1 | // (C) Copyright John Maddock 2005. |
---|
2 | // Distributed under the Boost Software License, Version 1.0. (See accompanying |
---|
3 | // file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt) |
---|
4 | |
---|
5 | #ifndef BOOST_MATH_COMPLEX_ASIN_INCLUDED |
---|
6 | #define BOOST_MATH_COMPLEX_ASIN_INCLUDED |
---|
7 | |
---|
8 | #ifndef BOOST_MATH_COMPLEX_DETAILS_INCLUDED |
---|
9 | # include <boost/math/complex/details.hpp> |
---|
10 | #endif |
---|
11 | #ifndef BOOST_MATH_LOG1P_INCLUDED |
---|
12 | # include <boost/math/special_functions/log1p.hpp> |
---|
13 | #endif |
---|
14 | #include <boost/assert.hpp> |
---|
15 | |
---|
16 | #ifdef BOOST_NO_STDC_NAMESPACE |
---|
17 | namespace std{ using ::sqrt; using ::fabs; using ::acos; using ::asin; using ::atan; using ::atan2; } |
---|
18 | #endif |
---|
19 | |
---|
20 | namespace boost{ namespace math{ |
---|
21 | |
---|
22 | template<class T> |
---|
23 | inline std::complex<T> asin(const std::complex<T>& z) |
---|
24 | { |
---|
25 | // |
---|
26 | // This implementation is a transcription of the pseudo-code in: |
---|
27 | // |
---|
28 | // "Implementing the complex Arcsine and Arccosine Functions using Exception Handling." |
---|
29 | // T E Hull, Thomas F Fairgrieve and Ping Tak Peter Tang. |
---|
30 | // ACM Transactions on Mathematical Software, Vol 23, No 3, Sept 1997. |
---|
31 | // |
---|
32 | |
---|
33 | // |
---|
34 | // These static constants should really be in a maths constants library: |
---|
35 | // |
---|
36 | static const T one = static_cast<T>(1); |
---|
37 | //static const T two = static_cast<T>(2); |
---|
38 | static const T half = static_cast<T>(0.5L); |
---|
39 | static const T a_crossover = static_cast<T>(1.5L); |
---|
40 | static const T b_crossover = static_cast<T>(0.6417L); |
---|
41 | //static const T pi = static_cast<T>(3.141592653589793238462643383279502884197L); |
---|
42 | static const T half_pi = static_cast<T>(1.57079632679489661923132169163975144L); |
---|
43 | static const T log_two = static_cast<T>(0.69314718055994530941723212145817657L); |
---|
44 | static const T quarter_pi = static_cast<T>(0.78539816339744830961566084581987572L); |
---|
45 | |
---|
46 | // |
---|
47 | // Get real and imaginary parts, discard the signs as we can |
---|
48 | // figure out the sign of the result later: |
---|
49 | // |
---|
50 | T x = std::fabs(z.real()); |
---|
51 | T y = std::fabs(z.imag()); |
---|
52 | T real, imag; // our results |
---|
53 | |
---|
54 | // |
---|
55 | // Begin by handling the special cases for infinities and nan's |
---|
56 | // specified in C99, most of this is handled by the regular logic |
---|
57 | // below, but handling it as a special case prevents overflow/underflow |
---|
58 | // arithmetic which may trip up some machines: |
---|
59 | // |
---|
60 | if(detail::test_is_nan(x)) |
---|
61 | { |
---|
62 | if(detail::test_is_nan(y)) |
---|
63 | return std::complex<T>(x, x); |
---|
64 | if(std::numeric_limits<T>::has_infinity && (y == std::numeric_limits<T>::infinity())) |
---|
65 | { |
---|
66 | real = x; |
---|
67 | imag = std::numeric_limits<T>::infinity(); |
---|
68 | } |
---|
69 | else |
---|
70 | return std::complex<T>(x, x); |
---|
71 | } |
---|
72 | else if(detail::test_is_nan(y)) |
---|
73 | { |
---|
74 | if(x == 0) |
---|
75 | { |
---|
76 | real = 0; |
---|
77 | imag = y; |
---|
78 | } |
---|
79 | else if(std::numeric_limits<T>::has_infinity && (x == std::numeric_limits<T>::infinity())) |
---|
80 | { |
---|
81 | real = y; |
---|
82 | imag = std::numeric_limits<T>::infinity(); |
---|
83 | } |
---|
84 | else |
---|
85 | return std::complex<T>(y, y); |
---|
86 | } |
---|
87 | else if(std::numeric_limits<T>::has_infinity && (x == std::numeric_limits<T>::infinity())) |
---|
88 | { |
---|
89 | if(y == std::numeric_limits<T>::infinity()) |
---|
90 | { |
---|
91 | real = quarter_pi; |
---|
92 | imag = std::numeric_limits<T>::infinity(); |
---|
93 | } |
---|
94 | else |
---|
95 | { |
---|
96 | real = half_pi; |
---|
97 | imag = std::numeric_limits<T>::infinity(); |
---|
98 | } |
---|
99 | } |
---|
100 | else if(std::numeric_limits<T>::has_infinity && (y == std::numeric_limits<T>::infinity())) |
---|
101 | { |
---|
102 | real = 0; |
---|
103 | imag = std::numeric_limits<T>::infinity(); |
---|
104 | } |
---|
105 | else |
---|
106 | { |
---|
107 | // |
---|
108 | // special case for real numbers: |
---|
109 | // |
---|
110 | if((y == 0) && (x <= one)) |
---|
111 | return std::complex<T>(std::asin(z.real())); |
---|
112 | // |
---|
113 | // Figure out if our input is within the "safe area" identified by Hull et al. |
---|
114 | // This would be more efficient with portable floating point exception handling; |
---|
115 | // fortunately the quantities M and u identified by Hull et al (figure 3), |
---|
116 | // match with the max and min methods of numeric_limits<T>. |
---|
117 | // |
---|
118 | T safe_max = detail::safe_max(static_cast<T>(8)); |
---|
119 | T safe_min = detail::safe_min(static_cast<T>(4)); |
---|
120 | |
---|
121 | T xp1 = one + x; |
---|
122 | T xm1 = x - one; |
---|
123 | |
---|
124 | if((x < safe_max) && (x > safe_min) && (y < safe_max) && (y > safe_min)) |
---|
125 | { |
---|
126 | T yy = y * y; |
---|
127 | T r = std::sqrt(xp1*xp1 + yy); |
---|
128 | T s = std::sqrt(xm1*xm1 + yy); |
---|
129 | T a = half * (r + s); |
---|
130 | T b = x / a; |
---|
131 | |
---|
132 | if(b <= b_crossover) |
---|
133 | { |
---|
134 | real = std::asin(b); |
---|
135 | } |
---|
136 | else |
---|
137 | { |
---|
138 | T apx = a + x; |
---|
139 | if(x <= one) |
---|
140 | { |
---|
141 | real = std::atan(x/std::sqrt(half * apx * (yy /(r + xp1) + (s-xm1)))); |
---|
142 | } |
---|
143 | else |
---|
144 | { |
---|
145 | real = std::atan(x/(y * std::sqrt(half * (apx/(r + xp1) + apx/(s+xm1))))); |
---|
146 | } |
---|
147 | } |
---|
148 | |
---|
149 | if(a <= a_crossover) |
---|
150 | { |
---|
151 | T am1; |
---|
152 | if(x < one) |
---|
153 | { |
---|
154 | am1 = half * (yy/(r + xp1) + yy/(s - xm1)); |
---|
155 | } |
---|
156 | else |
---|
157 | { |
---|
158 | am1 = half * (yy/(r + xp1) + (s + xm1)); |
---|
159 | } |
---|
160 | imag = boost::math::log1p(am1 + std::sqrt(am1 * (a + one))); |
---|
161 | } |
---|
162 | else |
---|
163 | { |
---|
164 | imag = std::log(a + std::sqrt(a*a - one)); |
---|
165 | } |
---|
166 | } |
---|
167 | else |
---|
168 | { |
---|
169 | // |
---|
170 | // This is the Hull et al exception handling code from Fig 3 of their paper: |
---|
171 | // |
---|
172 | if(y <= (std::numeric_limits<T>::epsilon() * std::fabs(xm1))) |
---|
173 | { |
---|
174 | if(x < one) |
---|
175 | { |
---|
176 | real = std::asin(x); |
---|
177 | imag = y / std::sqrt(xp1*xm1); |
---|
178 | } |
---|
179 | else |
---|
180 | { |
---|
181 | real = half_pi; |
---|
182 | if(((std::numeric_limits<T>::max)() / xp1) > xm1) |
---|
183 | { |
---|
184 | // xp1 * xm1 won't overflow: |
---|
185 | imag = boost::math::log1p(xm1 + std::sqrt(xp1*xm1)); |
---|
186 | } |
---|
187 | else |
---|
188 | { |
---|
189 | imag = log_two + std::log(x); |
---|
190 | } |
---|
191 | } |
---|
192 | } |
---|
193 | else if(y <= safe_min) |
---|
194 | { |
---|
195 | // There is an assumption in Hull et al's analysis that |
---|
196 | // if we get here then x == 1. This is true for all "good" |
---|
197 | // machines where : |
---|
198 | // |
---|
199 | // E^2 > 8*sqrt(u); with: |
---|
200 | // |
---|
201 | // E = std::numeric_limits<T>::epsilon() |
---|
202 | // u = (std::numeric_limits<T>::min)() |
---|
203 | // |
---|
204 | // Hull et al provide alternative code for "bad" machines |
---|
205 | // but we have no way to test that here, so for now just assert |
---|
206 | // on the assumption: |
---|
207 | // |
---|
208 | BOOST_ASSERT(x == 1); |
---|
209 | real = half_pi - std::sqrt(y); |
---|
210 | imag = std::sqrt(y); |
---|
211 | } |
---|
212 | else if(std::numeric_limits<T>::epsilon() * y - one >= x) |
---|
213 | { |
---|
214 | real = x/y; // This can underflow! |
---|
215 | imag = log_two + std::log(y); |
---|
216 | } |
---|
217 | else if(x > one) |
---|
218 | { |
---|
219 | real = std::atan(x/y); |
---|
220 | T xoy = x/y; |
---|
221 | imag = log_two + std::log(y) + half * boost::math::log1p(xoy*xoy); |
---|
222 | } |
---|
223 | else |
---|
224 | { |
---|
225 | T a = std::sqrt(one + y*y); |
---|
226 | real = x/a; // This can underflow! |
---|
227 | imag = half * boost::math::log1p(static_cast<T>(2)*y*(y+a)); |
---|
228 | } |
---|
229 | } |
---|
230 | } |
---|
231 | |
---|
232 | // |
---|
233 | // Finish off by working out the sign of the result: |
---|
234 | // |
---|
235 | if(z.real() < 0) |
---|
236 | real = -real; |
---|
237 | if(z.imag() < 0) |
---|
238 | imag = -imag; |
---|
239 | |
---|
240 | return std::complex<T>(real, imag); |
---|
241 | } |
---|
242 | |
---|
243 | } } // namespaces |
---|
244 | |
---|
245 | #endif // BOOST_MATH_COMPLEX_ASIN_INCLUDED |
---|