integrate.h

Go to the documentation of this file.

/* This file is part of Cloudy and is copyright (C)1978-2017 by Gary J. Ferland and
 * others.  For conditions of distribution and use see copyright notice in license.txt */

#ifndef INTEGRATE_H_
#define INTEGRATE_H_

#include "vectorize.h"

// Define integration methods
typedef enum { Gaussian32, VecGaussian32, Legendre } methods;

static const int numPoints = 32;

ALIGNED(CD_ALIGN) static const double qg32_w[numPoints] = {
        .35093050047350483e-2, .81371973654528350e-2, .12696032654631030e-1, .17136931456510717e-1,
        .21417949011113340e-1, .25499029631188088e-1, .29342046739267774e-1, .32911111388180923e-1,
        .36172897054424253e-1, .39096947893535153e-1, .41655962113473378e-1, .43826046502201906e-1,
        .45586939347881942e-1, .46922199540402283e-1, .47819360039637430e-1, .48270044257363900e-1,
        .48270044257363900e-1, .47819360039637430e-1, .46922199540402283e-1, .45586939347881942e-1,
        .43826046502201906e-1, .41655962113473378e-1, .39096947893535153e-1, .36172897054424253e-1,
        .32911111388180923e-1, .29342046739267774e-1, .25499029631188088e-1, .21417949011113340e-1,
        .17136931456510717e-1, .12696032654631030e-1, .81371973654528350e-2, .35093050047350483e-2
};

ALIGNED(CD_ALIGN) static const double qg32_a[numPoints] = {
        -.498631930924740780,  -.49280575577263417,  -.4823811277937532200, -.46745303796886984000,
        -.448160577883026060,  -.42468380686628499,  -.3972418979839712000, -.36609105937014484000,
        -.331522133465107600,  -.29385787862038116,  -.2534499544661147000, -.21067563806531767000,
        -.165934301141063820,  -.11964368112606854,  -.7223598079139825e-1, -.24153832843869158e-1,
         .24153832843869158e-1, .7223598079139825e-1, .11964368112606854,    .165934301141063820,
         .21067563806531767000, .2534499544661147000, .29385787862038116,    .331522133465107600,
         .36609105937014484000, .3972418979839712000, .42468380686628499,    .448160577883026060,
         .46745303796886984000, .4823811277937532200, .49280575577263417,    .498631930924740780
};

// define an integrator class.  Currently hard-wired to 32-point Gaussian
template<typename Integrand, methods Method>
class Integrator
{
        const double *p_w, *p_a;
public:
        Integrand func;
        explicit Integrator( const Integrand& fun ) : p_w(qg32_w), p_a(qg32_a), func(fun) {}
        double sum(double min, double max) const
        {
                if( Method == Gaussian32 )
                {
                        double a = 0.5*(max+min);
                        double b = max-min;
                        double total = 0.;
                        for( long i=0; i < numPoints; i++ )
                                total += b * p_w[i] * func(a + b*p_a[i]);
                        return total;
                }
                else
                        TotalInsanity();
        }
};

// define a vectorized integrator class.  Currently hard-wired to 32-point Gaussian
template<typename Integrand, methods Method>
class VecIntegrator
{
        const double *p_w, *p_a;
public:
        Integrand func;
        explicit VecIntegrator( const Integrand& fun ) : p_w(qg32_w), p_a(qg32_a), func(fun) {}
        double sum(double min, double max) const
        {
                if( Method == VecGaussian32 )
                {
                        double a = 0.5*(max+min);
                        double b = max-min;
                        ALIGNED(CD_ALIGN) double x[numPoints], y[numPoints];
                        for( long i=0; i < numPoints; i++ )
                                x[i] = a + b*p_a[i];
                        func(x, y, numPoints);
                        return b * reduce_ab(p_w, y, 0, numPoints);
                }
                else
                        TotalInsanity();
        }
};

double qg32( double, double, double(*)(double) );
/* declar of optimize_func, the last arg, changed from double(*)() to above,
 * seemed to fix flags that were raised */

// Open and closed integration routines, developed on qromb etc. in
// Press et al.
namespace integrate
{
        template <class T>
        class Trapezium
        {
                T m_f;
                double m_a, m_b, m_sum;
                long m_n;
        public:
                static const int NREF=2;
                Trapezium(const T& f, double a, double b) : m_f(f), m_a(a), m_b(b), m_sum(0.0), m_n(0) {}
                void step()
                {
                        if (m_n == 0)
                        {
                                m_sum = 0.5*(m_b-m_a)*(m_f(m_a)+m_f(m_b));
                                m_n = 1;
                        }
                        else
                        {
                                double rn = 1.0/m_n;
                                double si = 0.0;
                                for (int i=0; i<m_n; ++i)
                                {
                                        double x = (m_a*(m_n-i-0.5)+m_b*(i+0.5))*rn;
                                        si += m_f(x);
                                }
                                m_sum = (m_sum+si*(m_b-m_a)*rn)/NREF;
                                m_n *= NREF;
                        }
                }
                double sum() const
                {
                        return m_sum;
                }
                long evals() const
                {
                return m_n;
                }
        };

        template <class T>
        class Midpoint
        {
                T m_f;
                double m_a, m_b, m_sum;
                long m_n;
        public:
                static const int NREF=3;
                Midpoint(const T& f, double a, double b) : m_f(f), m_a(a), m_b(b), m_sum(0.0), m_n(0) {}
                void step()
                {
                        if (m_n == 0)
                        {
                                m_sum = (m_b-m_a)*m_f(0.5*(m_a+m_b));
                                m_n = 1;
                        }
                        else
                        {
                                double rn = 1.0/m_n;
                                const double sixth = 1./6.;
                                double si = 0.0;
                                for (int i=0; i<m_n; ++i)
                                {
                                        double x1 = (m_a*(m_n-i-sixth)+m_b*(i+sixth))*rn;
                                        si += m_f(x1);
                                        double x2 = (m_a*(m_n-i-1+sixth)+m_b*(i+1-sixth))*rn;
                                        si += m_f(x2);
                                }
                                m_sum = (m_sum+si*(m_b-m_a)*rn)/NREF;
                                m_n *= NREF;
                        }
                }
                double sum() const
                {
                        return m_sum;
                }
                long evals() const
                {
                        return m_n;
                }
        };
        
        template <class T>
        class Romberg
        {
                T m_f;
                double m_sum, m_dy;
        public:
                explicit Romberg(const T& f) : m_f(f), m_sum(0.0), m_dy(0.0) {}
                void update(double eps);
                double sum() const
                {
                        return m_sum;
                }
                double error() const
                {
                        return m_dy;
                }
                long evals() const
                {
                        return m_f.evals();
                }
        };
        
        template <class T>
        inline void Romberg<T>::update(double eps)
        {
                const int itmax=40/m_f.NREF, npt=5;
                double d1[npt], d0[npt-1];
                double y = 0.0;
                const double w1 = m_f.NREF*m_f.NREF, w2 = 1.0/w1;
                for (int i=0; i<npt; ++i)
                {
                        d1[i] = 0.;
                }
                for (int i=0; i<itmax; ++i)
                {
                        m_f.step();
                        y = m_f.sum();
                        
                        int l = (i<npt-1) ? i : npt-1;
                        for (int m=0; m<l; ++m)
                        {
                                d0[m] = d1[m];
                        }
                        d1[0] = y;
                        double fr = 1.0;
                        for (int m=0; m<l; ++m)
                        {
                                d1[m+1] = (d1[m]-d0[m]*fr)/(w1-fr);
                                y += d1[m+1];
                                fr *= w2;
                        }
                        
                        m_dy = fabs(m_sum-y);
                        m_sum = y;
                        //printf("Level %d value %g->%g error %g samples %ld\n",l,d1[0],m_sum,m_dy,m_f.evals());
                        if ( i > 2 && m_dy <= eps*fabs(y) )
                        {
                                return;
                        }
                }
                return;
        }

        template <class T>
        class Simple
        {
                T m_f;
                double m_sum, m_dy;
        public:
                explicit Simple(const T& f) : m_f(f), m_sum(0.0), m_dy(0.0) {}
                void update(double eps);
                double sum() const
                {
                        return m_sum;
                }
                double error() const
                {
                        return m_dy;
                }
                long evals() const
                {
                        return m_f.evals();
                }
        };
        
        template <class T>
        inline void Simple<T>::update(double eps)
        {
                const int itmax=40/m_f.NREF;
                double coarse=0.,fine=0.;
                for (int i=0; i<itmax; ++i)
                {
                        coarse = fine;
                        m_f.step();
                        fine = m_f.sum();
                                                
                        m_dy = fabs(coarse-fine);
                        m_sum = fine;
                        // printf("Level %d value %g error %g samples %ld\n",i,m_sum,m_dy,m_f.evals());
                        if ( i > 2 && m_dy <= eps*fabs(fine) )
                        {
                                return;
                        }
                }
                return;
        }
}
#endif /* INTEGRATE_H_ */