HiVector DriftCorrect::Solve(const HiVector &d) {
ostringstream hist;
_data = d;
if ( _skipFit || (!gotGoodLines(d))) {
_b2 = _data;
_coefs = HiVector(4, 0.0);
_uncert = _coefs;
_cc = HiVector(2, 0.0);
_chisq = 0.0;
if ( !gotGoodLines(d) ) {
hist << "NotEnoughLines(GoodLines[" << goodLines(d)
<< "],MinimumLines[" << _minLines << "]);";
}
hist << "SkipFit(TRUE: Not using LMFit)";
_history.add(hist.str());
}
else {
hist << "Fit(";
_b2 = HiVector(goodLines(_data));
if ( success(curvefit()) ) {
_coefs = coefs();
_uncert = uncert();
hist << "Solved,#Iters[" << nIterations() << "],ChiSq[" << Chisq()
<< "],DoF[" << DoF() << "])";
_history.add(hist.str());
_history.add("a0("+ToString(_coefs[0])+"+-"+ToString(_uncert[0])+")");
_history.add("a1("+ToString(_coefs[1])+"+-"+ToString(_uncert[1])+")");
_history.add("a2("+ToString(_coefs[2])+"+-"+ToString(_uncert[2])+")");
_history.add("a3("+ToString(_coefs[3])+"+-"+ToString(_uncert[3])+")");
}
else {
// Punt, fit a straight line to the data
_cc = poly_fit(d);
HiVector a(4);
a[0] = _cc[0];
a[1] = _cc[1];
a[2] = 0.0;
a[3] = 0.0;
_coefs = a;
hist << "Failed::Reason("<< statusstr() << "),#Iters["
<< nIterations() << "])";
_history.add(hist.str());
_history.add("a0("+ToString(_coefs[0])+")");
_history.add("a1("+ToString(_coefs[1])+")");
_history.add("a2("+ToString(_coefs[2])+")");
_history.add("a3("+ToString(_coefs[3])+")");
if ( _useLinFit ) {
_history.add("OnFailureUse(LinearFit(Zf))");
}
else {
_skipFit = true;
_history.add("OnFailureUse(ZfBuffer)");
}
}
}
return (Yfit());
}
bool Foam::SolverPerformance<Type>::operator!=
(
const SolverPerformance<Type>& sp
) const
{
return
(
solverName() != sp.solverName()
|| fieldName() != sp.fieldName()
|| initialResidual() != sp.initialResidual()
|| finalResidual() != sp.finalResidual()
|| nIterations() != sp.nIterations()
|| converged() != sp.converged()
|| singular() != sp.singular()
);
}
scalar surfaceOptimizer::optimiseSteepestDescent(const scalar tol)
{
point& pOpt = pts_[trias_[0][0]];
//- find the bounding box
const scalar avgEdge = Foam::mag(pMax_ - pMin_);
//- find the minimum value on the 5 x 5 raster
scalar K = evaluateStabilisationFactor();
scalar funcBefore, funcAfter(evaluateFunc(K));
//- start with steepest descent optimisation
vector gradF;
tensor gradGradF;
vector disp;
disp.z() = 0.0;
direction nIterations(0);
do
{
funcBefore = funcAfter;
evaluateGradients(K, gradF, gradGradF);
//- store data into a matrix
matrix2D mat;
mat[0][0] = gradGradF.xx();
mat[0][1] = gradGradF.xy();
mat[1][0] = gradGradF.yx();
mat[1][1] = gradGradF.yy();
FixedList<scalar, 2> source;
source[0] = gradF.x();
source[1] = gradF.y();
//- calculate the determinant
const scalar det = mat.determinant();
if( mag(det) < VSMALL )
{
disp = vector::zero;
}
else
{
disp.x() = mat.solveFirst(source);
disp.y() = mat.solveSecond(source);
if( mag(disp) > 0.2 * avgEdge )
{
vector dir = disp / mag(disp);
disp = dir * 0.2 * avgEdge;
}
}
# ifdef DEBUGSmooth
Info << "Second gradient " << gradGradF << endl;
Info << "Gradient " << gradF << endl;
Info << "Displacement " << disp << endl;
Info << "K = " << K << endl;
# endif
pOpt -= disp;
K = evaluateStabilisationFactor();
funcAfter = evaluateFunc(K);
if( mag(funcAfter - funcBefore) / funcBefore < tol )
break;
#ifdef DEBUGSmooth
Info << "New coordinates " << pOpt << endl;
# endif
} while( ++nIterations < 100 );
return funcAfter;
}
