| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 8/2025. All rights reserved. |
| 3 |
* // |
| 4 |
* // Redistribution and use in source and binary forms, with or without modification, |
| 5 |
* // are permitted provided that the following conditions are met: |
| 6 |
* // |
| 7 |
* // 1. Redistributions of source code must retain the above copyright notice, this |
| 8 |
* // list of conditions and the following disclaimer. |
| 9 |
* // |
| 10 |
* // 2. Redistributions in binary form must reproduce the above copyright notice, |
| 11 |
* // this list of conditions and the following disclaimer in the documentation |
| 12 |
* // and/or other materials provided with the distribution. |
| 13 |
* // |
| 14 |
* // 3. Neither the name of the copyright holder nor the names of its |
| 15 |
* // contributors may be used to endorse or promote products derived from |
| 16 |
* // this software without specific prior written permission. |
| 17 |
* // |
| 18 |
* // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 19 |
* // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 20 |
* // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 21 |
* // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE |
| 22 |
* // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 23 |
* // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR |
| 24 |
* // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
| 25 |
* // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 26 |
* // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 27 |
* // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 28 |
*/ |
| 29 |
use crate::common::is_integer; |
| 30 |
use crate::double_double::DoubleDouble; |
| 31 |
use crate::sincospi::f_fast_sinpi_dd; |
| 32 |
|
| 33 |
// Generated in Wolfram Mathematica: |
| 34 |
// <<FunctionApproximations` |
| 35 |
// ClearAll["Global`*"] |
| 36 |
// f[x_]:=PolyGamma[1, x+1] |
| 37 |
// {err0,approx}=MiniMaxApproximation[f[z],{z,{-0.99999999,0},11,11},WorkingPrecision->75,MaxIterations->100] |
| 38 |
// num=Numerator[approx][[1]]; |
| 39 |
// poly=num; |
| 40 |
// coeffs=CoefficientList[poly,z]; |
| 41 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 42 |
static P_1: [(u64, u64); 12] = [ |
| 43 |
(0x3c81873d89121fec, 0x3ffa51a6625307d3), |
| 44 |
(0x3cb78bf7af507504, 0x4016a65cbac476ca), |
| 45 |
(0xbc94179c65e021c2, 0x40218234a0a79582), |
| 46 |
(0x3cb842a8ab5e0994, 0x401fde32175f8515), |
| 47 |
(0xbc8768b33f5776b7, 0x4012de6bde49abff), |
| 48 |
(0x3c8e06354a27f081, 0x3ffe5d6ef3a2eac6), |
| 49 |
(0xbc8b391d09fed17a, 0x3fe0d186688252cf), |
| 50 |
(0xbc59a5c46bb8b8cc, 0x3fb958f9a0f156b7), |
| 51 |
(0x3c2ac44c6a197244, 0x3f88e605f24e1a89), |
| 52 |
(0x3bd2d05fa8be27f2, 0x3f4cd369f5d68104), |
| 53 |
(0x3b84ad0a748fdd22, 0x3efde955ebb17874), |
| 54 |
(0xb96aa8b9a65e0899, 0xbce053d04459ead7), |
| 55 |
]; |
| 56 |
|
| 57 |
// Generated by Wolfram Mathematica: |
| 58 |
// <<FunctionApproximations` |
| 59 |
// ClearAll["Global`*"] |
| 60 |
// f[x_]:=PolyGamma[1, x+1] |
| 61 |
// {err0,approx}=MiniMaxApproximation[f[z],{z,{0,3},11,11},WorkingPrecision->75,MaxIterations->100] |
| 62 |
// num=Numerator[approx][[1]]; |
| 63 |
// poly=num; |
| 64 |
// coeffs=CoefficientList[poly,z]; |
| 65 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 66 |
static P_2: [(u64, u64); 12] = [ |
| 67 |
(0x3c81873d8912236c, 0x3ffa51a6625307d3), |
| 68 |
(0xbcbc731a62342288, 0x40176c23f7ea51e6), |
| 69 |
(0xbcc45a6fd00e67a8, 0x4022cb5ae67657ef), |
| 70 |
(0x3cc0876fde7fe4e6, 0x4021d766062b9550), |
| 71 |
(0xbcaec4a4859cba1d, 0x401629f91cd4f291), |
| 72 |
(0x3c76184014e4d7e3, 0x4002d43da3352004), |
| 73 |
(0x3c812c7609483e0e, 0x3fe62e3266eef8c7), |
| 74 |
(0xbc5f991047f52d2b, 0x3fc1eacb910b951c), |
| 75 |
(0x3c28b9f38d603f2f, 0x3f930960a301df34), |
| 76 |
(0x3bf9b620eb930504, 0x3f5814f8e057b14b), |
| 77 |
(0xbb990860b88b54e4, 0x3f0b9f67c71aa3bf), |
| 78 |
(0x38e5cb6acfbaab77, 0xbc4194b8c01afe9a), |
| 79 |
]; |
| 80 |
|
| 81 |
// Generated by Wolfram Mathematica: |
| 82 |
// <<FunctionApproximations` |
| 83 |
// ClearAll["Global`*"] |
| 84 |
// f[x_]:=PolyGamma[1, x+1] |
| 85 |
// {err0,approx,err1}=MiniMaxApproximation[f[z],{z,{3,9},11,11},WorkingPrecision->75,MaxIterations->100] |
| 86 |
// num=Numerator[approx]; |
| 87 |
// poly=num; |
| 88 |
// coeffs=CoefficientList[poly,z]; |
| 89 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 90 |
static P_3: [(u64, u64); 12] = [ |
| 91 |
(0x3c9cd56dbb295efc, 0x3ffa51a662556db9), |
| 92 |
(0x3c9f4ee74f5f9daf, 0x4018ff913088cb34), |
| 93 |
(0x3ccf08737350609c, 0x402593d55686b8b1), |
| 94 |
(0xbcc6cd4ed33afebb, 0x402641d10de4def5), |
| 95 |
(0xbcb24d1957c1303c, 0x401e682c37e8e2cf), |
| 96 |
(0x3ca30ac79162ceb2, 0x400ccfc7c4566f55), |
| 97 |
(0x3c9efea5ff293dc9, 0x3ff33eb2c6e89d0b), |
| 98 |
(0x3c74670a11068abc, 0x3fd1fbf456e5c6f0), |
| 99 |
(0x3c47b5dcdea19c36, 0x3fa6a6a2148c482c), |
| 100 |
(0xbc14642012a1cc1e, 0x3f71851e927f52e7), |
| 101 |
(0x3bc7db88a4ec5478, 0x3f29a45059a43475), |
| 102 |
(0xb7bc31e55271eab0, 0xbb375518529c52fb), |
| 103 |
]; |
| 104 |
|
| 105 |
// Generated in Wolfram Mathematica: |
| 106 |
// <<FunctionApproximations` |
| 107 |
// ClearAll["Global`*"] |
| 108 |
// f[x_]:=PolyGamma[1, x+1] |
| 109 |
// {err0,approx}=MiniMaxApproximation[f[z],{z,{-0.99999999,0},11,11},WorkingPrecision->75,MaxIterations->100] |
| 110 |
// den=Denominator[approx][[1]]; |
| 111 |
// poly=den; |
| 112 |
// coeffs=CoefficientList[poly,z]; |
| 113 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 114 |
static Q_1: [(u64, u64); 12] = [ |
| 115 |
(0x0000000000000000, 0x3ff0000000000000), |
| 116 |
(0xbcb84c43a11fc28a, 0x40139d9587da0fb5), |
| 117 |
(0x3ca1cf3dbcddbb57, 0x402507cb6225f0f0), |
| 118 |
(0x3cb01aa6ddcc3cfd, 0x402a1b416d0ed4e6), |
| 119 |
(0xbcbc31c216b5ff66, 0x4024ec8829e535d4), |
| 120 |
(0xbcb335c23022f43e, 0x4016d2ba6d1a18e6), |
| 121 |
(0x3cafbfffc03ad28a, 0x400158c4611ed51f), |
| 122 |
(0xbc8d1fb10a031a27, 0x3fe26f15bb52f89b), |
| 123 |
(0x3c56a9fea160eecb, 0x3fbaec13f663049d), |
| 124 |
(0xbc2f4ee869ba9364, 0x3f89cde0500cd68f), |
| 125 |
(0x3be3f23afc9398b6, 0x3f4d4b0f4dcf3eb8), |
| 126 |
(0x3b67a3ed4795d33e, 0x3efde955eafac9c2), |
| 127 |
]; |
| 128 |
|
| 129 |
// Generated by Wolfram Mathematica: |
| 130 |
// <<FunctionApproximations` |
| 131 |
// ClearAll["Global`*"] |
| 132 |
// f[x_]:=PolyGamma[1, x+1] |
| 133 |
// {err0,approx}=MiniMaxApproximation[f[z],{z,{0,3},11,11},WorkingPrecision->75,MaxIterations->100] |
| 134 |
// den=Denominator[approx][[1]]; |
| 135 |
// poly=den; |
| 136 |
// coeffs=CoefficientList[poly,z]; |
| 137 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 138 |
static Q_2: [(u64, u64); 12] = [ |
| 139 |
(0x0000000000000000, 0x3ff0000000000000), |
| 140 |
(0xbc81a3e4e026b7b1, 0x401415d1a20a9339), |
| 141 |
(0x3cc279576dfe3ec9, 0x402627c1a95d33d2), |
| 142 |
(0x3c9a94b5cf0cae88, 0x402c724dc5cf4577), |
| 143 |
(0xbc8aa1fa0c3820a8, 0x4027b7a332bb07f4), |
| 144 |
(0xbc96968367088d66, 0x401b14376177bdd7), |
| 145 |
(0x3ca2d3dfa5847f4d, 0x4005b0511cd98f2c), |
| 146 |
(0xbc8cfad394d41dd1, 0x3fe877bc2d02c7f3), |
| 147 |
(0xbc51592b8ec81a92, 0x3fc31f52afc72b95), |
| 148 |
(0x3c2cbef277d587e9, 0x3f93cb2f0e574376), |
| 149 |
(0xbbfbb670fd94f6ba, 0x3f5883767f745a92), |
| 150 |
(0xbb931b04d74e5893, 0x3f0b9f67c71a60f3), |
| 151 |
]; |
| 152 |
|
| 153 |
// Generated by Wolfram Mathematica: |
| 154 |
// <<FunctionApproximations` |
| 155 |
// ClearAll["Global`*"] |
| 156 |
// f[x_]:=PolyGamma[1, x+1] |
| 157 |
// {err0,approx,err1}=MiniMaxApproximation[f[z],{z,{3,9},11,11},WorkingPrecision->75,MaxIterations->100] |
| 158 |
// den=Denominator[approx]; |
| 159 |
// poly=den; |
| 160 |
// coeffs=CoefficientList[poly,z]; |
| 161 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 162 |
static Q_3: [(u64, u64); 12] = [ |
| 163 |
(0x0000000000000000, 0x3ff0000000000000), |
| 164 |
(0x3cbafcb4b6d646d9, 0x40150b12a79fc9cf), |
| 165 |
(0x3c989ef814b8dd2a, 0x40288c1d26ffdca5), |
| 166 |
(0x3cb0282cfea9c473, 0x4030d737c893cd5f), |
| 167 |
(0x3cc955b8aaadb37d, 0x402e5c289b6de3e0), |
| 168 |
(0x3cb377161f8861d2, 0x4022fb66d87bd522), |
| 169 |
(0xbcb4b0e4cff46ad6, 0x4010e5d13c2a5907), |
| 170 |
(0xbc8824539e4b1bd6, 0x3ff58c8fe8f26fca), |
| 171 |
(0xbc7d34220d810ea0, 0x3fd36c1351f43e66), |
| 172 |
(0xbc4cbdbe85570017, 0x3fa7c1170466605e), |
| 173 |
(0xbc0c3afb98775c53, 0x3f71ebafd3e5e3b9), |
| 174 |
(0x3bc0b0b7f16afd0a, 0x3f29a45059a43475), |
| 175 |
]; |
| 176 |
|
| 177 |
#[inline] |
| 178 |
fn approx_trigamma(x: f64) -> DoubleDouble { |
| 179 |
if x <= 10. { |
| 180 |
let (p, q) = if x <= 1. { |
| 181 |
(&P_1, &Q_1) |
| 182 |
} else if x <= 4. { |
| 183 |
(&P_2, &Q_2) |
| 184 |
} else { |
| 185 |
(&P_3, &Q_3) |
| 186 |
}; |
| 187 |
let x2 = DoubleDouble::from_exact_mult(x, x); |
| 188 |
let x4 = DoubleDouble::quick_mult(x2, x2); |
| 189 |
let x8 = DoubleDouble::quick_mult(x4, x4); |
| 190 |
|
| 191 |
let e0 = DoubleDouble::mul_f64_add( |
| 192 |
DoubleDouble::from_bit_pair(p[1]), |
| 193 |
x, |
| 194 |
DoubleDouble::from_bit_pair(p[0]), |
| 195 |
); |
| 196 |
let e1 = DoubleDouble::mul_f64_add( |
| 197 |
DoubleDouble::from_bit_pair(p[3]), |
| 198 |
x, |
| 199 |
DoubleDouble::from_bit_pair(p[2]), |
| 200 |
); |
| 201 |
let e2 = DoubleDouble::mul_f64_add( |
| 202 |
DoubleDouble::from_bit_pair(p[5]), |
| 203 |
x, |
| 204 |
DoubleDouble::from_bit_pair(p[4]), |
| 205 |
); |
| 206 |
let e3 = DoubleDouble::mul_f64_add( |
| 207 |
DoubleDouble::from_bit_pair(p[7]), |
| 208 |
x, |
| 209 |
DoubleDouble::from_bit_pair(p[6]), |
| 210 |
); |
| 211 |
let e4 = DoubleDouble::mul_f64_add( |
| 212 |
DoubleDouble::from_bit_pair(p[9]), |
| 213 |
x, |
| 214 |
DoubleDouble::from_bit_pair(p[8]), |
| 215 |
); |
| 216 |
let e5 = DoubleDouble::mul_f64_add( |
| 217 |
DoubleDouble::from_bit_pair(p[11]), |
| 218 |
x, |
| 219 |
DoubleDouble::from_bit_pair(p[10]), |
| 220 |
); |
| 221 |
|
| 222 |
let f0 = DoubleDouble::mul_add(x2, e1, e0); |
| 223 |
let f1 = DoubleDouble::mul_add(x2, e3, e2); |
| 224 |
let f2 = DoubleDouble::mul_add(x2, e5, e4); |
| 225 |
|
| 226 |
let g0 = DoubleDouble::mul_add(x4, f1, f0); |
| 227 |
|
| 228 |
let p_num = DoubleDouble::mul_add(x8, f2, g0); |
| 229 |
|
| 230 |
let rcp = DoubleDouble::from_quick_recip(x); |
| 231 |
let rcp2 = DoubleDouble::quick_mult(rcp, rcp); |
| 232 |
|
| 233 |
let e0 = DoubleDouble::mul_f64_add_f64( |
| 234 |
DoubleDouble::from_bit_pair(q[1]), |
| 235 |
x, |
| 236 |
f64::from_bits(0x3ff0000000000000), |
| 237 |
); |
| 238 |
let e1 = DoubleDouble::mul_f64_add( |
| 239 |
DoubleDouble::from_bit_pair(q[3]), |
| 240 |
x, |
| 241 |
DoubleDouble::from_bit_pair(q[2]), |
| 242 |
); |
| 243 |
let e2 = DoubleDouble::mul_f64_add( |
| 244 |
DoubleDouble::from_bit_pair(q[5]), |
| 245 |
x, |
| 246 |
DoubleDouble::from_bit_pair(q[4]), |
| 247 |
); |
| 248 |
let e3 = DoubleDouble::mul_f64_add( |
| 249 |
DoubleDouble::from_bit_pair(q[7]), |
| 250 |
x, |
| 251 |
DoubleDouble::from_bit_pair(q[6]), |
| 252 |
); |
| 253 |
let e4 = DoubleDouble::mul_f64_add( |
| 254 |
DoubleDouble::from_bit_pair(q[9]), |
| 255 |
x, |
| 256 |
DoubleDouble::from_bit_pair(q[8]), |
| 257 |
); |
| 258 |
let e5 = DoubleDouble::mul_f64_add( |
| 259 |
DoubleDouble::from_bit_pair(q[11]), |
| 260 |
x, |
| 261 |
DoubleDouble::from_bit_pair(q[10]), |
| 262 |
); |
| 263 |
|
| 264 |
let f0 = DoubleDouble::mul_add(x2, e1, e0); |
| 265 |
let f1 = DoubleDouble::mul_add(x2, e3, e2); |
| 266 |
let f2 = DoubleDouble::mul_add(x2, e5, e4); |
| 267 |
|
| 268 |
let g0 = DoubleDouble::mul_add(x4, f1, f0); |
| 269 |
|
| 270 |
let p_den = DoubleDouble::mul_add(x8, f2, g0); |
| 271 |
|
| 272 |
let q = DoubleDouble::div(p_num, p_den); |
| 273 |
let r = DoubleDouble::quick_dd_add(q, rcp2); |
| 274 |
return r; |
| 275 |
} |
| 276 |
// asymptotic expansion Trigamma[x] = 1/x + 1/x^2 + sum(Bernoulli(2*k)/x^(2*k + 1)) |
| 277 |
// Generated in SageMath: |
| 278 |
// var('x') |
| 279 |
// def bernoulli_terms(x, N): |
| 280 |
// S = 0 |
| 281 |
// for k in range(1, N+1): |
| 282 |
// B = bernoulli(2*k) |
| 283 |
// term = B*x**(-(2*k+1)) |
| 284 |
// S += term |
| 285 |
// return S |
| 286 |
// |
| 287 |
// terms = bernoulli_terms(x, 10) |
| 288 |
// coeffs = [RealField(150)(terms.coefficient(x, n)) for n in range(0, terms.degree(x)+1, 1)] |
| 289 |
// for k in range(0, 14): |
| 290 |
// c = terms.coefficient(x, -k) # coefficient of x^(-k) |
| 291 |
// if c == 0: |
| 292 |
// continue |
| 293 |
// print("f64::from_bits(" + double_to_hex(c) + "),") |
| 294 |
const C: [(u64, u64); 10] = [ |
| 295 |
(0x3c65555555555555, 0x3fc5555555555555), |
| 296 |
(0xbc21111111111111, 0xbfa1111111111111), |
| 297 |
(0x3c38618618618618, 0x3f98618618618618), |
| 298 |
(0xbc21111111111111, 0xbfa1111111111111), |
| 299 |
(0xbc4364d9364d9365, 0x3fb364d9364d9365), |
| 300 |
(0xbc6981981981981a, 0xbfd0330330330330), |
| 301 |
(0xbc95555555555555, 0x3ff2aaaaaaaaaaab), |
| 302 |
(0xbcb7979797979798, 0xc01c5e5e5e5e5e5e), |
| 303 |
(0xbcac3b070ec1c3b0, 0x404b7c4f8f13e3c5), |
| 304 |
(0x3cc8d3018d3018d3, 0xc08088fe72cfe72d), |
| 305 |
]; |
| 306 |
|
| 307 |
let rcp = DoubleDouble::from_quick_recip(x); |
| 308 |
|
| 309 |
let q = DoubleDouble::quick_mult(rcp, rcp); |
| 310 |
|
| 311 |
let q2 = DoubleDouble::quick_mult(q, q); |
| 312 |
let q4 = q2 * q2; |
| 313 |
let q8 = q4 * q4; |
| 314 |
|
| 315 |
let e0 = DoubleDouble::quick_mul_add( |
| 316 |
DoubleDouble::from_bit_pair(C[1]), |
| 317 |
q, |
| 318 |
DoubleDouble::from_bit_pair(C[0]), |
| 319 |
); |
| 320 |
let e1 = DoubleDouble::quick_mul_add( |
| 321 |
DoubleDouble::from_bit_pair(C[3]), |
| 322 |
q, |
| 323 |
DoubleDouble::from_bit_pair(C[2]), |
| 324 |
); |
| 325 |
let e2 = DoubleDouble::quick_mul_add( |
| 326 |
DoubleDouble::from_bit_pair(C[5]), |
| 327 |
q, |
| 328 |
DoubleDouble::from_bit_pair(C[4]), |
| 329 |
); |
| 330 |
let e3 = DoubleDouble::quick_mul_add( |
| 331 |
DoubleDouble::from_bit_pair(C[7]), |
| 332 |
q, |
| 333 |
DoubleDouble::from_bit_pair(C[6]), |
| 334 |
); |
| 335 |
let e4 = DoubleDouble::quick_mul_add( |
| 336 |
DoubleDouble::from_bit_pair(C[9]), |
| 337 |
q, |
| 338 |
DoubleDouble::from_bit_pair(C[8]), |
| 339 |
); |
| 340 |
|
| 341 |
let q0 = DoubleDouble::quick_mul_add(q2, e1, e0); |
| 342 |
let q1 = DoubleDouble::quick_mul_add(q2, e3, e2); |
| 343 |
|
| 344 |
let r0 = DoubleDouble::quick_mul_add(q4, q1, q0); |
| 345 |
let mut p = DoubleDouble::quick_mul_add(q8, e4, r0); |
| 346 |
|
| 347 |
let q_over_2 = DoubleDouble::quick_mult_f64(q, 0.5); |
| 348 |
p = DoubleDouble::quick_mult(p, q); |
| 349 |
p = DoubleDouble::quick_mult(p, rcp); |
| 350 |
p = DoubleDouble::quick_dd_add(q_over_2, p); |
| 351 |
p = DoubleDouble::quick_dd_add(p, rcp); |
| 352 |
p |
| 353 |
} |
| 354 |
|
| 355 |
#[inline] |
| 356 |
fn approx_trigamma_dd(x: DoubleDouble) -> DoubleDouble { |
| 357 |
if x.hi <= 10. { |
| 358 |
let (p, q) = if x.hi <= 1. { |
| 359 |
(&P_1, &Q_1) |
| 360 |
} else if x.hi <= 4. { |
| 361 |
(&P_2, &Q_2) |
| 362 |
} else { |
| 363 |
(&P_3, &Q_3) |
| 364 |
}; |
| 365 |
let x2 = DoubleDouble::quick_mult(x, x); |
| 366 |
let x4 = DoubleDouble::quick_mult(x2, x2); |
| 367 |
let x8 = DoubleDouble::quick_mult(x4, x4); |
| 368 |
|
| 369 |
let e0 = DoubleDouble::mul_add( |
| 370 |
DoubleDouble::from_bit_pair(p[1]), |
| 371 |
x, |
| 372 |
DoubleDouble::from_bit_pair(p[0]), |
| 373 |
); |
| 374 |
let e1 = DoubleDouble::mul_add( |
| 375 |
DoubleDouble::from_bit_pair(p[3]), |
| 376 |
x, |
| 377 |
DoubleDouble::from_bit_pair(p[2]), |
| 378 |
); |
| 379 |
let e2 = DoubleDouble::mul_add( |
| 380 |
DoubleDouble::from_bit_pair(p[5]), |
| 381 |
x, |
| 382 |
DoubleDouble::from_bit_pair(p[4]), |
| 383 |
); |
| 384 |
let e3 = DoubleDouble::mul_add( |
| 385 |
DoubleDouble::from_bit_pair(p[7]), |
| 386 |
x, |
| 387 |
DoubleDouble::from_bit_pair(p[6]), |
| 388 |
); |
| 389 |
let e4 = DoubleDouble::mul_add( |
| 390 |
DoubleDouble::from_bit_pair(p[9]), |
| 391 |
x, |
| 392 |
DoubleDouble::from_bit_pair(p[8]), |
| 393 |
); |
| 394 |
let e5 = DoubleDouble::mul_add( |
| 395 |
DoubleDouble::from_bit_pair(p[11]), |
| 396 |
x, |
| 397 |
DoubleDouble::from_bit_pair(p[10]), |
| 398 |
); |
| 399 |
|
| 400 |
let f0 = DoubleDouble::mul_add(x2, e1, e0); |
| 401 |
let f1 = DoubleDouble::mul_add(x2, e3, e2); |
| 402 |
let f2 = DoubleDouble::mul_add(x2, e5, e4); |
| 403 |
|
| 404 |
let g0 = DoubleDouble::mul_add(x4, f1, f0); |
| 405 |
|
| 406 |
let p_num = DoubleDouble::mul_add(x8, f2, g0); |
| 407 |
|
| 408 |
let rcp = x.recip(); |
| 409 |
let rcp2 = DoubleDouble::quick_mult(rcp, rcp); |
| 410 |
|
| 411 |
let e0 = DoubleDouble::mul_add_f64( |
| 412 |
DoubleDouble::from_bit_pair(q[1]), |
| 413 |
x, |
| 414 |
f64::from_bits(0x3ff0000000000000), |
| 415 |
); |
| 416 |
let e1 = DoubleDouble::mul_add( |
| 417 |
DoubleDouble::from_bit_pair(q[3]), |
| 418 |
x, |
| 419 |
DoubleDouble::from_bit_pair(q[2]), |
| 420 |
); |
| 421 |
let e2 = DoubleDouble::mul_add( |
| 422 |
DoubleDouble::from_bit_pair(q[5]), |
| 423 |
x, |
| 424 |
DoubleDouble::from_bit_pair(q[4]), |
| 425 |
); |
| 426 |
let e3 = DoubleDouble::mul_add( |
| 427 |
DoubleDouble::from_bit_pair(q[7]), |
| 428 |
x, |
| 429 |
DoubleDouble::from_bit_pair(q[6]), |
| 430 |
); |
| 431 |
let e4 = DoubleDouble::mul_add( |
| 432 |
DoubleDouble::from_bit_pair(q[9]), |
| 433 |
x, |
| 434 |
DoubleDouble::from_bit_pair(q[8]), |
| 435 |
); |
| 436 |
let e5 = DoubleDouble::mul_add( |
| 437 |
DoubleDouble::from_bit_pair(q[11]), |
| 438 |
x, |
| 439 |
DoubleDouble::from_bit_pair(q[10]), |
| 440 |
); |
| 441 |
|
| 442 |
let f0 = DoubleDouble::mul_add(x2, e1, e0); |
| 443 |
let f1 = DoubleDouble::mul_add(x2, e3, e2); |
| 444 |
let f2 = DoubleDouble::mul_add(x2, e5, e4); |
| 445 |
|
| 446 |
let g0 = DoubleDouble::mul_add(x4, f1, f0); |
| 447 |
|
| 448 |
let p_den = DoubleDouble::mul_add(x8, f2, g0); |
| 449 |
|
| 450 |
let q = DoubleDouble::div(p_num, p_den); |
| 451 |
let r = DoubleDouble::quick_dd_add(q, rcp2); |
| 452 |
return r; |
| 453 |
} |
| 454 |
// asymptotic expansion Trigamma[x] = 1/x + 1/x^2 + sum(Bernoulli(2*k)/x^(2*k + 1)) |
| 455 |
// Generated in SageMath: |
| 456 |
// var('x') |
| 457 |
// def bernoulli_terms(x, N): |
| 458 |
// S = 0 |
| 459 |
// for k in range(1, N+1): |
| 460 |
// B = bernoulli(2*k) |
| 461 |
// term = B*x**(-(2*k+1)) |
| 462 |
// S += term |
| 463 |
// return S |
| 464 |
// |
| 465 |
// terms = bernoulli_terms(x, 10) |
| 466 |
// coeffs = [RealField(150)(terms.coefficient(x, n)) for n in range(0, terms.degree(x)+1, 1)] |
| 467 |
// for k in range(0, 14): |
| 468 |
// c = terms.coefficient(x, -k) # coefficient of x^(-k) |
| 469 |
// if c == 0: |
| 470 |
// continue |
| 471 |
// print("f64::from_bits(" + double_to_hex(c) + "),") |
| 472 |
const C: [(u64, u64); 10] = [ |
| 473 |
(0x3c65555555555555, 0x3fc5555555555555), |
| 474 |
(0xbc21111111111111, 0xbfa1111111111111), |
| 475 |
(0x3c38618618618618, 0x3f98618618618618), |
| 476 |
(0xbc21111111111111, 0xbfa1111111111111), |
| 477 |
(0xbc4364d9364d9365, 0x3fb364d9364d9365), |
| 478 |
(0xbc6981981981981a, 0xbfd0330330330330), |
| 479 |
(0xbc95555555555555, 0x3ff2aaaaaaaaaaab), |
| 480 |
(0xbcb7979797979798, 0xc01c5e5e5e5e5e5e), |
| 481 |
(0xbcac3b070ec1c3b0, 0x404b7c4f8f13e3c5), |
| 482 |
(0x3cc8d3018d3018d3, 0xc08088fe72cfe72d), |
| 483 |
]; |
| 484 |
|
| 485 |
let rcp = x.recip(); |
| 486 |
|
| 487 |
let q = DoubleDouble::quick_mult(rcp, rcp); |
| 488 |
|
| 489 |
let q2 = DoubleDouble::quick_mult(q, q); |
| 490 |
let q4 = q2 * q2; |
| 491 |
let q8 = q4 * q4; |
| 492 |
|
| 493 |
let e0 = DoubleDouble::quick_mul_add( |
| 494 |
DoubleDouble::from_bit_pair(C[1]), |
| 495 |
q, |
| 496 |
DoubleDouble::from_bit_pair(C[0]), |
| 497 |
); |
| 498 |
let e1 = DoubleDouble::quick_mul_add( |
| 499 |
DoubleDouble::from_bit_pair(C[3]), |
| 500 |
q, |
| 501 |
DoubleDouble::from_bit_pair(C[2]), |
| 502 |
); |
| 503 |
let e2 = DoubleDouble::quick_mul_add( |
| 504 |
DoubleDouble::from_bit_pair(C[5]), |
| 505 |
q, |
| 506 |
DoubleDouble::from_bit_pair(C[4]), |
| 507 |
); |
| 508 |
let e3 = DoubleDouble::quick_mul_add( |
| 509 |
DoubleDouble::from_bit_pair(C[7]), |
| 510 |
q, |
| 511 |
DoubleDouble::from_bit_pair(C[6]), |
| 512 |
); |
| 513 |
let e4 = DoubleDouble::quick_mul_add( |
| 514 |
DoubleDouble::from_bit_pair(C[9]), |
| 515 |
q, |
| 516 |
DoubleDouble::from_bit_pair(C[8]), |
| 517 |
); |
| 518 |
|
| 519 |
let q0 = DoubleDouble::quick_mul_add(q2, e1, e0); |
| 520 |
let q1 = DoubleDouble::quick_mul_add(q2, e3, e2); |
| 521 |
|
| 522 |
let r0 = DoubleDouble::quick_mul_add(q4, q1, q0); |
| 523 |
let mut p = DoubleDouble::quick_mul_add(q8, e4, r0); |
| 524 |
|
| 525 |
let q_over_2 = DoubleDouble::quick_mult_f64(q, 0.5); |
| 526 |
p = DoubleDouble::quick_mult(p, q); |
| 527 |
p = DoubleDouble::quick_mult(p, rcp); |
| 528 |
p = DoubleDouble::quick_dd_add(q_over_2, p); |
| 529 |
p = DoubleDouble::quick_dd_add(p, rcp); |
| 530 |
p |
| 531 |
} |
| 532 |
|
| 533 |
/// Computes the trigamma function ψ₁(x). |
| 534 |
/// |
| 535 |
/// The trigamma function is the second derivative of the logarithm of the gamma function. |
| 536 |
pub fn f_trigamma(x: f64) -> f64 { |
| 537 |
let xb = x.to_bits(); |
| 538 |
if !x.is_normal() { |
| 539 |
if x.is_infinite() { |
| 540 |
return if x.is_sign_negative() { |
| 541 |
f64::NEG_INFINITY |
| 542 |
} else { |
| 543 |
0. |
| 544 |
}; |
| 545 |
} |
| 546 |
if x.is_nan() { |
| 547 |
return f64::NAN; |
| 548 |
} |
| 549 |
if xb == 0 { |
| 550 |
return f64::INFINITY; |
| 551 |
} |
| 552 |
} |
| 553 |
|
| 554 |
let x_e = (x.to_bits() >> 52) & 0x7ff; |
| 555 |
|
| 556 |
const E_BIAS: u64 = (1u64 << (11 - 1u64)) - 1u64; |
| 557 |
|
| 558 |
if x_e < E_BIAS - 52 { |
| 559 |
// |x| < 2^-52 |
| 560 |
let dx = x; |
| 561 |
return 1. / (dx * dx); |
| 562 |
} |
| 563 |
|
| 564 |
if x < 0. { |
| 565 |
if is_integer(x) { |
| 566 |
return f64::INFINITY; |
| 567 |
} |
| 568 |
// reflection formula |
| 569 |
// Trigamma[1-x] + Trigamma[x] = PI^2 / sinpi^2(x) |
| 570 |
const SQR_PI: DoubleDouble = |
| 571 |
DoubleDouble::from_bit_pair((0x3cc692b71366cc05, 0x4023bd3cc9be45de)); // pi^2 |
| 572 |
let sinpi_ax = f_fast_sinpi_dd(-x); |
| 573 |
let dx = DoubleDouble::from_full_exact_sub(1., x); |
| 574 |
let result = DoubleDouble::div(SQR_PI, DoubleDouble::quick_mult(sinpi_ax, sinpi_ax)); |
| 575 |
let trigamma_x = approx_trigamma_dd(dx); |
| 576 |
return DoubleDouble::quick_dd_sub(result, trigamma_x).to_f64(); |
| 577 |
} |
| 578 |
|
| 579 |
approx_trigamma(x).to_f64() |
| 580 |
} |
| 581 |
|
| 582 |
#[cfg(test)] |
| 583 |
mod tests { |
| 584 |
|
| 585 |
use super::*; |
| 586 |
|
| 587 |
#[test] |
| 588 |
fn test_trigamma() { |
| 589 |
assert_eq!(f_trigamma(-27.058018), 300.35629698636757); |
| 590 |
assert_eq!(f_trigamma(27.058018), 0.037648965757704725); |
| 591 |
assert_eq!(f_trigamma(8.058018), 0.13211796975281037); |
| 592 |
assert_eq!(f_trigamma(-8.058018), 300.2758629255111); |
| 593 |
assert_eq!(f_trigamma(2.23432), 0.5621320243666134); |
| 594 |
assert_eq!(f_trigamma(-2.4653), 9.653674003034206); |
| 595 |
assert_eq!(f_trigamma(0.123541), 66.91128231455282); |
| 596 |
assert_eq!(f_trigamma(-0.54331), 9.154415950366596); |
| 597 |
assert_eq!(f_trigamma(-5.), f64::INFINITY); |
| 598 |
assert_eq!(f_trigamma(f64::INFINITY), 0.0); |
| 599 |
assert_eq!(f_trigamma(f64::NEG_INFINITY), f64::NEG_INFINITY); |
| 600 |
assert!(f_trigamma(f64::NAN).is_nan()); |
| 601 |
} |
| 602 |
} |