PolynomialSolver.h

Go to the documentation of this file.

/********************************************************************
**                Image Component Library (ICL)                    **
**                                                                 **
** Copyright (C) 2006-2013 CITEC, University of Bielefeld          **
**                         Neuroinformatics Group                  **
** Website: www.iclcv.org and                                      **
**          http://opensource.cit-ec.de/projects/icl               **
**                                                                 **
** File   : ICLUtils/src/ICLMath/PolynomialSolver.h                **
** Module : ICLMath                                                **
** Authors: Sergius Gaulik                                         **
**                                                                 **
**                                                                 **
** GNU LESSER GENERAL PUBLIC LICENSE                               **
** This file may be used under the terms of the GNU Lesser General **
** Public License version 3.0 as published by the                  **
**                                                                 **
** Free Software Foundation and appearing in the file LICENSE.LGPL **
** included in the packaging of this file.  Please review the      **
** following information to ensure the license requirements will   **
** be met: http://www.gnu.org/licenses/lgpl-3.0.txt                **
**                                                                 **
** The development of this software was supported by the           **
** Excellence Cluster EXC 277 Cognitive Interaction Technology.    **
** The Excellence Cluster EXC 277 is a grant of the Deutsche       **
** Forschungsgemeinschaft (DFG) in the context of the German       **
** Excellence Initiative.                                          **
**                                                                 **
********************************************************************/

#pragma once

#include <ICLUtils/BasicTypes.h>
#include <float.h>

#if (defined _MSC_VER && _MSC_VER < 1800)
#pragma WARNING("This compiler does not support the functions ilogb and scalbln.")
#else

namespace icl{
  namespace math{

  typedef std::complex<double> xcomplex;

    /****************************************************************************
     *
     * cpoly.C                                                       René Hartung
     *                                                          Laurent Bartholdi
     *
     *   @(#)$id: fr_dll.c,v 1.18 2010/10/26 05:19:40 gap exp $
     *
     * Copyright (c) 2011, Laurent Bartholdi
     *
     ****************************************************************************
     *
     * template for root-finding of complex polynomial
     *
     * The ICL developers applied only minor changes to this file
     ****************************************************************************/

    typedef double xreal;
    static const struct { double ZERO, INFIN; int MIN_EXP, MAX_EXP; }
    xdata = { 0.0, DBL_MAX, DBL_MIN_EXP, DBL_MAX_EXP };
    static xreal xnorm(xcomplex z) { return std::norm(z); }
    static xreal xabs(xcomplex z) { return std::abs(z); }
    static xreal xroot(xreal x, int n) { return pow(x, 1.0 / n); }
    static int xlogb(xcomplex z) { return ilogb(xnorm(z)) / 2; }
    #define xbits(z) DBL_MANT_DIG
    #define xeta(z) DBL_EPSILON
    static void xscalbln(xcomplex *z, int e) {
      z->real(scalbln(z->real(), e));
      z->imag(scalbln(z->imag(), e));
    }

    // CAUCHY COMPUTES A LOWER BOUND ON THE MODULI OF THE ZEROS OF A
    // POLYNOMIAL - PT IS THE MODULUS OF THE COEFFICIENTS.
    //
    static xcomplex cauchy(const int deg, xcomplex *P)
    {
      xreal x, xm, f, dx, df;
      xreal *tmp = new xreal[deg + 1];

      for(int i = 0; i<=deg; i++){ tmp[i] = xabs(P[i]); };

      // Compute upper estimate bound
      x = xroot(tmp[deg],deg) / xroot(tmp[0],deg);
      if(tmp[deg - 1] != 0.0) {
        // Newton step at the origin is better, use it
        xm = tmp[deg] / tmp[deg-1];
        if (xm < x) x = xm;
      }

      tmp[deg] = -tmp[deg];

      // Chop the interval (0,x) until f < 0
      while(1) {
        xm = x * 0.1;
        // Evaluate the polynomial <tmp> at <xm>
        f = tmp[0];
        for(int i = 1; i <= deg; i++)
          f = f * xm + tmp[i];

        if(f <= 0.0) break;
        x = xm;
      }
      dx = x;

       // Do Newton iteration until x converges to two decimal places
      while(fabs(dx / x) > 0.005) {
        f  = tmp[0];
        df = 0.0;
        for(int i = 1; i <= deg; i++){
          df = df * x + f;
          f  =  f * x + tmp[i];
        }

        dx = f / df;
        x -= dx;                                // Newton step
      }

      delete tmp;

      return (xcomplex)(x);
    }

    // RETURNS A SCALE FACTOR TO MULTIPLY THE COEFFICIENTS OF THE POLYNOMIAL.
    // THE SCALING IS DONE TO AVOID OVERFLOW AND TO AVOID UNDETECTED UNDERFLOW
    // INTERFERING WITH THE CONVERGENCE CRITERION.  THE FACTOR IS A POWER OF THE
    //int BASE.
    // PT - MODULUS OF COEFFICIENTS OF P
    // ETA, INFIN, SMALNO, BASE - CONSTANTS DESCRIBING THE FLOATING POINT ARITHMETIC.
    //
    static void scale(const int deg, xcomplex* P)
    {
       int hi, lo, max, min, x, sc;

       // Find largest and smallest moduli of coefficients
       hi = (int)(xdata.MAX_EXP / 2.0);
       lo = (int)(xdata.MIN_EXP - xbits(P[0]));
       max = xlogb(P[0]); // leading coefficient does not vanish!
       min = xlogb(P[0]);

       for(int i = 0; i <= deg; i++) {
          if (P[i] != xdata.ZERO){
            x = xlogb(P[i]);
            if(x > max) max = x;
            if(x < min) min = x;
          }
        }

       // Scale only if there are very large or very small components
       if(min >= lo && max <= hi) return;

       x = lo - min;
       if(x <= 0)
          sc = -(max+min) / 2;
       else {
          sc = x;
          if(xdata.MAX_EXP - sc > max) sc = 0;
          }

       // Scale the polynomial
       for(int i = 0; i<= deg; i++){ xscalbln(&P[i],sc); }
    }

    // COMPUTES  THE DERIVATIVE  POLYNOMIAL AS THE INITIAL H
    // POLYNOMIAL AND COMPUTES L1 NO-SHIFT H POLYNOMIALS.
    //
    static void noshft(const int l1, int deg, xcomplex *P, xcomplex *H)
    {
      int i, j, jj;
      //#ifndef ICL_SYSTEM_WINDOWS removed by CE
      //#warning "why is tmp not iniitialized here?"
      //#endif
      xcomplex t, tmp;

      // compute the first H-polynomial as the (normed) derivative of P
      for (i = 0; i < deg; i++) {
        // CE: shouldn't it be this here (I add it since it seems to not make sense
        // to use the uninitiylized tmp here a multipliler ...
        //
        tmp = P[i]; // added by CE
        tmp *= (deg - i);
        tmp /= deg;
        H[i] = tmp;
      }
#if 0
      // possible fix? found here: https://github.com/gap-packages/float/blob/master/src/cpoly.C
      for(i = 0; i < deg; i++)
        H[i] = (P[i] * (deg-i)) / deg;
#endif

      for(jj = 1; jj <= l1; jj++) {
        if(xnorm(H[deg - 1]) > xeta(P[deg-1])*xeta(P[deg-1])* 10*10 * xnorm(P[deg - 1])) {
          t = -P[deg] / H[deg-1];
          for(i = 0; i < deg-1; i++){
      j = deg - i - 1;
      H[j] = t * H[j-1] + P[j];
          }
          H[0] = P[0];
        } else {
          // if the constant term is essentially zero, shift H coefficients
          for(i = 0; i < deg-1; i++) {
      j = deg - i - 1;
      H[j] = H[j - 1];
          }
          H[0] = xdata.ZERO;
        }
      }
    }

    // EVALUATES A POLYNOMIAL  P  AT  S  BY THE HORNER RECURRENCE
    // PLACING THE PARTIAL SUMS IN Q AND THE COMPUTED VALUE IN PV.
    //
    static xcomplex polyev(const int deg, const xcomplex s, const xcomplex *Q, xcomplex *q) {
       q[0] = Q[0];
       for(int i = 1; i <= deg; i++){ q[i] = q[i-1] * s + Q[i]; };
       return q[deg];
    }

    // COMPUTES  T = -P(S)/H(S).
    // BOOL   - LOGICAL, SET TRUE IF H(S) IS ESSENTIALLY ZERO.
    //
    static xcomplex calct(bool *bol, int deg, xcomplex Ps, xcomplex *H, xcomplex *h, xcomplex s){
      xcomplex Hs;
      Hs = polyev(deg-1, s, H, h);
      *bol = xnorm(Hs) <= xeta(H[deg-1])*xeta(H[deg-1]) * 10*10 * xnorm(H[deg-1]);
      if(!*bol)
        return -Ps / Hs;
      else
        return xdata.ZERO;
    }

    // CALCULATES THE NEXT SHIFTED H POLYNOMIAL.
    // BOOL   -  LOGICAL, IF .TRUE. H(S) IS ESSENTIALLY ZERO
    //
    static void nexth(const bool bol, int deg, xcomplex t, xcomplex *H, xcomplex *h, xcomplex *p){
       if(!bol){
          for(int j = 1; j < deg; j++)
             H[j] = t * h[j-1] + p[j];
          H[0] = p[0];
          }
       else {
         // If h[s] is zero replace H with qh
         for(int j = 1; j < deg; j++)
            H[j] = h[j - 1];
         h[0] = xdata.ZERO;
       }
    }

    // BOUNDS THE ERROR IN EVALUATING THE POLYNOMIAL BY THE HORNER RECURRENCE.
    // QR,QI - THE PARTIAL SUMS
    // MS    -MODULUS OF THE POINT
    // MP    -MODULUS OF POLYNOMIAL VALUE
    // ARE, MRE -ERROR BOUNDS ON COMPLEX ADDITION AND MULTIPLICATION
    //
    static xreal errev(const int deg, const xcomplex *p, const xreal ms, const xreal mp){
       xreal MRE = 2.0 * sqrt(2.0) * xeta(p[0]);
       xreal e =  xabs(p[0]) * MRE / (xeta(p[0]) + MRE);

       for(int i = 0; i <= deg; i++)
          e = e * ms + xabs(p[i]);

       return e * (xeta(p[0]) + MRE) - MRE * mp;
    }

    // CARRIES OUT THE THIRD STAGE ITERATION.
    // L3 - LIMIT OF STEPS IN STAGE 3.
    // ZR,ZI   - ON ENTRY CONTAINS THE INITIAL ITERATE, IF THE
    //           ITERATION CONVERGES IT CONTAINS THE FINAL ITERATE ON EXIT.
    // CONV    -  .TRUE. IF ITERATION CONVERGES
    //
    static bool vrshft(const int l3, int deg, xcomplex *P, xcomplex *p, xcomplex *H, xcomplex *h, xcomplex *zero, xcomplex *s){
      bool bol, conv, b;
      int i, j;
      xcomplex Ps, t;
      xreal mp, ms, omp = 0.0, relstp = 0.0, tp;

      conv = b = false;
      *s = *zero;

      // Main loop for stage three
      for(i = 1; i <= l3; i++) {
        // Evaluate P at S and test for convergence
        Ps = polyev(deg, *s, P, p);
        mp = xabs(Ps);
        ms = xabs(*s);
        if(mp <= 20 * errev(deg, p, ms, mp)) {
          // Polynomial value is smaller in value than a bound on the error
          // in evaluating P, terminate the iteration
          conv = true;
          *zero = *s;
          return conv;
        }

        if(i != 1) {
          if(!(b || mp < omp || relstp >= 0.05)){
      //       if(!(b || xlogb(mp) < omp || real(relstp) >= 0.05)){
      // Iteration has stalled. Probably a cluster of zeros. Do 5 fixed
      // shift steps into the cluster to force one zero to dominate
      tp = relstp;
      b = true;
      if(relstp < xeta(P[0])) tp = xeta(P[0]);

      *s *= 1.0 + xcomplex(1.0, 1.0)*sqrt(tp);

      Ps = polyev(deg, *s, P, p);
      for(j = 1; j <= 5; j++){
        t = calct(&bol, deg, Ps, H, h, *s);
        nexth(bol, deg, t, H, h, p);
      }
      omp = xdata.INFIN;
      goto _20;
          }

          // Exit if polynomial value increase significantly
          if(mp * 0.1 > omp) return conv;
        }

        omp = mp;

        // Calculate next iterate
      _20:  t = calct(&bol, deg, Ps, H, h, *s);
        nexth(bol, deg, t, H, h, p);
        t = calct(&bol, deg, Ps, H, h, *s);
        if(!bol) {
          relstp = xabs(t) / xabs(*s);
          *s += t;
        }
      } // end for

      return conv;
    }

    // COMPUTES L2 FIXED-SHIFT H POLYNOMIALS AND TESTS FOR CONVERGENCE.
    // INITIATES A VARIABLE-SHIFT ITERATION AND RETURNS WITH THE
    // APPROXIMATE ZERO IF SUCCESSFUL.
    // L2 - LIMIT OF FIXED SHIFT STEPS
    // ZR,ZI - APPROXIMATE ZERO IF CONV IS .TRUE.
    // CONV  - LOGICAL INDICATING CONVERGENCE OF STAGE 3 ITERATION
    //
    static bool fxshft(const int l2, int deg, xcomplex *P, xcomplex *p, xcomplex *H, xcomplex *h, xcomplex *zero, xcomplex *s){
       bool bol, conv;                 // boolean for convergence of stage 2
       bool test, pasd;
       xcomplex old_T, old_S, Ps, t;
       xcomplex *Tmp = new xcomplex[deg+1];

       Ps = polyev(deg, *s, P, p);
       test = true;
       pasd = false;

       // Calculate first T = -P(S)/H(S)
       t = calct(&bol, deg, Ps, H, h, *s);

       // Main loop for second stage
       for(int j = 1; j <= l2; j++){
          old_T = t;

          // Compute the next H Polynomial and new t
          nexth(bol, deg, t, H, h, p);
          t = calct(&bol, deg, Ps, H, h, *s);
          *zero = *s + t;

          // Test for convergence unless stage 3 has failed once or this
          // is the last H Polynomial
          if(!(bol || !test || j == l2)){
             if(xabs(t - old_T) < 0.5 * xabs(*zero)) {
                if(pasd) {
                   // The weak convergence test has been passwed twice, start the third stage
                   // Iteration, after saving the current H polynomial and shift
                   for(int i = 0; i < deg; i++)
                      Tmp[i] = H[i];
                   old_S = *s;

                   conv = vrshft(10, deg, P, p, H, h, zero, s);
                   if(conv) return conv;

                   //The iteration failed to converge. Turn off testing and restore h,s,pv and T
                   test = false;
                   for(int i = 0; i < deg; i++)
                      H[i] = Tmp[i];
                   *s = old_S;

                   Ps = polyev(deg, *s, P, p);
                   t = calct(&bol, deg, Ps, H, h, *s);
                   continue;
                   }
                pasd = true;
                }
             else
                pasd = false;
          }
       }

       // Attempt an iteration with final H polynomial from second stage
       conv = vrshft(10, deg, P, p, H, h, zero, s);

       delete Tmp;

       return conv;
    }

    // Main function
    //
    int cpoly(int degree, const xcomplex poly[], xcomplex Roots[])
    {
      xcomplex PhiDiff = xcomplex(-0.069756473, 0.99756405);
      xcomplex PhiRand = xcomplex(1.0, -1.0) / sqrt(2.0);
      xcomplex *P = new xcomplex[degree + 1], *H = new xcomplex[degree + 1];
      xcomplex *h = new xcomplex[degree + 1], *p = new xcomplex[degree + 1];
      xcomplex zero, s, bnd;
      unsigned int conv = 0;

      while(poly[0] == xdata.ZERO) {
        poly++;
        degree--;
        if (degree < 0)
          return -1;
      };

      int deg = degree;

      // Remove the zeros at the origin if any
      while(poly[deg] == xdata.ZERO){
        Roots[degree - deg] = xdata.ZERO;
        deg--;
      }

      if (deg == 0) return degree;

      // Make a copy of the coefficients
      for(int i = 0; i <= deg; i++) { P[i] = poly[i]; }

      scale(deg, P);

     search:

      if(deg <= 1){
        Roots[degree-1] = - P[1] / P[0];
        return degree;
      }

      // compute a bound of the moduli of the roots (Newton-Raphson)
      bnd = cauchy(deg, P);

      // Outer loop to control 2 Major passes with different sequences of shifts
      for(int cnt1 = 1; cnt1 <= 2; cnt1++) {
        // First stage  calculation , no shift
        noshft(5, deg, P, H);

        // Inner loop to select a shift
        for(int cnt2 = 1; cnt2 <= 9; cnt2++) {
          // Shift is chosen with modulus bnd and amplitude rotated by 94 degree from the previous shif
          PhiRand = PhiDiff * PhiRand;
          s = bnd * PhiRand;

          // Second stage calculation, fixed shift
          conv = fxshft(10 * cnt2, deg, P, p, H, h, &zero, &s);
          if(conv) {
      // The second stage jumps directly to the third stage iteration
      // If successful the zero is stored and the polynomial deflated
      Roots[degree - deg] = zero;

      // continue with the remaining polynomial
      deg--;
      for(int i = 0; i <= deg; i++){ P[i] = p[i]; };
      goto search;
          }
          // if the iteration is unsuccessful another shift is chosen
        }

        // if 9 shifts fail, the outer loop is repeated with another sequence of shifts
      }

      delete P;
      delete H;
      delete h;
      delete p;

      // The zerofinder has failed on two major passes
      // return empty handed with the number of roots found (less than the original degree)
      return degree - deg;
    }

    // Wrapper function:
    // This function calls the one made by Laurent Bartholdi
    int solve_poly(int degree, const double *in_real, const double *in_imag, double *out_real, double *out_imag)
    {
      int nroots;
      xcomplex *poly  = new xcomplex[degree+1];
      xcomplex *roots = new xcomplex[degree+1];

      for (int i = 0; i <= degree; ++i) {
        poly[i].real(in_real[i]);
        poly[i].imag(in_imag[i]);
      }

      nroots = cpoly(degree, poly, roots);

      for (int i = 0; i < nroots; ++i) {
        out_real[i] = roots[i].real();
        out_imag[i] = roots[i].imag();
      }

      delete[] poly;
      delete[] roots;

      return nroots;
    }
  }
}

#endif