math/example/daubechies_coefficients.cpp
2019-03-02 11:35:04 -07:00

222 lines
8.1 KiB
C++

/*
* Copyright Nick Thompson, 2018
* Use, modification and distribution are subject to the
* Boost Software License, Version 1.0. (See accompanying file
* LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
#include <iostream>
#include <vector>
#include <string>
#include <complex>
#include <bitset>
#include <boost/assert.hpp>
#include <boost/multiprecision/cpp_bin_float.hpp>
#include <boost/math/constants/constants.hpp>
#include <boost/math/tools/polynomial.hpp>
#include <boost/math/tools/roots.hpp>
#include <boost/math/special_functions/binomial.hpp>
#include <boost/multiprecision/cpp_complex.hpp>
#include <boost/multiprecision/complex128.hpp>
#include <boost/math/quadrature/gauss_kronrod.hpp>
using std::string;
using boost::math::tools::polynomial;
using boost::math::binomial_coefficient;
using boost::math::tools::schroder_iterate;
using boost::math::tools::halley_iterate;
using boost::math::tools::newton_raphson_iterate;
using boost::math::tools::complex_newton;
using boost::math::constants::half;
using boost::math::constants::root_two;
using boost::math::constants::pi;
using boost::math::quadrature::gauss_kronrod;
using boost::multiprecision::cpp_bin_float_100;
using boost::multiprecision::cpp_complex_100;
template<class Complex>
std::vector<std::pair<Complex, Complex>> find_roots(size_t p)
{
// Initialize the polynomial; see Mallat, A Wavelet Tour of Signal Processing, equation 7.96
BOOST_ASSERT(p>0);
typedef typename Complex::value_type Real;
std::vector<Complex> coeffs(p);
for (size_t k = 0; k < coeffs.size(); ++k)
{
coeffs[k] = Complex(binomial_coefficient<Real>(p-1+k, k), 0);
}
polynomial<Complex> P(std::move(coeffs));
polynomial<Complex> Pcopy = P;
polynomial<Complex> Pcopy_prime = P.prime();
auto orig = [&](Complex z) { return std::make_pair<Complex, Complex>(Pcopy(z), Pcopy_prime(z)); };
polynomial<Complex> P_prime = P.prime();
// Polynomial is of degree p-1.
std::vector<Complex> roots(p-1, {std::numeric_limits<Real>::quiet_NaN(),std::numeric_limits<Real>::quiet_NaN()});
size_t i = 0;
while(P.size() > 1)
{
Complex guess = {0.0, 1.0};
std::cout << std::setprecision(std::numeric_limits<Real>::digits10+3);
auto f = [&](Complex x)->std::pair<Complex, Complex>
{
return std::make_pair<Complex, Complex>(P(x), P_prime(x));
};
Complex r = complex_newton(f, guess);
using std::isnan;
if(isnan(r.real()))
{
int i = 50;
do {
// Try a different guess
guess *= Complex(1.0,-1.0);
r = complex_newton(f, guess);
std::cout << "New guess: " << guess << ", result? " << r << std::endl;
} while (isnan(r.real()) && i-- > 0);
if (isnan(r.real()))
{
std::cout << "Polynomial that killed the process: " << P << std::endl;
throw std::logic_error("Newton iteration did not converge");
}
}
// Refine r with the original function.
// We only use the polynomial division to ensure we don't get the same root over and over.
// However, the division induces error which can grow quickly-or slowly! See Numerical Recipes, section 9.5.1.
r = complex_newton(orig, r);
if (isnan(r.real()))
{
throw std::logic_error("Found a root for the deflated polynomial which is not a root for the original. Indicative of catastrophic numerical error.");
}
// Test the root:
using std::sqrt;
Real tol = sqrt(sqrt(std::numeric_limits<Real>::epsilon()));
if (norm(Pcopy(r)) > tol)
{
std::cout << "This is a bad root: P" << r << " = " << Pcopy(r) << std::endl;
std::cout << "Reduced polynomial leading to bad root: " << P << std::endl;
throw std::logic_error("Donezo.");
}
BOOST_ASSERT(i < roots.size());
roots[i] = r;
++i;
polynomial<Complex> q{-r, {1,0}};
// This optimization breaks at p = 11. I have no clue why.
// Unfortunate, because I expect it to be considerably more stable than
// repeatedly dividing by the complex root.
/*polynomial<Complex> q;
if (r.imag() > sqrt(std::numeric_limits<Real>::epsilon()))
{
// Then the complex conjugate is also a root:
using std::conj;
using std::norm;
BOOST_ASSERT(i < roots.size());
roots[i] = conj(r);
++i;
q = polynomial<Complex>({{norm(r), 0}, {-2*r.real(),0}, {1,0}});
}
else
{
// The imaginary part is numerical noise:
r.imag() = 0;
q = polynomial<Complex>({-r, {1,0}});
}*/
auto PR = quotient_remainder(P, q);
// I should validate that the remainder is small, but . . .
//std::cout << "Remainder = " << PR.second<< std::endl;
P = PR.first;
P_prime = P.prime();
}
std::vector<std::pair<Complex, Complex>> Qroots(p-1);
for (size_t i = 0; i < Qroots.size(); ++i)
{
Complex y = roots[i];
Complex z1 = static_cast<Complex>(1) - static_cast<Complex>(2)*y + static_cast<Complex>(2)*sqrt(y*(y-static_cast<Complex>(1)));
Complex z2 = static_cast<Complex>(1) - static_cast<Complex>(2)*y - static_cast<Complex>(2)*sqrt(y*(y-static_cast<Complex>(1)));
Qroots[i] = {z1, z2};
}
return Qroots;
}
template<class Complex>
std::vector<typename Complex::value_type> daubechies_coefficients(std::vector<std::pair<Complex, Complex>> const & Qroots)
{
typedef typename Complex::value_type Real;
size_t p = Qroots.size() + 1;
// Choose the minimum abs root; see Mallat, discussion just after equation 7.98
std::vector<Complex> chosen_roots(p-1);
for (size_t i = 0; i < p - 1; ++i)
{
if(norm(Qroots[i].first) <= 1)
{
chosen_roots[i] = Qroots[i].first;
}
else
{
BOOST_ASSERT(norm(Qroots[i].second) <= 1);
chosen_roots[i] = Qroots[i].second;
}
}
polynomial<Complex> R{1};
for (size_t i = 0; i < p-1; ++i)
{
Complex ak = chosen_roots[i];
R *= polynomial<Complex>({-ak/(static_cast<Complex>(1)-ak), static_cast<Complex>(1)/(static_cast<Complex>(1)-ak)});
}
polynomial<Complex> a{{half<Real>(), 0}, {half<Real>(),0}};
polynomial<Complex> poly = root_two<Real>()*pow(a, p)*R;
std::vector<Complex> result = poly.data();
// If we reverse, we get the Numerical Recipes and Daubechies convention.
// If we don't reverse, we get the Pywavelets and Mallat convention.
// I believe this is because of the sign convention on the DFT, which differs between Daubechies and Mallat.
// You implement a dot product in Daubechies/NR convention, and a convolution in PyWavelets/Mallat convention.
// I won't reverse so I can spot check against Pywavelets: http://wavelets.pybytes.com/wavelet/
//std::reverse(result.begin(), result.end());
std::vector<Real> h(result.size());
for (size_t i = 0; i < result.size(); ++i)
{
Complex r = result[i];
BOOST_ASSERT(r.imag() < sqrt(std::numeric_limits<Real>::epsilon()));
h[i] = r.real();
}
// Quick sanity check: We could check all vanishing moments, but that sum is horribly ill-conditioned too!
Real sum = 0;
Real scale = 0;
for (size_t i = 0; i < h.size(); ++i)
{
sum += h[i];
scale += h[i]*h[i];
}
BOOST_ASSERT(abs(scale -1) < sqrt(std::numeric_limits<Real>::epsilon()));
BOOST_ASSERT(abs(sum - root_two<Real>()) < sqrt(std::numeric_limits<Real>::epsilon()));
return h;
}
int main()
{
typedef boost::multiprecision::cpp_complex<100> Complex;
for(size_t p = 1; p < 200; ++p)
{
auto roots = find_roots<Complex>(p);
auto h = daubechies_coefficients(roots);
std::cout << "h_" << p << "[] = {";
for (auto& x : h) {
std::cout << x << ", ";
}
std::cout << "} // = h_" << p << "\n\n\n\n";
}
}