Actual source code: ciss.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] T. Sakurai and H. Sugiura, "A projection method for generalized
26: eigenvalue problems", J. Comput. Appl. Math. 159:119-128, 2003.
28: [2] T. Sakurai and H. Tadano, "CIRR: a Rayleigh-Ritz type method with
29: contour integral for generalized eigenvalue problems", Hokkaido
30: Math. J. 36:745-757, 2007.
31: */
33: #include <slepc/private/epsimpl.h>
34: #include <slepc/private/slepccontour.h>
35: #include <slepcblaslapack.h>
37: typedef struct {
38: /* user parameters */
39: PetscInt N; /* number of integration points (32) */
40: PetscInt L; /* block size (16) */
41: PetscInt M; /* moment degree (N/4 = 4) */
42: PetscReal delta; /* threshold of singular value (1e-12) */
43: PetscInt L_max; /* maximum number of columns of the source matrix V */
44: PetscReal spurious_threshold; /* discard spurious eigenpairs */
45: PetscBool isreal; /* A and B are real */
46: PetscInt npart; /* number of partitions */
47: PetscInt refine_inner;
48: PetscInt refine_blocksize;
49: EPSCISSQuadRule quad;
50: EPSCISSExtraction extraction;
51: PetscBool usest;
52: /* private data */
53: SlepcContourData contour;
54: PetscReal *sigma; /* threshold for numerical rank */
55: PetscScalar *weight;
56: PetscScalar *omega;
57: PetscScalar *pp;
58: BV V;
59: BV S;
60: BV pV;
61: BV Y;
62: PetscBool useconj;
63: PetscBool usest_set; /* whether the user set the usest flag or not */
64: PetscObjectId rgid;
65: PetscObjectState rgstate;
66: } EPS_CISS;
68: /*
69: Set up KSP solvers for every integration point, only called if !ctx->usest
70: */
71: static PetscErrorCode EPSCISSSetUp(EPS eps,Mat A,Mat B,Mat Pa,Mat Pb)
72: {
73: EPS_CISS *ctx = (EPS_CISS*)eps->data;
74: SlepcContourData contour;
75: PetscInt i,p_id,nsplit;
76: Mat Amat,Pmat;
77: MatStructure str,strp;
79: PetscFunctionBegin;
80: if (!ctx->contour || !ctx->contour->ksp) PetscCall(EPSCISSGetKSPs(eps,NULL,NULL));
81: PetscAssert(ctx->contour && ctx->contour->ksp,PetscObjectComm((PetscObject)eps),PETSC_ERR_PLIB,"Something went wrong with EPSCISSGetKSPs()");
82: contour = ctx->contour;
83: PetscCall(STGetMatStructure(eps->st,&str));
84: PetscCall(STGetSplitPreconditionerInfo(eps->st,&nsplit,&strp));
85: for (i=0;i<contour->npoints;i++) {
86: p_id = i*contour->subcomm->n + contour->subcomm->color;
87: PetscCall(MatDuplicate(A,MAT_COPY_VALUES,&Amat));
88: if (B) PetscCall(MatAXPY(Amat,-ctx->omega[p_id],B,str));
89: else PetscCall(MatShift(Amat,-ctx->omega[p_id]));
90: if (nsplit) {
91: PetscCall(MatDuplicate(Pa,MAT_COPY_VALUES,&Pmat));
92: if (Pb) PetscCall(MatAXPY(Pmat,-ctx->omega[p_id],Pb,strp));
93: else PetscCall(MatShift(Pmat,-ctx->omega[p_id]));
94: } else Pmat = Amat;
95: PetscCall(EPS_KSPSetOperators(contour->ksp[i],Amat,Amat));
96: PetscCall(MatDestroy(&Amat));
97: if (nsplit) PetscCall(MatDestroy(&Pmat));
98: }
99: PetscFunctionReturn(PETSC_SUCCESS);
100: }
102: /*
103: Y_i = (A-z_i B)^{-1}BV for every integration point, Y=[Y_i] is in the context
104: */
105: static PetscErrorCode EPSCISSSolve(EPS eps,Mat B,BV V,PetscInt L_start,PetscInt L_end)
106: {
107: EPS_CISS *ctx = (EPS_CISS*)eps->data;
108: SlepcContourData contour;
109: PetscInt i,p_id;
110: Mat MV,BMV=NULL,MC;
111: KSP ksp;
113: PetscFunctionBegin;
114: if (!ctx->contour || !ctx->contour->ksp) PetscCall(EPSCISSGetKSPs(eps,NULL,NULL));
115: contour = ctx->contour;
116: PetscAssert(ctx->contour && ctx->contour->ksp,PetscObjectComm((PetscObject)eps),PETSC_ERR_PLIB,"Something went wrong with EPSCISSGetKSPs()");
117: PetscCall(BVSetActiveColumns(V,L_start,L_end));
118: PetscCall(BVGetMat(V,&MV));
119: for (i=0;i<contour->npoints;i++) {
120: p_id = i*contour->subcomm->n + contour->subcomm->color;
121: if (ctx->usest) {
122: PetscCall(STSetShift(eps->st,ctx->omega[p_id]));
123: PetscCall(STGetKSP(eps->st,&ksp));
124: } else ksp = contour->ksp[i];
125: PetscCall(BVSetActiveColumns(ctx->Y,i*ctx->L+L_start,i*ctx->L+L_end));
126: PetscCall(BVGetMat(ctx->Y,&MC));
127: if (B) {
128: if (!i) {
129: PetscCall(MatProductCreate(B,MV,NULL,&BMV));
130: PetscCall(MatProductSetType(BMV,MATPRODUCT_AB));
131: PetscCall(MatProductSetFromOptions(BMV));
132: PetscCall(MatProductSymbolic(BMV));
133: }
134: PetscCall(MatProductNumeric(BMV));
135: PetscCall(KSPMatSolve(ksp,BMV,MC));
136: } else PetscCall(KSPMatSolve(ksp,MV,MC));
137: PetscCall(BVRestoreMat(ctx->Y,&MC));
138: if (ctx->usest && i<contour->npoints-1) PetscCall(KSPReset(ksp));
139: }
140: PetscCall(MatDestroy(&BMV));
141: PetscCall(BVRestoreMat(V,&MV));
142: PetscFunctionReturn(PETSC_SUCCESS);
143: }
145: static PetscErrorCode rescale_eig(EPS eps,PetscInt nv)
146: {
147: EPS_CISS *ctx = (EPS_CISS*)eps->data;
148: PetscInt i;
149: PetscScalar center;
150: PetscReal radius,a,b,c,d,rgscale;
151: #if defined(PETSC_USE_COMPLEX)
152: PetscReal start_ang,end_ang,vscale,theta;
153: #endif
154: PetscBool isring,isellipse,isinterval;
156: PetscFunctionBegin;
157: PetscCall(PetscObjectTypeCompare((PetscObject)eps->rg,RGELLIPSE,&isellipse));
158: PetscCall(PetscObjectTypeCompare((PetscObject)eps->rg,RGRING,&isring));
159: PetscCall(PetscObjectTypeCompare((PetscObject)eps->rg,RGINTERVAL,&isinterval));
160: PetscCall(RGGetScale(eps->rg,&rgscale));
161: if (isinterval) {
162: PetscCall(RGIntervalGetEndpoints(eps->rg,NULL,NULL,&c,&d));
163: if (c==d) {
164: for (i=0;i<nv;i++) {
165: #if defined(PETSC_USE_COMPLEX)
166: eps->eigr[i] = PetscRealPart(eps->eigr[i]);
167: #else
168: eps->eigi[i] = 0;
169: #endif
170: }
171: }
172: }
173: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) {
174: if (isellipse) {
175: PetscCall(RGEllipseGetParameters(eps->rg,¢er,&radius,NULL));
176: for (i=0;i<nv;i++) eps->eigr[i] = rgscale*(center + radius*eps->eigr[i]);
177: } else if (isinterval) {
178: PetscCall(RGIntervalGetEndpoints(eps->rg,&a,&b,&c,&d));
179: if (ctx->quad == EPS_CISS_QUADRULE_CHEBYSHEV) {
180: for (i=0;i<nv;i++) {
181: if (c==d) eps->eigr[i] = ((eps->eigr[i]+1.0)*(b-a)/2.0+a)*rgscale;
182: if (a==b) {
183: #if defined(PETSC_USE_COMPLEX)
184: eps->eigr[i] = ((eps->eigr[i]+1.0)*(d-c)/2.0+c)*rgscale*PETSC_i;
185: #else
186: SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Integration points on a vertical line require complex arithmetic");
187: #endif
188: }
189: }
190: } else {
191: center = (b+a)/2.0+(d+c)/2.0*PETSC_PI;
192: radius = PetscSqrtReal(PetscPowRealInt((b-a)/2.0,2)+PetscPowRealInt((d-c)/2.0,2));
193: for (i=0;i<nv;i++) eps->eigr[i] = center + radius*eps->eigr[i];
194: }
195: } else if (isring) { /* only supported in complex scalars */
196: #if defined(PETSC_USE_COMPLEX)
197: PetscCall(RGRingGetParameters(eps->rg,¢er,&radius,&vscale,&start_ang,&end_ang,NULL));
198: if (ctx->quad == EPS_CISS_QUADRULE_CHEBYSHEV) {
199: for (i=0;i<nv;i++) {
200: theta = (start_ang*2.0+(end_ang-start_ang)*(PetscRealPart(eps->eigr[i])+1.0))*PETSC_PI;
201: eps->eigr[i] = rgscale*center + (rgscale*radius+PetscImaginaryPart(eps->eigr[i]))*PetscCMPLX(PetscCosReal(theta),vscale*PetscSinReal(theta));
202: }
203: } else {
204: for (i=0;i<nv;i++) eps->eigr[i] = rgscale*(center + radius*eps->eigr[i]);
205: }
206: #endif
207: }
208: }
209: PetscFunctionReturn(PETSC_SUCCESS);
210: }
212: static PetscErrorCode EPSSetUp_CISS(EPS eps)
213: {
214: EPS_CISS *ctx = (EPS_CISS*)eps->data;
215: SlepcContourData contour;
216: PetscBool istrivial,isring,isellipse,isinterval,flg;
217: PetscReal c,d;
218: PetscInt nsplit;
219: PetscRandom rand;
220: PetscObjectId id;
221: PetscObjectState state;
222: Mat A[2],Psplit[2];
223: Vec v0;
225: PetscFunctionBegin;
226: EPSCheckNotStructured(eps);
227: if (eps->ncv==PETSC_DETERMINE) {
228: if (eps->nev==0) eps->nev = 1;
229: eps->ncv = ctx->L_max*ctx->M;
230: if (eps->ncv>eps->n) {
231: eps->ncv = eps->n;
232: ctx->L_max = eps->ncv/ctx->M;
233: PetscCheck(ctx->L_max,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"Cannot adjust solver parameters, try setting a smaller value of M (moment size)");
234: }
235: } else {
236: PetscCall(EPSSetDimensions_Default(eps,&eps->nev,&eps->ncv,&eps->mpd));
237: ctx->L_max = eps->ncv/ctx->M;
238: if (!ctx->L_max) {
239: ctx->L_max = 1;
240: eps->ncv = ctx->L_max*ctx->M;
241: }
242: }
243: ctx->L = PetscMin(ctx->L,ctx->L_max);
244: if (eps->max_it==PETSC_DETERMINE) eps->max_it = 5;
245: if (eps->mpd==PETSC_DETERMINE) eps->mpd = eps->ncv;
246: if (!eps->which) eps->which = EPS_ALL;
247: PetscCheck(eps->which==EPS_ALL,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"This solver supports only computing all eigenvalues");
248: EPSCheckUnsupported(eps,EPS_FEATURE_BALANCE | EPS_FEATURE_ARBITRARY | EPS_FEATURE_EXTRACTION | EPS_FEATURE_STOPPING | EPS_FEATURE_TWOSIDED);
250: /* check region */
251: PetscCall(RGIsTrivial(eps->rg,&istrivial));
252: PetscCheck(!istrivial,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"CISS requires a nontrivial region, e.g. -rg_type ellipse ...");
253: PetscCall(RGGetComplement(eps->rg,&flg));
254: PetscCheck(!flg,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"A region with complement flag set is not allowed");
255: PetscCall(PetscObjectTypeCompare((PetscObject)eps->rg,RGELLIPSE,&isellipse));
256: PetscCall(PetscObjectTypeCompare((PetscObject)eps->rg,RGRING,&isring));
257: PetscCall(PetscObjectTypeCompare((PetscObject)eps->rg,RGINTERVAL,&isinterval));
258: PetscCheck(isellipse || isring || isinterval,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"Currently only implemented for interval, elliptic or ring regions");
260: /* if the region has changed, then reset contour data */
261: PetscCall(PetscObjectGetId((PetscObject)eps->rg,&id));
262: PetscCall(PetscObjectStateGet((PetscObject)eps->rg,&state));
263: if (ctx->rgid && (id != ctx->rgid || state != ctx->rgstate)) {
264: PetscCall(SlepcContourDataDestroy(&ctx->contour));
265: PetscCall(PetscInfo(eps,"Resetting the contour data structure due to a change of region\n"));
266: ctx->rgid = id; ctx->rgstate = state;
267: }
269: #if !defined(PETSC_USE_COMPLEX)
270: PetscCheck(!isring,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"Ring region only supported for complex scalars");
271: #endif
272: if (isinterval) {
273: PetscCall(RGIntervalGetEndpoints(eps->rg,NULL,NULL,&c,&d));
274: #if !defined(PETSC_USE_COMPLEX)
275: PetscCheck(c==d && c==0.0,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"In real scalars, endpoints of the imaginary axis must be both zero");
276: #endif
277: if (!ctx->quad && c==d) ctx->quad = EPS_CISS_QUADRULE_CHEBYSHEV;
278: }
279: if (!ctx->quad) ctx->quad = EPS_CISS_QUADRULE_TRAPEZOIDAL;
281: /* create contour data structure */
282: if (!ctx->contour) {
283: PetscCall(RGCanUseConjugates(eps->rg,ctx->isreal,&ctx->useconj));
284: PetscCall(SlepcContourDataCreate(ctx->useconj?ctx->N/2:ctx->N,ctx->npart,(PetscObject)eps,&ctx->contour));
285: }
287: PetscCall(EPSAllocateSolution(eps,0));
288: PetscCall(BVGetRandomContext(eps->V,&rand)); /* make sure the random context is available when duplicating */
289: if (ctx->weight) PetscCall(PetscFree4(ctx->weight,ctx->omega,ctx->pp,ctx->sigma));
290: PetscCall(PetscMalloc4(ctx->N,&ctx->weight,ctx->N+1,&ctx->omega,ctx->N,&ctx->pp,ctx->L_max*ctx->M,&ctx->sigma));
292: /* allocate basis vectors */
293: PetscCall(BVDestroy(&ctx->S));
294: PetscCall(BVDuplicateResize(eps->V,ctx->L*ctx->M,&ctx->S));
295: PetscCall(BVDestroy(&ctx->V));
296: PetscCall(BVDuplicateResize(eps->V,ctx->L,&ctx->V));
298: PetscCall(STGetMatrix(eps->st,0,&A[0]));
299: PetscCall(MatIsShell(A[0],&flg));
300: PetscCheck(!flg,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"Matrix type shell is not supported in this solver");
301: if (eps->isgeneralized) PetscCall(STGetMatrix(eps->st,1,&A[1]));
303: if (!ctx->usest_set) ctx->usest = (ctx->npart>1)? PETSC_FALSE: PETSC_TRUE;
304: PetscCheck(!ctx->usest || ctx->npart==1,PetscObjectComm((PetscObject)eps),PETSC_ERR_SUP,"The usest flag is not supported when partitions > 1");
306: /* check if a user-defined split preconditioner has been set */
307: PetscCall(STGetSplitPreconditionerInfo(eps->st,&nsplit,NULL));
308: if (nsplit) {
309: PetscCall(STGetSplitPreconditionerTerm(eps->st,0,&Psplit[0]));
310: if (eps->isgeneralized) PetscCall(STGetSplitPreconditionerTerm(eps->st,1,&Psplit[1]));
311: }
313: contour = ctx->contour;
314: PetscCall(SlepcContourRedundantMat(contour,eps->isgeneralized?2:1,A,nsplit?Psplit:NULL));
315: if (contour->pA) {
316: PetscCall(BVGetColumn(ctx->V,0,&v0));
317: PetscCall(SlepcContourScatterCreate(contour,v0));
318: PetscCall(BVRestoreColumn(ctx->V,0,&v0));
319: PetscCall(BVDestroy(&ctx->pV));
320: PetscCall(BVCreate(PetscObjectComm((PetscObject)contour->xsub),&ctx->pV));
321: PetscCall(BVSetSizesFromVec(ctx->pV,contour->xsub,eps->n));
322: PetscCall(BVSetFromOptions(ctx->pV));
323: PetscCall(BVResize(ctx->pV,ctx->L,PETSC_FALSE));
324: }
326: EPSCheckDefinite(eps);
327: EPSCheckSinvertCondition(eps,ctx->usest," (with the usest flag set)");
329: PetscCall(BVDestroy(&ctx->Y));
330: if (contour->pA) {
331: PetscCall(BVCreate(PetscObjectComm((PetscObject)contour->xsub),&ctx->Y));
332: PetscCall(BVSetSizesFromVec(ctx->Y,contour->xsub,eps->n));
333: PetscCall(BVSetFromOptions(ctx->Y));
334: PetscCall(BVResize(ctx->Y,contour->npoints*ctx->L,PETSC_FALSE));
335: } else PetscCall(BVDuplicateResize(eps->V,contour->npoints*ctx->L,&ctx->Y));
337: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) PetscCall(DSSetType(eps->ds,DSGNHEP));
338: else if (eps->isgeneralized) {
339: if (eps->ishermitian && eps->ispositive) PetscCall(DSSetType(eps->ds,DSGHEP));
340: else PetscCall(DSSetType(eps->ds,DSGNHEP));
341: } else {
342: if (eps->ishermitian) PetscCall(DSSetType(eps->ds,DSHEP));
343: else PetscCall(DSSetType(eps->ds,DSNHEP));
344: }
345: PetscCall(DSAllocate(eps->ds,eps->ncv));
347: #if !defined(PETSC_USE_COMPLEX)
348: PetscCall(EPSSetWorkVecs(eps,3));
349: if (!eps->ishermitian) PetscCall(PetscInfo(eps,"Warning: complex eigenvalues are not calculated exactly without --with-scalar-type=complex in PETSc\n"));
350: #else
351: PetscCall(EPSSetWorkVecs(eps,2));
352: #endif
353: PetscFunctionReturn(PETSC_SUCCESS);
354: }
356: static PetscErrorCode EPSSetUpSort_CISS(EPS eps)
357: {
358: SlepcSC sc;
360: PetscFunctionBegin;
361: /* fill sorting criterion context */
362: eps->sc->comparison = SlepcCompareSmallestReal;
363: eps->sc->comparisonctx = NULL;
364: eps->sc->map = NULL;
365: eps->sc->mapobj = NULL;
367: /* fill sorting criterion for DS */
368: PetscCall(DSGetSlepcSC(eps->ds,&sc));
369: sc->comparison = SlepcCompareLargestMagnitude;
370: sc->comparisonctx = NULL;
371: sc->map = NULL;
372: sc->mapobj = NULL;
373: PetscFunctionReturn(PETSC_SUCCESS);
374: }
376: static PetscErrorCode EPSSolve_CISS(EPS eps)
377: {
378: EPS_CISS *ctx = (EPS_CISS*)eps->data;
379: SlepcContourData contour = ctx->contour;
380: Mat A,B,X,M,pA,pB,T,J,Pa=NULL,Pb=NULL;
381: BV V;
382: PetscInt i,j,ld,nmat,L_add=0,nv=0,L_base=ctx->L,inner,nlocal,*inside,nsplit;
383: PetscScalar *Mu,*H0,*H1=NULL,*rr,*temp;
384: PetscReal error,max_error,norm;
385: PetscBool *fl1;
386: Vec si,si1=NULL,w[3];
387: PetscRandom rand;
388: #if defined(PETSC_USE_COMPLEX)
389: PetscBool isellipse;
390: PetscReal est_eig,eta;
391: #else
392: PetscReal normi;
393: #endif
395: PetscFunctionBegin;
396: w[0] = eps->work[0];
397: #if defined(PETSC_USE_COMPLEX)
398: w[1] = NULL;
399: #else
400: w[1] = eps->work[2];
401: #endif
402: w[2] = eps->work[1];
403: PetscCall(VecGetLocalSize(w[0],&nlocal));
404: PetscCall(DSGetLeadingDimension(eps->ds,&ld));
405: PetscCall(RGComputeQuadrature(eps->rg,ctx->quad==EPS_CISS_QUADRULE_CHEBYSHEV?RG_QUADRULE_CHEBYSHEV:RG_QUADRULE_TRAPEZOIDAL,ctx->N,ctx->omega,ctx->pp,ctx->weight));
406: PetscCall(STGetNumMatrices(eps->st,&nmat));
407: PetscCall(STGetMatrix(eps->st,0,&A));
408: if (nmat>1) PetscCall(STGetMatrix(eps->st,1,&B));
409: else B = NULL;
410: J = (contour->pA && nmat>1)? contour->pA[1]: B;
411: V = contour->pA? ctx->pV: ctx->V;
412: if (!ctx->usest) {
413: T = contour->pA? contour->pA[0]: A;
414: PetscCall(STGetSplitPreconditionerInfo(eps->st,&nsplit,NULL));
415: if (nsplit) {
416: if (contour->pA) {
417: Pa = contour->pP[0];
418: if (nsplit>1) Pb = contour->pP[1];
419: } else {
420: PetscCall(STGetSplitPreconditionerTerm(eps->st,0,&Pa));
421: if (nsplit>1) PetscCall(STGetSplitPreconditionerTerm(eps->st,1,&Pb));
422: }
423: }
424: PetscCall(EPSCISSSetUp(eps,T,J,Pa,Pb));
425: }
426: PetscCall(BVSetActiveColumns(ctx->V,0,ctx->L));
427: PetscCall(BVSetRandomSign(ctx->V));
428: PetscCall(BVGetRandomContext(ctx->V,&rand));
430: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
431: PetscCall(EPSCISSSolve(eps,J,V,0,ctx->L));
432: #if defined(PETSC_USE_COMPLEX)
433: PetscCall(PetscObjectTypeCompare((PetscObject)eps->rg,RGELLIPSE,&isellipse));
434: if (isellipse) {
435: PetscCall(BVTraceQuadrature(ctx->Y,ctx->V,ctx->L,ctx->L,ctx->weight,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj,&est_eig));
436: PetscCall(PetscInfo(eps,"Estimated eigenvalue count: %f\n",(double)est_eig));
437: eta = PetscPowReal(10.0,-PetscLog10Real(eps->tol)/ctx->N);
438: L_add = PetscMax(0,(PetscInt)PetscCeilReal((est_eig*eta)/ctx->M)-ctx->L);
439: if (L_add>ctx->L_max-ctx->L) {
440: PetscCall(PetscInfo(eps,"Number of eigenvalues inside the contour path may be too large\n"));
441: L_add = ctx->L_max-ctx->L;
442: }
443: }
444: #endif
445: if (L_add>0) {
446: PetscCall(PetscInfo(eps,"Changing L %" PetscInt_FMT " -> %" PetscInt_FMT " by Estimate #Eig\n",ctx->L,ctx->L+L_add));
447: PetscCall(BVCISSResizeBases(ctx->S,contour->pA?ctx->pV:ctx->V,ctx->Y,ctx->L,ctx->L+L_add,ctx->M,contour->npoints));
448: PetscCall(BVSetActiveColumns(ctx->V,ctx->L,ctx->L+L_add));
449: PetscCall(BVSetRandomSign(ctx->V));
450: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
451: ctx->L += L_add;
452: PetscCall(EPSCISSSolve(eps,J,V,ctx->L-L_add,ctx->L));
453: }
454: PetscCall(PetscMalloc2(ctx->L*ctx->L*ctx->M*2,&Mu,ctx->L*ctx->M*ctx->L*ctx->M,&H0));
455: for (i=0;i<ctx->refine_blocksize;i++) {
456: PetscCall(BVDotQuadrature(ctx->Y,(contour->pA)?ctx->pV:ctx->V,Mu,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->subcomm,contour->npoints,ctx->useconj));
457: PetscCall(CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0));
458: PetscCall(PetscLogEventBegin(EPS_CISS_SVD,eps,0,0,0));
459: PetscCall(SlepcCISS_BH_SVD(H0,ctx->L*ctx->M,ctx->delta,ctx->sigma,&nv));
460: PetscCall(PetscLogEventEnd(EPS_CISS_SVD,eps,0,0,0));
461: if (ctx->sigma[0]<=ctx->delta || nv < ctx->L*ctx->M || ctx->L == ctx->L_max) break;
462: L_add = L_base;
463: if (ctx->L+L_add>ctx->L_max) L_add = ctx->L_max-ctx->L;
464: PetscCall(PetscInfo(eps,"Changing L %" PetscInt_FMT " -> %" PetscInt_FMT " by SVD(H0)\n",ctx->L,ctx->L+L_add));
465: PetscCall(BVCISSResizeBases(ctx->S,contour->pA?ctx->pV:ctx->V,ctx->Y,ctx->L,ctx->L+L_add,ctx->M,contour->npoints));
466: PetscCall(BVSetActiveColumns(ctx->V,ctx->L,ctx->L+L_add));
467: PetscCall(BVSetRandomSign(ctx->V));
468: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
469: ctx->L += L_add;
470: PetscCall(EPSCISSSolve(eps,J,V,ctx->L-L_add,ctx->L));
471: if (L_add) {
472: PetscCall(PetscFree2(Mu,H0));
473: PetscCall(PetscMalloc2(ctx->L*ctx->L*ctx->M*2,&Mu,ctx->L*ctx->M*ctx->L*ctx->M,&H0));
474: }
475: }
476: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) PetscCall(PetscMalloc1(ctx->L*ctx->M*ctx->L*ctx->M,&H1));
478: while (eps->reason == EPS_CONVERGED_ITERATING) {
479: eps->its++;
480: for (inner=0;inner<=ctx->refine_inner;inner++) {
481: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) {
482: PetscCall(BVDotQuadrature(ctx->Y,(contour->pA)?ctx->pV:ctx->V,Mu,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->subcomm,contour->npoints,ctx->useconj));
483: PetscCall(CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0));
484: PetscCall(PetscLogEventBegin(EPS_CISS_SVD,eps,0,0,0));
485: PetscCall(SlepcCISS_BH_SVD(H0,ctx->L*ctx->M,ctx->delta,ctx->sigma,&nv));
486: PetscCall(PetscLogEventEnd(EPS_CISS_SVD,eps,0,0,0));
487: break;
488: } else {
489: PetscCall(BVSumQuadrature(ctx->S,ctx->Y,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj));
490: PetscCall(BVSetActiveColumns(ctx->S,0,ctx->L));
491: PetscCall(BVSetActiveColumns(ctx->V,0,ctx->L));
492: PetscCall(BVCopy(ctx->S,ctx->V));
493: PetscCall(BVSVDAndRank(ctx->S,ctx->M,ctx->L,ctx->delta,BV_SVD_METHOD_REFINE,H0,ctx->sigma,&nv));
494: if (ctx->sigma[0]>ctx->delta && nv==ctx->L*ctx->M && inner!=ctx->refine_inner) {
495: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
496: PetscCall(EPSCISSSolve(eps,J,V,0,ctx->L));
497: } else break;
498: }
499: }
500: eps->nconv = 0;
501: if (nv == 0) eps->reason = EPS_CONVERGED_TOL;
502: else {
503: PetscCall(DSSetDimensions(eps->ds,nv,0,0));
504: PetscCall(DSSetState(eps->ds,DS_STATE_RAW));
506: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) {
507: PetscCall(CISS_BlockHankel(Mu,0,ctx->L,ctx->M,H0));
508: PetscCall(CISS_BlockHankel(Mu,1,ctx->L,ctx->M,H1));
509: PetscCall(DSGetArray(eps->ds,DS_MAT_A,&temp));
510: for (j=0;j<nv;j++) {
511: for (i=0;i<nv;i++) {
512: temp[i+j*ld] = H1[i+j*ctx->L*ctx->M];
513: }
514: }
515: PetscCall(DSRestoreArray(eps->ds,DS_MAT_A,&temp));
516: PetscCall(DSGetArray(eps->ds,DS_MAT_B,&temp));
517: for (j=0;j<nv;j++) {
518: for (i=0;i<nv;i++) {
519: temp[i+j*ld] = H0[i+j*ctx->L*ctx->M];
520: }
521: }
522: PetscCall(DSRestoreArray(eps->ds,DS_MAT_B,&temp));
523: } else {
524: PetscCall(BVSetActiveColumns(ctx->S,0,nv));
525: PetscCall(DSGetMat(eps->ds,DS_MAT_A,&pA));
526: PetscCall(MatZeroEntries(pA));
527: PetscCall(BVMatProject(ctx->S,A,ctx->S,pA));
528: PetscCall(DSRestoreMat(eps->ds,DS_MAT_A,&pA));
529: if (B) {
530: PetscCall(DSGetMat(eps->ds,DS_MAT_B,&pB));
531: PetscCall(MatZeroEntries(pB));
532: PetscCall(BVMatProject(ctx->S,B,ctx->S,pB));
533: PetscCall(DSRestoreMat(eps->ds,DS_MAT_B,&pB));
534: }
535: }
537: PetscCall(DSSolve(eps->ds,eps->eigr,eps->eigi));
538: PetscCall(DSSynchronize(eps->ds,eps->eigr,eps->eigi));
540: PetscCall(PetscMalloc3(nv,&fl1,nv,&inside,nv,&rr));
541: PetscCall(rescale_eig(eps,nv));
542: PetscCall(DSVectors(eps->ds,DS_MAT_X,NULL,NULL));
543: PetscCall(DSGetMat(eps->ds,DS_MAT_X,&X));
544: PetscCall(SlepcCISS_isGhost(X,nv,ctx->sigma,ctx->spurious_threshold,fl1));
545: PetscCall(DSRestoreMat(eps->ds,DS_MAT_X,&X));
546: PetscCall(RGCheckInside(eps->rg,nv,eps->eigr,eps->eigi,inside));
547: for (i=0;i<nv;i++) {
548: if (fl1[i] && inside[i]>=0) {
549: rr[i] = 1.0;
550: eps->nconv++;
551: } else rr[i] = 0.0;
552: }
553: PetscCall(DSSort(eps->ds,eps->eigr,eps->eigi,rr,NULL,&eps->nconv));
554: PetscCall(DSSynchronize(eps->ds,eps->eigr,eps->eigi));
555: PetscCall(rescale_eig(eps,nv));
556: PetscCall(PetscFree3(fl1,inside,rr));
557: PetscCall(BVSetActiveColumns(eps->V,0,nv));
558: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) {
559: PetscCall(BVSumQuadrature(ctx->S,ctx->Y,ctx->M,ctx->L,ctx->L,ctx->weight,ctx->pp,contour->scatterin,contour->subcomm,contour->npoints,ctx->useconj));
560: PetscCall(BVSetActiveColumns(ctx->S,0,ctx->L));
561: PetscCall(BVCopy(ctx->S,ctx->V));
562: PetscCall(BVSetActiveColumns(ctx->S,0,nv));
563: }
564: PetscCall(BVCopy(ctx->S,eps->V));
566: PetscCall(DSVectors(eps->ds,DS_MAT_X,NULL,NULL));
567: PetscCall(DSGetMat(eps->ds,DS_MAT_X,&X));
568: PetscCall(BVMultInPlace(ctx->S,X,0,eps->nconv));
569: if (eps->ishermitian) PetscCall(BVMultInPlace(eps->V,X,0,eps->nconv));
570: PetscCall(DSRestoreMat(eps->ds,DS_MAT_X,&X));
571: max_error = 0.0;
572: for (i=0;i<eps->nconv;i++) {
573: PetscCall(BVGetColumn(ctx->S,i,&si));
574: #if !defined(PETSC_USE_COMPLEX)
575: if (eps->eigi[i]!=0.0) PetscCall(BVGetColumn(ctx->S,i+1,&si1));
576: #endif
577: PetscCall(EPSComputeResidualNorm_Private(eps,PETSC_FALSE,eps->eigr[i],eps->eigi[i],si,si1,w,&error));
578: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) { /* vector is not normalized */
579: PetscCall(VecNorm(si,NORM_2,&norm));
580: #if !defined(PETSC_USE_COMPLEX)
581: if (eps->eigi[i]!=0.0) {
582: PetscCall(VecNorm(si1,NORM_2,&normi));
583: norm = SlepcAbsEigenvalue(norm,normi);
584: }
585: #endif
586: error /= norm;
587: }
588: PetscCall((*eps->converged)(eps,eps->eigr[i],eps->eigi[i],error,&error,eps->convergedctx));
589: PetscCall(BVRestoreColumn(ctx->S,i,&si));
590: #if !defined(PETSC_USE_COMPLEX)
591: if (eps->eigi[i]!=0.0) {
592: PetscCall(BVRestoreColumn(ctx->S,i+1,&si1));
593: i++;
594: }
595: #endif
596: max_error = PetscMax(max_error,error);
597: }
599: if (max_error <= eps->tol) eps->reason = EPS_CONVERGED_TOL;
600: else if (eps->its >= eps->max_it) eps->reason = EPS_DIVERGED_ITS;
601: else {
602: if (eps->nconv > ctx->L) nv = eps->nconv;
603: else if (ctx->L > nv) nv = ctx->L;
604: nv = PetscMin(nv,ctx->L*ctx->M);
605: PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,nv,ctx->L,NULL,&M));
606: PetscCall(MatSetRandom(M,rand));
607: PetscCall(BVSetActiveColumns(ctx->S,0,nv));
608: PetscCall(BVMultInPlace(ctx->S,M,0,ctx->L));
609: PetscCall(MatDestroy(&M));
610: PetscCall(BVSetActiveColumns(ctx->S,0,ctx->L));
611: PetscCall(BVSetActiveColumns(ctx->V,0,ctx->L));
612: PetscCall(BVCopy(ctx->S,ctx->V));
613: if (contour->pA) PetscCall(BVScatter(ctx->V,ctx->pV,contour->scatterin,contour->xdup));
614: PetscCall(EPSCISSSolve(eps,J,V,0,ctx->L));
615: }
616: }
617: }
618: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) PetscCall(PetscFree(H1));
619: PetscCall(PetscFree2(Mu,H0));
620: PetscFunctionReturn(PETSC_SUCCESS);
621: }
623: static PetscErrorCode EPSComputeVectors_CISS(EPS eps)
624: {
625: EPS_CISS *ctx = (EPS_CISS*)eps->data;
626: PetscInt n;
627: Mat Z,B=NULL;
629: PetscFunctionBegin;
630: if (eps->ishermitian) {
631: if (eps->isgeneralized && !eps->ispositive) PetscCall(EPSComputeVectors_Indefinite(eps));
632: else PetscCall(EPSComputeVectors_Hermitian(eps));
633: if (eps->isgeneralized && eps->ispositive && ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) {
634: /* normalize to have unit B-norm */
635: PetscCall(STGetMatrix(eps->st,1,&B));
636: PetscCall(BVSetMatrix(eps->V,B,PETSC_FALSE));
637: PetscCall(BVNormalize(eps->V,NULL));
638: PetscCall(BVSetMatrix(eps->V,NULL,PETSC_FALSE));
639: }
640: PetscFunctionReturn(PETSC_SUCCESS);
641: }
642: PetscCall(DSGetDimensions(eps->ds,&n,NULL,NULL,NULL));
643: PetscCall(BVSetActiveColumns(eps->V,0,n));
645: /* right eigenvectors */
646: PetscCall(DSVectors(eps->ds,DS_MAT_X,NULL,NULL));
648: /* V = V * Z */
649: PetscCall(DSGetMat(eps->ds,DS_MAT_X,&Z));
650: PetscCall(BVMultInPlace(eps->V,Z,0,n));
651: PetscCall(DSRestoreMat(eps->ds,DS_MAT_X,&Z));
652: PetscCall(BVSetActiveColumns(eps->V,0,eps->nconv));
654: /* normalize */
655: if (ctx->extraction == EPS_CISS_EXTRACTION_HANKEL) PetscCall(BVNormalize(eps->V,NULL));
656: PetscFunctionReturn(PETSC_SUCCESS);
657: }
659: static PetscErrorCode EPSCISSSetSizes_CISS(EPS eps,PetscInt ip,PetscInt bs,PetscInt ms,PetscInt npart,PetscInt bsmax,PetscBool realmats)
660: {
661: EPS_CISS *ctx = (EPS_CISS*)eps->data;
662: PetscInt oN,oL,oM,oLmax,onpart;
663: PetscMPIInt size;
665: PetscFunctionBegin;
666: oN = ctx->N;
667: if (ip == PETSC_DETERMINE) {
668: if (ctx->N!=32) { ctx->N =32; ctx->M = ctx->N/4; }
669: } else if (ip != PETSC_CURRENT) {
670: PetscCheck(ip>0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The ip argument must be > 0");
671: PetscCheck(ip%2==0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The ip argument must be an even number");
672: if (ctx->N!=ip) { ctx->N = ip; ctx->M = ctx->N/4; }
673: }
674: oL = ctx->L;
675: if (bs == PETSC_DETERMINE) {
676: ctx->L = 16;
677: } else if (bs != PETSC_CURRENT) {
678: PetscCheck(bs>0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The bs argument must be > 0");
679: ctx->L = bs;
680: }
681: oM = ctx->M;
682: if (ms == PETSC_DETERMINE) {
683: ctx->M = ctx->N/4;
684: } else if (ms != PETSC_CURRENT) {
685: PetscCheck(ms>0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The ms argument must be > 0");
686: PetscCheck(ms<=ctx->N,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The ms argument must be less than or equal to the number of integration points");
687: ctx->M = ms;
688: }
689: onpart = ctx->npart;
690: if (npart == PETSC_DETERMINE) {
691: ctx->npart = 1;
692: } else if (npart != PETSC_CURRENT) {
693: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)eps),&size));
694: PetscCheck(npart>0 && npart<=size,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"Illegal value of npart");
695: ctx->npart = npart;
696: }
697: oLmax = ctx->L_max;
698: if (bsmax == PETSC_DETERMINE) {
699: ctx->L_max = 64;
700: } else if (bsmax != PETSC_CURRENT) {
701: PetscCheck(bsmax>0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The bsmax argument must be > 0");
702: ctx->L_max = PetscMax(bsmax,ctx->L);
703: }
704: if (onpart != ctx->npart || oN != ctx->N || realmats != ctx->isreal) {
705: PetscCall(SlepcContourDataDestroy(&ctx->contour));
706: PetscCall(PetscInfo(eps,"Resetting the contour data structure due to a change of parameters\n"));
707: eps->state = EPS_STATE_INITIAL;
708: }
709: ctx->isreal = realmats;
710: if (oL != ctx->L || oM != ctx->M || oLmax != ctx->L_max) eps->state = EPS_STATE_INITIAL;
711: PetscFunctionReturn(PETSC_SUCCESS);
712: }
714: /*@
715: EPSCISSSetSizes - Sets the values of various size parameters in the CISS solver.
717: Logically Collective
719: Input Parameters:
720: + eps - the linear eigensolver context
721: . ip - number of integration points
722: . bs - block size
723: . ms - moment size
724: . npart - number of partitions when splitting the communicator
725: . bsmax - max block size
726: - realmats - $A$ and $B$ are real
728: Options Database Keys:
729: + -eps_ciss_integration_points \<ip\> - sets the number of integration points
730: . -eps_ciss_blocksize \<bs\> - sets the block size
731: . -eps_ciss_moments \<ms\> - sets the moment size
732: . -eps_ciss_partitions \<npart\> - sets the number of partitions
733: . -eps_ciss_maxblocksize \<bsmax\> - sets the maximum block size
734: - -eps_ciss_realmats - $A$ and $B$ are real
736: Notes:
737: For all integer arguments, you can use `PETSC_CURRENT` to keep the current value, and
738: `PETSC_DETERMINE` to set them to a default value.
740: The default number of partitions is 1. This means the internal `KSP` object is shared
741: among all processes of the `EPS` communicator. Otherwise, the communicator is split
742: into `npart` communicators, so that `npart` `KSP` solves proceed simultaneously.
744: For a detailed description of the parameters see {cite:p}`Mae16`.
746: Level: advanced
748: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSGetSizes()`, `EPSCISSSetThreshold()`, `EPSCISSSetRefinement()`
749: @*/
750: PetscErrorCode EPSCISSSetSizes(EPS eps,PetscInt ip,PetscInt bs,PetscInt ms,PetscInt npart,PetscInt bsmax,PetscBool realmats)
751: {
752: PetscFunctionBegin;
760: PetscTryMethod(eps,"EPSCISSSetSizes_C",(EPS,PetscInt,PetscInt,PetscInt,PetscInt,PetscInt,PetscBool),(eps,ip,bs,ms,npart,bsmax,realmats));
761: PetscFunctionReturn(PETSC_SUCCESS);
762: }
764: static PetscErrorCode EPSCISSGetSizes_CISS(EPS eps,PetscInt *ip,PetscInt *bs,PetscInt *ms,PetscInt *npart,PetscInt *bsmax,PetscBool *realmats)
765: {
766: EPS_CISS *ctx = (EPS_CISS*)eps->data;
768: PetscFunctionBegin;
769: if (ip) *ip = ctx->N;
770: if (bs) *bs = ctx->L;
771: if (ms) *ms = ctx->M;
772: if (npart) *npart = ctx->npart;
773: if (bsmax) *bsmax = ctx->L_max;
774: if (realmats) *realmats = ctx->isreal;
775: PetscFunctionReturn(PETSC_SUCCESS);
776: }
778: /*@
779: EPSCISSGetSizes - Gets the values of various size parameters in the CISS solver.
781: Not Collective
783: Input Parameter:
784: . eps - the linear eigensolver context
786: Output Parameters:
787: + ip - number of integration points
788: . bs - block size
789: . ms - moment size
790: . npart - number of partitions when splitting the communicator
791: . bsmax - max block size
792: - realmats - $A$ and $B$ are real
794: Level: advanced
796: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSSetSizes()`
797: @*/
798: PetscErrorCode EPSCISSGetSizes(EPS eps,PetscInt *ip,PetscInt *bs,PetscInt *ms,PetscInt *npart,PetscInt *bsmax,PetscBool *realmats)
799: {
800: PetscFunctionBegin;
802: PetscUseMethod(eps,"EPSCISSGetSizes_C",(EPS,PetscInt*,PetscInt*,PetscInt*,PetscInt*,PetscInt*,PetscBool*),(eps,ip,bs,ms,npart,bsmax,realmats));
803: PetscFunctionReturn(PETSC_SUCCESS);
804: }
806: static PetscErrorCode EPSCISSSetThreshold_CISS(EPS eps,PetscReal delta,PetscReal spur)
807: {
808: EPS_CISS *ctx = (EPS_CISS*)eps->data;
810: PetscFunctionBegin;
811: if (delta == (PetscReal)PETSC_DETERMINE) {
812: ctx->delta = SLEPC_DEFAULT_TOL*1e-4;
813: } else if (delta != (PetscReal)PETSC_CURRENT) {
814: PetscCheck(delta>0.0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The delta argument must be > 0.0");
815: ctx->delta = delta;
816: }
817: if (spur == (PetscReal)PETSC_DETERMINE) {
818: ctx->spurious_threshold = PetscSqrtReal(SLEPC_DEFAULT_TOL);
819: } else if (spur != (PetscReal)PETSC_CURRENT) {
820: PetscCheck(spur>0.0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The spurious threshold argument must be > 0.0");
821: ctx->spurious_threshold = spur;
822: }
823: PetscFunctionReturn(PETSC_SUCCESS);
824: }
826: /*@
827: EPSCISSSetThreshold - Sets the values of various threshold parameters in
828: the CISS solver.
830: Logically Collective
832: Input Parameters:
833: + eps - the linear eigensolver context
834: . delta - threshold for numerical rank
835: - spur - spurious threshold (to discard spurious eigenpairs)
837: Options Database Keys:
838: + -eps_ciss_delta \<delta\> - sets the delta
839: - -eps_ciss_spurious_threshold \<spur\> - sets the spurious threshold
841: Notes:
842: `PETSC_CURRENT` can be used to preserve the current value of any of the
843: arguments, and `PETSC_DETERMINE` to set them to a default value.
845: For a detailed description of the parameters see {cite:p}`Mae16`.
847: Level: advanced
849: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSGetThreshold()`, `EPSCISSSetSizes()`, `EPSCISSSetRefinement()`
850: @*/
851: PetscErrorCode EPSCISSSetThreshold(EPS eps,PetscReal delta,PetscReal spur)
852: {
853: PetscFunctionBegin;
857: PetscTryMethod(eps,"EPSCISSSetThreshold_C",(EPS,PetscReal,PetscReal),(eps,delta,spur));
858: PetscFunctionReturn(PETSC_SUCCESS);
859: }
861: static PetscErrorCode EPSCISSGetThreshold_CISS(EPS eps,PetscReal *delta,PetscReal *spur)
862: {
863: EPS_CISS *ctx = (EPS_CISS*)eps->data;
865: PetscFunctionBegin;
866: if (delta) *delta = ctx->delta;
867: if (spur) *spur = ctx->spurious_threshold;
868: PetscFunctionReturn(PETSC_SUCCESS);
869: }
871: /*@
872: EPSCISSGetThreshold - Gets the values of various threshold parameters
873: in the CISS solver.
875: Not Collective
877: Input Parameter:
878: . eps - the linear eigensolver context
880: Output Parameters:
881: + delta - threshold for numerical rank
882: - spur - spurious threshold (to discard spurious eigenpairs)
884: Level: advanced
886: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSSetThreshold()`
887: @*/
888: PetscErrorCode EPSCISSGetThreshold(EPS eps,PetscReal *delta,PetscReal *spur)
889: {
890: PetscFunctionBegin;
892: PetscUseMethod(eps,"EPSCISSGetThreshold_C",(EPS,PetscReal*,PetscReal*),(eps,delta,spur));
893: PetscFunctionReturn(PETSC_SUCCESS);
894: }
896: static PetscErrorCode EPSCISSSetRefinement_CISS(EPS eps,PetscInt inner,PetscInt blsize)
897: {
898: EPS_CISS *ctx = (EPS_CISS*)eps->data;
900: PetscFunctionBegin;
901: if (inner == PETSC_DETERMINE) {
902: ctx->refine_inner = 0;
903: } else if (inner != PETSC_CURRENT) {
904: PetscCheck(inner>=0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The refine inner argument must be >= 0");
905: ctx->refine_inner = inner;
906: }
907: if (blsize == PETSC_DETERMINE) {
908: ctx->refine_blocksize = 0;
909: } else if (blsize != PETSC_CURRENT) {
910: PetscCheck(blsize>=0,PetscObjectComm((PetscObject)eps),PETSC_ERR_ARG_OUTOFRANGE,"The refine blocksize argument must be >= 0");
911: ctx->refine_blocksize = blsize;
912: }
913: PetscFunctionReturn(PETSC_SUCCESS);
914: }
916: /*@
917: EPSCISSSetRefinement - Sets the values of various refinement parameters
918: in the CISS solver.
920: Logically Collective
922: Input Parameters:
923: + eps - the linear eigensolver context
924: . inner - number of iterative refinement iterations (inner loop)
925: - blsize - number of iterative refinement iterations (blocksize loop)
927: Options Database Keys:
928: + -eps_ciss_refine_inner \<inner\> - sets number of inner iterations
929: - -eps_ciss_refine_blocksize \<blsize\> - sets number of blocksize iterations
931: Notes:
932: `PETSC_CURRENT` can be used to preserve the current value of any of the
933: arguments, and `PETSC_DETERMINE` to set them to a default of 0 (no refinement).
935: For a detailed description of the parameters see {cite:p}`Mae16`.
937: Level: advanced
939: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSGetRefinement()`, `EPSCISSSetSizes()`, `EPSCISSSetThreshold()`
940: @*/
941: PetscErrorCode EPSCISSSetRefinement(EPS eps,PetscInt inner,PetscInt blsize)
942: {
943: PetscFunctionBegin;
947: PetscTryMethod(eps,"EPSCISSSetRefinement_C",(EPS,PetscInt,PetscInt),(eps,inner,blsize));
948: PetscFunctionReturn(PETSC_SUCCESS);
949: }
951: static PetscErrorCode EPSCISSGetRefinement_CISS(EPS eps,PetscInt *inner,PetscInt *blsize)
952: {
953: EPS_CISS *ctx = (EPS_CISS*)eps->data;
955: PetscFunctionBegin;
956: if (inner) *inner = ctx->refine_inner;
957: if (blsize) *blsize = ctx->refine_blocksize;
958: PetscFunctionReturn(PETSC_SUCCESS);
959: }
961: /*@
962: EPSCISSGetRefinement - Gets the values of various refinement parameters
963: in the CISS solver.
965: Not Collective
967: Input Parameter:
968: . eps - the linear eigensolver context
970: Output Parameters:
971: + inner - number of iterative refinement iterations (inner loop)
972: - blsize - number of iterative refinement iterations (blocksize loop)
974: Level: advanced
976: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSSetRefinement()`
977: @*/
978: PetscErrorCode EPSCISSGetRefinement(EPS eps, PetscInt *inner, PetscInt *blsize)
979: {
980: PetscFunctionBegin;
982: PetscUseMethod(eps,"EPSCISSGetRefinement_C",(EPS,PetscInt*,PetscInt*),(eps,inner,blsize));
983: PetscFunctionReturn(PETSC_SUCCESS);
984: }
986: static PetscErrorCode EPSCISSSetUseST_CISS(EPS eps,PetscBool usest)
987: {
988: EPS_CISS *ctx = (EPS_CISS*)eps->data;
990: PetscFunctionBegin;
991: ctx->usest = usest;
992: ctx->usest_set = PETSC_TRUE;
993: eps->state = EPS_STATE_INITIAL;
994: PetscFunctionReturn(PETSC_SUCCESS);
995: }
997: /*@
998: EPSCISSSetUseST - Sets a flag indicating that the CISS solver will
999: use the `ST` object for the linear solves.
1001: Logically Collective
1003: Input Parameters:
1004: + eps - the linear eigensolver context
1005: - usest - boolean flag to use the `ST` object or not
1007: Options Database Key:
1008: . -eps_ciss_usest - whether the `ST` object will be used or not
1010: Note:
1011: When this option is set, the linear solves can be configured by
1012: setting options for the `KSP` object obtained with `STGetKSP()`.
1013: Otherwise, several `KSP` objects are created, which can be accessed
1014: with `EPSCISSGetKSPs()`.
1016: The default is to use the `ST`, unless several partitions have been
1017: specified, see `EPSCISSSetSizes()`.
1019: Level: advanced
1021: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSGetUseST()`, `EPSCISSSetSizes()`, `EPSCISSGetKSPs()`, `STGetKSP()`
1022: @*/
1023: PetscErrorCode EPSCISSSetUseST(EPS eps,PetscBool usest)
1024: {
1025: PetscFunctionBegin;
1028: PetscTryMethod(eps,"EPSCISSSetUseST_C",(EPS,PetscBool),(eps,usest));
1029: PetscFunctionReturn(PETSC_SUCCESS);
1030: }
1032: static PetscErrorCode EPSCISSGetUseST_CISS(EPS eps,PetscBool *usest)
1033: {
1034: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1036: PetscFunctionBegin;
1037: *usest = ctx->usest;
1038: PetscFunctionReturn(PETSC_SUCCESS);
1039: }
1041: /*@
1042: EPSCISSGetUseST - Gets the flag for using the `ST` object
1043: in the CISS solver.
1045: Not Collective
1047: Input Parameter:
1048: . eps - the linear eigensolver context
1050: Output Parameter:
1051: . usest - boolean flag indicating if the `ST` object is being used
1053: Level: advanced
1055: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSSetUseST()`
1056: @*/
1057: PetscErrorCode EPSCISSGetUseST(EPS eps,PetscBool *usest)
1058: {
1059: PetscFunctionBegin;
1061: PetscAssertPointer(usest,2);
1062: PetscUseMethod(eps,"EPSCISSGetUseST_C",(EPS,PetscBool*),(eps,usest));
1063: PetscFunctionReturn(PETSC_SUCCESS);
1064: }
1066: static PetscErrorCode EPSCISSSetQuadRule_CISS(EPS eps,EPSCISSQuadRule quad)
1067: {
1068: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1070: PetscFunctionBegin;
1071: if (ctx->quad != quad) {
1072: ctx->quad = quad;
1073: eps->state = EPS_STATE_INITIAL;
1074: }
1075: PetscFunctionReturn(PETSC_SUCCESS);
1076: }
1078: /*@
1079: EPSCISSSetQuadRule - Sets the quadrature rule used in the CISS solver.
1081: Logically Collective
1083: Input Parameters:
1084: + eps - the linear eigensolver context
1085: - quad - the quadrature rule, see `EPSCISSQuadRule` for possible values
1087: Options Database Key:
1088: . -eps_ciss_quadrule \<quad\> - sets the quadrature rule, either `trapezoidal` or `chebyshev`
1090: Notes:
1091: By default, the trapezoidal rule is used (`EPS_CISS_QUADRULE_TRAPEZOIDAL`).
1093: If the `chebyshev` option is specified (`EPS_CISS_QUADRULE_CHEBYSHEV`), then
1094: Chebyshev points are used as quadrature points.
1096: Level: advanced
1098: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSGetQuadRule()`, `EPSCISSQuadRule`
1099: @*/
1100: PetscErrorCode EPSCISSSetQuadRule(EPS eps,EPSCISSQuadRule quad)
1101: {
1102: PetscFunctionBegin;
1105: PetscTryMethod(eps,"EPSCISSSetQuadRule_C",(EPS,EPSCISSQuadRule),(eps,quad));
1106: PetscFunctionReturn(PETSC_SUCCESS);
1107: }
1109: static PetscErrorCode EPSCISSGetQuadRule_CISS(EPS eps,EPSCISSQuadRule *quad)
1110: {
1111: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1113: PetscFunctionBegin;
1114: *quad = ctx->quad;
1115: PetscFunctionReturn(PETSC_SUCCESS);
1116: }
1118: /*@
1119: EPSCISSGetQuadRule - Gets the quadrature rule used in the CISS solver.
1121: Not Collective
1123: Input Parameter:
1124: . eps - the linear eigensolver context
1126: Output Parameter:
1127: . quad - quadrature rule
1129: Level: advanced
1131: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSSetQuadRule()`, `EPSCISSQuadRule`
1132: @*/
1133: PetscErrorCode EPSCISSGetQuadRule(EPS eps,EPSCISSQuadRule *quad)
1134: {
1135: PetscFunctionBegin;
1137: PetscAssertPointer(quad,2);
1138: PetscUseMethod(eps,"EPSCISSGetQuadRule_C",(EPS,EPSCISSQuadRule*),(eps,quad));
1139: PetscFunctionReturn(PETSC_SUCCESS);
1140: }
1142: static PetscErrorCode EPSCISSSetExtraction_CISS(EPS eps,EPSCISSExtraction extraction)
1143: {
1144: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1146: PetscFunctionBegin;
1147: if (ctx->extraction != extraction) {
1148: ctx->extraction = extraction;
1149: eps->state = EPS_STATE_INITIAL;
1150: }
1151: PetscFunctionReturn(PETSC_SUCCESS);
1152: }
1154: /*@
1155: EPSCISSSetExtraction - Sets the extraction technique used in the CISS solver.
1157: Logically Collective
1159: Input Parameters:
1160: + eps - the linear eigensolver context
1161: - extraction - the extraction technique, see `EPSCISSExtraction` for possible values
1163: Options Database Key:
1164: . -eps_ciss_extraction \<extraction\> - sets the extraction technique, either `ritz` or `hankel`
1166: Notes:
1167: By default, the Rayleigh-Ritz extraction is used (`EPS_CISS_EXTRACTION_RITZ`),
1168: see {cite:p}`Sak07`.
1170: If the `hankel` option is specified (`EPS_CISS_EXTRACTION_HANKEL`), then
1171: the block Hankel method is used for extracting eigenpairs {cite:p}`Sak03`.
1173: Level: advanced
1175: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSGetExtraction()`, `EPSCISSExtraction`
1176: @*/
1177: PetscErrorCode EPSCISSSetExtraction(EPS eps,EPSCISSExtraction extraction)
1178: {
1179: PetscFunctionBegin;
1182: PetscTryMethod(eps,"EPSCISSSetExtraction_C",(EPS,EPSCISSExtraction),(eps,extraction));
1183: PetscFunctionReturn(PETSC_SUCCESS);
1184: }
1186: static PetscErrorCode EPSCISSGetExtraction_CISS(EPS eps,EPSCISSExtraction *extraction)
1187: {
1188: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1190: PetscFunctionBegin;
1191: *extraction = ctx->extraction;
1192: PetscFunctionReturn(PETSC_SUCCESS);
1193: }
1195: /*@
1196: EPSCISSGetExtraction - Gets the extraction technique used in the CISS solver.
1198: Not Collective
1200: Input Parameter:
1201: . eps - the linear eigensolver context
1203: Output Parameter:
1204: . extraction - extraction technique
1206: Level: advanced
1208: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSSetExtraction()`, `EPSCISSExtraction`
1209: @*/
1210: PetscErrorCode EPSCISSGetExtraction(EPS eps,EPSCISSExtraction *extraction)
1211: {
1212: PetscFunctionBegin;
1214: PetscAssertPointer(extraction,2);
1215: PetscUseMethod(eps,"EPSCISSGetExtraction_C",(EPS,EPSCISSExtraction*),(eps,extraction));
1216: PetscFunctionReturn(PETSC_SUCCESS);
1217: }
1219: static PetscErrorCode EPSCISSGetKSPs_CISS(EPS eps,PetscInt *nsolve,KSP **ksp)
1220: {
1221: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1222: SlepcContourData contour;
1223: PetscInt i,nsplit;
1224: PC pc;
1225: MPI_Comm child;
1227: PetscFunctionBegin;
1228: if (!ctx->contour) { /* initialize contour data structure first */
1229: PetscCall(RGCanUseConjugates(eps->rg,ctx->isreal,&ctx->useconj));
1230: PetscCall(SlepcContourDataCreate(ctx->useconj?ctx->N/2:ctx->N,ctx->npart,(PetscObject)eps,&ctx->contour));
1231: }
1232: contour = ctx->contour;
1233: if (!contour->ksp) {
1234: PetscCall(PetscMalloc1(contour->npoints,&contour->ksp));
1235: PetscCall(EPSGetST(eps,&eps->st));
1236: PetscCall(STGetSplitPreconditionerInfo(eps->st,&nsplit,NULL));
1237: PetscCall(PetscSubcommGetChild(contour->subcomm,&child));
1238: for (i=0;i<contour->npoints;i++) {
1239: PetscCall(KSPCreate(child,&contour->ksp[i]));
1240: PetscCall(PetscObjectIncrementTabLevel((PetscObject)contour->ksp[i],(PetscObject)eps,1));
1241: PetscCall(KSPSetOptionsPrefix(contour->ksp[i],((PetscObject)eps)->prefix));
1242: PetscCall(KSPAppendOptionsPrefix(contour->ksp[i],"eps_ciss_"));
1243: PetscCall(PetscObjectSetOptions((PetscObject)contour->ksp[i],((PetscObject)eps)->options));
1244: PetscCall(KSPSetErrorIfNotConverged(contour->ksp[i],PETSC_TRUE));
1245: PetscCall(KSPSetTolerances(contour->ksp[i],SlepcDefaultTol(eps->tol),PETSC_CURRENT,PETSC_CURRENT,PETSC_CURRENT));
1246: PetscCall(KSPGetPC(contour->ksp[i],&pc));
1247: if (nsplit) {
1248: PetscCall(KSPSetType(contour->ksp[i],KSPBCGS));
1249: PetscCall(PCSetType(pc,PCBJACOBI));
1250: } else {
1251: PetscCall(KSPSetType(contour->ksp[i],KSPPREONLY));
1252: PetscCall(PCSetType(pc,PCLU));
1253: }
1254: }
1255: }
1256: if (nsolve) *nsolve = contour->npoints;
1257: if (ksp) *ksp = contour->ksp;
1258: PetscFunctionReturn(PETSC_SUCCESS);
1259: }
1261: /*@C
1262: EPSCISSGetKSPs - Retrieve the array of linear solver objects associated with
1263: the CISS solver.
1265: Not Collective
1267: Input Parameter:
1268: . eps - the linear eigensolver context
1270: Output Parameters:
1271: + nsolve - number of solver objects
1272: - ksp - array of linear solver object
1274: Note:
1275: The number of `KSP` solvers is equal to the number of integration points divided by
1276: the number of partitions, see `EPSCISSSetSizes()`. This value is halved in the case
1277: of real matrices with a region centered at the real axis.
1279: Level: advanced
1281: .seealso: [](ch:eps), `EPSCISS`, `EPSCISSSetSizes()`
1282: @*/
1283: PetscErrorCode EPSCISSGetKSPs(EPS eps,PetscInt *nsolve,KSP **ksp)
1284: {
1285: PetscFunctionBegin;
1287: PetscUseMethod(eps,"EPSCISSGetKSPs_C",(EPS,PetscInt*,KSP**),(eps,nsolve,ksp));
1288: PetscFunctionReturn(PETSC_SUCCESS);
1289: }
1291: static PetscErrorCode EPSReset_CISS(EPS eps)
1292: {
1293: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1295: PetscFunctionBegin;
1296: PetscCall(BVDestroy(&ctx->S));
1297: PetscCall(BVDestroy(&ctx->V));
1298: PetscCall(BVDestroy(&ctx->Y));
1299: if (!ctx->usest) PetscCall(SlepcContourDataReset(ctx->contour));
1300: PetscCall(BVDestroy(&ctx->pV));
1301: PetscFunctionReturn(PETSC_SUCCESS);
1302: }
1304: static PetscErrorCode EPSSetFromOptions_CISS(EPS eps,PetscOptionItems PetscOptionsObject)
1305: {
1306: PetscReal r3,r4;
1307: PetscInt i,i1,i2,i3,i4,i5,i6,i7;
1308: PetscBool b1,b2,flg,flg2,flg3,flg4,flg5,flg6;
1309: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1310: EPSCISSQuadRule quad;
1311: EPSCISSExtraction extraction;
1313: PetscFunctionBegin;
1314: PetscOptionsHeadBegin(PetscOptionsObject,"EPS CISS Options");
1316: PetscCall(EPSCISSGetSizes(eps,&i1,&i2,&i3,&i4,&i5,&b1));
1317: PetscCall(PetscOptionsInt("-eps_ciss_integration_points","Number of integration points","EPSCISSSetSizes",i1,&i1,&flg));
1318: PetscCall(PetscOptionsInt("-eps_ciss_blocksize","Block size","EPSCISSSetSizes",i2,&i2,&flg2));
1319: PetscCall(PetscOptionsInt("-eps_ciss_moments","Moment size","EPSCISSSetSizes",i3,&i3,&flg3));
1320: PetscCall(PetscOptionsInt("-eps_ciss_partitions","Number of partitions","EPSCISSSetSizes",i4,&i4,&flg4));
1321: PetscCall(PetscOptionsInt("-eps_ciss_maxblocksize","Maximum block size","EPSCISSSetSizes",i5,&i5,&flg5));
1322: PetscCall(PetscOptionsBool("-eps_ciss_realmats","True if A and B are real","EPSCISSSetSizes",b1,&b1,&flg6));
1323: if (flg || flg2 || flg3 || flg4 || flg5 || flg6) PetscCall(EPSCISSSetSizes(eps,i1,i2,i3,i4,i5,b1));
1325: PetscCall(EPSCISSGetThreshold(eps,&r3,&r4));
1326: PetscCall(PetscOptionsReal("-eps_ciss_delta","Threshold for numerical rank","EPSCISSSetThreshold",r3,&r3,&flg));
1327: PetscCall(PetscOptionsReal("-eps_ciss_spurious_threshold","Threshold for the spurious eigenpairs","EPSCISSSetThreshold",r4,&r4,&flg2));
1328: if (flg || flg2) PetscCall(EPSCISSSetThreshold(eps,r3,r4));
1330: PetscCall(EPSCISSGetRefinement(eps,&i6,&i7));
1331: PetscCall(PetscOptionsInt("-eps_ciss_refine_inner","Number of inner iterative refinement iterations","EPSCISSSetRefinement",i6,&i6,&flg));
1332: PetscCall(PetscOptionsInt("-eps_ciss_refine_blocksize","Number of blocksize iterative refinement iterations","EPSCISSSetRefinement",i7,&i7,&flg2));
1333: if (flg || flg2) PetscCall(EPSCISSSetRefinement(eps,i6,i7));
1335: PetscCall(EPSCISSGetUseST(eps,&b2));
1336: PetscCall(PetscOptionsBool("-eps_ciss_usest","Use ST for linear solves","EPSCISSSetUseST",b2,&b2,&flg));
1337: if (flg) PetscCall(EPSCISSSetUseST(eps,b2));
1339: PetscCall(PetscOptionsEnum("-eps_ciss_quadrule","Quadrature rule","EPSCISSSetQuadRule",EPSCISSQuadRules,(PetscEnum)ctx->quad,(PetscEnum*)&quad,&flg));
1340: if (flg) PetscCall(EPSCISSSetQuadRule(eps,quad));
1342: PetscCall(PetscOptionsEnum("-eps_ciss_extraction","Extraction technique","EPSCISSSetExtraction",EPSCISSExtractions,(PetscEnum)ctx->extraction,(PetscEnum*)&extraction,&flg));
1343: if (flg) PetscCall(EPSCISSSetExtraction(eps,extraction));
1345: PetscOptionsHeadEnd();
1347: if (!eps->rg) PetscCall(EPSGetRG(eps,&eps->rg));
1348: PetscCall(RGSetFromOptions(eps->rg)); /* this is necessary here to set useconj */
1349: if (!ctx->contour || !ctx->contour->ksp) PetscCall(EPSCISSGetKSPs(eps,NULL,NULL));
1350: PetscAssert(ctx->contour && ctx->contour->ksp,PetscObjectComm((PetscObject)eps),PETSC_ERR_PLIB,"Something went wrong with EPSCISSGetKSPs()");
1351: for (i=0;i<ctx->contour->npoints;i++) PetscCall(KSPSetFromOptions(ctx->contour->ksp[i]));
1352: PetscCall(PetscSubcommSetFromOptions(ctx->contour->subcomm));
1353: PetscFunctionReturn(PETSC_SUCCESS);
1354: }
1356: static PetscErrorCode EPSDestroy_CISS(EPS eps)
1357: {
1358: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1360: PetscFunctionBegin;
1361: PetscCall(SlepcContourDataDestroy(&ctx->contour));
1362: PetscCall(PetscFree4(ctx->weight,ctx->omega,ctx->pp,ctx->sigma));
1363: PetscCall(PetscFree(eps->data));
1364: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetSizes_C",NULL));
1365: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetSizes_C",NULL));
1366: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetThreshold_C",NULL));
1367: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetThreshold_C",NULL));
1368: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetRefinement_C",NULL));
1369: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetRefinement_C",NULL));
1370: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetUseST_C",NULL));
1371: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetUseST_C",NULL));
1372: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetQuadRule_C",NULL));
1373: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetQuadRule_C",NULL));
1374: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetExtraction_C",NULL));
1375: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetExtraction_C",NULL));
1376: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetKSPs_C",NULL));
1377: PetscFunctionReturn(PETSC_SUCCESS);
1378: }
1380: static PetscErrorCode EPSView_CISS(EPS eps,PetscViewer viewer)
1381: {
1382: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1383: PetscBool isascii;
1384: PetscViewer sviewer;
1386: PetscFunctionBegin;
1387: PetscCall(PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&isascii));
1388: if (isascii) {
1389: 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));
1390: if (ctx->isreal) PetscCall(PetscViewerASCIIPrintf(viewer," exploiting symmetry of integration points\n"));
1391: PetscCall(PetscViewerASCIIPrintf(viewer," threshold { delta: %g, spurious threshold: %g }\n",(double)ctx->delta,(double)ctx->spurious_threshold));
1392: PetscCall(PetscViewerASCIIPrintf(viewer," iterative refinement { inner: %" PetscInt_FMT ", blocksize: %" PetscInt_FMT " }\n",ctx->refine_inner, ctx->refine_blocksize));
1393: PetscCall(PetscViewerASCIIPrintf(viewer," extraction: %s\n",EPSCISSExtractions[ctx->extraction]));
1394: PetscCall(PetscViewerASCIIPrintf(viewer," quadrature rule: %s\n",EPSCISSQuadRules[ctx->quad]));
1395: if (ctx->usest) PetscCall(PetscViewerASCIIPrintf(viewer," using ST for linear solves\n"));
1396: else {
1397: if (!ctx->contour || !ctx->contour->ksp) PetscCall(EPSCISSGetKSPs(eps,NULL,NULL));
1398: PetscAssert(ctx->contour && ctx->contour->ksp,PetscObjectComm((PetscObject)eps),PETSC_ERR_PLIB,"Something went wrong with EPSCISSGetKSPs()");
1399: PetscCall(PetscViewerASCIIPushTab(viewer));
1400: if (ctx->npart>1 && ctx->contour->subcomm) {
1401: PetscCall(PetscViewerGetSubViewer(viewer,ctx->contour->subcomm->child,&sviewer));
1402: if (!ctx->contour->subcomm->color) PetscCall(KSPView(ctx->contour->ksp[0],sviewer));
1403: PetscCall(PetscViewerFlush(sviewer));
1404: PetscCall(PetscViewerRestoreSubViewer(viewer,ctx->contour->subcomm->child,&sviewer));
1405: /* extra call needed because of the two calls to PetscViewerASCIIPushSynchronized() in PetscViewerGetSubViewer() */
1406: PetscCall(PetscViewerASCIIPopSynchronized(viewer));
1407: } else PetscCall(KSPView(ctx->contour->ksp[0],viewer));
1408: PetscCall(PetscViewerASCIIPopTab(viewer));
1409: }
1410: }
1411: PetscFunctionReturn(PETSC_SUCCESS);
1412: }
1414: static PetscErrorCode EPSSetDefaultST_CISS(EPS eps)
1415: {
1416: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1417: PetscBool usest = ctx->usest;
1418: KSP ksp;
1419: PC pc;
1421: PetscFunctionBegin;
1422: if (!((PetscObject)eps->st)->type_name) {
1423: if (!ctx->usest_set) usest = (ctx->npart>1)? PETSC_FALSE: PETSC_TRUE;
1424: if (usest) PetscCall(STSetType(eps->st,STSINVERT));
1425: else {
1426: /* we are not going to use ST, so avoid factorizing the matrix */
1427: PetscCall(STSetType(eps->st,STSHIFT));
1428: if (eps->isgeneralized) {
1429: PetscCall(STGetKSP(eps->st,&ksp));
1430: PetscCall(KSPGetPC(ksp,&pc));
1431: PetscCall(PCSetType(pc,PCNONE));
1432: }
1433: }
1434: }
1435: PetscFunctionReturn(PETSC_SUCCESS);
1436: }
1438: /*MC
1439: EPSCISS - EPSCISS = "ciss" - A contour integral eigensolver based on the
1440: Sakurai-Sugiura scheme.
1442: Notes:
1443: This solver is based on the numerical contour integration idea
1444: proposed initially by {cite:t}`Sak03` and improved later by adding
1445: a Rayleigh-Ritz projection step {cite:p}`Sak07`.
1447: Contour integral methods are able to compute all eigenvalues
1448: lying inside a region of the complex plane. Use `EPSGetRG()` to
1449: specify the region. However, the computational cost is usually high
1450: because multiple linear systems must be solved. For this, we can
1451: use the `KSP` object inside `ST`, or several independent `KSP`s,
1452: see `EPSCISSSetUseST()`.
1454: Details of the implementation in SLEPc can be found in {cite:p}`Mae16`.
1456: Level: beginner
1458: .seealso: [](ch:eps), `EPS`, `EPSType`, `EPSSetType()`, `EPSGetRG()`
1459: M*/
1460: SLEPC_EXTERN PetscErrorCode EPSCreate_CISS(EPS eps)
1461: {
1462: EPS_CISS *ctx = (EPS_CISS*)eps->data;
1464: PetscFunctionBegin;
1465: PetscCall(PetscNew(&ctx));
1466: eps->data = ctx;
1468: eps->useds = PETSC_TRUE;
1469: eps->categ = EPS_CATEGORY_CONTOUR;
1471: eps->ops->solve = EPSSolve_CISS;
1472: eps->ops->setup = EPSSetUp_CISS;
1473: eps->ops->setupsort = EPSSetUpSort_CISS;
1474: eps->ops->setfromoptions = EPSSetFromOptions_CISS;
1475: eps->ops->destroy = EPSDestroy_CISS;
1476: eps->ops->reset = EPSReset_CISS;
1477: eps->ops->view = EPSView_CISS;
1478: eps->ops->computevectors = EPSComputeVectors_CISS;
1479: eps->ops->setdefaultst = EPSSetDefaultST_CISS;
1481: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetSizes_C",EPSCISSSetSizes_CISS));
1482: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetSizes_C",EPSCISSGetSizes_CISS));
1483: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetThreshold_C",EPSCISSSetThreshold_CISS));
1484: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetThreshold_C",EPSCISSGetThreshold_CISS));
1485: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetRefinement_C",EPSCISSSetRefinement_CISS));
1486: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetRefinement_C",EPSCISSGetRefinement_CISS));
1487: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetUseST_C",EPSCISSSetUseST_CISS));
1488: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetUseST_C",EPSCISSGetUseST_CISS));
1489: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetQuadRule_C",EPSCISSSetQuadRule_CISS));
1490: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetQuadRule_C",EPSCISSGetQuadRule_CISS));
1491: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSSetExtraction_C",EPSCISSSetExtraction_CISS));
1492: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetExtraction_C",EPSCISSGetExtraction_CISS));
1493: PetscCall(PetscObjectComposeFunction((PetscObject)eps,"EPSCISSGetKSPs_C",EPSCISSGetKSPs_CISS));
1495: /* set default values of parameters */
1496: ctx->N = 32;
1497: ctx->L = 16;
1498: ctx->M = ctx->N/4;
1499: ctx->delta = SLEPC_DEFAULT_TOL*1e-4;
1500: ctx->L_max = 64;
1501: ctx->spurious_threshold = PetscSqrtReal(SLEPC_DEFAULT_TOL);
1502: ctx->usest = PETSC_TRUE;
1503: ctx->usest_set = PETSC_FALSE;
1504: ctx->isreal = PETSC_FALSE;
1505: ctx->refine_inner = 0;
1506: ctx->refine_blocksize = 0;
1507: ctx->npart = 1;
1508: ctx->quad = (EPSCISSQuadRule)0;
1509: ctx->extraction = EPS_CISS_EXTRACTION_RITZ;
1510: PetscFunctionReturn(PETSC_SUCCESS);
1511: }