summaryrefslogtreecommitdiff
path: root/src/erf.c
blob: 672378e399115a4f099e86610c17c49da1bd46d3 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
/* mpfr_erf -- error function of a floating-point number

Copyright 2001, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012 Free Software Foundation, Inc.
Contributed by the Arenaire and Caramel projects, INRIA.

This file is part of the GNU MPFR Library.

The GNU MPFR Library is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 3 of the License, or (at your
option) any later version.

The GNU MPFR Library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
License for more details.

You should have received a copy of the GNU Lesser General Public License
along with the GNU MPFR Library; see the file COPYING.LESSER.  If not, see
http://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc.,
51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. */

#define MPFR_NEED_LONGLONG_H
#include "mpfr-impl.h"

#define EXP1 2.71828182845904523536 /* exp(1) */

static int mpfr_erf_0 (mpfr_ptr, mpfr_srcptr, double, mpfr_rnd_t);

int
mpfr_erf (mpfr_ptr y, mpfr_srcptr x, mpfr_rnd_t rnd_mode)
{
  mpfr_t xf;
  int inex, large;
  MPFR_SAVE_EXPO_DECL (expo);

  MPFR_LOG_FUNC
    (("x[%Pu]=%.*Rg rnd=%d", mpfr_get_prec (x), mpfr_log_prec, x, rnd_mode),
     ("y[%Pu]=%.*Rg inexact=%d", mpfr_get_prec (y), mpfr_log_prec, y, inex));

  if (MPFR_UNLIKELY (MPFR_IS_SINGULAR (x)))
    {
      if (MPFR_IS_NAN (x))
        {
          MPFR_SET_NAN (y);
          MPFR_RET_NAN;
        }
      else if (MPFR_IS_INF (x)) /* erf(+inf) = +1, erf(-inf) = -1 */
        return mpfr_set_si (y, MPFR_INT_SIGN (x), MPFR_RNDN);
      else /* erf(+0) = +0, erf(-0) = -0 */
        {
          MPFR_ASSERTD (MPFR_IS_ZERO (x));
          return mpfr_set (y, x, MPFR_RNDN); /* should keep the sign of x */
        }
    }

  /* now x is neither NaN, Inf nor 0 */

  /* first try expansion at x=0 when x is small, or asymptotic expansion
     where x is large */

  MPFR_SAVE_EXPO_MARK (expo);

  /* around x=0, we have erf(x) = 2x/sqrt(Pi) (1 - x^2/3 + ...),
     with 1 - x^2/3 <= sqrt(Pi)*erf(x)/2/x <= 1 for x >= 0. This means that
     if x^2/3 < 2^(-PREC(y)-1) we can decide of the correct rounding,
     unless we have a worst-case for 2x/sqrt(Pi). */
  if (MPFR_EXP(x) < - (mpfr_exp_t) (MPFR_PREC(y) / 2))
    {
      /* we use 2x/sqrt(Pi) (1 - x^2/3) <= erf(x) <= 2x/sqrt(Pi) for x > 0
         and 2x/sqrt(Pi) <= erf(x) <= 2x/sqrt(Pi) (1 - x^2/3) for x < 0.
         In both cases |2x/sqrt(Pi) (1 - x^2/3)| <= |erf(x)| <= |2x/sqrt(Pi)|.
         We will compute l and h such that l <= |2x/sqrt(Pi) (1 - x^2/3)|
         and |2x/sqrt(Pi)| <= h. If l and h round to the same value to
         precision PREC(y) and rounding rnd_mode, then we are done. */
      mpfr_t l, h; /* lower and upper bounds for erf(x) */
      int ok, inex2;

      mpfr_init2 (l, MPFR_PREC(y) + 17);
      mpfr_init2 (h, MPFR_PREC(y) + 17);
      /* first compute l */
      mpfr_mul (l, x, x, MPFR_RNDU);
      mpfr_div_ui (l, l, 3, MPFR_RNDU); /* upper bound on x^2/3 */
      mpfr_ui_sub (l, 1, l, MPFR_RNDZ); /* lower bound on 1 - x^2/3 */
      mpfr_const_pi (h, MPFR_RNDU); /* upper bound of Pi */
      mpfr_sqrt (h, h, MPFR_RNDU); /* upper bound on sqrt(Pi) */
      mpfr_div (l, l, h, MPFR_RNDZ); /* lower bound on 1/sqrt(Pi) (1 - x^2/3) */
      mpfr_mul_2ui (l, l, 1, MPFR_RNDZ); /* 2/sqrt(Pi) (1 - x^2/3) */
      mpfr_mul (l, l, x, MPFR_RNDZ); /* |l| is a lower bound on
                                       |2x/sqrt(Pi) (1 - x^2/3)| */
      /* now compute h */
      mpfr_const_pi (h, MPFR_RNDD); /* lower bound on Pi */
      mpfr_sqrt (h, h, MPFR_RNDD); /* lower bound on sqrt(Pi) */
      mpfr_div_2ui (h, h, 1, MPFR_RNDD); /* lower bound on sqrt(Pi)/2 */
      /* since sqrt(Pi)/2 < 1, the following should not underflow */
      mpfr_div (h, x, h, MPFR_IS_POS(x) ? MPFR_RNDU : MPFR_RNDD);
      /* round l and h to precision PREC(y) */
      inex = mpfr_prec_round (l, MPFR_PREC(y), rnd_mode);
      inex2 = mpfr_prec_round (h, MPFR_PREC(y), rnd_mode);
      /* Caution: we also need inex=inex2 (inex might be 0). */
      ok = SAME_SIGN (inex, inex2) && mpfr_cmp (l, h) == 0;
      if (ok)
        mpfr_set (y, h, rnd_mode);
      mpfr_clear (l);
      mpfr_clear (h);
      if (ok)
        goto end;
      /* this test can still fail for small precision, for example
         for x=-0.100E-2 with a target precision of 3 bits, since
         the error term x^2/3 is not that small. */
    }

  mpfr_init2 (xf, 53);
  mpfr_const_log2 (xf, MPFR_RNDU);
  mpfr_div (xf, x, xf, MPFR_RNDZ); /* round to zero ensures we get a lower
                                     bound of |x/log(2)| */
  mpfr_mul (xf, xf, x, MPFR_RNDZ);
  large = mpfr_cmp_ui (xf, MPFR_PREC (y) + 1) > 0;
  mpfr_clear (xf);

  /* when x goes to infinity, we have erf(x) = 1 - 1/sqrt(Pi)/exp(x^2)/x + ...
     and |erf(x) - 1| <= exp(-x^2) is true for any x >= 0, thus if
     exp(-x^2) < 2^(-PREC(y)-1) the result is 1 or 1-epsilon.
     This rewrites as x^2/log(2) > p+1. */
  if (MPFR_UNLIKELY (large))
    /* |erf x| = 1 or 1- */
    {
      mpfr_rnd_t rnd2 = MPFR_IS_POS (x) ? rnd_mode : MPFR_INVERT_RND(rnd_mode);
      if (rnd2 == MPFR_RNDN || rnd2 == MPFR_RNDU || rnd2 == MPFR_RNDA)
        {
          inex = MPFR_INT_SIGN (x);
          mpfr_set_si (y, inex, rnd2);
        }
      else /* round to zero */
        {
          inex = -MPFR_INT_SIGN (x);
          mpfr_setmax (y, 0); /* warning: setmax keeps the old sign of y */
          MPFR_SET_SAME_SIGN (y, x);
        }
    }
  else  /* use Taylor */
    {
      double xf2;

      /* FIXME: get rid of doubles/mpfr_get_d here */
      xf2 = mpfr_get_d (x, MPFR_RNDN);
      xf2 = xf2 * xf2; /* xf2 ~ x^2 */
      inex = mpfr_erf_0 (y, x, xf2, rnd_mode);
    }

 end:
  MPFR_SAVE_EXPO_FREE (expo);
  return mpfr_check_range (y, inex, rnd_mode);
}

/* return x*2^e */
static double
mul_2exp (double x, mpfr_exp_t e)
{
  if (e > 0)
    {
      while (e--)
        x *= 2.0;
    }
  else
    {
      while (e++)
        x /= 2.0;
    }

  return x;
}

/* evaluates erf(x) using the expansion at x=0:

   erf(x) = 2/sqrt(Pi) * sum((-1)^k*x^(2k+1)/k!/(2k+1), k=0..infinity)

   Assumes x is neither NaN nor infinite nor zero.
   Assumes also that e*x^2 <= n (target precision).
 */
static int
mpfr_erf_0 (mpfr_ptr res, mpfr_srcptr x, double xf2, mpfr_rnd_t rnd_mode)
{
  mpfr_prec_t n, m;
  mpfr_exp_t nuk, sigmak;
  double tauk;
  mpfr_t y, s, t, u;
  unsigned int k;
  int log2tauk;
  int inex;
  MPFR_ZIV_DECL (loop);

  n = MPFR_PREC (res); /* target precision */

  /* initial working precision */
  m = n + (mpfr_prec_t) (xf2 / LOG2) + 8 + MPFR_INT_CEIL_LOG2 (n);

  mpfr_init2 (y, m);
  mpfr_init2 (s, m);
  mpfr_init2 (t, m);
  mpfr_init2 (u, m);

  MPFR_ZIV_INIT (loop, m);
  for (;;)
    {
      mpfr_mul (y, x, x, MPFR_RNDU); /* err <= 1 ulp */
      mpfr_set_ui (s, 1, MPFR_RNDN);
      mpfr_set_ui (t, 1, MPFR_RNDN);
      tauk = 0.0;

      for (k = 1; ; k++)
        {
          mpfr_mul (t, y, t, MPFR_RNDU);
          mpfr_div_ui (t, t, k, MPFR_RNDU);
          mpfr_div_ui (u, t, 2 * k + 1, MPFR_RNDU);
          sigmak = MPFR_GET_EXP (s);
          if (k % 2)
            mpfr_sub (s, s, u, MPFR_RNDN);
          else
            mpfr_add (s, s, u, MPFR_RNDN);
          sigmak -= MPFR_GET_EXP(s);
          nuk = MPFR_GET_EXP(u) - MPFR_GET_EXP(s);

          if ((nuk < - (mpfr_exp_t) m) && ((double) k >= xf2))
            break;

          /* tauk <- 1/2 + tauk * 2^sigmak + (1+8k)*2^nuk */
          tauk = 0.5 + mul_2exp (tauk, sigmak)
            + mul_2exp (1.0 + 8.0 * (double) k, nuk);
        }

      mpfr_mul (s, x, s, MPFR_RNDU);
      MPFR_SET_EXP (s, MPFR_GET_EXP (s) + 1);

      mpfr_const_pi (t, MPFR_RNDZ);
      mpfr_sqrt (t, t, MPFR_RNDZ);
      mpfr_div (s, s, t, MPFR_RNDN);
      tauk = 4.0 * tauk + 11.0; /* final ulp-error on s */
      log2tauk = __gmpfr_ceil_log2 (tauk);

      if (MPFR_LIKELY (MPFR_CAN_ROUND (s, m - log2tauk, n, rnd_mode)))
        break;

      /* Actualisation of the precision */
      MPFR_ZIV_NEXT (loop, m);
      mpfr_set_prec (y, m);
      mpfr_set_prec (s, m);
      mpfr_set_prec (t, m);
      mpfr_set_prec (u, m);

    }
  MPFR_ZIV_FREE (loop);

  inex = mpfr_set (res, s, rnd_mode);

  mpfr_clear (y);
  mpfr_clear (t);
  mpfr_clear (u);
  mpfr_clear (s);

  return inex;
}