1 | #include <stdio.h>
|
---|
2 | #include <math.h>
|
---|
3 |
|
---|
4 | #include "astro.h"
|
---|
5 |
|
---|
6 | static void precess_hiprec (double mjd1, double mjd2, double *ra, double *dec);
|
---|
7 |
|
---|
8 |
|
---|
9 | #define DCOS(x) cos(degrad(x))
|
---|
10 | #define DSIN(x) sin(degrad(x))
|
---|
11 | #define DASIN(x) raddeg(asin(x))
|
---|
12 | #define DATAN2(y,x) raddeg(atan2((y),(x)))
|
---|
13 |
|
---|
14 | /* corrects ra and dec, both in radians, for precession from epoch 1 to epoch 2.
|
---|
15 | * the epochs are given by their modified JDs, mjd1 and mjd2, respectively.
|
---|
16 | * N.B. ra and dec are modifed IN PLACE.
|
---|
17 | */
|
---|
18 | void
|
---|
19 | precess (
|
---|
20 | double mjd1, double mjd2, /* initial and final epoch modified JDs */
|
---|
21 | double *ra, double *dec) /* ra/dec for mjd1 in, for mjd2 out */
|
---|
22 | {
|
---|
23 | precess_hiprec (mjd1, mjd2, ra, dec);
|
---|
24 | }
|
---|
25 |
|
---|
26 | /*
|
---|
27 | * Copyright (c) 1990 by Craig Counterman. All rights reserved.
|
---|
28 | *
|
---|
29 | * This software may be redistributed freely, not sold.
|
---|
30 | * This copyright notice and disclaimer of warranty must remain
|
---|
31 | * unchanged.
|
---|
32 | *
|
---|
33 | * No representation is made about the suitability of this
|
---|
34 | * software for any purpose. It is provided "as is" without express or
|
---|
35 | * implied warranty, to the extent permitted by applicable law.
|
---|
36 | *
|
---|
37 | * Rigorous precession. From Astronomical Ephemeris 1989, p. B18
|
---|
38 | *
|
---|
39 | * 96-06-20 Hayo Hase <hase@wettzell.ifag.de>: theta_a corrected
|
---|
40 | */
|
---|
41 | static void
|
---|
42 | precess_hiprec (
|
---|
43 | double mjd1, double mjd2, /* initial and final epoch modified JDs */
|
---|
44 | double *ra, double *dec) /* ra/dec for mjd1 in, for mjd2 out */
|
---|
45 | {
|
---|
46 | static double last_mjd1 = -213.432, last_from;
|
---|
47 | static double last_mjd2 = -213.432, last_to;
|
---|
48 | double zeta_A, z_A, theta_A;
|
---|
49 | double T;
|
---|
50 | double A, B, C;
|
---|
51 | double alpha, delta;
|
---|
52 | double alpha_in, delta_in;
|
---|
53 | double from_equinox, to_equinox;
|
---|
54 | double alpha2000, delta2000;
|
---|
55 |
|
---|
56 | /* convert mjds to years;
|
---|
57 | * avoid the remarkably expensive calls to mjd_year()
|
---|
58 | */
|
---|
59 | if (last_mjd1 == mjd1)
|
---|
60 | from_equinox = last_from;
|
---|
61 | else {
|
---|
62 | mjd_year (mjd1, &from_equinox);
|
---|
63 | last_mjd1 = mjd1;
|
---|
64 | last_from = from_equinox;
|
---|
65 | }
|
---|
66 | if (last_mjd2 == mjd2)
|
---|
67 | to_equinox = last_to;
|
---|
68 | else {
|
---|
69 | mjd_year (mjd2, &to_equinox);
|
---|
70 | last_mjd2 = mjd2;
|
---|
71 | last_to = to_equinox;
|
---|
72 | }
|
---|
73 |
|
---|
74 | /* convert coords in rads to degs */
|
---|
75 | alpha_in = raddeg(*ra);
|
---|
76 | delta_in = raddeg(*dec);
|
---|
77 |
|
---|
78 | /* precession progresses about 1 arc second in .047 years */
|
---|
79 | /* From from_equinox to 2000.0 */
|
---|
80 | if (fabs (from_equinox-2000.0) > .02) {
|
---|
81 | T = (from_equinox - 2000.0)/100.0;
|
---|
82 | zeta_A = 0.6406161* T + 0.0000839* T*T + 0.0000050* T*T*T;
|
---|
83 | z_A = 0.6406161* T + 0.0003041* T*T + 0.0000051* T*T*T;
|
---|
84 | theta_A = 0.5567530* T - 0.0001185* T*T - 0.0000116* T*T*T;
|
---|
85 |
|
---|
86 | A = DSIN(alpha_in - z_A) * DCOS(delta_in);
|
---|
87 | B = DCOS(alpha_in - z_A) * DCOS(theta_A) * DCOS(delta_in)
|
---|
88 | + DSIN(theta_A) * DSIN(delta_in);
|
---|
89 | C = -DCOS(alpha_in - z_A) * DSIN(theta_A) * DCOS(delta_in)
|
---|
90 | + DCOS(theta_A) * DSIN(delta_in);
|
---|
91 |
|
---|
92 | alpha2000 = DATAN2(A,B) - zeta_A;
|
---|
93 | range (&alpha2000, 360.0);
|
---|
94 | delta2000 = DASIN(C);
|
---|
95 | } else {
|
---|
96 | /* should get the same answer, but this could improve accruacy */
|
---|
97 | alpha2000 = alpha_in;
|
---|
98 | delta2000 = delta_in;
|
---|
99 | };
|
---|
100 |
|
---|
101 |
|
---|
102 | /* From 2000.0 to to_equinox */
|
---|
103 | if (fabs (to_equinox - 2000.0) > .02) {
|
---|
104 | T = (to_equinox - 2000.0)/100.0;
|
---|
105 | zeta_A = 0.6406161* T + 0.0000839* T*T + 0.0000050* T*T*T;
|
---|
106 | z_A = 0.6406161* T + 0.0003041* T*T + 0.0000051* T*T*T;
|
---|
107 | theta_A = 0.5567530* T - 0.0001185* T*T - 0.0000116* T*T*T;
|
---|
108 |
|
---|
109 | A = DSIN(alpha2000 + zeta_A) * DCOS(delta2000);
|
---|
110 | B = DCOS(alpha2000 + zeta_A) * DCOS(theta_A) * DCOS(delta2000)
|
---|
111 | - DSIN(theta_A) * DSIN(delta2000);
|
---|
112 | C = DCOS(alpha2000 + zeta_A) * DSIN(theta_A) * DCOS(delta2000)
|
---|
113 | + DCOS(theta_A) * DSIN(delta2000);
|
---|
114 |
|
---|
115 | alpha = DATAN2(A,B) + z_A;
|
---|
116 | range(&alpha, 360.0);
|
---|
117 | delta = DASIN(C);
|
---|
118 | } else {
|
---|
119 | /* should get the same answer, but this could improve accruacy */
|
---|
120 | alpha = alpha2000;
|
---|
121 | delta = delta2000;
|
---|
122 | };
|
---|
123 |
|
---|
124 | *ra = degrad(alpha);
|
---|
125 | *dec = degrad(delta);
|
---|
126 | }
|
---|
127 |
|
---|
128 | #if 0
|
---|
129 | static void
|
---|
130 | precess_fast (
|
---|
131 | double mjd1, double mjd2, /* initial and final epoch modified JDs */
|
---|
132 | double *ra, double *dec) /* ra/dec for mjd1 in, for mjd2 out */
|
---|
133 | {
|
---|
134 | #define N degrad (20.0468/3600.0)
|
---|
135 | #define M hrrad (3.07234/3600.0)
|
---|
136 | double nyrs;
|
---|
137 |
|
---|
138 | nyrs = (mjd2 - mjd1)/365.2425;
|
---|
139 | *dec += N * cos(*ra) * nyrs;
|
---|
140 | *ra += (M + (N * sin(*ra) * tan(*dec))) * nyrs;
|
---|
141 | range (ra, 2.0*PI);
|
---|
142 | }
|
---|
143 | #endif
|
---|
144 |
|
---|
145 | /* For RCS Only -- Do Not Edit */
|
---|
146 | static char *rcsid[2] = {(char *)rcsid, "@(#) $RCSfile: precess.c,v $ $Date: 2006-11-22 13:53:30 $ $Revision: 1.6 $ $Name: not supported by cvs2svn $"};
|
---|