| 1 | /* -*- mode: c++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ |
| 2 | |
| 3 | // NOTE: The following copyright notice |
| 4 | // applies only to the (modified) code of erff. |
| 5 | // |
| 6 | |
| 7 | // erff |
| 8 | // ==== |
| 9 | // |
| 10 | // Based on code from the gnu C library, originally written by Sun. |
| 11 | // Modified to remove reliance on features of gcc and 64-bit width |
| 12 | // of doubles. No doubt this results in some slight deterioration |
| 13 | // of efficiency, but this is not really noticeable in testing. |
| 14 | // |
| 15 | |
| 16 | // |
| 17 | // ==================================================== |
| 18 | // Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. |
| 19 | // |
| 20 | // Developed at SunPro, a Sun Microsystems, Inc. business. |
| 21 | // Permission to use, copy, modify, and distribute this |
| 22 | // software is freely granted, provided that this notice |
| 23 | // is preserved. |
| 24 | // ==================================================== |
| 25 | |
| 26 | |
| 27 | #include <ql/math/errorfunction.hpp> |
| 28 | #include <cfloat> |
| 29 | |
| 30 | namespace QuantLib { |
| 31 | |
| 32 | // x |
| 33 | // 2 | |
| 34 | // erf(x) = --------- | exp(-t*t)dt |
| 35 | // sqrt(pi) \| |
| 36 | // 0 |
| 37 | // |
| 38 | // erfc(x) = 1-erf(x) |
| 39 | // Note that |
| 40 | // erf(-x) = -erf(x) |
| 41 | // erfc(-x) = 2 - erfc(x) |
| 42 | // |
| 43 | // Method: |
| 44 | // 1. For |x| in [0, 0.84375] |
| 45 | // erf(x) = x + x*R(x^2) |
| 46 | // erfc(x) = 1 - erf(x) if x in [-.84375,0.25] |
| 47 | // = 0.5 + ((0.5-x)-x*R) if x in [0.25,0.84375] |
| 48 | // where R = P/Q where P is an odd poly of degree 8 and |
| 49 | // Q is an odd poly of degree 10. |
| 50 | // -57.90 |
| 51 | // | R - (erf(x)-x)/x | <= 2 |
| 52 | // |
| 53 | // |
| 54 | // Remark. The formula is derived by noting |
| 55 | // erf(x) = (2/sqrt(pi))*(x - x^3/3 + x^5/10 - x^7/42 + ....) |
| 56 | // and that |
| 57 | // 2/sqrt(pi) = 1.128379167095512573896158903121545171688 |
| 58 | // is close to one. The interval is chosen because the fix |
| 59 | // point of erf(x) is near 0.6174 (i.e., erf(x)=x when x is |
| 60 | // near 0.6174), and by some experiment, 0.84375 is chosen to |
| 61 | // guarantee the error is less than one ulp for erf. |
| 62 | // |
| 63 | // 2. For |x| in [0.84375,1.25], let s = |x| - 1, and |
| 64 | // c = 0.84506291151 rounded to single (24 bits) |
| 65 | // erf(x) = sign(x) * (c + P1(s)/Q1(s)) |
| 66 | // erfc(x) = (1-c) - P1(s)/Q1(s) if x > 0 |
| 67 | // 1+(c+P1(s)/Q1(s)) if x < 0 |
| 68 | // |P1/Q1 - (erf(|x|)-c)| <= 2**-59.06 |
| 69 | // Remark: here we use the taylor series expansion at x=1. |
| 70 | // erf(1+s) = erf(1) + s*Poly(s) |
| 71 | // = 0.845.. + P1(s)/Q1(s) |
| 72 | // That is, we use rational approximation to approximate |
| 73 | // erf(1+s) - (c = (single)0.84506291151) |
| 74 | // Note that |P1/Q1|< 0.078 for x in [0.84375,1.25] |
| 75 | // where |
| 76 | // P1(s) = degree 6 poly in s |
| 77 | // Q1(s) = degree 6 poly in s |
| 78 | // |
| 79 | // 3. For x in [1.25,1/0.35(~2.857143)], |
| 80 | // erfc(x) = (1/x)*exp(-x*x-0.5625+R1/S1) |
| 81 | // erf(x) = 1 - erfc(x) |
| 82 | // where |
| 83 | // R1(z) = degree 7 poly in z, (z=1/x^2) |
| 84 | // S1(z) = degree 8 poly in z |
| 85 | // |
| 86 | // 4. For x in [1/0.35,28] |
| 87 | // erfc(x) = (1/x)*exp(-x*x-0.5625+R2/S2) if x > 0 |
| 88 | // = 2.0 - (1/x)*exp(-x*x-0.5625+R2/S2) if -6<x<0 |
| 89 | // = 2.0 - tiny (if x <= -6) |
| 90 | // erf(x) = sign(x)*(1.0 - erfc(x)) if x < 6, else |
| 91 | // erf(x) = sign(x)*(1.0 - tiny) |
| 92 | // where |
| 93 | // R2(z) = degree 6 poly in z, (z=1/x^2) |
| 94 | // S2(z) = degree 7 poly in z |
| 95 | // |
| 96 | // Note1: |
| 97 | // To compute exp(-x*x-0.5625+R/S), let s be a single |
| 98 | // precision number and s := x; then |
| 99 | // -x*x = -s*s + (s-x)*(s+x) |
| 100 | // exp(-x*x-0.5626+R/S) = |
| 101 | // exp(-s*s-0.5625)*exp((s-x)*(s+x)+R/S); |
| 102 | // Note2: |
| 103 | // Here 4 and 5 make use of the asymptotic series |
| 104 | // exp(-x*x) |
| 105 | // erfc(x) ~ ---------- * ( 1 + Poly(1/x^2) ) |
| 106 | // x*sqrt(pi) |
| 107 | // We use rational approximation to approximate |
| 108 | // g(s)=f(1/x^2) = log(erfc(x)*x) - x*x + 0.5625 |
| 109 | // Here is the error bound for R1/S1 and R2/S2 |
| 110 | // |R1/S1 - f(x)| < 2**(-62.57) |
| 111 | // |R2/S2 - f(x)| < 2**(-61.52) |
| 112 | // |
| 113 | // 5. For inf > x >= 28 |
| 114 | // erf(x) = sign(x) *(1 - tiny) (raise inexact) |
| 115 | // erfc(x) = tiny*tiny (raise underflow) if x > 0 |
| 116 | // = 2 - tiny if x<0 |
| 117 | // |
| 118 | // 7. Special case: |
| 119 | // erf(0) = 0, erf(inf) = 1, erf(-inf) = -1, |
| 120 | // erfc(0) = 1, erfc(inf) = 0, erfc(-inf) = 2, |
| 121 | // erfc/erf(NaN) is NaN |
| 122 | |
| 123 | const Real |
| 124 | ErrorFunction::tiny = QL_EPSILON, |
| 125 | ErrorFunction::one = 1.00000000000000000000e+00, /* 0x3FF00000, 0x00000000 */ |
| 126 | /* c = (float)0.84506291151 */ |
| 127 | ErrorFunction::erx = 8.45062911510467529297e-01, /* 0x3FEB0AC1, 0x60000000 */ |
| 128 | // |
| 129 | // Coefficients for approximation to erf on [0,0.84375] |
| 130 | // |
| 131 | ErrorFunction::efx = 1.28379167095512586316e-01, /* 0x3FC06EBA, 0x8214DB69 */ |
| 132 | ErrorFunction::efx8 = 1.02703333676410069053e+00, /* 0x3FF06EBA, 0x8214DB69 */ |
| 133 | ErrorFunction::pp0 = 1.28379167095512558561e-01, /* 0x3FC06EBA, 0x8214DB68 */ |
| 134 | ErrorFunction::pp1 = -3.25042107247001499370e-01, /* 0xBFD4CD7D, 0x691CB913 */ |
| 135 | ErrorFunction::pp2 = -2.84817495755985104766e-02, /* 0xBF9D2A51, 0xDBD7194F */ |
| 136 | ErrorFunction::pp3 = -5.77027029648944159157e-03, /* 0xBF77A291, 0x236668E4 */ |
| 137 | ErrorFunction::pp4 = -2.37630166566501626084e-05, /* 0xBEF8EAD6, 0x120016AC */ |
| 138 | ErrorFunction::qq1 = 3.97917223959155352819e-01, /* 0x3FD97779, 0xCDDADC09 */ |
| 139 | ErrorFunction::qq2 = 6.50222499887672944485e-02, /* 0x3FB0A54C, 0x5536CEBA */ |
| 140 | ErrorFunction::qq3 = 5.08130628187576562776e-03, /* 0x3F74D022, 0xC4D36B0F */ |
| 141 | ErrorFunction::qq4 = 1.32494738004321644526e-04, /* 0x3F215DC9, 0x221C1A10 */ |
| 142 | ErrorFunction::qq5 = -3.96022827877536812320e-06, /* 0xBED09C43, 0x42A26120 */ |
| 143 | // |
| 144 | // Coefficients for approximation to erf in [0.84375,1.25] |
| 145 | // |
| 146 | ErrorFunction::pa0 = -2.36211856075265944077e-03, /* 0xBF6359B8, 0xBEF77538 */ |
| 147 | ErrorFunction::pa1 = 4.14856118683748331666e-01, /* 0x3FDA8D00, 0xAD92B34D */ |
| 148 | ErrorFunction::pa2 = -3.72207876035701323847e-01, /* 0xBFD7D240, 0xFBB8C3F1 */ |
| 149 | ErrorFunction::pa3 = 3.18346619901161753674e-01, /* 0x3FD45FCA, 0x805120E4 */ |
| 150 | ErrorFunction::pa4 = -1.10894694282396677476e-01, /* 0xBFBC6398, 0x3D3E28EC */ |
| 151 | ErrorFunction::pa5 = 3.54783043256182359371e-02, /* 0x3FA22A36, 0x599795EB */ |
| 152 | ErrorFunction::pa6 = -2.16637559486879084300e-03, /* 0xBF61BF38, 0x0A96073F */ |
| 153 | ErrorFunction::qa1 = 1.06420880400844228286e-01, /* 0x3FBB3E66, 0x18EEE323 */ |
| 154 | ErrorFunction::qa2 = 5.40397917702171048937e-01, /* 0x3FE14AF0, 0x92EB6F33 */ |
| 155 | ErrorFunction::qa3 = 7.18286544141962662868e-02, /* 0x3FB2635C, 0xD99FE9A7 */ |
| 156 | ErrorFunction::qa4 = 1.26171219808761642112e-01, /* 0x3FC02660, 0xE763351F */ |
| 157 | ErrorFunction::qa5 = 1.36370839120290507362e-02, /* 0x3F8BEDC2, 0x6B51DD1C */ |
| 158 | ErrorFunction::qa6 = 1.19844998467991074170e-02, /* 0x3F888B54, 0x5735151D */ |
| 159 | // |
| 160 | // Coefficients for approximation to erfc in [1.25,1/0.35] |
| 161 | // |
| 162 | ErrorFunction::ra0 = -9.86494403484714822705e-03, /* 0xBF843412, 0x600D6435 */ |
| 163 | ErrorFunction::ra1 = -6.93858572707181764372e-01, /* 0xBFE63416, 0xE4BA7360 */ |
| 164 | ErrorFunction::ra2 = -1.05586262253232909814e+01, /* 0xC0251E04, 0x41B0E726 */ |
| 165 | ErrorFunction::ra3 = -6.23753324503260060396e+01, /* 0xC04F300A, 0xE4CBA38D */ |
| 166 | ErrorFunction::ra4 = -1.62396669462573470355e+02, /* 0xC0644CB1, 0x84282266 */ |
| 167 | ErrorFunction::ra5 = -1.84605092906711035994e+02, /* 0xC067135C, 0xEBCCABB2 */ |
| 168 | ErrorFunction::ra6 = -8.12874355063065934246e+01, /* 0xC0545265, 0x57E4D2F2 */ |
| 169 | ErrorFunction::ra7 = -9.81432934416914548592e+00, /* 0xC023A0EF, 0xC69AC25C */ |
| 170 | ErrorFunction::sa1 = 1.96512716674392571292e+01, /* 0x4033A6B9, 0xBD707687 */ |
| 171 | ErrorFunction::sa2 = 1.37657754143519042600e+02, /* 0x4061350C, 0x526AE721 */ |
| 172 | ErrorFunction::sa3 = 4.34565877475229228821e+02, /* 0x407B290D, 0xD58A1A71 */ |
| 173 | ErrorFunction::sa4 = 6.45387271733267880336e+02, /* 0x40842B19, 0x21EC2868 */ |
| 174 | ErrorFunction::sa5 = 4.29008140027567833386e+02, /* 0x407AD021, 0x57700314 */ |
| 175 | ErrorFunction::sa6 = 1.08635005541779435134e+02, /* 0x405B28A3, 0xEE48AE2C */ |
| 176 | ErrorFunction::sa7 = 6.57024977031928170135e+00, /* 0x401A47EF, 0x8E484A93 */ |
| 177 | ErrorFunction::sa8 = -6.04244152148580987438e-02, /* 0xBFAEEFF2, 0xEE749A62 */ |
| 178 | // |
| 179 | // Coefficients for approximation to erfc in [1/.35,28] |
| 180 | // |
| 181 | ErrorFunction::rb0 = -9.86494292470009928597e-03, /* 0xBF843412, 0x39E86F4A */ |
| 182 | ErrorFunction::rb1 = -7.99283237680523006574e-01, /* 0xBFE993BA, 0x70C285DE */ |
| 183 | ErrorFunction::rb2 = -1.77579549177547519889e+01, /* 0xC031C209, 0x555F995A */ |
| 184 | ErrorFunction::rb3 = -1.60636384855821916062e+02, /* 0xC064145D, 0x43C5ED98 */ |
| 185 | ErrorFunction::rb4 = -6.37566443368389627722e+02, /* 0xC083EC88, 0x1375F228 */ |
| 186 | ErrorFunction::rb5 = -1.02509513161107724954e+03, /* 0xC0900461, 0x6A2E5992 */ |
| 187 | ErrorFunction::rb6 = -4.83519191608651397019e+02, /* 0xC07E384E, 0x9BDC383F */ |
| 188 | ErrorFunction::sb1 = 3.03380607434824582924e+01, /* 0x403E568B, 0x261D5190 */ |
| 189 | ErrorFunction::sb2 = 3.25792512996573918826e+02, /* 0x40745CAE, 0x221B9F0A */ |
| 190 | ErrorFunction::sb3 = 1.53672958608443695994e+03, /* 0x409802EB, 0x189D5118 */ |
| 191 | ErrorFunction::sb4 = 3.19985821950859553908e+03, /* 0x40A8FFB7, 0x688C246A */ |
| 192 | ErrorFunction::sb5 = 2.55305040643316442583e+03, /* 0x40A3F219, 0xCEDF3BE6 */ |
| 193 | ErrorFunction::sb6 = 4.74528541206955367215e+02, /* 0x407DA874, 0xE79FE763 */ |
| 194 | ErrorFunction::sb7 = -2.24409524465858183362e+01; /* 0xC03670E2, 0x42712D62 */ |
| 195 | |
| 196 | Real ErrorFunction::operator()(Real x) const { |
| 197 | |
| 198 | Real R,S,P,Q,s,y,z,r, ax; |
| 199 | |
| 200 | if (!std::isfinite(x: x)) { |
| 201 | if (std::isnan(x: x)) |
| 202 | return x; |
| 203 | else |
| 204 | return ( x > 0 ? 1 : -1); |
| 205 | } |
| 206 | |
| 207 | ax = std::fabs(x: x); |
| 208 | |
| 209 | if(ax < 0.84375) { /* |x|<0.84375 */ |
| 210 | if(ax < 3.7252902984e-09) { /* |x|<2**-28 */ |
| 211 | if (ax < DBL_MIN*16) |
| 212 | return 0.125*(8.0*x+efx8*x); /*avoid underflow */ |
| 213 | return x + efx*x; |
| 214 | } |
| 215 | z = x*x; |
| 216 | r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); |
| 217 | s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); |
| 218 | y = r/s; |
| 219 | return x + x*y; |
| 220 | } |
| 221 | if(ax <1.25) { /* 0.84375 <= |x| < 1.25 */ |
| 222 | s = ax-one; |
| 223 | P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); |
| 224 | Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); |
| 225 | if(x>=0) return erx + P/Q; else return -erx - P/Q; |
| 226 | } |
| 227 | if (ax >= 6) { /* inf>|x|>=6 */ |
| 228 | if(x>=0) return one-tiny; else return tiny-one; |
| 229 | } |
| 230 | |
| 231 | /* Starts to lose accuracy when ax~5 */ |
| 232 | s = one/(ax*ax); |
| 233 | |
| 234 | if(ax < 2.85714285714285) { /* |x| < 1/0.35 */ |
| 235 | R = ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(ra5+s*(ra6+s*ra7)))))); |
| 236 | S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(sa5+s*(sa6+s*(sa7+s*sa8))))))); |
| 237 | } else { /* |x| >= 1/0.35 */ |
| 238 | R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(rb5+s*rb6))))); |
| 239 | S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(sb5+s*(sb6+s*sb7)))))); |
| 240 | } |
| 241 | r = std::exp( x: -ax*ax-0.5625 +R/S); |
| 242 | if(x>=0) return one-r/ax; else return r/ax-one; |
| 243 | |
| 244 | } |
| 245 | |
| 246 | } |
| 247 | |