jyn_asympt.c revision 1.1.1.4
1/* mpfr_jn_asympt, mpfr_yn_asympt -- shared code for mpfr_jn and mpfr_yn
2
3Copyright 2007-2020 Free Software Foundation, Inc.
4Contributed by the AriC and Caramba projects, INRIA.
5
6This file is part of the GNU MPFR Library.
7
8The GNU MPFR Library is free software; you can redistribute it and/or modify
9it under the terms of the GNU Lesser General Public License as published by
10the Free Software Foundation; either version 3 of the License, or (at your
11option) any later version.
12
13The GNU MPFR Library is distributed in the hope that it will be useful, but
14WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
15or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
16License for more details.
17
18You should have received a copy of the GNU Lesser General Public License
19along with the GNU MPFR Library; see the file COPYING.LESSER.  If not, see
20https://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc.,
2151 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. */
22
23#ifdef MPFR_JN
24# define FUNCTION mpfr_jn_asympt
25#else
26# ifdef MPFR_YN
27#  define FUNCTION mpfr_yn_asympt
28# else
29#  error "neither MPFR_JN nor MPFR_YN is defined"
30# endif
31#endif
32
33/* Implements asymptotic expansion for jn or yn (formulae 9.2.5 and 9.2.6
34   from Abramowitz & Stegun).
35   Assumes |z| > p log(2)/2, where p is the target precision
36   (z can be negative only for jn).
37   Return 0 if the expansion does not converge enough (the value 0 as inexact
38   flag should not happen for normal input).
39   Note: for MPFR_RNDF, it returns 0 if the expansion failed, and a non-zero
40   value otherwise (with no other meaning).
41*/
42static int
43FUNCTION (mpfr_ptr res, long n, mpfr_srcptr z, mpfr_rnd_t r)
44{
45  mpfr_t s, c, P, Q, t, iz, err_t, err_s, err_u;
46  mpfr_prec_t w;
47  long k;
48  int inex, stop, diverge = 0;
49  mpfr_exp_t err2, err;
50  MPFR_ZIV_DECL (loop);
51
52  mpfr_init2 (c, 64);
53
54  /* The terms of the asymptotic expansion grow like mu^(2k)/(8z)^(2k), where
55     mu = 4n^2, thus we need mu < 8|z| so that it converges,
56     i.e., n^2/2 < |z| */
57  MPFR_ASSERTD (n >= 0);
58  mpfr_set_ui (c, n, MPFR_RNDU);
59  mpfr_mul_ui (c, c, n, MPFR_RNDU);
60  mpfr_div_2ui (c, c, 1, MPFR_RNDU);
61  if (mpfr_cmpabs (c, z) >= 0)
62    {
63      mpfr_clear (c);
64      return 0; /* asymptotic expansion failed */
65    }
66
67  w = MPFR_PREC(res) + MPFR_INT_CEIL_LOG2(MPFR_PREC(res)) + 4;
68
69  MPFR_ZIV_INIT (loop, w);
70  for (;;)
71    {
72      mpfr_set_prec (c, w);
73      mpfr_init2 (s, w);
74      mpfr_init2 (P, w);
75      mpfr_init2 (Q, w);
76      mpfr_init2 (t, w);
77      mpfr_init2 (iz, w);
78      mpfr_init2 (err_t, 31);
79      mpfr_init2 (err_s, 31);
80      mpfr_init2 (err_u, 31);
81
82      /* Approximate sin(z) and cos(z). In the following, err <= k means that
83         the approximate value y and the true value x are related by
84         y = x * (1 + u)^k with |u| <= 2^(-w), following Higham's method. */
85      mpfr_sin_cos (s, c, z, MPFR_RNDN);
86      if (MPFR_IS_NEG(z))
87        mpfr_neg (s, s, MPFR_RNDN); /* compute jn/yn(|z|), fix sign later */
88      /* The absolute error on s/c is bounded by 1/2 ulp(1/2) <= 2^(-w-1). */
89      mpfr_add (t, s, c, MPFR_RNDN);
90      mpfr_sub (c, s, c, MPFR_RNDN);
91      mpfr_swap (s, t);
92      /* now s approximates sin(z)+cos(z), and c approximates sin(z)-cos(z),
93         with total absolute error bounded by 2^(1-w). */
94
95      /* precompute 1/(8|z|) */
96      mpfr_si_div (iz, MPFR_IS_POS(z) ? 1 : -1, z, MPFR_RNDN);   /* err <= 1 */
97      mpfr_div_2ui (iz, iz, 3, MPFR_RNDN);
98
99      /* compute P and Q */
100      mpfr_set_ui (P, 1, MPFR_RNDN);
101      mpfr_set_ui (Q, 0, MPFR_RNDN);
102      mpfr_set_ui (t, 1, MPFR_RNDN); /* current term */
103      mpfr_set_ui (err_t, 0, MPFR_RNDN); /* error on t */
104      mpfr_set_ui (err_s, 0, MPFR_RNDN); /* error on P and Q (sum of errors) */
105      for (k = 1, stop = 0; stop < 4; k++)
106        {
107          /* compute next term: t(k)/t(k-1) = (2n+2k-1)(2n-2k+1)/(8kz) */
108          MPFR_LOG_MSG (("loop (k,stop) = (%ld,%d)\n", k, stop));
109          mpfr_mul_si (t, t, 2 * (n + k) - 1, MPFR_RNDN); /* err <= err_k + 1 */
110          mpfr_mul_si (t, t, 2 * (n - k) + 1, MPFR_RNDN); /* err <= err_k + 2 */
111          mpfr_div_ui (t, t, k, MPFR_RNDN);               /* err <= err_k + 3 */
112          mpfr_mul (t, t, iz, MPFR_RNDN);                 /* err <= err_k + 5 */
113          /* the relative error on t is bounded by (1+u)^(5k)-1, which is
114             bounded by 6ku for 6ku <= 0.02: first |5 log(1+u)| <= |5.5u|
115             for |u| <= 0.15, then |exp(5.5u)-1| <= 6u for |u| <= 0.02. */
116          mpfr_mul_ui (err_t, t, 6 * k, MPFR_IS_POS(t) ? MPFR_RNDU : MPFR_RNDD);
117          mpfr_abs (err_t, err_t, MPFR_RNDN); /* exact */
118          /* the absolute error on t is bounded by err_t * 2^(-w) */
119          mpfr_abs (err_u, t, MPFR_RNDU);
120          mpfr_mul_2ui (err_u, err_u, w, MPFR_RNDU); /* t * 2^w */
121          mpfr_add (err_u, err_u, err_t, MPFR_RNDU); /* max|t| * 2^w */
122          if (stop >= 2)
123            {
124              /* take into account the neglected terms: t * 2^w */
125              mpfr_div_2ui (err_s, err_s, w, MPFR_RNDU);
126              if (MPFR_IS_POS(t))
127                mpfr_add (err_s, err_s, t, MPFR_RNDU);
128              else
129                mpfr_sub (err_s, err_s, t, MPFR_RNDU);
130              mpfr_mul_2ui (err_s, err_s, w, MPFR_RNDU);
131              stop ++;
132            }
133          /* if k is odd, add to Q, otherwise to P */
134          else if (k & 1)
135            {
136              /* if k = 1 mod 4, add, otherwise subtract */
137              if ((k & 2) == 0)
138                mpfr_add (Q, Q, t, MPFR_RNDN);
139              else
140                mpfr_sub (Q, Q, t, MPFR_RNDN);
141              /* check if the next term is smaller than ulp(Q): if EXP(err_u)
142                 <= EXP(Q), since the current term is bounded by
143                 err_u * 2^(-w), it is bounded by ulp(Q) */
144              if (MPFR_GET_EXP (err_u) <= MPFR_GET_EXP (Q))
145                stop ++;
146              else
147                stop = 0;
148            }
149          else
150            {
151              /* if k = 0 mod 4, add, otherwise subtract */
152              if ((k & 2) == 0)
153                mpfr_add (P, P, t, MPFR_RNDN);
154              else
155                mpfr_sub (P, P, t, MPFR_RNDN);
156              /* check if the next term is smaller than ulp(P) */
157              if (MPFR_GET_EXP (err_u) <= MPFR_GET_EXP (P))
158                stop ++;
159              else
160                stop = 0;
161            }
162          mpfr_add (err_s, err_s, err_t, MPFR_RNDU);
163          /* the sum of the rounding errors on P and Q is bounded by
164             err_s * 2^(-w) */
165
166          /* stop when start to diverge */
167          if (stop < 2 &&
168              ((MPFR_IS_POS(z) && mpfr_cmp_ui (z, (k + 1) / 2) < 0) ||
169               (MPFR_IS_NEG(z) && mpfr_cmp_si (z, - ((k + 1) / 2)) > 0)))
170            {
171              /* if we have to stop the series because it diverges, then
172                 increasing the precision will most probably fail, since
173                 we will stop to the same point, and thus compute a very
174                 similar approximation */
175              diverge = 1;
176              stop = 2; /* force stop */
177            }
178        }
179      /* the sum of the total errors on P and Q is bounded by err_s * 2^(-w) */
180
181      /* Now combine: the sum of the rounding errors on P and Q is bounded by
182         err_s * 2^(-w), and the absolute error on s/c is bounded by 2^(1-w) */
183      if ((n & 1) == 0) /* n even: P * (sin + cos) + Q (cos - sin) for jn
184                                   Q * (sin + cos) + P (sin - cos) for yn */
185        {
186#ifdef MPFR_JN
187          mpfr_mul (c, c, Q, MPFR_RNDN); /* Q * (sin - cos) */
188          mpfr_mul (s, s, P, MPFR_RNDN); /* P * (sin + cos) */
189#else
190          mpfr_mul (c, c, P, MPFR_RNDN); /* P * (sin - cos) */
191          mpfr_mul (s, s, Q, MPFR_RNDN); /* Q * (sin + cos) */
192#endif
193          err = MPFR_GET_EXP (c);
194          if (MPFR_GET_EXP (s) > err)
195            err = MPFR_EXP (s);
196#ifdef MPFR_JN
197          mpfr_sub (s, s, c, MPFR_RNDN);
198#else
199          mpfr_add (s, s, c, MPFR_RNDN);
200#endif
201        }
202      else /* n odd: P * (sin - cos) + Q (cos + sin) for jn,
203                     Q * (sin - cos) - P (cos + sin) for yn */
204        {
205#ifdef MPFR_JN
206          mpfr_mul (c, c, P, MPFR_RNDN); /* P * (sin - cos) */
207          mpfr_mul (s, s, Q, MPFR_RNDN); /* Q * (sin + cos) */
208#else
209          mpfr_mul (c, c, Q, MPFR_RNDN); /* Q * (sin - cos) */
210          mpfr_mul (s, s, P, MPFR_RNDN); /* P * (sin + cos) */
211#endif
212          err = MPFR_GET_EXP (c);
213          if (MPFR_GET_EXP (s) > err)
214            err = MPFR_EXP (s);
215#ifdef MPFR_JN
216          mpfr_add (s, s, c, MPFR_RNDN);
217#else
218          mpfr_sub (s, c, s, MPFR_RNDN);
219#endif
220        }
221      if ((n & 2) != 0)
222        mpfr_neg (s, s, MPFR_RNDN);
223      if (MPFR_GET_EXP (s) > err)
224        err = MPFR_EXP (s);
225      /* the absolute error on s is bounded by P*err(s/c) + Q*err(s/c)
226         + err(P)*(s/c) + err(Q)*(s/c) + 3 * 2^(err - w - 1)
227         <= (|P|+|Q|) * 2^(1-w) + err_s * 2^(1-w) + 2^err * 2^(1-w),
228         since |c|, |old_s| <= 2. */
229      err2 = (MPFR_GET_EXP (P) >= MPFR_GET_EXP (Q))
230        ? MPFR_EXP (P) + 2 : MPFR_EXP (Q) + 2;
231      /* (|P| + |Q|) * 2^(1 - w) <= 2^(err2 - w) */
232      err = MPFR_GET_EXP (err_s) >= err ? MPFR_EXP (err_s) + 2 : err + 2;
233      /* err_s * 2^(1-w) + 2^old_err * 2^(1-w) <= 2^err * 2^(-w) */
234      err2 = (err >= err2) ? err + 1 : err2 + 1;
235      /* now the absolute error on s is bounded by 2^(err2 - w) */
236
237      /* multiply by sqrt(1/(Pi*z)) */
238      mpfr_const_pi (c, MPFR_RNDN);     /* Pi, err <= 1 */
239      mpfr_mul (c, c, z, MPFR_RNDN);    /* err <= 2 */
240      mpfr_si_div (c, MPFR_IS_POS(z) ? 1 : -1, c, MPFR_RNDN); /* err <= 3 */
241      mpfr_sqrt (c, c, MPFR_RNDN);      /* err<=5/2, thus the absolute error is
242                                          bounded by 3*u*|c| for |u| <= 0.25 */
243      mpfr_mul (err_t, c, s, MPFR_SIGN(c)==MPFR_SIGN(s) ? MPFR_RNDU : MPFR_RNDD);
244      mpfr_abs (err_t, err_t, MPFR_RNDU);
245      mpfr_mul_ui (err_t, err_t, 3, MPFR_RNDU);
246      /* 3*2^(-w)*|old_c|*|s| [see below] is bounded by err_t * 2^(-w) */
247      err2 += MPFR_GET_EXP (c);
248      /* |old_c| * 2^(err2 - w) [see below] is bounded by 2^(err2-w) */
249      mpfr_mul (c, c, s, MPFR_RNDN);    /* the absolute error on c is bounded by
250                                          1/2 ulp(c) + 3*2^(-w)*|old_c|*|s|
251                                          + |old_c| * 2^(err2 - w) */
252      /* compute err_t * 2^(-w) + 1/2 ulp(c) = (err_t + 2^EXP(c)) * 2^(-w) */
253      err = (MPFR_GET_EXP (err_t) > MPFR_GET_EXP (c)) ?
254        MPFR_EXP (err_t) + 1 : MPFR_EXP (c) + 1;
255      /* err_t * 2^(-w) + 1/2 ulp(c) <= 2^(err - w) */
256      /* now err_t * 2^(-w) bounds 1/2 ulp(c) + 3*2^(-w)*|old_c|*|s| */
257      err = (err >= err2) ? err + 1 : err2 + 1;
258      /* the absolute error on c is bounded by 2^(err - w) */
259
260      mpfr_clear (s);
261      mpfr_clear (P);
262      mpfr_clear (Q);
263      mpfr_clear (t);
264      mpfr_clear (iz);
265      mpfr_clear (err_t);
266      mpfr_clear (err_s);
267      mpfr_clear (err_u);
268
269      err -= MPFR_GET_EXP (c);
270      if (MPFR_LIKELY (MPFR_CAN_ROUND (c, w - err, MPFR_PREC(res), r)))
271        break;
272      if (diverge != 0)
273        {
274          MPFR_ZIV_FREE (loop);
275          mpfr_clear (c);
276          return 0; /* means that the asymptotic expansion failed */
277        }
278      MPFR_ZIV_NEXT (loop, w);
279    }
280  MPFR_ZIV_FREE (loop);
281
282  inex = (MPFR_IS_POS(z) || ((n & 1) == 0)) ? mpfr_set (res, c, r)
283    : mpfr_neg (res, c, r);
284  mpfr_clear (c);
285
286  /* for RNDF, mpfr_set or mpfr_neg may return 0, but if we return 0, it
287     would mean the asymptotic expansion failed, thus we return 1 instead */
288  return (r != MPFR_RNDF) ? inex : 1;
289}
290