| 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::{f_fmla, is_integerf}; |
| 30 |
use crate::polyeval::{f_polyeval6, f_polyeval10}; |
| 31 |
use crate::sin_cosf::fast_sinpif; |
| 32 |
|
| 33 |
#[inline] |
| 34 |
fn approx_trigamma(x: f64) -> f64 { |
| 35 |
if x <= 1. { |
| 36 |
// Polynomial for Trigamma[x+1] |
| 37 |
// <<FunctionApproximations` |
| 38 |
// ClearAll["Global`*"] |
| 39 |
// f[x_]:=PolyGamma[1, x+1] |
| 40 |
// {err0,approx,err1}=MiniMaxApproximation[f[z],{z,{-0.99999999,0},9,9},WorkingPrecision->75,MaxIterations->100] |
| 41 |
// num=Numerator[approx]; |
| 42 |
// den=Denominator[approx]; |
| 43 |
// poly=num; |
| 44 |
// coeffs=CoefficientList[poly,z]; |
| 45 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 46 |
// poly=den; |
| 47 |
// coeffs=CoefficientList[poly,z]; |
| 48 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 49 |
let p_num = f_polyeval10( |
| 50 |
x, |
| 51 |
f64::from_bits(0x3ffa51a6625307d3), |
| 52 |
f64::from_bits(0x40142fc9c4f02b3f), |
| 53 |
f64::from_bits(0x401b30a762805b44), |
| 54 |
f64::from_bits(0x4014dc84c95656bd), |
| 55 |
f64::from_bits(0x4003e44f4c820b4c), |
| 56 |
f64::from_bits(0x3fe81f37523197d3), |
| 57 |
f64::from_bits(0x3fc22bffe2490221), |
| 58 |
f64::from_bits(0x3f8f221a6329ea36), |
| 59 |
f64::from_bits(0x3f47406930b9563c), |
| 60 |
f64::from_bits(0xbd99cd44c6ad497a), |
| 61 |
); |
| 62 |
let p_den = f_polyeval10( |
| 63 |
x, |
| 64 |
f64::from_bits(0x3ff0000000000000), |
| 65 |
f64::from_bits(0x40121e3db4e0a2f3), |
| 66 |
f64::from_bits(0x40218e97a5430c4f), |
| 67 |
f64::from_bits(0x402329897737b159), |
| 68 |
f64::from_bits(0x401a0fdc27807c2d), |
| 69 |
f64::from_bits(0x4006ff242e1f3a51), |
| 70 |
f64::from_bits(0x3fea6eda129c4e85), |
| 71 |
f64::from_bits(0x3fc32700b2ae2e88), |
| 72 |
f64::from_bits(0x3f8fdc1dc6116d41), |
| 73 |
f64::from_bits(0x3f4740690261cfbc), |
| 74 |
); |
| 75 |
return p_num / p_den + 1. / (x * x); |
| 76 |
} else if x <= 4. { |
| 77 |
// Polynomial for Trigamma[x+1] |
| 78 |
// <<FunctionApproximations` |
| 79 |
// ClearAll["Global`*"] |
| 80 |
// f[x_]:=PolyGamma[1, x+1] |
| 81 |
// {err0,approx,err1}=MiniMaxApproximation[f[z],{z,{0,3},9,9},WorkingPrecision->75,MaxIterations->100] |
| 82 |
// num=Numerator[approx]; |
| 83 |
// den=Denominator[approx]; |
| 84 |
// poly=num; |
| 85 |
// coeffs=CoefficientList[poly,z]; |
| 86 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 87 |
// poly=den; |
| 88 |
// coeffs=CoefficientList[poly,z]; |
| 89 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 90 |
let p_num = f_polyeval10( |
| 91 |
x, |
| 92 |
f64::from_bits(0x3ffa51a6625307d3), |
| 93 |
f64::from_bits(0x4015167fa2d2b5a8), |
| 94 |
f64::from_bits(0x401dd40865d5985e), |
| 95 |
f64::from_bits(0x4018353d9425fb58), |
| 96 |
f64::from_bits(0x4008a12aa45851fa), |
| 97 |
f64::from_bits(0x3ff018736d0c5dbe), |
| 98 |
f64::from_bits(0x3fca715702bdb519), |
| 99 |
f64::from_bits(0x3f9908a9d73d983c), |
| 100 |
f64::from_bits(0x3f54fd9cbbb46314), |
| 101 |
f64::from_bits(0xbd00b8a28c578ab5), |
| 102 |
); |
| 103 |
let p_den = f_polyeval10( |
| 104 |
x, |
| 105 |
f64::from_bits(0x3ff0000000000000), |
| 106 |
f64::from_bits(0x4012aa7f041a768b), |
| 107 |
f64::from_bits(0x4022c2604e5f9c7a), |
| 108 |
f64::from_bits(0x4025655b63c2db22), |
| 109 |
f64::from_bits(0x401eaa8e59c8295d), |
| 110 |
f64::from_bits(0x400cc8724a58809c), |
| 111 |
f64::from_bits(0x3ff1c7a91c8e3c40), |
| 112 |
f64::from_bits(0x3fcc05613a11183e), |
| 113 |
f64::from_bits(0x3f99b096bd3ce542), |
| 114 |
f64::from_bits(0x3f54fd9cbb9c6167), |
| 115 |
); |
| 116 |
return p_num / p_den + 1. / (x * x); |
| 117 |
} else if x <= 10. { |
| 118 |
// Polynomial for Trigamma[x+1] |
| 119 |
// <<FunctionApproximations` |
| 120 |
// ClearAll["Global`*"] |
| 121 |
// f[x_]:=PolyGamma[1, x+1] |
| 122 |
// {err0,approx,err1}=MiniMaxApproximation[f[z],{z,{3,9},9,9},WorkingPrecision->75,MaxIterations->100] |
| 123 |
// num=Numerator[approx]; |
| 124 |
// den=Denominator[approx]; |
| 125 |
// poly=num; |
| 126 |
// coeffs=CoefficientList[poly,z]; |
| 127 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 128 |
// poly=den; |
| 129 |
// coeffs=CoefficientList[poly,z]; |
| 130 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 131 |
let p_num = f_polyeval10( |
| 132 |
x, |
| 133 |
f64::from_bits(0x3ffa51a664b1b211), |
| 134 |
f64::from_bits(0x4016d7f75881312a), |
| 135 |
f64::from_bits(0x4021b10defb47bcc), |
| 136 |
f64::from_bits(0x401fe9633665e1bf), |
| 137 |
f64::from_bits(0x4012601cce6766d7), |
| 138 |
f64::from_bits(0x3ffbd0ece1c435f1), |
| 139 |
f64::from_bits(0x3fdb3fd0e233c485), |
| 140 |
f64::from_bits(0x3faffdedea90b870), |
| 141 |
f64::from_bits(0x3f71a4bbf0d00147), |
| 142 |
f64::from_bits(0xbc0aae286498a357), |
| 143 |
); |
| 144 |
let p_den = f_polyeval10( |
| 145 |
x, |
| 146 |
f64::from_bits(0x3ff0000000000000), |
| 147 |
f64::from_bits(0x4013bbbd604d685d), |
| 148 |
f64::from_bits(0x40253a4fca05438e), |
| 149 |
f64::from_bits(0x402a4aba4634f14f), |
| 150 |
f64::from_bits(0x4024cdd23bd6284a), |
| 151 |
f64::from_bits(0x4015fbe371275c3f), |
| 152 |
f64::from_bits(0x3fff4d7ebf7d1ed0), |
| 153 |
f64::from_bits(0x3fdd459154d7bc72), |
| 154 |
f64::from_bits(0x3fb08c1cd4cedca3), |
| 155 |
f64::from_bits(0x3f71a4bbf0d0012d), |
| 156 |
); |
| 157 |
return p_num / p_den + 1. / (x * x); |
| 158 |
} |
| 159 |
// asymptotic expansion Trigamma[x] = 1/x + 1/x^2 + sum(Bernoulli(2*k)/x^(2*k + 1)) |
| 160 |
// Generated in SageMath: |
| 161 |
// var('x') |
| 162 |
// def bernoulli_terms(x, N): |
| 163 |
// S = 0 |
| 164 |
// for k in range(1, N+1): |
| 165 |
// B = bernoulli(2*k) |
| 166 |
// term = B*x**(-(2*k+1)) |
| 167 |
// S += term |
| 168 |
// return S |
| 169 |
// |
| 170 |
// terms = bernoulli_terms(x, 7) |
| 171 |
// coeffs = [RealField(150)(terms.coefficient(x, n)) for n in range(0, terms.degree(x)+1, 1)] |
| 172 |
// for k in range(0, 14): |
| 173 |
// c = terms.coefficient(x, -k) # coefficient of x^(-k) |
| 174 |
// if c == 0: |
| 175 |
// continue |
| 176 |
// print("f64::from_bits(" + double_to_hex(c) + "),") |
| 177 |
let r = 1. / x; |
| 178 |
let r2 = r * r; |
| 179 |
let p = f_polyeval6( |
| 180 |
r2, |
| 181 |
f64::from_bits(0x3fc5555555555555), |
| 182 |
f64::from_bits(0xbfa1111111111111), |
| 183 |
f64::from_bits(0x3f98618618618618), |
| 184 |
f64::from_bits(0xbfa1111111111111), |
| 185 |
f64::from_bits(0x3fb364d9364d9365), |
| 186 |
f64::from_bits(0xbfd0330330330330), |
| 187 |
); |
| 188 |
f_fmla(p, r2 * r, f_fmla(r2, 0.5, r)) |
| 189 |
} |
| 190 |
|
| 191 |
/// Computes the trigamma function ψ₁(x). |
| 192 |
/// |
| 193 |
/// The trigamma function is the second derivative of the logarithm of the gamma function. |
| 194 |
pub fn f_trigammaf(x: f32) -> f32 { |
| 195 |
let xb = x.to_bits(); |
| 196 |
// filter out exceptional cases |
| 197 |
if xb >= 0xffu32 << 23 || xb == 0 { |
| 198 |
if x.is_infinite() { |
| 199 |
return if x.is_sign_negative() { |
| 200 |
f32::NEG_INFINITY |
| 201 |
} else { |
| 202 |
0. |
| 203 |
}; |
| 204 |
} |
| 205 |
if x.is_nan() { |
| 206 |
return f32::NAN; |
| 207 |
} |
| 208 |
if xb.wrapping_shl(1) == 0 { |
| 209 |
return f32::INFINITY; |
| 210 |
} |
| 211 |
} |
| 212 |
|
| 213 |
let ax = x.to_bits() & 0x7fff_ffff; |
| 214 |
|
| 215 |
if ax <= 0x34000000u32 { |
| 216 |
// |x| < f32::EPSILON |
| 217 |
let dx = x as f64; |
| 218 |
return (1. / (dx * dx)) as f32; |
| 219 |
} |
| 220 |
|
| 221 |
let mut dx = x as f64; |
| 222 |
|
| 223 |
let mut result = 0.; |
| 224 |
let mut sum_parity: f64 = 1.0; |
| 225 |
|
| 226 |
if x < 0. { |
| 227 |
// singularity at negative integers |
| 228 |
if is_integerf(x) { |
| 229 |
return f32::INFINITY; |
| 230 |
} |
| 231 |
// reflection formula |
| 232 |
// Trigamma[1-x] + Trigamma[x] = PI^2 / sinpi^2(x) |
| 233 |
const SQR_PI: f64 = f64::from_bits(0x4023bd3cc9be45de); // pi^2 |
| 234 |
let sinpi_ax = fast_sinpif(-x); |
| 235 |
dx = 1. - dx; |
| 236 |
result = SQR_PI / (sinpi_ax * sinpi_ax); |
| 237 |
sum_parity = -1.; |
| 238 |
} |
| 239 |
|
| 240 |
let r = approx_trigamma(dx) * sum_parity; |
| 241 |
result += r; |
| 242 |
result as f32 |
| 243 |
} |
| 244 |
|
| 245 |
#[cfg(test)] |
| 246 |
mod tests { |
| 247 |
use crate::f_trigammaf; |
| 248 |
|
| 249 |
#[test] |
| 250 |
fn test_trigamma() { |
| 251 |
assert_eq!(f_trigammaf(-27.058018), 300.35904); |
| 252 |
assert_eq!(f_trigammaf(27.058018), 0.037648965); |
| 253 |
assert_eq!(f_trigammaf(8.058018), 0.13211797); |
| 254 |
assert_eq!(f_trigammaf(-8.058018), 300.27863); |
| 255 |
assert_eq!(f_trigammaf(2.23432), 0.56213206); |
| 256 |
assert_eq!(f_trigammaf(-2.4653), 9.653673); |
| 257 |
assert_eq!(f_trigammaf(0.123541), 66.911285); |
| 258 |
assert_eq!(f_trigammaf(-0.54331), 9.154416); |
| 259 |
assert_eq!(f_trigammaf(-5.), f32::INFINITY); |
| 260 |
assert_eq!(f_trigammaf(-0.), f32::INFINITY); |
| 261 |
assert_eq!(f_trigammaf(0.), f32::INFINITY); |
| 262 |
assert_eq!(f_trigammaf(f32::INFINITY), 0.0); |
| 263 |
assert_eq!(f_trigammaf(f32::NEG_INFINITY), f32::NEG_INFINITY); |
| 264 |
assert!(f_trigammaf(f32::NAN).is_nan()); |
| 265 |
} |
| 266 |
} |