1 | |
---|
2 | #include <math.h> |
---|
3 | #include "parameters.hh" |
---|
4 | #include <stdio.h> |
---|
5 | #include <stdlib.h> |
---|
6 | #include "refl_adv.h" |
---|
7 | |
---|
8 | extern "C" { |
---|
9 | #include "libmultifunc/librefl.h" |
---|
10 | } |
---|
11 | |
---|
12 | using namespace std; |
---|
13 | |
---|
14 | #define lamda 4.62 |
---|
15 | |
---|
16 | |
---|
17 | double re_adv_kernel(double dp[], double q) { |
---|
18 | int n = dp[0]; |
---|
19 | int i,j; |
---|
20 | double nsl; |
---|
21 | |
---|
22 | double scale = dp[1]; |
---|
23 | double thick_inter_sub = dp[2]; |
---|
24 | double sld_sub = dp[4]; |
---|
25 | double sld_super = dp[5]; |
---|
26 | double background = dp[6]; |
---|
27 | double npts = dp[69]; //number of sub_layers in each interface |
---|
28 | |
---|
29 | double total_thick=0.0; |
---|
30 | |
---|
31 | int n_s; |
---|
32 | double sld_i,sldim_i,dz,phi,R,ko2; |
---|
33 | double pi; |
---|
34 | |
---|
35 | int* fun_type; |
---|
36 | double* sld; |
---|
37 | double* sld_im; |
---|
38 | double* thick_inter; |
---|
39 | double* thick; |
---|
40 | double* fun_coef; |
---|
41 | complex phi1,alpha,alpha2,kn,fnm,fnp,rn,Xn,nn,nn2,an,nnp1,one,two,n_sub,n_sup,knp1,Xnp1; |
---|
42 | |
---|
43 | fun_type = (int*)malloc((n+2)*sizeof(int)); |
---|
44 | sld = (double*)malloc((n+2)*sizeof(double)); |
---|
45 | sld_im = (double*)malloc((n+2)*sizeof(double)); |
---|
46 | thick_inter = (double*)malloc((n+2)*sizeof(double)); |
---|
47 | thick = (double*)malloc((n+2)*sizeof(double)); |
---|
48 | fun_coef = (double*)malloc((n+2)*sizeof(double)); |
---|
49 | |
---|
50 | fun_type[0] = dp[3]; |
---|
51 | fun_coef[0] = fabs(dp[70]); |
---|
52 | for (i =1; i<=n; i++){ |
---|
53 | sld[i] = dp[i+6]; |
---|
54 | thick_inter[i]= dp[i+16]; |
---|
55 | thick[i] = dp[i+26]; |
---|
56 | fun_type[i] = dp[i+36]; |
---|
57 | sld_im[i] = dp[i+46]; |
---|
58 | fun_coef[i] = fabs(dp[i+56]); |
---|
59 | //printf("type_func2 =%g\n",fun_coef[i]); |
---|
60 | |
---|
61 | total_thick += thick[i] + thick_inter[i]; |
---|
62 | } |
---|
63 | sld[0] = sld_sub; |
---|
64 | sld[n+1] = sld_super; |
---|
65 | sld_im[0] = fabs(dp[1+66]); |
---|
66 | sld_im[n+1] = fabs(dp[2+66]); |
---|
67 | thick[0] = total_thick/5.0; |
---|
68 | thick[n+1] = total_thick/5.0; |
---|
69 | thick_inter[0] = thick_inter_sub; |
---|
70 | thick_inter[n+1] = 0.0; |
---|
71 | fun_coef[n+1] = 0.0; |
---|
72 | |
---|
73 | nsl=npts;//21.0; //nsl = Num_sub_layer: MUST ODD number in double //no other number works now |
---|
74 | |
---|
75 | pi = 4.0*atan(1.0); |
---|
76 | one = cassign(1.0,0.0); |
---|
77 | Xn = cassign(0.0,0.0); |
---|
78 | two = cassign(0.0,-2.0); |
---|
79 | |
---|
80 | //Checking if floor is available. |
---|
81 | //no imaginary sld inputs in this function yet |
---|
82 | n_sub=cassign(1.0-sld_sub*pow(lamda,2.0)/(2.0*pi),pow(lamda,2.0)/(2.0*pi)*sld_im[0]); |
---|
83 | n_sup=cassign(1.0-sld_super*pow(lamda,2.0)/(2.0*pi),pow(lamda,2.0)/(2.0*pi)*sld_im[n+1]); |
---|
84 | ko2 = pow(2.0*pi/lamda,2.0); |
---|
85 | |
---|
86 | phi = asin(lamda*q/(4.0*pi)); |
---|
87 | phi1 = cplx_div(rcmult(phi,one),n_sup); |
---|
88 | alpha = cplx_mult(n_sup,cplx_cos(phi1)); |
---|
89 | alpha2 = cplx_mult(alpha,alpha); |
---|
90 | |
---|
91 | nnp1=n_sub; |
---|
92 | knp1=cplx_sqrt(rcmult(ko2,cplx_sub(cplx_mult(nnp1,nnp1),alpha2))); //nnp1*ko*sin(phinp1) |
---|
93 | Xnp1=cassign(0.0,0.0); |
---|
94 | dz = 0.0; |
---|
95 | // iteration for # of layers +sub from the top |
---|
96 | for (i=1;i<=n+1; i++){ |
---|
97 | //if (fun_coef[i]==0.0) |
---|
98 | // // this condition protects an error in numerical multiplication |
---|
99 | // fun_coef[i] = 1e-14; |
---|
100 | //iteration for 9 sub-layers |
---|
101 | for (j=0;j<2;j++){ |
---|
102 | for (n_s=0;n_s<nsl; n_s++){ |
---|
103 | // for flat layer |
---|
104 | if (j==1){ |
---|
105 | if (i==n+1) |
---|
106 | break; |
---|
107 | dz = thick[i]; |
---|
108 | sld_i = sld[i]; |
---|
109 | sldim_i = sld_im[i]; |
---|
110 | } |
---|
111 | // for interface |
---|
112 | else{ |
---|
113 | dz = thick_inter[i-1]/nsl; |
---|
114 | if (sld[i-1] == sld[i]){ |
---|
115 | sld_i = sld[i]; |
---|
116 | } |
---|
117 | else{ |
---|
118 | sld_i = intersldfunc(fun_type[i-1],nsl, n_s+0.5, fun_coef[i-1], sld[i-1], sld[i]); |
---|
119 | } |
---|
120 | if (sld_im[i-1] == sld_im[i]){ |
---|
121 | sldim_i = sld_im[i]; |
---|
122 | } |
---|
123 | else{ |
---|
124 | sldim_i = intersldfunc(fun_type[i-1],nsl, n_s+0.5, fun_coef[i-1], sld_im[i-1], sld_im[i]); |
---|
125 | } |
---|
126 | } |
---|
127 | nn = cassign(1.0-sld_i*pow(lamda,2.0)/(2.0*pi),pow(lamda,2.0)/(2.0*pi)*sldim_i); |
---|
128 | nn2=cplx_mult(nn,nn); |
---|
129 | |
---|
130 | kn=cplx_sqrt(rcmult(ko2,cplx_sub(nn2,alpha2))); //nn*ko*sin(phin) |
---|
131 | an=cplx_exp(rcmult(dz,cplx_mult(two,kn))); |
---|
132 | |
---|
133 | fnm=cplx_sub(kn,knp1); |
---|
134 | fnp=cplx_add(kn,knp1); |
---|
135 | rn=cplx_div(fnm,fnp); |
---|
136 | Xn=cplx_mult(an,cplx_div(cplx_add(rn,Xnp1),cplx_add(one,cplx_mult(rn,Xnp1)))); //Xn=an*((rn+Xnp1*anp1)/(1+rn*Xnp1*anp1)) |
---|
137 | |
---|
138 | Xnp1=Xn; |
---|
139 | knp1=kn; |
---|
140 | // no for-loop for flat layer |
---|
141 | if (j==1) |
---|
142 | break; |
---|
143 | } |
---|
144 | } |
---|
145 | } |
---|
146 | R=pow(Xn.re,2.0)+pow(Xn.im,2.0); |
---|
147 | // This temperarily fixes the total reflection for Rfunction and linear. |
---|
148 | // ToDo: Show why it happens that Xn.re=0 and Xn.im >1! |
---|
149 | if (Xn.im == 0.0 || R > 1){ |
---|
150 | R=1.0; |
---|
151 | } |
---|
152 | R *= scale; |
---|
153 | R += background; |
---|
154 | |
---|
155 | free(fun_type); |
---|
156 | free(sld); |
---|
157 | free(sld_im); |
---|
158 | free(thick_inter); |
---|
159 | free(thick); |
---|
160 | free(fun_coef); |
---|
161 | |
---|
162 | return R; |
---|
163 | |
---|
164 | } |
---|
165 | ReflAdvModel :: ReflAdvModel() { |
---|
166 | n_layers = Parameter(1.0); |
---|
167 | scale = Parameter(1.0); |
---|
168 | thick_inter0 = Parameter(1.0); |
---|
169 | func_inter0 = Parameter(0); |
---|
170 | sld_bottom0 = Parameter(2.07e-06); |
---|
171 | sld_medium = Parameter(1.0e-06); |
---|
172 | background = Parameter(0.0); |
---|
173 | |
---|
174 | |
---|
175 | sld_flat1 = Parameter(2.7e-06); |
---|
176 | sld_flat2 = Parameter(3.5e-06); |
---|
177 | sld_flat3 = Parameter(4.0e-06); |
---|
178 | sld_flat4 = Parameter(3.5e-06); |
---|
179 | sld_flat5 = Parameter(4.0e-06); |
---|
180 | sld_flat6 = Parameter(3.5e-06); |
---|
181 | sld_flat7 = Parameter(4.0e-06); |
---|
182 | sld_flat8 = Parameter(3.5e-06); |
---|
183 | sld_flat9 = Parameter(4.0e-06); |
---|
184 | sld_flat10 = Parameter(3.5e-06); |
---|
185 | |
---|
186 | |
---|
187 | thick_inter1 = Parameter(1.0); |
---|
188 | thick_inter2 = Parameter(1.0); |
---|
189 | thick_inter3 = Parameter(1.0); |
---|
190 | thick_inter4 = Parameter(1.0); |
---|
191 | thick_inter5 = Parameter(1.0); |
---|
192 | thick_inter6 = Parameter(1.0); |
---|
193 | thick_inter7 = Parameter(1.0); |
---|
194 | thick_inter8 = Parameter(1.0); |
---|
195 | thick_inter9 = Parameter(1.0); |
---|
196 | thick_inter10 = Parameter(1.0); |
---|
197 | |
---|
198 | |
---|
199 | thick_flat1 = Parameter(15.0); |
---|
200 | thick_flat2 = Parameter(100.0); |
---|
201 | thick_flat3 = Parameter(100.0); |
---|
202 | thick_flat4 = Parameter(100.0); |
---|
203 | thick_flat5 = Parameter(100.0); |
---|
204 | thick_flat6 = Parameter(100.0); |
---|
205 | thick_flat7 = Parameter(100.0); |
---|
206 | thick_flat8 = Parameter(100.0); |
---|
207 | thick_flat9 = Parameter(100.0); |
---|
208 | thick_flat10 = Parameter(100.0); |
---|
209 | |
---|
210 | |
---|
211 | func_inter1 = Parameter(0); |
---|
212 | func_inter2 = Parameter(0); |
---|
213 | func_inter3 = Parameter(0); |
---|
214 | func_inter4 = Parameter(0); |
---|
215 | func_inter5 = Parameter(0); |
---|
216 | func_inter6 = Parameter(0); |
---|
217 | func_inter7 = Parameter(0); |
---|
218 | func_inter8 = Parameter(0); |
---|
219 | func_inter9 = Parameter(0); |
---|
220 | func_inter10 = Parameter(0); |
---|
221 | |
---|
222 | sldIM_flat1 = Parameter(0); |
---|
223 | sldIM_flat2 = Parameter(0); |
---|
224 | sldIM_flat3 = Parameter(0); |
---|
225 | sldIM_flat4 = Parameter(0); |
---|
226 | sldIM_flat5 = Parameter(0); |
---|
227 | sldIM_flat6 = Parameter(0); |
---|
228 | sldIM_flat7 = Parameter(0); |
---|
229 | sldIM_flat8 = Parameter(0); |
---|
230 | sldIM_flat9 = Parameter(0); |
---|
231 | sldIM_flat10 = Parameter(0); |
---|
232 | |
---|
233 | nu_inter1 = Parameter(2.5); |
---|
234 | nu_inter2 = Parameter(2.5); |
---|
235 | nu_inter3 = Parameter(2.5); |
---|
236 | nu_inter4 = Parameter(2.5); |
---|
237 | nu_inter5 = Parameter(2.5); |
---|
238 | nu_inter6 = Parameter(2.5); |
---|
239 | nu_inter7 = Parameter(2.5); |
---|
240 | nu_inter8 = Parameter(2.5); |
---|
241 | nu_inter9 = Parameter(2.5); |
---|
242 | nu_inter10 = Parameter(2.5); |
---|
243 | |
---|
244 | sldIM_sub0 = Parameter(0.0); |
---|
245 | sldIM_medium = Parameter(0.0); |
---|
246 | npts_inter = Parameter(21.0); |
---|
247 | nu_inter0 = Parameter(2.5); |
---|
248 | } |
---|
249 | |
---|
250 | /** |
---|
251 | * Function to evaluate 1D NR function |
---|
252 | * @param q: q-value |
---|
253 | * @return: function value |
---|
254 | */ |
---|
255 | double ReflAdvModel :: operator()(double q) { |
---|
256 | double dp[71]; |
---|
257 | // Fill parameter array for IGOR library |
---|
258 | // Add the background after averaging |
---|
259 | dp[0] = n_layers(); |
---|
260 | dp[1] = scale(); |
---|
261 | dp[2] = thick_inter0(); |
---|
262 | dp[3] = func_inter0(); |
---|
263 | dp[4] = sld_bottom0(); |
---|
264 | dp[5] = sld_medium(); |
---|
265 | dp[6] = background(); |
---|
266 | |
---|
267 | dp[7] = sld_flat1(); |
---|
268 | dp[8] = sld_flat2(); |
---|
269 | dp[9] = sld_flat3(); |
---|
270 | dp[10] = sld_flat4(); |
---|
271 | dp[11] = sld_flat5(); |
---|
272 | dp[12] = sld_flat6(); |
---|
273 | dp[13] = sld_flat7(); |
---|
274 | dp[14] = sld_flat8(); |
---|
275 | dp[15] = sld_flat9(); |
---|
276 | dp[16] = sld_flat10(); |
---|
277 | |
---|
278 | dp[17] = thick_inter1(); |
---|
279 | dp[18] = thick_inter2(); |
---|
280 | dp[19] = thick_inter3(); |
---|
281 | dp[20] = thick_inter4(); |
---|
282 | dp[21] = thick_inter5(); |
---|
283 | dp[22] = thick_inter6(); |
---|
284 | dp[23] = thick_inter7(); |
---|
285 | dp[24] = thick_inter8(); |
---|
286 | dp[25] = thick_inter9(); |
---|
287 | dp[26] = thick_inter10(); |
---|
288 | |
---|
289 | dp[27] = thick_flat1(); |
---|
290 | dp[28] = thick_flat2(); |
---|
291 | dp[29] = thick_flat3(); |
---|
292 | dp[30] = thick_flat4(); |
---|
293 | dp[31] = thick_flat5(); |
---|
294 | dp[32] = thick_flat6(); |
---|
295 | dp[33] = thick_flat7(); |
---|
296 | dp[34] = thick_flat8(); |
---|
297 | dp[35] = thick_flat9(); |
---|
298 | dp[36] = thick_flat10(); |
---|
299 | |
---|
300 | dp[37] = func_inter1(); |
---|
301 | dp[38] = func_inter2(); |
---|
302 | dp[39] = func_inter3(); |
---|
303 | dp[40] = func_inter4(); |
---|
304 | dp[41] = func_inter5(); |
---|
305 | dp[42] = func_inter6(); |
---|
306 | dp[43] = func_inter7(); |
---|
307 | dp[44] = func_inter8(); |
---|
308 | dp[45] = func_inter9(); |
---|
309 | dp[46] = func_inter10(); |
---|
310 | |
---|
311 | dp[47] = sldIM_flat1(); |
---|
312 | dp[48] = sldIM_flat2(); |
---|
313 | dp[49] = sldIM_flat3(); |
---|
314 | dp[50] = sldIM_flat4(); |
---|
315 | dp[51] = sldIM_flat5(); |
---|
316 | dp[52] = sldIM_flat6(); |
---|
317 | dp[53] = sldIM_flat7(); |
---|
318 | dp[54] = sldIM_flat8(); |
---|
319 | dp[55] = sldIM_flat9(); |
---|
320 | dp[56] = sldIM_flat10(); |
---|
321 | |
---|
322 | dp[57] = nu_inter1(); |
---|
323 | dp[58] = nu_inter2(); |
---|
324 | dp[59] = nu_inter3(); |
---|
325 | dp[60] = nu_inter4(); |
---|
326 | dp[61] = nu_inter5(); |
---|
327 | dp[62] = nu_inter6(); |
---|
328 | dp[63] = nu_inter7(); |
---|
329 | dp[64] = nu_inter8(); |
---|
330 | dp[65] = nu_inter9(); |
---|
331 | dp[66] = nu_inter10(); |
---|
332 | |
---|
333 | dp[67] = sldIM_sub0(); |
---|
334 | dp[68] = sldIM_medium(); |
---|
335 | dp[69] = npts_inter(); |
---|
336 | dp[70] = nu_inter0(); |
---|
337 | // Get the dispersion points for the radius |
---|
338 | //vector<WeightPoint> weights_thick; |
---|
339 | //thick_inter0.get_weights(weights_thick_inter0); |
---|
340 | |
---|
341 | |
---|
342 | return re_adv_kernel(dp,q); |
---|
343 | } |
---|
344 | |
---|
345 | /** |
---|
346 | * Function to evaluate 2D NR function |
---|
347 | * @param q_x: value of Q along x |
---|
348 | * @param q_y: value of Q along y |
---|
349 | * @return: function value |
---|
350 | */ |
---|
351 | double ReflAdvModel :: operator()(double qx, double qy) { |
---|
352 | // For 2D set qy as q, ignoring qx. |
---|
353 | double q = qy;//sqrt(qx*qx + qy*qy); |
---|
354 | if (q < 0.0){ |
---|
355 | return 0.0; |
---|
356 | } |
---|
357 | return (*this).operator()(q); |
---|
358 | } |
---|
359 | |
---|
360 | /** |
---|
361 | * Function to evaluate 2D NR function |
---|
362 | * @param pars: parameters of the sphere |
---|
363 | * @param q: q-value |
---|
364 | * @param phi: angle phi |
---|
365 | * @return: function value |
---|
366 | */ |
---|
367 | double ReflAdvModel :: evaluate_rphi(double q, double phi) { |
---|
368 | return (*this).operator()(q); |
---|
369 | } |
---|
370 | |
---|
371 | /** |
---|
372 | * Function to calculate effective radius |
---|
373 | * @return: effective radius value |
---|
374 | */ |
---|
375 | double ReflAdvModel :: calculate_ER() { |
---|
376 | //NOT implemented yet!!! |
---|
377 | return 0.0; |
---|
378 | } |
---|
379 | double ReflAdvModel :: calculate_VR() { |
---|
380 | return 1.0; |
---|
381 | } |
---|