Actual source code: nepsetup.c

slepc-3.12.2 2020-01-13
Report Typos and Errors
  1: /*
  2:    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
  3:    SLEPc - Scalable Library for Eigenvalue Problem Computations
  4:    Copyright (c) 2002-2019, 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:    NEP routines related to problem setup
 12: */

 14: #include <slepc/private/nepimpl.h>       /*I "slepcnep.h" I*/

 16: /*@
 17:    NEPSetUp - Sets up all the internal data structures necessary for the
 18:    execution of the NEP solver.

 20:    Collective on nep

 22:    Input Parameter:
 23: .  nep   - solver context

 25:    Notes:
 26:    This function need not be called explicitly in most cases, since NEPSolve()
 27:    calls it. It can be useful when one wants to measure the set-up time
 28:    separately from the solve time.

 30:    Level: developer

 32: .seealso: NEPCreate(), NEPSolve(), NEPDestroy()
 33: @*/
 34: PetscErrorCode NEPSetUp(NEP nep)
 35: {
 37:   PetscInt       k;
 38:   SlepcSC        sc;
 39:   Mat            T;
 40:   PetscBool      flg;
 41:   KSP            ksp;
 42:   PC             pc;
 43:   PetscMPIInt    size;
 44:   MatSolverType  stype;

 48:   NEPCheckProblem(nep,1);
 49:   if (nep->state) return(0);
 50:   PetscLogEventBegin(NEP_SetUp,nep,0,0,0);

 52:   /* reset the convergence flag from the previous solves */
 53:   nep->reason = NEP_CONVERGED_ITERATING;

 55:   /* set default solver type (NEPSetFromOptions was not called) */
 56:   if (!((PetscObject)nep)->type_name) {
 57:     NEPSetType(nep,NEPRII);
 58:   }
 59:   if (nep->useds && !nep->ds) { NEPGetDS(nep,&nep->ds); }
 60:   if (!nep->rg) { NEPGetRG(nep,&nep->rg); }
 61:   if (!((PetscObject)nep->rg)->type_name) {
 62:     RGSetType(nep->rg,RGINTERVAL);
 63:   }
 64:   if (nep->twosided && !nep->hasts) SETERRQ(PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"This solver does not support computing left eigenvectors (no two-sided variant)");

 66:   /* set problem dimensions */
 67:   switch (nep->fui) {
 68:   case NEP_USER_INTERFACE_CALLBACK:
 69:     NEPGetFunction(nep,&T,NULL,NULL,NULL);
 70:     MatGetSize(T,&nep->n,NULL);
 71:     MatGetLocalSize(T,&nep->nloc,NULL);
 72:     break;
 73:   case NEP_USER_INTERFACE_SPLIT:
 74:     MatDuplicate(nep->A[0],MAT_DO_NOT_COPY_VALUES,&nep->function);
 75:     MatDuplicate(nep->A[0],MAT_DO_NOT_COPY_VALUES,&nep->jacobian);
 76:     PetscLogObjectParent((PetscObject)nep,(PetscObject)nep->function);
 77:     PetscLogObjectParent((PetscObject)nep,(PetscObject)nep->jacobian);
 78:     MatGetSize(nep->A[0],&nep->n,NULL);
 79:     MatGetLocalSize(nep->A[0],&nep->nloc,NULL);
 80:     break;
 81:   }

 83:   /* set default problem type */
 84:   if (!nep->problem_type) {
 85:     NEPSetProblemType(nep,NEP_GENERAL);
 86:   }

 88:   /* check consistency of refinement options */
 89:   if (nep->refine) {
 90:     if (nep->fui!=NEP_USER_INTERFACE_SPLIT) SETERRQ(PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"Iterative refinement only implemented in split form");
 91:     if (!nep->scheme) {  /* set default scheme */
 92:       NEPRefineGetKSP(nep,&ksp);
 93:       KSPGetPC(ksp,&pc);
 94:       PetscObjectTypeCompare((PetscObject)ksp,KSPPREONLY,&flg);
 95:       if (flg) {
 96:         PetscObjectTypeCompareAny((PetscObject)pc,&flg,PCLU,PCCHOLESKY,"");
 97:       }
 98:       nep->scheme = flg? NEP_REFINE_SCHEME_MBE: NEP_REFINE_SCHEME_SCHUR;
 99:     }
100:     if (nep->scheme==NEP_REFINE_SCHEME_MBE) {
101:       NEPRefineGetKSP(nep,&ksp);
102:       KSPGetPC(ksp,&pc);
103:       PetscObjectTypeCompare((PetscObject)ksp,KSPPREONLY,&flg);
104:       if (flg) {
105:         PetscObjectTypeCompareAny((PetscObject)pc,&flg,PCLU,PCCHOLESKY,"");
106:       }
107:       if (!flg) SETERRQ(PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"The MBE scheme for refinement requires a direct solver in KSP");
108:       MPI_Comm_size(PetscObjectComm((PetscObject)pc),&size);
109:       if (size>1) {   /* currently selected PC is a factorization */
110:         PCFactorGetMatSolverType(pc,&stype);
111:         PetscStrcmp(stype,MATSOLVERPETSC,&flg);
112:         if (flg) SETERRQ(PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"For Newton refinement, you chose to solve linear systems with a factorization, but in parallel runs you need to select an external package");
113:       }
114:     }
115:     if (nep->scheme==NEP_REFINE_SCHEME_SCHUR) {
116:       if (nep->npart>1) SETERRQ(PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"The Schur scheme for refinement does not support subcommunicators");
117:     }
118:   }
119:   /* call specific solver setup */
120:   (*nep->ops->setup)(nep);

122:   /* by default, compute eigenvalues close to target */
123:   /* nep->target should contain the initial guess for the eigenvalue */
124:   if (!nep->which) nep->which = NEP_TARGET_MAGNITUDE;

126:   /* set tolerance if not yet set */
127:   if (nep->tol==PETSC_DEFAULT) nep->tol = SLEPC_DEFAULT_TOL;
128:   if (nep->refine) {
129:     if (nep->rtol==PETSC_DEFAULT) nep->rtol = PetscMax(nep->tol/1000,PETSC_MACHINE_EPSILON);
130:     if (nep->rits==PETSC_DEFAULT) nep->rits = (nep->refine==NEP_REFINE_SIMPLE)? 10: 1;
131:   }

133:   /* fill sorting criterion context */
134:   switch (nep->which) {
135:     case NEP_LARGEST_MAGNITUDE:
136:       nep->sc->comparison    = SlepcCompareLargestMagnitude;
137:       nep->sc->comparisonctx = NULL;
138:       break;
139:     case NEP_SMALLEST_MAGNITUDE:
140:       nep->sc->comparison    = SlepcCompareSmallestMagnitude;
141:       nep->sc->comparisonctx = NULL;
142:       break;
143:     case NEP_LARGEST_REAL:
144:       nep->sc->comparison    = SlepcCompareLargestReal;
145:       nep->sc->comparisonctx = NULL;
146:       break;
147:     case NEP_SMALLEST_REAL:
148:       nep->sc->comparison    = SlepcCompareSmallestReal;
149:       nep->sc->comparisonctx = NULL;
150:       break;
151:     case NEP_LARGEST_IMAGINARY:
152:       nep->sc->comparison    = SlepcCompareLargestImaginary;
153:       nep->sc->comparisonctx = NULL;
154:       break;
155:     case NEP_SMALLEST_IMAGINARY:
156:       nep->sc->comparison    = SlepcCompareSmallestImaginary;
157:       nep->sc->comparisonctx = NULL;
158:       break;
159:     case NEP_TARGET_MAGNITUDE:
160:       nep->sc->comparison    = SlepcCompareTargetMagnitude;
161:       nep->sc->comparisonctx = &nep->target;
162:       break;
163:     case NEP_TARGET_REAL:
164:       nep->sc->comparison    = SlepcCompareTargetReal;
165:       nep->sc->comparisonctx = &nep->target;
166:       break;
167:     case NEP_TARGET_IMAGINARY:
168: #if defined(PETSC_USE_COMPLEX)
169:       nep->sc->comparison    = SlepcCompareTargetImaginary;
170:       nep->sc->comparisonctx = &nep->target;
171: #endif
172:       break;
173:     case NEP_ALL:
174:       nep->sc->comparison    = SlepcCompareSmallestReal;
175:       nep->sc->comparisonctx = NULL;
176:       break;
177:     case NEP_WHICH_USER:
178:       break;
179:   }

181:   nep->sc->map    = NULL;
182:   nep->sc->mapobj = NULL;

184:   /* fill sorting criterion for DS */
185:   if (nep->useds) {
186:     DSGetSlepcSC(nep->ds,&sc);
187:     sc->comparison    = nep->sc->comparison;
188:     sc->comparisonctx = nep->sc->comparisonctx;
189:     PetscObjectTypeCompare((PetscObject)nep,NEPNLEIGS,&flg);
190:     if (!flg) {
191:       sc->map    = NULL;
192:       sc->mapobj = NULL;
193:     }
194:   }
195:   if (nep->nev > nep->ncv) SETERRQ(PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"nev bigger than ncv");

197:   /* process initial vectors */
198:   if (nep->nini<0) {
199:     k = -nep->nini;
200:     if (k>nep->ncv) SETERRQ(PetscObjectComm((PetscObject)nep),1,"The number of initial vectors is larger than ncv");
201:     BVInsertVecs(nep->V,0,&k,nep->IS,PETSC_TRUE);
202:     SlepcBasisDestroy_Private(&nep->nini,&nep->IS);
203:     nep->nini = k;
204:   }
205:   PetscLogEventEnd(NEP_SetUp,nep,0,0,0);
206:   nep->state = NEP_STATE_SETUP;
207:   return(0);
208: }

210: /*@C
211:    NEPSetInitialSpace - Specify a basis of vectors that constitute the initial
212:    space, that is, the subspace from which the solver starts to iterate.

214:    Collective on nep

216:    Input Parameter:
217: +  nep   - the nonlinear eigensolver context
218: .  n     - number of vectors
219: -  is    - set of basis vectors of the initial space

221:    Notes:
222:    Some solvers start to iterate on a single vector (initial vector). In that case,
223:    the other vectors are ignored.

225:    These vectors do not persist from one NEPSolve() call to the other, so the
226:    initial space should be set every time.

228:    The vectors do not need to be mutually orthonormal, since they are explicitly
229:    orthonormalized internally.

231:    Common usage of this function is when the user can provide a rough approximation
232:    of the wanted eigenspace. Then, convergence may be faster.

234:    Level: intermediate
235: @*/
236: PetscErrorCode NEPSetInitialSpace(NEP nep,PetscInt n,Vec is[])
237: {

243:   if (n<0) SETERRQ(PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"Argument n cannot be negative");
244:   if (n>0) {
247:   }
248:   SlepcBasisReference_Private(n,is,&nep->nini,&nep->IS);
249:   if (n>0) nep->state = NEP_STATE_INITIAL;
250:   return(0);
251: }

253: /*
254:   NEPSetDimensions_Default - Set reasonable values for ncv, mpd if not set
255:   by the user. This is called at setup.
256:  */
257: PetscErrorCode NEPSetDimensions_Default(NEP nep,PetscInt nev,PetscInt *ncv,PetscInt *mpd)
258: {
260:   if (*ncv) { /* ncv set */
261:     if (*ncv<nev) SETERRQ(PetscObjectComm((PetscObject)nep),1,"The value of ncv must be at least nev");
262:   } else if (*mpd) { /* mpd set */
263:     *ncv = PetscMin(nep->n,nev+(*mpd));
264:   } else { /* neither set: defaults depend on nev being small or large */
265:     if (nev<500) *ncv = PetscMin(nep->n,PetscMax(2*nev,nev+15));
266:     else {
267:       *mpd = 500;
268:       *ncv = PetscMin(nep->n,nev+(*mpd));
269:     }
270:   }
271:   if (!*mpd) *mpd = *ncv;
272:   return(0);
273: }

275: /*@
276:    NEPAllocateSolution - Allocate memory storage for common variables such
277:    as eigenvalues and eigenvectors.

279:    Collective on nep

281:    Input Parameters:
282: +  nep   - eigensolver context
283: -  extra - number of additional positions, used for methods that require a
284:            working basis slightly larger than ncv

286:    Developers Note:
287:    This is SLEPC_EXTERN because it may be required by user plugin NEP
288:    implementations.

290:    Level: developer
291: @*/
292: PetscErrorCode NEPAllocateSolution(NEP nep,PetscInt extra)
293: {
295:   PetscInt       oldsize,newc,requested;
296:   PetscLogDouble cnt;
297:   Mat            T;
298:   Vec            t;

301:   requested = nep->ncv + extra;

303:   /* oldsize is zero if this is the first time setup is called */
304:   BVGetSizes(nep->V,NULL,NULL,&oldsize);
305:   newc = PetscMax(0,requested-oldsize);

307:   /* allocate space for eigenvalues and friends */
308:   if (requested != oldsize || !nep->eigr) {
309:     PetscFree4(nep->eigr,nep->eigi,nep->errest,nep->perm);
310:     PetscMalloc4(requested,&nep->eigr,requested,&nep->eigi,requested,&nep->errest,requested,&nep->perm);
311:     cnt = newc*sizeof(PetscScalar) + newc*sizeof(PetscReal) + newc*sizeof(PetscInt);
312:     PetscLogObjectMemory((PetscObject)nep,cnt);
313:   }

315:   /* allocate V */
316:   if (!nep->V) { NEPGetBV(nep,&nep->V); }
317:   if (!oldsize) {
318:     if (!((PetscObject)(nep->V))->type_name) {
319:       BVSetType(nep->V,BVSVEC);
320:     }
321:     if (nep->fui==NEP_USER_INTERFACE_SPLIT) T = nep->A[0];
322:     else {
323:       NEPGetFunction(nep,&T,NULL,NULL,NULL);
324:     }
325:     MatCreateVecsEmpty(T,&t,NULL);
326:     BVSetSizesFromVec(nep->V,t,requested);
327:     VecDestroy(&t);
328:   } else {
329:     BVResize(nep->V,requested,PETSC_FALSE);
330:   }

332:   /* allocate W */
333:   if (nep->twosided) {
334:     BVDestroy(&nep->W);
335:     BVDuplicate(nep->V,&nep->W);
336:   }
337:   return(0);
338: }