Page Menu
Home
HEPForge
Search
Configure Global Search
Log In
Files
F7879014
No One
Temporary
Actions
View File
Edit File
Delete File
View Transforms
Subscribe
Mute Notifications
Award Token
Flag For Later
Size
16 KB
Subscribers
None
View Options
diff --git a/vjet/delta.C b/vjet/delta.C
index 6929c90..bc3ce4a 100644
--- a/vjet/delta.C
+++ b/vjet/delta.C
@@ -1,81 +1,81 @@
#include "vjint.h"
#include "codata.h"
#include "phasespace.h"
#include "settings.h"
#include "coupling.h"
#include "luminosity.h"
#include "mesq.h"
#include "resconst.h"
#include <iostream>
double vjint::logx2min;
double vjint::delta(double x)
{
double fac = gevfb/1000.*coupling::aemmz*asp_.asp_*resconst::Cf;
double q2 = phasespace::m2;
//change of variable to allow integration over 0-1
//"esp" introduced for better behavior at small qt
int esp = 8;
double x2 = exp((1.-pow(x,esp))*logx2min);
double xjac = x2*logx2min*esp*pow(x,(esp-1));
//x2=x2min*exp(1d0/x2min*t)
//xjac=xjac*x2(1d0-x2min)
//compute x1 as a function of x2
double x1 = (opts.sroot*phasespace::mt*phasespace::exppy*x2-q2)/(x2*pow(opts.sroot,2)-opts.sroot*phasespace::mt*phasespace::expmy);
//check x1,x2<1
double tiny = 0.;// !1d-8
if (x1 > 1.-tiny || x2 > 1.-tiny)
return 0.;
//Partonic mandelstam invariants (bug fix in DYqT: th <-> uh)
double sh = x1*x2*pow(opts.sroot,2);
double th = q2-opts.sroot*x1*phasespace::mt*phasespace::expmy;
double uh = q2-opts.sroot*x2*phasespace::mt*phasespace::exppy;
//1/xjj is the jacobian of the integration in dx1 of delta(q2-sh-uh-th) dx1
//following from the change of variable t = q2-sh-uh-th
//dx1 = dt * dx1(t)/dt -> 1/xjj = dx1(t)/dt
double xjj = abs(x2*pow(opts.sroot,2)-opts.sroot*phasespace::mt*phasespace::expmy);
//common factor for all the contributions
double factor = fac/sh;
//compute parton luminosity
luminosity::pdf1(x1);
luminosity::pdf2(x2);
luminosity::calc();
//leading order delta(s2) contributions
double xloqg=(factor/(pow(coupling::NC,2)-1.))*(aqg0_(sh,th,uh,q2)+agq0_(sh,th,uh,q2));
double xloqqb=(factor/coupling::NC)*aqqb0_(sh,th,uh,q2);
double xlo=xloqg+xloqqb;
//next to leading order delta(s2) contributions
double xnlo = 0.;
- if (opts.order == 2)
+ if (opts.order >= 2)
{
double s2 = 0.;
utils_fu_(uh,q2);
utils_(sh,th,uh,q2,s2);
utils_dilog_(sh,th,uh,q2);
double xnloqg = factor/(pow(coupling::NC,2)-1.)
*(bqg1_(sh,th,uh,q2)+bgq1_(sh,th,uh,q2)+
bqg2_(sh,th,uh,q2)+bgq2_(sh,th,uh,q2)+
cqg1_(sh,th,uh,q2)+cgq1_(sh,th,uh,q2)+
cqg2_(sh,th,uh,q2)+cgq2_(sh,th,uh,q2)+
bqg3_(sh,th,uh,q2)+bgq3_(sh,th,uh,q2));
double xnloqqb = factor/coupling::NC
*(bqqb1_(sh,th,uh,q2)+bqqb2_(sh,th,uh,q2)+
cqqb1_(sh,th,uh,q2)+d0aa_(sh,th,uh,q2)+
bqqb3_(sh,th,uh,q2));
xnlo = xnloqg+xnloqqb;
}
return -M_PI*xjac/xjj * (xlo+(asp_.asp_/2./M_PI)*xnlo);
}
diff --git a/vjet/vjint.C b/vjet/vjint.C
index fef125f..6f64fc9 100644
--- a/vjet/vjint.C
+++ b/vjet/vjint.C
@@ -1,449 +1,449 @@
#include "vjint.h"
#include "settings.h"
#include "interface.h"
#include "resconst.h"
#include "cubature.h"
#include "gaussrules.h"
#include "mesq.h"
#include "coupling.h"
#include "scales.h"
#include "phasespace.h"
#include "luminosity.h"
#include "LHAPDF/LHAPDF.h"
#include <iomanip>
#include <math.h>
const double vjint::zmin = 1e-13;//1e-8;
const double vjint::zmax = 1.;//1.-1e-10;//1.-1e-8;
const double vjint::lz = log(zmax/zmin);
int vjint::z1rule;
int vjint::z2rule;
double *vjint::t1;
double *vjint::t2;
double vjint::brz;
double vjint::brw;
void vjint::release()
{
delete[] t1;
delete[] t2;
}
void vjint::init()
{
z1rule = opts.zrule;
z2rule = opts.zrule;
t1 = new double [z1rule];
t2 = new double [z2rule];
double az = 0.;
double bz = 1.;
double cz = 0.5*(az+bz);
double mz = 0.5*(bz-az);
for (int j = 0; j < z1rule; j++)
t1[j] = zmin*pow(zmax/zmin, cz+mz*gr::xxx[z1rule-1][j]);
for (int j = 0; j < z2rule; j++)
t2[j] = zmin*pow(zmax/zmin, cz+mz*gr::xxx[z2rule-1][j]);
//gevpb_.gevpb_ = 3.8937966e8; //MCFM 6.8 value
//gevpb_.gevpb_ = 0.389379e9; //dyres value
//gevpb_.gevpb_ = gevfb/1000.;
// flagch_.flagch_ = 0;
// vegint_.iter_ = 20;
// vegint_.locall_ = 10000;
// vegint_.nlocall_ = 20000;
// dypara_.ss_ = opts.sroot;
// dypara_.s_ = pow(dypara_.ss_,2);
// dypdf_.ih1_ = opts.ih1;
// dypdf_.ih2_ = opts.ih2;
//vjorder_.iord_ = opts.order - 1;
// if (opts.nproc == 3)
// {
// if (opts.useGamma && !opts.zerowidth)
// prodflag_.prodflag_ = 5;
// else
// prodflag_.prodflag_ = 3;
// }
// else if (opts.nproc == 1)
// prodflag_.prodflag_ = 21;
// else if (opts.nproc == 2)
// prodflag_.prodflag_ = 22;
ckm_.vud_ = cabib_.Vud_;
ckm_.vus_ = cabib_.Vus_;
ckm_.vub_ = cabib_.Vub_;
ckm_.vcd_ = cabib_.Vcd_;
ckm_.vcs_ = cabib_.Vcs_;
ckm_.vcb_ = cabib_.Vcb_;
ckm_.vtd_ = 0.;
ckm_.vts_ = 0.;
ckm_.vtb_ = 0.;
double cw2 = 1.- coupling::xw;
// em_.aemmz_ = coupling::aemmz; //sqrt(2.)* ewcouple_.Gf_ *pow(coupling::wmass,2)* ewcouple_.xw_ /M_PI;
brz = 1./dymasses_.zwidth_*coupling::aemmz/24.*coupling::zmass *(pow(-1.+2.*coupling::xw,2)+pow(2.*coupling::xw,2))/(coupling::xw*cw2); //=0.033638
brw = 1./dymasses_.wwidth_*coupling::aemmz*coupling::wmass/(12.*coupling::xw); //=0.10906
//quarks are ordered according to mass:
//1,2,3,4,5,6
//u,d,s,c,b,t
//double equ = 2./3.; //up-quarks electric charge
//double eqd = -1./3.; //down-quarks electric charge
//quarks_.eq_[0]=equ;
//quarks_.eq_[1]=eqd;
//quarks_.eq_[2]=eqd;
//quarks_.eq_[3]=equ;
//quarks_.eq_[4]=eqd;
quarks_.eq_[0] = ewcharge_.Q_[MAXNF+2];
quarks_.eq_[1] = ewcharge_.Q_[MAXNF+1];
quarks_.eq_[2] = ewcharge_.Q_[MAXNF+3];
quarks_.eq_[3] = ewcharge_.Q_[MAXNF+4];
quarks_.eq_[4] = ewcharge_.Q_[MAXNF+5];
//definition of 'generalized' ckm matrix:
// (uu ud us uc ub ut)
// (du dd ds dc db dt)
//ckm=(su sd ss sc sb st)
// (cu cd cs cc cb ct)
// (bu bd bs bc bb bt)
// (tu td ts tc tb tt)
for (int i = 0; i < 6; i++)
for (int j = 0; j < 6; j++)
quarks_.ckm_[j][i] = 0.;
quarks_.ckm_[1][0]=ckm_.vud_;
quarks_.ckm_[2][0]=ckm_.vus_;
quarks_.ckm_[4][0]=ckm_.vub_;
quarks_.ckm_[0][1]=ckm_.vud_;
quarks_.ckm_[3][1]=ckm_.vcd_;
quarks_.ckm_[5][1]=ckm_.vtd_;
quarks_.ckm_[0][2]=ckm_.vus_;
quarks_.ckm_[3][2]=ckm_.vcs_;
quarks_.ckm_[5][2]=ckm_.vts_;
quarks_.ckm_[1][3]=ckm_.vcd_;
quarks_.ckm_[2][3]=ckm_.vcs_;
quarks_.ckm_[4][3]=ckm_.vcb_;
quarks_.ckm_[0][4]=ckm_.vub_;
quarks_.ckm_[3][4]=ckm_.vcb_;
quarks_.ckm_[5][4]=ckm_.vtb_;
quarks_.ckm_[1][5]=ckm_.vtd_;
quarks_.ckm_[2][5]=ckm_.vts_;
quarks_.ckm_[4][5]=ckm_.vtb_;
//definition of 'delta' matrix
for (int i = 0; i < MAXNF; i++)
for (int j = 0; j < MAXNF; j++)
quarks_.delta_[j][i] = 0.;
for (int i = 0; i < MAXNF; i++)
quarks_.delta_[i][i] = 1.;
//definition of tau3's Pauli matrix
for (int i = 0; i < MAXNF; i++)
for (int j = 0; j < MAXNF; j++)
quarks_.tau3_[j][i] = 0.;
//quarks_.tau3_[0][0]=1.;
//quarks_.tau3_[1][1]=-1.;
//quarks_.tau3_[2][2]=-1.;
//quarks_.tau3_[3][3]=1.;
//quarks_.tau3_[4][4]=-1.;
quarks_.tau3_[0][0] = ewcharge_.tau_[MAXNF+2];
quarks_.tau3_[1][1] = ewcharge_.tau_[MAXNF+1];
quarks_.tau3_[2][2] = ewcharge_.tau_[MAXNF+3];
quarks_.tau3_[3][3] = ewcharge_.tau_[MAXNF+4];
quarks_.tau3_[4][4] = ewcharge_.tau_[MAXNF+5];
//definition of constants and couplings
// const2_.ca_ = resconst::CA;
// const2_.xnc_ = coupling::NC;
// const2_.cf_ = resconst::Cf;
// const2_.tr_ = resconst::NF/2.;
// const2_.pi_ = M_PI;
dyqcd_.pi_ = M_PI;
dyqcd_.cf_ = resconst::Cf;
dyqcd_.ca_ = resconst::CA;
dyqcd_.tr_ = resconst::NF/2.;
dyqcd_.xnc_ = coupling::NC;
dyqcd_.nf_ = resconst::NF;
// dycouplings_.alpha0_ = coupling::aemmz;
// dycouplings_.xw_ = coupling::xw;
// dycouplings_.sw_ = sqrt(coupling::xw); // sin_w
// dycouplings_.cw_ = sqrt(1.-coupling::xw); // cos_w
luminosity::init();
}
double vjint::vint(double m, double pt, double y)
{
double q2 = m*m;
//set scales and alpha strong
scales::set(m, pt);
scales::vjet();
/*
if (opts.dynamicscale)
{
scales2_.xmur_ = sqrt(pow(opts.kmuren*m,2) + pow(opts.kpt_muren*pt,2));
scales2_.xmuf_ = sqrt(pow(opts.kmufac*m,2) + pow(opts.kpt_mufac*pt,2));
}
else
{
scales2_.xmur_ = sqrt(pow(opts.kmuren*opts.rmass,2) + pow(opts.kpt_muren*pt,2));
scales2_.xmuf_ = sqrt(pow(opts.kmufac*opts.rmass,2) + pow(opts.kpt_mufac*pt,2));
}
scales2_.xmur2_ = pow(scales2_.xmur_,2);
scales2_.xmuf2_ = pow(scales2_.xmuf_,2);
asnew_.as_ = LHAPDF::alphasPDF(scales2_.xmur_)/M_PI;
*/
//calculate logs of scales
utils_scales_(q2);
//calculate propagators
double cw2 = 1.- coupling::xw;
//if (opts.zerowidth)
// {
// sigs_.sigz_ = brz; // /(16.*cw2)
// sigs_.sigw_ = brw; // /4.
// sigs_.siggamma_ = 0.;
// sigs_.sigint_ = 0.;
// }
//else
// {
//sigs_.sigz_ = brz*dymasses_.zwidth_*q2/(M_PI*coupling::zmass) / (pow(q2-pow(coupling::zmass,2),2)+pow(coupling::zmass,2)*pow(dymasses_.zwidth_,2)); // /(16.*cw2)
//sigs_.sigw_ = brw*dymasses_.wwidth_*q2/(M_PI*coupling::wmass) / (pow(q2-pow(coupling::wmass,2),2)+pow(coupling::wmass,2)*pow(dymasses_.wwidth_,2)); // !/4.
//sigs_.siggamma_ = em_.aemmz_/(3.*M_PI*q2);
//sigs_.sigint_ = -em_.aemmz_/(6.*M_PI)*(q2-pow(coupling::zmass,2)) /(pow(q2-pow(coupling::zmass,2),2)+pow(coupling::zmass,2)*pow(dymasses_.zwidth_,2)) * (-1.+4.*coupling::xw)/(2.*sqrt(coupling::xw*cw2)); // !sqrt(sw2/cw2)/2.
mesq::setpropagators(m);
if (opts.nproc == 3)
sigs_.sigz_ = brz*dymasses_.zwidth_/(M_PI*coupling::zmass)*mesq::propZ;
else
sigs_.sigw_ = brw*dymasses_.wwidth_/(M_PI*coupling::wmass)*mesq::propW;
if (opts.useGamma)
{
sigs_.siggamma_ = coupling::aemmz/(3.*M_PI) * mesq::propG;
sigs_.sigint_ = -coupling::aemmz/(6.*M_PI)* mesq::propZG * (-1.+4.*coupling::xw)/(2.*sqrt(coupling::xw*cw2));
}
//}
//set phase space variables
// internal_.q_ = m;
// internal_.q2_ = pow(m,2);
// internal_.qt_ = pt;
// yv_.yv_ = y;
// yv_.expyp_ = phasespace::exppy;//exp(y);
// yv_.expym_ = phasespace::expmy;//exp(-y);//1./yv_.expyp_;
// if (opts.zerowidth)
// internal_.q_ = opts.rmass;
//call cross section calculation
// int ord = opts.order - 1;
// double res, err, chi2a;
// qtdy_(res,err,chi2a,y,y,ord);
// cout << setprecision(16) << "fortran " << m << " " << pt << " " << " " << y << " " << res << endl;
double res = calc(m, pt, y);
//cout << setprecision(16) << "C++ " << m << " " << pt << " " << " " << y << " " << res << endl;
//Apply conversion factors:
//ds/dqt = ds/dqt2 * dqt2/dqt
//fb = 1000 * pb
return 2*pt*res*1000.;
}
int xdelta_cubature(unsigned ndim, const double x[], void *data, unsigned ncomp, double f[])
{
//f[0] = xdelta_(x);
f[0] = vjint::delta(x[0]);
return 0;
}
//int sing_cubature(unsigned ndim, const double x[], void *data, unsigned ncomp, double f[])
//{
// double zz[2];
// zz[0] = vjint::zmin*pow(vjint::zmax/vjint::zmin,x[0]);
// double jacz1 = zz[0]*vjint::lz;
//
// zz[1] = vjint::zmin*pow(vjint::zmax/vjint::zmin,x[1]);
// double jacz2 = zz[1]*vjint::lz;
//
// f[0] = sing_(zz)*jacz1*jacz2;
// return 0;
//}
double vjint::calc(double m, double pt, double y)
{
clock_t begin_time, end_time;
double q2 = phasespace::m2;
// tm_.tm_ = sqrt(q2+pt*pt);
//kinematical limits on qt (at y = 0) --> possibly not needed, since the phace space is generated within kinematical limits
double z = q2/pow(opts.sroot,2);
double xr = pow(1-z,2)-4*z*pow(pt/m,2);
if (xr < 0)
return 0.;
//kinematical limits on y including qt
double tmpx = (q2+pow(opts.sroot,2))/opts.sroot/phasespace::mt;
double ymax = log((tmpx+sqrt(max(0.,pow(tmpx,2)-4.)))/2.);
double ay = fabs(y);
if (fabs(y) > ymax)
return 0.;
if (fabs(*(long*)& ay - *(long*)& ymax) < 2 ) //check also equality
return 0.;
//...compute as at order=nloop
asp_.asp_ = asnew_.as_*M_PI;
luminosity::cache();
//integration of the delta term
//lower limit of integration for x2
double x2min=(q2-opts.sroot*phasespace::mt*phasespace::expmy)/(opts.sroot*phasespace::mt*phasespace::exppy-pow(opts.sroot,2));
logx2min = log(x2min);
if (x2min >= 1. || x2min < 0.)
{
//cout << "error in x2min " << x2min << " m " << phasespace::m << " pt " << phasespace::qt << " y " << phasespace::y << endl;
return 0.;
}
/*
//pcubature integration
const int ndimx = 1; //dimensions of the integral
const int ncomp = 1; //components of the integrand
void *userdata = NULL;
double integral[1];
double error[1];
const int eval = 0;
const double epsrel = min(1e-4,opts.pcubaccuracy);
const double epsabs = 0.;
//boundaries of integration
double xmin[1] = {0.};
double xmax[1] = {1.};
begin_time = clock();
pcubature(ncomp, xdelta_cubature, userdata,
ndimx, xmin, xmax,
eval, epsabs, epsrel, ERROR_INDIVIDUAL, integral, error);
end_time = clock();
double rdelta = integral[0];
double err = error[0];
if (opts.timeprofile)
cout << setw(10) << "xdelta time" << setw(10) << float( end_time - begin_time ) / CLOCKS_PER_SEC << setw(10) << "result" << setw(10) << rdelta
<< endl;
*/
begin_time = clock();
double rdelta = 0.;
int xrule = opts.xrule;
//start integration
double ax = 0.;
double bx = 1.;
double cx = 0.5*(ax+bx);
double mx = 0.5*(bx-ax);
double jac = mx;
for (int i = 0; i < xrule; i++)
{
double x = cx+mx*gr::xxx[xrule-1][i];
rdelta += delta(x) * jac * gr::www[xrule-1][i];
}
end_time = clock();
if (opts.timeprofile)
cout << setw(10) << "xdelta time" << setw(10) << float( end_time - begin_time ) / CLOCKS_PER_SEC << setw(10) << "result" << setw(10) << rdelta
<< endl;
//integration of non-delta(s2) terms (only at NLO)
//initialise
double rsing = 0.;
begin_time = clock();
- if (opts.order == 2)
+ if (opts.order >= 2)
{
rsing = sing();
/*
////pcubature integration
//const int ndims = 2; //dimensions of the integral
//// boundaries of integration
//double zmn[2] = {zmin,zmin};
//double zmx[2] = {zmax,zmax};
//pcubature(ncomp, sing_cubature, userdata,
// ndims, zmn, zmx,
// eval, epsabs, 1e-4, ERROR_INDIVIDUAL, integral, error);
//rsing = integral[0];
//err = sqrt(err*err + error[0]*error[0]);
//// cout << m << " " << pt << " " << y << " " << integral[0] << " " << error[0] << endl;
//gaussian quadrature
double zz[2];
double az1 = zmin;
double bz1 = zmax;
double cz1=0.5*(az1+bz1);
double mz1=0.5*(bz1-az1);
for (int j = 0; j < z1rule; j++)
{
double z1 = cz1 + mz1*gr::xxx[z1rule-1][j];
double t = t1[j];//zmin*pow(zmax/zmin,z1);
double jacz1=t*lz;
zz[0]=t;
double az2 = zmin;
double bz2 = zmax;
double cz2 = 0.5*(az2+bz2);
double mz2 = 0.5*(bz2-az2);
for (int jj = 0; jj < z2rule; jj++)
{
double z2 = cz2 + mz2*gr::xxx[z2rule-1][jj];
double t = t2[jj];//zmin*pow(zmax/zmin,z2);
double jacz2 = t*lz;
zz[1] = t;
rsing=rsing+sing_(zz)*gr::www[z1rule-1][j]*gr::www[z2rule-1][jj]
//rsing=rsing+sing(zz[0],zz[1])*gr::www[z1rule-1][j]*gr::www[z2rule-1][jj]
*jacz1*jacz2*mz1*mz2;
}
}
//cout << m << " " << pt << " " << y << " " << rsing << endl;
*/
rsing = rsing/2./M_PI;//*(asp_.asp_/2./M_PI);
}
else if (opts.order == 1)
rsing = 0.;
end_time = clock();
if (opts.timeprofile)
cout << setw(10) << "rsing time" << setw(10) << float( end_time - begin_time ) / CLOCKS_PER_SEC << setw(10) << "result" << setw(10) << rsing
<< endl;
//final result
double res = rdelta+rsing;
//cout << m << " " << pt << " " << y << " " << res << endl;
return res;
}
File Metadata
Details
Attached
Mime Type
text/x-diff
Expires
Tue, Nov 19, 7:15 PM (1 d, 10 h)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
3805791
Default Alt Text
(16 KB)
Attached To
R460 DYTurbo
Event Timeline
Log In to Comment