shithub: riscv

ref: 8c6c2b7216b6039188e76d26f9a62c4cb8ee4fca
dir: /sys/src/ape/lib/ap/math/j0.c/

View raw version
#include <math.h>
#include <errno.h>
/*
	floating point Bessel's function
	of the first and second kinds
	of order zero

	j0(x) returns the value of J0(x)
	for all real values of x.

	There are no error returns.
	Calls sin, cos, sqrt.

	There is a niggling bug in J0 which
	causes errors up to 2e-16 for x in the
	interval [-8,8].
	The bug is caused by an inappropriate order
	of summation of the series.  rhm will fix it
	someday.

	Coefficients are from Hart & Cheney.
	#5849 (19.22D)
	#6549 (19.25D)
	#6949 (19.41D)

	y0(x) returns the value of Y0(x)
	for positive real values of x.
	For x<=0, error number EDOM is set and a
	large negative value is returned.

	Calls sin, cos, sqrt, log, j0.

	The values of Y0 have not been checked
	to more than ten places.

	Coefficients are from Hart & Cheney.
	#6245 (18.78D)
	#6549 (19.25D)
	#6949 (19.41D)
*/

static double pzero, qzero;
static double tpi	= .6366197723675813430755350535e0;
static double pio4	= .7853981633974483096156608458e0;
static double p1[] = {
	0.4933787251794133561816813446e21,
	-.1179157629107610536038440800e21,
	0.6382059341072356562289432465e19,
	-.1367620353088171386865416609e18,
	0.1434354939140344111664316553e16,
	-.8085222034853793871199468171e13,
	0.2507158285536881945555156435e11,
	-.4050412371833132706360663322e8,
	0.2685786856980014981415848441e5,
};
static double q1[] = {
	0.4933787251794133562113278438e21,
	0.5428918384092285160200195092e19,
	0.3024635616709462698627330784e17,
	0.1127756739679798507056031594e15,
	0.3123043114941213172572469442e12,
	0.6699987672982239671814028660e9,
	0.1114636098462985378182402543e7,
	0.1363063652328970604442810507e4,
	1.0
};
static double p2[] = {
	0.5393485083869438325262122897e7,
	0.1233238476817638145232406055e8,
	0.8413041456550439208464315611e7,
	0.2016135283049983642487182349e7,
	0.1539826532623911470917825993e6,
	0.2485271928957404011288128951e4,
	0.0,
};
static double q2[] = {
	0.5393485083869438325560444960e7,
	0.1233831022786324960844856182e8,
	0.8426449050629797331554404810e7,
	0.2025066801570134013891035236e7,
	0.1560017276940030940592769933e6,
	0.2615700736920839685159081813e4,
	1.0,
};
static double p3[] = {
	-.3984617357595222463506790588e4,
	-.1038141698748464093880530341e5,
	-.8239066313485606568803548860e4,
	-.2365956170779108192723612816e4,
	-.2262630641933704113967255053e3,
	-.4887199395841261531199129300e1,
	0.0,
};
static double q3[] = {
	0.2550155108860942382983170882e6,
	0.6667454239319826986004038103e6,
	0.5332913634216897168722255057e6,
	0.1560213206679291652539287109e6,
	0.1570489191515395519392882766e5,
	0.4087714673983499223402830260e3,
	1.0,
};
static double p4[] = {
	-.2750286678629109583701933175e20,
	0.6587473275719554925999402049e20,
	-.5247065581112764941297350814e19,
	0.1375624316399344078571335453e18,
	-.1648605817185729473122082537e16,
	0.1025520859686394284509167421e14,
	-.3436371222979040378171030138e11,
	0.5915213465686889654273830069e8,
	-.4137035497933148554125235152e5,
};
static double q4[] = {
	0.3726458838986165881989980e21,
	0.4192417043410839973904769661e19,
	0.2392883043499781857439356652e17,
	0.9162038034075185262489147968e14,
	0.2613065755041081249568482092e12,
	0.5795122640700729537480087915e9,
	0.1001702641288906265666651753e7,
	0.1282452772478993804176329391e4,
	1.0,
};

static void
asympt(double arg)
{
	double zsq, n, d;
	int i;

	zsq = 64 / (arg*arg);
	for(n=0,d=0,i=6;i>=0;i--) {
		n = n*zsq + p2[i];
		d = d*zsq + q2[i];
	}
	pzero = n/d;
	for(n=0,d=0,i=6;i>=0;i--) {
		n = n*zsq + p3[i];
		d = d*zsq + q3[i];
	}
	qzero = (8/arg)*(n/d);
}

double
j0(double arg)
{
	double argsq, n, d;
	int i;

	if(arg < 0)
		arg = -arg;
	if(arg > 8) {
		asympt(arg);
		n = arg - pio4;
		return sqrt(tpi/arg)*(pzero*cos(n) - qzero*sin(n));
	}
	argsq = arg*arg;
	for(n=0,d=0,i=8;i>=0;i--) {
		n = n*argsq + p1[i];
		d = d*argsq + q1[i];
	}
	return n/d;
}

double
y0(double arg)
{
	double argsq, n, d;
	int i;

	errno = 0;
	if(arg <= 0) {
		errno = EDOM;
		return(-HUGE_VAL);
	}
	if(arg > 8) {
		asympt(arg);
		n = arg - pio4;
		return sqrt(tpi/arg)*(pzero*sin(n) + qzero*cos(n));
	}
	argsq = arg*arg;
	for(n=0,d=0,i=8;i>=0;i--) {
		n = n*argsq + p4[i];
		d = d*argsq + q4[i];
	}
	return n/d + tpi*j0(arg)*log(arg);
}