Actual source code: nciss.c
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 nonlinear eigenvalue problems using contour
27: integrals", JSIAM Lett. 1:52-55, 2009.
29: [2] S. Yokota and T. Sakurai, "A projection method for nonlinear
30: eigenvalue problems using contour integrals", JSIAM Lett.
31: 5:41-44, 2013.
32: */
34: #include <slepc/private/nepimpl.h>
35: #include <slepc/private/slepccontour.h>
37: typedef struct _n_nep_ciss_project *NEP_CISS_PROJECT;
38: typedef struct {
39: /* parameters */
40: PetscInt N; /* number of integration points (32) */
41: PetscInt L; /* block size (16) */
42: PetscInt M; /* moment degree (N/4 = 4) */
43: PetscReal delta; /* threshold of singular value (1e-12) */
44: PetscInt L_max; /* maximum number of columns of the source matrix V */
45: PetscReal spurious_threshold; /* discard spurious eigenpairs */
46: PetscBool isreal; /* T(z) is real for real z */
47: PetscInt npart; /* number of partitions */
48: PetscInt refine_inner;
49: PetscInt refine_blocksize;
50: NEPCISSExtraction extraction;
51: /* private data */
52: SlepcContourData contour;
53: PetscReal *sigma; /* threshold for numerical rank */
54: PetscScalar *weight;
55: PetscScalar *omega;
56: PetscScalar *pp;
57: BV V;
58: BV S;
59: BV Y;
60: PetscBool useconj;
61: Mat J; /* auxiliary matrix when using subcomm */
62: BV pV;
63: NEP_CISS_PROJECT dsctxf;
64: PetscObjectId rgid;
65: PetscObjectState rgstate;
66: } NEP_CISS;
68: struct _n_nep_ciss_project {
69: NEP nep;
70: BV Q;
71: };
73: static PetscErrorCode NEPContourDSComputeMatrix(DS ds,PetscScalar lambda,PetscBool deriv,DSMatType mat,void *ctx)
74: {
75: NEP_CISS_PROJECT proj = (NEP_CISS_PROJECT)ctx;
76: Mat M,fun;
78: PetscFunctionBegin;
79: if (!deriv) {
80: PetscCall(NEPComputeFunction(proj->nep,lambda,proj->nep->function,proj->nep->function));
81: fun = proj->nep->function;
82: } else {
83: PetscCall(NEPComputeJacobian(proj->nep,lambda,proj->nep->jacobian));
84: fun = proj->nep->jacobian;
85: }
86: PetscCall(DSGetMat(ds,mat,&M));
87: PetscCall(BVMatProject(proj->Q,fun,proj->Q,M));
88: PetscCall(DSRestoreMat(ds,mat,&M));
89: PetscFunctionReturn(PETSC_SUCCESS);
90: }
92: static PetscErrorCode NEPComputeFunctionSubcomm(NEP nep,PetscScalar lambda,Mat T,Mat P,PetscBool deriv)
93: {
94: PetscInt i;
95: PetscScalar alpha;
96: NEP_CISS *ctx = (NEP_CISS*)nep->data;
98: PetscFunctionBegin;
99: PetscAssert(nep->fui!=NEP_USER_INTERFACE_CALLBACK,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_WRONGSTATE,"Should not arrive here with callbacks");
100: PetscCall(MatZeroEntries(T));
101: if (!deriv && T != P) PetscCall(MatZeroEntries(P));
102: for (i=0;i<nep->nt;i++) {
103: if (!deriv) PetscCall(FNEvaluateFunction(nep->f[i],lambda,&alpha));
104: else PetscCall(FNEvaluateDerivative(nep->f[i],lambda,&alpha));
105: PetscCall(MatAXPY(T,alpha,ctx->contour->pA[i],nep->mstr));
106: if (!deriv && T != P) PetscCall(MatAXPY(P,alpha,ctx->contour->pP[i],nep->mstrp));
107: }
108: PetscFunctionReturn(PETSC_SUCCESS);
109: }
111: /*
112: Set up KSP solvers for every integration point
113: */
114: static PetscErrorCode NEPCISSSetUp(NEP nep,Mat T,Mat P)
115: {
116: NEP_CISS *ctx = (NEP_CISS*)nep->data;
117: SlepcContourData contour;
118: PetscInt i,p_id;
119: Mat Amat,Pmat;
121: PetscFunctionBegin;
122: if (!ctx->contour || !ctx->contour->ksp) PetscCall(NEPCISSGetKSPs(nep,NULL,NULL));
123: contour = ctx->contour;
124: PetscAssert(ctx->contour && ctx->contour->ksp,PetscObjectComm((PetscObject)nep),PETSC_ERR_PLIB,"Something went wrong with NEPCISSGetKSPs()");
125: for (i=0;i<contour->npoints;i++) {
126: p_id = i*contour->subcomm->n + contour->subcomm->color;
127: PetscCall(MatDuplicate(T,MAT_DO_NOT_COPY_VALUES,&Amat));
128: if (T != P) PetscCall(MatDuplicate(P,MAT_DO_NOT_COPY_VALUES,&Pmat)); else Pmat = Amat;
129: if (contour->subcomm->n == 1 || nep->fui==NEP_USER_INTERFACE_CALLBACK) PetscCall(NEPComputeFunction(nep,ctx->omega[p_id],Amat,Pmat));
130: else PetscCall(NEPComputeFunctionSubcomm(nep,ctx->omega[p_id],Amat,Pmat,PETSC_FALSE));
131: PetscCall(NEP_KSPSetOperators(contour->ksp[i],Amat,Pmat));
132: PetscCall(MatDestroy(&Amat));
133: if (T != P) PetscCall(MatDestroy(&Pmat));
134: }
135: PetscFunctionReturn(PETSC_SUCCESS);
136: }
138: /*
139: Y_i = F(z_i)^{-1}Fp(z_i)V for every integration point, Y=[Y_i] is in the context
140: */
141: static PetscErrorCode NEPCISSSolve(NEP nep,Mat dT,BV V,PetscInt L_start,PetscInt L_end)
142: {
143: NEP_CISS *ctx = (NEP_CISS*)nep->data;
144: SlepcContourData contour = ctx->contour;
145: PetscInt i,p_id;
146: Mat MV,BMV=NULL,MC;
148: PetscFunctionBegin;
149: PetscCall(BVSetActiveColumns(V,L_start,L_end));
150: PetscCall(BVGetMat(V,&MV));
151: for (i=0;i<contour->npoints;i++) {
152: p_id = i*contour->subcomm->n + contour->subcomm->color;
153: if (contour->subcomm->n==1 || nep->fui==NEP_USER_INTERFACE_CALLBACK) PetscCall(NEPComputeJacobian(nep,ctx->omega[p_id],dT));
154: else PetscCall(NEPComputeFunctionSubcomm(nep,ctx->omega[p_id],dT,NULL,PETSC_TRUE));
155: PetscCall(BVSetActiveColumns(ctx->Y,i*ctx->L+L_start,i*ctx->L+L_end));
156: PetscCall(BVGetMat(ctx->Y,&MC));
157: if (!i) {
158: PetscCall(MatProductCreate(dT,MV,NULL,&BMV));
159: PetscCall(MatProductSetType(BMV,MATPRODUCT_AB));
160: PetscCall(MatProductSetFromOptions(BMV));
161: PetscCall(MatProductSymbolic(BMV));
162: }
163: PetscCall(MatProductNumeric(BMV));
164: PetscCall(KSPMatSolve(contour->ksp[i],BMV,MC));
165: PetscCall(BVRestoreMat(ctx->Y,&MC));
166: }
167: PetscCall(MatDestroy(&BMV));
168: PetscCall(BVRestoreMat(V,&MV));
169: PetscFunctionReturn(PETSC_SUCCESS);
170: }
172: static PetscErrorCode NEPSetUp_CISS(NEP nep)
173: {
174: NEP_CISS *ctx = (NEP_CISS*)nep->data;
175: SlepcContourData contour;
176: PetscInt nwork;
177: PetscBool istrivial,isellipse,flg;
178: NEP_CISS_PROJECT dsctxf;
179: PetscObjectId id;
180: PetscObjectState state;
181: Vec v0;
183: PetscFunctionBegin;
184: if (nep->ncv==PETSC_DETERMINE) nep->ncv = ctx->L_max*ctx->M;
185: else {
186: ctx->L_max = nep->ncv/ctx->M;
187: if (!ctx->L_max) {
188: ctx->L_max = 1;
189: nep->ncv = ctx->L_max*ctx->M;
190: }
191: }
192: ctx->L = PetscMin(ctx->L,ctx->L_max);
193: if (nep->max_it==PETSC_DETERMINE) nep->max_it = 5;
194: if (nep->mpd==PETSC_DETERMINE) nep->mpd = nep->ncv;
195: if (!nep->which) nep->which = NEP_ALL;
196: PetscCheck(nep->which==NEP_ALL,PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"This solver supports only computing all eigenvalues");
197: NEPCheckUnsupported(nep,NEP_FEATURE_STOPPING | NEP_FEATURE_TWOSIDED);
199: /* check region */
200: PetscCall(RGIsTrivial(nep->rg,&istrivial));
201: PetscCheck(!istrivial,PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"CISS requires a nontrivial region, e.g. -rg_type ellipse ...");
202: PetscCall(RGGetComplement(nep->rg,&flg));
203: PetscCheck(!flg,PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"A region with complement flag set is not allowed");
204: PetscCall(PetscObjectTypeCompare((PetscObject)nep->rg,RGELLIPSE,&isellipse));
205: PetscCheck(isellipse,PetscObjectComm((PetscObject)nep),PETSC_ERR_SUP,"Currently only implemented for elliptic regions");
207: /* if the region has changed, then reset contour data */
208: PetscCall(PetscObjectGetId((PetscObject)nep->rg,&id));
209: PetscCall(PetscObjectStateGet((PetscObject)nep->rg,&state));
210: if (ctx->rgid && (id != ctx->rgid || state != ctx->rgstate)) {
211: PetscCall(SlepcContourDataDestroy(&ctx->contour));
212: PetscCall(PetscInfo(nep,"Resetting the contour data structure due to a change of region\n"));
213: ctx->rgid = id; ctx->rgstate = state;
214: }
216: /* create contour data structure */
217: if (!ctx->contour) {
218: PetscCall(RGCanUseConjugates(nep->rg,ctx->isreal,&ctx->useconj));
219: PetscCall(SlepcContourDataCreate(ctx->useconj?ctx->N/2:ctx->N,ctx->npart,(PetscObject)nep,&ctx->contour));
220: }
222: PetscCall(NEPAllocateSolution(nep,0));
223: if (ctx->weight) PetscCall(PetscFree4(ctx->weight,ctx->omega,ctx->pp,ctx->sigma));
224: PetscCall(PetscMalloc4(ctx->N,&ctx->weight,ctx->N,&ctx->omega,ctx->N,&ctx->pp,ctx->L_max*ctx->M,&ctx->sigma));
226: /* allocate basis vectors */
227: PetscCall(BVDestroy(&ctx->S));
228: PetscCall(BVDuplicateResize(nep->V,ctx->L*ctx->M,&ctx->S));
229: PetscCall(BVDestroy(&ctx->V));
230: PetscCall(BVDuplicateResize(nep->V,ctx->L,&ctx->V));
232: contour = ctx->contour;
233: if (contour->subcomm && contour->subcomm->n != 1 && nep->fui==NEP_USER_INTERFACE_CALLBACK) {
234: PetscCall(NEPComputeFunction(nep,0,nep->function,nep->function_pre));
235: PetscCall(SlepcContourRedundantMat(contour,1,&nep->function,(nep->function!=nep->function_pre)?&nep->function_pre:NULL));
236: } else PetscCall(SlepcContourRedundantMat(contour,nep->nt,nep->A,nep->P));
237: if (contour->pA) {
238: if (!ctx->J) PetscCall(MatDuplicate(contour->pA[0],MAT_DO_NOT_COPY_VALUES,&ctx->J));
239: PetscCall(BVGetColumn(ctx->V,0,&v0));
240: PetscCall(SlepcContourScatterCreate(contour,v0));
241: PetscCall(BVRestoreColumn(ctx->V,0,&v0));
242: PetscCall(BVDestroy(&ctx->pV));
243: PetscCall(BVCreate(PetscObjectComm((PetscObject)contour->xsub),&ctx->pV));
244: PetscCall(BVSetSizesFromVec(ctx->pV,contour->xsub,nep->n));
245: PetscCall(BVSetFromOptions(ctx->pV));
246: PetscCall(BVResize(ctx->pV,ctx->L,PETSC_FALSE));
247: }
249: PetscCall(BVDestroy(&ctx->Y));
250: if (contour->pA) {
251: PetscCall(BVCreate(PetscObjectComm((PetscObject)contour->xsub),&ctx->Y));
252: PetscCall(BVSetSizesFromVec(ctx->Y,contour->xsub,nep->n));
253: PetscCall(BVSetFromOptions(ctx->Y));
254: PetscCall(BVResize(ctx->Y,contour->npoints*ctx->L,PETSC_FALSE));
255: } else PetscCall(BVDuplicateResize(nep->V,contour->npoints*ctx->L,&ctx->Y));
257: if (ctx->extraction == NEP_CISS_EXTRACTION_RITZ) {
258: PetscCall(DSSetMethod(nep->ds,1));
259: PetscCall(DSNEPSetRG(nep->ds,nep->rg));
260: if (nep->fui==NEP_USER_INTERFACE_SPLIT) PetscCall(DSNEPSetFN(nep->ds,nep->nt,nep->f));
261: else {
262: PetscCall(PetscNew(&dsctxf));
263: PetscCall(DSNEPSetComputeMatrixFunction(nep->ds,NEPContourDSComputeMatrix,dsctxf));
264: dsctxf->nep = nep;
265: ctx->dsctxf = dsctxf;
266: }
267: }
268: PetscCall(DSAllocate(nep->ds,nep->ncv));
269: nwork = (nep->fui==NEP_USER_INTERFACE_SPLIT)? 2: 1;
270: PetscCall(NEPSetWorkVecs(nep,nwork));
271: PetscFunctionReturn(PETSC_SUCCESS);
272: }
274: static PetscErrorCode NEPSolve_CISS(NEP nep)
275: {
276: NEP_CISS *ctx = (NEP_CISS*)nep->data;
277: SlepcContourData contour = ctx->contour;
278: Mat X,M,E,T,P,J;
279: BV V;
280: PetscInt i,j,ld,L_add=0,nv=0,L_base=ctx->L,inner,*inside;
281: PetscScalar *Mu,*H0,*H1,*rr,*temp,center;
282: PetscReal error,max_error,radius,rgscale,est_eig,eta;
283: PetscBool isellipse,*fl1;
284: Vec si;
285: SlepcSC sc;
286: PetscRandom rand;
288: PetscFunctionBegin;
289: PetscCall(DSGetSlepcSC(nep->ds,&sc));
290: sc->comparison = SlepcCompareLargestMagnitude;
291: sc->comparisonctx = NULL;
292: sc->map = NULL;
293: sc->mapobj = NULL;
294: PetscCall(DSGetLeadingDimension(nep->ds,&ld));
295: PetscCall(RGComputeQuadrature(nep->rg,RG_QUADRULE_TRAPEZOIDAL,ctx->N,ctx->omega,ctx->pp,ctx->weight));
296: if (contour->pA) {
297: T = contour->pA[0];
298: P = ((nep->fui==NEP_USER_INTERFACE_SPLIT && nep->P) || (nep->fui==NEP_USER_INTERFACE_CALLBACK && contour->pP))? contour->pP[0]: T;
299: } else {
300: T = nep->function;
301: P = nep->function_pre? nep->function_pre: nep->function;
302: }
303: PetscCall(NEPCISSSetUp(nep,T,P));
304: PetscCall(BVSetActiveColumns(ctx->V,0,ctx->L));
305: PetscCall(BVSetRandomSign(ctx->V));
306: PetscCall(BVGetRandomContext(ctx->V,&rand));
307: if (contour->pA) {
308: J = ctx->J;
309: V = ctx->pV;
310: PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
311: } else {
312: J = nep->jacobian;
313: V = ctx->V;
314: }
315: PetscCall(NEPCISSSolve(nep,J,V,0,ctx->L));
316: PetscCall(PetscObjectTypeCompare((PetscObject)nep->rg,RGELLIPSE,&isellipse));
317: if (isellipse) {
318: PetscCall(BVTraceQuadrature(ctx->Y,ctx->V,ctx->L,ctx->L,ctx->weight,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj,&est_eig));
319: PetscCall(PetscInfo(nep,"Estimated eigenvalue count: %f\n",(double)est_eig));
320: eta = PetscPowReal(10.0,-PetscLog10Real(nep->tol)/ctx->N);
321: L_add = PetscMax(0,(PetscInt)PetscCeilReal((est_eig*eta)/ctx->M)-ctx->L);
322: if (L_add>ctx->L_max-ctx->L) {
323: PetscCall(PetscInfo(nep,"Number of eigenvalues inside the contour path may be too large\n"));
324: L_add = ctx->L_max-ctx->L;
325: }
326: }
327: /* Updates L after estimate the number of eigenvalue */
328: if (L_add>0) {
329: PetscCall(PetscInfo(nep,"Changing L %" PetscInt_FMT " -> %" PetscInt_FMT " by Estimate #Eig\n",ctx->L,ctx->L+L_add));
330: PetscCall(BVCISSResizeBases(ctx->S,contour->pA?ctx->pV:ctx->V,ctx->Y,ctx->L,ctx->L+L_add,ctx->M,contour->npoints));
331: PetscCall(BVSetActiveColumns(ctx->V,ctx->L,ctx->L+L_add));
332: PetscCall(BVSetRandomSign(ctx->V));
333: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
334: ctx->L += L_add;
335: PetscCall(NEPCISSSolve(nep,J,V,ctx->L-L_add,ctx->L));
336: }
338: PetscCall(PetscMalloc2(ctx->L*ctx->L*ctx->M*2,&Mu,ctx->L*ctx->M*ctx->L*ctx->M,&H0));
339: for (i=0;i<ctx->refine_blocksize;i++) {
340: PetscCall(BVDotQuadrature(ctx->Y,V,Mu,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->subcomm,contour->npoints,ctx->useconj));
341: PetscCall(CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0));
342: PetscCall(PetscLogEventBegin(NEP_CISS_SVD,nep,0,0,0));
343: PetscCall(SlepcCISS_BH_SVD(H0,ctx->L*ctx->M,ctx->delta,ctx->sigma,&nv));
344: PetscCall(PetscLogEventEnd(NEP_CISS_SVD,nep,0,0,0));
345: if (ctx->sigma[0]<=ctx->delta || nv < ctx->L*ctx->M || ctx->L == ctx->L_max) break;
346: L_add = L_base;
347: if (ctx->L+L_add>ctx->L_max) L_add = ctx->L_max-ctx->L;
348: PetscCall(PetscInfo(nep,"Changing L %" PetscInt_FMT " -> %" PetscInt_FMT " by SVD(H0)\n",ctx->L,ctx->L+L_add));
349: PetscCall(BVCISSResizeBases(ctx->S,contour->pA?ctx->pV:ctx->V,ctx->Y,ctx->L,ctx->L+L_add,ctx->M,contour->npoints));
350: PetscCall(BVSetActiveColumns(ctx->V,ctx->L,ctx->L+L_add));
351: PetscCall(BVSetRandomSign(ctx->V));
352: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
353: ctx->L += L_add;
354: PetscCall(NEPCISSSolve(nep,J,V,ctx->L-L_add,ctx->L));
355: if (L_add) {
356: PetscCall(PetscFree2(Mu,H0));
357: PetscCall(PetscMalloc2(ctx->L*ctx->L*ctx->M*2,&Mu,ctx->L*ctx->M*ctx->L*ctx->M,&H0));
358: }
359: }
361: PetscCall(RGGetScale(nep->rg,&rgscale));
362: PetscCall(RGEllipseGetParameters(nep->rg,¢er,&radius,NULL));
364: if (ctx->extraction == NEP_CISS_EXTRACTION_HANKEL) PetscCall(PetscMalloc1(ctx->L*ctx->M*ctx->L*ctx->M,&H1));
366: while (nep->reason == NEP_CONVERGED_ITERATING) {
367: nep->its++;
368: for (inner=0;inner<=ctx->refine_inner;inner++) {
369: if (ctx->extraction == NEP_CISS_EXTRACTION_HANKEL) {
370: PetscCall(BVDotQuadrature(ctx->Y,V,Mu,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->subcomm,contour->npoints,ctx->useconj));
371: PetscCall(CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0));
372: PetscCall(PetscLogEventBegin(NEP_CISS_SVD,nep,0,0,0));
373: PetscCall(SlepcCISS_BH_SVD(H0,ctx->L*ctx->M,ctx->delta,ctx->sigma,&nv));
374: PetscCall(PetscLogEventEnd(NEP_CISS_SVD,nep,0,0,0));
375: } else {
376: PetscCall(BVSumQuadrature(ctx->S,ctx->Y,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj));
377: /* compute SVD of S */
378: PetscCall(BVSVDAndRank(ctx->S,ctx->M,ctx->L,ctx->delta,(ctx->extraction==NEP_CISS_EXTRACTION_CAA)?BV_SVD_METHOD_QR_CAA:BV_SVD_METHOD_QR,H0,ctx->sigma,&nv));
379: }
380: PetscCall(PetscInfo(nep,"Estimated rank: nv = %" PetscInt_FMT "\n",nv));
381: if (ctx->sigma[0]>ctx->delta && nv==ctx->L*ctx->M && inner!=ctx->refine_inner) {
382: PetscCall(BVSumQuadrature(ctx->S,ctx->Y,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj));
383: PetscCall(BVSetActiveColumns(ctx->S,0,ctx->L));
384: PetscCall(BVSetActiveColumns(ctx->V,0,ctx->L));
385: PetscCall(BVCopy(ctx->S,ctx->V));
386: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
387: PetscCall(NEPCISSSolve(nep,J,V,0,ctx->L));
388: } else break;
389: }
390: nep->nconv = 0;
391: if (nv == 0) { nep->reason = NEP_CONVERGED_TOL; break; }
392: else {
393: /* Extracting eigenpairs */
394: PetscCall(DSSetDimensions(nep->ds,nv,0,0));
395: PetscCall(DSSetState(nep->ds,DS_STATE_RAW));
396: if (ctx->extraction == NEP_CISS_EXTRACTION_HANKEL) {
397: PetscCall(CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0));
398: PetscCall(CISS_BlockHankel(Mu,1,ctx->L,ctx->M,H1));
399: PetscCall(DSGetArray(nep->ds,DS_MAT_A,&temp));
400: for (j=0;j<nv;j++)
401: for (i=0;i<nv;i++)
402: temp[i+j*ld] = H1[i+j*ctx->L*ctx->M];
403: PetscCall(DSRestoreArray(nep->ds,DS_MAT_A,&temp));
404: PetscCall(DSGetArray(nep->ds,DS_MAT_B,&temp));
405: for (j=0;j<nv;j++)
406: for (i=0;i<nv;i++)
407: temp[i+j*ld] = H0[i+j*ctx->L*ctx->M];
408: PetscCall(DSRestoreArray(nep->ds,DS_MAT_B,&temp));
409: } else if (ctx->extraction == NEP_CISS_EXTRACTION_CAA) {
410: PetscCall(BVSetActiveColumns(ctx->S,0,nv));
411: PetscCall(DSGetArray(nep->ds,DS_MAT_A,&temp));
412: for (i=0;i<nv;i++) PetscCall(PetscArraycpy(temp+i*ld,H0+i*nv,nv));
413: PetscCall(DSRestoreArray(nep->ds,DS_MAT_A,&temp));
414: } else {
415: PetscCall(BVSetActiveColumns(ctx->S,0,nv));
416: if (nep->fui==NEP_USER_INTERFACE_SPLIT) {
417: for (i=0;i<nep->nt;i++) {
418: PetscCall(DSGetMat(nep->ds,DSMatExtra[i],&E));
419: PetscCall(BVMatProject(ctx->S,nep->A[i],ctx->S,E));
420: PetscCall(DSRestoreMat(nep->ds,DSMatExtra[i],&E));
421: }
422: } else { ctx->dsctxf->Q = ctx->S; }
423: }
424: PetscCall(DSSolve(nep->ds,nep->eigr,nep->eigi));
425: PetscCall(DSSynchronize(nep->ds,nep->eigr,nep->eigi));
426: PetscCall(DSGetDimensions(nep->ds,NULL,NULL,NULL,&nv));
427: if (ctx->extraction == NEP_CISS_EXTRACTION_CAA || ctx->extraction == NEP_CISS_EXTRACTION_HANKEL) {
428: for (i=0;i<nv;i++) {
429: nep->eigr[i] = (nep->eigr[i]*radius+center)*rgscale;
430: }
431: }
432: PetscCall(PetscMalloc3(nv,&fl1,nv,&inside,nv,&rr));
433: PetscCall(DSVectors(nep->ds,DS_MAT_X,NULL,NULL));
434: PetscCall(DSGetMat(nep->ds,DS_MAT_X,&X));
435: PetscCall(SlepcCISS_isGhost(X,nv,ctx->sigma,ctx->spurious_threshold,fl1));
436: PetscCall(DSRestoreMat(nep->ds,DS_MAT_X,&X));
437: PetscCall(RGCheckInside(nep->rg,nv,nep->eigr,nep->eigi,inside));
438: for (i=0;i<nv;i++) {
439: if (fl1[i] && inside[i]>=0) {
440: rr[i] = 1.0;
441: nep->nconv++;
442: } else rr[i] = 0.0;
443: }
444: PetscCall(DSSort(nep->ds,nep->eigr,nep->eigi,rr,NULL,&nep->nconv));
445: PetscCall(DSSynchronize(nep->ds,nep->eigr,nep->eigi));
446: if (ctx->extraction == NEP_CISS_EXTRACTION_CAA || ctx->extraction == NEP_CISS_EXTRACTION_HANKEL) {
447: for (i=0;i<nv;i++) nep->eigr[i] = (nep->eigr[i]*radius+center)*rgscale;
448: }
449: PetscCall(PetscFree3(fl1,inside,rr));
450: PetscCall(BVSetActiveColumns(nep->V,0,nv));
451: PetscCall(DSVectors(nep->ds,DS_MAT_X,NULL,NULL));
452: if (ctx->extraction == NEP_CISS_EXTRACTION_HANKEL) {
453: PetscCall(BVSumQuadrature(ctx->S,ctx->Y,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj));
454: PetscCall(BVSetActiveColumns(ctx->S,0,nv));
455: PetscCall(BVCopy(ctx->S,nep->V));
456: PetscCall(DSGetMat(nep->ds,DS_MAT_X,&X));
457: PetscCall(BVMultInPlace(ctx->S,X,0,nep->nconv));
458: PetscCall(BVMultInPlace(nep->V,X,0,nep->nconv));
459: PetscCall(DSRestoreMat(nep->ds,DS_MAT_X,&X));
460: } else {
461: PetscCall(DSGetMat(nep->ds,DS_MAT_X,&X));
462: PetscCall(BVMultInPlace(ctx->S,X,0,nep->nconv));
463: PetscCall(DSRestoreMat(nep->ds,DS_MAT_X,&X));
464: PetscCall(BVSetActiveColumns(ctx->S,0,nep->nconv));
465: PetscCall(BVCopy(ctx->S,nep->V));
466: }
467: max_error = 0.0;
468: for (i=0;i<nep->nconv;i++) {
469: PetscCall(BVGetColumn(nep->V,i,&si));
470: PetscCall(VecNormalize(si,NULL));
471: PetscCall(NEPComputeResidualNorm_Private(nep,PETSC_FALSE,nep->eigr[i],si,nep->work,&error));
472: PetscCall((*nep->converged)(nep,nep->eigr[i],0,error,&error,nep->convergedctx));
473: PetscCall(BVRestoreColumn(nep->V,i,&si));
474: max_error = PetscMax(max_error,error);
475: }
476: if (max_error <= nep->tol) nep->reason = NEP_CONVERGED_TOL;
477: else if (nep->its > nep->max_it) nep->reason = NEP_DIVERGED_ITS;
478: else {
479: if (nep->nconv > ctx->L) nv = nep->nconv;
480: else if (ctx->L > nv) nv = ctx->L;
481: nv = PetscMin(nv,ctx->L*ctx->M);
482: PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,nv,ctx->L,NULL,&M));
483: PetscCall(MatSetRandom(M,rand));
484: PetscCall(BVSetActiveColumns(ctx->S,0,nv));
485: PetscCall(BVMultInPlace(ctx->S,M,0,ctx->L));
486: PetscCall(MatDestroy(&M));
487: PetscCall(BVSetActiveColumns(ctx->S,0,ctx->L));
488: PetscCall(BVSetActiveColumns(ctx->V,0,ctx->L));
489: PetscCall(BVCopy(ctx->S,ctx->V));
490: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
491: PetscCall(NEPCISSSolve(nep,J,V,0,ctx->L));
492: }
493: }
494: }
495: PetscCall(PetscFree2(Mu,H0));
496: if (ctx->extraction == NEP_CISS_EXTRACTION_HANKEL) PetscCall(PetscFree(H1));
497: PetscFunctionReturn(PETSC_SUCCESS);
498: }
500: static PetscErrorCode NEPCISSSetSizes_CISS(NEP nep,PetscInt ip,PetscInt bs,PetscInt ms,PetscInt npart,PetscInt bsmax,PetscBool realmats)
501: {
502: NEP_CISS *ctx = (NEP_CISS*)nep->data;
503: PetscInt oN,oL,oM,oLmax,onpart;
504: PetscMPIInt size;
506: PetscFunctionBegin;
507: oN = ctx->N;
508: if (ip == PETSC_DETERMINE) {
509: if (ctx->N!=32) { ctx->N =32; ctx->M = ctx->N/4; }
510: } else if (ip != PETSC_CURRENT) {
511: PetscCheck(ip>0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The ip argument must be > 0");
512: PetscCheck(ip%2==0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The ip argument must be an even number");
513: if (ctx->N!=ip) { ctx->N = ip; ctx->M = ctx->N/4; }
514: }
515: oL = ctx->L;
516: if (bs == PETSC_DETERMINE) {
517: ctx->L = 16;
518: } else if (bs != PETSC_CURRENT) {
519: PetscCheck(bs>0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The bs argument must be > 0");
520: ctx->L = bs;
521: }
522: oM = ctx->M;
523: if (ms == PETSC_DETERMINE) {
524: ctx->M = ctx->N/4;
525: } else if (ms != PETSC_CURRENT) {
526: PetscCheck(ms>0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The ms argument must be > 0");
527: PetscCheck(ms<=ctx->N,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The ms argument must be less than or equal to the number of integration points");
528: ctx->M = ms;
529: }
530: onpart = ctx->npart;
531: if (npart == PETSC_DETERMINE) {
532: ctx->npart = 1;
533: } else if (npart != PETSC_CURRENT) {
534: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)nep),&size));
535: PetscCheck(npart>0 && npart<=size,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"Illegal value of npart");
536: ctx->npart = npart;
537: }
538: oLmax = ctx->L_max;
539: if (bsmax == PETSC_DETERMINE) {
540: ctx->L_max = 64;
541: } else if (bsmax != PETSC_CURRENT) {
542: PetscCheck(bsmax>0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The bsmax argument must be > 0");
543: ctx->L_max = PetscMax(bsmax,ctx->L);
544: }
545: if (onpart != ctx->npart || oN != ctx->N || realmats != ctx->isreal) {
546: PetscCall(SlepcContourDataDestroy(&ctx->contour));
547: PetscCall(PetscInfo(nep,"Resetting the contour data structure due to a change of parameters\n"));
548: nep->state = NEP_STATE_INITIAL;
549: }
550: ctx->isreal = realmats;
551: if (oL != ctx->L || oM != ctx->M || oLmax != ctx->L_max) nep->state = NEP_STATE_INITIAL;
552: PetscFunctionReturn(PETSC_SUCCESS);
553: }
555: /*@
556: NEPCISSSetSizes - Sets the values of various size parameters in the CISS solver.
558: Logically Collective
560: Input Parameters:
561: + nep - the nonlinear eigensolver context
562: . ip - number of integration points
563: . bs - block size
564: . ms - moment size
565: . npart - number of partitions when splitting the communicator
566: . bsmax - max block size
567: - realmats - $T(z)$ is real for real $z$
569: Options Database Keys:
570: + -nep_ciss_integration_points \<ip\> - sets the number of integration points
571: . -nep_ciss_blocksize \<bs\> - sets the block size
572: . -nep_ciss_moments \<ms\> - sets the moment size
573: . -nep_ciss_partitions \<npart\> - sets the number of partitions
574: . -nep_ciss_maxblocksize \<bsmax\> - sets the maximum block size
575: - -nep_ciss_realmats - $T(z)$ is real for real $z$
577: Notes:
578: For all integer arguments, you can use `PETSC_CURRENT` to keep the current value, and
579: `PETSC_DETERMINE` to set them to a default value.
581: The default number of partitions is 1. This means the internal `KSP` object is shared
582: among all processes of the `NEP` communicator. Otherwise, the communicator is split
583: into `npart` communicators, so that `npart` `KSP` solves proceed simultaneously.
585: The `realmats` flag can be set to `PETSC_TRUE` when $T(\cdot)$ is guaranteed to be real
586: when the argument is a real value, for example, when all matrices in
587: the split form are real. When set to `PETSC_TRUE`, the solver avoids some computations.
589: For a detailed description of the parameters see {cite:p}`Mae16`.
591: Level: advanced
593: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSGetSizes()`
594: @*/
595: PetscErrorCode NEPCISSSetSizes(NEP nep,PetscInt ip,PetscInt bs,PetscInt ms,PetscInt npart,PetscInt bsmax,PetscBool realmats)
596: {
597: PetscFunctionBegin;
605: PetscTryMethod(nep,"NEPCISSSetSizes_C",(NEP,PetscInt,PetscInt,PetscInt,PetscInt,PetscInt,PetscBool),(nep,ip,bs,ms,npart,bsmax,realmats));
606: PetscFunctionReturn(PETSC_SUCCESS);
607: }
609: static PetscErrorCode NEPCISSGetSizes_CISS(NEP nep,PetscInt *ip,PetscInt *bs,PetscInt *ms,PetscInt *npart,PetscInt *bsmax,PetscBool *realmats)
610: {
611: NEP_CISS *ctx = (NEP_CISS*)nep->data;
613: PetscFunctionBegin;
614: if (ip) *ip = ctx->N;
615: if (bs) *bs = ctx->L;
616: if (ms) *ms = ctx->M;
617: if (npart) *npart = ctx->npart;
618: if (bsmax) *bsmax = ctx->L_max;
619: if (realmats) *realmats = ctx->isreal;
620: PetscFunctionReturn(PETSC_SUCCESS);
621: }
623: /*@
624: NEPCISSGetSizes - Gets the values of various size parameters in the CISS solver.
626: Not Collective
628: Input Parameter:
629: . nep - the nonlinear eigensolver context
631: Output Parameters:
632: + ip - number of integration points
633: . bs - block size
634: . ms - moment size
635: . npart - number of partitions when splitting the communicator
636: . bsmax - max block size
637: - realmats - $T(z)$ is real for real $z$
639: Level: advanced
641: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSSetSizes()`
642: @*/
643: PetscErrorCode NEPCISSGetSizes(NEP nep,PetscInt *ip,PetscInt *bs,PetscInt *ms,PetscInt *npart,PetscInt *bsmax,PetscBool *realmats)
644: {
645: PetscFunctionBegin;
647: PetscUseMethod(nep,"NEPCISSGetSizes_C",(NEP,PetscInt*,PetscInt*,PetscInt*,PetscInt*,PetscInt*,PetscBool*),(nep,ip,bs,ms,npart,bsmax,realmats));
648: PetscFunctionReturn(PETSC_SUCCESS);
649: }
651: static PetscErrorCode NEPCISSSetThreshold_CISS(NEP nep,PetscReal delta,PetscReal spur)
652: {
653: NEP_CISS *ctx = (NEP_CISS*)nep->data;
655: PetscFunctionBegin;
656: if (delta == (PetscReal)PETSC_DETERMINE) {
657: ctx->delta = SLEPC_DEFAULT_TOL*1e-4;
658: } else if (delta != (PetscReal)PETSC_CURRENT) {
659: PetscCheck(delta>0.0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The delta argument must be > 0.0");
660: ctx->delta = delta;
661: }
662: if (spur == (PetscReal)PETSC_DETERMINE) {
663: ctx->spurious_threshold = PetscSqrtReal(SLEPC_DEFAULT_TOL);
664: } else if (spur != (PetscReal)PETSC_CURRENT) {
665: PetscCheck(spur>0.0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The spurious threshold argument must be > 0.0");
666: ctx->spurious_threshold = spur;
667: }
668: PetscFunctionReturn(PETSC_SUCCESS);
669: }
671: /*@
672: NEPCISSSetThreshold - Sets the values of various threshold parameters in
673: the CISS solver.
675: Logically Collective
677: Input Parameters:
678: + nep - the nonlinear eigensolver context
679: . delta - threshold for numerical rank
680: - spur - spurious threshold (to discard spurious eigenpairs)
682: Options Database Keys:
683: + -nep_ciss_delta \<delta\> - sets the delta
684: - -nep_ciss_spurious_threshold \<spur\> - sets the spurious threshold
686: Notes:
687: `PETSC_CURRENT` can be used to preserve the current value of any of the
688: arguments, and `PETSC_DETERMINE` to set them to a default value.
690: For a detailed description of the parameters see {cite:p}`Mae16`.
692: Level: advanced
694: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSGetThreshold()`
695: @*/
696: PetscErrorCode NEPCISSSetThreshold(NEP nep,PetscReal delta,PetscReal spur)
697: {
698: PetscFunctionBegin;
702: PetscTryMethod(nep,"NEPCISSSetThreshold_C",(NEP,PetscReal,PetscReal),(nep,delta,spur));
703: PetscFunctionReturn(PETSC_SUCCESS);
704: }
706: static PetscErrorCode NEPCISSGetThreshold_CISS(NEP nep,PetscReal *delta,PetscReal *spur)
707: {
708: NEP_CISS *ctx = (NEP_CISS*)nep->data;
710: PetscFunctionBegin;
711: if (delta) *delta = ctx->delta;
712: if (spur) *spur = ctx->spurious_threshold;
713: PetscFunctionReturn(PETSC_SUCCESS);
714: }
716: /*@
717: NEPCISSGetThreshold - Gets the values of various threshold parameters in
718: the CISS solver.
720: Not Collective
722: Input Parameter:
723: . nep - the nonlinear eigensolver context
725: Output Parameters:
726: + delta - threshold for numerical rank
727: - spur - spurious threshold (to discard spurious eigenpairs)
729: Level: advanced
731: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSSetThreshold()`
732: @*/
733: PetscErrorCode NEPCISSGetThreshold(NEP nep,PetscReal *delta,PetscReal *spur)
734: {
735: PetscFunctionBegin;
737: PetscUseMethod(nep,"NEPCISSGetThreshold_C",(NEP,PetscReal*,PetscReal*),(nep,delta,spur));
738: PetscFunctionReturn(PETSC_SUCCESS);
739: }
741: static PetscErrorCode NEPCISSSetRefinement_CISS(NEP nep,PetscInt inner,PetscInt blsize)
742: {
743: NEP_CISS *ctx = (NEP_CISS*)nep->data;
745: PetscFunctionBegin;
746: if (inner == PETSC_DETERMINE) {
747: ctx->refine_inner = 0;
748: } else if (inner != PETSC_CURRENT) {
749: PetscCheck(inner>=0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The refine inner argument must be >= 0");
750: ctx->refine_inner = inner;
751: }
752: if (blsize == PETSC_DETERMINE) {
753: ctx->refine_blocksize = 0;
754: } else if (blsize != PETSC_CURRENT) {
755: PetscCheck(blsize>=0,PetscObjectComm((PetscObject)nep),PETSC_ERR_ARG_OUTOFRANGE,"The refine blocksize argument must be >= 0");
756: ctx->refine_blocksize = blsize;
757: }
758: PetscFunctionReturn(PETSC_SUCCESS);
759: }
761: /*@
762: NEPCISSSetRefinement - Sets the values of various refinement parameters
763: in the CISS solver.
765: Logically Collective
767: Input Parameters:
768: + nep - the nonlinear eigensolver context
769: . inner - number of iterative refinement iterations (inner loop)
770: - blsize - number of iterative refinement iterations (blocksize loop)
772: Options Database Keys:
773: + -nep_ciss_refine_inner \<inner\> - sets number of inner iterations
774: - -nep_ciss_refine_blocksize \<blsize\> - sets number of blocksize iterations
776: Note:
777: `PETSC_CURRENT` can be used to preserve the current value of any of the
778: arguments, and `PETSC_DETERMINE` to set them to a default of 0 (no refinement).
780: Level: advanced
782: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSGetRefinement()`
783: @*/
784: PetscErrorCode NEPCISSSetRefinement(NEP nep,PetscInt inner,PetscInt blsize)
785: {
786: PetscFunctionBegin;
790: PetscTryMethod(nep,"NEPCISSSetRefinement_C",(NEP,PetscInt,PetscInt),(nep,inner,blsize));
791: PetscFunctionReturn(PETSC_SUCCESS);
792: }
794: static PetscErrorCode NEPCISSGetRefinement_CISS(NEP nep,PetscInt *inner,PetscInt *blsize)
795: {
796: NEP_CISS *ctx = (NEP_CISS*)nep->data;
798: PetscFunctionBegin;
799: if (inner) *inner = ctx->refine_inner;
800: if (blsize) *blsize = ctx->refine_blocksize;
801: PetscFunctionReturn(PETSC_SUCCESS);
802: }
804: /*@
805: NEPCISSGetRefinement - Gets the values of various refinement parameters
806: in the CISS solver.
808: Not Collective
810: Input Parameter:
811: . nep - the nonlinear eigensolver context
813: Output Parameters:
814: + inner - number of iterative refinement iterations (inner loop)
815: - blsize - number of iterative refinement iterations (blocksize loop)
817: Level: advanced
819: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSSetRefinement()`
820: @*/
821: PetscErrorCode NEPCISSGetRefinement(NEP nep, PetscInt *inner, PetscInt *blsize)
822: {
823: PetscFunctionBegin;
825: PetscUseMethod(nep,"NEPCISSGetRefinement_C",(NEP,PetscInt*,PetscInt*),(nep,inner,blsize));
826: PetscFunctionReturn(PETSC_SUCCESS);
827: }
829: static PetscErrorCode NEPCISSSetExtraction_CISS(NEP nep,NEPCISSExtraction extraction)
830: {
831: NEP_CISS *ctx = (NEP_CISS*)nep->data;
833: PetscFunctionBegin;
834: if (ctx->extraction != extraction) {
835: ctx->extraction = extraction;
836: nep->state = NEP_STATE_INITIAL;
837: }
838: PetscFunctionReturn(PETSC_SUCCESS);
839: }
841: /*@
842: NEPCISSSetExtraction - Sets the extraction technique used in the CISS solver.
844: Logically Collective
846: Input Parameters:
847: + nep - the nonlinear eigensolver context
848: - extraction - the extraction technique, see `NEPCISSExtraction` for possible values
850: Options Database Key:
851: . -nep_ciss_extraction \<extraction\> - Sets the extraction technique, either `ritz`, `hankel` or `caa`
853: Notes:
854: By default, the Rayleigh-Ritz extraction is used (`NEP_CISS_EXTRACTION_RITZ`).
856: If the `hankel` or the `caa` option is specified (`NEP_CISS_EXTRACTION_HANKEL` or
857: `NEP_CISS_EXTRACTION_CAA`), then the block Hankel method, or the communication-avoiding
858: Arnoldi method, respectively, is used for extracting eigenpairs.
860: Level: advanced
862: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSGetExtraction()`, `NEPCISSExtraction`
863: @*/
864: PetscErrorCode NEPCISSSetExtraction(NEP nep,NEPCISSExtraction extraction)
865: {
866: PetscFunctionBegin;
869: PetscTryMethod(nep,"NEPCISSSetExtraction_C",(NEP,NEPCISSExtraction),(nep,extraction));
870: PetscFunctionReturn(PETSC_SUCCESS);
871: }
873: static PetscErrorCode NEPCISSGetExtraction_CISS(NEP nep,NEPCISSExtraction *extraction)
874: {
875: NEP_CISS *ctx = (NEP_CISS*)nep->data;
877: PetscFunctionBegin;
878: *extraction = ctx->extraction;
879: PetscFunctionReturn(PETSC_SUCCESS);
880: }
882: /*@
883: NEPCISSGetExtraction - Gets the extraction technique used in the CISS solver.
885: Not Collective
887: Input Parameter:
888: . nep - the nonlinear eigensolver context
890: Output Parameter:
891: . extraction - extraction technique
893: Level: advanced
895: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSSetExtraction()`, `NEPCISSExtraction`
896: @*/
897: PetscErrorCode NEPCISSGetExtraction(NEP nep,NEPCISSExtraction *extraction)
898: {
899: PetscFunctionBegin;
901: PetscAssertPointer(extraction,2);
902: PetscUseMethod(nep,"NEPCISSGetExtraction_C",(NEP,NEPCISSExtraction*),(nep,extraction));
903: PetscFunctionReturn(PETSC_SUCCESS);
904: }
906: static PetscErrorCode NEPCISSGetKSPs_CISS(NEP nep,PetscInt *nsolve,KSP **ksp)
907: {
908: NEP_CISS *ctx = (NEP_CISS*)nep->data;
909: SlepcContourData contour;
910: PetscInt i;
911: PC pc;
912: MPI_Comm child;
914: PetscFunctionBegin;
915: if (!ctx->contour) { /* initialize contour data structure first */
916: PetscCall(RGCanUseConjugates(nep->rg,ctx->isreal,&ctx->useconj));
917: PetscCall(SlepcContourDataCreate(ctx->useconj?ctx->N/2:ctx->N,ctx->npart,(PetscObject)nep,&ctx->contour));
918: }
919: contour = ctx->contour;
920: if (!contour->ksp) {
921: PetscCall(PetscMalloc1(contour->npoints,&contour->ksp));
922: PetscCall(PetscSubcommGetChild(contour->subcomm,&child));
923: for (i=0;i<contour->npoints;i++) {
924: PetscCall(KSPCreate(child,&contour->ksp[i]));
925: PetscCall(PetscObjectIncrementTabLevel((PetscObject)contour->ksp[i],(PetscObject)nep,1));
926: PetscCall(KSPSetOptionsPrefix(contour->ksp[i],((PetscObject)nep)->prefix));
927: PetscCall(KSPAppendOptionsPrefix(contour->ksp[i],"nep_ciss_"));
928: PetscCall(PetscObjectSetOptions((PetscObject)contour->ksp[i],((PetscObject)nep)->options));
929: PetscCall(KSPSetErrorIfNotConverged(contour->ksp[i],PETSC_TRUE));
930: PetscCall(KSPSetTolerances(contour->ksp[i],SlepcDefaultTol(nep->tol),PETSC_CURRENT,PETSC_CURRENT,PETSC_CURRENT));
931: PetscCall(KSPGetPC(contour->ksp[i],&pc));
932: if ((nep->fui==NEP_USER_INTERFACE_SPLIT && nep->P) || (nep->fui==NEP_USER_INTERFACE_CALLBACK && nep->function_pre!=nep->function)) {
933: PetscCall(KSPSetType(contour->ksp[i],KSPBCGS));
934: PetscCall(PCSetType(pc,PCBJACOBI));
935: } else {
936: PetscCall(KSPSetType(contour->ksp[i],KSPPREONLY));
937: PetscCall(PCSetType(pc,PCLU));
938: }
939: }
940: }
941: if (nsolve) *nsolve = contour->npoints;
942: if (ksp) *ksp = contour->ksp;
943: PetscFunctionReturn(PETSC_SUCCESS);
944: }
946: /*@C
947: NEPCISSGetKSPs - Retrieve the array of linear solver objects associated with
948: the CISS solver.
950: Collective
952: Input Parameter:
953: . nep - the nonlinear eigensolver context
955: Output Parameters:
956: + nsolve - number of solver objects
957: - ksp - array of linear solver object
959: Notes:
960: The number of `KSP` solvers is equal to the number of integration points divided by
961: the number of partitions. This value is halved in the case of real matrices with
962: a region centered at the real axis.
964: Level: advanced
966: .seealso: [](ch:nep), `NEPCISS`, `NEPCISSSetSizes()`
967: @*/
968: PetscErrorCode NEPCISSGetKSPs(NEP nep,PetscInt *nsolve,KSP **ksp)
969: {
970: PetscFunctionBegin;
972: PetscUseMethod(nep,"NEPCISSGetKSPs_C",(NEP,PetscInt*,KSP**),(nep,nsolve,ksp));
973: PetscFunctionReturn(PETSC_SUCCESS);
974: }
976: static PetscErrorCode NEPReset_CISS(NEP nep)
977: {
978: NEP_CISS *ctx = (NEP_CISS*)nep->data;
980: PetscFunctionBegin;
981: PetscCall(BVDestroy(&ctx->S));
982: PetscCall(BVDestroy(&ctx->V));
983: PetscCall(BVDestroy(&ctx->Y));
984: PetscCall(SlepcContourDataReset(ctx->contour));
985: PetscCall(MatDestroy(&ctx->J));
986: PetscCall(BVDestroy(&ctx->pV));
987: if (ctx->extraction == NEP_CISS_EXTRACTION_RITZ && nep->fui==NEP_USER_INTERFACE_CALLBACK) PetscCall(PetscFree(ctx->dsctxf));
988: PetscFunctionReturn(PETSC_SUCCESS);
989: }
991: static PetscErrorCode NEPSetFromOptions_CISS(NEP nep,PetscOptionItems PetscOptionsObject)
992: {
993: NEP_CISS *ctx = (NEP_CISS*)nep->data;
994: PetscReal r1,r2;
995: PetscInt i,i1,i2,i3,i4,i5,i6,i7;
996: PetscBool b1,flg,flg2,flg3,flg4,flg5,flg6;
997: NEPCISSExtraction extraction;
999: PetscFunctionBegin;
1000: PetscOptionsHeadBegin(PetscOptionsObject,"NEP CISS Options");
1002: PetscCall(NEPCISSGetSizes(nep,&i1,&i2,&i3,&i4,&i5,&b1));
1003: PetscCall(PetscOptionsInt("-nep_ciss_integration_points","Number of integration points","NEPCISSSetSizes",i1,&i1,&flg));
1004: PetscCall(PetscOptionsInt("-nep_ciss_blocksize","Block size","NEPCISSSetSizes",i2,&i2,&flg2));
1005: PetscCall(PetscOptionsInt("-nep_ciss_moments","Moment size","NEPCISSSetSizes",i3,&i3,&flg3));
1006: PetscCall(PetscOptionsInt("-nep_ciss_partitions","Number of partitions","NEPCISSSetSizes",i4,&i4,&flg4));
1007: PetscCall(PetscOptionsInt("-nep_ciss_maxblocksize","Maximum block size","NEPCISSSetSizes",i5,&i5,&flg5));
1008: PetscCall(PetscOptionsBool("-nep_ciss_realmats","True if T(z) is real for real z","NEPCISSSetSizes",b1,&b1,&flg6));
1009: if (flg || flg2 || flg3 || flg4 || flg5 || flg6) PetscCall(NEPCISSSetSizes(nep,i1,i2,i3,i4,i5,b1));
1011: PetscCall(NEPCISSGetThreshold(nep,&r1,&r2));
1012: PetscCall(PetscOptionsReal("-nep_ciss_delta","Threshold for numerical rank","NEPCISSSetThreshold",r1,&r1,&flg));
1013: PetscCall(PetscOptionsReal("-nep_ciss_spurious_threshold","Threshold for the spurious eigenpairs","NEPCISSSetThreshold",r2,&r2,&flg2));
1014: if (flg || flg2) PetscCall(NEPCISSSetThreshold(nep,r1,r2));
1016: PetscCall(NEPCISSGetRefinement(nep,&i6,&i7));
1017: PetscCall(PetscOptionsInt("-nep_ciss_refine_inner","Number of inner iterative refinement iterations","NEPCISSSetRefinement",i6,&i6,&flg));
1018: PetscCall(PetscOptionsInt("-nep_ciss_refine_blocksize","Number of blocksize iterative refinement iterations","NEPCISSSetRefinement",i7,&i7,&flg2));
1019: if (flg || flg2) PetscCall(NEPCISSSetRefinement(nep,i6,i7));
1021: PetscCall(PetscOptionsEnum("-nep_ciss_extraction","Extraction technique","NEPCISSSetExtraction",NEPCISSExtractions,(PetscEnum)ctx->extraction,(PetscEnum*)&extraction,&flg));
1022: if (flg) PetscCall(NEPCISSSetExtraction(nep,extraction));
1024: PetscOptionsHeadEnd();
1026: if (!nep->rg) PetscCall(NEPGetRG(nep,&nep->rg));
1027: PetscCall(RGSetFromOptions(nep->rg)); /* this is necessary here to set useconj */
1028: if (!ctx->contour || !ctx->contour->ksp) PetscCall(NEPCISSGetKSPs(nep,NULL,NULL));
1029: PetscAssert(ctx->contour && ctx->contour->ksp,PetscObjectComm((PetscObject)nep),PETSC_ERR_PLIB,"Something went wrong with NEPCISSGetKSPs()");
1030: for (i=0;i<ctx->contour->npoints;i++) PetscCall(KSPSetFromOptions(ctx->contour->ksp[i]));
1031: PetscCall(PetscSubcommSetFromOptions(ctx->contour->subcomm));
1032: PetscFunctionReturn(PETSC_SUCCESS);
1033: }
1035: static PetscErrorCode NEPDestroy_CISS(NEP nep)
1036: {
1037: NEP_CISS *ctx = (NEP_CISS*)nep->data;
1039: PetscFunctionBegin;
1040: PetscCall(SlepcContourDataDestroy(&ctx->contour));
1041: PetscCall(PetscFree4(ctx->weight,ctx->omega,ctx->pp,ctx->sigma));
1042: PetscCall(PetscFree(nep->data));
1043: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSSetSizes_C",NULL));
1044: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetSizes_C",NULL));
1045: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSSetThreshold_C",NULL));
1046: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetThreshold_C",NULL));
1047: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSSetRefinement_C",NULL));
1048: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetRefinement_C",NULL));
1049: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSSetExtraction_C",NULL));
1050: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetExtraction_C",NULL));
1051: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetKSPs_C",NULL));
1052: PetscFunctionReturn(PETSC_SUCCESS);
1053: }
1055: static PetscErrorCode NEPView_CISS(NEP nep,PetscViewer viewer)
1056: {
1057: NEP_CISS *ctx = (NEP_CISS*)nep->data;
1058: PetscBool isascii;
1059: PetscViewer sviewer;
1061: PetscFunctionBegin;
1062: PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&isascii));
1063: if (isascii) {
1064: PetscCall(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));
1065: if (ctx->isreal) PetscCall(PetscViewerASCIIPrintf(viewer," exploiting symmetry of integration points\n"));
1066: PetscCall(PetscViewerASCIIPrintf(viewer," threshold { delta: %g, spurious threshold: %g }\n",(double)ctx->delta,(double)ctx->spurious_threshold));
1067: PetscCall(PetscViewerASCIIPrintf(viewer," iterative refinement { inner: %" PetscInt_FMT ", blocksize: %" PetscInt_FMT " }\n",ctx->refine_inner, ctx->refine_blocksize));
1068: PetscCall(PetscViewerASCIIPrintf(viewer," extraction: %s\n",NEPCISSExtractions[ctx->extraction]));
1069: if (!ctx->contour || !ctx->contour->ksp) PetscCall(NEPCISSGetKSPs(nep,NULL,NULL));
1070: PetscAssert(ctx->contour && ctx->contour->ksp,PetscObjectComm((PetscObject)nep),PETSC_ERR_PLIB,"Something went wrong with NEPCISSGetKSPs()");
1071: PetscCall(PetscViewerASCIIPushTab(viewer));
1072: if (ctx->npart>1 && ctx->contour->subcomm) {
1073: PetscCall(PetscViewerGetSubViewer(viewer,ctx->contour->subcomm->child,&sviewer));
1074: if (!ctx->contour->subcomm->color) PetscCall(KSPView(ctx->contour->ksp[0],sviewer));
1075: PetscCall(PetscViewerFlush(sviewer));
1076: PetscCall(PetscViewerRestoreSubViewer(viewer,ctx->contour->subcomm->child,&sviewer));
1077: /* extra call needed because of the two calls to PetscViewerASCIIPushSynchronized() in PetscViewerGetSubViewer() */
1078: PetscCall(PetscViewerASCIIPopSynchronized(viewer));
1079: } else PetscCall(KSPView(ctx->contour->ksp[0],viewer));
1080: PetscCall(PetscViewerASCIIPopTab(viewer));
1081: }
1082: PetscFunctionReturn(PETSC_SUCCESS);
1083: }
1085: static PetscErrorCode NEPSetDSType_CISS(NEP nep)
1086: {
1087: NEP_CISS *ctx = (NEP_CISS*)nep->data;
1089: PetscFunctionBegin;
1090: switch (ctx->extraction) {
1091: case NEP_CISS_EXTRACTION_RITZ: PetscCall(DSSetType(nep->ds,DSNEP)); break;
1092: case NEP_CISS_EXTRACTION_HANKEL: PetscCall(DSSetType(nep->ds,DSGNHEP)); break;
1093: case NEP_CISS_EXTRACTION_CAA: PetscCall(DSSetType(nep->ds,DSNHEP)); break;
1094: }
1095: PetscFunctionReturn(PETSC_SUCCESS);
1096: }
1098: /*MC
1099: NEPCISS - NEPCISS = "ciss" - A contour integral eigensolver based on the
1100: Sakurai-Sugiura scheme.
1102: Notes:
1103: This solver is based on the numerical contour integration idea
1104: proposed initially for linear problems by {cite:t}`Sak03`. In nonlinear
1105: eigenproblems, a Rayleigh-Ritz projection is done, resulting in
1106: a small dense nonlinear eigenproblem {cite:p}`Asa09,Yok13`.
1108: Contour integral methods are able to compute all eigenvalues
1109: lying inside a region of the complex plane. Use `NEPGetRG()` to
1110: specify the region. However, the computational cost is usually high
1111: because multiple linear systems must be solved. Use `NEPCISSGetKSPs()`
1112: to configure the `KSP` objects for this.
1114: Details of the implementation in SLEPc can be found in {cite:p}`Mae16`.
1116: Level: beginner
1118: .seealso: [](ch:nep), `NEP`, `NEPType`, `NEPSetType()`, `NEPGetRG()`
1119: M*/
1120: SLEPC_EXTERN PetscErrorCode NEPCreate_CISS(NEP nep)
1121: {
1122: NEP_CISS *ctx = (NEP_CISS*)nep->data;
1124: PetscFunctionBegin;
1125: PetscCall(PetscNew(&ctx));
1126: nep->data = ctx;
1127: /* set default values of parameters */
1128: ctx->N = 32;
1129: ctx->L = 16;
1130: ctx->M = ctx->N/4;
1131: ctx->delta = SLEPC_DEFAULT_TOL*1e-4;
1132: ctx->L_max = 64;
1133: ctx->spurious_threshold = PetscSqrtReal(SLEPC_DEFAULT_TOL);
1134: ctx->isreal = PETSC_FALSE;
1135: ctx->npart = 1;
1137: nep->useds = PETSC_TRUE;
1139: nep->ops->solve = NEPSolve_CISS;
1140: nep->ops->setup = NEPSetUp_CISS;
1141: nep->ops->setfromoptions = NEPSetFromOptions_CISS;
1142: nep->ops->reset = NEPReset_CISS;
1143: nep->ops->destroy = NEPDestroy_CISS;
1144: nep->ops->view = NEPView_CISS;
1145: nep->ops->setdstype = NEPSetDSType_CISS;
1147: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSSetSizes_C",NEPCISSSetSizes_CISS));
1148: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetSizes_C",NEPCISSGetSizes_CISS));
1149: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSSetThreshold_C",NEPCISSSetThreshold_CISS));
1150: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetThreshold_C",NEPCISSGetThreshold_CISS));
1151: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSSetRefinement_C",NEPCISSSetRefinement_CISS));
1152: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetRefinement_C",NEPCISSGetRefinement_CISS));
1153: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSSetExtraction_C",NEPCISSSetExtraction_CISS));
1154: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetExtraction_C",NEPCISSGetExtraction_CISS));
1155: PetscCall(PetscObjectComposeFunction((PetscObject)nep,"NEPCISSGetKSPs_C",NEPCISSGetKSPs_CISS));
1156: PetscFunctionReturn(PETSC_SUCCESS);
1157: }