[18f2ca1] | 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 2010, 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 | |
---|
| 21 | #include <math.h> |
---|
| 22 | #include "parameters.hh" |
---|
| 23 | #include <stdio.h> |
---|
| 24 | using namespace std; |
---|
| 25 | |
---|
| 26 | extern "C" { |
---|
[c637521] | 27 | #include "libCylinder.h" |
---|
| 28 | #include "libStructureFactor.h" |
---|
[18f2ca1] | 29 | } |
---|
[c637521] | 30 | #include "csparallelepiped.h" |
---|
[18f2ca1] | 31 | |
---|
[df88829] | 32 | // Convenience parameter structure |
---|
| 33 | typedef struct { |
---|
[c637521] | 34 | double scale; |
---|
| 35 | double shortA; |
---|
| 36 | double midB; |
---|
| 37 | double longC; |
---|
| 38 | double rimA; |
---|
| 39 | double rimB; |
---|
| 40 | double rimC; |
---|
| 41 | double sld_rimA; |
---|
| 42 | double sld_rimB; |
---|
| 43 | double sld_rimC; |
---|
| 44 | double sld_pcore; |
---|
| 45 | double sld_solv; |
---|
| 46 | double background; |
---|
| 47 | double parallel_theta; |
---|
| 48 | double parallel_phi; |
---|
| 49 | double parallel_psi; |
---|
[df88829] | 50 | } CSParallelepipedParameters; |
---|
| 51 | |
---|
| 52 | static double cspkernel(double dp[],double q, double ala, double alb, double alc){ |
---|
[c637521] | 53 | // mu passed in is really mu*sqrt(1-sig^2) |
---|
| 54 | double argA,argB,argC,argtA,argtB,argtC,tmp1,tmp2,tmp3,tmpt1,tmpt2,tmpt3; //local variables |
---|
[df88829] | 55 | |
---|
| 56 | double aa,bb,cc, ta,tb,tc; |
---|
| 57 | double Vin,Vot,V1,V2,V3; |
---|
| 58 | double rhoA,rhoB,rhoC, rhoP, rhosolv; |
---|
| 59 | double dr0, drA,drB, drC; |
---|
| 60 | double retVal; |
---|
| 61 | |
---|
| 62 | aa = dp[1]; |
---|
| 63 | bb = dp[2]; |
---|
| 64 | cc = dp[3]; |
---|
| 65 | ta = dp[4]; |
---|
| 66 | tb = dp[5]; |
---|
| 67 | tc = dp[6]; |
---|
| 68 | rhoA=dp[7]; |
---|
| 69 | rhoB=dp[8]; |
---|
| 70 | rhoC=dp[9]; |
---|
| 71 | rhoP=dp[10]; |
---|
| 72 | rhosolv=dp[11]; |
---|
| 73 | dr0=rhoP-rhosolv; |
---|
| 74 | drA=rhoA-rhosolv; |
---|
| 75 | drB=rhoB-rhosolv; |
---|
| 76 | drC=rhoC-rhosolv; |
---|
| 77 | Vin=(aa*bb*cc); |
---|
| 78 | Vot=(aa*bb*cc+2.0*ta*bb*cc+2.0*aa*tb*cc+2.0*aa*bb*tc); |
---|
| 79 | V1=(2.0*ta*bb*cc); // incorrect V1 (aa*bb*cc+2*ta*bb*cc) |
---|
| 80 | V2=(2.0*aa*tb*cc); // incorrect V2(aa*bb*cc+2*aa*tb*cc) |
---|
| 81 | V3=(2.0*aa*bb*tc); |
---|
| 82 | //aa = aa/bb; |
---|
| 83 | ta=(aa+2.0*ta);///bb; |
---|
| 84 | tb=(aa+2.0*tb);///bb; |
---|
| 85 | tc=(aa+2.0*tc); |
---|
[c637521] | 86 | //handle arg=0 separately, as sin(t)/t -> 1 as t->0 |
---|
| 87 | argA = q*aa*ala/2.0; |
---|
| 88 | argB = q*bb*alb/2.0; |
---|
| 89 | argC = q*cc*alc/2.0; |
---|
| 90 | argtA = q*ta*ala/2.0; |
---|
[df88829] | 91 | argtB = q*tb*alb/2.0; |
---|
| 92 | argtC = q*tc*alc/2.0; |
---|
| 93 | |
---|
[c637521] | 94 | if(argA==0.0) { |
---|
[df88829] | 95 | tmp1 = 1.0; |
---|
| 96 | } else { |
---|
| 97 | tmp1 = sin(argA)/argA; |
---|
[c637521] | 98 | } |
---|
| 99 | if (argB==0.0) { |
---|
[df88829] | 100 | tmp2 = 1.0; |
---|
| 101 | } else { |
---|
| 102 | tmp2 = sin(argB)/argB; |
---|
[c637521] | 103 | } |
---|
[df88829] | 104 | |
---|
[c637521] | 105 | if (argC==0.0) { |
---|
[df88829] | 106 | tmp3 = 1.0; |
---|
| 107 | } else { |
---|
| 108 | tmp3 = sin(argC)/argC; |
---|
[c637521] | 109 | } |
---|
| 110 | if(argtA==0.0) { |
---|
[df88829] | 111 | tmpt1 = 1.0; |
---|
| 112 | } else { |
---|
| 113 | tmpt1 = sin(argtA)/argtA; |
---|
[c637521] | 114 | } |
---|
| 115 | if (argtB==0.0) { |
---|
[df88829] | 116 | tmpt2 = 1.0; |
---|
| 117 | } else { |
---|
| 118 | tmpt2 = sin(argtB)/argtB; |
---|
[c637521] | 119 | } |
---|
| 120 | if (argtC==0.0) { |
---|
[df88829] | 121 | tmpt3 = 1.0; |
---|
| 122 | } else { |
---|
| 123 | tmpt3 = sin(argtC)*sin(argtC)/argtC/argtC; |
---|
[c637521] | 124 | } |
---|
| 125 | // This expression is different from NIST/IGOR package (I strongly believe the IGOR is wrong!!!). 10/15/2010. |
---|
| 126 | retVal =( dr0*tmp1*tmp2*tmp3*Vin + drA*(tmpt1-tmp1)*tmp2*tmp3*V1+ drB*tmp1*(tmpt2-tmp2)*tmp3*V2 + drC*tmp1*tmp2*(tmpt3-tmp3)*V3)* |
---|
| 127 | ( dr0*tmp1*tmp2*tmp3*Vin + drA*(tmpt1-tmp1)*tmp2*tmp3*V1+ drB*tmp1*(tmpt2-tmp2)*tmp3*V2 + drC*tmp1*tmp2*(tmpt3-tmp3)*V3); // correct FF : square of sum of phase factors |
---|
| 128 | //retVal *= (tmp3*tmp3); |
---|
| 129 | retVal /= Vot; |
---|
[df88829] | 130 | |
---|
[c637521] | 131 | return (retVal); |
---|
[df88829] | 132 | |
---|
| 133 | }//Function cspkernel() |
---|
| 134 | |
---|
| 135 | /** |
---|
| 136 | * Function to evaluate 2D scattering function |
---|
| 137 | * @param pars: parameters of the CSparallelepiped |
---|
| 138 | * @param q: q-value |
---|
| 139 | * @param q_x: q_x / q |
---|
| 140 | * @param q_y: q_y / q |
---|
| 141 | * @return: function value |
---|
| 142 | */ |
---|
| 143 | static double csparallelepiped_analytical_2D_scaled(CSParallelepipedParameters *pars, double q, double q_x, double q_y) { |
---|
| 144 | double dp[13]; |
---|
| 145 | double cparallel_x, cparallel_y, cparallel_z, bparallel_x, bparallel_y, parallel_x, parallel_y; |
---|
| 146 | double q_z; |
---|
| 147 | double alpha, cos_val_c, cos_val_b, cos_val_a, edgeA, edgeB, edgeC; |
---|
| 148 | |
---|
| 149 | double answer; |
---|
| 150 | //convert angle degree to radian |
---|
| 151 | double pi = 4.0*atan(1.0); |
---|
| 152 | double theta = pars->parallel_theta * pi/180.0; |
---|
| 153 | double phi = pars->parallel_phi * pi/180.0; |
---|
| 154 | double psi = pars->parallel_psi* pi/180.0; |
---|
| 155 | |
---|
| 156 | // Fill paramater array |
---|
| 157 | dp[0] = 1.0; |
---|
| 158 | dp[1] = pars->shortA; |
---|
| 159 | dp[2] = pars->midB; |
---|
| 160 | dp[3] = pars->longC; |
---|
| 161 | dp[4] = pars->rimA; |
---|
| 162 | dp[5] = pars->rimB; |
---|
| 163 | dp[6] = pars->rimC; |
---|
| 164 | dp[7] = pars->sld_rimA; |
---|
| 165 | dp[8] = pars->sld_rimB; |
---|
| 166 | dp[9] = pars->sld_rimC; |
---|
| 167 | dp[10] = pars->sld_pcore; |
---|
| 168 | dp[11] = pars->sld_solv; |
---|
| 169 | dp[12] = 0.0; |
---|
| 170 | |
---|
| 171 | |
---|
| 172 | edgeA = pars->shortA; |
---|
| 173 | edgeB = pars->midB; |
---|
| 174 | edgeC = pars->longC; |
---|
| 175 | |
---|
| 176 | |
---|
[c637521] | 177 | // parallelepiped c axis orientation |
---|
| 178 | cparallel_x = sin(theta) * cos(phi); |
---|
| 179 | cparallel_y = sin(theta) * sin(phi); |
---|
| 180 | cparallel_z = cos(theta); |
---|
[df88829] | 181 | |
---|
[c637521] | 182 | // q vector |
---|
| 183 | q_z = 0.0; |
---|
[df88829] | 184 | |
---|
[c637521] | 185 | // Compute the angle btw vector q and the |
---|
| 186 | // axis of the parallelepiped |
---|
| 187 | cos_val_c = cparallel_x*q_x + cparallel_y*q_y + cparallel_z*q_z; |
---|
| 188 | alpha = acos(cos_val_c); |
---|
[df88829] | 189 | |
---|
[c637521] | 190 | // parallelepiped a axis orientation |
---|
| 191 | parallel_x = sin(psi);//cos(pars->parallel_theta) * sin(pars->parallel_phi)*sin(pars->parallel_psi); |
---|
| 192 | parallel_y = cos(psi);//cos(pars->parallel_theta) * cos(pars->parallel_phi)*cos(pars->parallel_psi); |
---|
[df88829] | 193 | |
---|
[c637521] | 194 | cos_val_a = parallel_x*q_x + parallel_y*q_y; |
---|
[df88829] | 195 | |
---|
| 196 | |
---|
| 197 | |
---|
[c637521] | 198 | // parallelepiped b axis orientation |
---|
| 199 | bparallel_x = sqrt(1.0-sin(theta)*cos(phi))*cos(psi);//cos(pars->parallel_theta) * cos(pars->parallel_phi)* cos(pars->parallel_psi); |
---|
| 200 | bparallel_y = sqrt(1.0-sin(theta)*cos(phi))*sin(psi);//cos(pars->parallel_theta) * sin(pars->parallel_phi)* sin(pars->parallel_psi); |
---|
| 201 | // axis of the parallelepiped |
---|
| 202 | cos_val_b = sin(acos(cos_val_a)) ; |
---|
[df88829] | 203 | |
---|
| 204 | |
---|
| 205 | |
---|
[c637521] | 206 | // The following test should always pass |
---|
| 207 | if (fabs(cos_val_c)>1.0) { |
---|
| 208 | printf("parallel_ana_2D: Unexpected error: cos(alpha)>1\n"); |
---|
| 209 | return 0; |
---|
| 210 | } |
---|
[df88829] | 211 | |
---|
| 212 | // Call the IGOR library function to get the kernel |
---|
| 213 | answer = cspkernel( dp,q, sin(alpha)*cos_val_a,sin(alpha)*cos_val_b,cos_val_c); |
---|
| 214 | |
---|
| 215 | //convert to [cm-1] |
---|
| 216 | answer *= 1.0e8; |
---|
| 217 | |
---|
| 218 | //Scale |
---|
| 219 | answer *= pars->scale; |
---|
| 220 | |
---|
| 221 | // add in the background |
---|
| 222 | answer += pars->background; |
---|
| 223 | |
---|
| 224 | return answer; |
---|
| 225 | } |
---|
| 226 | |
---|
| 227 | /** |
---|
| 228 | * Function to evaluate 2D scattering function |
---|
| 229 | * @param pars: parameters of the CSparallelepiped |
---|
| 230 | * @param q: q-value |
---|
| 231 | * @return: function value |
---|
| 232 | */ |
---|
| 233 | static double csparallelepiped_analytical_2DXY(CSParallelepipedParameters *pars, double qx, double qy) { |
---|
| 234 | double q; |
---|
| 235 | q = sqrt(qx*qx+qy*qy); |
---|
[c637521] | 236 | return csparallelepiped_analytical_2D_scaled(pars, q, qx/q, qy/q); |
---|
[df88829] | 237 | } |
---|
| 238 | |
---|
| 239 | |
---|
| 240 | |
---|
| 241 | |
---|
[18f2ca1] | 242 | CSParallelepipedModel :: CSParallelepipedModel() { |
---|
[c637521] | 243 | scale = Parameter(1.0); |
---|
| 244 | shortA = Parameter(35.0, true); |
---|
| 245 | shortA.set_min(1.0); |
---|
| 246 | midB = Parameter(75.0, true); |
---|
| 247 | midB.set_min(1.0); |
---|
| 248 | longC = Parameter(400.0, true); |
---|
| 249 | longC.set_min(1.0); |
---|
| 250 | rimA = Parameter(10.0, true); |
---|
| 251 | rimB = Parameter(10.0, true); |
---|
| 252 | rimC = Parameter(10.0, true); |
---|
| 253 | sld_rimA = Parameter(2.0e-6, true); |
---|
| 254 | sld_rimB = Parameter(4.0e-6, true); |
---|
| 255 | sld_rimC = Parameter(2.0e-6, true); |
---|
| 256 | sld_pcore = Parameter(1.0e-6); |
---|
| 257 | sld_solv = Parameter(6.0e-6); |
---|
| 258 | background = Parameter(0.06); |
---|
| 259 | parallel_theta = Parameter(0.0, true); |
---|
| 260 | parallel_phi = Parameter(0.0, true); |
---|
| 261 | parallel_psi = Parameter(0.0, true); |
---|
[18f2ca1] | 262 | } |
---|
| 263 | |
---|
| 264 | /** |
---|
| 265 | * Function to evaluate 1D scattering function |
---|
| 266 | * The NIST IGOR library is used for the actual calculation. |
---|
| 267 | * @param q: q-value |
---|
| 268 | * @return: function value |
---|
| 269 | */ |
---|
| 270 | double CSParallelepipedModel :: operator()(double q) { |
---|
[c637521] | 271 | double dp[13]; |
---|
| 272 | |
---|
| 273 | // Fill parameter array for IGOR library |
---|
| 274 | // Add the background after averaging |
---|
| 275 | dp[0] = scale(); |
---|
| 276 | dp[1] = shortA(); |
---|
| 277 | dp[2] = midB(); |
---|
| 278 | dp[3] = longC(); |
---|
| 279 | dp[4] = rimA(); |
---|
| 280 | dp[5] = rimB(); |
---|
| 281 | dp[6] = rimC(); |
---|
| 282 | dp[7] = sld_rimA(); |
---|
| 283 | dp[8] = sld_rimB(); |
---|
| 284 | dp[9] = sld_rimC(); |
---|
| 285 | dp[10] = sld_pcore(); |
---|
| 286 | dp[11] = sld_solv(); |
---|
| 287 | dp[12] = 0.0; |
---|
| 288 | |
---|
| 289 | // Get the dispersion points for the short_edgeA |
---|
| 290 | vector<WeightPoint> weights_shortA; |
---|
| 291 | shortA.get_weights(weights_shortA); |
---|
| 292 | |
---|
| 293 | // Get the dispersion points for the longer_edgeB |
---|
| 294 | vector<WeightPoint> weights_midB; |
---|
| 295 | midB.get_weights(weights_midB); |
---|
| 296 | |
---|
| 297 | // Get the dispersion points for the longuest_edgeC |
---|
| 298 | vector<WeightPoint> weights_longC; |
---|
| 299 | longC.get_weights(weights_longC); |
---|
| 300 | |
---|
| 301 | |
---|
| 302 | |
---|
| 303 | // Perform the computation, with all weight points |
---|
| 304 | double sum = 0.0; |
---|
| 305 | double norm = 0.0; |
---|
| 306 | double vol = 0.0; |
---|
| 307 | |
---|
| 308 | // Loop over short_edgeA weight points |
---|
| 309 | for(int i=0; i< (int)weights_shortA.size(); i++) { |
---|
| 310 | dp[1] = weights_shortA[i].value; |
---|
| 311 | |
---|
| 312 | // Loop over longer_edgeB weight points |
---|
| 313 | for(int j=0; j< (int)weights_midB.size(); j++) { |
---|
| 314 | dp[2] = weights_midB[j].value; |
---|
| 315 | |
---|
| 316 | // Loop over longuest_edgeC weight points |
---|
| 317 | for(int k=0; k< (int)weights_longC.size(); k++) { |
---|
| 318 | dp[3] = weights_longC[k].value; |
---|
| 319 | //Un-normalize by volume |
---|
| 320 | sum += weights_shortA[i].weight * weights_midB[j].weight |
---|
| 321 | * weights_longC[k].weight * CSParallelepiped(dp, q) |
---|
| 322 | * weights_shortA[i].value*weights_midB[j].value |
---|
| 323 | * weights_longC[k].value; |
---|
| 324 | //Find average volume |
---|
| 325 | vol += weights_shortA[i].weight * weights_midB[j].weight |
---|
| 326 | * weights_longC[k].weight |
---|
| 327 | * weights_shortA[i].value * weights_midB[j].value |
---|
| 328 | * weights_longC[k].value; |
---|
| 329 | |
---|
| 330 | norm += weights_shortA[i].weight |
---|
| 331 | * weights_midB[j].weight * weights_longC[k].weight; |
---|
| 332 | } |
---|
| 333 | } |
---|
| 334 | } |
---|
| 335 | if (vol != 0.0 && norm != 0.0) { |
---|
| 336 | //Re-normalize by avg volume |
---|
| 337 | sum = sum/(vol/norm);} |
---|
| 338 | |
---|
| 339 | return sum/norm + background(); |
---|
[18f2ca1] | 340 | } |
---|
| 341 | /** |
---|
| 342 | * Function to evaluate 2D scattering function |
---|
| 343 | * @param q_x: value of Q along x |
---|
| 344 | * @param q_y: value of Q along y |
---|
| 345 | * @return: function value |
---|
| 346 | */ |
---|
| 347 | double CSParallelepipedModel :: operator()(double qx, double qy) { |
---|
[c637521] | 348 | CSParallelepipedParameters dp; |
---|
| 349 | // Fill parameter array |
---|
| 350 | dp.scale = scale(); |
---|
| 351 | dp.shortA = shortA(); |
---|
| 352 | dp.midB = midB(); |
---|
| 353 | dp.longC = longC(); |
---|
| 354 | dp.rimA = rimA(); |
---|
| 355 | dp.rimB = rimB(); |
---|
| 356 | dp.rimC = rimC(); |
---|
| 357 | dp.sld_rimA = sld_rimA(); |
---|
| 358 | dp.sld_rimB = sld_rimB(); |
---|
| 359 | dp.sld_rimC = sld_rimC(); |
---|
| 360 | dp.sld_pcore = sld_pcore(); |
---|
| 361 | dp.sld_solv = sld_solv(); |
---|
| 362 | dp.background = 0.0; |
---|
| 363 | //dp.background = background(); |
---|
| 364 | dp.parallel_theta = parallel_theta(); |
---|
| 365 | dp.parallel_phi = parallel_phi(); |
---|
| 366 | dp.parallel_psi = parallel_psi(); |
---|
| 367 | |
---|
| 368 | |
---|
| 369 | |
---|
| 370 | // Get the dispersion points for the short_edgeA |
---|
| 371 | vector<WeightPoint> weights_shortA; |
---|
| 372 | shortA.get_weights(weights_shortA); |
---|
| 373 | |
---|
| 374 | // Get the dispersion points for the longer_edgeB |
---|
| 375 | vector<WeightPoint> weights_midB; |
---|
| 376 | midB.get_weights(weights_midB); |
---|
| 377 | |
---|
| 378 | // Get the dispersion points for the longuest_edgeC |
---|
| 379 | vector<WeightPoint> weights_longC; |
---|
| 380 | longC.get_weights(weights_longC); |
---|
| 381 | |
---|
| 382 | // Get angular averaging for theta |
---|
| 383 | vector<WeightPoint> weights_parallel_theta; |
---|
| 384 | parallel_theta.get_weights(weights_parallel_theta); |
---|
| 385 | |
---|
| 386 | // Get angular averaging for phi |
---|
| 387 | vector<WeightPoint> weights_parallel_phi; |
---|
| 388 | parallel_phi.get_weights(weights_parallel_phi); |
---|
| 389 | |
---|
| 390 | // Get angular averaging for psi |
---|
| 391 | vector<WeightPoint> weights_parallel_psi; |
---|
| 392 | parallel_psi.get_weights(weights_parallel_psi); |
---|
| 393 | |
---|
| 394 | // Perform the computation, with all weight points |
---|
| 395 | double sum = 0.0; |
---|
| 396 | double norm = 0.0; |
---|
| 397 | double norm_vol = 0.0; |
---|
| 398 | double vol = 0.0; |
---|
| 399 | double pi = 4.0*atan(1.0); |
---|
| 400 | |
---|
| 401 | // Loop over radius weight points |
---|
| 402 | for(int i=0; i< (int)weights_shortA.size(); i++) { |
---|
| 403 | dp.shortA = weights_shortA[i].value; |
---|
| 404 | |
---|
| 405 | // Loop over longer_edgeB weight points |
---|
| 406 | for(int j=0; j< (int)weights_midB.size(); j++) { |
---|
| 407 | dp.midB = weights_midB[j].value; |
---|
| 408 | |
---|
| 409 | // Average over longuest_edgeC distribution |
---|
| 410 | for(int k=0; k< (int)weights_longC.size(); k++) { |
---|
| 411 | dp.longC = weights_longC[k].value; |
---|
| 412 | |
---|
| 413 | // Average over theta distribution |
---|
| 414 | for(int l=0; l< (int)weights_parallel_theta.size(); l++) { |
---|
| 415 | dp.parallel_theta = weights_parallel_theta[l].value; |
---|
| 416 | |
---|
| 417 | // Average over phi distribution |
---|
| 418 | for(int m=0; m< (int)weights_parallel_phi.size(); m++) { |
---|
| 419 | dp.parallel_phi = weights_parallel_phi[m].value; |
---|
| 420 | |
---|
| 421 | // Average over phi distribution |
---|
| 422 | for(int n=0; n< (int)weights_parallel_psi.size(); n++) { |
---|
| 423 | dp.parallel_psi = weights_parallel_psi[n].value; |
---|
| 424 | //Un-normalize by volume |
---|
| 425 | double _ptvalue = weights_shortA[i].weight |
---|
| 426 | * weights_midB[j].weight |
---|
| 427 | * weights_longC[k].weight |
---|
| 428 | * weights_parallel_theta[l].weight |
---|
| 429 | * weights_parallel_phi[m].weight |
---|
| 430 | * weights_parallel_psi[n].weight |
---|
| 431 | * csparallelepiped_analytical_2DXY(&dp, qx, qy) |
---|
| 432 | * weights_shortA[i].value*weights_midB[j].value |
---|
| 433 | * weights_longC[k].value; |
---|
| 434 | |
---|
| 435 | if (weights_parallel_theta.size()>1) { |
---|
| 436 | _ptvalue *= fabs(sin(weights_parallel_theta[l].value*pi/180.0)); |
---|
| 437 | } |
---|
| 438 | sum += _ptvalue; |
---|
| 439 | //Find average volume |
---|
| 440 | vol += weights_shortA[i].weight |
---|
| 441 | * weights_midB[j].weight |
---|
| 442 | * weights_longC[k].weight |
---|
| 443 | * weights_shortA[i].value*weights_midB[j].value |
---|
| 444 | * weights_longC[k].value; |
---|
| 445 | //Find norm for volume |
---|
| 446 | norm_vol += weights_shortA[i].weight |
---|
| 447 | * weights_midB[j].weight |
---|
| 448 | * weights_longC[k].weight; |
---|
| 449 | |
---|
| 450 | norm += weights_shortA[i].weight |
---|
| 451 | * weights_midB[j].weight |
---|
| 452 | * weights_longC[k].weight |
---|
| 453 | * weights_parallel_theta[l].weight |
---|
| 454 | * weights_parallel_phi[m].weight |
---|
| 455 | * weights_parallel_psi[n].weight; |
---|
| 456 | } |
---|
| 457 | } |
---|
| 458 | |
---|
| 459 | } |
---|
| 460 | } |
---|
| 461 | } |
---|
| 462 | } |
---|
| 463 | // Averaging in theta needs an extra normalization |
---|
| 464 | // factor to account for the sin(theta) term in the |
---|
| 465 | // integration (see documentation). |
---|
| 466 | if (weights_parallel_theta.size()>1) norm = norm / asin(1.0); |
---|
| 467 | |
---|
| 468 | if (vol != 0.0 && norm_vol != 0.0) { |
---|
| 469 | //Re-normalize by avg volume |
---|
| 470 | sum = sum/(vol/norm_vol);} |
---|
| 471 | |
---|
| 472 | return sum/norm + background(); |
---|
[18f2ca1] | 473 | } |
---|
| 474 | |
---|
| 475 | |
---|
| 476 | /** |
---|
| 477 | * Function to evaluate 2D scattering function |
---|
| 478 | * @param pars: parameters of the cylinder |
---|
| 479 | * @param q: q-value |
---|
| 480 | * @param phi: angle phi |
---|
| 481 | * @return: function value |
---|
| 482 | */ |
---|
| 483 | double CSParallelepipedModel :: evaluate_rphi(double q, double phi) { |
---|
[c637521] | 484 | double qx = q*cos(phi); |
---|
| 485 | double qy = q*sin(phi); |
---|
| 486 | return (*this).operator()(qx, qy); |
---|
[18f2ca1] | 487 | } |
---|
| 488 | /** |
---|
| 489 | * Function to calculate effective radius |
---|
| 490 | * @return: effective radius value |
---|
| 491 | */ |
---|
| 492 | double CSParallelepipedModel :: calculate_ER() { |
---|
[c637521] | 493 | CSParallelepipedParameters dp; |
---|
| 494 | dp.shortA = shortA(); |
---|
| 495 | dp.midB = midB(); |
---|
| 496 | dp.longC = longC(); |
---|
| 497 | dp.rimA = rimA(); |
---|
| 498 | dp.rimB = rimB(); |
---|
| 499 | dp.rimC = rimC(); |
---|
| 500 | |
---|
| 501 | double rad_out = 0.0; |
---|
| 502 | double pi = 4.0*atan(1.0); |
---|
| 503 | double suf_rad = sqrt((dp.shortA*dp.midB+2.0*dp.rimA*dp.midB+2.0*dp.rimA*dp.shortA)/pi); |
---|
| 504 | double height =(dp.longC + 2.0*dp.rimC); |
---|
| 505 | // Perform the computation, with all weight points |
---|
| 506 | double sum = 0.0; |
---|
| 507 | double norm = 0.0; |
---|
| 508 | |
---|
| 509 | // Get the dispersion points for the short_edgeA |
---|
| 510 | vector<WeightPoint> weights_shortA; |
---|
| 511 | shortA.get_weights(weights_shortA); |
---|
| 512 | |
---|
| 513 | // Get the dispersion points for the longer_edgeB |
---|
| 514 | vector<WeightPoint> weights_midB; |
---|
| 515 | midB.get_weights(weights_midB); |
---|
| 516 | |
---|
| 517 | // Get angular averaging for the longuest_edgeC |
---|
| 518 | vector<WeightPoint> weights_longC; |
---|
| 519 | longC.get_weights(weights_longC); |
---|
| 520 | |
---|
| 521 | // Loop over radius weight points |
---|
| 522 | for(int i=0; i< (int)weights_shortA.size(); i++) { |
---|
| 523 | dp.shortA = weights_shortA[i].value; |
---|
| 524 | |
---|
| 525 | // Loop over longer_edgeB weight points |
---|
| 526 | for(int j=0; j< (int)weights_midB.size(); j++) { |
---|
| 527 | dp.midB = weights_midB[j].value; |
---|
| 528 | |
---|
| 529 | // Average over longuest_edgeC distribution |
---|
| 530 | for(int k=0; k< (int)weights_longC.size(); k++) { |
---|
| 531 | dp.longC = weights_longC[k].value; |
---|
| 532 | //Calculate surface averaged radius |
---|
| 533 | //This is rough approximation. |
---|
| 534 | suf_rad = sqrt((dp.shortA*dp.midB+2.0*dp.rimA*dp.midB+2.0*dp.rimA*dp.shortA)/pi); |
---|
| 535 | height =(dp.longC + 2.0*dp.rimC); |
---|
| 536 | //Note: output of "DiamCyl(dp.length,dp.radius)" is a DIAMETER. |
---|
| 537 | sum +=weights_shortA[i].weight* weights_midB[j].weight |
---|
| 538 | * weights_longC[k].weight*DiamCyl(height, suf_rad)/2.0; |
---|
| 539 | norm += weights_shortA[i].weight* weights_midB[j].weight*weights_longC[k].weight; |
---|
| 540 | } |
---|
| 541 | } |
---|
| 542 | } |
---|
| 543 | |
---|
| 544 | if (norm != 0){ |
---|
| 545 | //return the averaged value |
---|
| 546 | rad_out = sum/norm;} |
---|
| 547 | else{ |
---|
| 548 | //return normal value |
---|
| 549 | //Note: output of "DiamCyl(length,radius)" is DIAMETER. |
---|
| 550 | rad_out = DiamCyl(dp.longC, suf_rad)/2.0;} |
---|
| 551 | return rad_out; |
---|
[18f2ca1] | 552 | |
---|
| 553 | } |
---|
[e08bd5b] | 554 | double CSParallelepipedModel :: calculate_VR() { |
---|
| 555 | return 1.0; |
---|
| 556 | } |
---|