1 | r""" |
---|
2 | Definition |
---|
3 | ---------- |
---|
4 | |
---|
5 | This model employs the empirical multiple level unified Exponential/Power-law |
---|
6 | fit method developed by Beaucage. Four functions are included so that 1, 2, 3, |
---|
7 | or 4 levels can be used. In addition a 0 level has been added which simply |
---|
8 | calculates |
---|
9 | |
---|
10 | .. math:: |
---|
11 | |
---|
12 | I(q) = \text{scale} / q + \text{background} |
---|
13 | |
---|
14 | The Beaucage method is able to reasonably approximate the scattering from |
---|
15 | many different types of particles, including fractal clusters, random coils |
---|
16 | (Debye equation), ellipsoidal particles, etc. |
---|
17 | |
---|
18 | The model works best for mass fractal systems characterized by Porod exponents |
---|
19 | between 5/3 and 3. It should not be used for surface fractal systems. Hammouda |
---|
20 | (2010) has pointed out a deficiency in the way this model handles the |
---|
21 | transitioning between the Guinier and Porod regimes and which can create |
---|
22 | artefacts that appear as kinks in the fitted model function. |
---|
23 | |
---|
24 | Also see the Guinier_Porod model. |
---|
25 | |
---|
26 | The empirical fit function is: |
---|
27 | |
---|
28 | .. math:: |
---|
29 | |
---|
30 | I(q) = \text{background} |
---|
31 | + \sum_{i=1}^N \Bigl[ |
---|
32 | G_i \exp\Bigl(-\frac{q^2R_{gi}^2}{3}\Bigr) |
---|
33 | + B_i \exp\Bigl(-\frac{q^2R_{g(i+1)}^2}{3}\Bigr) |
---|
34 | \Bigl(\frac{1}{q_i^*}\Bigr)^{P_i} \Bigr] |
---|
35 | |
---|
36 | where |
---|
37 | |
---|
38 | .. math:: |
---|
39 | |
---|
40 | q_i^* = q \left[\operatorname{erf} |
---|
41 | \left(\frac{q R_{gi}}{\sqrt{6}}\right) |
---|
42 | \right]^{-3} |
---|
43 | |
---|
44 | |
---|
45 | For each level, the four parameters $G_i$, $R_{gi}$, $B_i$ and $P_i$ must |
---|
46 | be chosen. Beaucage has an additional factor $k$ in the definition of |
---|
47 | $q_i^*$ which is ignored here. |
---|
48 | |
---|
49 | For example, to approximate the scattering from random coils (Debye equation), |
---|
50 | set $R_{gi}$ as the Guinier radius, $P_i = 2$, and $B_i = 2 G_i / R_{gi}$ |
---|
51 | |
---|
52 | See the references for further information on choosing the parameters. |
---|
53 | |
---|
54 | For 2D data: The 2D scattering intensity is calculated in the same way as 1D, |
---|
55 | where the $q$ vector is defined as |
---|
56 | |
---|
57 | .. math:: |
---|
58 | |
---|
59 | q = \sqrt{q_x^2 + q_y^2} |
---|
60 | |
---|
61 | |
---|
62 | References |
---|
63 | ---------- |
---|
64 | |
---|
65 | G Beaucage, *J. Appl. Cryst.*, 28 (1995) 717-728 |
---|
66 | |
---|
67 | G Beaucage, *J. Appl. Cryst.*, 29 (1996) 134-146 |
---|
68 | |
---|
69 | B Hammouda, *Analysis of the Beaucage model, J. Appl. Cryst.*, (2010), 43, 1474-1478 |
---|
70 | """ |
---|
71 | |
---|
72 | from __future__ import division |
---|
73 | |
---|
74 | import numpy as np |
---|
75 | from numpy import inf, exp, sqrt, errstate |
---|
76 | from scipy.special import erf, gamma |
---|
77 | |
---|
78 | category = "shape-independent" |
---|
79 | name = "unified_power_Rg" |
---|
80 | title = "Unified Power Rg" |
---|
81 | description = """ |
---|
82 | The Beaucage model employs the empirical multiple level unified |
---|
83 | Exponential/Power-law fit method developed by G. Beaucage. Four functions |
---|
84 | are included so that 1, 2, 3, or 4 levels can be used. |
---|
85 | """ |
---|
86 | |
---|
87 | # pylint: disable=bad-whitespace, line-too-long |
---|
88 | parameters = [ |
---|
89 | ["level", "", 1, [1, 6], "", "Level number"], |
---|
90 | ["rg[level]", "Ang", 15.8, [0, inf], "", "Radius of gyration"], |
---|
91 | ["power[level]", "", 4, [-inf, inf], "", "Power"], |
---|
92 | ["B[level]", "1/cm", 4.5e-6, [-inf, inf], "", ""], |
---|
93 | ["G[level]", "1/cm", 400, [0, inf], "", ""], |
---|
94 | ] |
---|
95 | # pylint: enable=bad-whitespace, line-too-long |
---|
96 | |
---|
97 | def Iq(q, level, rg, power, B, G): |
---|
98 | level = int(level + 0.5) |
---|
99 | if level == 0: |
---|
100 | with errstate(divide='ignore'): |
---|
101 | return 1./q |
---|
102 | |
---|
103 | with errstate(divide='ignore', invalid='ignore'): |
---|
104 | result = np.zeros(q.shape, 'd') |
---|
105 | for i in range(level): |
---|
106 | exp_now = exp(-(q*rg[i])**2/3.) |
---|
107 | pow_now = (erf(q*rg[i]/sqrt(6.))**3/q)**power[i] |
---|
108 | if i < level-1: |
---|
109 | exp_next = exp(-(q*rg[i+1])**2/3.) |
---|
110 | else: |
---|
111 | exp_next = 1 |
---|
112 | result += G[i]*exp_now + B[i]*exp_next*pow_now |
---|
113 | |
---|
114 | result[q == 0] = np.sum(G[:level]) |
---|
115 | return result |
---|
116 | |
---|
117 | Iq.vectorized = True |
---|
118 | |
---|
119 | def random(): |
---|
120 | level = np.minimum(np.random.poisson(0.5) + 1, 6) |
---|
121 | n = level |
---|
122 | power = np.random.uniform(1.6, 3, n) |
---|
123 | rg = 10**np.random.uniform(1, 5, n) |
---|
124 | G = np.random.uniform(0.1, 10, n)**2 * 10**np.random.uniform(0.3, 3, n) |
---|
125 | B = G * power / rg**power * gamma(power/2) |
---|
126 | scale = 10**np.random.uniform(1, 4) |
---|
127 | pars = dict( |
---|
128 | #background=0, |
---|
129 | scale=scale, |
---|
130 | level=level, |
---|
131 | ) |
---|
132 | pars.update(("power%d"%(k+1), v) for k, v in enumerate(power)) |
---|
133 | pars.update(("rg%d"%(k+1), v) for k, v in enumerate(rg)) |
---|
134 | pars.update(("B%d"%(k+1), v) for k, v in enumerate(B)) |
---|
135 | pars.update(("G%d"%(k+1), v) for k, v in enumerate(G)) |
---|
136 | return pars |
---|
137 | |
---|
138 | demo = dict( |
---|
139 | level=2, |
---|
140 | rg=[15.8, 21], |
---|
141 | power=[4, 2], |
---|
142 | B=[4.5e-6, 0.0006], |
---|
143 | G=[400, 3], |
---|
144 | scale=1., |
---|
145 | background=0., |
---|
146 | ) |
---|