218 lines
6.8 KiB
C++
218 lines
6.8 KiB
C++
/** @file WaveSimulation.cpp
|
|
*
|
|
* @author Cory Alexander Balaton (coryab)
|
|
* @author Janita Ovidie Sandtrøen Willumsen (janitaws)
|
|
*
|
|
* @version 0.1
|
|
*
|
|
* @brief Implementation of the WaveSimulation class.
|
|
*
|
|
* @bug No known bugs
|
|
* */
|
|
#include "WaveSimulation.hpp"
|
|
|
|
#include "utils.hpp"
|
|
|
|
#include <cmath>
|
|
#include <complex>
|
|
#include <cstdint>
|
|
#include <cstdlib>
|
|
|
|
WaveSimulation::WaveSimulation(double h, double dt, double T, double x_c,
|
|
double y_c, double sigma_x, double sigma_y,
|
|
double p_x, double p_y, double thickness,
|
|
double pos_x, double ap_sep, double ap,
|
|
uint32_t slits)
|
|
{
|
|
this->dt = dt;
|
|
this->h = h;
|
|
this->T = T;
|
|
this->M = 1. / h;
|
|
this->N = M - 2;
|
|
this->build_V(thickness, pos_x, ap_sep, ap, slits);
|
|
this->initialize_U(x_c, y_c, sigma_x, sigma_y, p_x, p_y);
|
|
this->build_A();
|
|
this->build_B();
|
|
}
|
|
|
|
WaveSimulation::WaveSimulation(double h, double dt, double T, double x_c, double y_c,
|
|
double sigma_x, double sigma_y, double p_x, double p_y)
|
|
{
|
|
this->dt = dt;
|
|
this->h = h;
|
|
this->T = T;
|
|
this->M = 1. / h;
|
|
this->N = M - 2;
|
|
this->build_V(0.,0.,0.,0., 0);
|
|
this->initialize_U(x_c, y_c, sigma_x, sigma_y, p_x, p_y);
|
|
this->build_A();
|
|
this->build_B();
|
|
|
|
}
|
|
|
|
void WaveSimulation::solve(std::ofstream &ofile)
|
|
{
|
|
ofile << this->N << '\n';
|
|
uint32_t iterations = this->T / this->dt;
|
|
for (size_t i = 0; i < iterations; i++) {
|
|
this->write_U(ofile);
|
|
this->step();
|
|
}
|
|
}
|
|
|
|
void WaveSimulation::step()
|
|
{
|
|
arma::cx_vec tmp = this->B * this->U.as_col();
|
|
arma::spsolve(this->U, this->A, tmp);
|
|
}
|
|
|
|
void WaveSimulation::build_A()
|
|
{
|
|
// Create the diagonal
|
|
arma::cx_vec diagonal(this->N * this->N);
|
|
|
|
// Set diagonal values
|
|
std::complex<double> r = (1._i * this->dt) / (2 * h * h);
|
|
for (size_t i = 0; i < diagonal.size(); i++) {
|
|
diagonal(i) = 1. + 4. * r + (1._i * this->dt / 2.) * this->V(i);
|
|
}
|
|
|
|
// Create the submatrix
|
|
arma::cx_mat sub_matrix(this->N, this->N, arma::fill::zeros);
|
|
sub_matrix.diag(-1).fill(-r);
|
|
sub_matrix.diag(1).fill(-r);
|
|
|
|
// Set the size of A
|
|
this->A.set_size(this->N * this->N, this->N * this->N);
|
|
|
|
// Fill in the values in the submatrix diagonal
|
|
for (size_t i = 0; i < this->A.n_cols; i += this->N) {
|
|
this->A.submat(i, i, i + this->N - 1, i + this->N - 1) = sub_matrix;
|
|
}
|
|
|
|
// Fill the last sub/sup-diagonals
|
|
this->A.diag() = diagonal;
|
|
this->A.diag(-this->N).fill(-r);
|
|
this->A.diag(this->N).fill(-r);
|
|
}
|
|
|
|
void WaveSimulation::build_B()
|
|
{
|
|
std::complex<double> r = (1._i * this->dt) / (2 * h * h);
|
|
|
|
// Create the diagonal
|
|
arma::cx_vec diagonal(this->N * this->N);
|
|
|
|
for (size_t i = 0; i < diagonal.size(); i++) {
|
|
diagonal(i) = 1. - 4. * r - (1._i * this->dt / 2.) * this->V(i);
|
|
}
|
|
|
|
// Create the submatrix
|
|
arma::cx_mat sub_matrix(this->N, this->N, arma::fill::zeros);
|
|
sub_matrix.diag(-1).fill(r);
|
|
sub_matrix.diag(1).fill(r);
|
|
|
|
// Set the size of B
|
|
this->B.set_size(this->N * this->N, this->N * this->N);
|
|
|
|
// Fill in the values in the submatrix diagonal
|
|
for (size_t i = 0; i < this->B.n_cols; i += this->N) {
|
|
this->B.submat(i, i, i + this->N - 1, i + this->N - 1) = sub_matrix;
|
|
}
|
|
|
|
// Fill the last sub/sup-diagonals
|
|
this->B.diag() = diagonal;
|
|
this->B.diag(-this->N).fill(r);
|
|
this->B.diag(this->N).fill(r);
|
|
}
|
|
|
|
void WaveSimulation::initialize_U(double x_c, double y_c, double sigma_x,
|
|
double sigma_y, double p_x, double p_y)
|
|
{
|
|
this->U.set_size(this->N, this->N);
|
|
double x, y, diff_x, diff_y;
|
|
std::complex<double> sum = 0.;
|
|
for (size_t j = 0; j < this->U.n_cols; j++) {
|
|
x = j * h;
|
|
diff_x = x - x_c;
|
|
for (size_t i = 0; i < this->U.n_rows; i++) {
|
|
y = i * h;
|
|
diff_y = y - y_c;
|
|
this->U(i, j) =
|
|
std::exp(-(diff_x * diff_x) / (2. * sigma_x * sigma_x)
|
|
- (diff_y * diff_y) / (2. * sigma_y * sigma_y)
|
|
+ p_x * x * 1._i + p_y * y * 1._i);
|
|
sum += this->U(i, j) * std::conj(this->U(i, j));
|
|
}
|
|
}
|
|
if (std::abs(sum.imag()) > 1e-7) {
|
|
abort();
|
|
}
|
|
double norm = 1. / std::sqrt(sum.real());
|
|
|
|
this->U.for_each([norm](std::complex<double> &el) { el *= norm; });
|
|
}
|
|
|
|
void WaveSimulation::write_U(std::ofstream &ofile)
|
|
{
|
|
this->U.for_each(
|
|
[&ofile](std::complex<double> el) { ofile << el << '\t'; });
|
|
ofile << '\n';
|
|
}
|
|
|
|
void WaveSimulation::build_V(double thickness, double pos_x,
|
|
double aperture_separation, double aperture,
|
|
uint32_t slits)
|
|
{
|
|
this->V.set_size(this->N, this->N);
|
|
this->V.fill(0.);
|
|
if (slits == 0) {
|
|
return;
|
|
}
|
|
arma::cx_vec res;
|
|
|
|
if (slits % 2 == 0) {
|
|
res = arma::cx_vec(aperture_separation / this->h,
|
|
arma::fill::value(1e10));
|
|
for (size_t i = 0; i < slits; i += 2) {
|
|
res = arma::join_cols(
|
|
res, arma::cx_vec(aperture / this->h, arma::fill::zeros));
|
|
res = arma::join_cols(
|
|
arma::cx_vec(aperture / this->h, arma::fill::zeros), res);
|
|
res =
|
|
arma::join_cols(res, arma::cx_vec(aperture_separation / this->h,
|
|
arma::fill::value(1e10)));
|
|
res = arma::join_cols(arma::cx_vec(aperture_separation / this->h,
|
|
arma::fill::value(1e10)),
|
|
res);
|
|
}
|
|
}
|
|
else {
|
|
res = arma::cx_vec(aperture / this->h, arma::fill::value(0));
|
|
for (size_t i = 0; i < slits - 1; i += 2) {
|
|
res =
|
|
arma::join_cols(res, arma::cx_vec(aperture_separation / this->h,
|
|
arma::fill::value(1e10)));
|
|
res = arma::join_cols(arma::cx_vec(aperture_separation / this->h,
|
|
arma::fill::value(1e10)),
|
|
res);
|
|
res = arma::join_cols(
|
|
res, arma::cx_vec(aperture / this->h, arma::fill::zeros));
|
|
res = arma::join_cols(
|
|
arma::cx_vec(aperture / this->h, arma::fill::zeros), res);
|
|
}
|
|
}
|
|
if (res.size() > this->N) {
|
|
abort();
|
|
}
|
|
uint32_t fill = (this->N - res.size()) / 2;
|
|
res = arma::join_cols(arma::cx_vec(fill, arma::fill::value(1e10)), res);
|
|
res = arma::join_cols(res, arma::cx_vec(fill + ((this->N - res.size()) % 2),
|
|
arma::fill::value(1e10)));
|
|
|
|
uint32_t start = pos_x / this->h - thickness / this->h / 2;
|
|
for (size_t i = 0; i < thickness / this->h; i++) {
|
|
this->V.col(start + i) = res;
|
|
}
|
|
}
|