[02cc1ea] | 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 | */ |
---|
| 21 | // RKH 03Apr2014 a re-parametrised CoreShellEllipsoid with core axial ratio X and shell thickness T |
---|
| 22 | #include <math.h> |
---|
| 23 | #include "parameters.hh" |
---|
| 24 | #include <stdio.h> |
---|
| 25 | #include <stdlib.h> |
---|
| 26 | using namespace std; |
---|
| 27 | #include "spheroidXT.h" |
---|
| 28 | |
---|
| 29 | extern "C" { |
---|
| 30 | #include "libCylinder.h" |
---|
| 31 | #include "libStructureFactor.h" |
---|
| 32 | } |
---|
| 33 | |
---|
| 34 | typedef struct { |
---|
| 35 | double scale; |
---|
| 36 | double equat_core; |
---|
| 37 | double X_core; |
---|
| 38 | double T_shell; |
---|
| 39 | double XpolarShell; |
---|
| 40 | double sld_core; |
---|
| 41 | double sld_shell; |
---|
| 42 | double sld_solvent; |
---|
| 43 | double background; |
---|
| 44 | double axis_theta; |
---|
| 45 | double axis_phi; |
---|
| 46 | |
---|
| 47 | } SpheroidXTParameters; |
---|
| 48 | |
---|
| 49 | /** |
---|
| 50 | * Function to evaluate 2D scattering function |
---|
| 51 | * @param pars: parameters of the prolate |
---|
| 52 | * @param q: q-value |
---|
| 53 | * @param q_x: q_x / q |
---|
| 54 | * @param q_y: q_y / q |
---|
| 55 | * @return: function value |
---|
| 56 | */ |
---|
| 57 | static double spheroidXT_analytical_2D_scaled(SpheroidXTParameters *pars, double q, double q_x, double q_y) { |
---|
| 58 | |
---|
| 59 | double cyl_x, cyl_y;//, cyl_z; |
---|
| 60 | //double q_z; |
---|
| 61 | double alpha, vol, cos_val; |
---|
| 62 | double answer; |
---|
| 63 | double Pi = 4.0*atan(1.0); |
---|
| 64 | double sldcs,sldss; |
---|
| 65 | |
---|
| 66 | //convert angle degree to radian |
---|
| 67 | double theta = pars->axis_theta * Pi/180.0; |
---|
| 68 | double phi = pars->axis_phi * Pi/180.0; |
---|
| 69 | |
---|
| 70 | |
---|
| 71 | // ellipsoid orientation, the axis of the rotation is consistent with the ploar axis. |
---|
| 72 | cyl_x = cos(theta) * cos(phi); |
---|
| 73 | cyl_y = sin(theta); |
---|
| 74 | //cyl_z = -cos(theta) * sin(phi); |
---|
| 75 | //del sld |
---|
| 76 | sldcs = pars->sld_core - pars->sld_shell; |
---|
| 77 | sldss = pars->sld_shell- pars->sld_solvent; |
---|
| 78 | |
---|
| 79 | // q vector |
---|
| 80 | //q_z = 0; |
---|
| 81 | |
---|
| 82 | // Compute the angle btw vector q and the |
---|
| 83 | // axis of the cylinder |
---|
| 84 | cos_val = cyl_x*q_x + cyl_y*q_y;// + cyl_z*q_z; |
---|
| 85 | |
---|
| 86 | // The following test should always pass |
---|
| 87 | if (fabs(cos_val)>1.0) { |
---|
| 88 | printf("cyl_ana_2D: Unexpected error: cos(alpha)>1\n"); |
---|
| 89 | return 0; |
---|
| 90 | } |
---|
| 91 | |
---|
| 92 | // Note: cos(alpha) = 0 and 1 will get an |
---|
| 93 | // undefined value from CylKernel |
---|
| 94 | alpha = acos( cos_val ); |
---|
| 95 | |
---|
| 96 | // Call the IGOR library function to get the kernel: MUST use gfn4 not gf2 because of the def of params. |
---|
| 97 | // was answer = gfn4(cos_val,pars->equat_core,pars->polar_core,pars->equat_shell,pars->polar_shell,sldcs,sldss,q); |
---|
| 98 | answer = gfn4(cos_val,pars->equat_core, pars->equat_core*pars->X_core, pars->equat_core + pars->T_shell, pars->equat_core*pars->X_core + pars->T_shell*pars->XpolarShell ,sldcs,sldss,q); |
---|
| 99 | //It seems that it should be normalized somehow. How??? |
---|
| 100 | |
---|
| 101 | //normalize by cylinder volume |
---|
| 102 | //NOTE that for this (Fournet) definition of the integral, one must MULTIPLY by Vcyl |
---|
| 103 | // was vol = 4.0*Pi/3.0*pars->equat_shell*pars->equat_shell*pars->polar_shell; |
---|
| 104 | vol = 4.0*Pi/3.0*(pars->equat_core + pars->T_shell)* (pars->equat_core + pars->T_shell) * ( pars->equat_core*pars->X_core + pars->T_shell*pars->XpolarShell); |
---|
| 105 | answer /= vol; |
---|
| 106 | |
---|
| 107 | //convert to [cm-1] |
---|
| 108 | answer *= 1.0e8; |
---|
| 109 | |
---|
| 110 | //Scale |
---|
| 111 | answer *= pars->scale; |
---|
| 112 | |
---|
| 113 | // add in the background |
---|
| 114 | answer += pars->background; |
---|
| 115 | |
---|
| 116 | return answer; |
---|
| 117 | } |
---|
| 118 | |
---|
| 119 | CoreShellEllipsoidXTModel :: CoreShellEllipsoidXTModel() { |
---|
| 120 | scale = Parameter(1.0); |
---|
| 121 | equat_core = Parameter(20.0, true); |
---|
| 122 | equat_core.set_min(0.0); |
---|
| 123 | X_core = Parameter(3.0, true); |
---|
| 124 | X_core.set_min(0.001); |
---|
| 125 | T_shell = Parameter(30.0, true); |
---|
| 126 | T_shell.set_min(0.0); |
---|
| 127 | XpolarShell = Parameter(1.0, true); |
---|
| 128 | XpolarShell.set_min(0.0); |
---|
| 129 | sld_core = Parameter(2e-6); |
---|
| 130 | sld_shell = Parameter(1e-6); |
---|
| 131 | sld_solvent = Parameter(6.3e-6); |
---|
| 132 | background = Parameter(0.0); |
---|
| 133 | axis_theta = Parameter(0.0, true); |
---|
| 134 | axis_phi = Parameter(0.0, true); |
---|
| 135 | |
---|
| 136 | } |
---|
| 137 | |
---|
| 138 | |
---|
| 139 | /** |
---|
| 140 | * Function to evaluate 2D scattering function |
---|
| 141 | * @param pars: parameters of the prolate |
---|
| 142 | * @param q: q-value |
---|
| 143 | * @return: function value |
---|
| 144 | */ |
---|
| 145 | |
---|
| 146 | static double spheroidXT_analytical_2DXY(SpheroidXTParameters *pars, double qx, double qy) { |
---|
| 147 | double q; |
---|
| 148 | q = sqrt(qx*qx+qy*qy); |
---|
| 149 | return spheroidXT_analytical_2D_scaled(pars, q, qx/q, qy/q); |
---|
| 150 | } |
---|
| 151 | |
---|
| 152 | /** |
---|
| 153 | * Function to evaluate 1D scattering function |
---|
| 154 | * The NIST IGOR library is used for the actual calculation. |
---|
| 155 | * @param q: q-value |
---|
| 156 | * @return: function value |
---|
| 157 | */ |
---|
| 158 | double CoreShellEllipsoidXTModel :: operator()(double q) { |
---|
| 159 | double dp[9]; |
---|
| 160 | |
---|
| 161 | // Fill parameter array for IGOR library |
---|
| 162 | // Add the background after averaging |
---|
| 163 | dp[0] = scale(); |
---|
| 164 | dp[1] = equat_core(); |
---|
| 165 | //dp[2] = polar_core(); |
---|
| 166 | //dp[3] = equat_shell(); |
---|
| 167 | //dp[4] = polar_shell(); |
---|
| 168 | dp[2] = equat_core()*X_core(); |
---|
| 169 | dp[3] = equat_core() + T_shell(); |
---|
| 170 | dp[4] = equat_core()*X_core() + T_shell()*XpolarShell(); |
---|
| 171 | dp[5] = sld_core(); |
---|
| 172 | dp[6] = sld_shell(); |
---|
| 173 | dp[7] = sld_solvent(); |
---|
| 174 | dp[8] = 0.0; |
---|
| 175 | |
---|
| 176 | // Get the dispersion points for the major core |
---|
| 177 | vector<WeightPoint> weights_equat_core; |
---|
| 178 | equat_core.get_weights(weights_equat_core); |
---|
| 179 | |
---|
| 180 | // Get the dispersion points |
---|
| 181 | vector<WeightPoint> weights_X_core; |
---|
| 182 | X_core.get_weights(weights_X_core); |
---|
| 183 | |
---|
| 184 | // Get the dispersion points |
---|
| 185 | vector<WeightPoint> weights_T_shell; |
---|
| 186 | T_shell.get_weights(weights_T_shell); |
---|
| 187 | |
---|
| 188 | // Get the dispersion points |
---|
| 189 | vector<WeightPoint> weights_XpolarShell; |
---|
| 190 | XpolarShell.get_weights(weights_XpolarShell); |
---|
| 191 | |
---|
| 192 | |
---|
| 193 | // Perform the computation, with all weight points |
---|
| 194 | double sum = 0.0; |
---|
| 195 | double norm = 0.0; |
---|
| 196 | double vol = 0.0; |
---|
| 197 | double wproduct = 0.0; |
---|
| 198 | |
---|
| 199 | // Loop over equat core weight points |
---|
| 200 | for(int i=0; i<(int)weights_equat_core.size(); i++) { |
---|
| 201 | dp[1] = weights_equat_core[i].value; |
---|
| 202 | |
---|
| 203 | // Loop over polar core weight points |
---|
| 204 | for(int j=0; j<(int)weights_X_core.size(); j++) { |
---|
| 205 | dp[2] = dp[1]*weights_X_core[j].value; |
---|
| 206 | |
---|
| 207 | // Loop over equat outer weight points |
---|
| 208 | for(int k=0; k<(int)weights_T_shell.size(); k++) { |
---|
| 209 | dp[3] = dp[1] + weights_T_shell[k].value; |
---|
| 210 | |
---|
| 211 | // Loop over polar outer weight points |
---|
| 212 | for(int l=0; l<(int)weights_XpolarShell.size(); l++) { |
---|
| 213 | dp[4] = dp[2] + weights_XpolarShell[l].value*weights_T_shell[k].value; |
---|
| 214 | |
---|
| 215 | //Un-normalize by volume |
---|
| 216 | wproduct =weights_equat_core[i].weight* weights_X_core[j].weight * weights_T_shell[k].weight |
---|
| 217 | * weights_XpolarShell[l].weight; |
---|
| 218 | sum += wproduct * OblateForm(dp, q) * dp[3]*dp[3]*dp[4]; |
---|
| 219 | //Find average volume |
---|
| 220 | vol += wproduct * dp[3]*dp[3]*dp[4]; |
---|
| 221 | // was pow(weights_equat_shell[k].value,2)*weights_polar_shell[l].value; |
---|
| 222 | norm += wproduct; |
---|
| 223 | } |
---|
| 224 | } |
---|
| 225 | } |
---|
| 226 | } |
---|
| 227 | if (vol != 0.0 && norm != 0.0) { |
---|
| 228 | //Re-normalize by avg volume |
---|
| 229 | sum = sum/(vol/norm);} |
---|
| 230 | return sum/norm + background(); |
---|
| 231 | } |
---|
| 232 | |
---|
| 233 | /** |
---|
| 234 | * Function to evaluate 2D scattering function |
---|
| 235 | * @param q_x: value of Q along x |
---|
| 236 | * @param q_y: value of Q along y |
---|
| 237 | * @return: function value |
---|
| 238 | */ |
---|
| 239 | /* |
---|
| 240 | double OblateModel :: operator()(double qx, double qy) { |
---|
| 241 | double q = sqrt(qx*qx + qy*qy); |
---|
| 242 | |
---|
| 243 | return (*this).operator()(q); |
---|
| 244 | } |
---|
| 245 | */ |
---|
| 246 | |
---|
| 247 | /** |
---|
| 248 | * Function to evaluate 2D scattering function |
---|
| 249 | * @param pars: parameters of the oblate |
---|
| 250 | * @param q: q-value |
---|
| 251 | * @param phi: angle phi |
---|
| 252 | * @return: function value |
---|
| 253 | */ |
---|
| 254 | double CoreShellEllipsoidXTModel :: evaluate_rphi(double q, double phi) { |
---|
| 255 | double qx = q*cos(phi); |
---|
| 256 | double qy = q*sin(phi); |
---|
| 257 | return (*this).operator()(qx, qy); |
---|
| 258 | } |
---|
| 259 | |
---|
| 260 | /** |
---|
| 261 | * Function to evaluate 2D scattering function |
---|
| 262 | * @param q_x: value of Q along x |
---|
| 263 | * @param q_y: value of Q along y |
---|
| 264 | * @return: function value |
---|
| 265 | */ |
---|
| 266 | double CoreShellEllipsoidXTModel :: operator()(double qx, double qy) { |
---|
| 267 | SpheroidXTParameters dp; |
---|
| 268 | // Fill parameter array |
---|
| 269 | dp.scale = scale(); |
---|
| 270 | dp.equat_core = equat_core(); |
---|
| 271 | dp.X_core = X_core(); |
---|
| 272 | dp.T_shell = T_shell(); |
---|
| 273 | dp.XpolarShell = XpolarShell(); |
---|
| 274 | dp.sld_core = sld_core(); |
---|
| 275 | dp.sld_shell = sld_shell(); |
---|
| 276 | dp.sld_solvent = sld_solvent(); |
---|
| 277 | dp.background = 0.0; |
---|
| 278 | dp.axis_theta = axis_theta(); |
---|
| 279 | dp.axis_phi = axis_phi(); |
---|
| 280 | |
---|
| 281 | // Get the dispersion points for the major core |
---|
| 282 | vector<WeightPoint> weights_equat_core; |
---|
| 283 | equat_core.get_weights(weights_equat_core); |
---|
| 284 | |
---|
| 285 | // Get the dispersion points |
---|
| 286 | vector<WeightPoint> weights_X_core; |
---|
| 287 | X_core.get_weights(weights_X_core); |
---|
| 288 | |
---|
| 289 | // Get the dispersion points |
---|
| 290 | vector<WeightPoint> weights_T_shell; |
---|
| 291 | T_shell.get_weights(weights_T_shell); |
---|
| 292 | |
---|
| 293 | // Get the dispersion points |
---|
| 294 | vector<WeightPoint> weights_XpolarShell; |
---|
| 295 | XpolarShell.get_weights(weights_XpolarShell); |
---|
| 296 | |
---|
| 297 | |
---|
| 298 | // Get angular averaging for theta |
---|
| 299 | vector<WeightPoint> weights_theta; |
---|
| 300 | axis_theta.get_weights(weights_theta); |
---|
| 301 | |
---|
| 302 | // Get angular averaging for phi |
---|
| 303 | vector<WeightPoint> weights_phi; |
---|
| 304 | axis_phi.get_weights(weights_phi); |
---|
| 305 | |
---|
| 306 | // Perform the computation, with all weight points |
---|
| 307 | double sum = 0.0; |
---|
| 308 | double norm = 0.0; |
---|
| 309 | double norm_vol = 0.0; |
---|
| 310 | double vol = 0.0; |
---|
| 311 | double pi = 4.0*atan(1.0); |
---|
| 312 | double equat_outer = 0.0; |
---|
| 313 | double polar_outer = 0.0; |
---|
| 314 | |
---|
| 315 | // Loop over major core weight points |
---|
| 316 | for(int i=0; i< (int)weights_equat_core.size(); i++) { |
---|
| 317 | dp.equat_core = weights_equat_core[i].value; |
---|
| 318 | |
---|
| 319 | // Loop over minor core weight points |
---|
| 320 | for(int j=0; j< (int)weights_X_core.size(); j++) { |
---|
| 321 | dp.X_core = weights_X_core[j].value; |
---|
| 322 | |
---|
| 323 | // Loop over equat outer weight points |
---|
| 324 | for(int k=0; k< (int)weights_T_shell.size(); k++) { |
---|
| 325 | dp.T_shell = weights_T_shell[k].value; |
---|
| 326 | equat_outer = weights_equat_core[i].value + weights_T_shell[k].value; |
---|
| 327 | |
---|
| 328 | // Loop over polar outer weight points |
---|
| 329 | for(int l=0; l< (int)weights_XpolarShell.size(); l++) { |
---|
| 330 | dp.XpolarShell = weights_XpolarShell[l].value; |
---|
| 331 | polar_outer = weights_equat_core[i].value*weights_X_core[j].value + weights_T_shell[k].value*weights_XpolarShell[l].value; |
---|
| 332 | |
---|
| 333 | // Average over theta distribution |
---|
| 334 | for(int m=0; m< (int)weights_theta.size(); m++) { |
---|
| 335 | dp.axis_theta = weights_theta[m].value; |
---|
| 336 | |
---|
| 337 | // Average over phi distribution |
---|
| 338 | for(int n=0; n< (int)weights_phi.size(); n++) { |
---|
| 339 | dp.axis_phi = weights_phi[n].value; |
---|
| 340 | //Un-normalize by volume |
---|
| 341 | double _ptvalue = weights_equat_core[i].weight *weights_X_core[j].weight |
---|
| 342 | * weights_T_shell[k].weight * weights_XpolarShell[l].weight |
---|
| 343 | * weights_theta[m].weight |
---|
| 344 | * weights_phi[n].weight |
---|
| 345 | // rkh NOTE this passes the NEW parameters |
---|
| 346 | * spheroidXT_analytical_2DXY(&dp, qx, qy) * pow(equat_outer,2)*polar_outer; |
---|
| 347 | if (weights_theta.size()>1) { |
---|
| 348 | _ptvalue *= fabs(cos(weights_theta[m].value*pi/180.0)); |
---|
| 349 | } |
---|
| 350 | sum += _ptvalue; |
---|
| 351 | //Find average volume |
---|
| 352 | // rkh had to change this, original weighted by outer shell volume weights only, see spheroid.cpp, which looks odd, |
---|
| 353 | // (as has to assume that weights of other loops sume to unity) and here we need all four loops to get the outer size. |
---|
| 354 | vol += weights_equat_core[i].weight *weights_X_core[j].weight |
---|
| 355 | * weights_T_shell[k].weight * weights_XpolarShell[l].weight |
---|
| 356 | * pow(equat_outer,2)*polar_outer; |
---|
| 357 | //Find norm for volume |
---|
| 358 | norm_vol += weights_equat_core[i].weight *weights_X_core[j].weight |
---|
| 359 | * weights_T_shell[k].weight * weights_XpolarShell[l].weight; |
---|
| 360 | |
---|
| 361 | norm += weights_equat_core[i].weight *weights_X_core[j].weight |
---|
| 362 | * weights_T_shell[k].weight * weights_XpolarShell[l].weight |
---|
| 363 | * weights_theta[m].weight * weights_phi[n].weight; |
---|
| 364 | } |
---|
| 365 | } |
---|
| 366 | } |
---|
| 367 | } |
---|
| 368 | } |
---|
| 369 | } |
---|
| 370 | // Averaging in theta needs an extra normalization |
---|
| 371 | // factor to account for the sin(theta) term in the |
---|
| 372 | // integration (see documentation). |
---|
| 373 | if (weights_theta.size()>1) norm = norm / asin(1.0); |
---|
| 374 | |
---|
| 375 | if (vol != 0.0 && norm_vol != 0.0) { |
---|
| 376 | //Re-normalize by avg volume |
---|
| 377 | sum = sum/(vol/norm_vol);} |
---|
| 378 | |
---|
| 379 | return sum/norm + background(); |
---|
| 380 | } |
---|
| 381 | |
---|
| 382 | /** |
---|
| 383 | * Function to calculate effective radius |
---|
| 384 | * @return: effective radius value |
---|
| 385 | * rkh This now needs to integrate over all four variables as above not just two |
---|
| 386 | */ |
---|
| 387 | double CoreShellEllipsoidXTModel :: calculate_ER() { |
---|
| 388 | SpheroidXTParameters dp; |
---|
| 389 | |
---|
| 390 | dp.equat_core = equat_core(); |
---|
| 391 | dp.X_core = X_core(); |
---|
| 392 | dp.T_shell = T_shell(); |
---|
| 393 | dp.XpolarShell = XpolarShell(); |
---|
| 394 | |
---|
| 395 | double rad_out = 0.0; |
---|
| 396 | double equat_outer = 0.0; |
---|
| 397 | double polar_outer = 0.0; |
---|
| 398 | |
---|
| 399 | // Perform the computation, with all weight points |
---|
| 400 | double sum = 0.0; |
---|
| 401 | double norm = 0.0; |
---|
| 402 | |
---|
| 403 | // Get the dispersion points for the core |
---|
| 404 | vector<WeightPoint> weights_equat_core; |
---|
| 405 | equat_core.get_weights(weights_equat_core); |
---|
| 406 | |
---|
| 407 | // Get the dispersion points |
---|
| 408 | vector<WeightPoint> weights_X_core; |
---|
| 409 | X_core.get_weights(weights_X_core); |
---|
| 410 | |
---|
| 411 | // Get the dispersion points |
---|
| 412 | vector<WeightPoint> weights_T_shell; |
---|
| 413 | T_shell.get_weights(weights_T_shell); |
---|
| 414 | |
---|
| 415 | // Get the dispersion points |
---|
| 416 | vector<WeightPoint> weights_XpolarShell; |
---|
| 417 | XpolarShell.get_weights(weights_XpolarShell); |
---|
| 418 | |
---|
| 419 | |
---|
| 420 | // Loop over core weight points |
---|
| 421 | for(int i=0; i< (int)weights_equat_core.size(); i++) { |
---|
| 422 | dp.equat_core = weights_equat_core[i].value; |
---|
| 423 | // Loop over weight points |
---|
| 424 | for(int j=0; j< (int)weights_X_core.size(); j++) { |
---|
| 425 | // Loop over weight points |
---|
| 426 | for(int k=0; k< (int)weights_T_shell.size(); k++) { |
---|
| 427 | equat_outer = weights_equat_core[i].value + weights_T_shell[k].value; |
---|
| 428 | // Loop over polar outer weight points |
---|
| 429 | for(int l=0; l< (int)weights_XpolarShell.size(); l++) { |
---|
| 430 | polar_outer = weights_equat_core[i].value*weights_X_core[j].value + weights_T_shell[k].value*weights_XpolarShell[l].value; |
---|
| 431 | //Note: output of "DiamEllip(dp.polar_shell,dp.equat_shell)" is DIAMETER. |
---|
| 432 | sum +=weights_equat_core[i].weight *weights_X_core[j].weight |
---|
| 433 | * weights_T_shell[k].weight * weights_XpolarShell[l].weight*DiamEllip(polar_outer,equat_outer)/2.0; |
---|
| 434 | norm += weights_equat_core[i].weight *weights_X_core[j].weight |
---|
| 435 | * weights_T_shell[k].weight * weights_XpolarShell[l].weight; |
---|
| 436 | } |
---|
| 437 | } |
---|
| 438 | } |
---|
| 439 | } |
---|
| 440 | if (norm != 0){ |
---|
| 441 | //return the averaged value |
---|
| 442 | rad_out = sum/norm;} |
---|
| 443 | else{ |
---|
| 444 | //return normal value |
---|
| 445 | //Note: output of "DiamEllip(dp.polar_shell,dp.equat_shell)" is DIAMETER. |
---|
| 446 | rad_out = DiamEllip(dp.equat_core + dp.T_shell, dp.equat_core * dp.X_core + dp.T_shell*dp.XpolarShell)/2.0;} |
---|
| 447 | |
---|
| 448 | return rad_out; |
---|
| 449 | } |
---|
| 450 | double CoreShellEllipsoidXTModel :: calculate_VR() { |
---|
| 451 | return 1.0; |
---|
| 452 | } |
---|