1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
|
// Gmsh - Copyright (C) 1997-2020 C. Geuzaine, J.-F. Remacle
//
// See the LICENSE.txt file for license information. Please report all
// issues on https://gitlab.onelab.info/gmsh/gmsh/issues.
#include "eigenSolver.h"
#include "OS.h"
#if defined(HAVE_SLEPC)
#include <slepceps.h>
#if SLEPC_VERSION_RELEASE != 0 && \
(SLEPC_VERSION_MAJOR < 3 || \
(SLEPC_VERSION_MAJOR == 3 && SLEPC_VERSION_MINOR < 2))
#define EPSDestroy(e) EPSDestroy(*(e))
#endif
void eigenSolver::_check(int ierr) const { CHKERRABORT(PETSC_COMM_WORLD, ierr); }
eigenSolver::eigenSolver(dofManager<double> *manager, std::string A,
std::string B, bool hermitian)
: _sysA(0), _sysB(0), _hermitian(hermitian)
{
if(A.size()) {
_sysA = dynamic_cast<linearSystemPETSc<double> *>(manager->getLinearSystem(A));
if(!_sysA) Msg::Error("Could not find PETSc system '%s'", A.c_str());
}
if(B.size()) {
_sysB = dynamic_cast<linearSystemPETSc<double> *>(manager->getLinearSystem(B));
if(!_sysB) Msg::Error("Could not find PETSc system '%s'", B.c_str());
}
}
eigenSolver::eigenSolver(linearSystemPETSc<double> *A,
linearSystemPETSc<double> *B, bool hermitian)
: _sysA(A), _sysB(B), _hermitian(hermitian)
{
}
bool eigenSolver::solve(int numEigenValues, std::string which,
std::string method, double tolVal, int iterMax)
{
if(!_sysA) return false;
Mat A = _sysA->getMatrix();
Mat B = _sysB ? _sysB->getMatrix() : PETSC_NULL;
PetscInt N, M;
_check(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
_check(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));
_check(MatGetSize(A, &N, &M));
PetscInt N2, M2;
if(_sysB) {
_check(MatAssemblyBegin(B, MAT_FINAL_ASSEMBLY));
_check(MatAssemblyEnd(B, MAT_FINAL_ASSEMBLY));
_check(MatGetSize(B, &N2, &M2));
}
// generalized eigenvalue problem A x - \lambda B x = 0
EPS eps;
_check(EPSCreate(PETSC_COMM_WORLD, &eps));
_check(EPSSetOperators(eps, A, B));
if(_hermitian)
_check(EPSSetProblemType(eps, _sysB ? EPS_GHEP : EPS_HEP));
else
_check(EPSSetProblemType(eps, _sysB ? EPS_GNHEP : EPS_NHEP));
// set some default options
_check(EPSSetDimensions(eps, numEigenValues, PETSC_DECIDE, PETSC_DECIDE));
_check(EPSSetTolerances(eps, tolVal, iterMax));
if(method == "krylovschur")
_check(EPSSetType(eps, EPSKRYLOVSCHUR));
else if(method == "arnoldi")
_check(EPSSetType(eps, EPSARNOLDI));
else if(method == "arpack")
_check(EPSSetType(eps, EPSARPACK));
else if(method == "power")
_check(EPSSetType(eps, EPSPOWER));
else{
Msg::Error("eigenSolver: method '%s' not available", method.c_str());
_check(EPSSetType(eps, EPSKRYLOVSCHUR));
}
// override these options at runtime, petsc-style
_check(EPSSetFromOptions(eps));
// force options specified directly as arguments
if(numEigenValues)
_check(EPSSetDimensions(eps, numEigenValues, PETSC_DECIDE, PETSC_DECIDE));
if(which == "smallest")
_check(EPSSetWhichEigenpairs(eps, EPS_SMALLEST_MAGNITUDE));
else if(which == "smallestReal")
_check(EPSSetWhichEigenpairs(eps, EPS_SMALLEST_REAL));
else if(which == "largest")
_check(EPSSetWhichEigenpairs(eps, EPS_LARGEST_MAGNITUDE));
// print info
#if(SLEPC_VERSION_RELEASE == 0 || \
(SLEPC_VERSION_MAJOR > 3 || \
(SLEPC_VERSION_MAJOR == 3 && SLEPC_VERSION_MINOR >= 4)))
EPSType type;
#else
const EPSType type;
#endif
_check(EPSGetType(eps, &type));
Msg::Debug("SLEPc solution method: %s", type);
PetscInt nev;
_check(EPSGetDimensions(eps, &nev, PETSC_NULL, PETSC_NULL));
Msg::Debug("SLEPc number of requested eigenvalues: %d", nev);
PetscReal tol;
PetscInt maxit;
_check(EPSGetTolerances(eps, &tol, &maxit));
Msg::Debug("SLEPc stopping condition: tol=%g, maxit=%d", tol, maxit);
// solve
Msg::Info("SLEPc solving...");
double t1 = Cpu(), w1 = TimeOfDay();
_check(EPSSolve(eps));
// check convergence
int its;
_check(EPSGetIterationNumber(eps, &its));
EPSConvergedReason reason;
_check(EPSGetConvergedReason(eps, &reason));
if(reason == EPS_CONVERGED_TOL) {
double t2 = Cpu(), w2 = TimeOfDay();
Msg::Debug("SLEPc converged in %d iterations (Wall %gs, CPU %gs)", its,
w2 - w1, t2 - t1);
}
else if(reason == EPS_DIVERGED_ITS)
Msg::Error("SLEPc diverged after %d iterations", its);
else if(reason == EPS_DIVERGED_BREAKDOWN)
Msg::Error("SLEPc generic breakdown in method");
#if(SLEPC_VERSION_MAJOR < 3 || \
(SLEPC_VERSION_MAJOR == 3 && SLEPC_VERSION_MINOR < 2))
else if(reason == EPS_DIVERGED_NONSYMMETRIC)
Msg::Error("The operator is nonsymmetric");
#endif
// get number of converged approximate eigenpairs
PetscInt nconv;
_check(EPSGetConverged(eps, &nconv));
Msg::Debug("SLEPc number of converged eigenpairs: %d", nconv);
// ignore additional eigenvalues if we get more than what we asked
if(nconv > nev) nconv = nev;
if(nconv > 0) {
Vec xr, xi;
#if(PETSC_VERSION_MAJOR == 3) && (PETSC_VERSION_MINOR < 6)
_check(MatGetVecs(A, PETSC_NULL, &xr));
_check(MatGetVecs(A, PETSC_NULL, &xi));
#else
_check(MatCreateVecs(A, PETSC_NULL, &xr));
_check(MatCreateVecs(A, PETSC_NULL, &xi));
#endif
Msg::Debug(" Re[EigenValue] Im[EigenValue]"
" Relative error");
for(int i = 0; i < nconv; i++) {
PetscScalar kr, ki;
_check(EPSGetEigenpair(eps, i, &kr, &ki, xr, xi));
PetscReal error;
#if(PETSC_VERSION_MAJOR == 3) && (PETSC_VERSION_MINOR < 6)
_check(EPSComputeRelativeError(eps, i, &error));
#else
_check(EPSComputeError(eps, i, EPS_ERROR_RELATIVE, &error));
#endif
#if defined(PETSC_USE_COMPLEX)
PetscReal re = PetscRealPart(kr);
PetscReal im = PetscImaginaryPart(kr);
#else
PetscReal re = kr;
PetscReal im = ki;
#endif
Msg::Debug("EIG %03d %s%.16e %s%.16e %3.6e", i, (re < 0) ? "" : " ", re,
(im < 0) ? "" : " ", im, error);
// store eigenvalues and eigenvectors
_eigenValues.push_back(std::complex<double>(re, im));
PetscScalar *tmpr, *tmpi;
_check(VecGetArray(xr, &tmpr));
_check(VecGetArray(xi, &tmpi));
std::vector<std::complex<double> > ev(N);
for(int i = 0; i < N; i++) {
#if defined(PETSC_USE_COMPLEX)
ev[i] = tmpr[i];
#else
ev[i] = std::complex<double>(tmpr[i], tmpi[i]);
#endif
}
_eigenVectors.push_back(ev);
}
_check(VecDestroy(&xr));
_check(VecDestroy(&xi));
}
_check(EPSDestroy(&eps));
if(reason == EPS_CONVERGED_TOL) {
Msg::Debug("SLEPc done");
return true;
}
else {
Msg::Warning("SLEPc failed");
return false;
}
}
void eigenSolver::normalize_mode(std::vector<int> modeView, double scale)
{
Msg::Info("Normalize all eigenvectors");
for(std::size_t imode = 0; imode < modeView.size(); imode++) {
int i = modeView[imode];
double norm = 0.;
for(std::size_t j = 0; j < _eigenVectors[i].size(); j++) {
std::complex<double> val = _eigenVectors[i][j];
double normval = std::abs(val);
if(normval > norm) norm = normval;
}
if(norm == 0) {
Msg::Error("zero eigenvector");
return;
}
for(std::size_t j = 0; j < _eigenVectors[i].size(); j++) {
_eigenVectors[i][j] *= (scale / norm);
}
}
}
#endif
|
