LMU/musrfit
LMU/musrfit
5
0
Fork 0
Code Issues 6 Pull Requests Actions Packages Projects Releases 3 Activity

improved handling of LF, including bug fixing.

Browse Source
This commit is contained in:
nemu 2010-08-05 14:20:49 +00:00
parent 599c83a17c
commit e94d869534
2 changed files with 92 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

113
src/classes/PTheory.cpp
View File

@ -99,6 +99,7 @@ PTheory::PTheory(PMsrHandler *msrInfo, UInt_t runNo, const Bool_t hasParent)
fUserFcnClassName = TString("");
fUserFcn = 0;
fDynLFdt = 0.0;
fSamplingTime = 0.001; // default = 1ns (units in us)
static UInt_t lineNo = 1; // lineNo
static UInt_t depth = 0; // needed to handle '+' properly
@ -1177,8 +1178,12 @@ Double_t PTheory::StaticGaussKTLF(register Double_t t, const PDoubleVector& para
if (tt < 0.0) // for times < 0 return a function value of 1.0
return 1.0;
if (val[0] < 0.02) { // if smaller 20kHz ~ 0.27G use zero field formula
Double_t sigma_t_2 = tt*tt*val[1]*val[1];
Double_t sigma_t_2 = 0.0;
if (val[0] < 0.02) { // if smaller 20kHz ~ 0.27G use the ZF formula
sigma_t_2 = tt*tt*val[1]*val[1];
result = 0.333333333333333 * (1.0 + 2.0*(1.0 - sigma_t_2)*TMath::Exp(-0.5*sigma_t_2));
} else if (val[1]/val[0] > 79.5775) { // check if Delta/w0 > 500.0, in which case the ZF formula is used
sigma_t_2 = tt*tt*val[1]*val[1];
result = 0.333333333333333 * (1.0 + 2.0*(1.0 - sigma_t_2)*TMath::Exp(-0.5*sigma_t_2));
} else {
Double_t delta = val[1];
@ -1390,7 +1395,10 @@ Double_t PTheory::StaticLorentzKTLF(register Double_t t, const PDoubleVector& pa
if (tt < 0.0) // for times < 0 return a function value of 1.0
return 1.0;
if (val[0] < 0.02) { // if smaller 20kHz ~ 0.27G use zero field formula
if (val[0] < 0.02) { // if smaller 20kHz ~ 0.27G use the ZF formula
Double_t at = tt*val[1];
result = 0.333333333333333 * (1.0 + 2.0*(1.0 - at)*TMath::Exp(-at));
} else if (val[1]/val[0] > 159.1549) { // check if a/w0 > 1000.0, in which case the ZF formula is used
Double_t at = tt*val[1];
result = 0.333333333333333 * (1.0 + 2.0*(1.0 - at)*TMath::Exp(-at));
} else {
@ -2011,9 +2019,10 @@ Double_t PTheory::UserFcn(register Double_t t, const PDoubleVector& paramValues,
/**
* <p>Calculates the non-analytic integral of the static Gaussian Kubo-Toyabe in longitudinal field, i.e.
* \f[ G_{\rm G,LF}(t) = \int^t_0 \exp\left[-1/2 (\sigma\tau)^2\right]\sin(2\pi\nu\tau)\mathrm{d}\tau \f]
* and stores \f$G_{\rm G,LF}(t)\f$ in a vector, fLFIntegral, from \f$t=0\ldots 20\f$ us with a time resolution of 1 ns.
* and stores \f$G_{\rm G,LF}(t)\f$ in a vector, fLFIntegral, from \f$t=0\f$ up to 20us, if needed
*
* The fLFIntegral is given in steps of 1 ns, e.g. index i=100 coresponds to t=100ns
* The fLFIntegral is given in steps of fSamplingTime (default = 1 ns), e.g. for fSamplingTime = 0.001,
* index i=100 coresponds to t=100ns
*
* <b>meaning of val:</b> val[0]=\f$\nu\f$, val[1]=\f$\sigma\f$
*
@ -2023,15 +2032,39 @@ void PTheory::CalculateGaussLFIntegral(const Double_t *val) const
{
// val[0] = nu (field), val[1] = Delta
if (val[0] == 0.0) // field == 0.0, hence nothing to be done
if (val[0] == 0.0) { // field == 0.0, hence nothing to be done
return;
} else if (val[1]/val[0] > 79.5775) { // check if a/w0 > 500.0, in which case the ZF formula is used and here nothing has to be done
return;
}
const Double_t dt=0.001; // all times in usec
Double_t dt=0.001; // all times in usec
Double_t t, ft;
Double_t w0 = TMath::TwoPi()*val[0];
Double_t Delta = val[1];
Double_t preFactor = 2.0*TMath::Power(Delta, 4.0) / TMath::Power(w0, 3.0);
// check if the time resolution needs to be increased
const Int_t samplingPerPeriod = 20;
const Int_t samplingOnExp = 3000;
if ((Delta <= w0) && (1.0/val[0] < 20.0)) { // makes sure that the frequency sampling is fine enough
if (1.0/val[0]/samplingPerPeriod < 0.001) {
dt = 1.0/val[0]/samplingPerPeriod;
}
} else if ((Delta > w0) && (Delta <= 10.0)) {
if (Delta/w0 > 10.0) {
dt = 0.00005;
}
} else if ((Delta > w0) && (Delta > 10.0)) { // makes sure there is a fine enough sampling for large Delta's
if (1.0/Delta/samplingOnExp < 0.001) {
dt = 1.0/Delta/samplingOnExp;
}
}
// keep sampling time
fSamplingTime = dt;
// clear previously allocated vector
fLFIntegral.clear();
@ -2040,21 +2073,24 @@ void PTheory::CalculateGaussLFIntegral(const Double_t *val) const
fLFIntegral.push_back(0.0); // start value of the integral
ft = 0.0;
for (UInt_t i=1; i<2e4; i++) {
Double_t step = 0.0;
do {
t += dt;
ft += 0.5*dt*preFactor*(TMath::Exp(-0.5*TMath::Power(Delta * (t-dt), 2.0))*TMath::Sin(w0*(t-dt))+
TMath::Exp(-0.5*TMath::Power(Delta * t, 2.0))*TMath::Sin(w0*t));
step = 0.5*dt*preFactor*(exp(-0.5*pow(Delta * (t-dt), 2.0))*sin(w0*(t-dt))+
exp(-0.5*pow(Delta * t, 2.0))*sin(w0*t));
ft += step;
fLFIntegral.push_back(ft);
}
} while ((t <= 20.0) && (fabs(step) > 1.0e-10));
}
//--------------------------------------------------------------------------
/**
* <p>Calculates the non-analytic integral of the static Lorentzian Kubo-Toyabe in longitudinal field, i.e.
* \f[ G_{\rm L,LF}(t) = \int^t_0 \exp\left[-a \tau \right] j_0(2\pi\nu\tau)\mathrm{d}\tau \f]
* and stores \f$G_{\rm L,LF}(t)\f$ in a vector, fLFIntegral, from \f$t=0\ldots 20\f$ us with a time resolution of 1 ns.
* and stores \f$G_{\rm L,LF}(t)\f$ in a vector, fLFIntegral, from \f$t=0\f$ up to 20 us, if needed.
*
* The fLFIntegral is given in steps of 1 ns, e.g. index i=100 coresponds to t=100ns
* The fLFIntegral is given in steps of fSamplingTime (default = 1 ns), e.g. for fSamplingTime = 0.001,
* index i=100 coresponds to t=100ns
*
* <b>meaning of val:</b> val[0]=\f$\nu\f$, val[1]=\f$a\f$
*
@ -2064,15 +2100,39 @@ void PTheory::CalculateLorentzLFIntegral(const Double_t *val) const
{
// val[0] = nu, val[1] = a
if (val[0] == 0.0) // field == 0.0, hence nothing to be done
// a few checks if the integral actually needs to be calculated
if (val[0] < 0.02) { // if smaller 20kHz ~ 0.27G use the ZF formula and here nothing has to be done
return;
} else if (val[1]/val[0] > 159.1549) { // check if a/w0 > 1000.0, in which case the ZF formula is used and here nothing has to be done
return;
}
const Double_t dt=0.001; // all times in usec
Double_t dt=0.001; // all times in usec
Double_t t, ft;
Double_t w0 = TMath::TwoPi()*val[0];
Double_t a = val[1];
Double_t preFactor = a*(1+pow(a/w0,2.0));
// check if the time resolution needs to be increased
const Int_t samplingPerPeriod = 20;
const Int_t samplingOnExp = 3000;
if ((a <= w0) && (1.0/val[0] < 20.0)) { // makes sure that the frequency sampling is fine enough
if (1.0/val[0]/samplingPerPeriod < 0.001) {
dt = 1.0/val[0]/samplingPerPeriod;
}
} else if ((a > w0) && (a <= 10.0)) {
if (a/w0 > 10.0) {
dt = 0.00005;
}
} else if ((a > w0) && (a > 10.0)) { // makes sure there is a fine enough sampling for large a's
if (1.0/a/samplingOnExp < 0.001) {
dt = 1.0/a/samplingOnExp;
}
}
// keep sampling time
fSamplingTime = dt;
// clear previously allocated vector
fLFIntegral.clear();
@ -2086,12 +2146,15 @@ void PTheory::CalculateLorentzLFIntegral(const Double_t *val) const
ft += 0.5*dt*preFactor*(1.0+sin(w0*t)/(w0*t)*exp(-a*t));
fLFIntegral.push_back(ft);
// calculate all the other integral bin values
for (UInt_t i=2; i<2e4; i++) {
Double_t step = 0.0;
do {
t += dt;
ft += 0.5*dt*preFactor*(sin(w0*(t-dt))/(w0*(t-dt))*exp(-a*(t-dt))+sin(w0*t)/(w0*t)*exp(-a*t));
step = 0.5*dt*preFactor*(sin(w0*(t-dt))/(w0*(t-dt))*exp(-a*(t-dt))+sin(w0*t)/(w0*t)*exp(-a*t));
ft += step;
fLFIntegral.push_back(ft);
} while ((t <= 20.0) && (fabs(step) > 1.0e-10));
}
}
//--------------------------------------------------------------------------
/**
@ -2103,16 +2166,16 @@ void PTheory::CalculateLorentzLFIntegral(const Double_t *val) const
*/
Double_t PTheory::GetLFIntegralValue(const Double_t t) const
{
UInt_t idx = static_cast<UInt_t>(t/0.001); // since LF-integral is calculated in nsec
UInt_t idx = static_cast<UInt_t>(t/fSamplingTime);
if (idx < 0)
return 0.0;
if (idx > fLFIntegral.size()-1)
if (idx > fLFIntegral.size()-2)
return fLFIntegral[fLFIntegral.size()-1];
// liniarly interpolate between the two relvant function bins
Double_t df = (fLFIntegral[idx+1]-fLFIntegral[idx])/0.001*(t-idx*0.001);
Double_t df = (fLFIntegral[idx+1]-fLFIntegral[idx])/fSamplingTime*(t-idx*fSamplingTime);
return fLFIntegral[idx]+df;
}
@ -2185,6 +2248,9 @@ void PTheory::CalculateDynKTLF(const Double_t *val, Int_t tag) const
if (val[0] < 0.02) { // if smaller 20kHz ~ 0.27G use zero field formula
Double_t sigma_t_2 = t*t*val[1]*val[1];
p0exp[i] = 0.333333333333333 * (1.0 + 2.0*(1.0 - sigma_t_2)*TMath::Exp(-0.5*sigma_t_2));
} else if (val[1]/val[0] > 79.5775) { // check if Delta/w0 > 500.0, in which case the ZF formula is used
Double_t sigma_t_2 = t*t*val[1]*val[1];
p0exp[i] = 0.333333333333333 * (1.0 + 2.0*(1.0 - sigma_t_2)*TMath::Exp(-0.5*sigma_t_2));
} else {
Double_t delta = val[1];
Double_t w0 = TWO_PI*val[0];
@ -2198,6 +2264,9 @@ void PTheory::CalculateDynKTLF(const Double_t *val, Int_t tag) const
if (val[0] < 0.02) { // if smaller 20kHz ~ 0.27G use zero field formula
Double_t at = t*val[1];
p0exp[i] = 0.333333333333333 * (1.0 + 2.0*(1.0 - at)*TMath::Exp(-at));
} else if (val[1]/val[0] > 159.1549) { // check if a/w0 > 1000.0, in which case the ZF formula is used
Double_t at = t*val[1];
p0exp[i] = 0.333333333333333 * (1.0 + 2.0*(1.0 - at)*TMath::Exp(-at));
} else {
Double_t a = val[1];
Double_t at = a*t;
@ -2260,7 +2329,7 @@ Double_t PTheory::GetDynKTLFValue(const Double_t t) const
if (idx < 0)
return 0.0;
if (idx > static_cast<Int_t>(fDynLFFuncValue.size())-1)
if (idx > static_cast<Int_t>(fDynLFFuncValue.size())-2)
return fDynLFFuncValue[fDynLFFuncValue.size()-1];
// liniarly interpolate between the two relvant function bins

1
src/include/PTheory.h
View File

@ -247,6 +247,7 @@ class PTheory
PMsrHandler *fMsrInfo; ///< pointer to the msr-file handler
mutable Double_t fSamplingTime; ///< needed for LF. Keeps the sampling time of the non-analytic integral
mutable Double_t fPrevParam[THEORY_MAX_PARAM]; ///< needed for LF. Keeps the previous fitting parameters
mutable PDoubleVector fLFIntegral; ///< needed for LF. Keeps the non-analytic integral values
mutable Double_t fDynLFdt; ///< needed for LF. Keeps the time step for the dynamic LF calculation, used in the integral equation approach