/* gsl/cdf_gamma.c * * Copyright (C) 1996, 1997, 1998, 1999, 2000, 2002 Gerard Jungman and Jason Stover. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or (at * your option) any later version. * * This program is distributed in the hope that it will be useful, but * WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ /* * Author: G. Jungman * modified for gsl/cdf by J. Stover. */ #include #include #include #include #include #include "erf.h" #include "exp.h" #include "log.h" #include "gamma.h" #include "error.h" /* * Wrappers for the functions that do the work. */ int gsl_cdf_gamma_e ( double x, double alpha, double beta, gsl_cdf_tail_t tail, gsl_cdf_result *result) { int rc = 0; if ( tail == GSL_CDF_UPPER ) { rc = gamma_inc_P_e ( alpha, (x/beta), result); } else if (tail == GSL_CDF_LOWER ) { rc = gamma_inc_Q_e ( alpha, (x/beta), result); } return rc; } /* The dominant part, * D(a,x) := x^a e^(-x) / Gamma(a+1) */ static int gamma_inc_D(const double a, const double x, gsl_cdf_result * result) { if(a < 10.0) { double lnr; gsl_cdf_result lg; gsl_cdf_lngamma_e(a+1.0, &lg); lnr = a * log(x) - x - lg.val; result->val = exp(lnr); result->err = 2.0 * GSL_DBL_EPSILON * (fabs(lnr) + 1.0) * fabs(result->val); return GSL_SUCCESS; } else { double mu = (x-a)/a; double term1; gsl_cdf_result gstar; gsl_cdf_result ln_term; gsl_cdf_log_1plusx_mx_e(mu, &ln_term); /* log(1+mu) - mu */ gsl_cdf_gammastar_e(a, &gstar); term1 = exp(a*ln_term.val)/sqrt(2.0*M_PI*a); result->val = term1/gstar.val; result->err = 2.0 * GSL_DBL_EPSILON * (fabs(a*ln_term.val) + 1.0) * fabs(result->val); result->err += gstar.err/fabs(gstar.val) * fabs(result->val); return GSL_SUCCESS; } } /* P series representation. */ static int gamma_inc_P_series(const double a, const double x, gsl_cdf_result * result) { const int nmax = 5000; gsl_cdf_result D; int stat_D = gamma_inc_D(a, x, &D); double sum = 1.0; double term = 1.0; int n; for(n=1; nval = D.val * sum; result->err = D.err * fabs(sum); result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); if(n == nmax) GSL_ERROR ("error", GSL_EMAXITER); else return stat_D; } /* Q large x asymptotic */ static int gamma_inc_Q_large_x(const double a, const double x, gsl_cdf_result * result) { const int nmax = 5000; gsl_cdf_result D; const int stat_D = gamma_inc_D(a, x, &D); double sum = 1.0; double term = 1.0; double last = 1.0; int n; for(n=1; n 1.0) break; if(fabs(term/sum) < GSL_DBL_EPSILON) break; sum += term; last = term; } result->val = D.val * (a/x) * sum; result->err = D.err * fabs((a/x) * sum); result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); if(n == nmax) GSL_ERROR ("error", GSL_EMAXITER); else return stat_D; } /* Uniform asymptotic for x near a, a and x large. * See [Temme, p. 285] * FIXME: need c1 coefficient */ static int gamma_inc_Q_asymp_unif(const double a, const double x, gsl_cdf_result * result) { const double rta = sqrt(a); const double eps = (x-a)/a; gsl_cdf_result ln_term; const int stat_ln = gsl_cdf_log_1plusx_mx_e(eps, &ln_term); /* log(1+eps) - eps */ const double eta = eps * sqrt(-2.0*ln_term.val/(eps*eps)); gsl_cdf_result erfc; double R; double c0, c1; gsl_cdf_erfc_e(eta*M_SQRT2*rta, &erfc); if(fabs(eps) < GSL_ROOT5_DBL_EPSILON) { c0 = -1.0/3.0 + eps*(1.0/12.0 - eps*(23.0/540.0 - eps*(353.0/12960.0 - eps*589.0/30240.0))); c1 = 0.0; } else { double rt_term; rt_term = sqrt(-2.0 * ln_term.val/(eps*eps)); c0 = (1.0 - 1.0/rt_term)/eps; c1 = 0.0; } R = exp(-0.5*a*eta*eta)/(M_SQRT2*M_SQRTPI*rta) * (c0 + c1/a); result->val = 0.5 * erfc.val + R; result->err = GSL_DBL_EPSILON * fabs(R * 0.5 * a*eta*eta) + 0.5 * erfc.err; result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); return stat_ln; } /* Continued fraction for Q. * * Q(a,x) = D(a,x) a/x F(a,x) * 1 (1-a)/x 1/x (2-a)/x 2/x (3-a)/x * F(a,x) = ---- ------- ----- -------- ----- -------- ... * 1 + 1 + 1 + 1 + 1 + 1 + * * Hans E. Plesser, 2002-01-22 (hans dot plesser at itf dot nlh dot no): * * Since the Gautschi equivalent series method for CF evaluation may lead * to singularities, I have replaced it with the modified Lentz algorithm * given in * * I J Thompson and A R Barnett * Coulomb and Bessel Functions of Complex Arguments and Order * J Computational Physics 64:490-509 (1986) * * In consequence, gamma_inc_Q_CF_protected() is now obsolete and has been * removed. * * Identification of terms between the above equation for F(a, x) and * the first equation in the appendix of Thompson&Barnett is as follows: * * b_0 = 0, b_n = 1 for all n > 0 * * a_1 = 1 * a_n = (n/2-a)/x for n even * a_n = (n-1)/(2x) for n odd * */ static int gamma_inc_Q_CF(const double a, const double x, gsl_cdf_result * result) { const int nmax = 5000; const double small = gsl_pow_3 (GSL_DBL_EPSILON); gsl_cdf_result D; const int stat_D = gamma_inc_D(a, x, &D); double hn = 1.0; /* convergent */ double Cn = 1.0 / small; double Dn = 1.0; int n; /* n == 1 has a_1, b_1, b_0 independent of a,x, so that has been done by hand */ for ( n = 2 ; n < nmax ; n++ ) { double an; double delta; if(GSL_IS_ODD(n)) an = 0.5*(n-1)/x; else an = (0.5*n-a)/x; Dn = 1.0 + an * Dn; if ( fabs(Dn) < small ) Dn = small; Cn = 1.0 + an/Cn; if ( fabs(Cn) < small ) Cn = small; Dn = 1.0 / Dn; delta = Cn * Dn; hn *= delta; if(fabs(delta-1) < GSL_DBL_EPSILON) break; } result->val = D.val * (a/x) * hn; result->err = D.err * fabs((a/x) * hn); result->err += GSL_DBL_EPSILON * (2.0 + 0.5*n) * fabs(result->val); if(n == nmax) GSL_ERROR ("error", GSL_EMAXITER); else return stat_D; } /* Useful for small a and x. Handles the subtraction analytically. */ static int gamma_inc_Q_series(const double a, const double x, gsl_cdf_result * result) { double term1; /* 1 - x^a/Gamma(a+1) */ double sum; /* 1 + (a+1)/(a+2)(-x)/2! + (a+1)/(a+3)(-x)^2/3! + ... */ int stat_sum; double term2; /* a temporary variable used at the end */ { /* Evaluate series for 1 - x^a/Gamma(a+1), small a */ const double pg21 = -2.404113806319188570799476; /* PolyGamma[2,1] */ const double lnx = log(x); const double el = M_EULER+lnx; const double c1 = -el; const double c2 = M_PI*M_PI/12.0 - 0.5*el*el; const double c3 = el*(M_PI*M_PI/12.0 - el*el/6.0) + pg21/6.0; const double c4 = -0.04166666666666666667 * (-1.758243446661483480 + lnx) * (-0.764428657272716373 + lnx) * ( 0.723980571623507657 + lnx) * ( 4.107554191916823640 + lnx); const double c5 = -0.0083333333333333333 * (-2.06563396085715900 + lnx) * (-1.28459889470864700 + lnx) * (-0.27583535756454143 + lnx) * ( 1.33677371336239618 + lnx) * ( 5.17537282427561550 + lnx); const double c6 = -0.0013888888888888889 * (-2.30814336454783200 + lnx) * (-1.65846557706987300 + lnx) * (-0.88768082560020400 + lnx) * ( 0.17043847751371778 + lnx) * ( 1.92135970115863890 + lnx) * ( 6.22578557795474900 + lnx); const double c7 = -0.00019841269841269841 * (-2.5078657901291800 + lnx) * (-1.9478900888958200 + lnx) * (-1.3194837322612730 + lnx) * (-0.5281322700249279 + lnx) * ( 0.5913834939078759 + lnx) * ( 2.4876819633378140 + lnx) * ( 7.2648160783762400 + lnx); const double c8 = -0.00002480158730158730 * (-2.677341544966400 + lnx) * (-2.182810448271700 + lnx) * (-1.649350342277400 + lnx) * (-1.014099048290790 + lnx) * (-0.191366955370652 + lnx) * ( 0.995403817918724 + lnx) * ( 3.041323283529310 + lnx) * ( 8.295966556941250 + lnx); const double c9 = -2.75573192239859e-6 * (-2.8243487670469080 + lnx) * (-2.3798494322701120 + lnx) * (-1.9143674728689960 + lnx) * (-1.3814529102920370 + lnx) * (-0.7294312810261694 + lnx) * ( 0.1299079285269565 + lnx) * ( 1.3873333251885240 + lnx) * ( 3.5857258865210760 + lnx) * ( 9.3214237073814600 + lnx); const double c10 = -2.75573192239859e-7 * (-2.9540329644556910 + lnx) * (-2.5491366926991850 + lnx) * (-2.1348279229279880 + lnx) * (-1.6741881076349450 + lnx) * (-1.1325949616098420 + lnx) * (-0.4590034650618494 + lnx) * ( 0.4399352987435699 + lnx) * ( 1.7702236517651670 + lnx) * ( 4.1231539047474080 + lnx) * ( 10.342627908148680 + lnx); term1 = a*(c1+a*(c2+a*(c3+a*(c4+a*(c5+a*(c6+a*(c7+a*(c8+a*(c9+a*c10))))))))); } { /* Evaluate the sum. */ const int nmax = 5000; double t = 1.0; int n; sum = 1.0; for(n=1; nval = term1 + term2; result->err = GSL_DBL_EPSILON * (fabs(term1) + 2.0*fabs(term2)); result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); return stat_sum; } /*-*-*-*-*-*-*-*-*-*-*-* Functions with Error Codes *-*-*-*-*-*-*-*-*-*-*-*/ /* * Normalized Incomplete Gamma Function * * Q(a,x) = 1/Gamma(a) Integral[ t^(a-1) e^(-t), {t,x,Infinity} ] * * a > 0, x >= 0 * * exceptions: GSL_EDOM */ static int gamma_inc_Q_e(const double a, const double x, gsl_cdf_result * result) { if(a <= 0.0) { DOMAIN_ERROR(result); } else if(x <= 0.0) { result->val = 1.0; result->err = 0.0; return GSL_SUCCESS; } else if(x <= 0.5*a) { /* If the series is quick, do that. It is * robust and simple. */ gsl_cdf_result P; int stat_P = gamma_inc_P_series(a, x, &P); result->val = 1.0 - P.val; result->err = P.err; result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); return stat_P; } else if(a >= 1.0e+06 && (x-a)*(x-a) < a) { /* Then try the difficult asymptotic regime. * This is the only way to do this region. */ return gamma_inc_Q_asymp_unif(a, x, result); } else if(a < 0.2 && x < 5.0) { /* Cancellations at small a must be handled * analytically; x should not be too big * either since the series terms grow * with x and log(x). */ return gamma_inc_Q_series(a, x, result); } else if(a <= x) { if(x <= 1.0e+06) { /* Continued fraction is excellent for x >~ a. * We do not let x be too large when x > a since * it is somewhat pointless to try this there; * the function is rapidly decreasing for * x large and x > a, and it will just * underflow in that region anyway. We * catch that case in the standard * large-x method. */ return gamma_inc_Q_CF(a, x, result); } else { return gamma_inc_Q_large_x(a, x, result); } } else { if(a < 0.8*x) { /* Continued fraction again. The convergence * is a little slower here, but that is fine. * We have to trade that off against the slow * convergence of the series, which is the * only other option. */ return gamma_inc_Q_CF(a, x, result); } else { gsl_cdf_result P; int stat_P = gamma_inc_P_series(a, x, &P); result->val = 1.0 - P.val; result->err = P.err; result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); return stat_P; } } } /* * Complementary Normalized Incomplete Gamma Function * * P(a,x) = 1/Gamma(a) Integral[ t^(a-1) e^(-t), {t,0,x} ] * * a > 0, x >= 0 * * exceptions: GSL_EDOM */ static int gamma_inc_P_e(const double a, const double x, gsl_cdf_result * result) { if(a <= 0.0 ) { DOMAIN_ERROR(result); } else if(x <= 0.0) { result->val = 0.0; result->err = 0.0; return GSL_SUCCESS; } else if(x < 20.0 || x < 0.5*a) { /* Do the easy series cases. Robust and quick. */ return gamma_inc_P_series(a, x, result); } else if(a > 1.0e+06 && (x-a)*(x-a) < a) { /* Crossover region. Note that Q and P are * roughly the same order of magnitude here, * so the subtraction is stable. */ gsl_cdf_result Q; int stat_Q = gamma_inc_Q_asymp_unif(a, x, &Q); result->val = 1.0 - Q.val; result->err = Q.err; result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); return stat_Q; } else if(a <= x) { /* Q <~ P in this area, so the * subtractions are stable. */ gsl_cdf_result Q; int stat_Q; if(a > 0.2*x) { stat_Q = gamma_inc_Q_CF(a, x, &Q); } else { stat_Q = gamma_inc_Q_large_x(a, x, &Q); } result->val = 1.0 - Q.val; result->err = Q.err; result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); return stat_Q; } else { if((x-a)*(x-a) < a) { /* This condition is meant to insure * that Q is not very close to 1, * so the subtraction is stable. */ gsl_cdf_result Q; int stat_Q = gamma_inc_Q_CF(a, x, &Q); result->val = 1.0 - Q.val; result->err = Q.err; result->err += 2.0 * GSL_DBL_EPSILON * fabs(result->val); return stat_Q; } else { return gamma_inc_P_series(a, x, result); } } } /*-*-*-*-*-*-*-*-*-* Functions w/ Natural Prototypes *-*-*-*-*-*-*-*-*-*-*/ #include "eval.h" double gsl_cdf_gamma_inc_P(const double a, const double x) { EVAL_RESULT(gamma_inc_P_e(a, x, &result)); } double gsl_cdf_gamma_inc_Q(const double a, const double x) { EVAL_RESULT(gamma_inc_Q_e(a, x, &result)); }