[cfa28d3] | 1 | from __future__ import division, print_function |
---|
| 2 | |
---|
| 3 | import time |
---|
| 4 | from copy import copy |
---|
| 5 | |
---|
| 6 | import numpy as np |
---|
| 7 | from numpy import pi, radians, sin, cos, sqrt |
---|
| 8 | from numpy.random import poisson, uniform |
---|
[a1c32c2] | 9 | from numpy.polynomial.legendre import leggauss |
---|
[cfa28d3] | 10 | from scipy.integrate import simps |
---|
| 11 | from scipy.special import j1 as J1 |
---|
| 12 | |
---|
| 13 | # Definition of rotation matrices comes from wikipedia: |
---|
| 14 | # https://en.wikipedia.org/wiki/Rotation_matrix#Basic_rotations |
---|
| 15 | def Rx(angle): |
---|
| 16 | """Construct a matrix to rotate points about *x* by *angle* degrees.""" |
---|
| 17 | a = radians(angle) |
---|
| 18 | R = [[1, 0, 0], |
---|
| 19 | [0, +cos(a), -sin(a)], |
---|
| 20 | [0, +sin(a), +cos(a)]] |
---|
| 21 | return np.matrix(R) |
---|
| 22 | |
---|
| 23 | def Ry(angle): |
---|
| 24 | """Construct a matrix to rotate points about *y* by *angle* degrees.""" |
---|
| 25 | a = radians(angle) |
---|
| 26 | R = [[+cos(a), 0, +sin(a)], |
---|
| 27 | [0, 1, 0], |
---|
| 28 | [-sin(a), 0, +cos(a)]] |
---|
| 29 | return np.matrix(R) |
---|
| 30 | |
---|
| 31 | def Rz(angle): |
---|
| 32 | """Construct a matrix to rotate points about *z* by *angle* degrees.""" |
---|
| 33 | a = radians(angle) |
---|
| 34 | R = [[+cos(a), -sin(a), 0], |
---|
| 35 | [+sin(a), +cos(a), 0], |
---|
| 36 | [0, 0, 1]] |
---|
| 37 | return np.matrix(R) |
---|
| 38 | |
---|
| 39 | def rotation(theta, phi, psi): |
---|
| 40 | """ |
---|
| 41 | Apply the jitter transform to a set of points. |
---|
| 42 | |
---|
| 43 | Points are stored in a 3 x n numpy matrix, not a numpy array or tuple. |
---|
| 44 | """ |
---|
| 45 | return Rx(phi)*Ry(theta)*Rz(psi) |
---|
| 46 | |
---|
[226473d] | 47 | def apply_view(points, view): |
---|
| 48 | """ |
---|
| 49 | Apply the view transform (theta, phi, psi) to a set of points. |
---|
| 50 | |
---|
| 51 | Points are stored in a 3 x n numpy array. |
---|
| 52 | |
---|
| 53 | View angles are in degrees. |
---|
| 54 | """ |
---|
| 55 | theta, phi, psi = view |
---|
| 56 | return np.asarray((Rz(phi)*Ry(theta)*Rz(psi))*np.matrix(points.T)).T |
---|
| 57 | |
---|
| 58 | |
---|
| 59 | def invert_view(qx, qy, view): |
---|
| 60 | """ |
---|
| 61 | Return (qa, qb, qc) for the (theta, phi, psi) view angle at detector |
---|
| 62 | pixel (qx, qy). |
---|
| 63 | |
---|
| 64 | View angles are in degrees. |
---|
| 65 | """ |
---|
| 66 | theta, phi, psi = view |
---|
| 67 | q = np.vstack((qx.flatten(), qy.flatten(), 0*qx.flatten())) |
---|
| 68 | return np.asarray((Rz(-psi)*Ry(-theta)*Rz(-phi))*np.matrix(q)) |
---|
| 69 | |
---|
| 70 | |
---|
[cfa28d3] | 71 | class Shape: |
---|
| 72 | rotation = np.matrix([[1., 0, 0], [0, 1, 0], [0, 0, 1]]) |
---|
| 73 | center = np.array([0., 0., 0.])[:, None] |
---|
| 74 | r_max = None |
---|
| 75 | |
---|
| 76 | def volume(self): |
---|
| 77 | # type: () -> float |
---|
| 78 | raise NotImplementedError() |
---|
| 79 | |
---|
| 80 | def sample(self, density): |
---|
| 81 | # type: (float) -> np.ndarray[N], np.ndarray[N, 3] |
---|
| 82 | raise NotImplementedError() |
---|
| 83 | |
---|
| 84 | def rotate(self, theta, phi, psi): |
---|
| 85 | self.rotation = rotation(theta, phi, psi) * self.rotation |
---|
| 86 | return self |
---|
| 87 | |
---|
| 88 | def shift(self, x, y, z): |
---|
| 89 | self.center = self.center + np.array([x, y, z])[:, None] |
---|
| 90 | return self |
---|
| 91 | |
---|
| 92 | def _adjust(self, points): |
---|
| 93 | points = np.asarray(self.rotation * np.matrix(points.T)) + self.center |
---|
| 94 | return points.T |
---|
| 95 | |
---|
| 96 | def r_bins(self, q, over_sampling=1, r_step=0.): |
---|
| 97 | r_max = min(2 * pi / q[0], self.r_max) |
---|
| 98 | if r_step == 0.: |
---|
| 99 | r_step = 2 * pi / q[-1] / over_sampling |
---|
| 100 | #r_step = 0.01 |
---|
| 101 | return np.arange(r_step, r_max, r_step) |
---|
| 102 | |
---|
| 103 | class Composite(Shape): |
---|
| 104 | def __init__(self, shapes, center=(0, 0, 0), orientation=(0, 0, 0)): |
---|
| 105 | self.shapes = shapes |
---|
| 106 | self.rotate(*orientation) |
---|
| 107 | self.shift(*center) |
---|
| 108 | |
---|
| 109 | # Find the worst case distance between any two points amongst a set |
---|
| 110 | # of shapes independent of orientation. This could easily be a |
---|
| 111 | # factor of two worse than necessary, e.g., a pair of thin rods |
---|
| 112 | # end-to-end vs the same pair side-by-side. |
---|
| 113 | distances = [((s1.r_max + s2.r_max)/2 |
---|
| 114 | + sqrt(np.sum((s1.center - s2.center)**2))) |
---|
| 115 | for s1 in shapes |
---|
| 116 | for s2 in shapes] |
---|
| 117 | self.r_max = max(distances + [s.r_max for s in shapes]) |
---|
| 118 | |
---|
| 119 | def volume(self): |
---|
| 120 | return sum(shape.volume() for shape in self.shapes) |
---|
| 121 | |
---|
| 122 | def sample(self, density): |
---|
| 123 | values, points = zip(*(shape.sample(density) for shape in self.shapes)) |
---|
| 124 | return np.hstack(values), self._adjust(np.vstack(points)) |
---|
| 125 | |
---|
| 126 | class Box(Shape): |
---|
| 127 | def __init__(self, a, b, c, |
---|
| 128 | value, center=(0, 0, 0), orientation=(0, 0, 0)): |
---|
| 129 | self.value = np.asarray(value) |
---|
| 130 | self.rotate(*orientation) |
---|
| 131 | self.shift(*center) |
---|
| 132 | self.a, self.b, self.c = a, b, c |
---|
| 133 | self._scale = np.array([a/2, b/2, c/2])[None, :] |
---|
| 134 | self.r_max = sqrt(a**2 + b**2 + c**2) |
---|
| 135 | |
---|
| 136 | def volume(self): |
---|
| 137 | return self.a*self.b*self.c |
---|
| 138 | |
---|
| 139 | def sample(self, density): |
---|
| 140 | num_points = poisson(density*self.a*self.b*self.c) |
---|
| 141 | points = self._scale*uniform(-1, 1, size=(num_points, 3)) |
---|
| 142 | values = self.value.repeat(points.shape[0]) |
---|
| 143 | return values, self._adjust(points) |
---|
| 144 | |
---|
| 145 | class EllipticalCylinder(Shape): |
---|
| 146 | def __init__(self, ra, rb, length, |
---|
| 147 | value, center=(0, 0, 0), orientation=(0, 0, 0)): |
---|
| 148 | self.value = np.asarray(value) |
---|
| 149 | self.rotate(*orientation) |
---|
| 150 | self.shift(*center) |
---|
| 151 | self.ra, self.rb, self.length = ra, rb, length |
---|
| 152 | self._scale = np.array([ra, rb, length/2])[None, :] |
---|
| 153 | self.r_max = sqrt(4*max(ra, rb)**2 + length**2) |
---|
| 154 | |
---|
| 155 | def volume(self): |
---|
| 156 | return pi*self.ra*self.rb*self.length |
---|
| 157 | |
---|
| 158 | def sample(self, density): |
---|
| 159 | # density of the bounding box |
---|
| 160 | num_points = poisson(density*4*self.ra*self.rb*self.length) |
---|
| 161 | points = uniform(-1, 1, size=(num_points, 3)) |
---|
| 162 | radius = points[:, 0]**2 + points[:, 1]**2 |
---|
| 163 | points = self._scale*points[radius <= 1] |
---|
| 164 | values = self.value.repeat(points.shape[0]) |
---|
| 165 | return values, self._adjust(points) |
---|
| 166 | |
---|
| 167 | class TriaxialEllipsoid(Shape): |
---|
| 168 | def __init__(self, ra, rb, rc, |
---|
| 169 | value, center=(0, 0, 0), orientation=(0, 0, 0)): |
---|
| 170 | self.value = np.asarray(value) |
---|
| 171 | self.rotate(*orientation) |
---|
| 172 | self.shift(*center) |
---|
| 173 | self.ra, self.rb, self.rc = ra, rb, rc |
---|
| 174 | self._scale = np.array([ra, rb, rc])[None, :] |
---|
| 175 | self.r_max = 2*max(ra, rb, rc) |
---|
| 176 | |
---|
| 177 | def volume(self): |
---|
| 178 | return 4*pi/3 * self.ra * self.rb * self.rc |
---|
| 179 | |
---|
| 180 | def sample(self, density): |
---|
| 181 | # randomly sample from a box of side length 2*r, excluding anything |
---|
| 182 | # not in the ellipsoid |
---|
| 183 | num_points = poisson(density*8*self.ra*self.rb*self.rc) |
---|
| 184 | points = uniform(-1, 1, size=(num_points, 3)) |
---|
| 185 | radius = np.sum(points**2, axis=1) |
---|
| 186 | points = self._scale*points[radius <= 1] |
---|
| 187 | values = self.value.repeat(points.shape[0]) |
---|
| 188 | return values, self._adjust(points) |
---|
| 189 | |
---|
| 190 | class Helix(Shape): |
---|
| 191 | def __init__(self, helix_radius, helix_pitch, tube_radius, tube_length, |
---|
| 192 | value, center=(0, 0, 0), orientation=(0, 0, 0)): |
---|
| 193 | self.value = np.asarray(value) |
---|
| 194 | self.rotate(*orientation) |
---|
| 195 | self.shift(*center) |
---|
| 196 | self.helix_radius, self.helix_pitch = helix_radius, helix_pitch |
---|
| 197 | self.tube_radius, self.tube_length = tube_radius, tube_length |
---|
| 198 | helix_length = helix_pitch * tube_length/sqrt(helix_radius**2 + helix_pitch**2) |
---|
| 199 | self.r_max = sqrt((2*helix_radius + 2*tube_radius)*2 |
---|
| 200 | + (helix_length + 2*tube_radius)**2) |
---|
| 201 | |
---|
| 202 | def volume(self): |
---|
| 203 | # small tube radius approximation; for larger tubes need to account |
---|
| 204 | # for the fact that the inner length is much shorter than the outer |
---|
| 205 | # length |
---|
| 206 | return pi*self.tube_radius**2*self.tube_length |
---|
| 207 | |
---|
| 208 | def points(self, density): |
---|
| 209 | num_points = poisson(density*4*self.tube_radius**2*self.tube_length) |
---|
| 210 | points = uniform(-1, 1, size=(num_points, 3)) |
---|
| 211 | radius = points[:, 0]**2 + points[:, 1]**2 |
---|
| 212 | points = points[radius <= 1] |
---|
| 213 | |
---|
| 214 | # Based on math stackexchange answer by Jyrki Lahtonen |
---|
| 215 | # https://math.stackexchange.com/a/461637 |
---|
| 216 | # with helix along z rather than x [so tuples in answer are (z, x, y)] |
---|
| 217 | # and with random points in the cross section (p1, p2) rather than |
---|
| 218 | # uniform points on the surface (cos u, sin u). |
---|
| 219 | a, R = self.tube_radius, self.helix_radius |
---|
| 220 | h = self.helix_pitch |
---|
| 221 | scale = 1/sqrt(R**2 + h**2) |
---|
| 222 | t = points[:, 3] * (self.tube_length * scale/2) |
---|
| 223 | cos_t, sin_t = cos(t), sin(t) |
---|
| 224 | |
---|
| 225 | # rx = R*cos_t |
---|
| 226 | # ry = R*sin_t |
---|
| 227 | # rz = h*t |
---|
| 228 | # nx = -a * cos_t * points[:, 1] |
---|
| 229 | # ny = -a * sin_t * points[:, 1] |
---|
| 230 | # nz = 0 |
---|
| 231 | # bx = (a * h/scale) * sin_t * points[:, 2] |
---|
| 232 | # by = (-a * h/scale) * cos_t * points[:, 2] |
---|
| 233 | # bz = a*R/scale |
---|
| 234 | # x = rx + nx + bx |
---|
| 235 | # y = ry + ny + by |
---|
| 236 | # z = rz + nz + bz |
---|
| 237 | u, v = (R - a*points[:, 1]), (a * h/scale)*points[:, 2] |
---|
| 238 | x = u * cos_t + v * sin_t |
---|
| 239 | y = u * sin_t - v * cos_t |
---|
| 240 | z = a*R/scale + h * t |
---|
| 241 | |
---|
| 242 | points = np.hstack((x, y, z)) |
---|
| 243 | values = self.value.repeat(points.shape[0]) |
---|
| 244 | return values, self._adjust(points) |
---|
| 245 | |
---|
[226473d] | 246 | def calc_Iqxy(qx, qy, rho, points, volume=1, view=(0, 0, 0)): |
---|
| 247 | x, y, z = points.T |
---|
| 248 | qx, qy = np.broadcast_arrays(qx, qy) |
---|
| 249 | qa, qb, qc = invert_view(qx, qy, view) |
---|
| 250 | rho, volume = np.broadcast_arrays(rho, volume) |
---|
| 251 | values = rho*volume |
---|
| 252 | |
---|
| 253 | # I(q) = |sum V(r) rho(r) e^(1j q.r)|^2 / sum V(r) |
---|
| 254 | Iq = [abs(np.sum(values*np.exp(1j*(qa_k*x + qb_k*y + qc_k*z))))**2 |
---|
| 255 | for qa_k, qb_k, qc_k in zip(qa.flat, qb.flat, qc.flat)] |
---|
| 256 | return np.array(Iq).reshape(qx.shape) / np.sum(volume) |
---|
| 257 | |
---|
[cfa28d3] | 258 | def _calc_Pr_nonuniform(r, rho, points): |
---|
| 259 | # Make Pr a little be bigger than necessary so that only distances |
---|
| 260 | # min < d < max end up in Pr |
---|
| 261 | n_max = len(r)+1 |
---|
| 262 | extended_Pr = np.zeros(n_max+1, 'd') |
---|
| 263 | # r refers to bin centers; find corresponding bin edges |
---|
| 264 | bins = bin_edges(r) |
---|
| 265 | t_next = time.clock() + 3 |
---|
| 266 | for k, rho_k in enumerate(rho[:-1]): |
---|
| 267 | distance = sqrt(np.sum((points[k] - points[k+1:])**2, axis=1)) |
---|
| 268 | weights = rho_k * rho[k+1:] |
---|
| 269 | index = np.searchsorted(bins, distance) |
---|
| 270 | # Note: indices may be duplicated, so "Pr[index] += w" will not work!! |
---|
| 271 | extended_Pr += np.bincount(index, weights, n_max+1) |
---|
| 272 | t = time.clock() |
---|
| 273 | if t > t_next: |
---|
| 274 | t_next = t + 3 |
---|
| 275 | print("processing %d of %d"%(k, len(rho)-1)) |
---|
| 276 | Pr = extended_Pr[1:-1] |
---|
| 277 | return Pr / Pr.max() |
---|
| 278 | |
---|
| 279 | def calc_Pr(r, rho, points): |
---|
| 280 | # P(r) with uniform steps in r is 3x faster; check if we are uniform |
---|
| 281 | # before continuing |
---|
| 282 | if np.max(np.abs(np.diff(r) - r[0])) > r[0]*0.01: |
---|
| 283 | return _calc_Pr_nonuniform(r, rho, points) |
---|
| 284 | |
---|
| 285 | # Make Pr a little be bigger than necessary so that only distances |
---|
| 286 | # min < d < max end up in Pr |
---|
| 287 | n_max = len(r) |
---|
| 288 | extended_Pr = np.zeros(n_max+1, 'd') |
---|
| 289 | t0 = time.clock() |
---|
| 290 | t_next = t0 + 3 |
---|
| 291 | row_zero = np.zeros(len(rho), 'i') |
---|
| 292 | for k, rho_k in enumerate(rho[:-1]): |
---|
| 293 | distances = sqrt(np.sum((points[k] - points[k+1:])**2, axis=1)) |
---|
| 294 | weights = rho_k * rho[k+1:] |
---|
| 295 | index = np.minimum(np.asarray(distances/r[0], 'i'), n_max) |
---|
| 296 | # Note: indices may be duplicated, so "Pr[index] += w" will not work!! |
---|
| 297 | extended_Pr += np.bincount(index, weights, n_max+1) |
---|
| 298 | t = time.clock() |
---|
| 299 | if t > t_next: |
---|
| 300 | t_next = t + 3 |
---|
| 301 | print("processing %d of %d"%(k, len(rho)-1)) |
---|
| 302 | #print("time py:", time.clock() - t0) |
---|
| 303 | Pr = extended_Pr[:-1] |
---|
| 304 | # Make Pr independent of sampling density. The factor of 2 comes because |
---|
| 305 | # we are only accumulating the upper triangular distances. |
---|
| 306 | #Pr = Pr * 2 / len(rho)**2 |
---|
| 307 | return Pr / Pr.max() |
---|
| 308 | |
---|
| 309 | # Can get an additional 2x by going to C. Cuda/OpenCL will allow even |
---|
| 310 | # more speedup, though still bounded by the n^2 cose. |
---|
| 311 | """ |
---|
| 312 | void pdfcalc(int n, const double *pts, const double *rho, |
---|
| 313 | int nPr, double *Pr, double rstep) |
---|
| 314 | { |
---|
| 315 | int i,j; |
---|
| 316 | |
---|
| 317 | for (i=0; i<n-2; i++) { |
---|
| 318 | for (j=i+1; j<=n-1; j++) { |
---|
| 319 | const double dxx=pts[3*i]-pts[3*j]; |
---|
| 320 | const double dyy=pts[3*i+1]-pts[3*j+1]; |
---|
| 321 | const double dzz=pts[3*i+2]-pts[3*j+2]; |
---|
| 322 | const double d=sqrt(dxx*dxx+dyy*dyy+dzz*dzz); |
---|
| 323 | const int k=rint(d/rstep); |
---|
| 324 | if (k < nPr) Pr[k]+=rho[i]*rho[j]; |
---|
| 325 | } |
---|
| 326 | } |
---|
| 327 | } |
---|
| 328 | """ |
---|
| 329 | |
---|
| 330 | def j0(x): |
---|
| 331 | return np.sinc(x/np.pi) |
---|
| 332 | |
---|
| 333 | def calc_Iq(q, r, Pr): |
---|
| 334 | Iq = np.array([simps(Pr * j0(qk*r), r) for qk in q]) |
---|
| 335 | #Iq = np.array([np.trapz(Pr * j0(qk*r), r) for qk in q]) |
---|
| 336 | Iq /= Iq[0] |
---|
| 337 | return Iq |
---|
| 338 | |
---|
| 339 | # NOTE: copied from sasmodels/resolution.py |
---|
| 340 | def bin_edges(x): |
---|
| 341 | """ |
---|
| 342 | Determine bin edges from bin centers, assuming that edges are centered |
---|
| 343 | between the bins. |
---|
| 344 | |
---|
| 345 | Note: this uses the arithmetic mean, which may not be appropriate for |
---|
| 346 | log-scaled data. |
---|
| 347 | """ |
---|
| 348 | if len(x) < 2 or (np.diff(x) < 0).any(): |
---|
| 349 | raise ValueError("Expected bins to be an increasing set") |
---|
| 350 | edges = np.hstack([ |
---|
| 351 | x[0] - 0.5*(x[1] - x[0]), # first point minus half first interval |
---|
| 352 | 0.5*(x[1:] + x[:-1]), # mid points of all central intervals |
---|
| 353 | x[-1] + 0.5*(x[-1] - x[-2]), # last point plus half last interval |
---|
| 354 | ]) |
---|
| 355 | return edges |
---|
| 356 | |
---|
| 357 | def plot_calc(r, Pr, q, Iq, theory=None): |
---|
| 358 | import matplotlib.pyplot as plt |
---|
| 359 | plt.subplot(211) |
---|
| 360 | plt.plot(r, Pr, '-', label="Pr") |
---|
| 361 | plt.xlabel('r (A)') |
---|
| 362 | plt.ylabel('Pr (1/A^2)') |
---|
| 363 | plt.subplot(212) |
---|
| 364 | plt.loglog(q, Iq, '-', label='from Pr') |
---|
| 365 | plt.xlabel('q (1/A') |
---|
| 366 | plt.ylabel('Iq') |
---|
| 367 | if theory is not None: |
---|
| 368 | plt.loglog(theory[0], theory[1], '-', label='analytic') |
---|
| 369 | plt.legend() |
---|
| 370 | |
---|
[226473d] | 371 | def plot_calc_2d(qx, qy, Iqxy, theory=None): |
---|
| 372 | import matplotlib.pyplot as plt |
---|
| 373 | qx, qy = bin_edges(qx), bin_edges(qy) |
---|
| 374 | #qx, qy = np.meshgrid(qx, qy) |
---|
| 375 | if theory is not None: |
---|
| 376 | plt.subplot(121) |
---|
| 377 | plt.pcolormesh(qx, qy, np.log10(Iqxy)) |
---|
| 378 | plt.xlabel('qx (1/A)') |
---|
| 379 | plt.ylabel('qy (1/A)') |
---|
| 380 | if theory is not None: |
---|
| 381 | plt.subplot(122) |
---|
| 382 | plt.pcolormesh(qx, qy, np.log10(theory)) |
---|
| 383 | plt.xlabel('qx (1/A)') |
---|
| 384 | |
---|
[cfa28d3] | 385 | def plot_points(rho, points): |
---|
| 386 | import mpl_toolkits.mplot3d |
---|
| 387 | import matplotlib.pyplot as plt |
---|
| 388 | |
---|
| 389 | ax = plt.axes(projection='3d') |
---|
| 390 | try: |
---|
| 391 | ax.axis('square') |
---|
| 392 | except Exception: |
---|
| 393 | pass |
---|
| 394 | n = len(points) |
---|
| 395 | index = np.random.choice(n, size=1000) if n > 1000 else slice(None, None) |
---|
| 396 | ax.scatter(points[index, 0], points[index, 1], points[index, 2], c=rho[index]) |
---|
| 397 | #low, high = points.min(axis=0), points.max(axis=0) |
---|
| 398 | #ax.axis([low[0], high[0], low[1], high[1], low[2], high[2]]) |
---|
| 399 | ax.autoscale(True) |
---|
| 400 | |
---|
| 401 | def sas_2J1x_x(x): |
---|
| 402 | with np.errstate(all='ignore'): |
---|
| 403 | retvalue = 2*J1(x)/x |
---|
| 404 | retvalue[x == 0] = 1. |
---|
| 405 | return retvalue |
---|
| 406 | |
---|
| 407 | def sas_3j1x_x(x): |
---|
| 408 | """return 3*j1(x)/x""" |
---|
| 409 | with np.errstate(all='ignore'): |
---|
| 410 | retvalue = 3*(sin(x) - x*cos(x))/x**3 |
---|
| 411 | retvalue[x == 0.] = 1. |
---|
| 412 | return retvalue |
---|
| 413 | |
---|
| 414 | def cylinder_Iq(q, radius, length): |
---|
[a1c32c2] | 415 | z, w = leggauss(76) |
---|
| 416 | cos_alpha = (z+1)/2 |
---|
| 417 | sin_alpha = sqrt(1.0 - cos_alpha**2) |
---|
[cfa28d3] | 418 | Iq = np.empty_like(q) |
---|
| 419 | for k, qk in enumerate(q): |
---|
[a1c32c2] | 420 | qab, qc = qk*sin_alpha, qk*cos_alpha |
---|
| 421 | Fq = sas_2J1x_x(qab*radius) * j0(qc*length/2) |
---|
| 422 | Iq[k] = np.sum(w*Fq**2) |
---|
[cfa28d3] | 423 | Iq = Iq/Iq[0] |
---|
| 424 | return Iq |
---|
| 425 | |
---|
[226473d] | 426 | def cylinder_Iqxy(qx, qy, radius, length, view=(0, 0, 0)): |
---|
| 427 | qa, qb, qc = invert_view(qx, qy, view) |
---|
| 428 | qab = np.sqrt(qa**2 + qb**2) |
---|
| 429 | Fq = sas_2J1x_x(qab*radius) * j0(qc*length/2) |
---|
| 430 | Iq = Fq**2 |
---|
| 431 | return Iq.reshape(qx.shape) |
---|
| 432 | |
---|
[cfa28d3] | 433 | def sphere_Iq(q, radius): |
---|
| 434 | Iq = sas_3j1x_x(q*radius)**2 |
---|
| 435 | return Iq/Iq[0] |
---|
| 436 | |
---|
| 437 | def csbox_Iq(q, a, b, c, da, db, dc, slda, sldb, sldc, sld_core): |
---|
[a1c32c2] | 438 | z, w = leggauss(76) |
---|
| 439 | |
---|
[cfa28d3] | 440 | sld_solvent = 0 |
---|
| 441 | overlapping = False |
---|
| 442 | dr0 = sld_core - sld_solvent |
---|
| 443 | drA, drB, drC = slda-sld_solvent, sldb-sld_solvent, sldc-sld_solvent |
---|
| 444 | tA, tB, tC = a + 2*da, b + 2*db, c + 2*dc |
---|
| 445 | |
---|
[a1c32c2] | 446 | outer_sum = np.zeros_like(q) |
---|
| 447 | for cos_alpha, outer_w in zip((z+1)/2, w): |
---|
[cfa28d3] | 448 | sin_alpha = sqrt(1.0-cos_alpha*cos_alpha) |
---|
| 449 | qc = q*cos_alpha |
---|
| 450 | siC = c*j0(c*qc/2) |
---|
| 451 | siCt = tC*j0(tC*qc/2) |
---|
[a1c32c2] | 452 | inner_sum = np.zeros_like(q) |
---|
| 453 | for beta, inner_w in zip((z + 1)*pi/4, w): |
---|
[cfa28d3] | 454 | qa, qb = q*sin_alpha*sin(beta), q*sin_alpha*cos(beta) |
---|
| 455 | siA = a*j0(a*qa/2) |
---|
| 456 | siB = b*j0(b*qb/2) |
---|
| 457 | siAt = tA*j0(tA*qa/2) |
---|
| 458 | siBt = tB*j0(tB*qb/2) |
---|
| 459 | if overlapping: |
---|
[a1c32c2] | 460 | Fq = (dr0*siA*siB*siC |
---|
| 461 | + drA*(siAt-siA)*siB*siC |
---|
| 462 | + drB*siAt*(siBt-siB)*siC |
---|
| 463 | + drC*siAt*siBt*(siCt-siC)) |
---|
[cfa28d3] | 464 | else: |
---|
[a1c32c2] | 465 | Fq = (dr0*siA*siB*siC |
---|
| 466 | + drA*(siAt-siA)*siB*siC |
---|
| 467 | + drB*siA*(siBt-siB)*siC |
---|
| 468 | + drC*siA*siB*(siCt-siC)) |
---|
| 469 | inner_sum += inner_w * Fq**2 |
---|
| 470 | outer_sum += outer_w * inner_sum |
---|
| 471 | Iq = outer_sum / 4 # = outer*um*zm*8.0/(4.0*M_PI) |
---|
[cfa28d3] | 472 | return Iq/Iq[0] |
---|
| 473 | |
---|
| 474 | def check_shape(shape, fn=None): |
---|
| 475 | rho_solvent = 0 |
---|
| 476 | q = np.logspace(-3, 0, 200) |
---|
| 477 | r = shape.r_bins(q, r_step=0.01) |
---|
| 478 | sampling_density = 15000 / shape.volume() |
---|
| 479 | rho, points = shape.sample(sampling_density) |
---|
| 480 | Pr = calc_Pr(r, rho-rho_solvent, points) |
---|
| 481 | Iq = calc_Iq(q, r, Pr) |
---|
| 482 | theory = (q, fn(q)) if fn is not None else None |
---|
| 483 | |
---|
| 484 | import pylab |
---|
| 485 | #plot_points(rho, points); pylab.figure() |
---|
| 486 | plot_calc(r, Pr, q, Iq, theory=theory) |
---|
| 487 | pylab.show() |
---|
| 488 | |
---|
[226473d] | 489 | def check_shape_2d(shape, fn=None, view=(0, 0, 0)): |
---|
| 490 | rho_solvent = 0 |
---|
| 491 | qx = qy = np.linspace(-1, 1, 100) |
---|
| 492 | Qx, Qy = np.meshgrid(qx, qy) |
---|
| 493 | sampling_density = 50000 / shape.volume() |
---|
| 494 | rho, points = shape.sample(sampling_density) |
---|
| 495 | Iqxy = calc_Iqxy(Qx, Qy, rho, points, view=view) |
---|
| 496 | Iqxy += 0.001 * Iqxy.max() |
---|
| 497 | theory = fn(Qx, Qy)+0.001 if fn is not None else None |
---|
| 498 | |
---|
| 499 | import pylab |
---|
| 500 | plot_calc_2d(qx, qy, Iqxy, theory=theory) |
---|
| 501 | pylab.show() |
---|
| 502 | |
---|
[cfa28d3] | 503 | def check_cylinder(radius=25, length=125, rho=2.): |
---|
| 504 | shape = EllipticalCylinder(radius, radius, length, rho) |
---|
| 505 | fn = lambda q: cylinder_Iq(q, radius, length) |
---|
| 506 | check_shape(shape, fn) |
---|
| 507 | |
---|
[226473d] | 508 | def check_cylinder_2d(radius=25, length=125, rho=2., view=(0, 0, 0)): |
---|
| 509 | shape = EllipticalCylinder(radius, radius, length, rho) |
---|
| 510 | fn = lambda qx, qy, view=view: cylinder_Iqxy(qx, qy, radius, length, view=view) |
---|
| 511 | check_shape_2d(shape, fn, view=view) |
---|
| 512 | |
---|
[cfa28d3] | 513 | def check_sphere(radius=125, rho=2): |
---|
| 514 | shape = TriaxialEllipsoid(radius, radius, radius, rho) |
---|
| 515 | fn = lambda q: sphere_Iq(q, radius) |
---|
| 516 | check_shape(shape, fn) |
---|
| 517 | |
---|
| 518 | def check_csbox(a=10, b=20, c=30, da=1, db=2, dc=3, slda=1, sldb=2, sldc=3, sld_core=4): |
---|
| 519 | core = Box(a, b, c, sld_core) |
---|
| 520 | side_a = Box(da, b, c, slda, center=((a+da)/2, 0, 0)) |
---|
| 521 | side_b = Box(a, db, c, sldb, center=(0, (b+db)/2, 0)) |
---|
| 522 | side_c = Box(a, b, dc, sldc, center=(0, 0, (c+dc)/2)) |
---|
| 523 | side_a2 = copy(side_a).shift(-a-da, 0, 0) |
---|
| 524 | side_b2 = copy(side_b).shift(0, -b-db, 0) |
---|
| 525 | side_c2 = copy(side_c).shift(0, 0, -c-dc) |
---|
| 526 | shape = Composite((core, side_a, side_b, side_c, side_a2, side_b2, side_c2)) |
---|
| 527 | fn = lambda q: csbox_Iq(q, a, b, c, da, db, dc, slda, sldb, sldc, sld_core) |
---|
| 528 | check_shape(shape, fn) |
---|
| 529 | |
---|
| 530 | if __name__ == "__main__": |
---|
| 531 | check_cylinder(radius=10, length=40) |
---|
[226473d] | 532 | #check_cylinder_2d(radius=10, length=40, view=(90,30,0)) |
---|
[cfa28d3] | 533 | #check_sphere() |
---|
| 534 | #check_csbox() |
---|
| 535 | #check_csbox(da=100, db=200, dc=300) |
---|