[8a48713] | 1 | /** |
---|
| 2 | This software was developed by the University of Tennessee as part of the |
---|
| 3 | Distributed Data Analysis of Neutron Scattering Experiments (DANSE) |
---|
| 4 | project funded by the US National Science Foundation. |
---|
| 5 | |
---|
| 6 | If you use DANSE applications to do scientific research that leads to |
---|
| 7 | publication, we ask that you acknowledge the use of the software with the |
---|
| 8 | following sentence: |
---|
| 9 | |
---|
| 10 | "This work benefited from DANSE software developed under NSF award DMR-0520547." |
---|
| 11 | |
---|
| 12 | copyright 2008, University of Tennessee |
---|
| 13 | */ |
---|
| 14 | |
---|
| 15 | /** |
---|
| 16 | * Scattering model classes |
---|
| 17 | * The classes use the IGOR library found in |
---|
| 18 | * sansmodels/src/libigor |
---|
| 19 | * |
---|
| 20 | * TODO: add 2D function |
---|
| 21 | */ |
---|
| 22 | |
---|
| 23 | #include <math.h> |
---|
| 24 | #include "parameters.hh" |
---|
| 25 | #include <stdio.h> |
---|
| 26 | using namespace std; |
---|
| 27 | |
---|
| 28 | extern "C" { |
---|
| 29 | #include "libCylinder.h" |
---|
[f9bf661] | 30 | #include "libStructureFactor.h" |
---|
[318b5bbb] | 31 | #include "libmultifunc/libfunc.h" |
---|
[8343e18] | 32 | } |
---|
| 33 | #include "parallelepiped.h" |
---|
| 34 | |
---|
| 35 | // Convenience parameter structure |
---|
| 36 | typedef struct { |
---|
| 37 | double scale; |
---|
| 38 | double short_a; |
---|
| 39 | double short_b; |
---|
| 40 | double long_c; |
---|
| 41 | double sldPipe; |
---|
| 42 | double sldSolv; |
---|
| 43 | double background; |
---|
| 44 | double parallel_theta; |
---|
| 45 | double parallel_phi; |
---|
| 46 | double parallel_psi; |
---|
[318b5bbb] | 47 | double M0_sld_pipe; |
---|
| 48 | double M_theta_pipe; |
---|
| 49 | double M_phi_pipe; |
---|
| 50 | double M0_sld_solv; |
---|
| 51 | double M_theta_solv; |
---|
| 52 | double M_phi_solv; |
---|
| 53 | double Up_frac_i; |
---|
| 54 | double Up_frac_f; |
---|
| 55 | double Up_theta; |
---|
[8343e18] | 56 | } ParallelepipedParameters; |
---|
| 57 | |
---|
| 58 | |
---|
| 59 | static double pkernel(double a, double b,double c, double ala, double alb, double alc){ |
---|
| 60 | // mu passed in is really mu*sqrt(1-sig^2) |
---|
| 61 | double argA,argB,argC,tmp1,tmp2,tmp3; //local variables |
---|
| 62 | |
---|
| 63 | //handle arg=0 separately, as sin(t)/t -> 1 as t->0 |
---|
| 64 | argA = a*ala/2.0; |
---|
| 65 | argB = b*alb/2.0; |
---|
| 66 | argC = c*alc/2.0; |
---|
| 67 | if(argA==0.0) { |
---|
| 68 | tmp1 = 1.0; |
---|
| 69 | } else { |
---|
| 70 | tmp1 = sin(argA)*sin(argA)/argA/argA; |
---|
| 71 | } |
---|
| 72 | if (argB==0.0) { |
---|
| 73 | tmp2 = 1.0; |
---|
| 74 | } else { |
---|
| 75 | tmp2 = sin(argB)*sin(argB)/argB/argB; |
---|
| 76 | } |
---|
| 77 | |
---|
| 78 | if (argC==0.0) { |
---|
| 79 | tmp3 = 1.0; |
---|
| 80 | } else { |
---|
| 81 | tmp3 = sin(argC)*sin(argC)/argC/argC; |
---|
| 82 | } |
---|
| 83 | |
---|
| 84 | return (tmp1*tmp2*tmp3); |
---|
| 85 | |
---|
| 86 | } |
---|
| 87 | |
---|
| 88 | /** |
---|
| 89 | * Function to evaluate 2D scattering function |
---|
| 90 | * @param pars: parameters of the parallelepiped |
---|
| 91 | * @param q: q-value |
---|
| 92 | * @param q_x: q_x / q |
---|
| 93 | * @param q_y: q_y / q |
---|
| 94 | * @return: function value |
---|
| 95 | */ |
---|
| 96 | static double parallelepiped_analytical_2D_scaled(ParallelepipedParameters *pars, double q, double q_x, double q_y) { |
---|
[318b5bbb] | 97 | double cparallel_x, cparallel_y, bparallel_x, bparallel_y, parallel_x, parallel_y; |
---|
| 98 | //double q_z; |
---|
| 99 | double vol, cos_val_c, cos_val_b, cos_val_a, edgeA, edgeB, edgeC; |
---|
[8343e18] | 100 | |
---|
[318b5bbb] | 101 | double answer = 0.0; |
---|
| 102 | double form = 0.0; |
---|
[8343e18] | 103 | double pi = 4.0*atan(1.0); |
---|
[318b5bbb] | 104 | |
---|
[8343e18] | 105 | //convert angle degree to radian |
---|
| 106 | double theta = pars->parallel_theta * pi/180.0; |
---|
| 107 | double phi = pars->parallel_phi * pi/180.0; |
---|
| 108 | double psi = pars->parallel_psi * pi/180.0; |
---|
[318b5bbb] | 109 | double sld_solv = pars->sldSolv; |
---|
| 110 | double sld_pipe = pars->sldPipe; |
---|
| 111 | double m_max = pars->M0_sld_pipe; |
---|
| 112 | double m_max_solv = pars->M0_sld_solv; |
---|
| 113 | double contrast = 0.0; |
---|
| 114 | |
---|
[8343e18] | 115 | edgeA = pars->short_a; |
---|
| 116 | edgeB = pars->short_b; |
---|
| 117 | edgeC = pars->long_c; |
---|
| 118 | |
---|
| 119 | |
---|
| 120 | // parallelepiped c axis orientation |
---|
[318b5bbb] | 121 | cparallel_x = cos(theta) * cos(phi); |
---|
| 122 | cparallel_y = sin(theta); |
---|
| 123 | //cparallel_z = -cos(theta) * sin(phi); |
---|
[8343e18] | 124 | |
---|
| 125 | // q vector |
---|
[318b5bbb] | 126 | //q_z = 0.0; |
---|
[8343e18] | 127 | |
---|
| 128 | // Compute the angle btw vector q and the |
---|
| 129 | // axis of the parallelepiped |
---|
[318b5bbb] | 130 | cos_val_c = cparallel_x*q_x + cparallel_y*q_y;// + cparallel_z*q_z; |
---|
| 131 | //alpha = acos(cos_val_c); |
---|
[8343e18] | 132 | |
---|
| 133 | // parallelepiped a axis orientation |
---|
[318b5bbb] | 134 | parallel_x = -cos(phi)*sin(psi) * sin(theta)+sin(phi)*cos(psi); |
---|
| 135 | parallel_y = sin(psi)*cos(theta); |
---|
| 136 | //parallel_z = sin(theta)*sin(phi)*sin(psi)+cos(phi)*cos(psi); |
---|
| 137 | |
---|
[8343e18] | 138 | cos_val_a = parallel_x*q_x + parallel_y*q_y; |
---|
| 139 | |
---|
| 140 | // parallelepiped b axis orientation |
---|
[318b5bbb] | 141 | bparallel_x = -sin(theta)*cos(psi)*cos(phi)-sin(psi)*sin(phi); |
---|
| 142 | bparallel_y = cos(theta)*cos(psi); |
---|
| 143 | //bparallel_z = sin(theta)*sin(phi)*cos(psi)-sin(psi)*cos(phi); |
---|
| 144 | |
---|
[8343e18] | 145 | // axis of the parallelepiped |
---|
[318b5bbb] | 146 | cos_val_b = bparallel_x*q_x + bparallel_y*q_y; |
---|
[8343e18] | 147 | |
---|
| 148 | // The following test should always pass |
---|
| 149 | if (fabs(cos_val_c)>1.0) { |
---|
[318b5bbb] | 150 | //printf("parallel_ana_2D: Unexpected error: cos(alpha)>1\n"); |
---|
| 151 | cos_val_c = 1.0; |
---|
| 152 | } |
---|
| 153 | if (fabs(cos_val_a)>1.0) { |
---|
| 154 | //printf("parallel_ana_2D: Unexpected error: cos(alpha)>1\n"); |
---|
| 155 | cos_val_a = 1.0; |
---|
| 156 | } |
---|
| 157 | if (fabs(cos_val_b)>1.0) { |
---|
| 158 | //printf("parallel_ana_2D: Unexpected error: cos(alpha)>1\n"); |
---|
| 159 | cos_val_b = 1.0; |
---|
[8343e18] | 160 | } |
---|
| 161 | // Call the IGOR library function to get the kernel |
---|
[318b5bbb] | 162 | form = pkernel( q*edgeA, q*edgeB, q*edgeC, cos_val_a, cos_val_b, cos_val_c); |
---|
| 163 | |
---|
| 164 | if (m_max < 1.0e-32 && m_max_solv < 1.0e-32){ |
---|
| 165 | // Multiply by contrast^2 |
---|
| 166 | contrast = (pars->sldPipe - pars->sldSolv); |
---|
| 167 | answer = contrast * contrast * form; |
---|
| 168 | } |
---|
| 169 | else{ |
---|
| 170 | double qx = q_x; |
---|
| 171 | double qy = q_y; |
---|
| 172 | double s_theta = pars->Up_theta; |
---|
| 173 | double m_phi = pars->M_phi_pipe; |
---|
| 174 | double m_theta = pars->M_theta_pipe; |
---|
| 175 | double m_phi_solv = pars->M_phi_solv; |
---|
| 176 | double m_theta_solv = pars->M_theta_solv; |
---|
| 177 | double in_spin = pars->Up_frac_i; |
---|
| 178 | double out_spin = pars->Up_frac_f; |
---|
| 179 | polar_sld p_sld; |
---|
| 180 | polar_sld p_sld_solv; |
---|
| 181 | p_sld = cal_msld(1, qx, qy, sld_pipe, m_max, m_theta, m_phi, |
---|
| 182 | in_spin, out_spin, s_theta); |
---|
| 183 | p_sld_solv = cal_msld(1, qx, qy, sld_solv, m_max_solv, m_theta_solv, m_phi_solv, |
---|
| 184 | in_spin, out_spin, s_theta); |
---|
| 185 | //up_up |
---|
| 186 | if (in_spin > 0.0 && out_spin > 0.0){ |
---|
| 187 | answer += ((p_sld.uu- p_sld_solv.uu) * (p_sld.uu- p_sld_solv.uu) * form); |
---|
| 188 | } |
---|
| 189 | //down_down |
---|
| 190 | if (in_spin < 1.0 && out_spin < 1.0){ |
---|
| 191 | answer += ((p_sld.dd - p_sld_solv.dd) * (p_sld.dd - p_sld_solv.dd) * form); |
---|
| 192 | } |
---|
| 193 | //up_down |
---|
| 194 | if (in_spin > 0.0 && out_spin < 1.0){ |
---|
| 195 | answer += ((p_sld.re_ud - p_sld_solv.re_ud) * (p_sld.re_ud - p_sld_solv.re_ud) * form); |
---|
| 196 | answer += ((p_sld.im_ud - p_sld_solv.im_ud) * (p_sld.im_ud - p_sld_solv.im_ud) * form); |
---|
| 197 | } |
---|
| 198 | //down_up |
---|
| 199 | if (in_spin < 1.0 && out_spin > 0.0){ |
---|
| 200 | answer += ((p_sld.re_du - p_sld_solv.re_du) * (p_sld.re_du - p_sld_solv.re_du) * form); |
---|
| 201 | answer += ((p_sld.im_du - p_sld_solv.im_du) * (p_sld.im_du - p_sld_solv.im_du) * form); |
---|
| 202 | } |
---|
| 203 | } |
---|
| 204 | |
---|
[8343e18] | 205 | |
---|
| 206 | //normalize by cylinder volume |
---|
| 207 | //NOTE that for this (Fournet) definition of the integral, one must MULTIPLY by Vparallel |
---|
[318b5bbb] | 208 | vol = edgeA* edgeB * edgeC; |
---|
[8343e18] | 209 | answer *= vol; |
---|
| 210 | |
---|
| 211 | //convert to [cm-1] |
---|
| 212 | answer *= 1.0e8; |
---|
| 213 | |
---|
| 214 | //Scale |
---|
| 215 | answer *= pars->scale; |
---|
| 216 | |
---|
| 217 | // add in the background |
---|
| 218 | answer += pars->background; |
---|
| 219 | |
---|
| 220 | return answer; |
---|
| 221 | } |
---|
| 222 | |
---|
| 223 | /** |
---|
| 224 | * Function to evaluate 2D scattering function |
---|
| 225 | * @param pars: parameters of the parallelepiped |
---|
| 226 | * @param q: q-value |
---|
| 227 | * @return: function value |
---|
| 228 | */ |
---|
| 229 | static double parallelepiped_analytical_2DXY(ParallelepipedParameters *pars, double qx, double qy) { |
---|
| 230 | double q; |
---|
| 231 | q = sqrt(qx*qx+qy*qy); |
---|
| 232 | return parallelepiped_analytical_2D_scaled(pars, q, qx/q, qy/q); |
---|
[8a48713] | 233 | } |
---|
| 234 | |
---|
| 235 | ParallelepipedModel :: ParallelepipedModel() { |
---|
| 236 | scale = Parameter(1.0); |
---|
[3c102d4] | 237 | short_a = Parameter(35.0, true); |
---|
[f9bf661] | 238 | short_a.set_min(1.0); |
---|
[8e36cdd] | 239 | short_b = Parameter(75.0, true); |
---|
| 240 | short_b.set_min(1.0); |
---|
| 241 | long_c = Parameter(400.0, true); |
---|
| 242 | long_c.set_min(1.0); |
---|
[f10063e] | 243 | sldPipe = Parameter(6.3e-6); |
---|
| 244 | sldSolv = Parameter(1.0e-6); |
---|
[8a48713] | 245 | background = Parameter(0.0); |
---|
| 246 | parallel_theta = Parameter(0.0, true); |
---|
| 247 | parallel_phi = Parameter(0.0, true); |
---|
[975ec8e] | 248 | parallel_psi = Parameter(0.0, true); |
---|
[318b5bbb] | 249 | M0_sld_pipe = Parameter(0.0e-6); |
---|
| 250 | M_theta_pipe = Parameter(0.0); |
---|
| 251 | M_phi_pipe = Parameter(0.0); |
---|
| 252 | M0_sld_solv = Parameter(0.0e-6); |
---|
| 253 | M_theta_solv = Parameter(0.0); |
---|
| 254 | M_phi_solv = Parameter(0.0); |
---|
| 255 | Up_frac_i = Parameter(0.5); |
---|
| 256 | Up_frac_f = Parameter(0.5); |
---|
| 257 | Up_theta = Parameter(0.0); |
---|
[8a48713] | 258 | } |
---|
| 259 | |
---|
| 260 | /** |
---|
| 261 | * Function to evaluate 1D scattering function |
---|
| 262 | * The NIST IGOR library is used for the actual calculation. |
---|
| 263 | * @param q: q-value |
---|
| 264 | * @return: function value |
---|
| 265 | */ |
---|
| 266 | double ParallelepipedModel :: operator()(double q) { |
---|
[f10063e] | 267 | double dp[7]; |
---|
[8a48713] | 268 | |
---|
| 269 | // Fill parameter array for IGOR library |
---|
| 270 | // Add the background after averaging |
---|
| 271 | dp[0] = scale(); |
---|
[3c102d4] | 272 | dp[1] = short_a(); |
---|
[8e36cdd] | 273 | dp[2] = short_b(); |
---|
| 274 | dp[3] = long_c(); |
---|
[f10063e] | 275 | dp[4] = sldPipe(); |
---|
| 276 | dp[5] = sldSolv(); |
---|
| 277 | dp[6] = 0.0; |
---|
[8a48713] | 278 | |
---|
| 279 | // Get the dispersion points for the short_edgeA |
---|
[3c102d4] | 280 | vector<WeightPoint> weights_short_a; |
---|
| 281 | short_a.get_weights(weights_short_a); |
---|
[975ec8e] | 282 | |
---|
[8a48713] | 283 | // Get the dispersion points for the longer_edgeB |
---|
[8e36cdd] | 284 | vector<WeightPoint> weights_short_b; |
---|
| 285 | short_b.get_weights(weights_short_b); |
---|
[8a48713] | 286 | |
---|
| 287 | // Get the dispersion points for the longuest_edgeC |
---|
[8e36cdd] | 288 | vector<WeightPoint> weights_long_c; |
---|
| 289 | long_c.get_weights(weights_long_c); |
---|
[8a48713] | 290 | |
---|
| 291 | |
---|
| 292 | |
---|
| 293 | // Perform the computation, with all weight points |
---|
| 294 | double sum = 0.0; |
---|
| 295 | double norm = 0.0; |
---|
[c451be9] | 296 | double vol = 0.0; |
---|
[975ec8e] | 297 | |
---|
[8a48713] | 298 | // Loop over short_edgeA weight points |
---|
[3c102d4] | 299 | for(int i=0; i< (int)weights_short_a.size(); i++) { |
---|
| 300 | dp[1] = weights_short_a[i].value; |
---|
[8a48713] | 301 | |
---|
| 302 | // Loop over longer_edgeB weight points |
---|
[8e36cdd] | 303 | for(int j=0; j< (int)weights_short_b.size(); j++) { |
---|
| 304 | dp[2] = weights_short_b[j].value; |
---|
[8a48713] | 305 | |
---|
| 306 | // Loop over longuest_edgeC weight points |
---|
[8e36cdd] | 307 | for(int k=0; k< (int)weights_long_c.size(); k++) { |
---|
| 308 | dp[3] = weights_long_c[k].value; |
---|
[c451be9] | 309 | //Un-normalize by volume |
---|
[8e36cdd] | 310 | sum += weights_short_a[i].weight * weights_short_b[j].weight |
---|
[c451be9] | 311 | * weights_long_c[k].weight * Parallelepiped(dp, q) |
---|
| 312 | * weights_short_a[i].value*weights_short_b[j].value |
---|
| 313 | * weights_long_c[k].value; |
---|
| 314 | //Find average volume |
---|
| 315 | vol += weights_short_a[i].weight * weights_short_b[j].weight |
---|
| 316 | * weights_long_c[k].weight |
---|
| 317 | * weights_short_a[i].value * weights_short_b[j].value |
---|
| 318 | * weights_long_c[k].value; |
---|
[8a48713] | 319 | |
---|
[3c102d4] | 320 | norm += weights_short_a[i].weight |
---|
[8e36cdd] | 321 | * weights_short_b[j].weight * weights_long_c[k].weight; |
---|
[8a48713] | 322 | } |
---|
| 323 | } |
---|
| 324 | } |
---|
[c451be9] | 325 | if (vol != 0.0 && norm != 0.0) { |
---|
| 326 | //Re-normalize by avg volume |
---|
| 327 | sum = sum/(vol/norm);} |
---|
[f9bf661] | 328 | |
---|
[8a48713] | 329 | return sum/norm + background(); |
---|
| 330 | } |
---|
| 331 | /** |
---|
| 332 | * Function to evaluate 2D scattering function |
---|
| 333 | * @param q_x: value of Q along x |
---|
| 334 | * @param q_y: value of Q along y |
---|
| 335 | * @return: function value |
---|
| 336 | */ |
---|
| 337 | double ParallelepipedModel :: operator()(double qx, double qy) { |
---|
| 338 | ParallelepipedParameters dp; |
---|
| 339 | // Fill parameter array |
---|
| 340 | dp.scale = scale(); |
---|
[3c102d4] | 341 | dp.short_a = short_a(); |
---|
[8e36cdd] | 342 | dp.short_b = short_b(); |
---|
| 343 | dp.long_c = long_c(); |
---|
[f10063e] | 344 | dp.sldPipe = sldPipe(); |
---|
| 345 | dp.sldSolv = sldSolv(); |
---|
[8a48713] | 346 | dp.background = 0.0; |
---|
| 347 | //dp.background = background(); |
---|
| 348 | dp.parallel_theta = parallel_theta(); |
---|
| 349 | dp.parallel_phi = parallel_phi(); |
---|
[975ec8e] | 350 | dp.parallel_psi = parallel_psi(); |
---|
[318b5bbb] | 351 | dp.Up_theta = Up_theta(); |
---|
| 352 | dp.M_phi_pipe = M_phi_pipe(); |
---|
| 353 | dp.M_theta_pipe = M_theta_pipe(); |
---|
| 354 | dp.M0_sld_pipe = M0_sld_pipe(); |
---|
| 355 | dp.M_phi_solv = M_phi_solv(); |
---|
| 356 | dp.M_theta_solv = M_theta_solv(); |
---|
| 357 | dp.M0_sld_solv = M0_sld_solv(); |
---|
| 358 | dp.Up_frac_i = Up_frac_i(); |
---|
| 359 | dp.Up_frac_f = Up_frac_f(); |
---|
[975ec8e] | 360 | |
---|
[8a48713] | 361 | |
---|
| 362 | // Get the dispersion points for the short_edgeA |
---|
[3c102d4] | 363 | vector<WeightPoint> weights_short_a; |
---|
| 364 | short_a.get_weights(weights_short_a); |
---|
[8a48713] | 365 | |
---|
| 366 | // Get the dispersion points for the longer_edgeB |
---|
[8e36cdd] | 367 | vector<WeightPoint> weights_short_b; |
---|
| 368 | short_b.get_weights(weights_short_b); |
---|
[8a48713] | 369 | |
---|
| 370 | // Get angular averaging for the longuest_edgeC |
---|
[8e36cdd] | 371 | vector<WeightPoint> weights_long_c; |
---|
| 372 | long_c.get_weights(weights_long_c); |
---|
[8a48713] | 373 | |
---|
| 374 | // Get angular averaging for theta |
---|
| 375 | vector<WeightPoint> weights_parallel_theta; |
---|
| 376 | parallel_theta.get_weights(weights_parallel_theta); |
---|
| 377 | |
---|
| 378 | // Get angular averaging for phi |
---|
| 379 | vector<WeightPoint> weights_parallel_phi; |
---|
| 380 | parallel_phi.get_weights(weights_parallel_phi); |
---|
| 381 | |
---|
[975ec8e] | 382 | // Get angular averaging for psi |
---|
| 383 | vector<WeightPoint> weights_parallel_psi; |
---|
| 384 | parallel_psi.get_weights(weights_parallel_psi); |
---|
[8a48713] | 385 | |
---|
| 386 | // Perform the computation, with all weight points |
---|
| 387 | double sum = 0.0; |
---|
| 388 | double norm = 0.0; |
---|
[c451be9] | 389 | double norm_vol = 0.0; |
---|
| 390 | double vol = 0.0; |
---|
[4628e31] | 391 | double pi = 4.0*atan(1.0); |
---|
[8a48713] | 392 | // Loop over radius weight points |
---|
[3c102d4] | 393 | for(int i=0; i< (int)weights_short_a.size(); i++) { |
---|
| 394 | dp.short_a = weights_short_a[i].value; |
---|
[8a48713] | 395 | |
---|
| 396 | // Loop over longer_edgeB weight points |
---|
[8e36cdd] | 397 | for(int j=0; j< (int)weights_short_b.size(); j++) { |
---|
| 398 | dp.short_b = weights_short_b[j].value; |
---|
[8a48713] | 399 | |
---|
| 400 | // Average over longuest_edgeC distribution |
---|
[8e36cdd] | 401 | for(int k=0; k< (int)weights_long_c.size(); k++) { |
---|
| 402 | dp.long_c = weights_long_c[k].value; |
---|
[8a48713] | 403 | |
---|
| 404 | // Average over theta distribution |
---|
| 405 | for(int l=0; l< (int)weights_parallel_theta.size(); l++) { |
---|
| 406 | dp.parallel_theta = weights_parallel_theta[l].value; |
---|
| 407 | |
---|
| 408 | // Average over phi distribution |
---|
| 409 | for(int m=0; m< (int)weights_parallel_phi.size(); m++) { |
---|
| 410 | dp.parallel_phi = weights_parallel_phi[m].value; |
---|
| 411 | |
---|
[975ec8e] | 412 | // Average over phi distribution |
---|
| 413 | for(int n=0; n< (int)weights_parallel_psi.size(); n++) { |
---|
| 414 | dp.parallel_psi = weights_parallel_psi[n].value; |
---|
[c451be9] | 415 | //Un-normalize by volume |
---|
[3c102d4] | 416 | double _ptvalue = weights_short_a[i].weight |
---|
[8e36cdd] | 417 | * weights_short_b[j].weight |
---|
| 418 | * weights_long_c[k].weight |
---|
[975ec8e] | 419 | * weights_parallel_theta[l].weight |
---|
| 420 | * weights_parallel_phi[m].weight |
---|
| 421 | * weights_parallel_psi[n].weight |
---|
[c451be9] | 422 | * parallelepiped_analytical_2DXY(&dp, qx, qy) |
---|
| 423 | * weights_short_a[i].value*weights_short_b[j].value |
---|
| 424 | * weights_long_c[k].value; |
---|
| 425 | |
---|
[975ec8e] | 426 | if (weights_parallel_theta.size()>1) { |
---|
[318b5bbb] | 427 | _ptvalue *= fabs(cos(weights_parallel_theta[l].value*pi/180.0)); |
---|
[975ec8e] | 428 | } |
---|
| 429 | sum += _ptvalue; |
---|
[c451be9] | 430 | //Find average volume |
---|
| 431 | vol += weights_short_a[i].weight |
---|
| 432 | * weights_short_b[j].weight |
---|
| 433 | * weights_long_c[k].weight |
---|
| 434 | * weights_short_a[i].value*weights_short_b[j].value |
---|
| 435 | * weights_long_c[k].value; |
---|
| 436 | //Find norm for volume |
---|
| 437 | norm_vol += weights_short_a[i].weight |
---|
| 438 | * weights_short_b[j].weight |
---|
| 439 | * weights_long_c[k].weight; |
---|
[975ec8e] | 440 | |
---|
[3c102d4] | 441 | norm += weights_short_a[i].weight |
---|
[8e36cdd] | 442 | * weights_short_b[j].weight |
---|
| 443 | * weights_long_c[k].weight |
---|
[975ec8e] | 444 | * weights_parallel_theta[l].weight |
---|
| 445 | * weights_parallel_phi[m].weight |
---|
| 446 | * weights_parallel_psi[n].weight; |
---|
[8a48713] | 447 | } |
---|
| 448 | } |
---|
| 449 | |
---|
| 450 | } |
---|
| 451 | } |
---|
| 452 | } |
---|
| 453 | } |
---|
| 454 | // Averaging in theta needs an extra normalization |
---|
| 455 | // factor to account for the sin(theta) term in the |
---|
| 456 | // integration (see documentation). |
---|
| 457 | if (weights_parallel_theta.size()>1) norm = norm / asin(1.0); |
---|
[c451be9] | 458 | |
---|
| 459 | if (vol != 0.0 && norm_vol != 0.0) { |
---|
| 460 | //Re-normalize by avg volume |
---|
| 461 | sum = sum/(vol/norm_vol);} |
---|
| 462 | |
---|
[8a48713] | 463 | return sum/norm + background(); |
---|
| 464 | } |
---|
| 465 | |
---|
| 466 | |
---|
| 467 | /** |
---|
| 468 | * Function to evaluate 2D scattering function |
---|
| 469 | * @param pars: parameters of the cylinder |
---|
| 470 | * @param q: q-value |
---|
| 471 | * @param phi: angle phi |
---|
| 472 | * @return: function value |
---|
| 473 | */ |
---|
| 474 | double ParallelepipedModel :: evaluate_rphi(double q, double phi) { |
---|
| 475 | double qx = q*cos(phi); |
---|
| 476 | double qy = q*sin(phi); |
---|
| 477 | return (*this).operator()(qx, qy); |
---|
| 478 | } |
---|
[5eb9154] | 479 | /** |
---|
| 480 | * Function to calculate effective radius |
---|
| 481 | * @return: effective radius value |
---|
| 482 | */ |
---|
| 483 | double ParallelepipedModel :: calculate_ER() { |
---|
[f9bf661] | 484 | ParallelepipedParameters dp; |
---|
| 485 | dp.short_a = short_a(); |
---|
| 486 | dp.short_b = short_b(); |
---|
| 487 | dp.long_c = long_c(); |
---|
| 488 | double rad_out = 0.0; |
---|
| 489 | double pi = 4.0*atan(1.0); |
---|
| 490 | double suf_rad = sqrt(dp.short_a*dp.short_b/pi); |
---|
| 491 | |
---|
| 492 | // Perform the computation, with all weight points |
---|
| 493 | double sum = 0.0; |
---|
| 494 | double norm = 0.0; |
---|
| 495 | |
---|
| 496 | // Get the dispersion points for the short_edgeA |
---|
| 497 | vector<WeightPoint> weights_short_a; |
---|
| 498 | short_a.get_weights(weights_short_a); |
---|
| 499 | |
---|
| 500 | // Get the dispersion points for the longer_edgeB |
---|
| 501 | vector<WeightPoint> weights_short_b; |
---|
| 502 | short_b.get_weights(weights_short_b); |
---|
| 503 | |
---|
| 504 | // Get angular averaging for the longuest_edgeC |
---|
| 505 | vector<WeightPoint> weights_long_c; |
---|
| 506 | long_c.get_weights(weights_long_c); |
---|
| 507 | |
---|
| 508 | // Loop over radius weight points |
---|
| 509 | for(int i=0; i< (int)weights_short_a.size(); i++) { |
---|
| 510 | dp.short_a = weights_short_a[i].value; |
---|
| 511 | |
---|
| 512 | // Loop over longer_edgeB weight points |
---|
| 513 | for(int j=0; j< (int)weights_short_b.size(); j++) { |
---|
| 514 | dp.short_b = weights_short_b[j].value; |
---|
| 515 | |
---|
| 516 | // Average over longuest_edgeC distribution |
---|
| 517 | for(int k=0; k< (int)weights_long_c.size(); k++) { |
---|
| 518 | dp.long_c = weights_long_c[k].value; |
---|
| 519 | //Calculate surface averaged radius |
---|
| 520 | //This is rough approximation. |
---|
| 521 | suf_rad = sqrt(dp.short_a*dp.short_b/pi); |
---|
| 522 | |
---|
| 523 | //Note: output of "DiamCyl(dp.length,dp.radius)" is DIAMETER. |
---|
| 524 | sum +=weights_short_a[i].weight* weights_short_b[j].weight |
---|
| 525 | * weights_long_c[k].weight*DiamCyl(dp.long_c, suf_rad)/2.0; |
---|
| 526 | norm += weights_short_a[i].weight* weights_short_b[j].weight*weights_long_c[k].weight; |
---|
| 527 | } |
---|
| 528 | } |
---|
| 529 | } |
---|
| 530 | |
---|
| 531 | if (norm != 0){ |
---|
| 532 | //return the averaged value |
---|
| 533 | rad_out = sum/norm;} |
---|
| 534 | else{ |
---|
| 535 | //return normal value |
---|
| 536 | //Note: output of "DiamCyl(length,radius)" is DIAMETER. |
---|
| 537 | rad_out = DiamCyl(dp.long_c, suf_rad)/2.0;} |
---|
| 538 | return rad_out; |
---|
| 539 | |
---|
[5eb9154] | 540 | } |
---|
[e08bd5b] | 541 | double ParallelepipedModel :: calculate_VR() { |
---|
| 542 | return 1.0; |
---|
| 543 | } |
---|