Mercurial > projects > ldc
view tango/tango/math/Bessel.d @ 293:ebfa488f4abc trunk
[svn r314] Get correct value type for newing of multidimensional dynamic arrays.
Fixes array_initialization_26_E.
author | ChristianK |
---|---|
date | Sun, 22 Jun 2008 15:21:34 +0200 |
parents | 1700239cab2e |
children |
line wrap: on
line source
/** * Cylindrical Bessel functions of integral order. * * Copyright: Based on the CEPHES math library, which is * Copyright (C) 1994 Stephen L. Moshier (moshier@world.std.com). * License: BSD style: $(LICENSE) * Authors: Stephen L. Moshier (original C code). Conversion to D by Don Clugston */ module tango.math.Bessel; import tango.math.Math; private import tango.math.IEEE; private { // Rational polynomial approximations to j0, y0, j1, y1. // sqrt(j0^2(1/x^2) + y0^2(1/x^2)) = z P(z**2)/Q(z**2), z(x) = 1/sqrt(x) // Peak error = 1.80e-20 const real j0modulusn[] = [ 0x1.154700ea96e79656p-7, 0x1.72244b6e998cd6fp-4, 0x1.6ebccf42e9c19fd2p-1, 0x1.6bd844e89cbd639ap+1, 0x1.e812b377c75ebc96p+2, 0x1.46d69ca24ce76686p+3, 0x1.b756f7234cc67146p+2, 0x1.943a7471eaa50ab2p-2 ]; const real j0modulusd[] = [ 0x1.5b84007c37011506p-7, 0x1.cfe76758639bdab4p-4, 0x1.cbfa09bf71bafc7ep-1, 0x1.c8eafb3836f2eeb4p+1, 0x1.339db78060eb706ep+3, 0x1.a06530916be8bc7ap+3, 0x1.23bfe7f67a54893p+3, 1.0 ]; // atan(y0(x)/j0(x)) = x - pi/4 + x P(x**2)/Q(x**2) // Peak error = 2.80e-21. Relative error spread = 5.5e-1 const real j0phasen[] = [ -0x1.ccbaf3865bb0985ep-22, -0x1.3a6b175e64bdb82ep-14, -0x1.06124b5310cdca28p-8, -0x1.3cebb7ab09cf1b14p-4, -0x1.00156ed209b43c6p-1, -0x1.78aa9ba4254ca20cp-1 ]; const real j0phased[] = [ 0x1.ccbaf3865bb09918p-19, 0x1.3b5b0e12900d58b8p-11, 0x1.0897373ff9906f7ep-5, 0x1.450a5b8c552ade4ap-1, 0x1.123e263e7f0e96d2p+2, 0x1.d82ecca5654be7d2p+2, 1.0 ]; // j1(x) = (x^2-r0^2)(x^2-r1^2)(x^2-r2^2) x P(x**2)/Q(x**2), 0 <= x <= 9 // Peak error = 2e-21 const real j1n[] = [ -0x1.2f494fa4e623b1bp+58, 0x1.8289f0a5f1e1a784p+52, -0x1.9d773ee29a52c6d8p+45, 0x1.e9394ff57a46071cp+37, -0x1.616c7939904a359p+29, 0x1.424414b9ee5671eap+20, -0x1.6db34a9892d653e6p+10, 0x1.dcd7412d90a0db86p-1, -0x1.1444a1643199ee5ep-12 ]; const real j1d[] = [ 0x1.5a1e0a45eb67bacep+75, 0x1.35ee485d62f0ccbap+68, 0x1.11ee7aad4e4bcd8p+60, 0x1.3adde5dead800244p+51, 0x1.041c413dfbab693p+42, 0x1.4066d12193fcc082p+32, 0x1.24309d0dc2c4d42ep+22, 0x1.7115bea028dd75f2p+11, 1.0 ]; // sqrt(j1^2(1/x^2) + y1^2(1/x^2)) = z P(z**2)/Q(z**2), z(x) = 1/sqrt(x) // Peak error = 1.35e=20, Relative error spread = 9.9e0 const real [] j1modulusn = [ 0x1.059262020bf7520ap-6, 0x1.012ffc4d1f5cdbc8p-3, 0x1.03c17ce18cae596p+0, 0x1.6e0414a7114ae3ccp+1, 0x1.cb047410d229cbc4p+2, 0x1.4385d04bb718faaap+1, 0x1.914074c30c746222p-2, -0x1.42abe77f6b307aa6p+2 ]; const real [] j1modulusd = [ 0x1.47d4e6ad98d8246ep-6, 0x1.42562f48058ff904p-3, 0x1.44985e2af35c6f9cp+0, 0x1.c6f4a03469c4ef6cp+1, 0x1.1829a060e8d604cp+3, 0x1.44111c892f9cc84p+1, -0x1.d7c36d7f1e5aef6ap-1, -0x1.8eeafb1ac81c4c06p+2, 1.0 ]; // atan(y1(x)/j1(x)) = x - 3pi/4 + z P(z**2)/Q(z**2), z(x) = 1/x // Peak error = 4.83e-21. Relative error spread = 1.9e0 const real [] j1phasen = [ 0x1.ca9f612d83aaa818p-20, 0x1.2e82fcfb7d0fee9ep-12, 0x1.e28858c1e947506p-7, 0x1.12b8f96e5173d20ep-2, 0x1.965e6a013154c0ap+0, 0x1.0156a25eaa0dd78p+1 ]; const real [] j1phased = [ 0x1.31bf961e57c71ae4p-18, 0x1.9464d8f2abf750a6p-11, 0x1.446a786bac2131fp-5, 0x1.76caa8513919873cp-1, 0x1.2130b56bc1a563e4p+2, 0x1.b3cc1a865259dfc6p+2, 0x1p+0 ]; } /*** * Bessel function of order zero * * Returns Bessel function of first kind, order zero of the argument. */ /* The domain is divided into the intervals [0, 9] and * (9, infinity). In the first interval the rational approximation * is (x^2 - r^2) (x^2 - s^2) (x^2 - t^2) P7(x^2) / Q8(x^2), * where r, s, t are the first three zeros of the function. * In the second interval the expansion is in terms of the * modulus M0(x) = sqrt(J0(x)^2 + Y0(x)^2) and phase P0(x) * = atan(Y0(x)/J0(x)). M0 is approximated by sqrt(1/x)P7(1/x)/Q7(1/x). * The approximation to J0 is M0 * cos(x - pi/4 + 1/x P5(1/x^2)/Q6(1/x^2)). */ real cylBessel_j0(real x) { // j0(x) = (x^2-JZ1)(x^2-JZ2)(x^2-JZ3)P(x**2)/Q(x**2), 0 <= x <= 9 // Peak error = 8.49e-22. Relative error spread = 2.2e-3 const real j0n[] = [ -0x1.3e8ff72b890d72d8p+59, 0x1.cc86e3755a4c803p+53, -0x1.0ea6f5bac6623616p+47, 0x1.532c6d94d36f2874p+39, -0x1.ef25a232f6c00118p+30, 0x1.aa0690536c11fc2p+21, -0x1.94e67651cc57535p+11, 0x1.4bfe47ac8411eeb2p+0 ]; const real j0d[] = [ 0x1.0096dec5f6560158p+73, 0x1.11705db14995fb9cp+66, 0x1.220a41c3daaa7a58p+58, 0x1.93c6b48d196c1082p+49, 0x1.9814684a10dbfda2p+40, 0x1.36f20ec527fccda4p+31, 0x1.634596b9247fc34p+21, 0x1.1d3eb73f90657bfcp+11, 1.0 ]; real xx, y, z, modulus, phase; xx = x * x; if ( xx < 81.0L ) { const real [] JZ = [5.783185962946784521176L, 30.47126234366208639908L, 7.488700679069518344489e1L]; y = (xx - JZ[0]) * (xx - JZ[1]) * (xx - JZ[2]); y *= rationalPoly( xx, j0n, j0d); return y; } y = fabs(x); xx = 1.0/xx; phase = rationalPoly( xx, j0phasen, j0phased); z = 1.0/y; modulus = rationalPoly( z, j0modulusn, j0modulusd); y = modulus * cos( y - PI_4 + z*phase) / sqrt(y); return y; } /** * Bessel function of the second kind, order zero * Also known as the cylindrical Neumann function, order zero. * * Returns Bessel function of the second kind, of order * zero, of the argument. */ real cylBessel_y0(real x) { /* The domain is divided into the intervals [0, 5>, [5,9> and * [9, infinity). In the first interval a rational approximation * R(x) is employed to compute y0(x) = R(x) + 2/pi * log(x) * j0(x). * * In the second interval, the approximation is * (x - p)(x - q)(x - r)(x - s)P7(x)/Q7(x) * where p, q, r, s are zeros of y0(x). * * The third interval uses the same approximations to modulus * and phase as j0(x), whence y0(x) = modulus * sin(phase). */ // y0(x) = 2/pi * log(x) * j0(x) + P(z**2)/Q(z**2), 0 <= x <= 5 // Peak error = 8.55e-22. Relative error spread = 2.7e-1 const real y0n[] = [ -0x1.068026b402e2bf7ap+54, 0x1.3a2f7be8c4c8a03ep+55, -0x1.89928488d6524792p+51, 0x1.3e3ea2846f756432p+46, -0x1.c8be8d9366867c78p+39, 0x1.43879530964e5fbap+32, -0x1.bee052fef72a5d8p+23, 0x1.e688c8fe417c24d8p+13 ]; const real y0d[] = [ 0x1.bc96c5351e564834p+57, 0x1.6821ac3b4c5209a6p+51, 0x1.27098b571836ce64p+44, 0x1.41870d2a9b90aa76p+36, 0x1.00394fd321f52f48p+28, 0x1.317ce3b16d65b27p+19, 0x1.0432b36efe4b20aep+10, 1.0 ]; // y0(x) = (x-Y0Z1)(x-Y0Z2)(x-Y0Z3)(x-Y0Z4)P(x)/Q(x), 4.5 <= x <= 9 // Peak error = 2.35e-20. Relative error spread = 7.8e-13 const real y059n[] = [ -0x1.0fce17d26a21f218p+19, -0x1.c6fc144765fdfaa8p+16, 0x1.3e20237c53c7180ep+19, 0x1.7d14055ff6a493c4p+17, 0x1.b8b694729689d1f4p+12, -0x1.1e24596784b6c5cp+12, 0x1.35189cb3ece7ab46p+6, 0x1.9428b3f406b4aa08p+4, -0x1.791187b68dd4240ep+0, 0x1.8417216d568b325ep-6 ]; const real y059d[] = [ 0x1.17af71a3d4167676p+30, 0x1.a36abbb668c79d6cp+31, -0x1.4ff64a14ed73c4d6p+29, 0x1.9d427af195244ffep+26, -0x1.4e85bbbc8d2fd914p+23, 0x1.ac59b523ae0bd16cp+19, -0x1.8ebda33eaac74518p+15, 0x1.16194a051cd55a12p+11, -0x1.f2d714ab48d1bd7ep+5, 1.0 ]; real xx, y, z, modulus, phase; if ( x < 0.0 ) return -real.max; xx = x * x; if ( xx < 81.0L ) { if ( xx < 20.25L ) { y = M_2_PI * log(x) * cylBessel_j0(x); y += rationalPoly( xx, y0n, y0d); } else { const real [] Y0Z = [3.957678419314857868376e0L, 7.086051060301772697624e0L, 1.022234504349641701900e1L, 1.336109747387276347827e1L]; y = (x - Y0Z[0])*(x - Y0Z[1])*(x - Y0Z[2])*(x - Y0Z[3]); y *= rationalPoly( x, y059n, y059d); } return y; } y = fabs(x); xx = 1.0/xx; phase = rationalPoly( xx, j0phasen, j0phased); z = 1.0/y; modulus = rationalPoly( z, j0modulusn, j0modulusd); y = modulus * sin( y - PI_4 + z*phase) / sqrt(y); return y; } /** * Bessel function of order one * * Returns Bessel function of order one of the argument. */ real cylBessel_j1(real x) { /* The domain is divided into the intervals [0, 9] and * (9, infinity). In the first interval the rational approximation * is (x^2 - r^2) (x^2 - s^2) (x^2 - t^2) x P8(x^2) / Q8(x^2), * where r, s, t are the first three zeros of the function. * In the second interval the expansion is in terms of the * modulus M1(x) = sqrt(J1(x)^2 + Y1(x)^2) and phase P1(x) * = atan(Y1(x)/J1(x)). M1 is approximated by sqrt(1/x)P7(1/x)/Q8(1/x). * The approximation to j1 is M1 * cos(x - 3 pi/4 + 1/x P5(1/x^2)/Q6(1/x^2)). */ real xx, y, z, modulus, phase; xx = x * x; if ( xx < 81.0L ) { const real [] JZ = [1.46819706421238932572e1L, 4.92184563216946036703e1L, 1.03499453895136580332e2L]; y = (xx - JZ[0]) * (xx - JZ[1]) * (xx - JZ[2]); y *= x * poly( xx, j1n) / poly( xx, j1d); return y; } y = fabs(x); xx = 1.0/xx; phase = rationalPoly( xx, j1phasen, j1phased); z = 1.0/y; modulus = rationalPoly( z, j1modulusn, j1modulusd); const real M_3PI_4 = 3 * PI_4; y = modulus * cos( y - M_3PI_4 + z*phase) / sqrt(y); if( x < 0 ) y = -y; return y; } /** * Bessel function of the second kind, order zero * * Returns Bessel function of the second kind, of order * zero, of the argument. */ real cylBessel_y1(real x) in { assert(x>=0.0); // TODO: should it return -infinity for x<0 ? } body { /* The domain is divided into the intervals [0, 4.5>, [4.5,9> and * [9, infinity). In the first interval a rational approximation * R(x) is employed to compute y0(x) = R(x) + 2/pi * log(x) * j0(x). * * In the second interval, the approximation is * (x - p)(x - q)(x - r)(x - s)P9(x)/Q10(x) * where p, q, r, s are zeros of y1(x). * * The third interval uses the same approximations to modulus * and phase as j1(x), whence y1(x) = modulus * sin(phase). * * ACCURACY: * * Absolute error, when y0(x) < 1; else relative error: * * arithmetic domain # trials peak rms * IEEE 0, 30 36000 2.7e-19 5.3e-20 * */ // y1(x) = 2/pi * (log(x) * j1(x) - 1/x) + R(x^2) z P(z**2)/Q(z**2) // 0 <= x <= 4.5, z(x) = x // Peak error = 7.25e-22. Relative error spread = 4.5e-2 const real [] y1n = [ -0x1.32cab2601090742p+54, 0x1.432ceb7a8eaeff16p+52, -0x1.bcebec5a2484d3fap+47, 0x1.cc58f3cb54d6ac66p+41, -0x1.b1255e154d0eec0ep+34, 0x1.7a337df43298a7c8p+26, -0x1.f77a1afdeff0b62cp+16 ]; const real [] y1d = [ 0x1.8733bcfd7236e604p+56, 0x1.5af412c672fd18d4p+50, 0x1.394ba130685755ep+43, 0x1.7b3321523b24afcp+35, 0x1.52946dac22f61d0cp+27, 0x1.c9040c6053de5318p+18, 0x1.be5156e6771dba34p+9, 1.0 ]; // y1(x) = (x-YZ1)(x-YZ2)(x-YZ3)(x-YZ4)R(x) P(z)/Q(z) // z(x) = x, 4.5 <= x <= 9 // Peak error = 3.27e-22. Relative error spread = 4.5e-2 const real y159n[] = [ 0x1.2fed87b1e60aa736p+18, -0x1.1a2b18cdb2d1ec5ep+20, -0x1.b848827f47b47022p+20, -0x1.b2e422305ea19a86p+20, -0x1.e3f82ac304534676p+16, 0x1.47a2cb5e852d657ep+14, 0x1.81b2fc6e44d7be8p+12, -0x1.cd861d7b090dd22ep+9, 0x1.588897d683cbfbe2p+5, -0x1.5c7feccf76856bcap-1 ]; const real y159d[] = [ 0x1.9b64f2a4d5614462p+26, -0x1.17501e0e38db675ap+30, 0x1.fe88b567c2911c1cp+31, -0x1.86b1781e04e748d4p+29, 0x1.ccd7d4396f2edbcap+26, -0x1.694110c682e5cbcap+23, 0x1.c20f7005b88c789ep+19, -0x1.983a5b4275ab7da8p+15, 0x1.17c60380490fa1fcp+11, -0x1.ee84c254392634d8p+5, 1.0 ]; real xx, y, z, modulus, phase; z = 1.0/x; xx = x * x; if ( xx < 81.0L ) { if ( xx < 20.25L ) { y = M_2_PI * (log(x) * cylBessel_j1(x) - z); y += x * poly( xx, y1n) / poly( xx, y1d); } else { const real [] Y1Z = [ 2.19714132603101703515e0L, 5.42968104079413513277e0L, 8.59600586833116892643e0L, 1.17491548308398812434e1L]; y = (x - Y1Z[0])*(x - Y1Z[1])*(x - Y1Z[2])*(x - Y1Z[3]); y *= rationalPoly( x, y159n, y159d); } return y; } xx = 1.0/xx; phase = rationalPoly( xx, j1phasen, j1phased); modulus = rationalPoly( z, j1modulusn, j1modulusd); const real M_3PI_4 = 3 * PI_4; z = modulus * sin( x - M_3PI_4 + z*phase) / sqrt(x); return z; } /** * Bessel function of integer order * * Returns Bessel function of order n, where n is a * (possibly negative) integer. * * The ratio of jn(x) to j0(x) is computed by backward * recurrence. First the ratio jn/jn-1 is found by a * continued fraction expansion. Then the recurrence * relating successive orders is applied until j0 or j1 is * reached. * * If n = 0 or 1 the routine for j0 or j1 is called * directly. * * BUGS: Not suitable for large n or x. * */ real cylBessel_jn(int n, real x ) { real pkm2, pkm1, pk, xk, r, ans; int k, sign; if ( n < 0 ) { n = -n; if ( (n & 1) == 0 ) /* -1**n */ sign = 1; else sign = -1; } else sign = 1; if ( x < 0.0L ) { if ( n & 1 ) sign = -sign; x = -x; } if ( n == 0 ) return sign * cylBessel_j0(x); if ( n == 1 ) return sign * cylBessel_j1(x); // BUG: This code from Cephes is fast, but it makes the Wronksian test fail. // (accuracy is 8 bits lower). // But, the problem might lie in the n = 2 case in cylBessel_yn(). // if ( n == 2 ) // return sign * (2.0L * cylBessel_j1(x) / x - cylBessel_j0(x)); if ( x < real.epsilon ) return 0; /* continued fraction */ k = 53; pk = 2 * (n + k); ans = pk; xk = x * x; do { pk -= 2.0L; ans = pk - (xk/ans); } while( --k > 0 ); ans = x/ans; /* backward recurrence */ pk = 1.0L; pkm1 = 1.0L/ans; k = n-1; r = 2 * k; do { pkm2 = (pkm1 * r - pk * x) / x; pk = pkm1; pkm1 = pkm2; r -= 2.0L; } while( --k > 0 ); if ( fabs(pk) > fabs(pkm1) ) ans = cylBessel_j1(x)/pk; else ans = cylBessel_j0(x)/pkm1; return sign * ans; } /** * Bessel function of second kind of integer order * * Returns Bessel function of order n, where n is a * (possibly negative) integer. * * The function is evaluated by forward recurrence on * n, starting with values computed by the routines * cylBessel_y0() and cylBessel_y1(). * * If n = 0 or 1 the routine for cylBessel_y0 or cylBessel_y1 is called * directly. */ real cylBessel_yn(int n, real x) in { assert(x>0); // TODO: should it return -infinity for x<=0 ? } body { real an, r; int k, sign; if ( n < 0 ) { n = -n; if ( (n & 1) == 0 ) /* -1**n */ sign = 1; else sign = -1; } else sign = 1; if ( n == 0 ) return sign * cylBessel_y0(x); if ( n == 1 ) return sign * cylBessel_y1(x); /* forward recurrence on n */ real anm2 = cylBessel_y0(x); real anm1 = cylBessel_y1(x); k = 1; r = 2 * k; do { an = r * anm1 / x - anm2; anm2 = anm1; anm1 = an; r += 2.0L; ++k; } while( k < n ); return sign * an; } private { // Evaluate Chebyshev series double evalCheby(double x, double [] poly) { double b0, b1, b2; b0 = poly[$-1]; b1 = 0.0; for (int i=poly.length-1; i>=0; --i) { b2 = b1; b1 = b0; b0 = x * b1 - b2 + poly[i]; } return 0.5*(b0-b2); } } /** * Modified Bessel function of order zero * * Returns modified Bessel function of order zero of the * argument. * * The function is defined as i0(x) = j0( ix ). * * The range is partitioned into the two intervals [0,8] and * (8, infinity). Chebyshev polynomial expansions are employed * in each interval. */ double cylBessel_i0(double x) { // Chebyshev coefficients for exp(-x) I0(x) in the interval [0,8]. // lim(x->0){ exp(-x) I0(x) } = 1. const double [] A = [ 0x1.5a84e9035a22ap-1, -0x1.37febc057cd8dp-2, 0x1.5f7ac77ac88c0p-3, -0x1.84b70342d06eap-4, 0x1.93e8acea8a32dp-5, -0x1.84e9ef121b6f0p-6, 0x1.59961f3dde3ddp-7, -0x1.1b65e201aa849p-8, 0x1.adc758a12100ep-10, -0x1.2e2fd1f15eb52p-11, 0x1.8b51b74107cabp-13, -0x1.e2b2659c41d5ap-15, 0x1.13f58be9a2859p-16, -0x1.2866fcba56427p-18, 0x1.2bf24978cf4acp-20, -0x1.1ec638f227f8dp-22, 0x1.03b769d4d6435p-24, -0x1.beaf68c0b30abp-27, 0x1.6d903a454cb34p-29, -0x1.1d4fe13ae9556p-31, 0x1.a98becc743c10p-34, -0x1.2fc957a946abcp-36, 0x1.9fe2fe19bd324p-39, -0x1.1164c62ee1af0p-41, 0x1.59b464b262627p-44, -0x1.a5022c297fbebp-47, 0x1.ee6d893f65ebap-50, -0x1.184eb721ebbb4p-52, 0x1.33362977da589p-55, -0x1.45cb72134d0efp-58 ]; // Chebyshev coefficients for exp(-x) sqrt(x) I0(x) // in the inverted interval [8,infinity]. // lim(x->inf){ exp(-x) sqrt(x) I0(x) } = 1/sqrt(2pi). const double [] B = [ 0x1.9be62aca809cbp-1, 0x1.b998ca2e59049p-9, 0x1.20fa378999e52p-14, 0x1.8412bc101c586p-19, 0x1.b8007d9cd616ep-23, 0x1.8569280d6d56dp-26, 0x1.d2c64a9225b87p-29, 0x1.0f9ccc0f46f75p-31, 0x1.a24feabe8004fp-37, -0x1.1511d08397425p-35, -0x1.d0fd7357e7bf2p-37, -0x1.f904303178d66p-40, 0x1.94347fa268cecp-41, 0x1.b1c8c6b83c073p-42, 0x1.156ff0d5fc545p-46, -0x1.75d99cf68bb32p-45, -0x1.583fe7e65629ap-47, 0x1.12a919094e6d7p-48, 0x1.fee7da3eafb1fp-50, -0x1.8aee7d908de38p-52, -0x1.4600babd21fe4p-52, 0x1.3f3dd076041cdp-55, 0x1.9be1812d98421p-55, -0x1.646da66119130p-58, -0x1.0adb754ca8b19p-57 ]; double y; if (x < 0) x = -x; if (x <= 8.0) { y = (x/2.0) - 2.0; return exp(x) * evalCheby( y, A); } return exp(x) * evalCheby( 32.0/x - 2.0, B) / sqrt(x); } /** * Modified Bessel function of order one * * Returns modified Bessel function of order one of the * argument. * * The function is defined as i1(x) = -i j1( ix ). * * The range is partitioned into the two intervals [0,8] and * (8, infinity). Chebyshev polynomial expansions are employed * in each interval. */ double cylBessel_i1(double x) { const double [] A = [ 0x1.02a63724a7ffap-2, -0x1.694d10469192ep-3, 0x1.a46dad536f53cp-4, -0x1.b1bbc537c9ebcp-5, 0x1.951e3e7bb2349p-6, -0x1.5a29f7913a26ap-7, 0x1.1065349d3a1b4p-8, -0x1.8cc620b3cd4a4p-10, 0x1.0c95db6c6df7dp-11, -0x1.533cad3d694fep-13, 0x1.911b542c70d0bp-15, -0x1.bd5f9b8debbcfp-17, 0x1.d1c4ed511afc5p-19, -0x1.cc0798363992ap-21, 0x1.ae344b347d108p-23, -0x1.7dd3e24b8c3e8p-25, 0x1.4258e02395010p-27, -0x1.0361b28ea67e6p-29, 0x1.8ea34b43fdf6cp-32, -0x1.2510397eb07dep-34, 0x1.9cee2b21d3154p-37, -0x1.173835fb70366p-39, 0x1.6af784779d955p-42, -0x1.c628e1c8f0b3bp-45, 0x1.11d7f0615290cp-47, -0x1.3eaaa7e0d1573p-50, 0x1.663e3e593bfacp-53, -0x1.857d0c38a0576p-56, 0x1.99f2a0c3c4014p-59 ]; // Chebyshev coefficients for exp(-x) sqrt(x) I1(x) // in the inverted interval [8,infinity]. // lim(x->inf){ exp(-x) sqrt(x) I1(x) } = 1/sqrt(2pi). const double [] B = [ 0x1.8ea18b55b1514p-1, -0x1.3fda053fcdb4cp-7, -0x1.cfd7f804aa9a6p-14, -0x1.048df49ca0373p-18, -0x1.0dbfd2e9e5443p-22, -0x1.c415394bb46c1p-26, -0x1.0790b9ad53528p-28, -0x1.334ca5423dd80p-31, -0x1.4dcf9d4504c0cp-36, 0x1.1e1a1f1587865p-35, 0x1.f101f653c457bp-37, 0x1.1e7d3f6439fa3p-39, -0x1.953e1076ab493p-41, -0x1.cbc458e73e255p-42, -0x1.7a9482e6d22a0p-46, 0x1.80d3c26b3281ep-45, 0x1.776e1762d31e8p-47, -0x1.12db5138afbc7p-48, -0x1.0efcd8bc4d22ap-49, 0x1.7d68e5f04a2d1p-52, 0x1.55915fceb588ap-52, -0x1.2806c9c773320p-55, -0x1.acea3b2532277p-55, 0x1.45b8aea87b950p-58, 0x1.1556db352e8e6p-57 ]; double y, z; z = fabs(x); if( z <= 8.0 ) { y = (z/2.0) - 2.0; z = evalCheby( y, A ) * z * exp(z); } else { z = exp(z) * evalCheby( 32.0/z - 2.0, B ) / sqrt(z); } if (x < 0.0 ) z = -z; return z; } debug(UnitTest) { unittest { // argument, result1, result2, derivative. Correct result is result1+result2. const real [4][] j0_test_points = [ [8.0L, 1.71646118164062500000E-1L, 4.68897349140609086941E-6L, -2.34636346853914624381E-1L], [4.54541015625L, -3.09783935546875000000E-1L, 7.07472668157686463367E-6L, 2.42993657373627558460E-1L], [2.85711669921875L, -2.07901000976562500000E-1L, 1.15237285263902751582E-5L, -3.90402225324501311651E-1L], [2.0L, 2.23876953125000000000E-1L, 1.38260162356680518275E-5L, -5.76724807756873387202E-1L], [1.16415321826934814453125e-10L, 9.99984741210937500000E-1L, 1.52587890624999966119E-5L, 9.99999999999999999997E-1L], [-2.0L, 2.23876953125000000000E-1L, 1.38260162356680518275E-5L, 5.76724807756873387202E-1L] ]; const real [4][] y0_test_points = [ [ 8.0L, 2.23510742187500000000E-1L, 1.07472000662205273234E-5L, 1.58060461731247494256E-1L], [4.54541015625L, -2.08114624023437500000E-1L, 1.45018823856668874574E-5L, -2.88887645307401250876E-1L], [2.85711669921875L, 4.20303344726562500000E-1L, 1.32781607563122276008E-5L, -2.82488638474982469213E-1], [2.0L, 5.10360717773437500000E-1L, 1.49548763076195966066E-5L, 1.07032431540937546888E-1L], [1.16415321826934814453125e-10L, -1.46357574462890625000E1L, 3.54110537011061127637E-6L, 5.46852220461145271913E9L] ]; const real [4][] j1_test_points = [ [ 8.0L, 2.34634399414062500000E-1L, 1.94743985212438127665E-6L,1.42321263780814578043E-1], [4.54541015625L, -2.42996215820312500000E-1L, 2.55844668494153980076E-6L, -2.56317734136211337012E-1], [2.85711669921875L, 3.90396118164062500000E-1L, 6.10716043881165077013E-6L, -3.44531507106757980441E-1L], [2.0L, 5.76721191406250000000E-1L, 3.61635062338720244824E-6L, -6.44716247372010255494E-2L], [1.16415321826934814453125e-10L, 5.820677273504770710133016109466552734375e-11L, 8.881784197001251337312921818461805735896e-16L, 4.99999999999999999997E-1L], [-2.0L, -5.76721191406250000000E-1L, -3.61635062338720244824E-6L, -6.44716247372010255494E-2L] ]; const real [4][] y1_test_points = [ [8.0L, -1.58065795898437500000E-1L, 5.33416719000574444473E-6L, 2.43279047103972157309E-1L], [4.54541015625L, 2.88879394531250000000E-1L, 8.25077615125087585195E-6L, -2.71656024771791736625E-1L], [2.85711669921875L, 2.82485961914062500000E-1, 2.67656091996921314433E-6L, 3.21444694221532719737E-1], [2.0L, -1.07040405273437500000E-1L, 7.97373249995311162923E-6L, 5.63891888420213893041E-1], [1.16415321826934814453125e-10L, -5.46852220500000000000E9L, 3.88547280871200700671E-1L, 4.69742480525120196168E19L] ]; foreach(real [4] t; j0_test_points) { assert(feqrel(cylBessel_j0(t[0]), t[1]+t[2]) >=real.mant_dig-3); } foreach(real [4] t; y0_test_points) { assert(feqrel(cylBessel_y0(t[0]), t[1]+t[2]) >=real.mant_dig-4); } foreach(real [4] t; j1_test_points) { assert(feqrel(cylBessel_j1(t[0]), t[1]+t[2]) >=real.mant_dig-3); } foreach(real [4] t; y1_test_points) { assert(feqrel(cylBessel_y1(t[0]), t[1]+t[2]) >=real.mant_dig-4); } // Values from MS Excel, of doubtful accuracy. assert(fabs(-0.060_409_940_421_649 - cylBessel_j0(173.5)) < 0.000_000_000_1); assert(fabs(-0.044_733_447_576_5866 - cylBessel_y0(313.25)) < 0.000_000_000_1); assert(fabs(0.00391280088318945 - cylBessel_j1(123.25)) < 0.000_000_000_1); assert(fabs(-0.0648628570878951 - cylBessel_j1(-91)) < 0.000_000_000_1); assert(fabs(-0.0759578537652805 - cylBessel_y1(107.75)) < 0.000_000_000_1); assert(fabs(13.442_456_516_6771-cylBessel_i0(4.2)) < 0.000_001); assert(fabs(1.6500020842093e+28-cylBessel_i0(-68)) < 0.000_001e+28); assert(fabs(4.02746515903173e+10-cylBessel_i1(27)) < 0.000_001e+10); assert(fabs(-2.83613942886386e-02-cylBessel_i1(-0.0567)) < 0.000_000_001e-2); } } debug(UnitTest) { unittest { // Wronksian test for Bessel functions void testWronksian(int n, real x) { real Jnp1 = cylBessel_jn(n + 1, x); real Jmn = cylBessel_jn(-n, x); real Jn = cylBessel_jn(n, x); real Jmnp1 = cylBessel_jn(-(n + 1), x); /* This should be trivially zero. */ assert( fabs(Jnp1 * Jmn + Jn * Jmnp1) == 0); if (x < 0.0) { x = -x; Jn = cylBessel_jn(n, x); Jnp1 = cylBessel_jn(n + 1, x); } real Yn = cylBessel_yn(n, x); real Ynp1 = cylBessel_yn(n + 1, x); /* The Wronksian. */ real w1 = Jnp1 * Yn - Jn * Ynp1; /* What the Wronksian should be. */ real w2 = 2.0 / (PI * x); real reldif = feqrel(w1, w2); assert(reldif >= real.mant_dig-6); } real delta; int n, i, j; delta = 0.6 / PI; for (n = -30; n <= 30; n++) { real x = -30.0; while (x < 30.0) { testWronksian (n, x); x += delta; } delta += .00123456; } assert(cylBessel_jn(20, 1e-80)==0); // NaN propagation assert(isIdentical(cylBessel_i1(NaN(0xDEF)), NaN(0xDEF))); assert(isIdentical(cylBessel_i0(NaN(0x846)), NaN(0x846))); } }