| 1 | /* | 
|---|
| 2 | *    vmg - a versatile multigrid solver | 
|---|
| 3 | *    Copyright (C) 2012 Institute for Numerical Simulation, University of Bonn | 
|---|
| 4 | * | 
|---|
| 5 | *  vmg is free software: you can redistribute it and/or modify | 
|---|
| 6 | *  it under the terms of the GNU General Public License as published by | 
|---|
| 7 | *  the Free Software Foundation, either version 3 of the License, or | 
|---|
| 8 | *  (at your option) any later version. | 
|---|
| 9 | * | 
|---|
| 10 | *  vmg is distributed in the hope that it will be useful, | 
|---|
| 11 | *  but WITHOUT ANY WARRANTY; without even the implied warranty of | 
|---|
| 12 | *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the | 
|---|
| 13 | *  GNU General Public License for more details. | 
|---|
| 14 | * | 
|---|
| 15 | *  You should have received a copy of the GNU General Public License | 
|---|
| 16 | *  along with this program.  If not, see <http://www.gnu.org/licenses/>. | 
|---|
| 17 | */ | 
|---|
| 18 |  | 
|---|
| 19 | #ifdef HAVE_CONFIG_H | 
|---|
| 20 | #include <libvmg_config.h> | 
|---|
| 21 | #endif | 
|---|
| 22 |  | 
|---|
| 23 | #include <vector> | 
|---|
| 24 |  | 
|---|
| 25 | #include "base/helper.hpp" | 
|---|
| 26 | #include "base/index.hpp" | 
|---|
| 27 | #include "base/vector.hpp" | 
|---|
| 28 | #include "grid/grid.hpp" | 
|---|
| 29 | #include "units/particle/interpolation.hpp" | 
|---|
| 30 | #include "units/particle/particle.hpp" | 
|---|
| 31 |  | 
|---|
| 32 | #include "mg.hpp" | 
|---|
| 33 | #include "comm/comm.hpp" | 
|---|
| 34 |  | 
|---|
| 35 | using namespace VMG; | 
|---|
| 36 |  | 
|---|
| 37 | Particle::Interpolation::Interpolation(const int& degree) : | 
|---|
| 38 | deg(degree), | 
|---|
| 39 | deg_1(degree+1), | 
|---|
| 40 | buffer(degree+1), | 
|---|
| 41 | buffer_diff(degree) | 
|---|
| 42 | { | 
|---|
| 43 | coeff = new vmg_float[Helper::intpow(deg_1, 3)]; | 
|---|
| 44 | coeff_buffer = new vmg_float[deg_1]; | 
|---|
| 45 | for (int i=0; i<degree; ++i) | 
|---|
| 46 | buffer_diff[i].resize(i+1); | 
|---|
| 47 | } | 
|---|
| 48 |  | 
|---|
| 49 | Particle::Interpolation::~Interpolation() | 
|---|
| 50 | { | 
|---|
| 51 | delete [] coeff; | 
|---|
| 52 | delete [] coeff_buffer; | 
|---|
| 53 | } | 
|---|
| 54 |  | 
|---|
| 55 | void Particle::Interpolation::ComputeCoefficients(const Grid& grid, const Index& index) | 
|---|
| 56 | { | 
|---|
| 57 | Index i; | 
|---|
| 58 |  | 
|---|
| 59 | const Index begin = index - deg/2; | 
|---|
| 60 |  | 
|---|
| 61 | h = grid.Extent().MeshWidth(); | 
|---|
| 62 |  | 
|---|
| 63 | for (i[0]=0; i[0]<deg_1; ++i[0]) | 
|---|
| 64 | for (i[1]=0; i[1]<deg_1; ++i[1]) | 
|---|
| 65 | for (i[2]=0; i[2]<deg_1; ++i[2]) | 
|---|
| 66 | _access_coeff(i) = grid.GetVal(i+begin); | 
|---|
| 67 |  | 
|---|
| 68 | pos_begin = grid.Extent().Begin() | 
|---|
| 69 | + (begin - grid.Local().Begin() + grid.Global().LocalBegin() - grid.Global().GlobalBegin()) * grid.Extent().MeshWidth(); | 
|---|
| 70 |  | 
|---|
| 71 | // compute coefficients x-direction | 
|---|
| 72 | for (i=0; i[1]<deg_1; ++i[1]) | 
|---|
| 73 | for (i[2]=0; i[2]<deg_1; ++i[2]) | 
|---|
| 74 | _compute_coefficients_1d(i, 0); | 
|---|
| 75 |  | 
|---|
| 76 | // compute coefficients y-direction | 
|---|
| 77 | for (i=0; i[0]<deg_1; ++i[0]) | 
|---|
| 78 | for (i[2]=0; i[2]<deg_1; ++i[2]) | 
|---|
| 79 | _compute_coefficients_1d(i, 1); | 
|---|
| 80 |  | 
|---|
| 81 | // compute coefficients z-direction | 
|---|
| 82 | for (i=0; i[0]<deg_1; ++i[0]) | 
|---|
| 83 | for (i[1]=0; i[1]<deg_1; ++i[1]) | 
|---|
| 84 | _compute_coefficients_1d(i, 2); | 
|---|
| 85 | } | 
|---|
| 86 |  | 
|---|
| 87 | void Particle::Interpolation::_compute_coefficients_1d(const Index& index, const unsigned int& direction) | 
|---|
| 88 | { | 
|---|
| 89 | vmg_float power = 1.0; | 
|---|
| 90 | unsigned long faculty = 1; | 
|---|
| 91 | int c; | 
|---|
| 92 | Index i; | 
|---|
| 93 |  | 
|---|
| 94 | for (i=index, c=0; c<deg_1; ++i[direction], ++c) | 
|---|
| 95 | coeff_buffer[c] = _access_coeff(i); | 
|---|
| 96 |  | 
|---|
| 97 | i=index; | 
|---|
| 98 | ++i[direction]; | 
|---|
| 99 | for (c=1; c<deg_1; ++i[direction], ++c) { | 
|---|
| 100 | for (int j=0; j<deg_1-c; ++j) | 
|---|
| 101 | coeff_buffer[j] = coeff_buffer[j+1] - coeff_buffer[j]; | 
|---|
| 102 | faculty *= c; | 
|---|
| 103 | power *= h[direction]; | 
|---|
| 104 | _access_coeff(i) = coeff_buffer[0] / (faculty*power); | 
|---|
| 105 | } | 
|---|
| 106 | } | 
|---|
| 107 |  | 
|---|
| 108 | void Particle::Interpolation::Evaluate(Particle& p) | 
|---|
| 109 | { | 
|---|
| 110 | const Vector& pos = p.Pos(); | 
|---|
| 111 | vmg_float& pot = p.Pot(); | 
|---|
| 112 | Vector& field = p.Field(); | 
|---|
| 113 |  | 
|---|
| 114 | pot = 0.0; | 
|---|
| 115 | field = 0.0; | 
|---|
| 116 |  | 
|---|
| 117 | Vector offset = pos - pos_begin; | 
|---|
| 118 | buffer[0] = 1.0; | 
|---|
| 119 | for (int i=0; i<deg; ++i) { | 
|---|
| 120 | buffer[i+1] = buffer[i] * offset; | 
|---|
| 121 | for (int j=0; j<i; ++j) | 
|---|
| 122 | buffer_diff[i][j] = buffer_diff[i-1][j] * offset; | 
|---|
| 123 | buffer_diff[i][i] = buffer[i]; | 
|---|
| 124 | offset -= h; | 
|---|
| 125 | } | 
|---|
| 126 |  | 
|---|
| 127 | for (int i=1; i<deg; ++i) | 
|---|
| 128 | for (int j=1; j<=i; ++j) | 
|---|
| 129 | buffer_diff[i][0] += buffer_diff[i][j]; | 
|---|
| 130 |  | 
|---|
| 131 | for (int i=0; i<deg_1; ++i) | 
|---|
| 132 | for (int j=0; j<deg_1; ++j) | 
|---|
| 133 | for (int k=0; k<deg_1; ++k) | 
|---|
| 134 | pot += _access_coeff(i,j,k) * buffer[i][0] * buffer[j][1] * buffer[k][2]; | 
|---|
| 135 |  | 
|---|
| 136 | for (int i=0; i<deg_1; ++i) | 
|---|
| 137 | for (int j=0; j<deg_1; ++j) | 
|---|
| 138 | for (int k=0; k<deg; ++k) { | 
|---|
| 139 | field[0] -= _access_coeff(k+1, i, j) * buffer[i][1] * buffer[j][2] * buffer_diff[k][0][0]; | 
|---|
| 140 | field[1] -= _access_coeff(i, k+1, j) * buffer[i][0] * buffer[j][2] * buffer_diff[k][0][1]; | 
|---|
| 141 | field[2] -= _access_coeff(i, j, k+1) * buffer[i][0] * buffer[j][1] * buffer_diff[k][0][2]; | 
|---|
| 142 | } | 
|---|
| 143 | } | 
|---|
| 144 |  | 
|---|
| 145 | vmg_float Particle::Interpolation::EvaluatePotentialLR(const Particle& p) | 
|---|
| 146 | { | 
|---|
| 147 | vmg_float result = 0.0; | 
|---|
| 148 | Vector prod, offset; | 
|---|
| 149 | Index i; | 
|---|
| 150 |  | 
|---|
| 151 | const Vector& pos = p.Pos(); | 
|---|
| 152 |  | 
|---|
| 153 | prod[0] = 1.0; | 
|---|
| 154 | offset[0] = pos[0] - pos_begin[0]; | 
|---|
| 155 | for (i[0]=0; i[0]<deg_1; ++i[0]) { | 
|---|
| 156 | prod[1] = prod[0]; | 
|---|
| 157 | offset[1] = pos[1] - pos_begin[1]; | 
|---|
| 158 | for (i[1]=0; i[1]<deg_1; ++i[1]) { | 
|---|
| 159 | prod[2] = prod[1]; | 
|---|
| 160 | offset[2] = pos[2] - pos_begin[2]; | 
|---|
| 161 | for (i[2]=0; i[2]<deg_1; ++i[2]) { | 
|---|
| 162 | result += _access_coeff(i) * prod[2]; | 
|---|
| 163 | prod[2] *= offset[2]; | 
|---|
| 164 | offset[2] -= h[2]; | 
|---|
| 165 | } | 
|---|
| 166 | prod[1] *= offset[1]; | 
|---|
| 167 | offset[1] -= h[1]; | 
|---|
| 168 | } | 
|---|
| 169 | prod[0] *= offset[0]; | 
|---|
| 170 | offset[0] -= h[0]; | 
|---|
| 171 | } | 
|---|
| 172 |  | 
|---|
| 173 | return result; | 
|---|
| 174 | } | 
|---|