1 | r""" |
---|
2 | Show numerical precision of cylinder form. |
---|
3 | |
---|
4 | Using:: |
---|
5 | |
---|
6 | qr = q r sin(t) |
---|
7 | qh = q h/2 cos(t) |
---|
8 | F = 2 J_1(qr)/qr sin(qh)/qh |
---|
9 | """ |
---|
10 | |
---|
11 | import numpy as np |
---|
12 | from sympy.mpmath import mp |
---|
13 | #import matplotlib; matplotlib.use('TkAgg') |
---|
14 | import pylab |
---|
15 | |
---|
16 | SHOW_DIFF = True # True if show diff rather than function value |
---|
17 | LINEAR_X = False # True if q is linearly spaced instead of log spaced |
---|
18 | |
---|
19 | RADIUS = 20 |
---|
20 | LENGTH = 300 |
---|
21 | CONTRAST = 5 |
---|
22 | THETA = 45 |
---|
23 | |
---|
24 | def mp_form(vec, bits=500): |
---|
25 | """ |
---|
26 | Direct calculation using sympy multiprecision library. |
---|
27 | """ |
---|
28 | with mp.workprec(bits): |
---|
29 | return [_mp_f(mp.mpf(x)) for x in vec] |
---|
30 | |
---|
31 | def _mp_f(x): |
---|
32 | """ |
---|
33 | Helper function for mp_j1c |
---|
34 | """ |
---|
35 | f = mp.mpf |
---|
36 | theta = f(THETA)*mp.pi/f(180) |
---|
37 | qr = x * f(RADIUS)*mp.sin(theta) |
---|
38 | qh = x * f(LENGTH)/f(2)*mp.cos(theta) |
---|
39 | return (f(2)*mp.j1(qr)/qr * mp.sin(qh)/qh)**f(2) |
---|
40 | |
---|
41 | def np_form(x, dtype): |
---|
42 | """ |
---|
43 | Direct calculation using scipy. |
---|
44 | """ |
---|
45 | from scipy.special import j1 |
---|
46 | f = np.float64 if np.dtype(dtype) == np.float64 else np.float32 |
---|
47 | x = np.asarray(x, dtype) |
---|
48 | theta = f(THETA)*f(np.pi)/f(180) |
---|
49 | qr = x * f(RADIUS)*np.sin(theta) |
---|
50 | qh = x * f(LENGTH)/f(2)*np.cos(theta) |
---|
51 | return (f(2)*j1(qr)/qr*np.sin(qh)/qh)**f(2) |
---|
52 | |
---|
53 | def sasmodels_form(x, dtype): |
---|
54 | f = np.float64 if np.dtype(dtype) == np.float64 else np.float32 |
---|
55 | x = np.asarray(x, dtype) |
---|
56 | theta = f(THETA)*f(np.pi)/f(180) |
---|
57 | qr = x * f(RADIUS)*np.sin(theta) |
---|
58 | qh = x * f(LENGTH)/f(2)*np.cos(theta) |
---|
59 | return (J1c(qr, dtype)*np.sin(qh)/qh)**f(2) |
---|
60 | |
---|
61 | def J1c(x, dtype): |
---|
62 | x = np.asarray(x, dtype) |
---|
63 | f = np.float64 if np.dtype(dtype) == np.float64 else np.float32 |
---|
64 | return np.asarray([_J1c(xi, f) for xi in x], dtype) |
---|
65 | |
---|
66 | def _J1c(x, f): |
---|
67 | ax = abs(x) |
---|
68 | if ax < f(8.0): |
---|
69 | y = x*x |
---|
70 | ans1 = f(2)*(f(72362614232.0) |
---|
71 | + y*(f(-7895059235.0) |
---|
72 | + y*(f(242396853.1) |
---|
73 | + y*(f(-2972611.439) |
---|
74 | + y*(f(15704.48260) |
---|
75 | + y*(f(-30.16036606))))))) |
---|
76 | ans2 = (f(144725228442.0) |
---|
77 | + y*(f(2300535178.0) |
---|
78 | + y*(f(18583304.74) |
---|
79 | + y*(f(99447.43394) |
---|
80 | + y*(f(376.9991397) |
---|
81 | + y))))) |
---|
82 | return ans1/ans2 |
---|
83 | else: |
---|
84 | y = f(64.0)/(ax*ax) |
---|
85 | xx = ax - f(2.356194491) |
---|
86 | ans1 = (f(1.0) |
---|
87 | + y*(f(0.183105e-2) |
---|
88 | + y*(f(-0.3516396496e-4) |
---|
89 | + y*(f(0.2457520174e-5) |
---|
90 | + y*f(-0.240337019e-6))))) |
---|
91 | ans2 = (f(0.04687499995) |
---|
92 | + y*(f(-0.2002690873e-3) |
---|
93 | + y*(f(0.8449199096e-5) |
---|
94 | + y*(f(-0.88228987e-6) |
---|
95 | + y*f(0.105787412e-6))))) |
---|
96 | sn, cn = np.sin(xx), np.cos(xx) |
---|
97 | ans = np.sqrt(f(0.636619772)/ax) * (cn*ans1 - (f(8.0)/ax)*sn*ans2) * f(2)/x |
---|
98 | return -ans if (x < f(0.0)) else ans |
---|
99 | |
---|
100 | def plotdiff(x, target, actual, label): |
---|
101 | """ |
---|
102 | Plot the computed value. |
---|
103 | |
---|
104 | Use relative error if SHOW_DIFF, otherwise just plot the value directly. |
---|
105 | """ |
---|
106 | if SHOW_DIFF: |
---|
107 | err = np.clip(abs((target-actual)/target), 0, 1) |
---|
108 | pylab.loglog(x, err, '-', label=label) |
---|
109 | else: |
---|
110 | limits = np.min(target), np.max(target) |
---|
111 | pylab.semilogx(x, np.clip(actual,*limits), '-', label=label) |
---|
112 | |
---|
113 | def compare(x, precision): |
---|
114 | r""" |
---|
115 | Compare the different computation methods using the given precision. |
---|
116 | """ |
---|
117 | target = np.asarray(mp_form(x), 'double') |
---|
118 | plotdiff(x, target, mp_form(x, bits=11), '11-bit') |
---|
119 | plotdiff(x, target, np_form(x, precision), 'direct '+precision) |
---|
120 | plotdiff(x, target, sasmodels_form(x, precision), 'sasmodels '+precision) |
---|
121 | pylab.xlabel("qr (1/Ang)") |
---|
122 | if SHOW_DIFF: |
---|
123 | pylab.ylabel("relative error") |
---|
124 | else: |
---|
125 | pylab.ylabel("2 J1(x)/x") |
---|
126 | pylab.semilogx(x, target, '-', label="true value") |
---|
127 | if LINEAR_X: |
---|
128 | pylab.xscale('linear') |
---|
129 | |
---|
130 | def main(): |
---|
131 | r""" |
---|
132 | Compare accuracy of different methods for computing $3 j_1(x)/x$. |
---|
133 | :return: |
---|
134 | """ |
---|
135 | if LINEAR_X: |
---|
136 | qr = np.linspace(1,1000,2000) |
---|
137 | else: |
---|
138 | qr = np.logspace(-3,5,400) |
---|
139 | pylab.subplot(121) |
---|
140 | compare(qr, 'single') |
---|
141 | pylab.legend(loc='best') |
---|
142 | pylab.subplot(122) |
---|
143 | compare(qr, 'double') |
---|
144 | pylab.legend(loc='best') |
---|
145 | pylab.suptitle('2 J1(x)/x') |
---|
146 | |
---|
147 | if __name__ == "__main__": |
---|
148 | #print "\n".join(str(x) for x in mp_J1c([1e-6,1e-5,1e-4,1e-3])) |
---|
149 | main() |
---|
150 | pylab.show() |
---|