Actual source code: pciss.c

slepc-3.18.0 2022-10-01
Report Typos and Errors
  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:    SLEPc eigensolver: "ciss"

 13:    Method: Contour Integral Spectral Slicing

 15:    Algorithm:

 17:        Contour integral based on Sakurai-Sugiura method to construct a
 18:        subspace, with various eigenpair extractions (Rayleigh-Ritz,
 19:        explicit moment).

 21:    Based on code contributed by Y. Maeda, T. Sakurai.

 23:    References:

 25:        [1] J. Asakura, T. Sakurai, H. Tadano, T. Ikegami, K. Kimura, "A
 26:            numerical method for polynomial eigenvalue problems using contour
 27:            integral", Japan J. Indust. Appl. Math. 27:73-90, 2010.
 28: */

 30: #include <slepc/private/pepimpl.h>
 31: #include <slepc/private/slepccontour.h>

 33: typedef struct {
 34:   /* parameters */
 35:   PetscInt          N;             /* number of integration points (32) */
 36:   PetscInt          L;             /* block size (16) */
 37:   PetscInt          M;             /* moment degree (N/4 = 4) */
 38:   PetscReal         delta;         /* threshold of singular value (1e-12) */
 39:   PetscInt          L_max;         /* maximum number of columns of the source matrix V */
 40:   PetscReal         spurious_threshold; /* discard spurious eigenpairs */
 41:   PetscBool         isreal;        /* T(z) is real for real z */
 42:   PetscInt          npart;         /* number of partitions */
 43:   PetscInt          refine_inner;
 44:   PetscInt          refine_blocksize;
 45:   PEPCISSExtraction extraction;
 46:   /* private data */
 47:   SlepcContourData  contour;
 48:   PetscReal         *sigma;        /* threshold for numerical rank */
 49:   PetscScalar       *weight;
 50:   PetscScalar       *omega;
 51:   PetscScalar       *pp;
 52:   BV                V;
 53:   BV                S;
 54:   BV                Y;
 55:   PetscBool         useconj;
 56:   Mat               J,*Psplit;     /* auxiliary matrices */
 57:   BV                pV;
 58:   PetscObjectId     rgid;
 59:   PetscObjectState  rgstate;
 60: } PEP_CISS;

 62: static PetscErrorCode PEPComputeFunction(PEP pep,PetscScalar lambda,Mat T,Mat P,PetscBool deriv)
 63: {
 64:   PetscInt         i;
 65:   PetscScalar      *coeff;
 66:   Mat              *A,*K;
 67:   MatStructure     str,strp;
 68:   PEP_CISS         *ctx = (PEP_CISS*)pep->data;
 69:   SlepcContourData contour = ctx->contour;

 71:   A = (contour->pA)?contour->pA:pep->A;
 72:   K = (contour->pP)?contour->pP:ctx->Psplit;
 73:   PetscMalloc1(pep->nmat,&coeff);
 74:   if (deriv) PEPEvaluateBasisDerivative(pep,lambda,0,coeff,NULL);
 75:   else PEPEvaluateBasis(pep,lambda,0,coeff,NULL);
 76:   STGetMatStructure(pep->st,&str);
 77:   MatZeroEntries(T);
 78:   if (!deriv && T != P) {
 79:     STGetSplitPreconditionerInfo(pep->st,NULL,&strp);
 80:     MatZeroEntries(P);
 81:   }
 82:   i = deriv?1:0;
 83:   for (;i<pep->nmat;i++) {
 84:     MatAXPY(T,coeff[i],A[i],str);
 85:     if (!deriv && T != P) MatAXPY(P,coeff[i],K[i],strp);
 86:   }
 87:   PetscFree(coeff);
 88:   return 0;
 89: }

 91: /*
 92:   Set up KSP solvers for every integration point
 93: */
 94: static PetscErrorCode PEPCISSSetUp(PEP pep,Mat T,Mat P)
 95: {
 96:   PEP_CISS         *ctx = (PEP_CISS*)pep->data;
 97:   SlepcContourData contour;
 98:   PetscInt         i,p_id;
 99:   Mat              Amat,Pmat;

101:   if (!ctx->contour || !ctx->contour->ksp) PEPCISSGetKSPs(pep,NULL,NULL);
102:   contour = ctx->contour;
103:   for (i=0;i<contour->npoints;i++) {
104:     p_id = i*contour->subcomm->n + contour->subcomm->color;
105:     MatDuplicate(T,MAT_DO_NOT_COPY_VALUES,&Amat);
106:     if (T != P) MatDuplicate(P,MAT_DO_NOT_COPY_VALUES,&Pmat); else Pmat = Amat;
107:     PEPComputeFunction(pep,ctx->omega[p_id],Amat,Pmat,PETSC_FALSE);
108:     PEP_KSPSetOperators(contour->ksp[i],Amat,Pmat);
109:     MatDestroy(&Amat);
110:     if (T != P) MatDestroy(&Pmat);
111:   }
112:   return 0;
113: }

115: /*
116:   Y_i = F(z_i)^{-1}Fp(z_i)V for every integration point, Y=[Y_i] is in the context
117: */
118: static PetscErrorCode PEPCISSSolve(PEP pep,Mat dT,BV V,PetscInt L_start,PetscInt L_end)
119: {
120:   PEP_CISS         *ctx = (PEP_CISS*)pep->data;
121:   SlepcContourData contour;
122:   PetscInt         i,p_id;
123:   Mat              MV,BMV=NULL,MC;

125:   contour = ctx->contour;
126:   BVSetActiveColumns(V,L_start,L_end);
127:   BVGetMat(V,&MV);
128:   for (i=0;i<contour->npoints;i++) {
129:     p_id = i*contour->subcomm->n + contour->subcomm->color;
130:     PEPComputeFunction(pep,ctx->omega[p_id],dT,NULL,PETSC_TRUE);
131:     BVSetActiveColumns(ctx->Y,i*ctx->L+L_start,i*ctx->L+L_end);
132:     BVGetMat(ctx->Y,&MC);
133:     if (!i) {
134:       MatProductCreate(dT,MV,NULL,&BMV);
135:       MatProductSetType(BMV,MATPRODUCT_AB);
136:       MatProductSetFromOptions(BMV);
137:       MatProductSymbolic(BMV);
138:     }
139:     MatProductNumeric(BMV);
140:     KSPMatSolve(contour->ksp[i],BMV,MC);
141:     BVRestoreMat(ctx->Y,&MC);
142:   }
143:   MatDestroy(&BMV);
144:   BVRestoreMat(V,&MV);
145:   return 0;
146: }

148: PetscErrorCode PEPSetUp_CISS(PEP pep)
149: {
150:   PEP_CISS         *ctx = (PEP_CISS*)pep->data;
151:   SlepcContourData contour;
152:   PetscInt         i,nwork,nsplit;
153:   PetscBool        istrivial,isellipse,flg;
154:   PetscObjectId    id;
155:   PetscObjectState state;
156:   Vec              v0;

158:   if (pep->ncv==PETSC_DEFAULT) pep->ncv = ctx->L_max*ctx->M;
159:   else {
160:     ctx->L_max = pep->ncv/ctx->M;
161:     if (!ctx->L_max) {
162:       ctx->L_max = 1;
163:       pep->ncv = ctx->L_max*ctx->M;
164:     }
165:   }
166:   ctx->L = PetscMin(ctx->L,ctx->L_max);
167:   if (pep->max_it==PETSC_DEFAULT) pep->max_it = 5;
168:   if (pep->mpd==PETSC_DEFAULT) pep->mpd = pep->ncv;
169:   if (!pep->which) pep->which = PEP_ALL;
171:   PEPCheckUnsupported(pep,PEP_FEATURE_STOPPING);
172:   PEPCheckIgnored(pep,PEP_FEATURE_SCALE);

174:   /* check region */
175:   RGIsTrivial(pep->rg,&istrivial);
177:   RGGetComplement(pep->rg,&flg);
179:   PetscObjectTypeCompare((PetscObject)pep->rg,RGELLIPSE,&isellipse);

182:   /* if the region has changed, then reset contour data */
183:   PetscObjectGetId((PetscObject)pep->rg,&id);
184:   PetscObjectStateGet((PetscObject)pep->rg,&state);
185:   if (ctx->rgid && (id != ctx->rgid || state != ctx->rgstate)) {
186:     SlepcContourDataDestroy(&ctx->contour);
187:     PetscInfo(pep,"Resetting the contour data structure due to a change of region\n");
188:     ctx->rgid = id; ctx->rgstate = state;
189:   }

191:   /* create contour data structure */
192:   if (!ctx->contour) {
193:     RGCanUseConjugates(pep->rg,ctx->isreal,&ctx->useconj);
194:     SlepcContourDataCreate(ctx->useconj?ctx->N/2:ctx->N,ctx->npart,(PetscObject)pep,&ctx->contour);
195:   }

197:   PEPAllocateSolution(pep,0);
198:   if (ctx->weight) PetscFree4(ctx->weight,ctx->omega,ctx->pp,ctx->sigma);
199:   PetscMalloc4(ctx->N,&ctx->weight,ctx->N,&ctx->omega,ctx->N,&ctx->pp,ctx->L_max*ctx->M,&ctx->sigma);

201:   /* allocate basis vectors */
202:   BVDestroy(&ctx->S);
203:   BVDuplicateResize(pep->V,ctx->L*ctx->M,&ctx->S);
204:   BVDestroy(&ctx->V);
205:   BVDuplicateResize(pep->V,ctx->L,&ctx->V);

207:   /* check if a user-defined split preconditioner has been set */
208:   STGetSplitPreconditionerInfo(pep->st,&nsplit,NULL);
209:   if (nsplit) {
210:     PetscFree(ctx->Psplit);
211:     PetscMalloc1(nsplit,&ctx->Psplit);
212:     for (i=0;i<nsplit;i++) STGetSplitPreconditionerTerm(pep->st,i,&ctx->Psplit[i]);
213:   }

215:   contour = ctx->contour;
216:   SlepcContourRedundantMat(contour,pep->nmat,pep->A,ctx->Psplit);
217:   if (!ctx->J) MatDuplicate(contour->pA?contour->pA[0]:pep->A[0],MAT_DO_NOT_COPY_VALUES,&ctx->J);
218:   if (contour->pA) {
219:     BVGetColumn(ctx->V,0,&v0);
220:     SlepcContourScatterCreate(contour,v0);
221:     BVRestoreColumn(ctx->V,0,&v0);
222:     BVDestroy(&ctx->pV);
223:     BVCreate(PetscObjectComm((PetscObject)contour->xsub),&ctx->pV);
224:     BVSetSizesFromVec(ctx->pV,contour->xsub,pep->n);
225:     BVSetFromOptions(ctx->pV);
226:     BVResize(ctx->pV,ctx->L,PETSC_FALSE);
227:   }

229:   BVDestroy(&ctx->Y);
230:   if (contour->pA) {
231:     BVCreate(PetscObjectComm((PetscObject)contour->xsub),&ctx->Y);
232:     BVSetSizesFromVec(ctx->Y,contour->xsub,pep->n);
233:     BVSetFromOptions(ctx->Y);
234:     BVResize(ctx->Y,contour->npoints*ctx->L,PETSC_FALSE);
235:   } else BVDuplicateResize(pep->V,contour->npoints*ctx->L,&ctx->Y);

237:   if (ctx->extraction == PEP_CISS_EXTRACTION_HANKEL) DSSetType(pep->ds,DSGNHEP);
238:   else if (ctx->extraction == PEP_CISS_EXTRACTION_CAA) DSSetType(pep->ds,DSNHEP);
239:   else {
240:     DSSetType(pep->ds,DSPEP);
241:     DSPEPSetDegree(pep->ds,pep->nmat-1);
242:     DSPEPSetCoefficients(pep->ds,pep->pbc);
243:   }
244:   DSAllocate(pep->ds,pep->ncv);
245:   nwork = 2;
246:   PEPSetWorkVecs(pep,nwork);
247:   return 0;
248: }

250: PetscErrorCode PEPSolve_CISS(PEP pep)
251: {
252:   PEP_CISS         *ctx = (PEP_CISS*)pep->data;
253:   SlepcContourData contour = ctx->contour;
254:   Mat              X,M,E,T,P;
255:   PetscInt         i,j,ld,L_add=0,nv=0,L_base=ctx->L,inner,*inside,nsplit;
256:   PetscScalar      *Mu,*H0,*H1,*rr,*temp,center;
257:   PetscReal        error,max_error,radius,rgscale,est_eig,eta;
258:   PetscBool        isellipse,*fl1;
259:   Vec              si;
260:   SlepcSC          sc;
261:   PetscRandom      rand;

263:   DSSetFromOptions(pep->ds);
264:   DSGetSlepcSC(pep->ds,&sc);
265:   sc->comparison    = SlepcCompareLargestMagnitude;
266:   sc->comparisonctx = NULL;
267:   sc->map           = NULL;
268:   sc->mapobj        = NULL;
269:   DSGetLeadingDimension(pep->ds,&ld);
270:   RGComputeQuadrature(pep->rg,RG_QUADRULE_TRAPEZOIDAL,ctx->N,ctx->omega,ctx->pp,ctx->weight);
271:   STGetSplitPreconditionerInfo(pep->st,&nsplit,NULL);
272:   if (contour->pA) {
273:     T = contour->pA[0];
274:     P = nsplit? contour->pP[0]: T;
275:   } else {
276:     T = pep->A[0];
277:     P = nsplit? ctx->Psplit[0]: T;
278:   }
279:   PEPCISSSetUp(pep,T,P);
280:   BVSetActiveColumns(ctx->V,0,ctx->L);
281:   BVSetRandomSign(ctx->V);
282:   BVGetRandomContext(ctx->V,&rand);
283:   if (contour->pA) BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup);
284:   PEPCISSSolve(pep,ctx->J,(contour->pA)?ctx->pV:ctx->V,0,ctx->L);
285:   PetscObjectTypeCompare((PetscObject)pep->rg,RGELLIPSE,&isellipse);
286:   if (isellipse) {
287:     BVTraceQuadrature(ctx->Y,ctx->V,ctx->L,ctx->L,ctx->weight,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj,&est_eig);
288:     PetscInfo(pep,"Estimated eigenvalue count: %f\n",(double)est_eig);
289:     eta = PetscPowReal(10.0,-PetscLog10Real(pep->tol)/ctx->N);
290:     L_add = PetscMax(0,(PetscInt)PetscCeilReal((est_eig*eta)/ctx->M)-ctx->L);
291:     if (L_add>ctx->L_max-ctx->L) {
292:       PetscInfo(pep,"Number of eigenvalues inside the contour path may be too large\n");
293:       L_add = ctx->L_max-ctx->L;
294:     }
295:   }
296:   /* Updates L after estimate the number of eigenvalue */
297:   if (L_add>0) {
298:     PetscInfo(pep,"Changing L %" PetscInt_FMT " -> %" PetscInt_FMT " by Estimate #Eig\n",ctx->L,ctx->L+L_add);
299:     BVCISSResizeBases(ctx->S,contour->pA?ctx->pV:ctx->V,ctx->Y,ctx->L,ctx->L+L_add,ctx->M,contour->npoints);
300:     BVSetActiveColumns(ctx->V,ctx->L,ctx->L+L_add);
301:     BVSetRandomSign(ctx->V);
302:     if (contour->pA) BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup);
303:     ctx->L += L_add;
304:     PEPCISSSolve(pep,ctx->J,(contour->pA)?ctx->pV:ctx->V,ctx->L-L_add,ctx->L);
305:   }

307:   PetscMalloc2(ctx->L*ctx->L*ctx->M*2,&Mu,ctx->L*ctx->M*ctx->L*ctx->M,&H0);
308:   for (i=0;i<ctx->refine_blocksize;i++) {
309:     BVDotQuadrature(ctx->Y,(contour->pA)?ctx->pV:ctx->V,Mu,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->subcomm,contour->npoints,ctx->useconj);
310:     CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0);
311:     PetscLogEventBegin(PEP_CISS_SVD,pep,0,0,0);
312:     SlepcCISS_BH_SVD(H0,ctx->L*ctx->M,ctx->delta,ctx->sigma,&nv);
313:     PetscLogEventEnd(PEP_CISS_SVD,pep,0,0,0);
314:     if (ctx->sigma[0]<=ctx->delta || nv < ctx->L*ctx->M || ctx->L == ctx->L_max) break;
315:     L_add = L_base;
316:     if (ctx->L+L_add>ctx->L_max) L_add = ctx->L_max-ctx->L;
317:     PetscInfo(pep,"Changing L %" PetscInt_FMT " -> %" PetscInt_FMT " by SVD(H0)\n",ctx->L,ctx->L+L_add);
318:     BVCISSResizeBases(ctx->S,contour->pA?ctx->pV:ctx->V,ctx->Y,ctx->L,ctx->L+L_add,ctx->M,contour->npoints);
319:     BVSetActiveColumns(ctx->V,ctx->L,ctx->L+L_add);
320:     BVSetRandomSign(ctx->V);
321:     if (contour->pA) BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup);
322:     ctx->L += L_add;
323:     PEPCISSSolve(pep,ctx->J,(contour->pA)?ctx->pV:ctx->V,ctx->L-L_add,ctx->L);
324:     if (L_add) {
325:       PetscFree2(Mu,H0);
326:       PetscMalloc2(ctx->L*ctx->L*ctx->M*2,&Mu,ctx->L*ctx->M*ctx->L*ctx->M,&H0);
327:     }
328:   }

330:   RGGetScale(pep->rg,&rgscale);
331:   RGEllipseGetParameters(pep->rg,&center,&radius,NULL);

333:   if (ctx->extraction == PEP_CISS_EXTRACTION_HANKEL) PetscMalloc1(ctx->L*ctx->M*ctx->L*ctx->M,&H1);

335:   while (pep->reason == PEP_CONVERGED_ITERATING) {
336:     pep->its++;
337:     for (inner=0;inner<=ctx->refine_inner;inner++) {
338:       if (ctx->extraction == PEP_CISS_EXTRACTION_HANKEL) {
339:         BVDotQuadrature(ctx->Y,(contour->pA)?ctx->pV:ctx->V,Mu,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->subcomm,contour->npoints,ctx->useconj);
340:         CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0);
341:         PetscLogEventBegin(PEP_CISS_SVD,pep,0,0,0);
342:         SlepcCISS_BH_SVD(H0,ctx->L*ctx->M,ctx->delta,ctx->sigma,&nv);
343:         PetscLogEventEnd(PEP_CISS_SVD,pep,0,0,0);
344:       } else {
345:         BVSumQuadrature(ctx->S,ctx->Y,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj);
346:         /* compute SVD of S */
347:         BVSVDAndRank(ctx->S,ctx->M,ctx->L,ctx->delta,(ctx->extraction==PEP_CISS_EXTRACTION_CAA)?BV_SVD_METHOD_QR_CAA:BV_SVD_METHOD_QR,H0,ctx->sigma,&nv);
348:       }
349:       PetscInfo(pep,"Estimated rank: nv = %" PetscInt_FMT "\n",nv);
350:       if (ctx->sigma[0]>ctx->delta && nv==ctx->L*ctx->M && inner!=ctx->refine_inner) {
351:         BVSumQuadrature(ctx->S,ctx->Y,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj);
352:         BVSetActiveColumns(ctx->S,0,ctx->L);
353:         BVSetActiveColumns(ctx->V,0,ctx->L);
354:         BVCopy(ctx->S,ctx->V);
355:         if (contour->pA) BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup);
356:         PEPCISSSolve(pep,ctx->J,(contour->pA)?ctx->pV:ctx->V,0,ctx->L);
357:       } else break;
358:     }
359:     pep->nconv = 0;
360:     if (nv == 0) { pep->reason = PEP_CONVERGED_TOL; break; }
361:     else {
362:       /* Extracting eigenpairs */
363:       DSSetDimensions(pep->ds,nv,0,0);
364:       DSSetState(pep->ds,DS_STATE_RAW);
365:       if (ctx->extraction == PEP_CISS_EXTRACTION_HANKEL) {
366:         CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0);
367:         CISS_BlockHankel(Mu,1,ctx->L,ctx->M,H1);
368:         DSGetArray(pep->ds,DS_MAT_A,&temp);
369:         for (j=0;j<nv;j++)
370:           for (i=0;i<nv;i++)
371:             temp[i+j*ld] = H1[i+j*ctx->L*ctx->M];
372:         DSRestoreArray(pep->ds,DS_MAT_A,&temp);
373:         DSGetArray(pep->ds,DS_MAT_B,&temp);
374:         for (j=0;j<nv;j++)
375:           for (i=0;i<nv;i++)
376:             temp[i+j*ld] = H0[i+j*ctx->L*ctx->M];
377:         DSRestoreArray(pep->ds,DS_MAT_B,&temp);
378:       } else if (ctx->extraction == PEP_CISS_EXTRACTION_CAA) {
379:         BVSetActiveColumns(ctx->S,0,nv);
380:         DSGetArray(pep->ds,DS_MAT_A,&temp);
381:         for (i=0;i<nv;i++) PetscArraycpy(temp+i*ld,H0+i*nv,nv);
382:         DSRestoreArray(pep->ds,DS_MAT_A,&temp);
383:       } else {
384:         BVSetActiveColumns(ctx->S,0,nv);
385:         for (i=0;i<pep->nmat;i++) {
386:           DSGetMat(pep->ds,DSMatExtra[i],&E);
387:           BVMatProject(ctx->S,pep->A[i],ctx->S,E);
388:           DSRestoreMat(pep->ds,DSMatExtra[i],&E);
389:         }
390:         nv = (pep->nmat-1)*nv;
391:       }
392:       DSSolve(pep->ds,pep->eigr,pep->eigi);
393:       DSSynchronize(pep->ds,pep->eigr,pep->eigi);
394:       if (ctx->extraction == PEP_CISS_EXTRACTION_CAA || ctx->extraction == PEP_CISS_EXTRACTION_HANKEL) {
395:         for (i=0;i<nv;i++) {
396:           pep->eigr[i] = (pep->eigr[i]*radius+center)*rgscale;
397:         }
398:       }
399:       PetscMalloc3(nv,&fl1,nv,&inside,nv,&rr);
400:       DSVectors(pep->ds,DS_MAT_X,NULL,NULL);
401:       DSGetMat(pep->ds,DS_MAT_X,&X);
402:       SlepcCISS_isGhost(X,nv,ctx->sigma,ctx->spurious_threshold,fl1);
403:       DSRestoreMat(pep->ds,DS_MAT_X,&X);
404:       RGCheckInside(pep->rg,nv,pep->eigr,pep->eigi,inside);
405:       for (i=0;i<nv;i++) {
406:         if (fl1[i] && inside[i]>=0) {
407:           rr[i] = 1.0;
408:           pep->nconv++;
409:         } else rr[i] = 0.0;
410:       }
411:       DSSort(pep->ds,pep->eigr,pep->eigi,rr,NULL,&pep->nconv);
412:       DSSynchronize(pep->ds,pep->eigr,pep->eigi);
413:       if (ctx->extraction == PEP_CISS_EXTRACTION_CAA || ctx->extraction == PEP_CISS_EXTRACTION_HANKEL) {
414:         for (i=0;i<nv;i++) pep->eigr[i] = (pep->eigr[i]*radius+center)*rgscale;
415:       }
416:       PetscFree3(fl1,inside,rr);
417:       BVSetActiveColumns(pep->V,0,nv);
418:       DSVectors(pep->ds,DS_MAT_X,NULL,NULL);
419:       if (ctx->extraction == PEP_CISS_EXTRACTION_HANKEL) {
420:         BVSumQuadrature(ctx->S,ctx->Y,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj);
421:         BVSetActiveColumns(ctx->S,0,nv);
422:         BVCopy(ctx->S,pep->V);
423:         DSGetMat(pep->ds,DS_MAT_X,&X);
424:         BVMultInPlace(ctx->S,X,0,pep->nconv);
425:         BVMultInPlace(pep->V,X,0,pep->nconv);
426:         DSRestoreMat(pep->ds,DS_MAT_X,&X);
427:       } else {
428:         DSGetMat(pep->ds,DS_MAT_X,&X);
429:         BVMultInPlace(ctx->S,X,0,pep->nconv);
430:         DSRestoreMat(pep->ds,DS_MAT_X,&X);
431:         BVSetActiveColumns(ctx->S,0,pep->nconv);
432:         BVCopy(ctx->S,pep->V);
433:       }
434:       max_error = 0.0;
435:       for (i=0;i<pep->nconv;i++) {
436:         BVGetColumn(pep->V,i,&si);
437:         VecNormalize(si,NULL);
438:         PEPComputeResidualNorm_Private(pep,pep->eigr[i],0,si,NULL,pep->work,&error);
439:         (*pep->converged)(pep,pep->eigr[i],0,error,&error,pep->convergedctx);
440:         BVRestoreColumn(pep->V,i,&si);
441:         max_error = PetscMax(max_error,error);
442:       }
443:       if (max_error <= pep->tol) pep->reason = PEP_CONVERGED_TOL;
444:       else if (pep->its > pep->max_it) pep->reason = PEP_DIVERGED_ITS;
445:       else {
446:         if (pep->nconv > ctx->L) nv = pep->nconv;
447:         else if (ctx->L > nv) nv = ctx->L;
448:         nv = PetscMin(nv,ctx->L*ctx->M);
449:         MatCreateSeqDense(PETSC_COMM_SELF,nv,ctx->L,NULL,&M);
450:         MatSetRandom(M,rand);
451:         BVSetActiveColumns(ctx->S,0,nv);
452:         BVMultInPlace(ctx->S,M,0,ctx->L);
453:         MatDestroy(&M);
454:         BVSetActiveColumns(ctx->S,0,ctx->L);
455:         BVSetActiveColumns(ctx->V,0,ctx->L);
456:         BVCopy(ctx->S,ctx->V);
457:         if (contour->pA) BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup);
458:         PEPCISSSolve(pep,ctx->J,(contour->pA)?ctx->pV:ctx->V,0,ctx->L);
459:       }
460:     }
461:   }
462:   PetscFree2(Mu,H0);
463:   if (ctx->extraction == PEP_CISS_EXTRACTION_HANKEL) PetscFree(H1);
464:   return 0;
465: }

467: static PetscErrorCode PEPCISSSetSizes_CISS(PEP pep,PetscInt ip,PetscInt bs,PetscInt ms,PetscInt npart,PetscInt bsmax,PetscBool realmats)
468: {
469:   PEP_CISS       *ctx = (PEP_CISS*)pep->data;
470:   PetscInt       oN,oL,oM,oLmax,onpart;
471:   PetscMPIInt    size;

473:   oN = ctx->N;
474:   if (ip == PETSC_DECIDE || ip == PETSC_DEFAULT) {
475:     if (ctx->N!=32) { ctx->N =32; ctx->M = ctx->N/4; }
476:   } else {
479:     if (ctx->N!=ip) { ctx->N = ip; ctx->M = ctx->N/4; }
480:   }
481:   oL = ctx->L;
482:   if (bs == PETSC_DECIDE || bs == PETSC_DEFAULT) {
483:     ctx->L = 16;
484:   } else {
486:     ctx->L = bs;
487:   }
488:   oM = ctx->M;
489:   if (ms == PETSC_DECIDE || ms == PETSC_DEFAULT) {
490:     ctx->M = ctx->N/4;
491:   } else {
494:     ctx->M = PetscMax(ms,2);
495:   }
496:   onpart = ctx->npart;
497:   if (npart == PETSC_DECIDE || npart == PETSC_DEFAULT) {
498:     ctx->npart = 1;
499:   } else {
500:     MPI_Comm_size(PetscObjectComm((PetscObject)pep),&size);
502:     ctx->npart = npart;
503:   }
504:   oLmax = ctx->L_max;
505:   if (bsmax == PETSC_DECIDE || bsmax == PETSC_DEFAULT) {
506:     ctx->L_max = 64;
507:   } else {
509:     ctx->L_max = PetscMax(bsmax,ctx->L);
510:   }
511:   if (onpart != ctx->npart || oN != ctx->N || realmats != ctx->isreal) {
512:     SlepcContourDataDestroy(&ctx->contour);
513:     PetscInfo(pep,"Resetting the contour data structure due to a change of parameters\n");
514:     pep->state = PEP_STATE_INITIAL;
515:   }
516:   ctx->isreal = realmats;
517:   if (oL != ctx->L || oM != ctx->M || oLmax != ctx->L_max) pep->state = PEP_STATE_INITIAL;
518:   return 0;
519: }

521: /*@
522:    PEPCISSSetSizes - Sets the values of various size parameters in the CISS solver.

524:    Logically Collective on pep

526:    Input Parameters:
527: +  pep   - the polynomial eigensolver context
528: .  ip    - number of integration points
529: .  bs    - block size
530: .  ms    - moment size
531: .  npart - number of partitions when splitting the communicator
532: .  bsmax - max block size
533: -  realmats - all coefficient matrices of P(.) are real

535:    Options Database Keys:
536: +  -pep_ciss_integration_points - Sets the number of integration points
537: .  -pep_ciss_blocksize - Sets the block size
538: .  -pep_ciss_moments - Sets the moment size
539: .  -pep_ciss_partitions - Sets the number of partitions
540: .  -pep_ciss_maxblocksize - Sets the maximum block size
541: -  -pep_ciss_realmats - all coefficient matrices of P(.) are real

543:    Notes:
544:    The default number of partitions is 1. This means the internal KSP object is shared
545:    among all processes of the PEP communicator. Otherwise, the communicator is split
546:    into npart communicators, so that npart KSP solves proceed simultaneously.

548:    Level: advanced

550: .seealso: PEPCISSGetSizes()
551: @*/
552: PetscErrorCode PEPCISSSetSizes(PEP pep,PetscInt ip,PetscInt bs,PetscInt ms,PetscInt npart,PetscInt bsmax,PetscBool realmats)
553: {
561:   PetscTryMethod(pep,"PEPCISSSetSizes_C",(PEP,PetscInt,PetscInt,PetscInt,PetscInt,PetscInt,PetscBool),(pep,ip,bs,ms,npart,bsmax,realmats));
562:   return 0;
563: }

565: static PetscErrorCode PEPCISSGetSizes_CISS(PEP pep,PetscInt *ip,PetscInt *bs,PetscInt *ms,PetscInt *npart,PetscInt *bsmax,PetscBool *realmats)
566: {
567:   PEP_CISS *ctx = (PEP_CISS*)pep->data;

569:   if (ip) *ip = ctx->N;
570:   if (bs) *bs = ctx->L;
571:   if (ms) *ms = ctx->M;
572:   if (npart) *npart = ctx->npart;
573:   if (bsmax) *bsmax = ctx->L_max;
574:   if (realmats) *realmats = ctx->isreal;
575:   return 0;
576: }

578: /*@
579:    PEPCISSGetSizes - Gets the values of various size parameters in the CISS solver.

581:    Not Collective

583:    Input Parameter:
584: .  pep - the polynomial eigensolver context

586:    Output Parameters:
587: +  ip    - number of integration points
588: .  bs    - block size
589: .  ms    - moment size
590: .  npart - number of partitions when splitting the communicator
591: .  bsmax - max block size
592: -  realmats - all coefficient matrices of P(.) are real

594:    Level: advanced

596: .seealso: PEPCISSSetSizes()
597: @*/
598: PetscErrorCode PEPCISSGetSizes(PEP pep,PetscInt *ip,PetscInt *bs,PetscInt *ms,PetscInt *npart,PetscInt *bsmax,PetscBool *realmats)
599: {
601:   PetscUseMethod(pep,"PEPCISSGetSizes_C",(PEP,PetscInt*,PetscInt*,PetscInt*,PetscInt*,PetscInt*,PetscBool*),(pep,ip,bs,ms,npart,bsmax,realmats));
602:   return 0;
603: }

605: static PetscErrorCode PEPCISSSetThreshold_CISS(PEP pep,PetscReal delta,PetscReal spur)
606: {
607:   PEP_CISS *ctx = (PEP_CISS*)pep->data;

609:   if (delta == PETSC_DEFAULT) {
610:     ctx->delta = SLEPC_DEFAULT_TOL*1e-4;
611:   } else {
613:     ctx->delta = delta;
614:   }
615:   if (spur == PETSC_DEFAULT) {
616:     ctx->spurious_threshold = PetscSqrtReal(SLEPC_DEFAULT_TOL);
617:   } else {
619:     ctx->spurious_threshold = spur;
620:   }
621:   return 0;
622: }

624: /*@
625:    PEPCISSSetThreshold - Sets the values of various threshold parameters in
626:    the CISS solver.

628:    Logically Collective on pep

630:    Input Parameters:
631: +  pep   - the polynomial eigensolver context
632: .  delta - threshold for numerical rank
633: -  spur  - spurious threshold (to discard spurious eigenpairs)

635:    Options Database Keys:
636: +  -pep_ciss_delta - Sets the delta
637: -  -pep_ciss_spurious_threshold - Sets the spurious threshold

639:    Level: advanced

641: .seealso: PEPCISSGetThreshold()
642: @*/
643: PetscErrorCode PEPCISSSetThreshold(PEP pep,PetscReal delta,PetscReal spur)
644: {
648:   PetscTryMethod(pep,"PEPCISSSetThreshold_C",(PEP,PetscReal,PetscReal),(pep,delta,spur));
649:   return 0;
650: }

652: static PetscErrorCode PEPCISSGetThreshold_CISS(PEP pep,PetscReal *delta,PetscReal *spur)
653: {
654:   PEP_CISS *ctx = (PEP_CISS*)pep->data;

656:   if (delta) *delta = ctx->delta;
657:   if (spur)  *spur = ctx->spurious_threshold;
658:   return 0;
659: }

661: /*@
662:    PEPCISSGetThreshold - Gets the values of various threshold parameters in
663:    the CISS solver.

665:    Not Collective

667:    Input Parameter:
668: .  pep - the polynomial eigensolver context

670:    Output Parameters:
671: +  delta - threshold for numerical rank
672: -  spur  - spurious threshold (to discard spurious eigenpairs)

674:    Level: advanced

676: .seealso: PEPCISSSetThreshold()
677: @*/
678: PetscErrorCode PEPCISSGetThreshold(PEP pep,PetscReal *delta,PetscReal *spur)
679: {
681:   PetscUseMethod(pep,"PEPCISSGetThreshold_C",(PEP,PetscReal*,PetscReal*),(pep,delta,spur));
682:   return 0;
683: }

685: static PetscErrorCode PEPCISSSetRefinement_CISS(PEP pep,PetscInt inner,PetscInt blsize)
686: {
687:   PEP_CISS *ctx = (PEP_CISS*)pep->data;

689:   if (inner == PETSC_DEFAULT) {
690:     ctx->refine_inner = 0;
691:   } else {
693:     ctx->refine_inner = inner;
694:   }
695:   if (blsize == PETSC_DEFAULT) {
696:     ctx->refine_blocksize = 0;
697:   } else {
699:     ctx->refine_blocksize = blsize;
700:   }
701:   return 0;
702: }

704: /*@
705:    PEPCISSSetRefinement - Sets the values of various refinement parameters
706:    in the CISS solver.

708:    Logically Collective on pep

710:    Input Parameters:
711: +  pep    - the polynomial eigensolver context
712: .  inner  - number of iterative refinement iterations (inner loop)
713: -  blsize - number of iterative refinement iterations (blocksize loop)

715:    Options Database Keys:
716: +  -pep_ciss_refine_inner - Sets number of inner iterations
717: -  -pep_ciss_refine_blocksize - Sets number of blocksize iterations

719:    Level: advanced

721: .seealso: PEPCISSGetRefinement()
722: @*/
723: PetscErrorCode PEPCISSSetRefinement(PEP pep,PetscInt inner,PetscInt blsize)
724: {
728:   PetscTryMethod(pep,"PEPCISSSetRefinement_C",(PEP,PetscInt,PetscInt),(pep,inner,blsize));
729:   return 0;
730: }

732: static PetscErrorCode PEPCISSGetRefinement_CISS(PEP pep,PetscInt *inner,PetscInt *blsize)
733: {
734:   PEP_CISS *ctx = (PEP_CISS*)pep->data;

736:   if (inner)  *inner = ctx->refine_inner;
737:   if (blsize) *blsize = ctx->refine_blocksize;
738:   return 0;
739: }

741: /*@
742:    PEPCISSGetRefinement - Gets the values of various refinement parameters
743:    in the CISS solver.

745:    Not Collective

747:    Input Parameter:
748: .  pep - the polynomial eigensolver context

750:    Output Parameters:
751: +  inner  - number of iterative refinement iterations (inner loop)
752: -  blsize - number of iterative refinement iterations (blocksize loop)

754:    Level: advanced

756: .seealso: PEPCISSSetRefinement()
757: @*/
758: PetscErrorCode PEPCISSGetRefinement(PEP pep, PetscInt *inner, PetscInt *blsize)
759: {
761:   PetscUseMethod(pep,"PEPCISSGetRefinement_C",(PEP,PetscInt*,PetscInt*),(pep,inner,blsize));
762:   return 0;
763: }

765: static PetscErrorCode PEPCISSSetExtraction_CISS(PEP pep,PEPCISSExtraction extraction)
766: {
767:   PEP_CISS *ctx = (PEP_CISS*)pep->data;

769:   if (ctx->extraction != extraction) {
770:     ctx->extraction = extraction;
771:     pep->state      = PEP_STATE_INITIAL;
772:   }
773:   return 0;
774: }

776: /*@
777:    PEPCISSSetExtraction - Sets the extraction technique used in the CISS solver.

779:    Logically Collective on pep

781:    Input Parameters:
782: +  pep        - the polynomial eigensolver context
783: -  extraction - the extraction technique

785:    Options Database Key:
786: .  -pep_ciss_extraction - Sets the extraction technique (either 'ritz', 'hankel' or 'caa')

788:    Notes:
789:    By default, the Rayleigh-Ritz extraction is used (PEP_CISS_EXTRACTION_RITZ).

791:    If the 'hankel' or the 'caa' option is specified (PEP_CISS_EXTRACTION_HANKEL or
792:    PEP_CISS_EXTRACTION_CAA), then the Block Hankel method, or the Communication-avoiding
793:    Arnoldi method, respectively, is used for extracting eigenpairs.

795:    Level: advanced

797: .seealso: PEPCISSGetExtraction(), PEPCISSExtraction
798: @*/
799: PetscErrorCode PEPCISSSetExtraction(PEP pep,PEPCISSExtraction extraction)
800: {
803:   PetscTryMethod(pep,"PEPCISSSetExtraction_C",(PEP,PEPCISSExtraction),(pep,extraction));
804:   return 0;
805: }

807: static PetscErrorCode PEPCISSGetExtraction_CISS(PEP pep,PEPCISSExtraction *extraction)
808: {
809:   PEP_CISS *ctx = (PEP_CISS*)pep->data;

811:   *extraction = ctx->extraction;
812:   return 0;
813: }

815: /*@
816:    PEPCISSGetExtraction - Gets the extraction technique used in the CISS solver.

818:    Not Collective

820:    Input Parameter:
821: .  pep - the polynomial eigensolver context

823:    Output Parameters:
824: .  extraction - extraction technique

826:    Level: advanced

828: .seealso: PEPCISSSetExtraction() PEPCISSExtraction
829: @*/
830: PetscErrorCode PEPCISSGetExtraction(PEP pep,PEPCISSExtraction *extraction)
831: {
834:   PetscUseMethod(pep,"PEPCISSGetExtraction_C",(PEP,PEPCISSExtraction*),(pep,extraction));
835:   return 0;
836: }

838: static PetscErrorCode PEPCISSGetKSPs_CISS(PEP pep,PetscInt *nsolve,KSP **ksp)
839: {
840:   PEP_CISS         *ctx = (PEP_CISS*)pep->data;
841:   SlepcContourData contour;
842:   PetscInt         i,nsplit;
843:   PC               pc;
844:   MPI_Comm         child;

846:   if (!ctx->contour) {  /* initialize contour data structure first */
847:     RGCanUseConjugates(pep->rg,ctx->isreal,&ctx->useconj);
848:     SlepcContourDataCreate(ctx->useconj?ctx->N/2:ctx->N,ctx->npart,(PetscObject)pep,&ctx->contour);
849:   }
850:   contour = ctx->contour;
851:   if (!contour->ksp) {
852:     PetscMalloc1(contour->npoints,&contour->ksp);
853:     PEPGetST(pep,&pep->st);
854:     STGetSplitPreconditionerInfo(pep->st,&nsplit,NULL);
855:     PetscSubcommGetChild(contour->subcomm,&child);
856:     for (i=0;i<contour->npoints;i++) {
857:       KSPCreate(child,&contour->ksp[i]);
858:       PetscObjectIncrementTabLevel((PetscObject)contour->ksp[i],(PetscObject)pep,1);
859:       KSPSetOptionsPrefix(contour->ksp[i],((PetscObject)pep)->prefix);
860:       KSPAppendOptionsPrefix(contour->ksp[i],"pep_ciss_");
861:       PetscObjectSetOptions((PetscObject)contour->ksp[i],((PetscObject)pep)->options);
862:       KSPSetErrorIfNotConverged(contour->ksp[i],PETSC_TRUE);
863:       KSPSetTolerances(contour->ksp[i],SlepcDefaultTol(pep->tol),PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT);
864:       KSPGetPC(contour->ksp[i],&pc);
865:       if (nsplit) {
866:         KSPSetType(contour->ksp[i],KSPBCGS);
867:         PCSetType(pc,PCBJACOBI);
868:       } else {
869:         KSPSetType(contour->ksp[i],KSPPREONLY);
870:         PCSetType(pc,PCLU);
871:       }
872:     }
873:   }
874:   if (nsolve) *nsolve = contour->npoints;
875:   if (ksp)    *ksp    = contour->ksp;
876:   return 0;
877: }

879: /*@C
880:    PEPCISSGetKSPs - Retrieve the array of linear solver objects associated with
881:    the CISS solver.

883:    Not Collective

885:    Input Parameter:
886: .  pep - polynomial eigenvalue solver

888:    Output Parameters:
889: +  nsolve - number of solver objects
890: -  ksp - array of linear solver object

892:    Notes:
893:    The number of KSP solvers is equal to the number of integration points divided by
894:    the number of partitions. This value is halved in the case of real matrices with
895:    a region centered at the real axis.

897:    Level: advanced

899: .seealso: PEPCISSSetSizes()
900: @*/
901: PetscErrorCode PEPCISSGetKSPs(PEP pep,PetscInt *nsolve,KSP **ksp)
902: {
904:   PetscUseMethod(pep,"PEPCISSGetKSPs_C",(PEP,PetscInt*,KSP**),(pep,nsolve,ksp));
905:   return 0;
906: }

908: PetscErrorCode PEPReset_CISS(PEP pep)
909: {
910:   PEP_CISS       *ctx = (PEP_CISS*)pep->data;

912:   BVDestroy(&ctx->S);
913:   BVDestroy(&ctx->V);
914:   BVDestroy(&ctx->Y);
915:   SlepcContourDataReset(ctx->contour);
916:   MatDestroy(&ctx->J);
917:   BVDestroy(&ctx->pV);
918:   PetscFree(ctx->Psplit);
919:   return 0;
920: }

922: PetscErrorCode PEPSetFromOptions_CISS(PEP pep,PetscOptionItems *PetscOptionsObject)
923: {
924:   PEP_CISS          *ctx = (PEP_CISS*)pep->data;
925:   PetscReal         r1,r2;
926:   PetscInt          i,i1,i2,i3,i4,i5,i6,i7;
927:   PetscBool         b1,flg,flg2,flg3,flg4,flg5,flg6;
928:   PEPCISSExtraction extraction;

930:   PetscOptionsHeadBegin(PetscOptionsObject,"PEP CISS Options");

932:     PEPCISSGetSizes(pep,&i1,&i2,&i3,&i4,&i5,&b1);
933:     PetscOptionsInt("-pep_ciss_integration_points","Number of integration points","PEPCISSSetSizes",i1,&i1,&flg);
934:     PetscOptionsInt("-pep_ciss_blocksize","Block size","PEPCISSSetSizes",i2,&i2,&flg2);
935:     PetscOptionsInt("-pep_ciss_moments","Moment size","PEPCISSSetSizes",i3,&i3,&flg3);
936:     PetscOptionsInt("-pep_ciss_partitions","Number of partitions","PEPCISSSetSizes",i4,&i4,&flg4);
937:     PetscOptionsInt("-pep_ciss_maxblocksize","Maximum block size","PEPCISSSetSizes",i5,&i5,&flg5);
938:     PetscOptionsBool("-pep_ciss_realmats","True if all coefficient matrices of P(.) are real","PEPCISSSetSizes",b1,&b1,&flg6);
939:     if (flg || flg2 || flg3 || flg4 || flg5 || flg6) PEPCISSSetSizes(pep,i1,i2,i3,i4,i5,b1);

941:     PEPCISSGetThreshold(pep,&r1,&r2);
942:     PetscOptionsReal("-pep_ciss_delta","Threshold for numerical rank","PEPCISSSetThreshold",r1,&r1,&flg);
943:     PetscOptionsReal("-pep_ciss_spurious_threshold","Threshold for the spurious eigenpairs","PEPCISSSetThreshold",r2,&r2,&flg2);
944:     if (flg || flg2) PEPCISSSetThreshold(pep,r1,r2);

946:     PEPCISSGetRefinement(pep,&i6,&i7);
947:     PetscOptionsInt("-pep_ciss_refine_inner","Number of inner iterative refinement iterations","PEPCISSSetRefinement",i6,&i6,&flg);
948:     PetscOptionsInt("-pep_ciss_refine_blocksize","Number of blocksize iterative refinement iterations","PEPCISSSetRefinement",i7,&i7,&flg2);
949:     if (flg || flg2) PEPCISSSetRefinement(pep,i6,i7);

951:     PetscOptionsEnum("-pep_ciss_extraction","Extraction technique","PEPCISSSetExtraction",PEPCISSExtractions,(PetscEnum)ctx->extraction,(PetscEnum*)&extraction,&flg);
952:     if (flg) PEPCISSSetExtraction(pep,extraction);

954:   PetscOptionsHeadEnd();

956:   if (!pep->rg) PEPGetRG(pep,&pep->rg);
957:   RGSetFromOptions(pep->rg); /* this is necessary here to set useconj */
958:   if (!ctx->contour || !ctx->contour->ksp) PEPCISSGetKSPs(pep,NULL,NULL);
959:   for (i=0;i<ctx->contour->npoints;i++) KSPSetFromOptions(ctx->contour->ksp[i]);
960:   PetscSubcommSetFromOptions(ctx->contour->subcomm);
961:   return 0;
962: }

964: PetscErrorCode PEPDestroy_CISS(PEP pep)
965: {
966:   PEP_CISS       *ctx = (PEP_CISS*)pep->data;

968:   SlepcContourDataDestroy(&ctx->contour);
969:   PetscFree4(ctx->weight,ctx->omega,ctx->pp,ctx->sigma);
970:   PetscFree(pep->data);
971:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSSetSizes_C",NULL);
972:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetSizes_C",NULL);
973:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSSetThreshold_C",NULL);
974:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetThreshold_C",NULL);
975:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSSetRefinement_C",NULL);
976:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetRefinement_C",NULL);
977:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSSetExtraction_C",NULL);
978:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetExtraction_C",NULL);
979:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetKSPs_C",NULL);
980:   return 0;
981: }

983: PetscErrorCode PEPView_CISS(PEP pep,PetscViewer viewer)
984: {
985:   PEP_CISS       *ctx = (PEP_CISS*)pep->data;
986:   PetscBool      isascii;
987:   PetscViewer    sviewer;

989:   PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&isascii);
990:   if (isascii) {
991:     PetscViewerASCIIPrintf(viewer,"  sizes { integration points: %" PetscInt_FMT ", block size: %" PetscInt_FMT ", moment size: %" PetscInt_FMT ", partitions: %" PetscInt_FMT ", maximum block size: %" PetscInt_FMT " }\n",ctx->N,ctx->L,ctx->M,ctx->npart,ctx->L_max);
992:     if (ctx->isreal) PetscViewerASCIIPrintf(viewer,"  exploiting symmetry of integration points\n");
993:     PetscViewerASCIIPrintf(viewer,"  threshold { delta: %g, spurious threshold: %g }\n",(double)ctx->delta,(double)ctx->spurious_threshold);
994:     PetscViewerASCIIPrintf(viewer,"  iterative refinement  { inner: %" PetscInt_FMT ", blocksize: %" PetscInt_FMT " }\n",ctx->refine_inner, ctx->refine_blocksize);
995:     PetscViewerASCIIPrintf(viewer,"  extraction: %s\n",PEPCISSExtractions[ctx->extraction]);
996:     if (!ctx->contour || !ctx->contour->ksp) PEPCISSGetKSPs(pep,NULL,NULL);
997:     PetscViewerASCIIPushTab(viewer);
998:     if (ctx->npart>1 && ctx->contour->subcomm) {
999:       PetscViewerGetSubViewer(viewer,ctx->contour->subcomm->child,&sviewer);
1000:       if (!ctx->contour->subcomm->color) KSPView(ctx->contour->ksp[0],sviewer);
1001:       PetscViewerFlush(sviewer);
1002:       PetscViewerRestoreSubViewer(viewer,ctx->contour->subcomm->child,&sviewer);
1003:       PetscViewerFlush(viewer);
1004:       /* extra call needed because of the two calls to PetscViewerASCIIPushSynchronized() in PetscViewerGetSubViewer() */
1005:       PetscViewerASCIIPopSynchronized(viewer);
1006:     } else KSPView(ctx->contour->ksp[0],viewer);
1007:     PetscViewerASCIIPopTab(viewer);
1008:   }
1009:   return 0;
1010: }

1012: SLEPC_EXTERN PetscErrorCode PEPCreate_CISS(PEP pep)
1013: {
1014:   PEP_CISS       *ctx = (PEP_CISS*)pep->data;

1016:   PetscNew(&ctx);
1017:   pep->data = ctx;
1018:   /* set default values of parameters */
1019:   ctx->N                  = 32;
1020:   ctx->L                  = 16;
1021:   ctx->M                  = ctx->N/4;
1022:   ctx->delta              = SLEPC_DEFAULT_TOL*1e-4;
1023:   ctx->L_max              = 64;
1024:   ctx->spurious_threshold = PetscSqrtReal(SLEPC_DEFAULT_TOL);
1025:   ctx->isreal             = PETSC_FALSE;
1026:   ctx->npart              = 1;

1028:   pep->ops->solve          = PEPSolve_CISS;
1029:   pep->ops->setup          = PEPSetUp_CISS;
1030:   pep->ops->setfromoptions = PEPSetFromOptions_CISS;
1031:   pep->ops->reset          = PEPReset_CISS;
1032:   pep->ops->destroy        = PEPDestroy_CISS;
1033:   pep->ops->view           = PEPView_CISS;

1035:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSSetSizes_C",PEPCISSSetSizes_CISS);
1036:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetSizes_C",PEPCISSGetSizes_CISS);
1037:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSSetThreshold_C",PEPCISSSetThreshold_CISS);
1038:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetThreshold_C",PEPCISSGetThreshold_CISS);
1039:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSSetRefinement_C",PEPCISSSetRefinement_CISS);
1040:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetRefinement_C",PEPCISSGetRefinement_CISS);
1041:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSSetExtraction_C",PEPCISSSetExtraction_CISS);
1042:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetExtraction_C",PEPCISSGetExtraction_CISS);
1043:   PetscObjectComposeFunction((PetscObject)pep,"PEPCISSGetKSPs_C",PEPCISSGetKSPs_CISS);
1044:   return 0;
1045: }