Actual source code: sinvert.c
slepc-3.17.0 2022-03-31
1: /*
2: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
3: SLEPc - Scalable Library for Eigenvalue Problem Computations
4: Copyright (c) 2002-, Universitat Politecnica de Valencia, Spain
6: This file is part of SLEPc.
7: SLEPc is distributed under a 2-clause BSD license (see LICENSE).
8: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
9: */
10: /*
11: Implements the shift-and-invert technique for eigenvalue problems
12: */
14: #include <slepc/private/stimpl.h>
16: PetscErrorCode STBackTransform_Sinvert(ST st,PetscInt n,PetscScalar *eigr,PetscScalar *eigi)
17: {
18: PetscInt j;
19: #if !defined(PETSC_USE_COMPLEX)
20: PetscScalar t;
21: #endif
23: #if !defined(PETSC_USE_COMPLEX)
24: for (j=0;j<n;j++) {
25: if (eigi[j] == 0) eigr[j] = 1.0 / eigr[j] + st->sigma;
26: else {
27: t = eigr[j] * eigr[j] + eigi[j] * eigi[j];
28: eigr[j] = eigr[j] / t + st->sigma;
29: eigi[j] = - eigi[j] / t;
30: }
31: }
32: #else
33: for (j=0;j<n;j++) {
34: eigr[j] = 1.0 / eigr[j] + st->sigma;
35: }
36: #endif
37: PetscFunctionReturn(0);
38: }
40: PetscErrorCode STPostSolve_Sinvert(ST st)
41: {
42: if (st->matmode == ST_MATMODE_INPLACE) {
43: if (st->nmat>1) MatAXPY(st->A[0],st->sigma,st->A[1],st->str);
44: else MatShift(st->A[0],st->sigma);
45: st->Astate[0] = ((PetscObject)st->A[0])->state;
46: st->state = ST_STATE_INITIAL;
47: st->opready = PETSC_FALSE;
48: }
49: PetscFunctionReturn(0);
50: }
52: /*
53: Operator (sinvert):
54: Op P M
55: if nmat=1: (A-sI)^-1 A-sI NULL
56: if nmat=2: (A-sB)^-1 B A-sB B
57: */
58: PetscErrorCode STComputeOperator_Sinvert(ST st)
59: {
60: /* if the user did not set the shift, use the target value */
61: if (!st->sigma_set) st->sigma = st->defsigma;
62: PetscObjectReference((PetscObject)st->A[1]);
63: MatDestroy(&st->T[0]);
64: st->T[0] = st->A[1];
65: STMatMAXPY_Private(st,-st->sigma,0.0,0,NULL,PetscNot(st->state==ST_STATE_UPDATED),PETSC_FALSE,&st->T[1]);
66: PetscObjectReference((PetscObject)st->T[1]);
67: MatDestroy(&st->P);
68: st->P = st->T[1];
69: st->M = (st->nmat>1)? st->T[0]: NULL;
70: if (st->Psplit) { /* build custom preconditioner from the split matrices */
71: STMatMAXPY_Private(st,-st->sigma,0.0,0,NULL,PETSC_TRUE,PETSC_TRUE,&st->Pmat);
72: }
73: PetscFunctionReturn(0);
74: }
76: PetscErrorCode STSetUp_Sinvert(ST st)
77: {
78: PetscInt k,nc,nmat=st->nmat;
79: PetscScalar *coeffs=NULL;
81: if (nmat>1) STSetWorkVecs(st,1);
82: /* if the user did not set the shift, use the target value */
83: if (!st->sigma_set) st->sigma = st->defsigma;
84: if (nmat>2) { /* set-up matrices for polynomial eigenproblems */
85: if (st->transform) {
86: nc = (nmat*(nmat+1))/2;
87: PetscMalloc1(nc,&coeffs);
88: /* Compute coeffs */
89: STCoeffs_Monomial(st,coeffs);
90: /* T[0] = A_n */
91: k = nmat-1;
92: PetscObjectReference((PetscObject)st->A[k]);
93: MatDestroy(&st->T[0]);
94: st->T[0] = st->A[k];
95: for (k=1;k<nmat;k++) STMatMAXPY_Private(st,nmat>2?st->sigma:-st->sigma,0.0,nmat-k-1,coeffs?coeffs+(k*(k+1))/2:NULL,PetscNot(st->state==ST_STATE_UPDATED),PETSC_FALSE,&st->T[k]);
96: PetscFree(coeffs);
97: PetscObjectReference((PetscObject)st->T[nmat-1]);
98: MatDestroy(&st->P);
99: st->P = st->T[nmat-1];
100: if (st->Psplit) { /* build custom preconditioner from the split matrices */
101: STMatMAXPY_Private(st,st->sigma,0.0,0,coeffs?coeffs+((nmat-1)*nmat)/2:NULL,PETSC_TRUE,PETSC_TRUE,&st->Pmat);
102: }
103: ST_KSPSetOperators(st,st->P,st->Pmat?st->Pmat:st->P);
104: } else {
105: for (k=0;k<nmat;k++) {
106: PetscObjectReference((PetscObject)st->A[k]);
107: MatDestroy(&st->T[k]);
108: st->T[k] = st->A[k];
109: }
110: }
111: }
112: if (st->P) KSPSetUp(st->ksp);
113: PetscFunctionReturn(0);
114: }
116: PetscErrorCode STSetShift_Sinvert(ST st,PetscScalar newshift)
117: {
118: PetscInt nmat=PetscMax(st->nmat,2),k,nc;
119: PetscScalar *coeffs=NULL;
121: if (st->transform) {
122: if (st->matmode == ST_MATMODE_COPY && nmat>2) {
123: nc = (nmat*(nmat+1))/2;
124: PetscMalloc1(nc,&coeffs);
125: /* Compute coeffs */
126: STCoeffs_Monomial(st,coeffs);
127: }
128: for (k=1;k<nmat;k++) STMatMAXPY_Private(st,nmat>2?newshift:-newshift,nmat>2?st->sigma:-st->sigma,nmat-k-1,coeffs?coeffs+(k*(k+1))/2:NULL,PETSC_FALSE,PETSC_FALSE,&st->T[k]);
129: if (st->matmode == ST_MATMODE_COPY && nmat>2) PetscFree(coeffs);
130: if (st->P!=st->T[nmat-1]) {
131: PetscObjectReference((PetscObject)st->T[nmat-1]);
132: MatDestroy(&st->P);
133: st->P = st->T[nmat-1];
134: }
135: if (st->Psplit) { /* build custom preconditioner from the split matrices */
136: STMatMAXPY_Private(st,nmat>2?newshift:-newshift,nmat>2?st->sigma:-st->sigma,0,coeffs?coeffs+((nmat-1)*nmat)/2:NULL,PETSC_FALSE,PETSC_TRUE,&st->Pmat);
137: }
138: }
139: if (st->P) {
140: ST_KSPSetOperators(st,st->P,st->Pmat?st->Pmat:st->P);
141: KSPSetUp(st->ksp);
142: }
143: PetscFunctionReturn(0);
144: }
146: SLEPC_EXTERN PetscErrorCode STCreate_Sinvert(ST st)
147: {
148: st->usesksp = PETSC_TRUE;
150: st->ops->apply = STApply_Generic;
151: st->ops->applytrans = STApplyTranspose_Generic;
152: st->ops->backtransform = STBackTransform_Sinvert;
153: st->ops->setshift = STSetShift_Sinvert;
154: st->ops->getbilinearform = STGetBilinearForm_Default;
155: st->ops->setup = STSetUp_Sinvert;
156: st->ops->computeoperator = STComputeOperator_Sinvert;
157: st->ops->postsolve = STPostSolve_Sinvert;
158: st->ops->checknullspace = STCheckNullSpace_Default;
159: st->ops->setdefaultksp = STSetDefaultKSP_Default;
160: PetscFunctionReturn(0);
161: }