Actual source code: iterativ.c
petsc-master 2020-01-17
1: /*
2: This file contains some simple default routines.
3: These routines should be SHORT, since they will be included in every
4: executable image that uses the iterative routines (note that, through
5: the registry system, we provide a way to load only the truly necessary
6: files)
7: */
8: #include <petsc/private/kspimpl.h>
9: #include <petscdmshell.h>
11: /*@
12: KSPGetResidualNorm - Gets the last (approximate preconditioned)
13: residual norm that has been computed.
15: Not Collective
17: Input Parameters:
18: . ksp - the iterative context
20: Output Parameters:
21: . rnorm - residual norm
23: Level: intermediate
25: .seealso: KSPBuildResidual()
26: @*/
27: PetscErrorCode KSPGetResidualNorm(KSP ksp,PetscReal *rnorm)
28: {
32: *rnorm = ksp->rnorm;
33: return(0);
34: }
36: /*@
37: KSPGetIterationNumber - Gets the current iteration number; if the
38: KSPSolve() is complete, returns the number of iterations
39: used.
41: Not Collective
43: Input Parameters:
44: . ksp - the iterative context
46: Output Parameters:
47: . its - number of iterations
49: Level: intermediate
51: Notes:
52: During the ith iteration this returns i-1
53: .seealso: KSPBuildResidual(), KSPGetResidualNorm(), KSPGetTotalIterations()
54: @*/
55: PetscErrorCode KSPGetIterationNumber(KSP ksp,PetscInt *its)
56: {
60: *its = ksp->its;
61: return(0);
62: }
64: /*@
65: KSPGetTotalIterations - Gets the total number of iterations this KSP object has performed since was created, counted over all linear solves
67: Not Collective
69: Input Parameters:
70: . ksp - the iterative context
72: Output Parameters:
73: . its - total number of iterations
75: Level: intermediate
77: Notes:
78: Use KSPGetIterationNumber() to get the count for the most recent solve only
79: If this is called within a linear solve (such as in a KSPMonitor routine) then it does not include iterations within that current solve
81: .seealso: KSPBuildResidual(), KSPGetResidualNorm(), KSPGetIterationNumber()
82: @*/
83: PetscErrorCode KSPGetTotalIterations(KSP ksp,PetscInt *its)
84: {
88: *its = ksp->totalits;
89: return(0);
90: }
92: /*@C
93: KSPMonitorSingularValue - Prints the two norm of the true residual and
94: estimation of the extreme singular values of the preconditioned problem
95: at each iteration.
97: Logically Collective on ksp
99: Input Parameters:
100: + ksp - the iterative context
101: . n - the iteration
102: - rnorm - the two norm of the residual
104: Options Database Key:
105: . -ksp_monitor_singular_value - Activates KSPMonitorSingularValue()
107: Notes:
108: The CG solver uses the Lanczos technique for eigenvalue computation,
109: while GMRES uses the Arnoldi technique; other iterative methods do
110: not currently compute singular values.
112: Level: intermediate
114: .seealso: KSPComputeExtremeSingularValues()
115: @*/
116: PetscErrorCode KSPMonitorSingularValue(KSP ksp,PetscInt n,PetscReal rnorm,PetscViewerAndFormat *dummy)
117: {
118: PetscReal emin,emax,c;
120: PetscViewer viewer = dummy->viewer;
125: PetscViewerPushFormat(viewer,dummy->format);
126: PetscViewerASCIIAddTab(viewer,((PetscObject)ksp)->tablevel);
127: if (!ksp->calc_sings) {
128: PetscViewerASCIIPrintf(viewer,"%3D KSP Residual norm %14.12e \n",n,(double)rnorm);
129: } else {
130: KSPComputeExtremeSingularValues(ksp,&emax,&emin);
131: c = emax/emin;
132: PetscViewerASCIIPrintf(viewer,"%3D KSP Residual norm %14.12e %% max %14.12e min %14.12e max/min %14.12e\n",n,(double)rnorm,(double)emax,(double)emin,(double)c);
133: }
134: PetscViewerASCIISubtractTab(viewer,((PetscObject)ksp)->tablevel);
135: PetscViewerPopFormat(viewer);
136: return(0);
137: }
139: /*@C
140: KSPMonitorSolution - Monitors progress of the KSP solvers by calling
141: VecView() for the approximate solution at each iteration.
143: Collective on ksp
145: Input Parameters:
146: + ksp - the KSP context
147: . its - iteration number
148: . fgnorm - 2-norm of residual (or gradient)
149: - dummy - a viewer
151: Level: intermediate
153: Notes:
154: For some Krylov methods such as GMRES constructing the solution at
155: each iteration is expensive, hence using this will slow the code.
157: .seealso: KSPMonitorSet(), KSPMonitorDefault(), VecView()
158: @*/
159: PetscErrorCode KSPMonitorSolution(KSP ksp,PetscInt its,PetscReal fgnorm,PetscViewerAndFormat *dummy)
160: {
162: Vec x;
163: PetscViewer viewer = dummy->viewer;
167: KSPBuildSolution(ksp,NULL,&x);
168: PetscViewerPushFormat(viewer,dummy->format);
169: VecView(x,viewer);
170: PetscViewerPopFormat(viewer);
171: return(0);
172: }
174: /*@C
175: KSPMonitorDefault - Print the residual norm at each iteration of an
176: iterative solver.
178: Collective on ksp
180: Input Parameters:
181: + ksp - iterative context
182: . n - iteration number
183: . rnorm - 2-norm (preconditioned) residual value (may be estimated).
184: - dummy - an ASCII PetscViewer
186: Level: intermediate
188: .seealso: KSPMonitorSet(), KSPMonitorTrueResidualNorm(), KSPMonitorLGResidualNormCreate()
189: @*/
190: PetscErrorCode KSPMonitorDefault(KSP ksp,PetscInt n,PetscReal rnorm,PetscViewerAndFormat *dummy)
191: {
193: PetscViewer viewer = dummy->viewer;
197: PetscViewerPushFormat(viewer,dummy->format);
198: PetscViewerASCIIAddTab(viewer,((PetscObject)ksp)->tablevel);
199: if (n == 0 && ((PetscObject)ksp)->prefix) {
200: PetscViewerASCIIPrintf(viewer," Residual norms for %s solve.\n",((PetscObject)ksp)->prefix);
201: }
202: PetscViewerASCIIPrintf(viewer,"%3D KSP Residual norm %14.12e \n",n,(double)rnorm);
203: PetscViewerASCIISubtractTab(viewer,((PetscObject)ksp)->tablevel);
204: PetscViewerPopFormat(viewer);
205: return(0);
206: }
208: /*@C
209: KSPMonitorTrueResidualNorm - Prints the true residual norm as well as the preconditioned
210: residual norm at each iteration of an iterative solver.
212: Collective on ksp
214: Input Parameters:
215: + ksp - iterative context
216: . n - iteration number
217: . rnorm - 2-norm (preconditioned) residual value (may be estimated).
218: - dummy - an ASCII PetscViewer
220: Options Database Key:
221: . -ksp_monitor_true_residual - Activates KSPMonitorTrueResidualNorm()
223: Notes:
224: When using right preconditioning, these values are equivalent.
226: Level: intermediate
228: .seealso: KSPMonitorSet(), KSPMonitorDefault(), KSPMonitorLGResidualNormCreate(),KSPMonitorTrueResidualMaxNorm()
229: @*/
230: PetscErrorCode KSPMonitorTrueResidualNorm(KSP ksp,PetscInt n,PetscReal rnorm,PetscViewerAndFormat *dummy)
231: {
233: Vec resid;
234: PetscReal truenorm,bnorm;
235: PetscViewer viewer = dummy->viewer;
236: char normtype[256];
240: PetscViewerPushFormat(viewer,dummy->format);
241: PetscViewerASCIIAddTab(viewer,((PetscObject)ksp)->tablevel);
242: if (n == 0 && ((PetscObject)ksp)->prefix) {
243: PetscViewerASCIIPrintf(viewer," Residual norms for %s solve.\n",((PetscObject)ksp)->prefix);
244: }
245: KSPBuildResidual(ksp,NULL,NULL,&resid);
246: VecNorm(resid,NORM_2,&truenorm);
247: VecDestroy(&resid);
248: VecNorm(ksp->vec_rhs,NORM_2,&bnorm);
249: PetscStrncpy(normtype,KSPNormTypes[ksp->normtype],sizeof(normtype));
250: PetscStrtolower(normtype);
251: PetscViewerASCIIPrintf(viewer,"%3D KSP %s resid norm %14.12e true resid norm %14.12e ||r(i)||/||b|| %14.12e\n",n,normtype,(double)rnorm,(double)truenorm,(double)(truenorm/bnorm));
252: PetscViewerASCIISubtractTab(viewer,((PetscObject)ksp)->tablevel);
253: PetscViewerPopFormat(viewer);
254: return(0);
255: }
257: /*@C
258: KSPMonitorTrueResidualMaxNorm - Prints the true residual max norm each iteration of an iterative solver.
260: Collective on ksp
262: Input Parameters:
263: + ksp - iterative context
264: . n - iteration number
265: . rnorm - norm (preconditioned) residual value (may be estimated).
266: - dummy - an ASCII viewer
268: Options Database Key:
269: . -ksp_monitor_max - Activates KSPMonitorTrueResidualMaxNorm()
271: Notes:
272: This could be implemented (better) with a flag in ksp.
274: Level: intermediate
276: .seealso: KSPMonitorSet(), KSPMonitorDefault(), KSPMonitorLGResidualNormCreate(),KSPMonitorTrueResidualNorm()
277: @*/
278: PetscErrorCode KSPMonitorTrueResidualMaxNorm(KSP ksp,PetscInt n,PetscReal rnorm,PetscViewerAndFormat *dummy)
279: {
281: Vec resid;
282: PetscReal truenorm,bnorm;
283: PetscViewer viewer = dummy->viewer;
284: char normtype[256];
288: PetscViewerPushFormat(viewer,dummy->format);
289: PetscViewerASCIIAddTab(viewer,((PetscObject)ksp)->tablevel);
290: if (n == 0 && ((PetscObject)ksp)->prefix) {
291: PetscViewerASCIIPrintf(viewer," Residual norms (max) for %s solve.\n",((PetscObject)ksp)->prefix);
292: }
293: KSPBuildResidual(ksp,NULL,NULL,&resid);
294: VecNorm(resid,NORM_INFINITY,&truenorm);
295: VecDestroy(&resid);
296: VecNorm(ksp->vec_rhs,NORM_INFINITY,&bnorm);
297: PetscStrncpy(normtype,KSPNormTypes[ksp->normtype],sizeof(normtype));
298: PetscStrtolower(normtype);
299: PetscViewerASCIIPrintf(viewer,"%3D KSP true resid max norm %14.12e ||r(i)||/||b|| %14.12e\n",n,(double)truenorm,(double)(truenorm/bnorm));
300: PetscViewerASCIISubtractTab(viewer,((PetscObject)ksp)->tablevel);
301: PetscViewerPopFormat(viewer);
302: return(0);
303: }
305: PetscErrorCode KSPMonitorRange_Private(KSP ksp,PetscInt it,PetscReal *per)
306: {
308: Vec resid;
309: PetscReal rmax,pwork;
310: PetscInt i,n,N;
311: const PetscScalar *r;
314: KSPBuildResidual(ksp,NULL,NULL,&resid);
315: VecNorm(resid,NORM_INFINITY,&rmax);
316: VecGetLocalSize(resid,&n);
317: VecGetSize(resid,&N);
318: VecGetArrayRead(resid,&r);
319: pwork = 0.0;
320: for (i=0; i<n; i++) pwork += (PetscAbsScalar(r[i]) > .20*rmax);
321: VecRestoreArrayRead(resid,&r);
322: VecDestroy(&resid);
323: MPIU_Allreduce(&pwork,per,1,MPIU_REAL,MPIU_SUM,PetscObjectComm((PetscObject)ksp));
324: *per = *per/N;
325: return(0);
326: }
328: /*@C
329: KSPMonitorRange - Prints the percentage of residual elements that are more then 10 percent of the maximum value.
331: Collective on ksp
333: Input Parameters:
334: + ksp - iterative context
335: . it - iteration number
336: . rnorm - 2-norm (preconditioned) residual value (may be estimated).
337: - dummy - an ASCII viewer
339: Options Database Key:
340: . -ksp_monitor_range - Activates KSPMonitorRange()
342: Level: intermediate
344: .seealso: KSPMonitorSet(), KSPMonitorDefault(), KSPMonitorLGResidualNormCreate()
345: @*/
346: PetscErrorCode KSPMonitorRange(KSP ksp,PetscInt it,PetscReal rnorm,PetscViewerAndFormat *dummy)
347: {
348: PetscErrorCode ierr;
349: PetscReal perc,rel;
350: PetscViewer viewer = dummy->viewer;
351: /* should be in a MonitorRangeContext */
352: static PetscReal prev;
356: PetscViewerPushFormat(viewer,dummy->format);
357: PetscViewerASCIIAddTab(viewer,((PetscObject)ksp)->tablevel);
358: if (!it) prev = rnorm;
359: if (it == 0 && ((PetscObject)ksp)->prefix) {
360: PetscViewerASCIIPrintf(viewer," Residual norms for %s solve.\n",((PetscObject)ksp)->prefix);
361: }
362: KSPMonitorRange_Private(ksp,it,&perc);
364: rel = (prev - rnorm)/prev;
365: prev = rnorm;
366: PetscViewerASCIIPrintf(viewer,"%3D KSP preconditioned resid norm %14.12e Percent values above 20 percent of maximum %5.2f relative decrease %5.2e ratio %5.2e \n",it,(double)rnorm,(double)(100.0*perc),(double)rel,(double)(rel/perc));
367: PetscViewerASCIISubtractTab(viewer,((PetscObject)ksp)->tablevel);
368: PetscViewerPopFormat(viewer);
369: return(0);
370: }
372: /*@C
373: KSPMonitorDynamicTolerance - Recompute the inner tolerance in every
374: outer iteration in an adaptive way.
376: Collective on ksp
378: Input Parameters:
379: + ksp - iterative context
380: . n - iteration number (not used)
381: . fnorm - the current residual norm
382: - dummy - some context as a C struct. fields:
383: coef: a scaling coefficient. default 1.0. can be passed through
384: -sub_ksp_dynamic_tolerance_param
385: bnrm: norm of the right-hand side. store it to avoid repeated calculation
387: Notes:
388: This may be useful for a flexibly preconditioner Krylov method to
389: control the accuracy of the inner solves needed to gaurantee the
390: convergence of the outer iterations.
392: Level: advanced
394: .seealso: KSPMonitorDynamicToleranceDestroy()
395: @*/
396: PetscErrorCode KSPMonitorDynamicTolerance(KSP ksp,PetscInt its,PetscReal fnorm,void *dummy)
397: {
399: PC pc;
400: PetscReal outer_rtol, outer_abstol, outer_dtol, inner_rtol;
401: PetscInt outer_maxits,nksp,first,i;
402: KSPDynTolCtx *scale = (KSPDynTolCtx*)dummy;
403: KSP *subksp = NULL;
404: KSP kspinner;
405: PetscBool flg;
408: KSPGetPC(ksp, &pc);
410: /* compute inner_rtol */
411: if (scale->bnrm < 0.0) {
412: Vec b;
413: KSPGetRhs(ksp, &b);
414: VecNorm(b, NORM_2, &(scale->bnrm));
415: }
416: KSPGetTolerances(ksp, &outer_rtol, &outer_abstol, &outer_dtol, &outer_maxits);
417: inner_rtol = PetscMin(scale->coef * scale->bnrm * outer_rtol / fnorm, 0.999);
418: /*PetscPrintf(PETSC_COMM_WORLD, " Inner rtol = %g\n", (double)inner_rtol);*/
420: /* if pc is ksp */
421: PetscObjectTypeCompare((PetscObject)pc,PCKSP,&flg);
422: if (flg) {
423: PCKSPGetKSP(pc, &kspinner);
424: KSPSetTolerances(kspinner, inner_rtol, outer_abstol, outer_dtol, outer_maxits);
425: return(0);
426: }
428: /* if pc is bjacobi */
429: PetscObjectTypeCompare((PetscObject)pc,PCBJACOBI,&flg);
430: if (flg) {
431: PCBJacobiGetSubKSP(pc, &nksp, &first, &subksp);
432: if (subksp) {
433: for (i=0; i<nksp; i++) {
434: KSPSetTolerances(subksp[i], inner_rtol, outer_abstol, outer_dtol, outer_maxits);
435: }
436: return(0);
437: }
438: }
440: /* if pc is deflation*/
441: PetscObjectTypeCompare((PetscObject)pc,PCDEFLATION,&flg);
442: if (flg) {
443: PCDeflationGetCoarseKSP(pc,&kspinner);
444: KSPSetTolerances(kspinner,inner_rtol,outer_abstol,outer_dtol,PETSC_DEFAULT);
445: return(0);
446: }
448: /* todo: dynamic tolerance may apply to other types of pc too */
449: return(0);
450: }
452: /*
453: Destroy the dummy context used in KSPMonitorDynamicTolerance()
454: */
455: PetscErrorCode KSPMonitorDynamicToleranceDestroy(void **dummy)
456: {
460: PetscFree(*dummy);
461: return(0);
462: }
464: /*
465: Default (short) KSP Monitor, same as KSPMonitorDefault() except
466: it prints fewer digits of the residual as the residual gets smaller.
467: This is because the later digits are meaningless and are often
468: different on different machines; by using this routine different
469: machines will usually generate the same output.
471: Deprecated: Intentionally has no manual page
472: */
473: PetscErrorCode KSPMonitorDefaultShort(KSP ksp,PetscInt its,PetscReal fnorm,PetscViewerAndFormat *dummy)
474: {
476: PetscViewer viewer = dummy->viewer;
480: PetscViewerPushFormat(viewer,dummy->format);
481: PetscViewerASCIIAddTab(viewer,((PetscObject)ksp)->tablevel);
482: if (its == 0 && ((PetscObject)ksp)->prefix) {
483: PetscViewerASCIIPrintf(viewer," Residual norms for %s solve.\n",((PetscObject)ksp)->prefix);
484: }
486: if (fnorm > 1.e-9) {
487: PetscViewerASCIIPrintf(viewer,"%3D KSP Residual norm %g \n",its,(double)fnorm);
488: } else if (fnorm > 1.e-11) {
489: PetscViewerASCIIPrintf(viewer,"%3D KSP Residual norm %5.3e \n",its,(double)fnorm);
490: } else {
491: PetscViewerASCIIPrintf(viewer,"%3D KSP Residual norm < 1.e-11\n",its);
492: }
493: PetscViewerASCIISubtractTab(viewer,((PetscObject)ksp)->tablevel);
494: PetscViewerPopFormat(viewer);
495: return(0);
496: }
498: /*@C
499: KSPConvergedSkip - Convergence test that do not return as converged
500: until the maximum number of iterations is reached.
502: Collective on ksp
504: Input Parameters:
505: + ksp - iterative context
506: . n - iteration number
507: . rnorm - 2-norm residual value (may be estimated)
508: - dummy - unused convergence context
510: Returns:
511: . reason - KSP_CONVERGED_ITERATING, KSP_CONVERGED_ITS
513: Notes:
514: This should be used as the convergence test with the option
515: KSPSetNormType(ksp,KSP_NORM_NONE), since norms of the residual are
516: not computed. Convergence is then declared after the maximum number
517: of iterations have been reached. Useful when one is using CG or
518: BiCGStab as a smoother.
520: Level: advanced
522: .seealso: KSPSetConvergenceTest(), KSPSetTolerances(), KSPSetNormType()
523: @*/
524: PetscErrorCode KSPConvergedSkip(KSP ksp,PetscInt n,PetscReal rnorm,KSPConvergedReason *reason,void *dummy)
525: {
529: *reason = KSP_CONVERGED_ITERATING;
530: if (n >= ksp->max_it) *reason = KSP_CONVERGED_ITS;
531: return(0);
532: }
535: /*@C
536: KSPConvergedDefaultCreate - Creates and initializes the space used by the KSPConvergedDefault() function context
538: Note Collective
540: Output Parameter:
541: . ctx - convergence context
543: Level: intermediate
545: .seealso: KSPConvergedDefault(), KSPConvergedDefaultDestroy(), KSPSetConvergenceTest(), KSPSetTolerances(),
546: KSPConvergedSkip(), KSPConvergedReason, KSPGetConvergedReason(), KSPConvergedDefaultSetUIRNorm(), KSPConvergedDefaultSetUMIRNorm()
547: @*/
548: PetscErrorCode KSPConvergedDefaultCreate(void **ctx)
549: {
550: PetscErrorCode ierr;
551: KSPConvergedDefaultCtx *cctx;
554: PetscNew(&cctx);
555: *ctx = cctx;
556: return(0);
557: }
559: /*@
560: KSPConvergedDefaultSetUIRNorm - makes the default convergence test use || B*(b - A*(initial guess))||
561: instead of || B*b ||. In the case of right preconditioner or if KSPSetNormType(ksp,KSP_NORM_UNPRECONDIITONED)
562: is used there is no B in the above formula. UIRNorm is short for Use Initial Residual Norm.
564: Collective on ksp
566: Input Parameters:
567: . ksp - iterative context
569: Options Database:
570: . -ksp_converged_use_initial_residual_norm
572: Notes:
573: Use KSPSetTolerances() to alter the defaults for rtol, abstol, dtol.
575: The precise values of reason are macros such as KSP_CONVERGED_RTOL, which
576: are defined in petscksp.h.
578: If the convergence test is not KSPConvergedDefault() then this is ignored.
580: If right preconditioning is being used then B does not appear in the above formula.
583: Level: intermediate
585: .seealso: KSPSetConvergenceTest(), KSPSetTolerances(), KSPConvergedSkip(), KSPConvergedReason, KSPGetConvergedReason(), KSPConvergedDefaultSetUMIRNorm()
586: @*/
587: PetscErrorCode KSPConvergedDefaultSetUIRNorm(KSP ksp)
588: {
589: KSPConvergedDefaultCtx *ctx = (KSPConvergedDefaultCtx*) ksp->cnvP;
593: if (ksp->converged != KSPConvergedDefault) return(0);
594: if (ctx->mininitialrtol) SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_WRONGSTATE,"Cannot use KSPConvergedDefaultSetUIRNorm() and KSPConvergedDefaultSetUMIRNorm() together");
595: ctx->initialrtol = PETSC_TRUE;
596: return(0);
597: }
599: /*@
600: KSPConvergedDefaultSetUMIRNorm - makes the default convergence test use min(|| B*(b - A*(initial guess))||,|| B*b ||)
601: In the case of right preconditioner or if KSPSetNormType(ksp,KSP_NORM_UNPRECONDIITONED)
602: is used there is no B in the above formula. UMIRNorm is short for Use Minimum Initial Residual Norm.
604: Collective on ksp
606: Input Parameters:
607: . ksp - iterative context
609: Options Database:
610: . -ksp_converged_use_min_initial_residual_norm
612: Use KSPSetTolerances() to alter the defaults for rtol, abstol, dtol.
614: The precise values of reason are macros such as KSP_CONVERGED_RTOL, which
615: are defined in petscksp.h.
617: Level: intermediate
619: .seealso: KSPSetConvergenceTest(), KSPSetTolerances(), KSPConvergedSkip(), KSPConvergedReason, KSPGetConvergedReason(), KSPConvergedDefaultSetUIRNorm()
620: @*/
621: PetscErrorCode KSPConvergedDefaultSetUMIRNorm(KSP ksp)
622: {
623: KSPConvergedDefaultCtx *ctx = (KSPConvergedDefaultCtx*) ksp->cnvP;
627: if (ksp->converged != KSPConvergedDefault) return(0);
628: if (ctx->initialrtol) SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_WRONGSTATE,"Cannot use KSPConvergedDefaultSetUIRNorm() and KSPConvergedDefaultSetUMIRNorm() together");
629: ctx->mininitialrtol = PETSC_TRUE;
630: return(0);
631: }
633: /*@C
634: KSPConvergedDefault - Determines convergence of the linear iterative solvers by default
636: Collective on ksp
638: Input Parameters:
639: + ksp - iterative context
640: . n - iteration number
641: . rnorm - residual norm (may be estimated, depending on the method may be the preconditioned residual norm)
642: - ctx - convergence context which must be created by KSPConvergedDefaultCreate()
644: Output Parameter:
645: + positive - if the iteration has converged;
646: . negative - if residual norm exceeds divergence threshold;
647: - 0 - otherwise.
649: Notes:
650: KSPConvergedDefault() reaches convergence when rnorm < MAX (rtol * rnorm_0, abstol);
651: Divergence is detected if rnorm > dtol * rnorm_0,
653: where:
654: + rtol = relative tolerance,
655: . abstol = absolute tolerance.
656: . dtol = divergence tolerance,
657: - rnorm_0 is the two norm of the right hand side (or the preconditioned norm, depending on what was set with
658: KSPSetNormType(). When initial guess is non-zero you
659: can call KSPConvergedDefaultSetUIRNorm() to use the norm of (b - A*(initial guess))
660: as the starting point for relative norm convergence testing, that is as rnorm_0
662: Use KSPSetTolerances() to alter the defaults for rtol, abstol, dtol.
664: Use KSPSetNormType() (or -ksp_norm_type <none,preconditioned,unpreconditioned,natural>) to change the norm used for computing rnorm
666: The precise values of reason are macros such as KSP_CONVERGED_RTOL, which are defined in petscksp.h.
668: This routine is used by KSP by default so the user generally never needs call it directly.
670: Use KSPSetConvergenceTest() to provide your own test instead of using this one.
672: Level: intermediate
674: .seealso: KSPSetConvergenceTest(), KSPSetTolerances(), KSPConvergedSkip(), KSPConvergedReason, KSPGetConvergedReason(),
675: KSPConvergedDefaultSetUIRNorm(), KSPConvergedDefaultSetUMIRNorm(), KSPConvergedDefaultCreate(), KSPConvergedDefaultDestroy()
676: @*/
677: PetscErrorCode KSPConvergedDefault(KSP ksp,PetscInt n,PetscReal rnorm,KSPConvergedReason *reason,void *ctx)
678: {
679: PetscErrorCode ierr;
680: KSPConvergedDefaultCtx *cctx = (KSPConvergedDefaultCtx*) ctx;
681: KSPNormType normtype;
686: *reason = KSP_CONVERGED_ITERATING;
688: KSPGetNormType(ksp,&normtype);
689: if (normtype == KSP_NORM_NONE) return(0);
691: if (!cctx) SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_NULL,"Convergence context must have been created with KSPConvergedDefaultCreate()");
692: if (!n) {
693: /* if user gives initial guess need to compute norm of b */
694: if (!ksp->guess_zero && !cctx->initialrtol) {
695: PetscReal snorm = 0.0;
696: if (ksp->normtype == KSP_NORM_UNPRECONDITIONED || ksp->pc_side == PC_RIGHT) {
697: PetscInfo(ksp,"user has provided nonzero initial guess, computing 2-norm of RHS\n");
698: VecNorm(ksp->vec_rhs,NORM_2,&snorm); /* <- b'*b */
699: } else {
700: Vec z;
701: /* Should avoid allocating the z vector each time but cannot stash it in cctx because if KSPReset() is called the vector size might change */
702: VecDuplicate(ksp->vec_rhs,&z);
703: KSP_PCApply(ksp,ksp->vec_rhs,z);
704: if (ksp->normtype == KSP_NORM_PRECONDITIONED) {
705: PetscInfo(ksp,"user has provided nonzero initial guess, computing 2-norm of preconditioned RHS\n");
706: VecNorm(z,NORM_2,&snorm); /* dp <- b'*B'*B*b */
707: } else if (ksp->normtype == KSP_NORM_NATURAL) {
708: PetscScalar norm;
709: PetscInfo(ksp,"user has provided nonzero initial guess, computing natural norm of RHS\n");
710: VecDot(ksp->vec_rhs,z,&norm);
711: snorm = PetscSqrtReal(PetscAbsScalar(norm)); /* dp <- b'*B*b */
712: }
713: VecDestroy(&z);
714: }
715: /* handle special case of zero RHS and nonzero guess */
716: if (!snorm) {
717: PetscInfo(ksp,"Special case, user has provided nonzero initial guess and zero RHS\n");
718: snorm = rnorm;
719: }
720: if (cctx->mininitialrtol) ksp->rnorm0 = PetscMin(snorm,rnorm);
721: else ksp->rnorm0 = snorm;
722: } else {
723: ksp->rnorm0 = rnorm;
724: }
725: ksp->ttol = PetscMax(ksp->rtol*ksp->rnorm0,ksp->abstol);
726: }
728: if (n <= ksp->chknorm) return(0);
730: if (PetscIsInfOrNanReal(rnorm)) {
731: PCFailedReason pcreason;
732: PetscInt sendbuf,pcreason_max;
733: PCGetFailedReason(ksp->pc,&pcreason);
734: sendbuf = (PetscInt)pcreason;
735: MPI_Allreduce(&sendbuf,&pcreason_max,1,MPIU_INT,MPIU_MAX,PetscObjectComm((PetscObject)ksp));
736: if (pcreason_max) {
737: *reason = KSP_DIVERGED_PC_FAILED;
738: VecSetInf(ksp->vec_sol);
739: PetscInfo(ksp,"Linear solver pcsetup fails, declaring divergence \n");
740: } else {
741: *reason = KSP_DIVERGED_NANORINF;
742: PetscInfo(ksp,"Linear solver has created a not a number (NaN) as the residual norm, declaring divergence \n");
743: }
744: } else if (rnorm <= ksp->ttol) {
745: if (rnorm < ksp->abstol) {
746: PetscInfo3(ksp,"Linear solver has converged. Residual norm %14.12e is less than absolute tolerance %14.12e at iteration %D\n",(double)rnorm,(double)ksp->abstol,n);
747: *reason = KSP_CONVERGED_ATOL;
748: } else {
749: if (cctx->initialrtol) {
750: PetscInfo4(ksp,"Linear solver has converged. Residual norm %14.12e is less than relative tolerance %14.12e times initial residual norm %14.12e at iteration %D\n",(double)rnorm,(double)ksp->rtol,(double)ksp->rnorm0,n);
751: } else {
752: PetscInfo4(ksp,"Linear solver has converged. Residual norm %14.12e is less than relative tolerance %14.12e times initial right hand side norm %14.12e at iteration %D\n",(double)rnorm,(double)ksp->rtol,(double)ksp->rnorm0,n);
753: }
754: *reason = KSP_CONVERGED_RTOL;
755: }
756: } else if (rnorm >= ksp->divtol*ksp->rnorm0) {
757: PetscInfo3(ksp,"Linear solver is diverging. Initial right hand size norm %14.12e, current residual norm %14.12e at iteration %D\n",(double)ksp->rnorm0,(double)rnorm,n);
758: *reason = KSP_DIVERGED_DTOL;
759: }
760: return(0);
761: }
763: /*@C
764: KSPConvergedDefaultDestroy - Frees the space used by the KSPConvergedDefault() function context
766: Not Collective
768: Input Parameters:
769: . ctx - convergence context
771: Level: intermediate
773: .seealso: KSPConvergedDefault(), KSPConvergedDefaultCreate(), KSPSetConvergenceTest(), KSPSetTolerances(), KSPConvergedSkip(),
774: KSPConvergedReason, KSPGetConvergedReason(), KSPConvergedDefaultSetUIRNorm(), KSPConvergedDefaultSetUMIRNorm()
775: @*/
776: PetscErrorCode KSPConvergedDefaultDestroy(void *ctx)
777: {
778: PetscErrorCode ierr;
779: KSPConvergedDefaultCtx *cctx = (KSPConvergedDefaultCtx*) ctx;
782: VecDestroy(&cctx->work);
783: PetscFree(ctx);
784: return(0);
785: }
787: /*
788: KSPBuildSolutionDefault - Default code to create/move the solution.
790: Collective on ksp
792: Input Parameters:
793: + ksp - iterative context
794: - v - pointer to the user's vector
796: Output Parameter:
797: . V - pointer to a vector containing the solution
799: Level: advanced
801: Developers Note: This is PETSC_EXTERN because it may be used by user written plugin KSP implementations
803: .seealso: KSPGetSolution(), KSPBuildResidualDefault()
804: */
805: PetscErrorCode KSPBuildSolutionDefault(KSP ksp,Vec v,Vec *V)
806: {
810: if (ksp->pc_side == PC_RIGHT) {
811: if (ksp->pc) {
812: if (v) {
813: KSP_PCApply(ksp,ksp->vec_sol,v); *V = v;
814: } else SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_SUP,"Not working with right preconditioner");
815: } else {
816: if (v) {
817: VecCopy(ksp->vec_sol,v); *V = v;
818: } else *V = ksp->vec_sol;
819: }
820: } else if (ksp->pc_side == PC_SYMMETRIC) {
821: if (ksp->pc) {
822: if (ksp->transpose_solve) SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_SUP,"Not working with symmetric preconditioner and transpose solve");
823: if (v) {
824: PCApplySymmetricRight(ksp->pc,ksp->vec_sol,v);
825: *V = v;
826: } else SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_SUP,"Not working with symmetric preconditioner");
827: } else {
828: if (v) {
829: VecCopy(ksp->vec_sol,v); *V = v;
830: } else *V = ksp->vec_sol;
831: }
832: } else {
833: if (v) {
834: VecCopy(ksp->vec_sol,v); *V = v;
835: } else *V = ksp->vec_sol;
836: }
837: return(0);
838: }
840: /*
841: KSPBuildResidualDefault - Default code to compute the residual.
843: Collecive on ksp
845: Input Parameters:
846: . ksp - iterative context
847: . t - pointer to temporary vector
848: . v - pointer to user vector
850: Output Parameter:
851: . V - pointer to a vector containing the residual
853: Level: advanced
855: Developers Note: This is PETSC_EXTERN because it may be used by user written plugin KSP implementations
857: .seealso: KSPBuildSolutionDefault()
858: */
859: PetscErrorCode KSPBuildResidualDefault(KSP ksp,Vec t,Vec v,Vec *V)
860: {
862: Mat Amat,Pmat;
865: if (!ksp->pc) {KSPGetPC(ksp,&ksp->pc);}
866: PCGetOperators(ksp->pc,&Amat,&Pmat);
867: KSPBuildSolution(ksp,t,NULL);
868: KSP_MatMult(ksp,Amat,t,v);
869: VecAYPX(v,-1.0,ksp->vec_rhs);
870: *V = v;
871: return(0);
872: }
874: /*@C
875: KSPCreateVecs - Gets a number of work vectors.
877: Collective on ksp
879: Input Parameters:
880: + ksp - iterative context
881: . rightn - number of right work vectors
882: - leftn - number of left work vectors to allocate
884: Output Parameter:
885: + right - the array of vectors created
886: - left - the array of left vectors
888: Note: The right vector has as many elements as the matrix has columns. The left
889: vector has as many elements as the matrix has rows.
891: The vectors are new vectors that are not owned by the KSP, they should be destroyed with calls to VecDestroyVecs() when no longer needed.
893: Developers Note: First tries to duplicate the rhs and solution vectors of the KSP, if they do not exist tries to get them from the matrix, if
894: that does not exist tries to get them from the DM (if it is provided).
896: Level: advanced
898: .seealso: MatCreateVecs(), VecDestroyVecs()
900: @*/
901: PetscErrorCode KSPCreateVecs(KSP ksp,PetscInt rightn, Vec **right,PetscInt leftn,Vec **left)
902: {
904: Vec vecr = NULL,vecl = NULL;
905: PetscBool matset,pmatset;
906: Mat mat = NULL;
909: if (rightn) {
910: if (!right) SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_INCOMP,"You asked for right vectors but did not pass a pointer to hold them");
911: if (ksp->vec_sol) vecr = ksp->vec_sol;
912: else {
913: if (ksp->pc) {
914: PCGetOperatorsSet(ksp->pc,&matset,&pmatset);
915: /* check for mat before pmat because for KSPLSQR pmat may be a different size than mat since pmat maybe mat'*mat */
916: if (matset) {
917: PCGetOperators(ksp->pc,&mat,NULL);
918: MatCreateVecs(mat,&vecr,NULL);
919: } else if (pmatset) {
920: PCGetOperators(ksp->pc,NULL,&mat);
921: MatCreateVecs(mat,&vecr,NULL);
922: }
923: }
924: if (!vecr) {
925: if (ksp->dm) {
926: DMGetGlobalVector(ksp->dm,&vecr);
927: } else SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_WRONGSTATE,"You requested a vector from a KSP that cannot provide one");
928: }
929: }
930: VecDuplicateVecs(vecr,rightn,right);
931: if (!ksp->vec_sol) {
932: if (mat) {
933: VecDestroy(&vecr);
934: } else if (ksp->dm) {
935: DMRestoreGlobalVector(ksp->dm,&vecr);
936: }
937: }
938: }
939: if (leftn) {
940: if (!left) SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_INCOMP,"You asked for left vectors but did not pass a pointer to hold them");
941: if (ksp->vec_rhs) vecl = ksp->vec_rhs;
942: else {
943: if (ksp->pc) {
944: PCGetOperatorsSet(ksp->pc,&matset,&pmatset);
945: /* check for mat before pmat because for KSPLSQR pmat may be a different size than mat since pmat maybe mat'*mat */
946: if (matset) {
947: PCGetOperators(ksp->pc,&mat,NULL);
948: MatCreateVecs(mat,NULL,&vecl);
949: } else if (pmatset) {
950: PCGetOperators(ksp->pc,NULL,&mat);
951: MatCreateVecs(mat,NULL,&vecl);
952: }
953: }
954: if (!vecl) {
955: if (ksp->dm) {
956: DMGetGlobalVector(ksp->dm,&vecl);
957: } else SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_WRONGSTATE,"You requested a vector from a KSP that cannot provide one");
958: }
959: }
960: VecDuplicateVecs(vecl,leftn,left);
961: if (!ksp->vec_rhs) {
962: if (mat) {
963: VecDestroy(&vecl);
964: } else if (ksp->dm) {
965: DMRestoreGlobalVector(ksp->dm,&vecl);
966: }
967: }
968: }
969: return(0);
970: }
972: /*@C
973: KSPSetWorkVecs - Sets a number of work vectors into a KSP object
975: Collective on ksp
977: Input Parameters:
978: + ksp - iterative context
979: - nw - number of work vectors to allocate
981: Level: developer
983: Developers Note: This is PETSC_EXTERN because it may be used by user written plugin KSP implementations
985: @*/
986: PetscErrorCode KSPSetWorkVecs(KSP ksp,PetscInt nw)
987: {
991: VecDestroyVecs(ksp->nwork,&ksp->work);
992: ksp->nwork = nw;
993: KSPCreateVecs(ksp,nw,&ksp->work,0,NULL);
994: PetscLogObjectParents(ksp,nw,ksp->work);
995: return(0);
996: }
998: /*
999: KSPDestroyDefault - Destroys a iterative context variable for methods with
1000: no separate context. Preferred calling sequence KSPDestroy().
1002: Input Parameter:
1003: . ksp - the iterative context
1005: Developers Note: This is PETSC_EXTERN because it may be used by user written plugin KSP implementations
1007: */
1008: PetscErrorCode KSPDestroyDefault(KSP ksp)
1009: {
1014: PetscFree(ksp->data);
1015: return(0);
1016: }
1018: /*@
1019: KSPGetConvergedReason - Gets the reason the KSP iteration was stopped.
1021: Not Collective
1023: Input Parameter:
1024: . ksp - the KSP context
1026: Output Parameter:
1027: . reason - negative value indicates diverged, positive value converged, see KSPConvergedReason
1029: Possible values for reason: See also manual page for each reason
1030: $ KSP_CONVERGED_RTOL (residual 2-norm decreased by a factor of rtol, from 2-norm of right hand side)
1031: $ KSP_CONVERGED_ATOL (residual 2-norm less than abstol)
1032: $ KSP_CONVERGED_ITS (used by the preonly preconditioner that always uses ONE iteration, or when the KSPConvergedSkip() convergence test routine is set.
1033: $ KSP_CONVERGED_CG_NEG_CURVE (see note below)
1034: $ KSP_CONVERGED_CG_CONSTRAINED (see note below)
1035: $ KSP_CONVERGED_STEP_LENGTH (see note below)
1036: $ KSP_CONVERGED_ITERATING (returned if the solver is not yet finished)
1037: $ KSP_DIVERGED_ITS (required more than its to reach convergence)
1038: $ KSP_DIVERGED_DTOL (residual norm increased by a factor of divtol)
1039: $ KSP_DIVERGED_NANORINF (residual norm became Not-a-number or Inf likely due to 0/0)
1040: $ KSP_DIVERGED_BREAKDOWN (generic breakdown in method)
1041: $ KSP_DIVERGED_BREAKDOWN_BICG (Initial residual is orthogonal to preconditioned initial residual. Try a different preconditioner, or a different initial Level.)
1043: Options Database:
1044: . -ksp_converged_reason - prints the reason to standard out
1046: Notes:
1047: If this routine is called before or doing the KSPSolve() the value of KSP_CONVERGED_ITERATING is returned
1049: The values KSP_CONVERGED_CG_NEG_CURVE, KSP_CONVERGED_CG_CONSTRAINED, and KSP_CONVERGED_STEP_LENGTH are returned only by the special KSPNASH, KSPSTCG, and KSPGLTR
1050: solvers which are used by the SNESNEWTONTR (trust region) solver.
1052: Level: intermediate
1054: .seealso: KSPSetConvergenceTest(), KSPConvergedDefault(), KSPSetTolerances(), KSPConvergedReason
1055: @*/
1056: PetscErrorCode KSPGetConvergedReason(KSP ksp,KSPConvergedReason *reason)
1057: {
1061: *reason = ksp->reason;
1062: return(0);
1063: }
1065: #include <petsc/private/dmimpl.h>
1066: /*@
1067: KSPSetDM - Sets the DM that may be used by some preconditioners
1069: Logically Collective on ksp
1071: Input Parameters:
1072: + ksp - the preconditioner context
1073: - dm - the dm, cannot be NULL
1075: Notes:
1076: If this is used then the KSP will attempt to use the DM to create the matrix and use the routine set with
1077: DMKSPSetComputeOperators(). Use KSPSetDMActive(ksp,PETSC_FALSE) to instead use the matrix you've provided with
1078: KSPSetOperators().
1080: A DM can only be used for solving one problem at a time because information about the problem is stored on the DM,
1081: even when not using interfaces like DMKSPSetComputeOperators(). Use DMClone() to get a distinct DM when solving
1082: different problems using the same function space.
1084: Level: intermediate
1086: .seealso: KSPGetDM(), KSPSetDMActive(), KSPSetComputeOperators(), KSPSetComputeRHS(), KSPSetComputeInitialGuess(), DMKSPSetComputeOperators(), DMKSPSetComputeRHS(), DMKSPSetComputeInitialGuess()
1087: @*/
1088: PetscErrorCode KSPSetDM(KSP ksp,DM dm)
1089: {
1091: PC pc;
1096: PetscObjectReference((PetscObject)dm);
1097: if (ksp->dm) { /* Move the DMSNES context over to the new DM unless the new DM already has one */
1098: if (ksp->dm->dmksp && !dm->dmksp) {
1099: DMKSP kdm;
1100: DMCopyDMKSP(ksp->dm,dm);
1101: DMGetDMKSP(ksp->dm,&kdm);
1102: if (kdm->originaldm == ksp->dm) kdm->originaldm = dm; /* Grant write privileges to the replacement DM */
1103: }
1104: DMDestroy(&ksp->dm);
1105: }
1106: ksp->dm = dm;
1107: ksp->dmAuto = PETSC_FALSE;
1108: KSPGetPC(ksp,&pc);
1109: PCSetDM(pc,dm);
1110: ksp->dmActive = PETSC_TRUE;
1111: return(0);
1112: }
1114: /*@
1115: KSPSetDMActive - Indicates the DM should be used to generate the linear system matrix and right hand side
1117: Logically Collective on ksp
1119: Input Parameters:
1120: + ksp - the preconditioner context
1121: - flg - use the DM
1123: Notes:
1124: By default KSPSetDM() sets the DM as active, call KSPSetDMActive(ksp,PETSC_FALSE); after KSPSetDM(ksp,dm) to not have the KSP object use the DM to generate the matrices.
1126: Level: intermediate
1128: .seealso: KSPGetDM(), KSPSetDM(), SNESSetDM(), KSPSetComputeOperators(), KSPSetComputeRHS(), KSPSetComputeInitialGuess()
1129: @*/
1130: PetscErrorCode KSPSetDMActive(KSP ksp,PetscBool flg)
1131: {
1135: ksp->dmActive = flg;
1136: return(0);
1137: }
1139: /*@
1140: KSPGetDM - Gets the DM that may be used by some preconditioners
1142: Not Collective
1144: Input Parameter:
1145: . ksp - the preconditioner context
1147: Output Parameter:
1148: . dm - the dm
1150: Level: intermediate
1153: .seealso: KSPSetDM(), KSPSetDMActive()
1154: @*/
1155: PetscErrorCode KSPGetDM(KSP ksp,DM *dm)
1156: {
1161: if (!ksp->dm) {
1162: DMShellCreate(PetscObjectComm((PetscObject)ksp),&ksp->dm);
1163: ksp->dmAuto = PETSC_TRUE;
1164: }
1165: *dm = ksp->dm;
1166: return(0);
1167: }
1169: /*@
1170: KSPSetApplicationContext - Sets the optional user-defined context for the linear solver.
1172: Logically Collective on ksp
1174: Input Parameters:
1175: + ksp - the KSP context
1176: - usrP - optional user context
1178: Fortran Notes:
1179: To use this from Fortran you must write a Fortran interface definition for this
1180: function that tells Fortran the Fortran derived data type that you are passing in as the ctx argument.
1182: Level: intermediate
1184: .seealso: KSPGetApplicationContext()
1185: @*/
1186: PetscErrorCode KSPSetApplicationContext(KSP ksp,void *usrP)
1187: {
1189: PC pc;
1193: ksp->user = usrP;
1194: KSPGetPC(ksp,&pc);
1195: PCSetApplicationContext(pc,usrP);
1196: return(0);
1197: }
1199: /*@
1200: KSPGetApplicationContext - Gets the user-defined context for the linear solver.
1202: Not Collective
1204: Input Parameter:
1205: . ksp - KSP context
1207: Output Parameter:
1208: . usrP - user context
1210: Fortran Notes:
1211: To use this from Fortran you must write a Fortran interface definition for this
1212: function that tells Fortran the Fortran derived data type that you are passing in as the ctx argument.
1214: Level: intermediate
1216: .seealso: KSPSetApplicationContext()
1217: @*/
1218: PetscErrorCode KSPGetApplicationContext(KSP ksp,void *usrP)
1219: {
1222: *(void**)usrP = ksp->user;
1223: return(0);
1224: }
1226: #include <petsc/private/pcimpl.h>
1228: /*@
1229: KSPCheckSolve - Checks if the PCSetUp() or KSPSolve() failed and set the error flag for the outer PC. A KSP_DIVERGED_ITS is
1230: not considered a failure in this context
1232: Collective on ksp
1234: Input Parameter:
1235: + ksp - the linear solver (KSP) context.
1236: . pc - the preconditioner context
1237: - vec - a vector that will be initialized with Inf to indicate lack of convergence
1239: Notes: this may be called by a subset of the processes in the PC
1241: Level: developer
1243: Developer Note: this is used to manage returning from preconditioners whose inner KSP solvers have failed in some way
1245: .seealso: KSPCreate(), KSPSetType(), KSP, KSPCheckNorm(), KSPCheckDot()
1246: @*/
1247: PetscErrorCode KSPCheckSolve(KSP ksp,PC pc,Vec vec)
1248: {
1249: PetscErrorCode ierr;
1250: PCFailedReason pcreason;
1251: PC subpc;
1254: KSPGetPC(ksp,&subpc);
1255: PCGetFailedReason(subpc,&pcreason);
1256: if (pcreason || (ksp->reason < 0 && ksp->reason != KSP_DIVERGED_ITS)) {
1257: if (pc->erroriffailure) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_NOT_CONVERGED,"Detected not converged in KSP inner solve: KSP reason %s PC reason %s",KSPConvergedReasons[ksp->reason],PCFailedReasons[pcreason]);
1258: else {
1259: PetscInfo2(ksp,"Detected not converged in KSP inner solve: KSP reason %s PC reason %s\n",KSPConvergedReasons[ksp->reason],PCFailedReasons[pcreason]);
1260: pc->failedreason = PC_SUBPC_ERROR;
1261: VecSetInf(vec);
1262: }
1263: }
1264: return(0);
1265: }
1266: