home *** CD-ROM | disk | FTP | other *** search
- /*
- * Copyright (c) 1985 Regents of the University of California.
- * All rights reserved. The Berkeley software License Agreement
- * specifies the terms and conditions for redistribution.
- */
-
- #ifndef lint
- static char sccsid[] = "@(#)lgamma.c 5.2 (Berkeley) 4/29/88";
- #endif /* not lint */
-
- /*
- C program for floating point log Gamma function
-
- lgamma(x) computes the log of the absolute
- value of the Gamma function.
- The sign of the Gamma function is returned in the
- external quantity signgam.
-
- The coefficients for expansion around zero
- are #5243 from Hart & Cheney; for expansion
- around infinity they are #5404.
-
- Calls log, floor and sin.
- */
-
- #include <math.h>
- #if defined(vax)||defined(tahoe)
- #include <errno.h>
- #endif /* defined(vax)||defined(tahoe) */
- int signgam = 0;
- static double goobie = 0.9189385332046727417803297; /* log(2*pi)/2 */
- static double pi = 3.1415926535897932384626434;
-
- #define M 6
- #define N 8
- static double p1[] = {
- 0.83333333333333101837e-1,
- -.277777777735865004e-2,
- 0.793650576493454e-3,
- -.5951896861197e-3,
- 0.83645878922e-3,
- -.1633436431e-2,
- };
- static double p2[] = {
- -.42353689509744089647e5,
- -.20886861789269887364e5,
- -.87627102978521489560e4,
- -.20085274013072791214e4,
- -.43933044406002567613e3,
- -.50108693752970953015e2,
- -.67449507245925289918e1,
- 0.0,
- };
- static double q2[] = {
- -.42353689509744090010e5,
- -.29803853309256649932e4,
- 0.99403074150827709015e4,
- -.15286072737795220248e4,
- -.49902852662143904834e3,
- 0.18949823415702801641e3,
- -.23081551524580124562e2,
- 0.10000000000000000000e1,
- };
-
- double
- lgamma(arg)
- double arg;
- {
- double log(), pos(), neg(), asym();
-
- signgam = 1.;
- if(arg <= 0.) return(neg(arg));
- if(arg > 8.) return(asym(arg));
- return(log(pos(arg)));
- }
-
- static double
- asym(arg)
- double arg;
- {
- double log();
- double n, argsq;
- int i;
-
- argsq = 1./(arg*arg);
- for(n=0,i=M-1; i>=0; i--){
- n = n*argsq + p1[i];
- }
- return((arg-.5)*log(arg) - arg + goobie + n/arg);
- }
-
- static double
- neg(arg)
- double arg;
- {
- double t;
- double log(), sin(), floor(), pos();
-
- arg = -arg;
- /*
- * to see if arg were a true integer, the old code used the
- * mathematically correct observation:
- * sin(n*pi) = 0 <=> n is an integer.
- * but in finite precision arithmetic, sin(n*PI) will NEVER
- * be zero simply because n*PI is a rational number. hence
- * it failed to work with our newer, more accurate sin()
- * which uses true pi to do the argument reduction...
- * temp = sin(pi*arg);
- */
- t = floor(arg);
- if (arg - t > 0.5e0)
- t += 1.e0; /* t := integer nearest arg */
- #if defined(vax)||defined(tahoe)
- if (arg == t) {
- extern double infnan();
- return(infnan(ERANGE)); /* +INF */
- }
- #endif /* defined(vax)||defined(tahoe) */
- signgam = (int) (t - 2*floor(t/2)); /* signgam = 1 if t was odd, */
- /* 0 if t was even */
- signgam = signgam - 1 + signgam; /* signgam = 1 if t was odd, */
- /* -1 if t was even */
- t = arg - t; /* -0.5 <= t <= 0.5 */
- if (t < 0.e0) {
- t = -t;
- signgam = -signgam;
- }
- return(-log(arg*pos(arg)*sin(pi*t)/pi));
- }
-
- static double
- pos(arg)
- double arg;
- {
- double n, d, s;
- register i;
-
- if(arg < 2.) return(pos(arg+1.)/arg);
- if(arg > 3.) return((arg-1.)*pos(arg-1.));
-
- s = arg - 2.;
- for(n=0,d=0,i=N-1; i>=0; i--){
- n = n*s + p2[i];
- d = d*s + q2[i];
- }
- return(n/d);
- }
-