Actual source code: pjd.c

slepc-3.11.2 2019-07-30
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:    SLEPc polynomial eigensolver: "jd"

 13:    Method: Jacobi-Davidson

 15:    Algorithm:

 17:        Jacobi-Davidson for polynomial eigenvalue problems.
 18:        Based on code contributed by the authors of [2] below.

 20:    References:

 22:        [1] G.L.G. Sleijpen et al., "Jacobi-Davidson type methods for
 23:            generalized eigenproblems and polynomial eigenproblems", BIT
 24:            36(3):595-633, 1996.

 26:        [2] Feng-Nan Hwang, Zih-Hao Wei, Tsung-Ming Huang, Weichung Wang,
 27:            "A Parallel Additive Schwarz Preconditioned Jacobi-Davidson
 28:            Algorithm for Polynomial Eigenvalue Problems in Quantum Dot
 29:            Simulation", J. Comput. Phys. 229(8):2932-2947, 2010.
 30: */

 32: #include <slepc/private/pepimpl.h>    /*I "slepcpep.h" I*/
 33: #include <slepcblaslapack.h>

 35: typedef struct {
 36:   PetscReal   keep;          /* restart parameter */
 37:   PetscReal   fix;           /* fix parameter */
 38:   PetscBool   reusepc;       /* flag indicating whether pc is rebuilt or not */
 39:   BV          V;             /* work basis vectors to store the search space */
 40:   BV          W;             /* work basis vectors to store the test space */
 41:   BV          *TV;           /* work basis vectors to store T*V (each TV[i] is the coefficient for \lambda^i of T*V for the extended T) */
 42:   BV          *AX;           /* work basis vectors to store A_i*X for locked eigenvectors */
 43:   BV          N[2];          /* auxiliary work BVs */
 44:   BV          X;             /* locked eigenvectors */
 45:   PetscScalar *T;            /* matrix of the invariant pair */
 46:   PetscScalar *Tj;           /* matrix containing the powers of the invariant pair matrix */
 47:   PetscScalar *XpX;          /* X^H*X */
 48:   PetscInt    ld;            /* leading dimension for Tj and XpX */
 49:   PC          pcshell;       /* preconditioner including basic precond+projector */
 50:   Mat         Pshell;        /* auxiliary shell matrix */
 51:   PetscInt    nlock;         /* number of locked vectors in the invariant pair */
 52:   Vec         vtempl;        /* reference nested vector */
 53:   PetscInt    midx;          /* minimality index */
 54:   PetscInt    mmidx;         /* maximum allowed minimality index */
 55:   PEPJDProjection proj;      /* projection type (orthogonal, harmonic) */
 56: } PEP_JD;

 58: typedef struct {
 59:   PEP         pep;
 60:   PC          pc;            /* basic preconditioner */
 61:   Vec         Bp[2];         /* preconditioned residual of derivative polynomial, B\p */
 62:   Vec         u[2];          /* Ritz vector */
 63:   PetscScalar gamma[2];      /* precomputed scalar u'*B\p */
 64:   PetscScalar theta;
 65:   PetscScalar *M;
 66:   PetscScalar *ps;
 67:   PetscInt    ld;
 68:   Vec         *work;
 69:   Mat         PPr;
 70:   BV          X;
 71:   PetscInt    n;
 72: } PEP_JD_PCSHELL;

 74: typedef struct {
 75:   Mat         Pr,Pi;         /* matrix polynomial evaluated at theta */
 76:   PEP         pep;
 77:   Vec         *work;
 78:   PetscScalar theta[2];
 79: } PEP_JD_MATSHELL;

 81: /*
 82:    Duplicate and resize auxiliary basis
 83: */
 84: static PetscErrorCode PEPJDDuplicateBasis(PEP pep,BV *basis)
 85: {
 86:   PetscErrorCode     ierr;
 87:   PEP_JD             *pjd = (PEP_JD*)pep->data;
 88:   PetscInt           nloc,m;
 89:   BVType             type;
 90:   BVOrthogType       otype;
 91:   BVOrthogRefineType oref;
 92:   PetscReal          oeta;
 93:   BVOrthogBlockType  oblock;

 96:   if (pjd->ld>1) {
 97:     BVCreate(PetscObjectComm((PetscObject)pep),basis);
 98:     BVGetSizes(pep->V,&nloc,NULL,&m);
 99:     nloc += pjd->ld-1;
100:     BVSetSizes(*basis,nloc,PETSC_DECIDE,m);
101:     BVGetType(pep->V,&type);
102:     BVSetType(*basis,type);
103:     BVGetOrthogonalization(pep->V,&otype,&oref,&oeta,&oblock);
104:     BVSetOrthogonalization(*basis,otype,oref,oeta,oblock);
105:     PetscObjectStateIncrease((PetscObject)*basis);
106:   } else {
107:     BVDuplicate(pep->V,basis);
108:   }
109:   return(0);
110: }

112: PetscErrorCode PEPSetUp_JD(PEP pep)
113: {
115:   PEP_JD         *pjd = (PEP_JD*)pep->data;
116:   PetscBool      isprecond,flg;
117:   PetscInt       i;

120:   pep->lineariz = PETSC_FALSE;
121:   PEPSetDimensions_Default(pep,pep->nev,&pep->ncv,&pep->mpd);
122:   if (!pep->max_it) pep->max_it = PetscMax(100,2*pep->n/pep->ncv);
123:   if (!pep->which) pep->which = PEP_TARGET_MAGNITUDE;
124:   if (pep->which!=PEP_TARGET_MAGNITUDE && pep->which!=PEP_TARGET_REAL && pep->which!=PEP_TARGET_IMAGINARY) SETERRQ(PetscObjectComm((PetscObject)pep),PETSC_ERR_SUP,"Wrong value of pep->which");

126:   PetscObjectTypeCompare((PetscObject)pep->st,STPRECOND,&isprecond);
127:   if (!isprecond) SETERRQ(PetscObjectComm((PetscObject)pep),PETSC_ERR_SUP,"JD only works with PRECOND spectral transformation");

129:   STGetTransform(pep->st,&flg);
130:   if (flg) SETERRQ(PetscObjectComm((PetscObject)pep),PETSC_ERR_SUP,"Solver requires the ST transformation flag unset, see STSetTransform()");

132:   if (!pjd->mmidx) pjd->mmidx = pep->nmat-1;
133:   pjd->mmidx = PetscMin(pjd->mmidx,pep->nmat-1);
134:   if (!pjd->keep) pjd->keep = 0.5;
135:   PEPBasisCoefficients(pep,pep->pbc);
136:   PEPAllocateSolution(pep,0);
137:   PEPSetWorkVecs(pep,5);
138:   pjd->ld = pep->nev;
139: #if !defined (PETSC_USE_COMPLEX)
140:   pjd->ld++;
141: #endif
142:   PetscMalloc2(pep->nmat,&pjd->TV,pep->nmat,&pjd->AX);
143:   for (i=0;i<pep->nmat;i++) {
144:     PEPJDDuplicateBasis(pep,pjd->TV+i);
145:   }
146:   if (pjd->ld>1) {
147:     PEPJDDuplicateBasis(pep,&pjd->V);
148:     BVSetFromOptions(pjd->V);
149:     for (i=0;i<pep->nmat;i++) {
150:       BVDuplicateResize(pep->V,pjd->ld-1,pjd->AX+i);
151:     }
152:     BVDuplicateResize(pep->V,pjd->ld-1,pjd->N);
153:     BVDuplicateResize(pep->V,pjd->ld-1,pjd->N+1);
154:     pjd->X = pep->V;
155:     PetscCalloc3((pjd->ld)*(pjd->ld),&pjd->XpX,pep->ncv*pep->ncv,&pjd->T,pjd->ld*pjd->ld*pep->nmat,&pjd->Tj);
156:   } else pjd->V = pep->V;
157:   if (pjd->proj==PEP_JD_PROJECTION_HARMONIC) { PEPJDDuplicateBasis(pep,&pjd->W); }
158:   else pjd->W = pjd->V;
159:   DSSetType(pep->ds,DSPEP);
160:   DSPEPSetDegree(pep->ds,pep->nmat-1);
161:   if (pep->basis!=PEP_BASIS_MONOMIAL) {
162:     DSPEPSetCoefficients(pep->ds,pep->pbc);
163:   }
164:   DSAllocate(pep->ds,pep->ncv);
165:   return(0);
166: }

168: /*
169:    Updates columns (low to (high-1)) of TV[i]
170: */
171: static PetscErrorCode PEPJDUpdateTV(PEP pep,PetscInt low,PetscInt high,Vec *w)
172: {
174:   PEP_JD         *pjd = (PEP_JD*)pep->data;
175:   PetscInt       pp,col,i,nloc,nconv;
176:   Vec            v1,v2,t1,t2;
177:   PetscScalar    *array1,*array2,*x2,*xx,*N,*Np,*y2=NULL,zero=0.0,sone=1.0,*pT,fact,*psc;
178:   PetscReal      *cg,*ca,*cb;
179:   PetscMPIInt    rk,np;
180:   PetscBLASInt   n_,ld_,one=1;
181:   Mat            T;
182:   BV             pbv;

185:   ca = pep->pbc; cb = ca+pep->nmat; cg = cb + pep->nmat;
186:   nconv = pjd->nlock;
187:   PetscMalloc5(nconv,&x2,nconv,&xx,nconv*nconv,&pT,nconv*nconv,&N,nconv*nconv,&Np);
188:   MPI_Comm_rank(PetscObjectComm((PetscObject)pep),&rk);
189:   MPI_Comm_size(PetscObjectComm((PetscObject)pep),&np);
190:   BVGetSizes(pep->V,&nloc,NULL,NULL);
191:   t1 = w[0];
192:   t2 = w[1];
193:   PetscBLASIntCast(pjd->nlock,&n_);
194:   PetscBLASIntCast(pjd->ld,&ld_);
195:   if (nconv){
196:     for (i=0;i<nconv;i++) {
197:       PetscMemcpy(pT+i*nconv,pjd->T+i*pep->ncv,nconv*sizeof(PetscScalar));
198:     }
199:     MatCreateSeqDense(PETSC_COMM_SELF,nconv,nconv,pT,&T);
200:   }
201:   for (col=low;col<high;col++) {
202:     BVGetColumn(pjd->V,col,&v1);
203:     VecGetArray(v1,&array1);
204:     if (nconv>0) {
205:       for (i=0;i<nconv;i++) x2[i] = array1[nloc+i]* PetscSqrtReal(np);
206:     }
207:     VecPlaceArray(t1,array1);
208:     if (nconv) {
209:       BVSetActiveColumns(pjd->N[0],0,nconv);
210:       BVSetActiveColumns(pjd->N[1],0,nconv);
211:       BVDotVec(pjd->X,t1,xx);
212:     }
213:     for (pp=pep->nmat-1;pp>=0;pp--) {
214:       BVGetColumn(pjd->TV[pp],col,&v2);
215:       VecGetArray(v2,&array2);
216:       VecPlaceArray(t2,array2);
217:       MatMult(pep->A[pp],t1,t2);
218:       if (nconv) {
219:         if (pp<pep->nmat-3) {
220:           BVMult(pjd->N[0],1.0,-cg[pp+2],pjd->AX[pp+1],NULL);
221:           MatShift(T,-cb[pp+1]);
222:           BVMult(pjd->N[0],1.0/ca[pp],1.0/ca[pp],pjd->N[1],T);
223:           pbv = pjd->N[0]; pjd->N[0] = pjd->N[1]; pjd->N[1] = pbv;
224:           BVMultVec(pjd->N[1],1.0,1.0,t2,x2);
225:           MatShift(T,cb[pp+1]);
226:         } else if (pp==pep->nmat-3) {
227:           BVCopy(pjd->AX[pp+2],pjd->N[0]);
228:           BVScale(pjd->N[0],1/ca[pp+1]);
229:           BVCopy(pjd->AX[pp+1],pjd->N[1]);
230:           MatShift(T,-cb[pp+1]);
231:           BVMult(pjd->N[1],1.0/ca[pp],1.0/ca[pp],pjd->N[0],T);
232:           BVMultVec(pjd->N[1],1.0,1.0,t2,x2);
233:           MatShift(T,cb[pp+1]);
234:         } else if (pp==pep->nmat-2) {
235:           BVMultVec(pjd->AX[pp+1],1.0/ca[pp],1.0,t2,x2);
236:         }
237:         if (pp<pjd->midx) {
238:           y2 = array2+nloc;
239:           PetscStackCallBLAS("BLASgemv",BLASgemv_("C",&n_,&n_,&sone,pjd->Tj+pjd->ld*pjd->ld*pp,&ld_,xx,&one,&zero,y2,&one));
240:           if (pp<pjd->midx-2) {
241:             fact = -cg[pp+2];
242:             PetscStackCallBLAS("BLASgemm",BLASgemm_("C","N",&n_,&n_,&n_,&sone,pjd->Tj+(pp+1)*pjd->ld*pjd->ld,&ld_,pjd->XpX,&ld_,&fact,Np,&n_));
243:             fact = 1/ca[pp];
244:             MatShift(T,-cb[pp+1]);
245:             PetscStackCallBLAS("BLASgemm",BLASgemm_("N","N",&n_,&n_,&n_,&fact,N,&n_,pT,&n_,&fact,Np,&n_));
246:             MatShift(T,cb[pp+1]);
247:             psc = Np; Np = N; N = psc;
248:             PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n_,&n_,&sone,N,&n_,x2,&one,&sone,y2,&one));
249:           } else if (pp==pjd->midx-2) {
250:             fact = 1/ca[pp];
251:             PetscStackCallBLAS("BLASgemm",BLASgemm_("C","N",&n_,&n_,&n_,&fact,pjd->Tj+(pp+1)*pjd->ld*pjd->ld,&ld_,pjd->XpX,&ld_,&zero,N,&n_));
252:             PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n_,&n_,&sone,N,&n_,x2,&one,&sone,y2,&one));
253:           } else if (pp==pjd->midx-1) {
254:             PetscMemzero(Np,nconv*nconv*sizeof(PetscScalar));
255:           }
256:         }
257:         for (i=0;i<nconv;i++) array2[nloc+i] /= PetscSqrtReal(np);
258:       }
259:       VecResetArray(t2);
260:       VecRestoreArray(v2,&array2);
261:       BVRestoreColumn(pjd->TV[pp],col,&v2);
262:     }
263:     VecResetArray(t1);
264:     VecRestoreArray(v1,&array1);
265:     BVRestoreColumn(pjd->V,col,&v1);
266:   }
267:   if (nconv) {MatDestroy(&T);}
268:   PetscFree5(x2,xx,pT,N,Np);
269:   return(0);
270: }

272: /*
273:    RRQR of X. Xin*P=Xou*R. Rank of R is rk
274: */
275: static PetscErrorCode PEPJDOrthogonalize(PetscInt row,PetscInt col,PetscScalar *X,PetscInt ldx,PetscInt *rk,PetscInt *P,PetscScalar *R,PetscInt ldr)
276: {
277: #if defined(SLEPC_MISSING_LAPACK_GEQP3) || defined(PETSC_MISSING_LAPACK_ORGQR)
279:   SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"GEQP3/QRGQR - Lapack routines are unavailable");
280: #else
282:   PetscInt       i,j,n,r;
283:   PetscBLASInt   row_,col_,ldx_,*p,lwork,info,n_;
284:   PetscScalar    *tau,*work;
285:   PetscReal      tol,*rwork;

288:   PetscBLASIntCast(row,&row_);
289:   PetscBLASIntCast(col,&col_);
290:   PetscBLASIntCast(ldx,&ldx_);
291:   n = PetscMin(row,col);
292:   PetscBLASIntCast(n,&n_);
293:   lwork = 3*col_+1;
294:   PetscMalloc4(col,&p,n,&tau,lwork,&work,2*col,&rwork);
295:   for (i=1;i<col;i++) p[i] = 0;
296:   p[0] = 1;

298:   /* rank revealing QR */
299: #if defined(PETSC_USE_COMPLEX)
300:   PetscStackCallBLAS("LAPACKgeqp3",LAPACKgeqp3_(&row_,&col_,X,&ldx_,p,tau,work,&lwork,rwork,&info));
301: #else
302:   PetscStackCallBLAS("LAPACKgeqp3",LAPACKgeqp3_(&row_,&col_,X,&ldx_,p,tau,work,&lwork,&info));
303: #endif
304:   SlepcCheckLapackInfo("geqp3",info);
305:   if (P) for (i=0;i<col;i++) P[i] = p[i]-1;

307:   /* rank computation */
308:   tol = PetscMax(row,col)*PETSC_MACHINE_EPSILON*PetscAbsScalar(X[0]);
309:   r = 1;
310:   for (i=1;i<n;i++) {
311:     if (PetscAbsScalar(X[i+ldx*i])>tol) r++;
312:     else break;
313:   }
314:   if (rk) *rk=r;

316:   /* copy upper triangular matrix if requested */
317:   if (R) {
318:      for (i=0;i<r;i++) {
319:        PetscMemzero(R+i*ldr,r*sizeof(PetscScalar));
320:        for (j=0;j<=i;j++) R[i*ldr+j] = X[i*ldx+j];
321:      }
322:   }
323:   PetscStackCallBLAS("LAPACKorgqr",LAPACKorgqr_(&row_,&n_,&n_,X,&ldx_,tau,work,&lwork,&info));
324:   SlepcCheckLapackInfo("orgqr",info);
325:   PetscFree4(p,tau,work,rwork);
326:   return(0);
327: #endif
328: }

330: /*
331:    Application of extended preconditioner
332: */
333: static PetscErrorCode PEPJDExtendedPCApply(PC pc,Vec x,Vec y)
334: {
335:   PetscInt          i,j,nloc,n,ld=0;
336:   PetscMPIInt       np;
337:   Vec               tx,ty;
338:   PEP_JD_PCSHELL    *ctx;
339:   PetscErrorCode    ierr;
340:   const PetscScalar *array1;
341:   PetscScalar       *x2=NULL,*t=NULL,*ps=NULL,*array2,zero=0.0,sone=1.0;
342:   PetscBLASInt      one=1.0,ld_,n_,ncv_;
343:   PEP_JD            *pjd=NULL;

346:   MPI_Comm_size(PetscObjectComm((PetscObject)pc),&np);
347:   PCShellGetContext(pc,(void**)&ctx);
348:   n  = ctx->n;
349:   if (n) {
350:     pjd = (PEP_JD*)ctx->pep->data;
351:     ps = ctx->ps;
352:     ld = pjd->ld;
353:     PetscMalloc2(n,&x2,n,&t);
354:     VecGetLocalSize(ctx->work[0],&nloc);
355:     VecGetArrayRead(x,&array1);
356:     for (i=0;i<n;i++) x2[i] = array1[nloc+i]* PetscSqrtReal(np);
357:     VecRestoreArrayRead(x,&array1);
358:   }

360:   /* y = B\x apply PC */
361:   tx = ctx->work[0];
362:   ty = ctx->work[1];
363:   VecGetArrayRead(x,&array1);
364:   VecPlaceArray(tx,array1);
365:   VecGetArray(y,&array2);
366:   VecPlaceArray(ty,array2);
367:   PCApply(ctx->pc,tx,ty);
368:   if (n) {
369:     PetscBLASIntCast(ld,&ld_);
370:     PetscBLASIntCast(n,&n_);
371:     for (i=0;i<n;i++) {
372:       t[i] = 0.0;
373:       for (j=0;j<n;j++) t[i] += ctx->M[i+j*ld]*x2[j];
374:     }
375:     if (pjd->midx==1) {
376:       PetscBLASIntCast(ctx->pep->ncv,&ncv_);
377:       for (i=0;i<n;i++) pjd->T[i*(1+ctx->pep->ncv)] -= ctx->theta;
378:       PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n_,&n_,&sone,pjd->T,&ncv_,t,&one,&zero,x2,&one));
379:       for (i=0;i<n;i++) pjd->T[i*(1+ctx->pep->ncv)] += ctx->theta;
380:       for (i=0;i<n;i++) array2[nloc+i] = x2[i];
381:       for (i=0;i<n;i++) x2[i] = -t[i];
382:     } else {
383:       for (i=0;i<n;i++) array2[nloc+i] = t[i];
384:       PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n_,&n_,&sone,ps,&ld_,t,&one,&zero,x2,&one));
385:     }
386:     for (i=0;i<n;i++) array2[nloc+i] /= PetscSqrtReal(np);
387:     BVSetActiveColumns(pjd->X,0,n);
388:     BVMultVec(pjd->X,-1.0,1.0,ty,x2);
389:     PetscFree2(x2,t);
390:   }
391:   VecResetArray(tx);
392:   VecResetArray(ty);
393:   VecRestoreArrayRead(x,&array1);
394:   VecRestoreArray(y,&array2);
395:   return(0);
396: }

398: /*
399:    Application of shell preconditioner:
400:       y = B\x - eta*B\p,  with eta = (u'*B\x)/(u'*B\p)
401: */
402: static PetscErrorCode PCShellApply_PEPJD(PC pc,Vec x,Vec y)
403: {
405:   PetscScalar    rr,rx,xr,xx,eta;
406:   PEP_JD_PCSHELL *ctx;
407:   PetscInt       sz;
408:   const Vec      *xs,*ys;

411:   PCShellGetContext(pc,(void**)&ctx);
412:   VecCompGetSubVecs(x,&sz,&xs);
413:   VecCompGetSubVecs(y,NULL,&ys);
414:   /* y = B\x apply extended PC */
415:   PEPJDExtendedPCApply(pc,xs[0],ys[0]);
416:   if (sz==2) {
417:     PEPJDExtendedPCApply(pc,xs[1],ys[1]);
418:   }

420:   /* Compute eta = u'*y / u'*Bp */
421:   VecDot(ys[0],ctx->u[0],&rr);
422:   eta  = -rr*ctx->gamma[0];

424:   if (sz==2) {
425:     VecDot(ys[0],ctx->u[1],&xr);
426:     VecDot(ys[1],ctx->u[0],&rx);
427:     VecDot(ys[1],ctx->u[1],&xx);
428:     eta += -ctx->gamma[0]*xx-ctx->gamma[1]*(-xr+rx);
429:   }
430:   eta /= ctx->gamma[0]*ctx->gamma[0]+ctx->gamma[1]*ctx->gamma[1];

432:   /* y = y - eta*Bp */
433:   VecAXPY(ys[0],eta,ctx->Bp[0]);
434:   if (sz==2) {
435:     VecAXPY(ys[1],eta,ctx->Bp[1]);
436:     eta = -ctx->gamma[1]*(rr+xx)+ctx->gamma[0]*(-xr+rx);
437:     eta /= ctx->gamma[0]*ctx->gamma[0]+ctx->gamma[1]*ctx->gamma[1];
438:     VecAXPY(ys[0],eta,ctx->Bp[1]);
439:     VecAXPY(ys[1],-eta,ctx->Bp[0]);
440:   }
441:   return(0);
442: }

444: static PetscErrorCode PEPJDCopyToExtendedVec(PEP pep,Vec v,PetscScalar *a,PetscInt na,PetscInt off,Vec vex,PetscBool back)
445: {
447:   PetscMPIInt    np,rk,count;
448:   PetscScalar    *array1,*array2;
449:   PetscInt       nloc;

452:   MPI_Comm_rank(PetscObjectComm((PetscObject)pep),&rk);
453:   MPI_Comm_size(PetscObjectComm((PetscObject)pep),&np);
454:   BVGetSizes(pep->V,&nloc,NULL,NULL);
455:   if (v) {
456:     VecGetArray(v,&array1);
457:     VecGetArray(vex,&array2);
458:     if (back) {
459:       PetscMemcpy(array1,array2,nloc*sizeof(PetscScalar));
460:     } else {
461:       PetscMemcpy(array2,array1,nloc*sizeof(PetscScalar));
462:     }
463:     VecRestoreArray(v,&array1);
464:     VecRestoreArray(vex,&array2);
465:   }
466:   if (a) {
467:     VecGetArray(vex,&array2);
468:     if (back) {
469:       PetscMemcpy(a,array2+nloc+off,na*sizeof(PetscScalar));
470:       PetscMPIIntCast(na,&count);
471:       MPI_Bcast(a,count,MPIU_SCALAR,np-1,PetscObjectComm((PetscObject)pep));
472:     } else {
473:       PetscMemcpy(array2+nloc+off,a,na*sizeof(PetscScalar));
474:       PetscMPIIntCast(na,&count);
475:       MPI_Bcast(array2+nloc+off,count,MPIU_SCALAR,np-1,PetscObjectComm((PetscObject)pep));
476:     }
477:     VecRestoreArray(vex,&array2);
478:   }
479:   return(0);
480: }

482: /* Computes Phi^hat(lambda) times a vector or its derivative (depends on beval)
483:      if no vector is provided returns a matrix
484:  */
485: static PetscErrorCode PEPJDEvaluateHatBasis(PEP pep,PetscInt n,PetscScalar *H,PetscInt ldh,PetscScalar *beval,PetscScalar *t,PetscInt idx,PetscScalar *qpp,PetscScalar *qp,PetscScalar *q)
486: {
488:   PetscInt       j,i;
489:   PetscBLASInt   n_,ldh_,one=1;
490:   PetscReal      *a,*b,*g;
491:   PetscScalar    sone=1.0,zero=0.0;

494:   a = pep->pbc; b=a+pep->nmat; g=b+pep->nmat;
495:   PetscBLASIntCast(n,&n_);
496:   PetscBLASIntCast(ldh,&ldh_);
497:   if (idx<1) {
498:     PetscMemzero(q,(t?n:n*n)*sizeof(PetscScalar));
499:   } else if (idx==1) {
500:     if (t) {for (j=0;j<n;j++) q[j] = t[j]*beval[idx-1]/a[0];}
501:     else {
502:       PetscMemzero(q,n*n*sizeof(PetscScalar));
503:       for (j=0;j<n;j++) q[(j+1)*n] = beval[idx-1]/a[0];
504:     }
505:   } else {
506:     if (t) {
507:       PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n_,&n_,&sone,H,&ldh_,qp,&one,&zero,q,&one));
508:       for (j=0;j<n;j++) {
509:         q[j] += beval[idx-1]*t[j]-b[idx-1]*qp[j]-g[idx-1]*qpp[j];
510:         q[j] /= a[idx-1];
511:       }
512:     } else {
513:       PetscStackCallBLAS("BLASgemm",BLASgemm_("N","N",&n_,&n_,&n_,&sone,H,&ldh_,qp,&n_,&zero,q,&n_));
514:       for (j=0;j<n;j++) {
515:         q[j+n*j] += beval[idx-1];
516:         for (i=0;i<n;i++) {
517:           q[i+n*j] += -b[idx-1]*qp[j*n+i]-g[idx-1]*qpp[j*n+i];
518:           q[i+n*j] /= a[idx-1];
519:         }
520:       }
521:     }
522:   }
523:   return(0);
524: }

526: static PetscErrorCode PEPJDComputeResidual(PEP pep,PetscBool derivative,PetscInt sz,Vec *u,PetscScalar *theta,Vec *p,Vec *work)
527: {
528:   PEP_JD         *pjd = (PEP_JD*)pep->data;
530:   PetscMPIInt    rk,np,count;
531:   Vec            tu,tui=NULL,tp,tpi=NULL,w;
532:   PetscScalar    *dval,*dvali,*array1,*array2,*arrayi1,*arrayi2;
533:   PetscScalar    *x2=NULL,*x2i=NULL,*y2,*y2i,*qj=NULL,*qji=NULL,*qq,*tt=NULL,*xx=NULL,*xxi=NULL,sone=1.0;
534:   PetscInt       i,j,nconv,nloc;
535:   PetscBLASInt   n,ld,one=1;

538:   nconv = pjd->nlock;
539:   if (!nconv) {
540:     PetscMalloc1(2*sz*pep->nmat,&dval);
541:   } else {
542:     PetscMalloc5(2*pep->nmat,&dval,2*nconv,&xx,nconv,&tt,sz*nconv,&x2,(sz==2?3:1)*nconv*pep->nmat,&qj);
543:     if (sz==2) x2i = x2+nconv;
544:     MPI_Comm_rank(PetscObjectComm((PetscObject)pep),&rk);
545:     MPI_Comm_size(PetscObjectComm((PetscObject)pep),&np);
546:     BVGetSizes(pep->V,&nloc,NULL,NULL);
547:     VecGetArray(u[0],&array1);
548:     for (i=0;i<nconv;i++) x2[i] = array1[nloc+i]* PetscSqrtReal(np);
549:     VecRestoreArray(u[0],&array1);
550:     if (sz==2) {
551:       VecGetArray(u[1],&arrayi1);
552:       for (i=0;i<nconv;i++) x2i[i] = arrayi1[nloc+i]* PetscSqrtReal(np);
553:       VecRestoreArray(u[1],&arrayi1);
554:     }
555:   }
556:   dvali = dval+pep->nmat;
557:   tu = work[0];
558:   tp = work[1];
559:   w  = work[2];
560:   VecGetArray(u[0],&array1);
561:   VecPlaceArray(tu,array1);
562:   VecGetArray(p[0],&array2);
563:   VecPlaceArray(tp,array2);
564:   VecSet(tp,0.0);
565:   if (sz==2) {
566:     tui = work[3];
567:     tpi = work[4];
568:     VecGetArray(u[1],&arrayi1);
569:     VecPlaceArray(tui,arrayi1);
570:     VecGetArray(p[1],&arrayi2);
571:     VecPlaceArray(tpi,arrayi2);
572:     VecSet(tpi,0.0);
573:   }
574:   if (derivative) {
575:     PEPEvaluateBasisDerivative(pep,theta[0],theta[1],dval,dvali);
576:   } else {
577:     PEPEvaluateBasis(pep,theta[0],theta[1],dval,dvali);
578:   }
579:   for (i=derivative?1:0;i<pep->nmat;i++) {
580:     MatMult(pep->A[i],tu,w);
581:     VecAXPY(tp,dval[i],w);
582:     if (sz==2) {
583:       VecAXPY(tpi,dvali[i],w);
584:       MatMult(pep->A[i],tui,w);
585:       VecAXPY(tpi,dval[i],w);
586:       VecAXPY(tp,-dvali[i],w);
587:     }
588:   }
589:   if (nconv) {
590:     for (i=0;i<pep->nmat;i++) {
591:       PEPJDEvaluateHatBasis(pep,nconv,pjd->T,pep->ncv,dval,x2,i,i>1?qj+(i-2)*nconv:NULL,i>0?qj+(i-1)*nconv:NULL,qj+i*nconv);
592:     }
593:     if (sz==2) {
594:       qji = qj+nconv*pep->nmat;
595:       qq = qji+nconv*pep->nmat;
596:       for (i=0;i<pep->nmat;i++) {
597:         PEPJDEvaluateHatBasis(pep,nconv,pjd->T,pep->ncv,dvali,x2i,i,i>1?qji+(i-2)*nconv:NULL,i>0?qji+(i-1)*nconv:NULL,qji+i*nconv);
598:       }
599:       for (i=0;i<nconv*pep->nmat;i++) qj[i] -= qji[i];
600:       for (i=0;i<pep->nmat;i++) {
601:         PEPJDEvaluateHatBasis(pep,nconv,pjd->T,pep->ncv,dval,x2i,i,i>1?qji+(i-2)*nconv:NULL,i>0?qji+(i-1)*nconv:NULL,qji+i*nconv);
602:         PEPJDEvaluateHatBasis(pep,nconv,pjd->T,pep->ncv,dvali,x2,i,i>1?qq+(i-2)*nconv:NULL,i>0?qq+(i-1)*nconv:NULL,qq+i*nconv);
603:       }
604:       for (i=0;i<nconv*pep->nmat;i++) qji[i] += qq[i];
605:       for (i=derivative?2:1;i<pep->nmat;i++) {
606:         BVMultVec(pjd->AX[i],1.0,1.0,tpi,qji+i*nconv);
607:       }
608:     }
609:     for (i=derivative?2:1;i<pep->nmat;i++) {
610:       BVMultVec(pjd->AX[i],1.0,1.0,tp,qj+i*nconv);
611:     }

613:     /* extended vector part */
614:     BVSetActiveColumns(pjd->X,0,nconv);
615:     BVDotVec(pjd->X,tu,xx);
616:     xxi = xx+nconv;
617:     if (sz==2) {
618:       BVDotVec(pjd->X,tui,xxi);
619:     } else {
620:       PetscMemzero(xxi,nconv*sizeof(PetscScalar));
621:     }
622:     if (rk==np-1) {
623:       PetscBLASIntCast(nconv,&n);
624:       PetscBLASIntCast(pjd->ld,&ld);
625:       y2  = array2+nloc;
626:       PetscMemzero(y2,nconv*sizeof(PetscScalar));
627:       for (j=derivative?1:0;j<pjd->midx;j++) {
628:         for (i=0;i<nconv;i++) tt[i] = dval[j]*xx[i]-dvali[j]*xxi[i];
629:         PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n,&n,&sone,pjd->XpX,&ld,qj+j*nconv,&one,&sone,tt,&one));
630:         PetscStackCallBLAS("BLASgemv",BLASgemv_("C",&n,&n,&sone,pjd->Tj+j*ld*ld,&ld,tt,&one,&sone,y2,&one));
631:       }
632:       for (i=0;i<nconv;i++) array2[nloc+i] /= PetscSqrtReal(np);
633:       if (sz==2) {
634:         y2i = arrayi2+nloc;
635:         PetscMemzero(y2i,nconv*sizeof(PetscScalar));
636:         for (j=derivative?1:0;j<pjd->midx;j++) {
637:           for (i=0;i<nconv;i++) tt[i] = dval[j]*xxi[i]+dvali[j]*xx[i];
638:           PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n,&n,&sone,pjd->XpX,&ld,qji+j*nconv,&one,&sone,tt,&one));
639:           PetscStackCallBLAS("BLASgemv",BLASgemv_("C",&n,&n,&sone,pjd->Tj+j*ld*ld,&ld,tt,&one,&sone,y2i,&one));
640:         }
641:         for (i=0;i<nconv;i++) arrayi2[nloc+i] /= PetscSqrtReal(np);
642:       }
643:     }
644:     PetscMPIIntCast(nconv,&count);
645:     MPI_Bcast(array2+nloc,count,MPIU_SCALAR,np-1,PetscObjectComm((PetscObject)pep));
646:     if (sz==2) {MPI_Bcast(arrayi2+nloc,count,MPIU_SCALAR,np-1,PetscObjectComm((PetscObject)pep));}
647:   }
648:   if (nconv) {
649:     PetscFree5(dval,xx,tt,x2,qj);
650:   } else {
651:     PetscFree(dval);
652:   }
653:   VecResetArray(tu);
654:   VecRestoreArray(u[0],&array1);
655:   VecResetArray(tp);
656:   VecRestoreArray(p[0],&array2);
657:   if (sz==2) {
658:     VecResetArray(tui);
659:     VecRestoreArray(u[1],&arrayi1);
660:     VecResetArray(tpi);
661:     VecRestoreArray(p[1],&arrayi2);
662:   }
663:   return(0);
664: }

666: static PetscErrorCode PEPJDProcessInitialSpace(PEP pep,Vec *w)
667: {
668:   PEP_JD         *pjd = (PEP_JD*)pep->data;
670:   PetscScalar    *tt,target[2];
671:   Vec            vg,wg;
672:   PetscInt       i;
673:   PetscReal      norm;

676:   PetscMalloc1(pjd->ld-1,&tt);
677:   if (pep->nini==0) {
678:     BVSetRandomColumn(pjd->V,0);
679:     for (i=0;i<pjd->ld-1;i++) tt[i] = 0.0;
680:     BVGetColumn(pjd->V,0,&vg);
681:     PEPJDCopyToExtendedVec(pep,NULL,tt,pjd->ld-1,0,vg,PETSC_FALSE);
682:     BVRestoreColumn(pjd->V,0,&vg);
683:     BVNormColumn(pjd->V,0,NORM_2,&norm);
684:     BVScaleColumn(pjd->V,0,1.0/norm);
685:     if (pjd->proj==PEP_JD_PROJECTION_HARMONIC) {
686:       BVGetColumn(pjd->V,0,&vg);
687:       BVGetColumn(pjd->W,0,&wg);
688:       VecSet(wg,0.0);
689:       target[0] = pep->target; target[1] = 0.0;
690:       PEPJDComputeResidual(pep,PETSC_TRUE,1,&vg,target,&wg,w);
691:       BVRestoreColumn(pjd->W,0,&wg);
692:       BVRestoreColumn(pjd->V,0,&vg);
693:       BVNormColumn(pjd->W,0,NORM_2,&norm);
694:       BVScaleColumn(pjd->W,0,1.0/norm);
695:     }
696:   } else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Support for initial vectors not implemented yet");
697:   PetscFree(tt);
698:   return(0);
699: }

701: static PetscErrorCode PEPJDShellMatMult(Mat P,Vec x,Vec y)
702: {
703:   PetscErrorCode    ierr;
704:   PEP_JD_MATSHELL   *matctx;
705:   PEP_JD            *pjd;
706:   PetscInt          i,j,nconv,nloc,nmat,ldt,ncv,sz;
707:   Vec               tx,ty,txi=NULL,tyi=NULL;
708:   const Vec         *xs,*ys;
709:   PetscScalar       *array2,*arrayi1,*arrayi2,*array1;
710:   PetscScalar       *x2=NULL,*x2i=NULL,*y2,*y2i,*tt=NULL,*xx=NULL,*xxi,theta[2],sone=1.0,*qj,*qji=NULL,*qq,*val,*vali=NULL;
711:   PetscBLASInt      n,ld,one=1;
712:   PetscMPIInt       np;

715:   MPI_Comm_size(PetscObjectComm((PetscObject)P),&np);
716:   MatShellGetContext(P,(void**)&matctx);
717:   pjd   = (PEP_JD*)(matctx->pep->data);
718:   nconv = pjd->nlock;
719:   nmat  = matctx->pep->nmat;
720:   ncv   = matctx->pep->ncv;
721:   ldt   = pjd->ld;
722:   VecCompGetSubVecs(x,&sz,&xs);
723:   VecCompGetSubVecs(y,NULL,&ys);
724:   theta[0] = matctx->theta[0];
725:   theta[1] = (sz==2)?matctx->theta[1]:0.0;
726:   if (nconv>0) {
727:     PetscMalloc5(nconv,&tt,sz*nconv,&x2,(sz==2?3:1)*nconv*nmat,&qj,2*nconv,&xx,2*nmat,&val);
728:     if (sz==2)  x2i = x2+nconv;
729:     BVGetSizes(matctx->pep->V,&nloc,NULL,NULL);
730:     VecGetArray(xs[0],&array1);
731:     for (i=0;i<nconv;i++) x2[i] = array1[nloc+i]* PetscSqrtReal(np);
732:     VecRestoreArray(xs[0],&array1);
733:     if (sz==2) {
734:       VecGetArray(xs[1],&arrayi1);
735:       for (i=0;i<nconv;i++) x2i[i] = arrayi1[nloc+i]* PetscSqrtReal(np);
736:       VecRestoreArray(xs[1],&arrayi1);
737:     }
738:     vali = val+nmat;
739:   }
740:   tx = matctx->work[0];
741:   ty = matctx->work[1];
742:   VecGetArray(xs[0],&array1);
743:   VecPlaceArray(tx,array1);
744:   VecGetArray(ys[0],&array2);
745:   VecPlaceArray(ty,array2);
746:   MatMult(matctx->Pr,tx,ty);
747:   if (sz==2) {
748:     txi = matctx->work[2];
749:     tyi = matctx->work[3];
750:     VecGetArray(xs[1],&arrayi1);
751:     VecPlaceArray(txi,arrayi1);
752:     VecGetArray(ys[1],&arrayi2);
753:     VecPlaceArray(tyi,arrayi2);
754:     MatMult(matctx->Pr,txi,tyi);
755:     if (theta[1]!=0.0) {
756:       MatMult(matctx->Pi,txi,matctx->work[4]);
757:       VecAXPY(ty,-1.0,matctx->work[4]);
758:       MatMult(matctx->Pi,tx,matctx->work[4]);
759:       VecAXPY(tyi,1.0,matctx->work[4]);
760:     }
761:   }
762:   if (nconv) {
763:     PEPEvaluateBasis(matctx->pep,theta[0],theta[1],val,vali);
764:     for (i=0;i<nmat;i++) {
765:       PEPJDEvaluateHatBasis(matctx->pep,nconv,pjd->T,ncv,val,x2,i,i>1?qj+(i-2)*nconv:NULL,i>0?qj+(i-1)*nconv:NULL,qj+i*nconv);
766:     }
767:     if (sz==2) {
768:       qji = qj+nconv*nmat;
769:       qq = qji+nconv*nmat;
770:       for (i=0;i<nmat;i++) {
771:         PEPJDEvaluateHatBasis(matctx->pep,nconv,pjd->T,matctx->pep->ncv,vali,x2i,i,i>1?qji+(i-2)*nconv:NULL,i>0?qji+(i-1)*nconv:NULL,qji+i*nconv);
772:       }
773:       for (i=0;i<nconv*nmat;i++) qj[i] -= qji[i];
774:       for (i=0;i<nmat;i++) {
775:         PEPJDEvaluateHatBasis(matctx->pep,nconv,pjd->T,matctx->pep->ncv,val,x2i,i,i>1?qji+(i-2)*nconv:NULL,i>0?qji+(i-1)*nconv:NULL,qji+i*nconv);
776:         PEPJDEvaluateHatBasis(matctx->pep,nconv,pjd->T,matctx->pep->ncv,vali,x2,i,i>1?qq+(i-2)*nconv:NULL,i>0?qq+(i-1)*nconv:NULL,qq+i*nconv);
777:       }
778:       for (i=0;i<nconv*nmat;i++) qji[i] += qq[i];
779:       for (i=1;i<matctx->pep->nmat;i++) {
780:         BVMultVec(pjd->AX[i],1.0,1.0,tyi,qji+i*nconv);
781:       }
782:     }
783:     for (i=1;i<nmat;i++) {
784:       BVMultVec(pjd->AX[i],1.0,1.0,ty,qj+i*nconv);
785:     }

787:     /* extended vector part */
788:     BVSetActiveColumns(pjd->X,0,nconv);
789:     BVDotVec(pjd->X,tx,xx);
790:     xxi = xx+nconv;
791:     if (sz==2) {
792:       BVDotVec(pjd->X,txi,xxi);
793:     } else {
794:       PetscMemzero(xxi,nconv*sizeof(PetscScalar));
795:     }
796:       PetscBLASIntCast(pjd->nlock,&n);
797:       PetscBLASIntCast(ldt,&ld);
798:       y2 = array2+nloc;
799:       PetscMemzero(y2,nconv*sizeof(PetscScalar));
800:       for (j=0;j<pjd->midx;j++) {
801:         for (i=0;i<nconv;i++) tt[i] = val[j]*xx[i]-vali[j]*xxi[i];
802:         PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n,&n,&sone,pjd->XpX,&ld,qj+j*nconv,&one,&sone,tt,&one));
803:         PetscStackCallBLAS("BLASgemv",BLASgemv_("C",&n,&n,&sone,pjd->Tj+j*ld*ld,&ld,tt,&one,&sone,y2,&one));
804:       }
805:       if (sz==2) {
806:         y2i = arrayi2+nloc;
807:         PetscMemzero(y2i,nconv*sizeof(PetscScalar));
808:         for (j=0;j<pjd->midx;j++) {
809:           for (i=0;i<nconv;i++) tt[i] = val[j]*xxi[i]+vali[j]*xx[i];
810:           PetscStackCallBLAS("BLASgemv",BLASgemv_("N",&n,&n,&sone,pjd->XpX,&ld,qji+j*nconv,&one,&sone,tt,&one));
811:           PetscStackCallBLAS("BLASgemv",BLASgemv_("C",&n,&n,&sone,pjd->Tj+j*ld*ld,&ld,tt,&one,&sone,y2i,&one));
812:         }
813:       }
814:     for (i=0;i<nconv;i++) array2[nloc+i] /= PetscSqrtReal(np);
815:     PetscFree5(tt,x2,qj,xx,val);
816:     if (sz==2) {
817:       for (i=0;i<nconv;i++) arrayi2[nloc+i] /= PetscSqrtReal(np);
818:     }
819:   }
820:   VecResetArray(tx);
821:   VecRestoreArray(xs[0],&array1);
822:   VecResetArray(ty);
823:   VecRestoreArray(ys[0],&array2);
824:   if (sz==2) {
825:     VecResetArray(txi);
826:     VecRestoreArray(xs[1],&arrayi1);
827:     VecResetArray(tyi);
828:     VecRestoreArray(ys[1],&arrayi2);
829:   }
830:   return(0);
831: }

833: static PetscErrorCode PEPJDSellMatCreateVecs(Mat A,Vec *right,Vec *left)
834: {
835:   PetscErrorCode  ierr;
836:   PEP_JD_MATSHELL *matctx;
837:   PEP_JD          *pjd;
838:   PetscInt        kspsf=1,i;
839:   Vec             v[2];

842:   MatShellGetContext(A,(void**)&matctx);
843:   pjd   = (PEP_JD*)(matctx->pep->data);
844: #if !defined (PETSC_USE_COMPLEX)
845:   kspsf = 2;
846: #endif
847:   for (i=0;i<kspsf;i++){
848:     BVCreateVec(pjd->V,v+i);
849:   }
850:   if (right) {
851:     VecCreateCompWithVecs(v,kspsf,pjd->vtempl,right);
852:   }
853:   if (left) {
854:     VecCreateCompWithVecs(v,kspsf,pjd->vtempl,left);
855:   }
856:   for (i=0;i<kspsf;i++) {
857:     VecDestroy(&v[i]);
858:   }
859:   return(0);
860: }

862: static PetscErrorCode PEPJDUpdateExtendedPC(PEP pep,PetscScalar theta)
863: {
864: #if defined(PETSC_MISSING_LAPACK_GESVD) || defined(PETSC_MISSING_LAPACK_GETRI) || defined(PETSC_MISSING_LAPACK_GETRF)
866:   SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"GESVD/GETRI/GETRF - Lapack routines are unavailable");
867: #else
869:   PEP_JD         *pjd = (PEP_JD*)pep->data;
870:   PEP_JD_PCSHELL *pcctx;
871:   PetscInt       i,j,k,n=pjd->nlock,ld=pjd->ld,deg=pep->nmat-1;
872:   PetscScalar    *M,*ps,*work,*U,*V,*S,*Sp,*Spp,snone=-1.0,sone=1.0,zero=0.0,*val;
873:   PetscReal      tol,maxeig=0.0,*sg,*rwork;
874:   PetscBLASInt   n_,info,ld_,*p,lw_,rk=0;

877:   if (n) {
878:     PCShellGetContext(pjd->pcshell,(void**)&pcctx);
879:     pcctx->theta = theta;
880:     pcctx->n = n;
881:     M  = pcctx->M;
882:     PetscBLASIntCast(n,&n_);
883:     PetscBLASIntCast(ld,&ld_);
884:     if (pjd->midx==1) {
885:       PetscMemcpy(M,pjd->XpX,ld*ld*sizeof(PetscScalar));
886:       PetscCalloc2(10*n,&work,n,&p);
887:     } else {
888:       ps = pcctx->ps;
889:       PetscCalloc7(2*n*n,&U,3*n*n,&S,n,&sg,10*n,&work,5*n,&rwork,n,&p,deg+1,&val);
890:       V = U+n*n;
891:       /* pseudo-inverse */
892:       for (j=0;j<n;j++) {
893:         for (i=0;i<n;i++) S[n*j+i] = -pjd->T[pep->ncv*j+i];
894:         S[n*j+j] += theta;
895:       }
896:       lw_ = 10*n_;
897: #if !defined (PETSC_USE_COMPLEX)
898:       PetscStackCallBLAS("LAPACKgesvd",LAPACKgesvd_("S","S",&n_,&n_,S,&n_,sg,U,&n_,V,&n_,work,&lw_,&info));
899: #else
900:       PetscStackCallBLAS("LAPACKgesvd",LAPACKgesvd_("S","S",&n_,&n_,S,&n_,sg,U,&n_,V,&n_,work,&lw_,rwork,&info));
901: #endif
902:       SlepcCheckLapackInfo("gesvd",info);
903:       for (i=0;i<n;i++) maxeig = PetscMax(maxeig,sg[i]);
904:       tol = 10*PETSC_MACHINE_EPSILON*n*maxeig;
905:       for (j=0;j<n;j++) {
906:         if (sg[j]>tol) {
907:           for (i=0;i<n;i++) U[j*n+i] /= sg[j];
908:           rk++;
909:         } else break;
910:       }
911:       PetscStackCallBLAS("BLASgemm",BLASgemm_("N","N",&n_,&n_,&rk,&sone,U,&n_,V,&n_,&zero,ps,&ld_));

913:       /* compute M */
914:       PEPEvaluateBasis(pep,theta,0.0,val,NULL);
915:       PetscStackCallBLAS("BLASgemm",BLASgemm_("N","N",&n_,&n_,&n_,&snone,pjd->XpX,&ld_,ps,&ld_,&zero,M,&ld_));
916:       PetscMemzero(S,2*n*n*sizeof(PetscScalar));
917:       Sp = S+n*n;
918:       for (j=0;j<n;j++) S[j*(n+1)] = 1.0;
919:       for (k=1;k<pjd->midx;k++) {
920:         for (j=0;j<n;j++) for (i=0;i<n;i++) V[j*n+i] = S[j*n+i] - ps[j*ld+i]*val[k];
921:         PetscStackCallBLAS("BLASgemm",BLASgemm_("N","N",&n_,&n_,&n_,&sone,pjd->XpX,&ld_,V,&n_,&zero,U,&n_));
922:         PetscStackCallBLAS("BLASgemm",BLASgemm_("C","N",&n_,&n_,&n_,&sone,pjd->Tj+k*ld*ld,&ld_,U,&n_,&sone,M,&ld_));
923:         Spp = Sp; Sp = S;
924:         PEPJDEvaluateHatBasis(pep,n,pjd->T,pep->ncv,val,NULL,k+1,Spp,Sp,S);
925:       }
926:     }
927:     /* inverse */
928:     PetscStackCallBLAS("LAPACKgetrf",LAPACKgetrf_(&n_,&n_,M,&ld_,p,&info));
929:     SlepcCheckLapackInfo("getrf",info);
930:     PetscStackCallBLAS("LAPACKgetri",LAPACKgetri_(&n_,M,&ld_,p,work,&n_,&info));
931:     SlepcCheckLapackInfo("getri",info);
932:     if (pjd->midx==1) {
933:       PetscFree2(work,p);
934:     } else {
935:       PetscFree7(U,S,sg,work,rwork,p,val);
936:     }
937:   }
938:   return(0);
939: #endif
940: }

942: static PetscErrorCode PEPJDMatSetUp(PEP pep,PetscInt sz,PetscScalar *theta)
943: {
944:   PetscErrorCode  ierr;
945:   PEP_JD          *pjd = (PEP_JD*)pep->data;
946:   PEP_JD_MATSHELL *matctx;
947:   PEP_JD_PCSHELL  *pcctx;
948:   MatStructure    str;
949:   PetscScalar     *vals,*valsi;
950:   PetscBool       skipmat=PETSC_FALSE;
951:   PetscInt        i;
952:   Mat             Pr=NULL;

955:   if (sz==2 && theta[1]==0.0) sz = 1;
956:   MatShellGetContext(pjd->Pshell,(void**)&matctx);
957:   PCShellGetContext(pjd->pcshell,(void**)&pcctx);
958:   if (matctx->Pr && matctx->theta[0]==theta[0] && ((!matctx->Pi && sz==1) || (sz==2 && matctx->theta[1]==theta[1]))) {
959:     if (pcctx->n == pjd->nlock) return(0);
960:     skipmat = PETSC_TRUE;
961:   }
962:   if (!skipmat) {
963:     PetscMalloc2(pep->nmat,&vals,pep->nmat,&valsi);
964:       STGetMatStructure(pep->st,&str);
965:     PEPEvaluateBasis(pep,theta[0],theta[1],vals,valsi);
966:     if (!matctx->Pr) {
967:       MatDuplicate(pep->A[0],MAT_COPY_VALUES,&matctx->Pr);
968:     } else {
969:       MatCopy(pep->A[0],matctx->Pr,str);
970:     }
971:     for (i=1;i<pep->nmat;i++) {
972:       MatAXPY(matctx->Pr,vals[i],pep->A[i],str);
973:     }
974:     if (!pjd->reusepc) {
975:       if (pcctx->PPr && sz==2) {
976:         MatCopy(matctx->Pr,pcctx->PPr,str);
977:         Pr = pcctx->PPr;
978:       } else Pr = matctx->Pr;
979:     }
980:     matctx->theta[0] = theta[0];
981:     if (sz==2) {
982:       if (!matctx->Pi ) {
983:         MatDuplicate(pep->A[0],MAT_COPY_VALUES,&matctx->Pi);
984:       } else {
985:         MatCopy(pep->A[1],matctx->Pi,str);
986:       }
987:       MatScale(matctx->Pi,valsi[1]);
988:       for (i=2;i<pep->nmat;i++) {
989:         MatAXPY(matctx->Pi,valsi[i],pep->A[i],str);
990:       }
991:       matctx->theta[1] = theta[1];
992:     }
993:     PetscFree2(vals,valsi);
994:   }
995:   if (!pjd->reusepc) {
996:     if (!skipmat) {
997:       PCSetOperators(pcctx->pc,Pr,Pr);
998:       PCSetUp(pcctx->pc);
999:     }
1000:     PEPJDUpdateExtendedPC(pep,theta[0]);
1001:   }
1002:   return(0);
1003: }

1005: static PetscErrorCode PEPJDCreateShellPC(PEP pep,Vec *ww)
1006: {
1007:   PEP_JD          *pjd = (PEP_JD*)pep->data;
1008:   PEP_JD_PCSHELL  *pcctx;
1009:   PEP_JD_MATSHELL *matctx;
1010:   KSP             ksp;
1011:   PetscInt        nloc,mloc,kspsf=1;
1012:   Vec             v[2];
1013:   PetscScalar     target[2];
1014:   PetscErrorCode  ierr;
1015:   Mat             Pr;

1018:   /* Create the reference vector */
1019:   BVGetColumn(pjd->V,0,&v[0]);
1020:   v[1] = v[0];
1021: #if !defined (PETSC_USE_COMPLEX)
1022:   kspsf = 2;
1023: #endif
1024:   VecCreateCompWithVecs(v,kspsf,NULL,&pjd->vtempl);
1025:   BVRestoreColumn(pjd->V,0,&v[0]);
1026:   PetscLogObjectParent((PetscObject)pep,(PetscObject)pjd->vtempl);

1028:   /* Replace preconditioner with one containing projectors */
1029:   PCCreate(PetscObjectComm((PetscObject)pep),&pjd->pcshell);
1030:   PCSetType(pjd->pcshell,PCSHELL);
1031:   PCShellSetName(pjd->pcshell,"PCPEPJD");
1032:   PCShellSetApply(pjd->pcshell,PCShellApply_PEPJD);
1033:   PetscNew(&pcctx);
1034:   PCShellSetContext(pjd->pcshell,pcctx);
1035:   STGetKSP(pep->st,&ksp);
1036:   BVCreateVec(pjd->V,&pcctx->Bp[0]);
1037:   VecDuplicate(pcctx->Bp[0],&pcctx->Bp[1]);
1038:   KSPGetPC(ksp,&pcctx->pc);
1039:   PetscObjectReference((PetscObject)pcctx->pc);
1040:   MatGetLocalSize(pep->A[0],&mloc,&nloc);
1041:   if (pjd->ld>1) {
1042:     nloc += pjd->ld-1; mloc += pjd->ld-1;
1043:   }
1044:   PetscNew(&matctx);
1045:   MatCreateShell(PetscObjectComm((PetscObject)pep),kspsf*nloc,kspsf*mloc,PETSC_DETERMINE,PETSC_DETERMINE,matctx,&pjd->Pshell);
1046:   MatShellSetOperation(pjd->Pshell,MATOP_MULT,(void(*)(void))PEPJDShellMatMult);
1047:   MatShellSetOperation(pjd->Pshell,MATOP_CREATE_VECS,(void(*)(void))PEPJDSellMatCreateVecs);
1048:   matctx->pep = pep;
1049:   target[0] = pep->target; target[1] = 0.0;
1050:   PEPJDMatSetUp(pep,1,target);
1051:   Pr = matctx->Pr;
1052:   pcctx->PPr = NULL;
1053: #if !defined(PETSC_USE_COMPLEX)
1054:   if (!pjd->reusepc) {
1055:     MatDuplicate(matctx->Pr,MAT_COPY_VALUES,&pcctx->PPr);
1056:     Pr = pcctx->PPr;
1057:   }
1058: #endif
1059:   PCSetOperators(pcctx->pc,Pr,Pr);
1060:   PCSetErrorIfFailure(pcctx->pc,PETSC_TRUE);
1061:   KSPSetPC(ksp,pjd->pcshell);
1062:   if (pjd->reusepc) {
1063:     PCSetReusePreconditioner(pcctx->pc,PETSC_TRUE);
1064:     KSPSetReusePreconditioner(ksp,PETSC_TRUE);
1065:   }
1066:   KSPSetOperators(ksp,pjd->Pshell,pjd->Pshell);
1067:   KSPSetUp(ksp);
1068:   if (pjd->ld>1) {
1069:     PetscMalloc2(pjd->ld*pjd->ld,&pcctx->M,pjd->ld*pjd->ld,&pcctx->ps);
1070:     pcctx->pep = pep;
1071:   }
1072:   matctx->work = ww;
1073:   pcctx->work  = ww;
1074:   return(0);
1075: }

1077: static PetscErrorCode PEPJDEigenvectors(PEP pep)
1078: {
1079: #if defined(SLEPC_MISSING_LAPACK_TREVC)
1081:   SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"TREVC - Lapack routine is unavailable");
1082: #else
1084:   PEP_JD         *pjd = (PEP_JD*)pep->data;
1085:   PetscBLASInt   ld,nconv,info,nc;
1086:   PetscScalar    *Z,*w;
1087:   PetscReal      *wr,norm;
1088:   PetscInt       i;
1089:   Mat            U;
1090: #if !defined(PETSC_USE_COMPLEX)
1091:   Vec            v,v1;
1092: #endif

1095:   PetscMalloc3(pep->nconv*pep->nconv,&Z,3*pep->ncv,&wr,2*pep->ncv,&w);
1096:   PetscBLASIntCast(pep->ncv,&ld);
1097:   PetscBLASIntCast(pep->nconv,&nconv);
1098: #if !defined(PETSC_USE_COMPLEX)
1099:   PetscStackCallBLAS("LAPACKtrevc",LAPACKtrevc_("R","A",NULL,&nconv,pjd->T,&ld,NULL,&nconv,Z,&nconv,&nconv,&nc,wr,&info));
1100: #else
1101:   PetscStackCallBLAS("LAPACKtrevc",LAPACKtrevc_("R","A",NULL,&nconv,pjd->T,&ld,NULL,&nconv,Z,&nconv,&nconv,&nc,w,wr,&info));
1102: #endif
1103:   SlepcCheckLapackInfo("trevc",info);
1104:   MatCreateSeqDense(PETSC_COMM_SELF,nconv,nconv,Z,&U);
1105:   BVSetActiveColumns(pjd->X,0,pep->nconv);
1106:   BVMultInPlace(pjd->X,U,0,pep->nconv);
1107:   for (i=0;i<pep->nconv;i++) {
1108: #if !defined(PETSC_USE_COMPLEX)
1109:     if (pep->eigi[i]!=0.0) {   /* first eigenvalue of a complex conjugate pair */
1110:       BVGetColumn(pjd->X,i,&v);
1111:       BVGetColumn(pjd->X,i+1,&v1);
1112:       VecNormalizeComplex(v,v1,PETSC_TRUE,NULL);
1113:       BVRestoreColumn(pjd->X,i,&v);
1114:       BVRestoreColumn(pjd->X,i+1,&v1);
1115:       i++;
1116:     } else   /* real eigenvalue */
1117: #endif
1118:     {
1119:       BVNormColumn(pjd->X,i,NORM_2,&norm);
1120:       BVScaleColumn(pjd->X,i,1.0/norm);
1121:     }
1122:   }
1123:   MatDestroy(&U);
1124:   PetscFree3(Z,wr,w);
1125:   return(0);
1126: #endif
1127: }

1129: static PetscErrorCode PEPJDLockConverged(PEP pep,PetscInt *nv,PetscInt sz)
1130: {
1132:   PEP_JD         *pjd = (PEP_JD*)pep->data;
1133:   PetscInt       j,i,*P,ldds,rk=0,nvv=*nv;
1134:   Vec            v,x,w;
1135:   PetscScalar    *R,*r,*pX,target[2];
1136:   Mat            X;
1137:   PetscBLASInt   sz_,rk_,nv_,info;
1138:   PetscMPIInt    np;

1141:   /* update AX and XpX */
1142:   for (i=sz;i>0;i--) {
1143:     BVGetColumn(pjd->X,pjd->nlock-i,&x);
1144:     for (j=0;j<pep->nmat;j++) {
1145:       BVGetColumn(pjd->AX[j],pjd->nlock-i,&v);
1146:       MatMult(pep->A[j],x,v);
1147:       BVRestoreColumn(pjd->AX[j],pjd->nlock-i,&v);
1148:       BVSetActiveColumns(pjd->AX[j],0,pjd->nlock-i+1);
1149:     }
1150:     BVRestoreColumn(pjd->X,pjd->nlock-i,&x);
1151:     BVDotColumn(pjd->X,(pjd->nlock-i),pjd->XpX+(pjd->nlock-i)*(pjd->ld));
1152:     pjd->XpX[(pjd->nlock-i)*(1+pjd->ld)] = 1.0;
1153:     for (j=0;j<pjd->nlock-i;j++) pjd->XpX[j*(pjd->ld)+pjd->nlock-i] = PetscConj(pjd->XpX[(pjd->nlock-i)*(pjd->ld)+j]);
1154:   }

1156:   /* minimality index */
1157:   pjd->midx = PetscMin(pjd->mmidx,pjd->nlock);

1159:   /* evaluate the polynomial basis in T */
1160:   PetscMemzero(pjd->Tj,pjd->ld*pjd->ld*pep->nmat*sizeof(PetscScalar));
1161:   for (j=0;j<pep->nmat;j++) {
1162:     PEPEvaluateBasisMat(pep,pjd->nlock,pjd->T,pep->ncv,j,(j>1)?pjd->Tj+(j-2)*pjd->ld*pjd->ld:NULL,pjd->ld,j?pjd->Tj+(j-1)*pjd->ld*pjd->ld:NULL,pjd->ld,pjd->Tj+j*pjd->ld*pjd->ld,pjd->ld);
1163:   }

1165:   /* Extend search space */
1166:   MPI_Comm_size(PETSC_COMM_WORLD,&np);
1167:   PetscCalloc3(nvv,&P,nvv*nvv,&R,nvv*sz,&r);
1168:   DSGetLeadingDimension(pep->ds,&ldds);
1169:   DSGetArray(pep->ds,DS_MAT_X,&pX);
1170:   PEPJDOrthogonalize(nvv,nvv,pX,ldds,&rk,P,R,nvv);
1171:   for (j=0;j<sz;j++) {
1172:     for (i=0;i<rk;i++) r[i*sz+j] = PetscConj(R[nvv*i+j]*pep->eigr[P[i]]); /* first row scaled with permuted diagonal */
1173:   }
1174:   PetscBLASIntCast(rk,&rk_);
1175:   PetscBLASIntCast(sz,&sz_);
1176:   PetscBLASIntCast(nvv,&nv_);
1177:   PetscStackCallBLAS("LAPACKtrtri",LAPACKtrtri_("U","N",&rk_,R,&nv_,&info));
1178:   SlepcCheckLapackInfo("trtri",info);
1179:   for (i=0;i<sz;i++) {
1180:     PetscStackCallBLAS("BLAStrmv",BLAStrmv_("U","C","N",&rk_,R,&nv_,r+i,&sz_));
1181:   }
1182:   for (i=0;i<sz*rk;i++) r[i] = PetscConj(r[i])/PetscSqrtReal(np); /* revert */
1183:   BVSetActiveColumns(pjd->V,0,nvv);
1184:   rk -= sz;
1185:   for (j=0;j<rk;j++) {
1186:     PetscMemcpy(R+j*nvv,pX+(j+sz)*ldds,nvv*sizeof(PetscScalar));
1187:   }
1188:   DSRestoreArray(pep->ds,DS_MAT_X,&pX);
1189:   MatCreateSeqDense(PETSC_COMM_SELF,nvv,rk,R,&X);
1190:   BVMultInPlace(pjd->V,X,0,rk);
1191:   MatDestroy(&X);
1192:   BVSetActiveColumns(pjd->V,0,rk);
1193:   for (j=0;j<rk;j++) {
1194:     BVGetColumn(pjd->V,j,&v);
1195:     PEPJDCopyToExtendedVec(pep,NULL,r+sz*(j+sz),sz,pjd->nlock-sz,v,PETSC_FALSE);
1196:     BVRestoreColumn(pjd->V,j,&v);
1197:   }
1198:   BVOrthogonalize(pjd->V,NULL);

1200:   if (pjd->proj==PEP_JD_PROJECTION_HARMONIC) {
1201:     for (j=0;j<rk;j++) {
1202:       /* W = P(target)*V */
1203:       BVGetColumn(pjd->W,j,&w);
1204:       BVGetColumn(pjd->V,j,&v);
1205:       target[0] = pep->target; target[1] = 0.0;
1206:       PEPJDComputeResidual(pep,PETSC_FALSE,1,&v,target,&w,pep->work);
1207:       BVRestoreColumn(pjd->V,j,&v);
1208:       BVRestoreColumn(pjd->W,j,&w);
1209:     }
1210:     BVSetActiveColumns(pjd->W,0,rk);
1211:     BVOrthogonalize(pjd->W,NULL);
1212:   }
1213:   *nv = rk;
1214:   PetscFree3(P,R,r);
1215:   return(0);
1216: }

1218: PetscErrorCode PEPJDSystemSetUp(PEP pep,PetscInt sz,PetscScalar *theta,Vec *u,Vec *p,Vec *ww)
1219: {
1220:   PetscErrorCode  ierr;
1221:   PetscScalar     s[2];
1222:   PEP_JD          *pjd = (PEP_JD*)pep->data;
1223:   PEP_JD_PCSHELL  *pcctx;

1226:   PCShellGetContext(pjd->pcshell,(void**)&pcctx);
1227:   PEPJDMatSetUp(pep,sz,theta);
1228:   pcctx->u[0] = u[0]; pcctx->u[1] = u[1];
1229:   /* Compute r'. p is a work space vector */
1230:   PEPJDComputeResidual(pep,PETSC_TRUE,sz,u,theta,p,ww);
1231:   PEPJDExtendedPCApply(pjd->pcshell,p[0],pcctx->Bp[0]);
1232:   VecDot(pcctx->Bp[0],u[0],pcctx->gamma);
1233:   if (sz==2) {
1234:     PEPJDExtendedPCApply(pjd->pcshell,p[1],pcctx->Bp[1]);
1235:     VecDot(pcctx->Bp[0],u[1],pcctx->gamma+1);
1236:     VecMDot(pcctx->Bp[1],2,u,s);
1237:     pcctx->gamma[0] += s[1];
1238:     pcctx->gamma[1] = -pcctx->gamma[1]+s[0];
1239:   } else {
1240:     VecZeroEntries(pcctx->Bp[1]);
1241:     pcctx->gamma[1] = 0.0;
1242:   }
1243:   return(0);
1244: }

1246: PetscErrorCode PEPSolve_JD(PEP pep)
1247: {
1248:   PetscErrorCode  ierr;
1249:   PEP_JD          *pjd = (PEP_JD*)pep->data;
1250:   PetscInt        k,nv,nvc,ld,minv,dim,bupdated=0,sz=1,kspsf=1,idx,off,maxits,nloc;
1251:   PetscMPIInt     np,count;
1252:   PetscScalar     theta[2]={0.0,0.0},ritz[2]={0.0,0.0},*pX,*eig,*eigi,*array;
1253:   PetscReal       norm,norm1,*res,tol=0.0,rtol,abstol, dtol;
1254:   PetscBool       lindep,ini=PETSC_TRUE;
1255:   Vec             tc,t[2]={NULL,NULL},u[2]={NULL,NULL},p[2]={NULL,NULL};
1256:   Vec             rc,rr[2],r[2]={NULL,NULL},*ww=pep->work,v[2];
1257:   Mat             G,X,Y;
1258:   KSP             ksp;
1259:   PEP_JD_PCSHELL  *pcctx;
1260:   PEP_JD_MATSHELL *matctx;

1263:   MPI_Comm_size(PetscObjectComm((PetscObject)pep),&np);
1264:   BVGetSizes(pep->V,&nloc,NULL,NULL);
1265:   DSGetLeadingDimension(pep->ds,&ld);
1266:   PetscCalloc3(pep->ncv+pep->nev,&eig,pep->ncv+pep->nev,&eigi,pep->ncv+pep->nev,&res);
1267:   pjd->nlock = 0;
1268:   STGetKSP(pep->st,&ksp);
1269:   KSPGetTolerances(ksp,&rtol,&abstol,&dtol,&maxits);
1270: #if !defined (PETSC_USE_COMPLEX)
1271:   kspsf = 2;
1272: #endif
1273:   PEPJDProcessInitialSpace(pep,ww);
1274:   nv = (pep->nini)?pep->nini:1;

1276:   /* Replace preconditioner with one containing projectors */
1277:   PEPJDCreateShellPC(pep,ww);
1278:   PCShellGetContext(pjd->pcshell,(void**)&pcctx);

1280:   /* Create auxiliar vectors */
1281:   BVCreateVec(pjd->V,&u[0]);
1282:   VecDuplicate(u[0],&p[0]);
1283:   VecDuplicate(u[0],&r[0]);
1284: #if !defined (PETSC_USE_COMPLEX)
1285:   VecDuplicate(u[0],&u[1]);
1286:   VecDuplicate(u[0],&p[1]);
1287:   VecDuplicate(u[0],&r[1]);
1288: #endif

1290:   /* Restart loop */
1291:   while (pep->reason == PEP_CONVERGED_ITERATING) {
1292:     pep->its++;
1293:     DSSetDimensions(pep->ds,nv,0,0,0);
1294:     BVSetActiveColumns(pjd->V,bupdated,nv);
1295:     PEPJDUpdateTV(pep,bupdated,nv,ww);
1296:     if (pjd->proj==PEP_JD_PROJECTION_HARMONIC) { BVSetActiveColumns(pjd->W,bupdated,nv); }
1297:     for (k=0;k<pep->nmat;k++) {
1298:       BVSetActiveColumns(pjd->TV[k],bupdated,nv);
1299:       DSGetMat(pep->ds,DSMatExtra[k],&G);
1300:       BVMatProject(pjd->TV[k],NULL,pjd->W,G);
1301:       DSRestoreMat(pep->ds,DSMatExtra[k],&G);
1302:     }
1303:     BVSetActiveColumns(pjd->V,0,nv);
1304:     BVSetActiveColumns(pjd->W,0,nv);

1306:     /* Solve projected problem */
1307:     DSSetState(pep->ds,DS_STATE_RAW);
1308:     DSSolve(pep->ds,pep->eigr,pep->eigi);
1309:     DSSort(pep->ds,pep->eigr,pep->eigi,NULL,NULL,NULL);
1310:     DSSynchronize(pep->ds,pep->eigr,pep->eigi);
1311:     idx = 0;
1312:     do {
1313:       ritz[0] = pep->eigr[idx];
1314: #if !defined(PETSC_USE_COMPLEX)
1315:       ritz[1] = pep->eigi[idx];
1316: #endif
1317:       sz = (ritz[1]==0.0)?1:2;
1318:       /* Compute Ritz vector u=V*X(:,1) */
1319:       DSGetArray(pep->ds,DS_MAT_X,&pX);
1320:       BVSetActiveColumns(pjd->V,0,nv);
1321:       BVMultVec(pjd->V,1.0,0.0,u[0],pX+idx*ld);
1322:       if (sz==2) {
1323:         BVMultVec(pjd->V,1.0,0.0,u[1],pX+(idx+1)*ld);
1324:       }
1325:       DSRestoreArray(pep->ds,DS_MAT_X,&pX);
1326:       PEPJDComputeResidual(pep,PETSC_FALSE,sz,u,ritz,r,ww);
1327:       /* Check convergence */
1328:       VecNorm(r[0],NORM_2,&norm);
1329:       if (sz==2) {
1330:         VecNorm(r[1],NORM_2,&norm1);
1331:         norm = SlepcAbs(norm,norm1);
1332:       }
1333:       (*pep->converged)(pep,ritz[0],ritz[1],norm,&pep->errest[pep->nconv],pep->convergedctx);
1334:       if (sz==2) pep->errest[pep->nconv+1] = pep->errest[pep->nconv];
1335:       if (ini) {
1336:         tol = PetscMin(.1,pep->errest[pep->nconv]); ini = PETSC_FALSE;
1337:       } else tol = PetscMin(pep->errest[pep->nconv],tol/2);
1338:       (*pep->stopping)(pep,pep->its,pep->max_it,(pep->errest[pep->nconv]<pep->tol)?pep->nconv+sz:pep->nconv,pep->nev,&pep->reason,pep->stoppingctx);
1339:       if (pep->errest[pep->nconv]<pep->tol) {
1340:         /* Ritz pair converged */
1341:         ini = PETSC_TRUE;
1342:         minv = PetscMin(nv,(PetscInt)(pjd->keep*pep->ncv));
1343:         if (pjd->ld>1) {
1344:           BVGetColumn(pjd->X,pep->nconv,&v[0]);
1345:           PEPJDCopyToExtendedVec(pep,v[0],pjd->T+pep->ncv*pep->nconv,pjd->ld-1,0,u[0],PETSC_TRUE);
1346:           BVRestoreColumn(pjd->X,pep->nconv,&v[0]);
1347:           BVSetActiveColumns(pjd->X,0,pep->nconv+1);
1348:           BVNormColumn(pjd->X,pep->nconv,NORM_2,&norm);
1349:           BVScaleColumn(pjd->X,pep->nconv,1.0/norm);
1350:           for (k=0;k<pep->nconv;k++) pjd->T[pep->ncv*pep->nconv+k] *= PetscSqrtReal(np)/norm;
1351:           pjd->T[(pep->ncv+1)*pep->nconv] = ritz[0];
1352:           eig[pep->nconv] = ritz[0];
1353:           idx++;
1354:           if (sz==2) {
1355:             BVGetColumn(pjd->X,pep->nconv+1,&v[0]);
1356:             PEPJDCopyToExtendedVec(pep,v[0],pjd->T+pep->ncv*(pep->nconv+1),pjd->ld-1,0,u[1],PETSC_TRUE);
1357:             BVRestoreColumn(pjd->X,pep->nconv+1,&v[0]);
1358:             BVSetActiveColumns(pjd->X,0,pep->nconv+2);
1359:             BVNormColumn(pjd->X,pep->nconv+1,NORM_2,&norm1);
1360:             BVScaleColumn(pjd->X,pep->nconv+1,1.0/norm1);
1361:             for (k=0;k<pep->nconv;k++) pjd->T[pep->ncv*(pep->nconv+1)+k] *= PetscSqrtReal(np)/norm1;
1362:             pjd->T[(pep->ncv+1)*(pep->nconv+1)] = ritz[0];
1363:             pjd->T[(pep->ncv+1)*pep->nconv+1] = -ritz[1]*norm1/norm;
1364:             pjd->T[(pep->ncv+1)*(pep->nconv+1)-1] = ritz[1]*norm/norm1;
1365:             eig[pep->nconv+1] = ritz[0];
1366:             eigi[pep->nconv] = ritz[1]; eigi[pep->nconv+1] = -ritz[1];
1367:             idx++;
1368:           }
1369:         } else {
1370:           BVInsertVec(pep->V,pep->nconv,u[0]);
1371:         }
1372:         pep->nconv += sz;
1373:       }
1374:     } while (pep->errest[pep->nconv]<pep->tol && pep->nconv<nv);

1376:     if (pep->reason==PEP_CONVERGED_ITERATING) {
1377:       nvc = nv;
1378:       if (idx) {
1379:         pjd->nlock +=idx;
1380:         PEPJDLockConverged(pep,&nv,idx);
1381:       }
1382:       if (nv+sz>=pep->ncv-1) {
1383:         /* Basis full, force restart */
1384:         minv = PetscMin(nv,(PetscInt)(pjd->keep*pep->ncv));
1385:         DSGetDimensions(pep->ds,&dim,NULL,NULL,NULL,NULL);
1386:         DSGetArray(pep->ds,DS_MAT_X,&pX);
1387:         PEPJDOrthogonalize(dim,minv,pX,ld,&minv,NULL,NULL,ld);
1388:         DSRestoreArray(pep->ds,DS_MAT_X,&pX);
1389:         DSGetArray(pep->ds,DS_MAT_Y,&pX);
1390:         PEPJDOrthogonalize(dim,minv,pX,ld,&minv,NULL,NULL,ld);
1391:         DSRestoreArray(pep->ds,DS_MAT_Y,&pX);
1392:         DSGetMat(pep->ds,DS_MAT_X,&X);
1393:         BVMultInPlace(pjd->V,X,0,minv);
1394:         if (pjd->proj==PEP_JD_PROJECTION_HARMONIC) {
1395:          DSGetMat(pep->ds,DS_MAT_Y,&Y);
1396:          BVMultInPlace(pjd->W,Y,pep->nconv,minv);
1397:          DSRestoreMat(pep->ds,DS_MAT_Y,&Y);
1398:         }
1399:         MatDestroy(&X);
1400:         nv = minv;
1401:         bupdated = 0;
1402:       } else {
1403:         if (!idx && pep->errest[pep->nconv]<pjd->fix) {theta[0] = ritz[0]; theta[1] = ritz[1];}
1404:         else {theta[0] = pep->target; theta[1] = 0.0;}
1405:         /* Update system mat */
1406:         PEPJDSystemSetUp(pep,sz,theta,u,p,ww);
1407:         /* Solve correction equation to expand basis */
1408:         BVGetColumn(pjd->V,nv,&t[0]);
1409:         rr[0] = r[0];
1410:         if (sz==2) {
1411:           BVGetColumn(pjd->V,nv+1,&t[1]);
1412:           rr[1] = r[1];
1413:         } else {
1414:           t[1] = NULL;
1415:           rr[1] = NULL;
1416:         }
1417:         VecCreateCompWithVecs(t,kspsf,pjd->vtempl,&tc);
1418:         VecCreateCompWithVecs(rr,kspsf,pjd->vtempl,&rc);
1419:         VecCompSetSubVecs(pjd->vtempl,sz,NULL);
1420:         tol  = PetscMax(rtol,tol/2);
1421:         KSPSetTolerances(ksp,tol,abstol,dtol,maxits);
1422:         KSPSolve(ksp,rc,tc);
1423:         VecDestroy(&tc);
1424:         VecDestroy(&rc);
1425:         VecGetArray(t[0],&array);
1426:         PetscMPIIntCast(pep->nconv,&count);
1427:         MPI_Bcast(array+nloc,count,MPIU_SCALAR,np-1,PetscObjectComm((PetscObject)pep));
1428:         VecRestoreArray(t[0],&array);
1429:         BVRestoreColumn(pjd->V,nv,&t[0]);
1430:         BVOrthogonalizeColumn(pjd->V,nv,NULL,&norm,&lindep);
1431:         if (lindep || norm==0.0) {
1432:           if (sz==1) SETERRQ(PETSC_COMM_SELF,1,"Linearly dependent continuation vector");
1433:           else off = 1;
1434:         } else {
1435:           off = 0;
1436:           BVScaleColumn(pjd->V,nv,1.0/norm);
1437:         }
1438:         if (sz==2) {
1439:           VecGetArray(t[1],&array);
1440:           MPI_Bcast(array+nloc,count,MPIU_SCALAR,np-1,PetscObjectComm((PetscObject)pep));
1441:           VecRestoreArray(t[1],&array);
1442:           BVRestoreColumn(pjd->V,nv+1,&t[1]);
1443:           if (off) {
1444:             BVCopyColumn(pjd->V,nv+1,nv);
1445:           }
1446:           BVOrthogonalizeColumn(pjd->V,nv+1-off,NULL,&norm,&lindep);
1447:           if (lindep || norm==0.0) {
1448:             if (off) SETERRQ(PETSC_COMM_SELF,1,"Linearly dependent continuation vector");
1449:             else off = 1;
1450:           } else {
1451:             BVScaleColumn(pjd->V,nv+1-off,1.0/norm);
1452:           }
1453:         }
1454:         if (pjd->proj==PEP_JD_PROJECTION_HARMONIC) {
1455:           BVInsertVec(pjd->W,nv,r[0]);
1456:           if (sz==2 && !off) {
1457:             BVInsertVec(pjd->W,nv+1,r[1]);
1458:           }
1459:           BVOrthogonalizeColumn(pjd->W,nv,NULL,&norm,&lindep);
1460:           if (lindep || norm==0.0) SETERRQ(PETSC_COMM_SELF,1,"Linearly dependent continuation vector");
1461:           BVScaleColumn(pjd->W,nv,1.0/norm);
1462:           if (sz==2 && !off) {
1463:             BVOrthogonalizeColumn(pjd->W,nv+1,NULL,&norm,&lindep);
1464:             if (lindep || norm==0.0) SETERRQ(PETSC_COMM_SELF,1,"Linearly dependent continuation vector");
1465:             BVScaleColumn(pjd->W,nv+1,1.0/norm);
1466:           }
1467:         }
1468:         bupdated = idx?0:nv;
1469:         nv += sz-off;
1470:       }
1471:       for (k=0;k<nvc;k++) {
1472:         eig[pep->nconv-idx+k] = pep->eigr[k];
1473: #if !defined(PETSC_USE_COMPLEX)
1474:         eigi[pep->nconv-idx+k] = pep->eigi[k];
1475: #endif
1476:       }
1477:       PEPMonitor(pep,pep->its,pep->nconv,eig,eigi,pep->errest,pep->nconv+1);
1478:     }
1479:   }
1480:   if (pjd->ld>1) {
1481:     for (k=0;k<pep->nconv;k++) {
1482:       pep->eigr[k] = eig[k];
1483:       pep->eigi[k] = eigi[k];
1484:     }
1485:     if (pep->nconv>0) { PEPJDEigenvectors(pep); }
1486:     PetscFree2(pcctx->M,pcctx->ps);
1487:   }
1488:   VecDestroy(&u[0]);
1489:   VecDestroy(&r[0]);
1490:   VecDestroy(&p[0]);
1491: #if !defined (PETSC_USE_COMPLEX)
1492:   VecDestroy(&u[1]);
1493:   VecDestroy(&r[1]);
1494:   VecDestroy(&p[1]);
1495: #endif
1496:   KSPSetTolerances(ksp,rtol,abstol,dtol,maxits);
1497:   KSPSetPC(ksp,pcctx->pc);
1498:   VecDestroy(&pcctx->Bp[0]);
1499:   VecDestroy(&pcctx->Bp[1]);
1500:   MatShellGetContext(pjd->Pshell,(void**)&matctx);
1501:   MatDestroy(&matctx->Pr);
1502:   MatDestroy(&matctx->Pi);
1503:   MatDestroy(&pjd->Pshell);
1504:   MatDestroy(&pcctx->PPr);
1505:   PCDestroy(&pcctx->pc);
1506:   PetscFree(pcctx);
1507:   PetscFree(matctx);
1508:   PCDestroy(&pjd->pcshell);
1509:   PetscFree3(eig,eigi,res);
1510:   VecDestroy(&pjd->vtempl);
1511:   return(0);
1512: }

1514: PetscErrorCode PEPJDSetRestart_JD(PEP pep,PetscReal keep)
1515: {
1516:   PEP_JD *pjd = (PEP_JD*)pep->data;

1519:   if (keep==PETSC_DEFAULT) pjd->keep = 0.5;
1520:   else {
1521:     if (keep<0.1 || keep>0.9) SETERRQ(PetscObjectComm((PetscObject)pep),PETSC_ERR_ARG_OUTOFRANGE,"The keep argument must be in the range [0.1,0.9]");
1522:     pjd->keep = keep;
1523:   }
1524:   return(0);
1525: }

1527: /*@
1528:    PEPJDSetRestart - Sets the restart parameter for the Jacobi-Davidson
1529:    method, in particular the proportion of basis vectors that must be kept
1530:    after restart.

1532:    Logically Collective on PEP

1534:    Input Parameters:
1535: +  pep  - the eigenproblem solver context
1536: -  keep - the number of vectors to be kept at restart

1538:    Options Database Key:
1539: .  -pep_jd_restart - Sets the restart parameter

1541:    Notes:
1542:    Allowed values are in the range [0.1,0.9]. The default is 0.5.

1544:    Level: advanced

1546: .seealso: PEPJDGetRestart()
1547: @*/
1548: PetscErrorCode PEPJDSetRestart(PEP pep,PetscReal keep)
1549: {

1555:   PetscTryMethod(pep,"PEPJDSetRestart_C",(PEP,PetscReal),(pep,keep));
1556:   return(0);
1557: }

1559: PetscErrorCode PEPJDGetRestart_JD(PEP pep,PetscReal *keep)
1560: {
1561:   PEP_JD *pjd = (PEP_JD*)pep->data;

1564:   *keep = pjd->keep;
1565:   return(0);
1566: }

1568: /*@
1569:    PEPJDGetRestart - Gets the restart parameter used in the Jacobi-Davidson method.

1571:    Not Collective

1573:    Input Parameter:
1574: .  pep - the eigenproblem solver context

1576:    Output Parameter:
1577: .  keep - the restart parameter

1579:    Level: advanced

1581: .seealso: PEPJDSetRestart()
1582: @*/
1583: PetscErrorCode PEPJDGetRestart(PEP pep,PetscReal *keep)
1584: {

1590:   PetscUseMethod(pep,"PEPJDGetRestart_C",(PEP,PetscReal*),(pep,keep));
1591:   return(0);
1592: }

1594: PetscErrorCode PEPJDSetFix_JD(PEP pep,PetscReal fix)
1595: {
1596:   PEP_JD *pjd = (PEP_JD*)pep->data;

1599:   if (fix == PETSC_DEFAULT || fix == PETSC_DECIDE) pjd->fix = 0.01;
1600:   else {
1601:     if (fix < 0.0) SETERRQ(PetscObjectComm((PetscObject)pep),PETSC_ERR_ARG_OUTOFRANGE,"Invalid fix value");
1602:     pjd->fix = fix;
1603:   }
1604:   return(0);
1605: }

1607: /*@
1608:    PEPJDSetFix - Sets the threshold for changing the target in the correction
1609:    equation.

1611:    Logically Collective on PEP

1613:    Input Parameters:
1614: +  pep - the eigenproblem solver context
1615: -  fix - threshold for changing the target

1617:    Options Database Key:
1618: .  -pep_jd_fix - the fix value

1620:    Note:
1621:    The target in the correction equation is fixed at the first iterations.
1622:    When the norm of the residual vector is lower than the fix value,
1623:    the target is set to the corresponding eigenvalue.

1625:    Level: advanced

1627: .seealso: PEPJDGetFix()
1628: @*/
1629: PetscErrorCode PEPJDSetFix(PEP pep,PetscReal fix)
1630: {

1636:   PetscTryMethod(pep,"PEPJDSetFix_C",(PEP,PetscReal),(pep,fix));
1637:   return(0);
1638: }

1640: PetscErrorCode PEPJDGetFix_JD(PEP pep,PetscReal *fix)
1641: {
1642:   PEP_JD *pjd = (PEP_JD*)pep->data;

1645:   *fix = pjd->fix;
1646:   return(0);
1647: }

1649: /*@
1650:    PEPJDGetFix - Returns the threshold for changing the target in the correction
1651:    equation.

1653:    Not Collective

1655:    Input Parameter:
1656: .  pep - the eigenproblem solver context

1658:    Output Parameter:
1659: .  fix - threshold for changing the target

1661:    Note:
1662:    The target in the correction equation is fixed at the first iterations.
1663:    When the norm of the residual vector is lower than the fix value,
1664:    the target is set to the corresponding eigenvalue.

1666:    Level: advanced

1668: .seealso: PEPJDSetFix()
1669: @*/
1670: PetscErrorCode PEPJDGetFix(PEP pep,PetscReal *fix)
1671: {

1677:   PetscUseMethod(pep,"PEPJDGetFix_C",(PEP,PetscReal*),(pep,fix));
1678:   return(0);
1679: }

1681: PetscErrorCode PEPJDSetReusePreconditioner_JD(PEP pep,PetscBool reusepc)
1682: {
1683:   PEP_JD *pjd = (PEP_JD*)pep->data;

1686:   pjd->reusepc = reusepc;
1687:   return(0);
1688: }

1690: /*@
1691:    PEPJDSetReusePreconditioner - Sets a flag indicating whether the preconditioner
1692:    must be reused or not.

1694:    Logically Collective on PEP

1696:    Input Parameters:
1697: +  pep     - the eigenproblem solver context
1698: -  reusepc - the reuse flag

1700:    Options Database Key:
1701: .  -pep_jd_reuse_preconditioner - the reuse flag

1703:    Note:
1704:    The default value is False. If set to True, the preconditioner is built
1705:    only at the beginning, using the target value. Otherwise, it may be rebuilt
1706:    (depending on the fix parameter) at each iteration from the Ritz value.

1708:    Level: advanced

1710: .seealso: PEPJDGetReusePreconditioner(), PEPJDSetFix()
1711: @*/
1712: PetscErrorCode PEPJDSetReusePreconditioner(PEP pep,PetscBool reusepc)
1713: {

1719:   PetscTryMethod(pep,"PEPJDSetReusePreconditioner_C",(PEP,PetscBool),(pep,reusepc));
1720:   return(0);
1721: }

1723: PetscErrorCode PEPJDGetReusePreconditioner_JD(PEP pep,PetscBool *reusepc)
1724: {
1725:   PEP_JD *pjd = (PEP_JD*)pep->data;

1728:   *reusepc = pjd->reusepc;
1729:   return(0);
1730: }

1732: /*@
1733:    PEPJDGetReusePreconditioner - Returns the flag for reusing the preconditioner.

1735:    Not Collective

1737:    Input Parameter:
1738: .  pep - the eigenproblem solver context

1740:    Output Parameter:
1741: .  reusepc - the reuse flag

1743:    Level: advanced

1745: .seealso: PEPJDSetReusePreconditioner()
1746: @*/
1747: PetscErrorCode PEPJDGetReusePreconditioner(PEP pep,PetscBool *reusepc)
1748: {

1754:   PetscUseMethod(pep,"PEPJDGetReusePreconditioner_C",(PEP,PetscBool*),(pep,reusepc));
1755:   return(0);
1756: }

1758: PetscErrorCode PEPJDSetMinimalityIndex_JD(PEP pep,PetscInt mmidx)
1759: {
1760:   PEP_JD *pjd = (PEP_JD*)pep->data;

1763:   if (mmidx == PETSC_DEFAULT || mmidx == PETSC_DECIDE) pjd->mmidx = 1;
1764:   else {
1765:     if (mmidx < 1) SETERRQ(PetscObjectComm((PetscObject)pep),PETSC_ERR_ARG_OUTOFRANGE,"Invalid mmidx value");
1766:     pjd->mmidx = mmidx;
1767:     pep->state = PEP_STATE_INITIAL;
1768:   }
1769:   return(0);
1770: }

1772: /*@
1773:    PEPJDSetMinimalityIndex - Sets the maximum allowed value for the minimality index.

1775:    Logically Collective on PEP

1777:    Input Parameters:
1778: +  pep   - the eigenproblem solver context
1779: -  mmidx - maximum minimality index

1781:    Options Database Key:
1782: .  -pep_jd_minimality_index - the minimality index value

1784:    Note:
1785:    The default value is equal to the degree of the polynomial. A smaller value
1786:    can be used if the wanted eigenvectors are known to be linearly independent.

1788:    Level: advanced

1790: .seealso: PEPJDGetMinimalityIndex()
1791: @*/
1792: PetscErrorCode PEPJDSetMinimalityIndex(PEP pep,PetscInt mmidx)
1793: {

1799:   PetscTryMethod(pep,"PEPJDSetMinimalityIndex_C",(PEP,PetscInt),(pep,mmidx));
1800:   return(0);
1801: }

1803: PetscErrorCode PEPJDGetMinimalityIndex_JD(PEP pep,PetscInt *mmidx)
1804: {
1805:   PEP_JD *pjd = (PEP_JD*)pep->data;

1808:   *mmidx = pjd->mmidx;
1809:   return(0);
1810: }

1812: /*@
1813:    PEPJDGetMinimalityIndex - Returns the maximum allowed value of the minimality
1814:    index.

1816:    Not Collective

1818:    Input Parameter:
1819: .  pep - the eigenproblem solver context

1821:    Output Parameter:
1822: .  mmidx - minimality index

1824:    Level: advanced

1826: .seealso: PEPJDSetMinimalityIndex()
1827: @*/
1828: PetscErrorCode PEPJDGetMinimalityIndex(PEP pep,PetscInt *mmidx)
1829: {

1835:   PetscUseMethod(pep,"PEPJDGetMinimalityIndex_C",(PEP,PetscInt*),(pep,mmidx));
1836:   return(0);
1837: }

1839: PetscErrorCode PEPJDSetProjection_JD(PEP pep,PEPJDProjection proj)
1840: {
1841:   PEP_JD *pjd = (PEP_JD*)pep->data;

1844:   switch (proj) {
1845:     case PEP_JD_PROJECTION_HARMONIC:
1846:     case PEP_JD_PROJECTION_ORTHOGONAL:
1847:       if (pjd->proj != proj) {
1848:         pep->state = PEP_STATE_INITIAL;
1849:         pjd->proj = proj;
1850:       }
1851:       break;
1852:     default:
1853:       SETERRQ(PetscObjectComm((PetscObject)pep),PETSC_ERR_ARG_OUTOFRANGE,"Invalid 'proj' value");
1854:   }
1855:   return(0);
1856: }

1858: /*@
1859:    PEPJDSetProjection - Sets the type of projection to be used in the Jacobi-Davidson solver.

1861:    Logically Collective on PEP

1863:    Input Parameters:
1864: +  pep  - the eigenproblem solver context
1865: -  proj - the type of projection

1867:    Options Database Key:
1868: .  -pep_jd_projection - the projection type, either orthogonal or harmonic

1870:    Level: advanced

1872: .seealso: PEPJDGetProjection()
1873: @*/
1874: PetscErrorCode PEPJDSetProjection(PEP pep,PEPJDProjection proj)
1875: {

1881:   PetscTryMethod(pep,"PEPJDSetProjection_C",(PEP,PEPJDProjection),(pep,proj));
1882:   return(0);
1883: }

1885: PetscErrorCode PEPJDGetProjection_JD(PEP pep,PEPJDProjection *proj)
1886: {
1887:   PEP_JD *pjd = (PEP_JD*)pep->data;

1890:   *proj = pjd->proj;
1891:   return(0);
1892: }

1894: /*@
1895:    PEPJDGetProjection - Returns the type of projection used by the Jacobi-Davidson solver.

1897:    Not Collective

1899:    Input Parameter:
1900: .  pep - the eigenproblem solver context

1902:    Output Parameter:
1903: .  proj - the type of projection

1905:    Level: advanced

1907: .seealso: PEPJDSetProjection()
1908: @*/
1909: PetscErrorCode PEPJDGetProjection(PEP pep,PEPJDProjection *proj)
1910: {

1916:   PetscUseMethod(pep,"PEPJDGetProjection_C",(PEP,PEPJDProjection*),(pep,proj));
1917:   return(0);
1918: }

1920: PetscErrorCode PEPSetFromOptions_JD(PetscOptionItems *PetscOptionsObject,PEP pep)
1921: {
1922:   PetscErrorCode  ierr;
1923:   PetscBool       flg,b1;
1924:   PetscReal       r1;
1925:   PetscInt        i1;
1926:   PEPJDProjection proj;

1929:   PetscOptionsHead(PetscOptionsObject,"PEP JD Options");

1931:     PetscOptionsReal("-pep_jd_restart","Proportion of vectors kept after restart","PEPJDSetRestart",0.5,&r1,&flg);
1932:     if (flg) { PEPJDSetRestart(pep,r1); }

1934:     PetscOptionsReal("-pep_jd_fix","Tolerance for changing the target in the correction equation","PEPJDSetFix",0.01,&r1,&flg);
1935:     if (flg) { PEPJDSetFix(pep,r1); }

1937:     PetscOptionsBool("-pep_jd_reuse_preconditioner","Whether to reuse the preconditioner","PEPJDSetReusePreconditoiner",PETSC_FALSE,&b1,&flg);
1938:     if (flg) { PEPJDSetReusePreconditioner(pep,b1); }

1940:     PetscOptionsInt("-pep_jd_minimality_index","Maximum allowed minimality index","PEPJDSetMinimalityIndex",1,&i1,&flg);
1941:     if (flg) { PEPJDSetMinimalityIndex(pep,i1); }

1943:     PetscOptionsEnum("-pep_jd_projection","Type of projection","PEPJDSetProjection",PEPJDProjectionTypes,(PetscEnum)PEP_JD_PROJECTION_HARMONIC,(PetscEnum*)&proj,&flg);
1944:     if (flg) { PEPJDSetProjection(pep,proj); }

1946:   PetscOptionsTail();
1947:   return(0);
1948: }

1950: PetscErrorCode PEPView_JD(PEP pep,PetscViewer viewer)
1951: {
1953:   PEP_JD         *pjd = (PEP_JD*)pep->data;
1954:   PetscBool      isascii;

1957:   PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&isascii);
1958:   if (isascii) {
1959:     PetscViewerASCIIPrintf(viewer,"  %d%% of basis vectors kept after restart\n",(int)(100*pjd->keep));
1960:     PetscViewerASCIIPrintf(viewer,"  threshold for changing the target in the correction equation (fix): %g\n",(double)pjd->fix);
1961:     PetscViewerASCIIPrintf(viewer,"  projection type: %s\n",PEPJDProjectionTypes[pjd->proj]);
1962:     PetscViewerASCIIPrintf(viewer,"  maximum allowed minimality index: %d\n",pjd->mmidx);
1963:     if (pjd->reusepc) { PetscViewerASCIIPrintf(viewer,"  reusing the preconditioner\n"); }
1964:   }
1965:   return(0);
1966: }

1968: PetscErrorCode PEPSetDefaultST_JD(PEP pep)
1969: {
1971:   KSP            ksp;

1974:   if (!((PetscObject)pep->st)->type_name) {
1975:     STSetType(pep->st,STPRECOND);
1976:     STPrecondSetKSPHasMat(pep->st,PETSC_TRUE);
1977:   }
1978:   STSetTransform(pep->st,PETSC_FALSE);
1979:   STGetKSP(pep->st,&ksp);
1980:   if (!((PetscObject)ksp)->type_name) {
1981:     KSPSetType(ksp,KSPBCGSL);
1982:     KSPSetTolerances(ksp,1e-5,PETSC_DEFAULT,PETSC_DEFAULT,100);
1983:   }
1984:   return(0);
1985: }

1987: PetscErrorCode PEPReset_JD(PEP pep)
1988: {
1990:   PEP_JD         *pjd = (PEP_JD*)pep->data;
1991:   PetscInt       i;

1994:   for (i=0;i<pep->nmat;i++) {
1995:     BVDestroy(pjd->TV+i);
1996:   }
1997:   if (pjd->proj==PEP_JD_PROJECTION_HARMONIC) { BVDestroy(&pjd->W); }
1998:   if (pjd->ld>1) {
1999:     BVDestroy(&pjd->V);
2000:     for (i=0;i<pep->nmat;i++) {
2001:       BVDestroy(pjd->AX+i);
2002:     }
2003:     BVDestroy(&pjd->N[0]);
2004:     BVDestroy(&pjd->N[1]);
2005:     PetscFree3(pjd->XpX,pjd->T,pjd->Tj);
2006:   }
2007:   PetscFree2(pjd->TV,pjd->AX);
2008:   return(0);
2009: }

2011: PetscErrorCode PEPDestroy_JD(PEP pep)
2012: {

2016:   PetscFree(pep->data);
2017:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetRestart_C",NULL);
2018:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetRestart_C",NULL);
2019:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetFix_C",NULL);
2020:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetFix_C",NULL);
2021:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetReusePreconditioner_C",NULL);
2022:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetReusePreconditioner_C",NULL);
2023:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetMinimalityIndex_C",NULL);
2024:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetMinimalityIndex_C",NULL);
2025:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetProjection_C",NULL);
2026:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetProjection_C",NULL);
2027:   return(0);
2028: }

2030: SLEPC_EXTERN PetscErrorCode PEPCreate_JD(PEP pep)
2031: {
2032:   PEP_JD         *pjd;

2036:   PetscNewLog(pep,&pjd);
2037:   pep->data = (void*)pjd;

2039:   pjd->fix   = 0.01;
2040:   pjd->mmidx = 0;

2042:   pep->ops->solve          = PEPSolve_JD;
2043:   pep->ops->setup          = PEPSetUp_JD;
2044:   pep->ops->setfromoptions = PEPSetFromOptions_JD;
2045:   pep->ops->destroy        = PEPDestroy_JD;
2046:   pep->ops->reset          = PEPReset_JD;
2047:   pep->ops->view           = PEPView_JD;
2048:   pep->ops->setdefaultst   = PEPSetDefaultST_JD;

2050:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetRestart_C",PEPJDSetRestart_JD);
2051:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetRestart_C",PEPJDGetRestart_JD);
2052:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetFix_C",PEPJDSetFix_JD);
2053:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetFix_C",PEPJDGetFix_JD);
2054:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetReusePreconditioner_C",PEPJDSetReusePreconditioner_JD);
2055:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetReusePreconditioner_C",PEPJDGetReusePreconditioner_JD);
2056:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetMinimalityIndex_C",PEPJDSetMinimalityIndex_JD);
2057:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetMinimalityIndex_C",PEPJDGetMinimalityIndex_JD);
2058:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDSetProjection_C",PEPJDSetProjection_JD);
2059:   PetscObjectComposeFunction((PetscObject)pep,"PEPJDGetProjection_C",PEPJDGetProjection_JD);
2060:   return(0);
2061: }