1) 677 { 702 { 678 G4double ran=R1*G4UniformRand(); 703 G4double ran=R1*G4UniformRand(); 679 if(ran>1.) ran=1.; 704 if(ran>1.) ran=1.; 680 q2=-G4Log(1.-ran)/theB1; << 705 q2=-std::log(1.-ran)/theB1; 681 if(std::fabs(tss)>1.e-7) q2=(std::sqrt(t 706 if(std::fabs(tss)>1.e-7) q2=(std::sqrt(theB1*(theB1+(tss+tss)*q2))-theB1)/tss; 682 } 707 } 683 else if(rand<I12) 708 else if(rand<I12) 684 { 709 { 685 G4double ran=R2*G4UniformRand(); 710 G4double ran=R2*G4UniformRand(); 686 if(ran>1.) ran=1.; 711 if(ran>1.) ran=1.; 687 q2=-G4Log(1.-ran)/theB2; << 712 q2=-std::log(1.-ran)/theB2; 688 if(q2<0.) q2=0.; 713 if(q2<0.) q2=0.; 689 if(a<6.5) q2=G4Pow::GetInstance()->powA( << 714 if(a<6.5) q2=std::pow(q2,third); 690 else q2=G4Pow::GetInstance()->powA( << 715 else q2=std::pow(q2,fifth); 691 } 716 } 692 else if(rand<I13) 717 else if(rand<I13) 693 { 718 { 694 G4double ran=R3*G4UniformRand(); 719 G4double ran=R3*G4UniformRand(); 695 if(ran>1.) ran=1.; 720 if(ran>1.) ran=1.; 696 q2=-G4Log(1.-ran)/theB3; << 721 q2=-std::log(1.-ran)/theB3; 697 if(q2<0.) q2=0.; 722 if(q2<0.) q2=0.; 698 if(a>6.5) q2=G4Pow::GetInstance()->powA( << 723 if(a>6.5) q2=std::pow(q2,sevth); 699 } 724 } 700 else 725 else 701 { 726 { 702 G4double ran=R4*G4UniformRand(); 727 G4double ran=R4*G4UniformRand(); 703 if(ran>1.) ran=1.; 728 if(ran>1.) ran=1.; 704 q2=-G4Log(1.-ran)/theB4; << 729 q2=-std::log(1.-ran)/theB4; 705 if(a<6.5) q2=lastTM-q2; 730 if(a<6.5) q2=lastTM-q2; // u reduced for lightA (starts from 0) 706 } 731 } 707 } 732 } 708 if(q2<0.) q2=0.; 733 if(q2<0.) q2=0.; 709 if(!(q2>=-1.||q2<=1.))G4cout<<"*NAN*G4QHyEla 734 if(!(q2>=-1.||q2<=1.))G4cout<<"*NAN*G4QHyElasticCrossSect::GetExchangeT:-t="<<q2<<G4endl; 710 if(q2>lastTM) 735 if(q2>lastTM) 711 { 736 { 712 q2=lastTM; 737 q2=lastTM; 713 } 738 } 714 return q2*GeVSQ; 739 return q2*GeVSQ; 715 } 740 } 716 741 717 // Returns B in independent units (MeV^-2) (al 742 // Returns B in independent units (MeV^-2) (all internal calculations are in GeV) see ExT 718 G4double G4ChipsHyperonElasticXS::GetSlope(G4i 743 G4double G4ChipsHyperonElasticXS::GetSlope(G4int tgZ, G4int tgN, G4int PDG) 719 { 744 { 720 static const G4double GeVSQ=gigaelectronvolt 745 static const G4double GeVSQ=gigaelectronvolt*gigaelectronvolt; 721 if(onlyCS)G4cout<<"*Warning*G4ChipsHyperonEl 746 if(onlyCS)G4cout<<"*Warning*G4ChipsHyperonElasticXS::GetSlope: onlCS=true"<<G4endl; 722 if(lastLP<-4.3) return 0.; // S-wav 747 if(lastLP<-4.3) return 0.; // S-wave for p<14 MeV/c (kinE<.1MeV) 723 //AR-04Jun2014 if(PDG==3222 || PDG<3000 || << 748 if(PDG==3222 || PDG<3000 || PDG>3334) 724 if(PDG<3000 || PDG>3334) << 725 { 749 { 726 // G4cout<<"*Error*G4ChipsHyperonElasticXS 750 // G4cout<<"*Error*G4ChipsHyperonElasticXS::GetSlope: PDG="<<PDG<<", Z="<<tgZ 727 // <<", N="<<tgN<<", while it is def 751 // <<", N="<<tgN<<", while it is defined only for Hyperons"<<G4endl; 728 // throw G4QException("G4ChipsHyperonElast 752 // throw G4QException("G4ChipsHyperonElasticXS::GetSlope: HypA are implemented"); 729 G4ExceptionDescription ed; 753 G4ExceptionDescription ed; 730 ed << "PDG = " << PDG << ", Z = " << tgZ < 754 ed << "PDG = " << PDG << ", Z = " << tgZ << ", N = " << tgN 731 << ", while it is defined only for Hype 755 << ", while it is defined only for Hyperons" << G4endl; 732 G4Exception("G4ChipsHyperonElasticXS::GetS 756 G4Exception("G4ChipsHyperonElasticXS::GetSlope()", "HAD_CHPS_0000", 733 FatalException, ed); 757 FatalException, ed); 734 } 758 } 735 if(theB1<0.) theB1=0.; 759 if(theB1<0.) theB1=0.; 736 if(!(theB1>=-1.||theB1<=1.)) G4cout<<"*NAN*G 760 if(!(theB1>=-1.||theB1<=1.)) G4cout<<"*NAN*G4QHyElasticCrossS::Getslope:"<<theB1<<G4endl; 737 return theB1/GeVSQ; 761 return theB1/GeVSQ; 738 } 762 } 739 763 740 // Returns half max(Q2=-t) in independent unit 764 // Returns half max(Q2=-t) in independent units (MeV^2) 741 G4double G4ChipsHyperonElasticXS::GetHMaxT() 765 G4double G4ChipsHyperonElasticXS::GetHMaxT() 742 { 766 { 743 static const G4double HGeVSQ=gigaelectronvol 767 static const G4double HGeVSQ=gigaelectronvolt*gigaelectronvolt/2.; 744 return lastTM*HGeVSQ; 768 return lastTM*HGeVSQ; 745 } 769 } 746 770 747 // lastLP is used, so calculating tables, one 771 // lastLP is used, so calculating tables, one need to remember and then recover lastLP 748 G4double G4ChipsHyperonElasticXS::GetTabValues 772 G4double G4ChipsHyperonElasticXS::GetTabValues(G4double lp, G4int PDG, G4int tgZ, 749 773 G4int tgN) 750 { 774 { 751 //AR-04Jun2014 if(PDG==3222 || PDG<3000 || << 775 if(PDG==3222 || PDG<3000 || PDG>3334) G4cout<<"*Warning*G4QHypElCS::GTV:P="<<PDG<<G4endl; 752 if(PDG<3000 || PDG>3334) G4cout<<"*Warning*G << 776 if(tgZ<0 || tgZ>92) 753 << 754 << 755 //AR-24Apr2018 Switch to allow transuranic e << 756 const G4bool isHeavyElementAllowed = true; << 757 if(tgZ<0 || ( !isHeavyElementAllowed && tgZ> << 758 { 777 { 759 G4cout<<"*Warning*G4QHyperonElastCS::GetTa 778 G4cout<<"*Warning*G4QHyperonElastCS::GetTabValue:(1-92) NoIsotopesFor Z="<<tgZ<<G4endl; 760 return 0.; 779 return 0.; 761 } 780 } 762 G4int iZ=tgZ-1; // Z index 781 G4int iZ=tgZ-1; // Z index 763 if(iZ<0) 782 if(iZ<0) 764 { 783 { 765 iZ=0; // conversion of the neutron 784 iZ=0; // conversion of the neutron target to the proton target 766 tgZ=1; 785 tgZ=1; 767 tgN=0; 786 tgN=0; 768 } 787 } 769 G4double p=G4Exp(lp); // moment << 788 G4double p=std::exp(lp); // momentum 770 G4double sp=std::sqrt(p); // sqr 789 G4double sp=std::sqrt(p); // sqrt(p) 771 G4double p2=p*p; 790 G4double p2=p*p; 772 G4double p3=p2*p; 791 G4double p3=p2*p; 773 G4double p4=p3*p; 792 G4double p4=p3*p; 774 if ( tgZ == 1 && tgN == 0 ) // Hyperon+P 793 if ( tgZ == 1 && tgN == 0 ) // Hyperon+P 775 { 794 { 776 G4double dl2=lp-lastPAR[9]; 795 G4double dl2=lp-lastPAR[9]; 777 theSS=lastPAR[32]; 796 theSS=lastPAR[32]; 778 theS1=(lastPAR[10]+lastPAR[11]*dl2*dl2)/(1 797 theS1=(lastPAR[10]+lastPAR[11]*dl2*dl2)/(1.+lastPAR[12]/p4/p)+ 779 (lastPAR[13]/p2+lastPAR[14]*p)/(p4+l 798 (lastPAR[13]/p2+lastPAR[14]*p)/(p4+lastPAR[15]*sp); 780 theB1=lastPAR[16]*G4Pow::GetInstance()->po << 799 theB1=lastPAR[16]*std::pow(p,lastPAR[17])/(1.+lastPAR[18]/p3); 781 theS2=lastPAR[19]+lastPAR[20]/(p4+lastPAR[ 800 theS2=lastPAR[19]+lastPAR[20]/(p4+lastPAR[21]*p); 782 theB2=lastPAR[22]+lastPAR[23]/(p4+lastPAR[ 801 theB2=lastPAR[22]+lastPAR[23]/(p4+lastPAR[24]/sp); 783 theS3=lastPAR[25]+lastPAR[26]/(p4*p4+lastP 802 theS3=lastPAR[25]+lastPAR[26]/(p4*p4+lastPAR[27]*p2+lastPAR[28]); 784 theB3=lastPAR[29]+lastPAR[30]/(p4+lastPAR[ 803 theB3=lastPAR[29]+lastPAR[30]/(p4+lastPAR[31]); 785 theS4=0.; 804 theS4=0.; 786 theB4=0.; 805 theB4=0.; 787 // Returns the total elastic pim-p cross-s 806 // Returns the total elastic pim-p cross-section (to avoid spoiling lastSIG) 788 G4double dp=lp-lastPAR[4]; 807 G4double dp=lp-lastPAR[4]; 789 return lastPAR[0]/(lastPAR[1]+p2*(lastPAR[ 808 return lastPAR[0]/(lastPAR[1]+p2*(lastPAR[2]+p2))+(lastPAR[3]*dp*dp+lastPAR[5]+ 790 lastPAR[6]/p2)/(1.+lastPAR[7]/sp+la 809 lastPAR[6]/p2)/(1.+lastPAR[7]/sp+lastPAR[8]/p4); 791 } 810 } 792 else 811 else 793 { 812 { 794 G4double p5=p4*p; 813 G4double p5=p4*p; 795 G4double p6=p5*p; 814 G4double p6=p5*p; 796 G4double p8=p6*p2; 815 G4double p8=p6*p2; 797 G4double p10=p8*p2; 816 G4double p10=p8*p2; 798 G4double p12=p10*p2; 817 G4double p12=p10*p2; 799 G4double p16=p8*p8; 818 G4double p16=p8*p8; 800 //G4double p24=p16*p8; 819 //G4double p24=p16*p8; 801 G4double dl=lp-5.; 820 G4double dl=lp-5.; 802 G4double a=tgZ+tgN; 821 G4double a=tgZ+tgN; 803 G4double pah=G4Pow::GetInstance()->powA(p, << 822 G4double pah=std::pow(p,a/2); 804 G4double pa=pah*pah; 823 G4double pa=pah*pah; 805 G4double pa2=pa*pa; 824 G4double pa2=pa*pa; 806 if(a<6.5) 825 if(a<6.5) 807 { 826 { 808 theS1=lastPAR[9]/(1.+lastPAR[10]*p4*pa)+ 827 theS1=lastPAR[9]/(1.+lastPAR[10]*p4*pa)+lastPAR[11]/(p4+lastPAR[12]*p4/pa2)+ 809 (lastPAR[13]*dl*dl+lastPAR[14])/(1 828 (lastPAR[13]*dl*dl+lastPAR[14])/(1.+lastPAR[15]/p2); 810 theB1=(lastPAR[16]+lastPAR[17]*p2)/(p4+l 829 theB1=(lastPAR[16]+lastPAR[17]*p2)/(p4+lastPAR[18]/pah)+lastPAR[19]; 811 theSS=lastPAR[20]/(1.+lastPAR[21]/p2)+la 830 theSS=lastPAR[20]/(1.+lastPAR[21]/p2)+lastPAR[22]/(p6/pa+lastPAR[23]/p16); 812 theS2=lastPAR[24]/(pa/p2+lastPAR[25]/p4) 831 theS2=lastPAR[24]/(pa/p2+lastPAR[25]/p4)+lastPAR[26]; 813 theB2=lastPAR[27]*G4Pow::GetInstance()-> << 832 theB2=lastPAR[27]*std::pow(p,lastPAR[28])+lastPAR[29]/(p8+lastPAR[30]/p16); 814 theS3=lastPAR[31]/(pa*p+lastPAR[32]/pa)+ 833 theS3=lastPAR[31]/(pa*p+lastPAR[32]/pa)+lastPAR[33]; 815 theB3=lastPAR[34]/(p3+lastPAR[35]/p6)+la 834 theB3=lastPAR[34]/(p3+lastPAR[35]/p6)+lastPAR[36]/(1.+lastPAR[37]/p2); 816 theS4=p2*(pah*lastPAR[38]*G4Exp(-pah*las << 835 theS4=p2*(pah*lastPAR[38]*std::exp(-pah*lastPAR[39])+ 817 lastPAR[40]/(1.+lastPAR[41]*G4 << 836 lastPAR[40]/(1.+lastPAR[41]*std::pow(p,lastPAR[42]))); 818 theB4=lastPAR[43]*pa/p2/(1.+pa*lastPAR[4 837 theB4=lastPAR[43]*pa/p2/(1.+pa*lastPAR[44]); 819 } 838 } 820 else 839 else 821 { 840 { 822 theS1=lastPAR[9]/(1.+lastPAR[10]/p4)+las 841 theS1=lastPAR[9]/(1.+lastPAR[10]/p4)+lastPAR[11]/(p4+lastPAR[12]/p2)+ 823 lastPAR[13]/(p5+lastPAR[14]/p16); 842 lastPAR[13]/(p5+lastPAR[14]/p16); 824 theB1=(lastPAR[15]/p8+lastPAR[19])/(p+la << 843 theB1=(lastPAR[15]/p8+lastPAR[19])/(p+lastPAR[16]/std::pow(p,lastPAR[20]))+ 825 lastPAR[17]/(1.+lastPAR[18]/p4); 844 lastPAR[17]/(1.+lastPAR[18]/p4); 826 theSS=lastPAR[21]/(p4/G4Pow::GetInstance << 845 theSS=lastPAR[21]/(p4/std::pow(p,lastPAR[23])+lastPAR[22]/p4); 827 theS2=lastPAR[24]/p4/(G4Pow::GetInstance << 846 theS2=lastPAR[24]/p4/(std::pow(p,lastPAR[25])+lastPAR[26]/p12)+lastPAR[27]; 828 theB2=lastPAR[28]/G4Pow::GetInstance()-> << 847 theB2=lastPAR[28]/std::pow(p,lastPAR[29])+lastPAR[30]/std::pow(p,lastPAR[31]); 829 theS3=lastPAR[32]/G4Pow::GetInstance()-> << 848 theS3=lastPAR[32]/std::pow(p,lastPAR[35])/(1.+lastPAR[36]/p12)+ 830 lastPAR[33]/(1.+lastPAR[34]/p6); 849 lastPAR[33]/(1.+lastPAR[34]/p6); 831 theB3=lastPAR[37]/p8+lastPAR[38]/p2+last 850 theB3=lastPAR[37]/p8+lastPAR[38]/p2+lastPAR[39]/(1.+lastPAR[40]/p8); 832 theS4=(lastPAR[41]/p4+lastPAR[46]/p)/(1. 851 theS4=(lastPAR[41]/p4+lastPAR[46]/p)/(1.+lastPAR[42]/p10)+ 833 (lastPAR[43]+lastPAR[44]*dl*dl)/(1 852 (lastPAR[43]+lastPAR[44]*dl*dl)/(1.+lastPAR[45]/p12); 834 theB4=lastPAR[47]/(1.+lastPAR[48]/p)+las 853 theB4=lastPAR[47]/(1.+lastPAR[48]/p)+lastPAR[49]*p4/(1.+lastPAR[50]*p5); 835 } 854 } 836 // Returns the total elastic (n/p)A cross- 855 // Returns the total elastic (n/p)A cross-section (to avoid spoiling lastSIG) 837 G4double dlp=lp-lastPAR[5]; // ax 856 G4double dlp=lp-lastPAR[5]; // ax 838 // p1 p2 p3 857 // p1 p2 p3 p4 p5 839 return (lastPAR[0]*dlp*dlp+lastPAR[1])/(1. 858 return (lastPAR[0]*dlp*dlp+lastPAR[1])/(1.+lastPAR[2]/p)+lastPAR[3]/(p3+lastPAR[4]); 840 } 859 } 841 return 0.; 860 return 0.; 842 } // End of GetTableValues 861 } // End of GetTableValues 843 862 844 // Returns max -t=Q2 (GeV^2) for the momentum 863 // Returns max -t=Q2 (GeV^2) for the momentum pP(GeV) and the target nucleus (tgN,tgZ) 845 G4double G4ChipsHyperonElasticXS::GetQ2max(G4i 864 G4double G4ChipsHyperonElasticXS::GetQ2max(G4int PDG, G4int tgZ, G4int tgN, 846 865 G4double pP) 847 { 866 { 848 static const G4double mLamb= G4Lambda::Lambd 867 static const G4double mLamb= G4Lambda::Lambda()->GetPDGMass()*.001; // MeV to GeV 849 static const G4double mLa2= mLamb*mLamb; 868 static const G4double mLa2= mLamb*mLamb; 850 G4double pP2=pP*pP; 869 G4double pP2=pP*pP; // squared momentum of the projectile 851 if(tgZ || tgN>-1) 870 if(tgZ || tgN>-1) // --> Hyperon-A 852 { 871 { 853 G4double mt=G4ParticleTable::GetParticleTa 872 G4double mt=G4ParticleTable::GetParticleTable()->GetIonTable()->GetIon(tgZ,tgZ+tgN,0)->GetPDGMass()*.001; // Target mass in GeV 854 873 855 G4double dmt=mt+mt; 874 G4double dmt=mt+mt; 856 G4double mds=dmt*std::sqrt(pP2+mLa2)+mLa2+ 875 G4double mds=dmt*std::sqrt(pP2+mLa2)+mLa2+mt*mt; // Mondelstam mds (@@ other hyperons?) 857 return dmt*dmt*pP2/mds; 876 return dmt*dmt*pP2/mds; 858 } 877 } 859 else 878 else 860 { 879 { 861 // G4cout<<"*Error*G4ChipsHyperonElasticXS 880 // G4cout<<"*Error*G4ChipsHyperonElasticXS::GetQ2ma:PDG="<<PDG<<",Z="<<tgZ<<",N=" 862 // <<tgN<<", while it is defined onl 881 // <<tgN<<", while it is defined only for p projectiles & Z_target>0"<<G4endl; 863 // throw G4QException("G4ChipsHyperonElast 882 // throw G4QException("G4ChipsHyperonElasticXS::GetQ2max: only HyperA implemented"); 864 G4ExceptionDescription ed; 883 G4ExceptionDescription ed; 865 ed << "PDG = " << PDG << ", Z = " << tgZ < 884 ed << "PDG = " << PDG << ", Z = " << tgZ << ", N = " << tgN 866 << ", while it is defined only for p pr 885 << ", while it is defined only for p projectiles & Z_target>0" << G4endl; 867 G4Exception("G4ChipsHyperonElasticXS::GetQ 886 G4Exception("G4ChipsHyperonElasticXS::GetQ2max()", "HAD_CHPS_0000", 868 FatalException, ed); 887 FatalException, ed); 869 return 0; 888 return 0; 870 } 889 } 871 } 890 } 872 891