Actual source code: itfunc.c
1: /*
2: Interface KSP routines that the user calls.
3: */
5: #include <petsc/private/kspimpl.h>
6: #include <petsc/private/matimpl.h>
7: #include <petscdm.h>
9: /* number of nested levels of KSPSetUp/Solve(). This is used to determine if KSP_DIVERGED_ITS should be fatal. */
10: static PetscInt level = 0;
12: static inline PetscErrorCode ObjectView(PetscObject obj, PetscViewer viewer, PetscViewerFormat format)
13: {
14: PetscViewerPushFormat(viewer, format);
15: PetscObjectView(obj, viewer);
16: PetscViewerPopFormat(viewer);
17: return (0);
18: }
20: /*@
21: KSPComputeExtremeSingularValues - Computes the extreme singular values
22: for the preconditioned operator. Called after or during `KSPSolve()`.
24: Not Collective
26: Input Parameter:
27: . ksp - iterative context obtained from `KSPCreate()`
29: Output Parameters:
30: . emin, emax - extreme singular values
32: Options Database Keys:
33: . -ksp_view_singularvalues - compute extreme singular values and print when `KSPSolve()` completes.
35: Notes:
36: One must call `KSPSetComputeSingularValues()` before calling `KSPSetUp()`
37: (or use the option -ksp_view_eigenvalues) in order for this routine to work correctly.
39: Many users may just want to use the monitoring routine
40: `KSPMonitorSingularValue()` (which can be set with option -ksp_monitor_singular_value)
41: to print the extreme singular values at each iteration of the linear solve.
43: Estimates of the smallest singular value may be very inaccurate, especially if the Krylov method has not converged.
44: The largest singular value is usually accurate to within a few percent if the method has converged, but is still not
45: intended for eigenanalysis. Consider the excellent package `SLEPc` if accurate values are required.
47: Disable restarts if using KSPGMRES, otherwise this estimate will only be using those iterations after the last
48: restart. See `KSPGMRESSetRestart()` for more details.
50: Level: advanced
52: .seealso: [](chapter_ksp), `KSPSetComputeSingularValues()`, `KSPMonitorSingularValue()`, `KSPComputeEigenvalues()`, `KSP`
53: @*/
54: PetscErrorCode KSPComputeExtremeSingularValues(KSP ksp, PetscReal *emax, PetscReal *emin)
55: {
61: if (ksp->ops->computeextremesingularvalues) PetscUseTypeMethod(ksp, computeextremesingularvalues, emax, emin);
62: else {
63: *emin = -1.0;
64: *emax = -1.0;
65: }
66: return 0;
67: }
69: /*@
70: KSPComputeEigenvalues - Computes the extreme eigenvalues for the
71: preconditioned operator. Called after or during `KSPSolve()`.
73: Not Collective
75: Input Parameters:
76: + ksp - iterative context obtained from `KSPCreate()`
77: - n - size of arrays r and c. The number of eigenvalues computed (neig) will, in
78: general, be less than this.
80: Output Parameters:
81: + r - real part of computed eigenvalues, provided by user with a dimension of at least n
82: . c - complex part of computed eigenvalues, provided by user with a dimension of at least n
83: - neig - actual number of eigenvalues computed (will be less than or equal to n)
85: Options Database Keys:
86: . -ksp_view_eigenvalues - Prints eigenvalues to stdout
88: Notes:
89: The number of eigenvalues estimated depends on the size of the Krylov space
90: generated during the `KSPSolve()` ; for example, with
91: CG it corresponds to the number of CG iterations, for GMRES it is the number
92: of GMRES iterations SINCE the last restart. Any extra space in r[] and c[]
93: will be ignored.
95: `KSPComputeEigenvalues()` does not usually provide accurate estimates; it is
96: intended only for assistance in understanding the convergence of iterative
97: methods, not for eigenanalysis. For accurate computation of eigenvalues we recommend using
98: the excellent package SLEPc.
100: One must call `KSPSetComputeEigenvalues()` before calling `KSPSetUp()`
101: in order for this routine to work correctly.
103: Many users may just want to use the monitoring routine
104: `KSPMonitorSingularValue()` (which can be set with option -ksp_monitor_singular_value)
105: to print the singular values at each iteration of the linear solve.
107: `KSPComputeRitz()` provides estimates for both the eigenvalues and their corresponding eigenvectors.
109: Level: advanced
111: .seealso: [](chapter_ksp), `KSPSetComputeEigenvalues()`, `KSPSetComputeSingularValues()`, `KSPMonitorSingularValue()`, `KSPComputeExtremeSingularValues()`, `KSP`, `KSPComputeRitz()`
112: @*/
113: PetscErrorCode KSPComputeEigenvalues(KSP ksp, PetscInt n, PetscReal r[], PetscReal c[], PetscInt *neig)
114: {
122: if (n && ksp->ops->computeeigenvalues) PetscUseTypeMethod(ksp, computeeigenvalues, n, r, c, neig);
123: else *neig = 0;
124: return 0;
125: }
127: /*@
128: KSPComputeRitz - Computes the Ritz or harmonic Ritz pairs associated with the
129: smallest or largest in modulus, for the preconditioned operator.
131: Not Collective
133: Input Parameters:
134: + ksp - iterative context obtained from `KSPCreate()`
135: . ritz - `PETSC_TRUE` or `PETSC_FALSE` for Ritz pairs or harmonic Ritz pairs, respectively
136: - small - `PETSC_TRUE` or `PETSC_FALSE` for smallest or largest (harmonic) Ritz values, respectively
138: Output Parameters:
139: + nrit - On input number of (harmonic) Ritz pairs to compute; on output, actual number of computed (harmonic) Ritz pairs
140: . S - an array of the Ritz vectors, pass in an array of vectors of size nrit
141: . tetar - real part of the Ritz values, pass in an array of size nrit
142: - tetai - imaginary part of the Ritz values, pass in an array of size nrit
144: Notes:
145: This only works with a `KSPType` of `KSPGMRES`.
147: One must call `KSPSetComputeRitz()` before calling `KSPSetUp()` in order for this routine to work correctly.
149: This routine must be called after `KSPSolve()`.
151: In GMRES, the (harmonic) Ritz pairs are computed from the Hessenberg matrix obtained during
152: the last complete cycle of the GMRES solve, or during the partial cycle if the solve ended before
153: a restart (that is a complete GMRES cycle was never achieved).
155: The number of actual (harmonic) Ritz pairs computed is less than or equal to the restart
156: parameter for GMRES if a complete cycle has been performed or less or equal to the number of GMRES
157: iterations.
159: `KSPComputeEigenvalues()` provides estimates for only the eigenvalues (Ritz values).
161: For real matrices, the (harmonic) Ritz pairs can be complex-valued. In such a case,
162: the routine selects the complex (harmonic) Ritz value and its conjugate, and two successive entries of the
163: vectors S are equal to the real and the imaginary parts of the associated vectors.
164: When PETSc has been built with complex scalars, the real and imaginary parts of the Ritz
165: values are still returned in tetar and tetai, as is done in `KSPComputeEigenvalues()`, but
166: the Ritz vectors S are complex.
168: The (harmonic) Ritz pairs are given in order of increasing (harmonic) Ritz values in modulus.
170: The Ritz pairs do not necessarily accurately reflect the eigenvalues and eigenvectors of the operator, consider the
171: excellent package `SLEPc` if accurate values are required.
173: Level: advanced
175: .seealso: [](chapter_ksp), `KSPSetComputeRitz()`, `KSP`, `KSPGMRES`, `KSPComputeEigenvalues()`, `KSPSetComputeSingularValues()`, `KSPMonitorSingularValue()`
176: @*/
177: PetscErrorCode KSPComputeRitz(KSP ksp, PetscBool ritz, PetscBool small, PetscInt *nrit, Vec S[], PetscReal tetar[], PetscReal tetai[])
178: {
181: PetscTryTypeMethod(ksp, computeritz, ritz, small, nrit, S, tetar, tetai);
182: return 0;
183: }
184: /*@
185: KSPSetUpOnBlocks - Sets up the preconditioner for each block in
186: the block Jacobi, block Gauss-Seidel, and overlapping Schwarz
187: methods.
189: Collective
191: Input Parameter:
192: . ksp - the `KSP` context
194: Notes:
195: `KSPSetUpOnBlocks()` is a routine that the user can optionally call for
196: more precise profiling (via -log_view) of the setup phase for these
197: block preconditioners. If the user does not call `KSPSetUpOnBlocks()`,
198: it will automatically be called from within `KSPSolve()`.
200: Calling `KSPSetUpOnBlocks()` is the same as calling `PCSetUpOnBlocks()`
201: on the PC context within the `KSP` context.
203: Level: advanced
205: .seealso: [](chapter_ksp), `PCSetUpOnBlocks()`, `KSPSetUp()`, `PCSetUp()`, `KSP`
206: @*/
207: PetscErrorCode KSPSetUpOnBlocks(KSP ksp)
208: {
209: PC pc;
210: PCFailedReason pcreason;
213: level++;
214: KSPGetPC(ksp, &pc);
215: PCSetUpOnBlocks(pc);
216: PCGetFailedReasonRank(pc, &pcreason);
217: level--;
218: /*
219: This is tricky since only a subset of MPI ranks may set this; each KSPSolve_*() is responsible for checking
220: this flag and initializing an appropriate vector with VecSetInf() so that the first norm computation can
221: produce a result at KSPCheckNorm() thus communicating the known problem to all MPI ranks so they may
222: terminate the Krylov solve. For many KSP implementations this is handled within KSPInitialResidual()
223: */
224: if (pcreason) ksp->reason = KSP_DIVERGED_PC_FAILED;
225: return 0;
226: }
228: /*@
229: KSPSetReusePreconditioner - reuse the current preconditioner, do not construct a new one even if the operator changes
231: Collective
233: Input Parameters:
234: + ksp - iterative context obtained from `KSPCreate()`
235: - flag - `PETSC_TRUE` to reuse the current preconditioner
237: Level: intermediate
239: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSolve()`, `KSPDestroy()`, `PCSetReusePreconditioner()`, `KSP`
240: @*/
241: PetscErrorCode KSPSetReusePreconditioner(KSP ksp, PetscBool flag)
242: {
243: PC pc;
246: KSPGetPC(ksp, &pc);
247: PCSetReusePreconditioner(pc, flag);
248: return 0;
249: }
251: /*@
252: KSPGetReusePreconditioner - Determines if the `KSP` reuses the current preconditioner even if the operator in the preconditioner has changed.
254: Collective
256: Input Parameters:
257: . ksp - iterative context obtained from `KSPCreate()`
259: Output Parameters:
260: . flag - the boolean flag
262: Level: intermediate
264: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSolve()`, `KSPDestroy()`, `KSPSetReusePreconditioner()`, `KSP`
265: @*/
266: PetscErrorCode KSPGetReusePreconditioner(KSP ksp, PetscBool *flag)
267: {
270: *flag = PETSC_FALSE;
271: if (ksp->pc) PCGetReusePreconditioner(ksp->pc, flag);
272: return 0;
273: }
275: /*@
276: KSPSetSkipPCSetFromOptions - prevents `KSPSetFromOptions()` from calling `PCSetFromOptions()`. This is used if the same `PC` is shared by more than one `KSP` so its options are not resettable for each `KSP`
278: Collective
280: Input Parameters:
281: + ksp - iterative context obtained from `KSPCreate()`
282: - flag - `PETSC_TRUE` to skip calling the `PCSetFromOptions()`
284: Level: intermediate
286: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSolve()`, `KSPDestroy()`, `PCSetReusePreconditioner()`, `KSP`
287: @*/
288: PetscErrorCode KSPSetSkipPCSetFromOptions(KSP ksp, PetscBool flag)
289: {
291: ksp->skippcsetfromoptions = flag;
292: return 0;
293: }
295: /*@
296: KSPSetUp - Sets up the internal data structures for the
297: later use of an iterative solver.
299: Collective
301: Input Parameter:
302: . ksp - iterative context obtained from `KSPCreate()`
304: Level: developer
306: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSolve()`, `KSPDestroy()`, `KSP`
307: @*/
308: PetscErrorCode KSPSetUp(KSP ksp)
309: {
310: Mat A, B;
311: Mat mat, pmat;
312: MatNullSpace nullsp;
313: PCFailedReason pcreason;
316: level++;
318: /* reset the convergence flag from the previous solves */
319: ksp->reason = KSP_CONVERGED_ITERATING;
321: if (!((PetscObject)ksp)->type_name) KSPSetType(ksp, KSPGMRES);
322: KSPSetUpNorms_Private(ksp, PETSC_TRUE, &ksp->normtype, &ksp->pc_side);
324: if (ksp->dmActive && !ksp->setupstage) {
325: /* first time in so build matrix and vector data structures using DM */
326: if (!ksp->vec_rhs) DMCreateGlobalVector(ksp->dm, &ksp->vec_rhs);
327: if (!ksp->vec_sol) DMCreateGlobalVector(ksp->dm, &ksp->vec_sol);
328: DMCreateMatrix(ksp->dm, &A);
329: KSPSetOperators(ksp, A, A);
330: PetscObjectDereference((PetscObject)A);
331: }
333: if (ksp->dmActive) {
334: DMKSP kdm;
335: DMGetDMKSP(ksp->dm, &kdm);
337: if (kdm->ops->computeinitialguess && ksp->setupstage != KSP_SETUP_NEWRHS) {
338: /* only computes initial guess the first time through */
339: PetscCallBack("KSP callback initial guess", (*kdm->ops->computeinitialguess)(ksp, ksp->vec_sol, kdm->initialguessctx));
340: KSPSetInitialGuessNonzero(ksp, PETSC_TRUE);
341: }
342: if (kdm->ops->computerhs) PetscCallBack("KSP callback rhs", (*kdm->ops->computerhs)(ksp, ksp->vec_rhs, kdm->rhsctx));
344: if (ksp->setupstage != KSP_SETUP_NEWRHS) {
345: if (kdm->ops->computeoperators) {
346: KSPGetOperators(ksp, &A, &B);
347: PetscCallBack("KSP callback operators", (*kdm->ops->computeoperators)(ksp, A, B, kdm->operatorsctx));
348: } else SETERRQ(PetscObjectComm((PetscObject)ksp), PETSC_ERR_ARG_WRONGSTATE, "You called KSPSetDM() but did not use DMKSPSetComputeOperators() or KSPSetDMActive(ksp,PETSC_FALSE);");
349: }
350: }
352: if (ksp->setupstage == KSP_SETUP_NEWRHS) {
353: level--;
354: return 0;
355: }
356: PetscLogEventBegin(KSP_SetUp, ksp, ksp->vec_rhs, ksp->vec_sol, 0);
358: switch (ksp->setupstage) {
359: case KSP_SETUP_NEW:
360: PetscUseTypeMethod(ksp, setup);
361: break;
362: case KSP_SETUP_NEWMATRIX: { /* This should be replaced with a more general mechanism */
363: if (ksp->setupnewmatrix) PetscUseTypeMethod(ksp, setup);
364: } break;
365: default:
366: break;
367: }
369: if (!ksp->pc) KSPGetPC(ksp, &ksp->pc);
370: PCGetOperators(ksp->pc, &mat, &pmat);
371: /* scale the matrix if requested */
372: if (ksp->dscale) {
373: PetscScalar *xx;
374: PetscInt i, n;
375: PetscBool zeroflag = PETSC_FALSE;
376: if (!ksp->pc) KSPGetPC(ksp, &ksp->pc);
377: if (!ksp->diagonal) { /* allocate vector to hold diagonal */
378: MatCreateVecs(pmat, &ksp->diagonal, NULL);
379: }
380: MatGetDiagonal(pmat, ksp->diagonal);
381: VecGetLocalSize(ksp->diagonal, &n);
382: VecGetArray(ksp->diagonal, &xx);
383: for (i = 0; i < n; i++) {
384: if (xx[i] != 0.0) xx[i] = 1.0 / PetscSqrtReal(PetscAbsScalar(xx[i]));
385: else {
386: xx[i] = 1.0;
387: zeroflag = PETSC_TRUE;
388: }
389: }
390: VecRestoreArray(ksp->diagonal, &xx);
391: if (zeroflag) PetscInfo(ksp, "Zero detected in diagonal of matrix, using 1 at those locations\n");
392: MatDiagonalScale(pmat, ksp->diagonal, ksp->diagonal);
393: if (mat != pmat) MatDiagonalScale(mat, ksp->diagonal, ksp->diagonal);
394: ksp->dscalefix2 = PETSC_FALSE;
395: }
396: PetscLogEventEnd(KSP_SetUp, ksp, ksp->vec_rhs, ksp->vec_sol, 0);
397: PCSetErrorIfFailure(ksp->pc, ksp->errorifnotconverged);
398: PCSetUp(ksp->pc);
399: PCGetFailedReasonRank(ksp->pc, &pcreason);
400: /* TODO: this code was wrong and is still wrong, there is no way to propagate the failure to all processes; their is no code to handle a ksp->reason on only some ranks */
401: if (pcreason) ksp->reason = KSP_DIVERGED_PC_FAILED;
403: MatGetNullSpace(mat, &nullsp);
404: if (nullsp) {
405: PetscBool test = PETSC_FALSE;
406: PetscOptionsGetBool(((PetscObject)ksp)->options, ((PetscObject)ksp)->prefix, "-ksp_test_null_space", &test, NULL);
407: if (test) MatNullSpaceTest(nullsp, mat, NULL);
408: }
409: ksp->setupstage = KSP_SETUP_NEWRHS;
410: level--;
411: return 0;
412: }
414: /*@C
415: KSPConvergedReasonView - Displays the reason a `KSP` solve converged or diverged to a viewer
417: Collective
419: Parameter:
420: + ksp - iterative context obtained from `KSPCreate()`
421: - viewer - the viewer to display the reason
423: Options Database Keys:
424: + -ksp_converged_reason - print reason for converged or diverged, also prints number of iterations
425: - -ksp_converged_reason ::failed - only print reason and number of iterations when diverged
427: Notes:
428: To change the format of the output call PetscViewerPushFormat(viewer,format) before this call. Use PETSC_VIEWER_DEFAULT for the default,
429: use PETSC_VIEWER_FAILED to only display a reason if it fails.
431: Level: beginner
433: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSetUp()`, `KSPDestroy()`, `KSPSetTolerances()`, `KSPConvergedDefault()`,
434: `KSPSolveTranspose()`, `KSPGetIterationNumber()`, `KSP`, `KSPGetConvergedReason()`, `PetscViewerPushFormat()`, `PetscViewerPopFormat()`
435: @*/
436: PetscErrorCode KSPConvergedReasonView(KSP ksp, PetscViewer viewer)
437: {
438: PetscBool isAscii;
439: PetscViewerFormat format;
441: if (!viewer) viewer = PETSC_VIEWER_STDOUT_(PetscObjectComm((PetscObject)ksp));
442: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isAscii);
443: if (isAscii) {
444: PetscViewerGetFormat(viewer, &format);
445: PetscViewerASCIIAddTab(viewer, ((PetscObject)ksp)->tablevel);
446: if (ksp->reason > 0 && format != PETSC_VIEWER_FAILED) {
447: if (((PetscObject)ksp)->prefix) {
448: PetscViewerASCIIPrintf(viewer, "Linear %s solve converged due to %s iterations %" PetscInt_FMT "\n", ((PetscObject)ksp)->prefix, KSPConvergedReasons[ksp->reason], ksp->its);
449: } else {
450: PetscViewerASCIIPrintf(viewer, "Linear solve converged due to %s iterations %" PetscInt_FMT "\n", KSPConvergedReasons[ksp->reason], ksp->its);
451: }
452: } else if (ksp->reason <= 0) {
453: if (((PetscObject)ksp)->prefix) {
454: PetscViewerASCIIPrintf(viewer, "Linear %s solve did not converge due to %s iterations %" PetscInt_FMT "\n", ((PetscObject)ksp)->prefix, KSPConvergedReasons[ksp->reason], ksp->its);
455: } else {
456: PetscViewerASCIIPrintf(viewer, "Linear solve did not converge due to %s iterations %" PetscInt_FMT "\n", KSPConvergedReasons[ksp->reason], ksp->its);
457: }
458: if (ksp->reason == KSP_DIVERGED_PC_FAILED) {
459: PCFailedReason reason;
460: PCGetFailedReason(ksp->pc, &reason);
461: PetscViewerASCIIPrintf(viewer, " PC failed due to %s \n", PCFailedReasons[reason]);
462: }
463: }
464: PetscViewerASCIISubtractTab(viewer, ((PetscObject)ksp)->tablevel);
465: }
466: return 0;
467: }
469: /*@C
470: KSPConvergedReasonViewSet - Sets an ADDITIONAL function that is to be used at the
471: end of the linear solver to display the convergence reason of the linear solver.
473: Logically Collective
475: Input Parameters:
476: + ksp - the `KSP` context
477: . f - the ksp converged reason view function
478: . vctx - [optional] user-defined context for private data for the
479: ksp converged reason view routine (use NULL if no context is desired)
480: - reasonviewdestroy - [optional] routine that frees reasonview context
481: (may be NULL)
483: Options Database Keys:
484: + -ksp_converged_reason - sets a default `KSPConvergedReasonView()`
485: - -ksp_converged_reason_view_cancel - cancels all converged reason viewers that have
486: been hardwired into a code by
487: calls to `KSPConvergedReasonViewSet()`, but
488: does not cancel those set via
489: the options database.
491: Notes:
492: Several different converged reason view routines may be set by calling
493: `KSPConvergedReasonViewSet()` multiple times; all will be called in the
494: order in which they were set.
496: Level: intermediate
498: .seealso: [](chapter_ksp), `KSPConvergedReasonView()`, `KSPConvergedReasonViewCancel()`
499: @*/
500: PetscErrorCode KSPConvergedReasonViewSet(KSP ksp, PetscErrorCode (*f)(KSP, void *), void *vctx, PetscErrorCode (*reasonviewdestroy)(void **))
501: {
502: PetscInt i;
503: PetscBool identical;
506: for (i = 0; i < ksp->numberreasonviews; i++) {
507: PetscMonitorCompare((PetscErrorCode(*)(void))f, vctx, reasonviewdestroy, (PetscErrorCode(*)(void))ksp->reasonview[i], ksp->reasonviewcontext[i], ksp->reasonviewdestroy[i], &identical);
508: if (identical) return 0;
509: }
511: ksp->reasonview[ksp->numberreasonviews] = f;
512: ksp->reasonviewdestroy[ksp->numberreasonviews] = reasonviewdestroy;
513: ksp->reasonviewcontext[ksp->numberreasonviews++] = (void *)vctx;
514: return 0;
515: }
517: /*@
518: KSPConvergedReasonViewCancel - Clears all the reasonview functions for a `KSP` object.
520: Collective
522: Input Parameter:
523: . ksp - iterative context obtained from `KSPCreate()`
525: Level: intermediate
527: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPDestroy()`, `KSPReset()`
528: @*/
529: PetscErrorCode KSPConvergedReasonViewCancel(KSP ksp)
530: {
531: PetscInt i;
534: for (i = 0; i < ksp->numberreasonviews; i++) {
535: if (ksp->reasonviewdestroy[i]) (*ksp->reasonviewdestroy[i])(&ksp->reasonviewcontext[i]);
536: }
537: ksp->numberreasonviews = 0;
538: return 0;
539: }
541: /*@
542: KSPConvergedReasonViewFromOptions - Processes command line options to determine if/how a KSPReason is to be viewed.
544: Collective
546: Input Parameters:
547: . ksp - the `KSP` object
549: Level: intermediate
551: .seealso: [](chapter_ksp), `KSPConvergedReasonView()`
552: @*/
553: PetscErrorCode KSPConvergedReasonViewFromOptions(KSP ksp)
554: {
555: PetscViewer viewer;
556: PetscBool flg;
557: PetscViewerFormat format;
558: PetscInt i;
561: /* Call all user-provided reason review routines */
562: for (i = 0; i < ksp->numberreasonviews; i++) (*ksp->reasonview[i])(ksp, ksp->reasonviewcontext[i]);
564: /* Call the default PETSc routine */
565: PetscOptionsGetViewer(PetscObjectComm((PetscObject)ksp), ((PetscObject)ksp)->options, ((PetscObject)ksp)->prefix, "-ksp_converged_reason", &viewer, &format, &flg);
566: if (flg) {
567: PetscViewerPushFormat(viewer, format);
568: KSPConvergedReasonView(ksp, viewer);
569: PetscViewerPopFormat(viewer);
570: PetscViewerDestroy(&viewer);
571: }
572: return 0;
573: }
575: /*@C
576: KSPConvergedRateView - Displays the reason a `KSP` solve converged or diverged to a viewer
578: Collective
580: Input Parameters:
581: + ksp - iterative context obtained from `KSPCreate()`
582: - viewer - the viewer to display the reason
584: Options Database Keys:
585: . -ksp_converged_rate - print reason for convergence or divergence and the convergence rate (or 0.0 for divergence)
587: Notes:
588: To change the format of the output, call PetscViewerPushFormat(viewer,format) before this call.
590: Suppose that the residual is reduced linearly, $r_k = c^k r_0$, which means $log r_k = log r_0 + k log c$. After linear regression,
591: the slope is $\log c$. The coefficient of determination is given by $1 - \frac{\sum_i (y_i - f(x_i))^2}{\sum_i (y_i - \bar y)}$,
592: see also https://en.wikipedia.org/wiki/Coefficient_of_determination
594: Level: intermediate
596: .seealso: [](chapter_ksp), `KSPConvergedReasonView()`, `KSPGetConvergedRate()`, `KSPSetTolerances()`, `KSPConvergedDefault()`
597: @*/
598: PetscErrorCode KSPConvergedRateView(KSP ksp, PetscViewer viewer)
599: {
600: PetscViewerFormat format;
601: PetscBool isAscii;
602: PetscReal rrate, rRsq, erate = 0.0, eRsq = 0.0;
603: PetscInt its;
604: const char *prefix, *reason = KSPConvergedReasons[ksp->reason];
606: KSPGetOptionsPrefix(ksp, &prefix);
607: KSPGetIterationNumber(ksp, &its);
608: KSPComputeConvergenceRate(ksp, &rrate, &rRsq, &erate, &eRsq);
609: if (!viewer) viewer = PETSC_VIEWER_STDOUT_(PetscObjectComm((PetscObject)ksp));
610: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isAscii);
611: if (isAscii) {
612: PetscViewerGetFormat(viewer, &format);
613: PetscViewerASCIIAddTab(viewer, ((PetscObject)ksp)->tablevel);
614: if (ksp->reason > 0) {
615: if (prefix) PetscViewerASCIIPrintf(viewer, "Linear %s solve converged due to %s iterations %" PetscInt_FMT, prefix, reason, its);
616: else PetscViewerASCIIPrintf(viewer, "Linear solve converged due to %s iterations %" PetscInt_FMT, reason, its);
617: PetscViewerASCIIUseTabs(viewer, PETSC_FALSE);
618: if (rRsq >= 0.0) PetscViewerASCIIPrintf(viewer, " res rate %g R^2 %g", (double)rrate, (double)rRsq);
619: if (eRsq >= 0.0) PetscViewerASCIIPrintf(viewer, " error rate %g R^2 %g", (double)erate, (double)eRsq);
620: PetscViewerASCIIPrintf(viewer, "\n");
621: PetscViewerASCIIUseTabs(viewer, PETSC_TRUE);
622: } else if (ksp->reason <= 0) {
623: if (prefix) PetscViewerASCIIPrintf(viewer, "Linear %s solve did not converge due to %s iterations %" PetscInt_FMT, prefix, reason, its);
624: else PetscViewerASCIIPrintf(viewer, "Linear solve did not converge due to %s iterations %" PetscInt_FMT, reason, its);
625: PetscViewerASCIIUseTabs(viewer, PETSC_FALSE);
626: if (rRsq >= 0.0) PetscViewerASCIIPrintf(viewer, " res rate %g R^2 %g", (double)rrate, (double)rRsq);
627: if (eRsq >= 0.0) PetscViewerASCIIPrintf(viewer, " error rate %g R^2 %g", (double)erate, (double)eRsq);
628: PetscViewerASCIIPrintf(viewer, "\n");
629: PetscViewerASCIIUseTabs(viewer, PETSC_TRUE);
630: if (ksp->reason == KSP_DIVERGED_PC_FAILED) {
631: PCFailedReason reason;
632: PCGetFailedReason(ksp->pc, &reason);
633: PetscViewerASCIIPrintf(viewer, " PC failed due to %s \n", PCFailedReasons[reason]);
634: }
635: }
636: PetscViewerASCIISubtractTab(viewer, ((PetscObject)ksp)->tablevel);
637: }
638: return 0;
639: }
641: #include <petscdraw.h>
643: static PetscErrorCode KSPViewEigenvalues_Internal(KSP ksp, PetscBool isExplicit, PetscViewer viewer, PetscViewerFormat format)
644: {
645: PetscReal *r, *c;
646: PetscInt n, i, neig;
647: PetscBool isascii, isdraw;
648: PetscMPIInt rank;
650: MPI_Comm_rank(PetscObjectComm((PetscObject)ksp), &rank);
651: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii);
652: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERDRAW, &isdraw);
653: if (isExplicit) {
654: VecGetSize(ksp->vec_sol, &n);
655: PetscMalloc2(n, &r, n, &c);
656: KSPComputeEigenvaluesExplicitly(ksp, n, r, c);
657: neig = n;
658: } else {
659: PetscInt nits;
661: KSPGetIterationNumber(ksp, &nits);
662: n = nits + 2;
663: if (!nits) {
664: PetscViewerASCIIPrintf(viewer, "Zero iterations in solver, cannot approximate any eigenvalues\n");
665: return 0;
666: }
667: PetscMalloc2(n, &r, n, &c);
668: KSPComputeEigenvalues(ksp, n, r, c, &neig);
669: }
670: if (isascii) {
671: PetscViewerASCIIPrintf(viewer, "%s computed eigenvalues\n", isExplicit ? "Explicitly" : "Iteratively");
672: for (i = 0; i < neig; ++i) {
673: if (c[i] >= 0.0) PetscViewerASCIIPrintf(viewer, "%g + %gi\n", (double)r[i], (double)c[i]);
674: else PetscViewerASCIIPrintf(viewer, "%g - %gi\n", (double)r[i], -(double)c[i]);
675: }
676: } else if (isdraw && rank == 0) {
677: PetscDraw draw;
678: PetscDrawSP drawsp;
680: if (format == PETSC_VIEWER_DRAW_CONTOUR) {
681: KSPPlotEigenContours_Private(ksp, neig, r, c);
682: } else {
683: if (!ksp->eigviewer) PetscViewerDrawOpen(PETSC_COMM_SELF, NULL, isExplicit ? "Explicitly Computed Eigenvalues" : "Iteratively Computed Eigenvalues", PETSC_DECIDE, PETSC_DECIDE, 400, 400, &ksp->eigviewer);
684: PetscViewerDrawGetDraw(ksp->eigviewer, 0, &draw);
685: PetscDrawSPCreate(draw, 1, &drawsp);
686: PetscDrawSPReset(drawsp);
687: for (i = 0; i < neig; ++i) PetscDrawSPAddPoint(drawsp, r + i, c + i);
688: PetscDrawSPDraw(drawsp, PETSC_TRUE);
689: PetscDrawSPSave(drawsp);
690: PetscDrawSPDestroy(&drawsp);
691: }
692: }
693: PetscFree2(r, c);
694: return 0;
695: }
697: static PetscErrorCode KSPViewSingularvalues_Internal(KSP ksp, PetscViewer viewer, PetscViewerFormat format)
698: {
699: PetscReal smax, smin;
700: PetscInt nits;
701: PetscBool isascii;
703: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii);
704: KSPGetIterationNumber(ksp, &nits);
705: if (!nits) {
706: PetscViewerASCIIPrintf(viewer, "Zero iterations in solver, cannot approximate any singular values\n");
707: return 0;
708: }
709: KSPComputeExtremeSingularValues(ksp, &smax, &smin);
710: if (isascii) PetscViewerASCIIPrintf(viewer, "Iteratively computed extreme singular values: max %g min %g max/min %g\n", (double)smax, (double)smin, (double)(smax / smin));
711: return 0;
712: }
714: static PetscErrorCode KSPViewFinalResidual_Internal(KSP ksp, PetscViewer viewer, PetscViewerFormat format)
715: {
716: PetscBool isascii;
718: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii);
720: if (isascii) {
721: Mat A;
722: Vec t;
723: PetscReal norm;
725: PCGetOperators(ksp->pc, &A, NULL);
726: VecDuplicate(ksp->vec_rhs, &t);
727: KSP_MatMult(ksp, A, ksp->vec_sol, t);
728: VecAYPX(t, -1.0, ksp->vec_rhs);
729: VecNorm(t, NORM_2, &norm);
730: VecDestroy(&t);
731: PetscViewerASCIIPrintf(viewer, "KSP final norm of residual %g\n", (double)norm);
732: }
733: return 0;
734: }
736: static PetscErrorCode KSPMonitorPauseFinal_Internal(KSP ksp)
737: {
738: PetscInt i;
740: if (!ksp->pauseFinal) return 0;
741: for (i = 0; i < ksp->numbermonitors; ++i) {
742: PetscViewerAndFormat *vf = (PetscViewerAndFormat *)ksp->monitorcontext[i];
743: PetscDraw draw;
744: PetscReal lpause;
746: if (!vf) continue;
747: if (vf->lg) {
749: if (((PetscObject)vf->lg)->classid != PETSC_DRAWLG_CLASSID) continue;
750: PetscDrawLGGetDraw(vf->lg, &draw);
751: PetscDrawGetPause(draw, &lpause);
752: PetscDrawSetPause(draw, -1.0);
753: PetscDrawPause(draw);
754: PetscDrawSetPause(draw, lpause);
755: } else {
756: PetscBool isdraw;
759: if (((PetscObject)vf->viewer)->classid != PETSC_VIEWER_CLASSID) continue;
760: PetscObjectTypeCompare((PetscObject)vf->viewer, PETSCVIEWERDRAW, &isdraw);
761: if (!isdraw) continue;
762: PetscViewerDrawGetDraw(vf->viewer, 0, &draw);
763: PetscDrawGetPause(draw, &lpause);
764: PetscDrawSetPause(draw, -1.0);
765: PetscDrawPause(draw);
766: PetscDrawSetPause(draw, lpause);
767: }
768: }
769: return 0;
770: }
772: static PetscErrorCode KSPSolve_Private(KSP ksp, Vec b, Vec x)
773: {
774: PetscBool flg = PETSC_FALSE, inXisinB = PETSC_FALSE, guess_zero;
775: Mat mat, pmat;
776: MPI_Comm comm;
777: MatNullSpace nullsp;
778: Vec btmp, vec_rhs = NULL;
780: level++;
781: comm = PetscObjectComm((PetscObject)ksp);
782: if (x && x == b) {
784: VecDuplicate(b, &x);
785: inXisinB = PETSC_TRUE;
786: }
787: if (b) {
788: PetscObjectReference((PetscObject)b);
789: VecDestroy(&ksp->vec_rhs);
790: ksp->vec_rhs = b;
791: }
792: if (x) {
793: PetscObjectReference((PetscObject)x);
794: VecDestroy(&ksp->vec_sol);
795: ksp->vec_sol = x;
796: }
798: if (ksp->viewPre) ObjectView((PetscObject)ksp, ksp->viewerPre, ksp->formatPre);
800: if (ksp->presolve) (*ksp->presolve)(ksp, ksp->vec_rhs, ksp->vec_sol, ksp->prectx);
802: /* reset the residual history list if requested */
803: if (ksp->res_hist_reset) ksp->res_hist_len = 0;
804: if (ksp->err_hist_reset) ksp->err_hist_len = 0;
806: /* KSPSetUp() scales the matrix if needed */
807: KSPSetUp(ksp);
808: KSPSetUpOnBlocks(ksp);
810: if (ksp->guess) {
811: PetscObjectState ostate, state;
813: KSPGuessSetUp(ksp->guess);
814: PetscObjectStateGet((PetscObject)ksp->vec_sol, &ostate);
815: KSPGuessFormGuess(ksp->guess, ksp->vec_rhs, ksp->vec_sol);
816: PetscObjectStateGet((PetscObject)ksp->vec_sol, &state);
817: if (state != ostate) {
818: ksp->guess_zero = PETSC_FALSE;
819: } else {
820: PetscInfo(ksp, "Using zero initial guess since the KSPGuess object did not change the vector\n");
821: ksp->guess_zero = PETSC_TRUE;
822: }
823: }
825: VecSetErrorIfLocked(ksp->vec_sol, 3);
827: PetscLogEventBegin(KSP_Solve, ksp, ksp->vec_rhs, ksp->vec_sol, 0);
828: PCGetOperators(ksp->pc, &mat, &pmat);
829: /* diagonal scale RHS if called for */
830: if (ksp->dscale) {
831: VecPointwiseMult(ksp->vec_rhs, ksp->vec_rhs, ksp->diagonal);
832: /* second time in, but matrix was scaled back to original */
833: if (ksp->dscalefix && ksp->dscalefix2) {
834: Mat mat, pmat;
836: PCGetOperators(ksp->pc, &mat, &pmat);
837: MatDiagonalScale(pmat, ksp->diagonal, ksp->diagonal);
838: if (mat != pmat) MatDiagonalScale(mat, ksp->diagonal, ksp->diagonal);
839: }
841: /* scale initial guess */
842: if (!ksp->guess_zero) {
843: if (!ksp->truediagonal) {
844: VecDuplicate(ksp->diagonal, &ksp->truediagonal);
845: VecCopy(ksp->diagonal, ksp->truediagonal);
846: VecReciprocal(ksp->truediagonal);
847: }
848: VecPointwiseMult(ksp->vec_sol, ksp->vec_sol, ksp->truediagonal);
849: }
850: }
851: PCPreSolve(ksp->pc, ksp);
853: if (ksp->guess_zero) VecSet(ksp->vec_sol, 0.0);
854: if (ksp->guess_knoll) { /* The Knoll trick is independent on the KSPGuess specified */
855: PCApply(ksp->pc, ksp->vec_rhs, ksp->vec_sol);
856: KSP_RemoveNullSpace(ksp, ksp->vec_sol);
857: ksp->guess_zero = PETSC_FALSE;
858: }
860: /* can we mark the initial guess as zero for this solve? */
861: guess_zero = ksp->guess_zero;
862: if (!ksp->guess_zero) {
863: PetscReal norm;
865: VecNormAvailable(ksp->vec_sol, NORM_2, &flg, &norm);
866: if (flg && !norm) ksp->guess_zero = PETSC_TRUE;
867: }
868: if (ksp->transpose_solve) {
869: MatGetNullSpace(pmat, &nullsp);
870: } else {
871: MatGetTransposeNullSpace(pmat, &nullsp);
872: }
873: if (nullsp) {
874: VecDuplicate(ksp->vec_rhs, &btmp);
875: VecCopy(ksp->vec_rhs, btmp);
876: MatNullSpaceRemove(nullsp, btmp);
877: vec_rhs = ksp->vec_rhs;
878: ksp->vec_rhs = btmp;
879: }
880: VecLockReadPush(ksp->vec_rhs);
881: PetscUseTypeMethod(ksp, solve);
882: KSPMonitorPauseFinal_Internal(ksp);
884: VecLockReadPop(ksp->vec_rhs);
885: if (nullsp) {
886: ksp->vec_rhs = vec_rhs;
887: VecDestroy(&btmp);
888: }
890: ksp->guess_zero = guess_zero;
893: ksp->totalits += ksp->its;
895: KSPConvergedReasonViewFromOptions(ksp);
897: if (ksp->viewRate) {
898: PetscViewerPushFormat(ksp->viewerRate, ksp->formatRate);
899: KSPConvergedRateView(ksp, ksp->viewerRate);
900: PetscViewerPopFormat(ksp->viewerRate);
901: }
902: PCPostSolve(ksp->pc, ksp);
904: /* diagonal scale solution if called for */
905: if (ksp->dscale) {
906: VecPointwiseMult(ksp->vec_sol, ksp->vec_sol, ksp->diagonal);
907: /* unscale right hand side and matrix */
908: if (ksp->dscalefix) {
909: Mat mat, pmat;
911: VecReciprocal(ksp->diagonal);
912: VecPointwiseMult(ksp->vec_rhs, ksp->vec_rhs, ksp->diagonal);
913: PCGetOperators(ksp->pc, &mat, &pmat);
914: MatDiagonalScale(pmat, ksp->diagonal, ksp->diagonal);
915: if (mat != pmat) MatDiagonalScale(mat, ksp->diagonal, ksp->diagonal);
916: VecReciprocal(ksp->diagonal);
917: ksp->dscalefix2 = PETSC_TRUE;
918: }
919: }
920: PetscLogEventEnd(KSP_Solve, ksp, ksp->vec_rhs, ksp->vec_sol, 0);
921: if (ksp->guess) KSPGuessUpdate(ksp->guess, ksp->vec_rhs, ksp->vec_sol);
922: if (ksp->postsolve) (*ksp->postsolve)(ksp, ksp->vec_rhs, ksp->vec_sol, ksp->postctx);
924: PCGetOperators(ksp->pc, &mat, &pmat);
925: if (ksp->viewEV) KSPViewEigenvalues_Internal(ksp, PETSC_FALSE, ksp->viewerEV, ksp->formatEV);
926: if (ksp->viewEVExp) KSPViewEigenvalues_Internal(ksp, PETSC_TRUE, ksp->viewerEVExp, ksp->formatEVExp);
927: if (ksp->viewSV) KSPViewSingularvalues_Internal(ksp, ksp->viewerSV, ksp->formatSV);
928: if (ksp->viewFinalRes) KSPViewFinalResidual_Internal(ksp, ksp->viewerFinalRes, ksp->formatFinalRes);
929: if (ksp->viewMat) ObjectView((PetscObject)mat, ksp->viewerMat, ksp->formatMat);
930: if (ksp->viewPMat) ObjectView((PetscObject)pmat, ksp->viewerPMat, ksp->formatPMat);
931: if (ksp->viewRhs) ObjectView((PetscObject)ksp->vec_rhs, ksp->viewerRhs, ksp->formatRhs);
932: if (ksp->viewSol) ObjectView((PetscObject)ksp->vec_sol, ksp->viewerSol, ksp->formatSol);
933: if (ksp->view) ObjectView((PetscObject)ksp, ksp->viewer, ksp->format);
934: if (ksp->viewDScale) ObjectView((PetscObject)ksp->diagonal, ksp->viewerDScale, ksp->formatDScale);
935: if (ksp->viewMatExp) {
936: Mat A, B;
938: PCGetOperators(ksp->pc, &A, NULL);
939: if (ksp->transpose_solve) {
940: Mat AT;
942: MatCreateTranspose(A, &AT);
943: MatComputeOperator(AT, MATAIJ, &B);
944: MatDestroy(&AT);
945: } else {
946: MatComputeOperator(A, MATAIJ, &B);
947: }
948: ObjectView((PetscObject)B, ksp->viewerMatExp, ksp->formatMatExp);
949: MatDestroy(&B);
950: }
951: if (ksp->viewPOpExp) {
952: Mat B;
954: KSPComputeOperator(ksp, MATAIJ, &B);
955: ObjectView((PetscObject)B, ksp->viewerPOpExp, ksp->formatPOpExp);
956: MatDestroy(&B);
957: }
959: if (inXisinB) {
960: VecCopy(x, b);
961: VecDestroy(&x);
962: }
963: PetscObjectSAWsBlock((PetscObject)ksp);
964: if (ksp->errorifnotconverged && ksp->reason < 0 && ((level == 1) || (ksp->reason != KSP_DIVERGED_ITS))) {
965: if (ksp->reason == KSP_DIVERGED_PC_FAILED) {
966: PCFailedReason reason;
967: PCGetFailedReason(ksp->pc, &reason);
968: SETERRQ(comm, PETSC_ERR_NOT_CONVERGED, "KSPSolve has not converged, reason %s PC failed due to %s", KSPConvergedReasons[ksp->reason], PCFailedReasons[reason]);
969: } else SETERRQ(comm, PETSC_ERR_NOT_CONVERGED, "KSPSolve has not converged, reason %s", KSPConvergedReasons[ksp->reason]);
970: }
971: level--;
972: return 0;
973: }
975: /*@
976: KSPSolve - Solves linear system.
978: Collective
980: Parameters:
981: + ksp - iterative context obtained from `KSPCreate()`
982: . b - the right hand side vector
983: - x - the solution (this may be the same vector as b, then b will be overwritten with answer)
985: Options Database Keys:
986: + -ksp_view_eigenvalues - compute preconditioned operators eigenvalues
987: . -ksp_view_eigenvalues_explicit - compute the eigenvalues by forming the dense operator and using LAPACK
988: . -ksp_view_mat binary - save matrix to the default binary viewer
989: . -ksp_view_pmat binary - save matrix used to build preconditioner to the default binary viewer
990: . -ksp_view_rhs binary - save right hand side vector to the default binary viewer
991: . -ksp_view_solution binary - save computed solution vector to the default binary viewer
992: (can be read later with src/ksp/tutorials/ex10.c for testing solvers)
993: . -ksp_view_mat_explicit - for matrix-free operators, computes the matrix entries and views them
994: . -ksp_view_preconditioned_operator_explicit - computes the product of the preconditioner and matrix as an explicit matrix and views it
995: . -ksp_converged_reason - print reason for converged or diverged, also prints number of iterations
996: . -ksp_view_final_residual - print 2-norm of true linear system residual at the end of the solution process
997: . -ksp_error_if_not_converged - stop the program as soon as an error is detected in a `KSPSolve()`
998: - -ksp_view - print the ksp data structure at the end of the system solution
1000: Notes:
1002: If one uses `KSPSetDM()` then x or b need not be passed. Use `KSPGetSolution()` to access the solution in this case.
1004: The operator is specified with `KSPSetOperators()`.
1006: `KSPSolve()` will normally return without generating an error regardless of whether the linear system was solved or if constructing the preconditioner failed.
1007: Call `KSPGetConvergedReason()` to determine if the solver converged or failed and why. The option -ksp_error_if_not_converged or function `KSPSetErrorIfNotConverged()`
1008: will cause `KSPSolve()` to error as soon as an error occurs in the linear solver. In inner KSPSolves() KSP_DIVERGED_ITS is not treated as an error because when using nested solvers
1009: it may be fine that inner solvers in the preconditioner do not converge during the solution process.
1011: The number of iterations can be obtained from `KSPGetIterationNumber()`.
1013: If you provide a matrix that has a `MatSetNullSpace()` and `MatSetTransposeNullSpace()` this will use that information to solve singular systems
1014: in the least squares sense with a norm minimizing solution.
1016: A x = b where b = b_p + b_t where b_t is not in the range of A (and hence by the fundamental theorem of linear algebra is in the nullspace(A') see `MatSetNullSpace()`
1018: `KSP` first removes b_t producing the linear system A x = b_p (which has multiple solutions) and solves this to find the ||x|| minimizing solution (and hence
1019: it finds the solution x orthogonal to the nullspace(A). The algorithm is simply in each iteration of the Krylov method we remove the nullspace(A) from the search
1020: direction thus the solution which is a linear combination of the search directions has no component in the nullspace(A).
1022: We recommend always using `KSPGMRES` for such singular systems.
1023: If nullspace(A) = nullspace(A') (note symmetric matrices always satisfy this property) then both left and right preconditioning will work
1024: If nullspace(A) != nullspace(A') then left preconditioning will work but right preconditioning may not work (or it may).
1026: Developer Note: The reason we cannot always solve nullspace(A) != nullspace(A') systems with right preconditioning is because we need to remove at each iteration
1027: the nullspace(AB) from the search direction. While we know the nullspace(A) the nullspace(AB) equals B^-1 times the nullspace(A) but except for trivial preconditioners
1028: such as diagonal scaling we cannot apply the inverse of the preconditioner to a vector and thus cannot compute the nullspace(AB).
1030: If using a direct method (e.g., via the `KSP` solver
1031: `KSPPREONLY` and a preconditioner such as `PCLU` or `PCILU`,
1032: then its=1. See `KSPSetTolerances()` and `KSPConvergedDefault()`
1033: for more details.
1035: Understanding Convergence:
1036: The routines `KSPMonitorSet()`, `KSPComputeEigenvalues()`, and
1037: `KSPComputeEigenvaluesExplicitly()` provide information on additional
1038: options to monitor convergence and print eigenvalue information.
1040: Level: beginner
1042: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSetUp()`, `KSPDestroy()`, `KSPSetTolerances()`, `KSPConvergedDefault()`,
1043: `KSPSolveTranspose()`, `KSPGetIterationNumber()`, `MatNullSpaceCreate()`, `MatSetNullSpace()`, `MatSetTransposeNullSpace()`, `KSP`,
1044: `KSPConvergedReasonView()`, `KSPCheckSolve()`, `KSPSetErrorIfNotConverged()`
1045: @*/
1046: PetscErrorCode KSPSolve(KSP ksp, Vec b, Vec x)
1047: {
1051: ksp->transpose_solve = PETSC_FALSE;
1052: KSPSolve_Private(ksp, b, x);
1053: return 0;
1054: }
1056: /*@
1057: KSPSolveTranspose - Solves the transpose of a linear system.
1059: Collective
1061: Input Parameters:
1062: + ksp - iterative context obtained from `KSPCreate()`
1063: . b - right hand side vector
1064: - x - solution vector
1066: Notes:
1067: For complex numbers this solve the non-Hermitian transpose system.
1069: Developer Notes:
1070: We need to implement a `KSPSolveHermitianTranspose()`
1072: Level: developer
1074: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSetUp()`, `KSPDestroy()`, `KSPSetTolerances()`, `KSPConvergedDefault()`,
1075: `KSPSolve()`, `KSP`
1076: @*/
1077: PetscErrorCode KSPSolveTranspose(KSP ksp, Vec b, Vec x)
1078: {
1082: if (ksp->transpose.use_explicittranspose) {
1083: Mat J, Jpre;
1084: KSPGetOperators(ksp, &J, &Jpre);
1085: if (!ksp->transpose.reuse_transpose) {
1086: MatTranspose(J, MAT_INITIAL_MATRIX, &ksp->transpose.AT);
1087: if (J != Jpre) MatTranspose(Jpre, MAT_INITIAL_MATRIX, &ksp->transpose.BT);
1088: ksp->transpose.reuse_transpose = PETSC_TRUE;
1089: } else {
1090: MatTranspose(J, MAT_REUSE_MATRIX, &ksp->transpose.AT);
1091: if (J != Jpre) MatTranspose(Jpre, MAT_REUSE_MATRIX, &ksp->transpose.BT);
1092: }
1093: if (J == Jpre && ksp->transpose.BT != ksp->transpose.AT) {
1094: PetscObjectReference((PetscObject)ksp->transpose.AT);
1095: ksp->transpose.BT = ksp->transpose.AT;
1096: }
1097: KSPSetOperators(ksp, ksp->transpose.AT, ksp->transpose.BT);
1098: } else {
1099: ksp->transpose_solve = PETSC_TRUE;
1100: }
1101: KSPSolve_Private(ksp, b, x);
1102: return 0;
1103: }
1105: static PetscErrorCode KSPViewFinalMatResidual_Internal(KSP ksp, Mat B, Mat X, PetscViewer viewer, PetscViewerFormat format, PetscInt shift)
1106: {
1107: Mat A, R;
1108: PetscReal *norms;
1109: PetscInt i, N;
1110: PetscBool flg;
1112: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &flg);
1113: if (flg) {
1114: PCGetOperators(ksp->pc, &A, NULL);
1115: MatMatMult(A, X, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &R);
1116: MatAYPX(R, -1.0, B, SAME_NONZERO_PATTERN);
1117: MatGetSize(R, NULL, &N);
1118: PetscMalloc1(N, &norms);
1119: MatGetColumnNorms(R, NORM_2, norms);
1120: MatDestroy(&R);
1121: for (i = 0; i < N; ++i) PetscViewerASCIIPrintf(viewer, "%s #%" PetscInt_FMT " %g\n", i == 0 ? "KSP final norm of residual" : " ", shift + i, (double)norms[i]);
1122: PetscFree(norms);
1123: }
1124: return 0;
1125: }
1127: /*@
1128: KSPMatSolve - Solves a linear system with multiple right-hand sides stored as a MATDENSE. Unlike `KSPSolve()`, B and X must be different matrices.
1130: Input Parameters:
1131: + ksp - iterative context
1132: - B - block of right-hand sides
1134: Output Parameter:
1135: . X - block of solutions
1137: Notes:
1138: This is a stripped-down version of `KSPSolve()`, which only handles -ksp_view, -ksp_converged_reason, and -ksp_view_final_residual.
1140: Level: intermediate
1142: .seealso: [](chapter_ksp), `KSPSolve()`, `MatMatSolve()`, `MATDENSE`, `KSPHPDDM`, `PCBJACOBI`, `PCASM`
1143: @*/
1144: PetscErrorCode KSPMatSolve(KSP ksp, Mat B, Mat X)
1145: {
1146: Mat A, P, vB, vX;
1147: Vec cb, cx;
1148: PetscInt n1, N1, n2, N2, Bbn = PETSC_DECIDE;
1149: PetscBool match;
1157: MatCheckPreallocated(X, 3);
1158: if (!X->assembled) {
1159: MatSetOption(X, MAT_NO_OFF_PROC_ENTRIES, PETSC_TRUE);
1160: MatAssemblyBegin(X, MAT_FINAL_ASSEMBLY);
1161: MatAssemblyEnd(X, MAT_FINAL_ASSEMBLY);
1162: }
1164: KSPGetOperators(ksp, &A, &P);
1165: MatGetLocalSize(B, NULL, &n2);
1166: MatGetLocalSize(X, NULL, &n1);
1167: MatGetSize(B, NULL, &N2);
1168: MatGetSize(X, NULL, &N1);
1170: PetscObjectBaseTypeCompareAny((PetscObject)B, &match, MATSEQDENSE, MATMPIDENSE, "");
1172: PetscObjectBaseTypeCompareAny((PetscObject)X, &match, MATSEQDENSE, MATMPIDENSE, "");
1174: KSPSetUp(ksp);
1175: KSPSetUpOnBlocks(ksp);
1176: if (ksp->ops->matsolve) {
1177: if (ksp->guess_zero) MatZeroEntries(X);
1178: PetscLogEventBegin(KSP_MatSolve, ksp, B, X, 0);
1179: KSPGetMatSolveBatchSize(ksp, &Bbn);
1180: /* by default, do a single solve with all columns */
1181: if (Bbn == PETSC_DECIDE) Bbn = N2;
1183: PetscInfo(ksp, "KSP type %s solving using batches of width at most %" PetscInt_FMT "\n", ((PetscObject)ksp)->type_name, Bbn);
1184: /* if -ksp_matsolve_batch_size is greater than the actual number of columns, do a single solve with all columns */
1185: if (Bbn >= N2) {
1186: PetscUseTypeMethod(ksp, matsolve, B, X);
1187: if (ksp->viewFinalRes) KSPViewFinalMatResidual_Internal(ksp, B, X, ksp->viewerFinalRes, ksp->formatFinalRes, 0);
1189: KSPConvergedReasonViewFromOptions(ksp);
1191: if (ksp->viewRate) {
1192: PetscViewerPushFormat(ksp->viewerRate, PETSC_VIEWER_DEFAULT);
1193: KSPConvergedRateView(ksp, ksp->viewerRate);
1194: PetscViewerPopFormat(ksp->viewerRate);
1195: }
1196: } else {
1197: for (n2 = 0; n2 < N2; n2 += Bbn) {
1198: MatDenseGetSubMatrix(B, PETSC_DECIDE, PETSC_DECIDE, n2, PetscMin(n2 + Bbn, N2), &vB);
1199: MatDenseGetSubMatrix(X, PETSC_DECIDE, PETSC_DECIDE, n2, PetscMin(n2 + Bbn, N2), &vX);
1200: PetscUseTypeMethod(ksp, matsolve, vB, vX);
1201: if (ksp->viewFinalRes) KSPViewFinalMatResidual_Internal(ksp, vB, vX, ksp->viewerFinalRes, ksp->formatFinalRes, n2);
1203: KSPConvergedReasonViewFromOptions(ksp);
1205: if (ksp->viewRate) {
1206: PetscViewerPushFormat(ksp->viewerRate, PETSC_VIEWER_DEFAULT);
1207: KSPConvergedRateView(ksp, ksp->viewerRate);
1208: PetscViewerPopFormat(ksp->viewerRate);
1209: }
1210: MatDenseRestoreSubMatrix(B, &vB);
1211: MatDenseRestoreSubMatrix(X, &vX);
1212: }
1213: }
1214: if (ksp->viewMat) ObjectView((PetscObject)A, ksp->viewerMat, ksp->formatMat);
1215: if (ksp->viewPMat) ObjectView((PetscObject)P, ksp->viewerPMat, ksp->formatPMat);
1216: if (ksp->viewRhs) ObjectView((PetscObject)B, ksp->viewerRhs, ksp->formatRhs);
1217: if (ksp->viewSol) ObjectView((PetscObject)X, ksp->viewerSol, ksp->formatSol);
1218: if (ksp->view) KSPView(ksp, ksp->viewer);
1219: PetscLogEventEnd(KSP_MatSolve, ksp, B, X, 0);
1220: } else {
1221: PetscInfo(ksp, "KSP type %s solving column by column\n", ((PetscObject)ksp)->type_name);
1222: for (n2 = 0; n2 < N2; ++n2) {
1223: MatDenseGetColumnVecRead(B, n2, &cb);
1224: MatDenseGetColumnVecWrite(X, n2, &cx);
1225: KSPSolve(ksp, cb, cx);
1226: MatDenseRestoreColumnVecWrite(X, n2, &cx);
1227: MatDenseRestoreColumnVecRead(B, n2, &cb);
1228: }
1229: }
1230: return 0;
1231: }
1233: /*@
1234: KSPSetMatSolveBatchSize - Sets the maximum number of columns treated simultaneously in `KSPMatSolve()`.
1236: Logically collective
1238: Input Parameters:
1239: + ksp - iterative context
1240: - bs - batch size
1242: Level: advanced
1244: .seealso: [](chapter_ksp), `KSPMatSolve()`, `KSPGetMatSolveBatchSize()`, `-mat_mumps_icntl_27`, `-matmatmult_Bbn`
1245: @*/
1246: PetscErrorCode KSPSetMatSolveBatchSize(KSP ksp, PetscInt bs)
1247: {
1250: ksp->nmax = bs;
1251: return 0;
1252: }
1254: /*@
1255: KSPGetMatSolveBatchSize - Gets the maximum number of columns treated simultaneously in `KSPMatSolve()`.
1257: Input Parameter:
1258: . ksp - iterative context
1260: Output Parameter:
1261: . bs - batch size
1263: Level: advanced
1265: .seealso: [](chapter_ksp), `KSPMatSolve()`, `KSPSetMatSolveBatchSize()`, `-mat_mumps_icntl_27`, `-matmatmult_Bbn`
1266: @*/
1267: PetscErrorCode KSPGetMatSolveBatchSize(KSP ksp, PetscInt *bs)
1268: {
1271: *bs = ksp->nmax;
1272: return 0;
1273: }
1275: /*@
1276: KSPResetViewers - Resets all the viewers set from the options database during `KSPSetFromOptions()`
1278: Collective
1280: Input Parameter:
1281: . ksp - iterative context obtained from `KSPCreate()`
1283: Level: beginner
1285: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSetUp()`, `KSPSolve()`, `KSPSetFromOptions()`, `KSP`
1286: @*/
1287: PetscErrorCode KSPResetViewers(KSP ksp)
1288: {
1290: if (!ksp) return 0;
1291: PetscViewerDestroy(&ksp->viewer);
1292: PetscViewerDestroy(&ksp->viewerPre);
1293: PetscViewerDestroy(&ksp->viewerRate);
1294: PetscViewerDestroy(&ksp->viewerMat);
1295: PetscViewerDestroy(&ksp->viewerPMat);
1296: PetscViewerDestroy(&ksp->viewerRhs);
1297: PetscViewerDestroy(&ksp->viewerSol);
1298: PetscViewerDestroy(&ksp->viewerMatExp);
1299: PetscViewerDestroy(&ksp->viewerEV);
1300: PetscViewerDestroy(&ksp->viewerSV);
1301: PetscViewerDestroy(&ksp->viewerEVExp);
1302: PetscViewerDestroy(&ksp->viewerFinalRes);
1303: PetscViewerDestroy(&ksp->viewerPOpExp);
1304: PetscViewerDestroy(&ksp->viewerDScale);
1305: ksp->view = PETSC_FALSE;
1306: ksp->viewPre = PETSC_FALSE;
1307: ksp->viewMat = PETSC_FALSE;
1308: ksp->viewPMat = PETSC_FALSE;
1309: ksp->viewRhs = PETSC_FALSE;
1310: ksp->viewSol = PETSC_FALSE;
1311: ksp->viewMatExp = PETSC_FALSE;
1312: ksp->viewEV = PETSC_FALSE;
1313: ksp->viewSV = PETSC_FALSE;
1314: ksp->viewEVExp = PETSC_FALSE;
1315: ksp->viewFinalRes = PETSC_FALSE;
1316: ksp->viewPOpExp = PETSC_FALSE;
1317: ksp->viewDScale = PETSC_FALSE;
1318: return 0;
1319: }
1321: /*@
1322: KSPReset - Resets a `KSP` context to the kspsetupcalled = 0 state and removes any allocated Vecs and Mats
1324: Collective
1326: Input Parameter:
1327: . ksp - iterative context obtained from `KSPCreate()`
1329: Level: beginner
1331: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSetUp()`, `KSPSolve()`, `KSP`
1332: @*/
1333: PetscErrorCode KSPReset(KSP ksp)
1334: {
1336: if (!ksp) return 0;
1337: PetscTryTypeMethod(ksp, reset);
1338: if (ksp->pc) PCReset(ksp->pc);
1339: if (ksp->guess) {
1340: KSPGuess guess = ksp->guess;
1341: PetscTryTypeMethod(guess, reset);
1342: }
1343: VecDestroyVecs(ksp->nwork, &ksp->work);
1344: VecDestroy(&ksp->vec_rhs);
1345: VecDestroy(&ksp->vec_sol);
1346: VecDestroy(&ksp->diagonal);
1347: VecDestroy(&ksp->truediagonal);
1349: KSPResetViewers(ksp);
1351: ksp->setupstage = KSP_SETUP_NEW;
1352: ksp->nmax = PETSC_DECIDE;
1353: return 0;
1354: }
1356: /*@C
1357: KSPDestroy - Destroys `KSP` context.
1359: Collective
1361: Input Parameter:
1362: . ksp - iterative context obtained from `KSPCreate()`
1364: Level: beginner
1366: .seealso: [](chapter_ksp), `KSPCreate()`, `KSPSetUp()`, `KSPSolve()`, `KSP`
1367: @*/
1368: PetscErrorCode KSPDestroy(KSP *ksp)
1369: {
1370: PC pc;
1372: if (!*ksp) return 0;
1374: if (--((PetscObject)(*ksp))->refct > 0) {
1375: *ksp = NULL;
1376: return 0;
1377: }
1379: PetscObjectSAWsViewOff((PetscObject)*ksp);
1381: /*
1382: Avoid a cascading call to PCReset(ksp->pc) from the following call:
1383: PCReset() shouldn't be called from KSPDestroy() as it is unprotected by pc's
1384: refcount (and may be shared, e.g., by other ksps).
1385: */
1386: pc = (*ksp)->pc;
1387: (*ksp)->pc = NULL;
1388: KSPReset((*ksp));
1389: (*ksp)->pc = pc;
1390: PetscTryTypeMethod((*ksp), destroy);
1392: if ((*ksp)->transpose.use_explicittranspose) {
1393: MatDestroy(&(*ksp)->transpose.AT);
1394: MatDestroy(&(*ksp)->transpose.BT);
1395: (*ksp)->transpose.reuse_transpose = PETSC_FALSE;
1396: }
1398: KSPGuessDestroy(&(*ksp)->guess);
1399: DMDestroy(&(*ksp)->dm);
1400: PCDestroy(&(*ksp)->pc);
1401: PetscFree((*ksp)->res_hist_alloc);
1402: PetscFree((*ksp)->err_hist_alloc);
1403: if ((*ksp)->convergeddestroy) (*(*ksp)->convergeddestroy)((*ksp)->cnvP);
1404: KSPMonitorCancel((*ksp));
1405: KSPConvergedReasonViewCancel((*ksp));
1406: PetscViewerDestroy(&(*ksp)->eigviewer);
1407: PetscHeaderDestroy(ksp);
1408: return 0;
1409: }
1411: /*@
1412: KSPSetPCSide - Sets the preconditioning side.
1414: Logically Collective
1416: Input Parameter:
1417: . ksp - iterative context obtained from `KSPCreate()`
1419: Output Parameter:
1420: . side - the preconditioning side, where side is one of
1421: .vb
1422: PC_LEFT - left preconditioning (default)
1423: PC_RIGHT - right preconditioning
1424: PC_SYMMETRIC - symmetric preconditioning
1425: .ve
1427: Options Database Keys:
1428: . -ksp_pc_side <right,left,symmetric> - `KSP` preconditioner side
1430: Notes:
1431: Left preconditioning is used by default for most Krylov methods except KSPFGMRES which only supports right preconditioning.
1433: For methods changing the side of the preconditioner changes the norm type that is used, see `KSPSetNormType()`.
1435: Symmetric preconditioning is currently available only for the KSPQCG method. Note, however, that
1436: symmetric preconditioning can be emulated by using either right or left
1437: preconditioning and a pre or post processing step.
1439: Setting the PC side often affects the default norm type. See `KSPSetNormType()` for details.
1441: Level: intermediate
1443: .seealso: [](chapter_ksp), `KSPGetPCSide()`, `KSPSetNormType()`, `KSPGetNormType()`, `KSP`
1444: @*/
1445: PetscErrorCode KSPSetPCSide(KSP ksp, PCSide side)
1446: {
1449: ksp->pc_side = ksp->pc_side_set = side;
1450: return 0;
1451: }
1453: /*@
1454: KSPGetPCSide - Gets the preconditioning side.
1456: Not Collective
1458: Input Parameter:
1459: . ksp - iterative context obtained from `KSPCreate()`
1461: Output Parameter:
1462: . side - the preconditioning side, where side is one of
1463: .vb
1464: PC_LEFT - left preconditioning (default)
1465: PC_RIGHT - right preconditioning
1466: PC_SYMMETRIC - symmetric preconditioning
1467: .ve
1469: Level: intermediate
1471: .seealso: [](chapter_ksp), `KSPSetPCSide()`, `KSP`
1472: @*/
1473: PetscErrorCode KSPGetPCSide(KSP ksp, PCSide *side)
1474: {
1477: KSPSetUpNorms_Private(ksp, PETSC_TRUE, &ksp->normtype, &ksp->pc_side);
1478: *side = ksp->pc_side;
1479: return 0;
1480: }
1482: /*@
1483: KSPGetTolerances - Gets the relative, absolute, divergence, and maximum
1484: iteration tolerances used by the default `KSP` convergence tests.
1486: Not Collective
1488: Input Parameter:
1489: . ksp - the Krylov subspace context
1491: Output Parameters:
1492: + rtol - the relative convergence tolerance
1493: . abstol - the absolute convergence tolerance
1494: . dtol - the divergence tolerance
1495: - maxits - maximum number of iterations
1497: Notes:
1498: The user can specify NULL for any parameter that is not needed.
1500: Level: intermediate
1502: maximum, iterations
1504: .seealso: [](chapter_ksp), `KSPSetTolerances()`, `KSP`
1505: @*/
1506: PetscErrorCode KSPGetTolerances(KSP ksp, PetscReal *rtol, PetscReal *abstol, PetscReal *dtol, PetscInt *maxits)
1507: {
1509: if (abstol) *abstol = ksp->abstol;
1510: if (rtol) *rtol = ksp->rtol;
1511: if (dtol) *dtol = ksp->divtol;
1512: if (maxits) *maxits = ksp->max_it;
1513: return 0;
1514: }
1516: /*@
1517: KSPSetTolerances - Sets the relative, absolute, divergence, and maximum
1518: iteration tolerances used by the default `KSP` convergence testers.
1520: Logically Collective
1522: Input Parameters:
1523: + ksp - the Krylov subspace context
1524: . rtol - the relative convergence tolerance, relative decrease in the (possibly preconditioned) residual norm
1525: . abstol - the absolute convergence tolerance absolute size of the (possibly preconditioned) residual norm
1526: . dtol - the divergence tolerance, amount (possibly preconditioned) residual norm can increase before `KSPConvergedDefault()` concludes that the method is diverging
1527: - maxits - maximum number of iterations to use
1529: Options Database Keys:
1530: + -ksp_atol <abstol> - Sets abstol
1531: . -ksp_rtol <rtol> - Sets rtol
1532: . -ksp_divtol <dtol> - Sets dtol
1533: - -ksp_max_it <maxits> - Sets maxits
1535: Notes:
1536: Use PETSC_DEFAULT to retain the default value of any of the tolerances.
1538: See `KSPConvergedDefault()` for details how these parameters are used in the default convergence test. See also `KSPSetConvergenceTest()`
1539: for setting user-defined stopping criteria.
1541: Level: intermediate
1543: convergence, maximum, iterations
1545: .seealso: [](chapter_ksp), `KSPGetTolerances()`, `KSPConvergedDefault()`, `KSPSetConvergenceTest()`, `KSP`
1546: @*/
1547: PetscErrorCode KSPSetTolerances(KSP ksp, PetscReal rtol, PetscReal abstol, PetscReal dtol, PetscInt maxits)
1548: {
1555: if (rtol != PETSC_DEFAULT) {
1557: ksp->rtol = rtol;
1558: }
1559: if (abstol != PETSC_DEFAULT) {
1561: ksp->abstol = abstol;
1562: }
1563: if (dtol != PETSC_DEFAULT) {
1565: ksp->divtol = dtol;
1566: }
1567: if (maxits != PETSC_DEFAULT) {
1569: ksp->max_it = maxits;
1570: }
1571: return 0;
1572: }
1574: /*@
1575: KSPSetInitialGuessNonzero - Tells the iterative solver that the
1576: initial guess is nonzero; otherwise `KSP` assumes the initial guess
1577: is to be zero (and thus zeros it out before solving).
1579: Logically Collective
1581: Input Parameters:
1582: + ksp - iterative context obtained from `KSPCreate()`
1583: - flg - ``PETSC_TRUE`` indicates the guess is non-zero, `PETSC_FALSE` indicates the guess is zero
1585: Options database keys:
1586: . -ksp_initial_guess_nonzero <true,false> - use nonzero initial guess
1588: Level: beginner
1590: Notes:
1591: If this is not called the X vector is zeroed in the call to `KSPSolve()`.
1593: .seealso: [](chapter_ksp), `KSPGetInitialGuessNonzero()`, `KSPSetGuessType()`, `KSPGuessType`, `KSP`
1594: @*/
1595: PetscErrorCode KSPSetInitialGuessNonzero(KSP ksp, PetscBool flg)
1596: {
1599: ksp->guess_zero = (PetscBool) !(int)flg;
1600: return 0;
1601: }
1603: /*@
1604: KSPGetInitialGuessNonzero - Determines whether the `KSP` solver is using
1605: a zero initial guess.
1607: Not Collective
1609: Input Parameter:
1610: . ksp - iterative context obtained from `KSPCreate()`
1612: Output Parameter:
1613: . flag - `PETSC_TRUE` if guess is nonzero, else `PETSC_FALSE`
1615: Level: intermediate
1617: .seealso: [](chapter_ksp), `KSPSetInitialGuessNonzero()`, `KSP`
1618: @*/
1619: PetscErrorCode KSPGetInitialGuessNonzero(KSP ksp, PetscBool *flag)
1620: {
1623: if (ksp->guess_zero) *flag = PETSC_FALSE;
1624: else *flag = PETSC_TRUE;
1625: return 0;
1626: }
1628: /*@
1629: KSPSetErrorIfNotConverged - Causes `KSPSolve()` to generate an error if the solver has not converged as soon as the error is detected.
1631: Logically Collective
1633: Input Parameters:
1634: + ksp - iterative context obtained from `KSPCreate()`
1635: - flg - `PETSC_TRUE` indicates you want the error generated
1637: Options database keys:
1638: . -ksp_error_if_not_converged <true,false> - generate an error and stop the program
1640: Level: intermediate
1642: Notes:
1643: Normally PETSc continues if a linear solver fails to converge, you can call `KSPGetConvergedReason()` after a `KSPSolve()`
1644: to determine if it has converged.
1646: A `KSP_DIVERGED_ITS` will not generate an error in a `KSPSolve()` inside a nested linear solver
1648: .seealso: [](chapter_ksp), `KSPGetErrorIfNotConverged()`, `KSP`
1649: @*/
1650: PetscErrorCode KSPSetErrorIfNotConverged(KSP ksp, PetscBool flg)
1651: {
1654: ksp->errorifnotconverged = flg;
1655: return 0;
1656: }
1658: /*@
1659: KSPGetErrorIfNotConverged - Will `KSPSolve()` generate an error if the solver does not converge?
1661: Not Collective
1663: Input Parameter:
1664: . ksp - iterative context obtained from KSPCreate()
1666: Output Parameter:
1667: . flag - `PETSC_TRUE` if it will generate an error, else `PETSC_FALSE`
1669: Level: intermediate
1671: .seealso: [](chapter_ksp), `KSPSetErrorIfNotConverged()`, `KSP`
1672: @*/
1673: PetscErrorCode KSPGetErrorIfNotConverged(KSP ksp, PetscBool *flag)
1674: {
1677: *flag = ksp->errorifnotconverged;
1678: return 0;
1679: }
1681: /*@
1682: KSPSetInitialGuessKnoll - Tells the iterative solver to use `PCApply()` to compute the initial guess (The Knoll trick)
1684: Logically Collective
1686: Input Parameters:
1687: + ksp - iterative context obtained from `KSPCreate()`
1688: - flg - `PETSC_TRUE` or `PETSC_FALSE`
1690: Level: advanced
1692: Developer Note: the Knoll trick is not currently implemented using the KSPGuess class
1694: .seealso: [](chapter_ksp), `KSPGetInitialGuessKnoll()`, `KSPSetInitialGuessNonzero()`, `KSPGetInitialGuessNonzero()`, `KSP`
1695: @*/
1696: PetscErrorCode KSPSetInitialGuessKnoll(KSP ksp, PetscBool flg)
1697: {
1700: ksp->guess_knoll = flg;
1701: return 0;
1702: }
1704: /*@
1705: KSPGetInitialGuessKnoll - Determines whether the `KSP` solver is using the Knoll trick (using PCApply(pc,b,...) to compute
1706: the initial guess
1708: Not Collective
1710: Input Parameter:
1711: . ksp - iterative context obtained from `KSPCreate()`
1713: Output Parameter:
1714: . flag - `PETSC_TRUE` if using Knoll trick, else `PETSC_FALSE`
1716: Level: advanced
1718: .seealso: [](chapter_ksp), `KSPSetInitialGuessKnoll()`, `KSPSetInitialGuessNonzero()`, `KSPGetInitialGuessNonzero()`, `KSP`
1719: @*/
1720: PetscErrorCode KSPGetInitialGuessKnoll(KSP ksp, PetscBool *flag)
1721: {
1724: *flag = ksp->guess_knoll;
1725: return 0;
1726: }
1728: /*@
1729: KSPGetComputeSingularValues - Gets the flag indicating whether the extreme singular
1730: values will be calculated via a Lanczos or Arnoldi process as the linear
1731: system is solved.
1733: Not Collective
1735: Input Parameter:
1736: . ksp - iterative context obtained from `KSPCreate()`
1738: Output Parameter:
1739: . flg - `PETSC_TRUE` or `PETSC_FALSE`
1741: Options Database Key:
1742: . -ksp_monitor_singular_value - Activates `KSPSetComputeSingularValues()`
1744: Notes:
1745: Currently this option is not valid for all iterative methods.
1747: Many users may just want to use the monitoring routine
1748: `KSPMonitorSingularValue()` (which can be set with option -ksp_monitor_singular_value)
1749: to print the singular values at each iteration of the linear solve.
1751: Level: advanced
1753: .seealso: [](chapter_ksp), `KSPComputeExtremeSingularValues()`, `KSPMonitorSingularValue()`, `KSP`
1754: @*/
1755: PetscErrorCode KSPGetComputeSingularValues(KSP ksp, PetscBool *flg)
1756: {
1759: *flg = ksp->calc_sings;
1760: return 0;
1761: }
1763: /*@
1764: KSPSetComputeSingularValues - Sets a flag so that the extreme singular
1765: values will be calculated via a Lanczos or Arnoldi process as the linear
1766: system is solved.
1768: Logically Collective
1770: Input Parameters:
1771: + ksp - iterative context obtained from `KSPCreate()`
1772: - flg - `PETSC_TRUE` or `PETSC_FALSE`
1774: Options Database Key:
1775: . -ksp_monitor_singular_value - Activates `KSPSetComputeSingularValues()`
1777: Notes:
1778: Currently this option is not valid for all iterative methods.
1780: Many users may just want to use the monitoring routine
1781: `KSPMonitorSingularValue()` (which can be set with option -ksp_monitor_singular_value)
1782: to print the singular values at each iteration of the linear solve.
1784: Level: advanced
1786: .seealso: [](chapter_ksp), `KSPComputeExtremeSingularValues()`, `KSPMonitorSingularValue()`, `KSP`
1787: @*/
1788: PetscErrorCode KSPSetComputeSingularValues(KSP ksp, PetscBool flg)
1789: {
1792: ksp->calc_sings = flg;
1793: return 0;
1794: }
1796: /*@
1797: KSPGetComputeEigenvalues - Gets the flag indicating that the extreme eigenvalues
1798: values will be calculated via a Lanczos or Arnoldi process as the linear
1799: system is solved.
1801: Not Collective
1803: Input Parameter:
1804: . ksp - iterative context obtained from `KSPCreate()`
1806: Output Parameter:
1807: . flg - `PETSC_TRUE` or `PETSC_FALSE`
1809: Notes:
1810: Currently this option is not valid for all iterative methods.
1812: Level: advanced
1814: .seealso: [](chapter_ksp), `KSPComputeEigenvalues()`, `KSPComputeEigenvaluesExplicitly()`, `KSP`
1815: @*/
1816: PetscErrorCode KSPGetComputeEigenvalues(KSP ksp, PetscBool *flg)
1817: {
1820: *flg = ksp->calc_sings;
1821: return 0;
1822: }
1824: /*@
1825: KSPSetComputeEigenvalues - Sets a flag so that the extreme eigenvalues
1826: values will be calculated via a Lanczos or Arnoldi process as the linear
1827: system is solved.
1829: Logically Collective
1831: Input Parameters:
1832: + ksp - iterative context obtained from `KSPCreate()`
1833: - flg - `PETSC_TRUE` or `PETSC_FALSE`
1835: Notes:
1836: Currently this option is not valid for all iterative methods.
1838: Level: advanced
1840: .seealso: [](chapter_ksp), `KSPComputeEigenvalues()`, `KSPComputeEigenvaluesExplicitly()`, `KSP`
1841: @*/
1842: PetscErrorCode KSPSetComputeEigenvalues(KSP ksp, PetscBool flg)
1843: {
1846: ksp->calc_sings = flg;
1847: return 0;
1848: }
1850: /*@
1851: KSPSetComputeRitz - Sets a flag so that the Ritz or harmonic Ritz pairs
1852: will be calculated via a Lanczos or Arnoldi process as the linear
1853: system is solved.
1855: Logically Collective
1857: Input Parameters:
1858: + ksp - iterative context obtained from `KSPCreate()`
1859: - flg - `PETSC_TRUE` or `PETSC_FALSE`
1861: Notes:
1862: Currently this option is only valid for the GMRES method.
1864: Level: advanced
1866: .seealso: [](chapter_ksp), `KSPComputeRitz()`, `KSP`
1867: @*/
1868: PetscErrorCode KSPSetComputeRitz(KSP ksp, PetscBool flg)
1869: {
1872: ksp->calc_ritz = flg;
1873: return 0;
1874: }
1876: /*@
1877: KSPGetRhs - Gets the right-hand-side vector for the linear system to
1878: be solved.
1880: Not Collective
1882: Input Parameter:
1883: . ksp - iterative context obtained from `KSPCreate()`
1885: Output Parameter:
1886: . r - right-hand-side vector
1888: Level: developer
1890: .seealso: [](chapter_ksp), `KSPGetSolution()`, `KSPSolve()`, `KSP`
1891: @*/
1892: PetscErrorCode KSPGetRhs(KSP ksp, Vec *r)
1893: {
1896: *r = ksp->vec_rhs;
1897: return 0;
1898: }
1900: /*@
1901: KSPGetSolution - Gets the location of the solution for the
1902: linear system to be solved. Note that this may not be where the solution
1903: is stored during the iterative process; see `KSPBuildSolution()`.
1905: Not Collective
1907: Input Parameters:
1908: . ksp - iterative context obtained from `KSPCreate()`
1910: Output Parameters:
1911: . v - solution vector
1913: Level: developer
1915: .seealso: [](chapter_ksp), `KSPGetRhs()`, `KSPBuildSolution()`, `KSPSolve()`, `KSP`
1916: @*/
1917: PetscErrorCode KSPGetSolution(KSP ksp, Vec *v)
1918: {
1921: *v = ksp->vec_sol;
1922: return 0;
1923: }
1925: /*@
1926: KSPSetPC - Sets the preconditioner to be used to calculate the
1927: application of the preconditioner on a vector.
1929: Collective
1931: Input Parameters:
1932: + ksp - iterative context obtained from `KSPCreate()`
1933: - pc - the preconditioner object (can be NULL)
1935: Notes:
1936: Use `KSPGetPC()` to retrieve the preconditioner context.
1938: Level: developer
1940: .seealso: [](chapter_ksp), `KSPGetPC()`, `KSP`
1941: @*/
1942: PetscErrorCode KSPSetPC(KSP ksp, PC pc)
1943: {
1945: if (pc) {
1948: }
1949: PetscObjectReference((PetscObject)pc);
1950: PCDestroy(&ksp->pc);
1951: ksp->pc = pc;
1952: return 0;
1953: }
1955: /*@
1956: KSPGetPC - Returns a pointer to the preconditioner context
1957: set with `KSPSetPC()`.
1959: Not Collective
1961: Input Parameters:
1962: . ksp - iterative context obtained from `KSPCreate()`
1964: Output Parameter:
1965: . pc - preconditioner context
1967: Level: developer
1969: .seealso: [](chapter_ksp), `KSPSetPC()`, `KSP`
1970: @*/
1971: PetscErrorCode KSPGetPC(KSP ksp, PC *pc)
1972: {
1975: if (!ksp->pc) {
1976: PCCreate(PetscObjectComm((PetscObject)ksp), &ksp->pc);
1977: PetscObjectIncrementTabLevel((PetscObject)ksp->pc, (PetscObject)ksp, 0);
1978: PetscObjectSetOptions((PetscObject)ksp->pc, ((PetscObject)ksp)->options);
1979: }
1980: *pc = ksp->pc;
1981: return 0;
1982: }
1984: /*@
1985: KSPMonitor - runs the user provided monitor routines, if they exist
1987: Collective
1989: Input Parameters:
1990: + ksp - iterative context obtained from `KSPCreate()`
1991: . it - iteration number
1992: - rnorm - relative norm of the residual
1994: Notes:
1995: This routine is called by the `KSP` implementations.
1996: It does not typically need to be called by the user.
1998: Level: developer
2000: .seealso: [](chapter_ksp), `KSPMonitorSet()`
2001: @*/
2002: PetscErrorCode KSPMonitor(KSP ksp, PetscInt it, PetscReal rnorm)
2003: {
2004: PetscInt i, n = ksp->numbermonitors;
2006: for (i = 0; i < n; i++) (*ksp->monitor[i])(ksp, it, rnorm, ksp->monitorcontext[i]);
2007: return 0;
2008: }
2010: /*@C
2011: KSPMonitorSet - Sets an ADDITIONAL function to be called at every iteration to monitor
2012: the residual/error etc.
2014: Logically Collective
2016: Input Parameters:
2017: + ksp - iterative context obtained from `KSPCreate()`
2018: . monitor - pointer to function (if this is NULL, it turns off monitoring
2019: . mctx - [optional] context for private data for the
2020: monitor routine (use NULL if no context is desired)
2021: - monitordestroy - [optional] routine that frees monitor context
2022: (may be NULL)
2024: Calling Sequence of monitor:
2025: $ monitor (KSP ksp, PetscInt it, PetscReal rnorm, void *mctx)
2027: + ksp - iterative context obtained from `KSPCreate()`
2028: . it - iteration number
2029: . rnorm - (estimated) 2-norm of (preconditioned) residual
2030: - mctx - optional monitoring context, as set by `KSPMonitorSet()`
2032: Options Database Keys:
2033: + -ksp_monitor - sets `KSPMonitorResidual()`
2034: . -ksp_monitor draw - sets `KSPMonitorResidualDraw()` and plots residual
2035: . -ksp_monitor draw::draw_lg - sets `KSPMonitorResidualDrawLG()` and plots residual
2036: . -ksp_monitor_pause_final - Pauses any graphics when the solve finishes (only works for internal monitors)
2037: . -ksp_monitor_true_residual - sets `KSPMonitorTrueResidual()`
2038: . -ksp_monitor_true_residual draw::draw_lg - sets `KSPMonitorTrueResidualDrawLG()` and plots residual
2039: . -ksp_monitor_max - sets `KSPMonitorTrueResidualMax()`
2040: . -ksp_monitor_singular_value - sets `KSPMonitorSingularValue()`
2041: - -ksp_monitor_cancel - cancels all monitors that have
2042: been hardwired into a code by
2043: calls to `KSPMonitorSet()`, but
2044: does not cancel those set via
2045: the options database.
2047: Notes:
2048: The default is to do nothing. To print the residual, or preconditioned
2049: residual if `KSPSetNormType`(ksp,`KSP_NORM_PRECONDITIONED`) was called, use
2050: `KSPMonitorResidual()` as the monitoring routine, with a `PETSCVIEWERASCII` as the
2051: context.
2053: Several different monitoring routines may be set by calling
2054: `KSPMonitorSet()` multiple times; all will be called in the
2055: order in which they were set.
2057: Fortran Notes:
2058: Only a single monitor function can be set for each `KSP` object
2060: Level: beginner
2062: .seealso: [](chapter_ksp), `KSPMonitorResidual()`, `KSPMonitorCancel()`, `KSP`
2063: @*/
2064: PetscErrorCode KSPMonitorSet(KSP ksp, PetscErrorCode (*monitor)(KSP, PetscInt, PetscReal, void *), void *mctx, PetscErrorCode (*monitordestroy)(void **))
2065: {
2066: PetscInt i;
2067: PetscBool identical;
2070: for (i = 0; i < ksp->numbermonitors; i++) {
2071: PetscMonitorCompare((PetscErrorCode(*)(void))monitor, mctx, monitordestroy, (PetscErrorCode(*)(void))ksp->monitor[i], ksp->monitorcontext[i], ksp->monitordestroy[i], &identical);
2072: if (identical) return 0;
2073: }
2075: ksp->monitor[ksp->numbermonitors] = monitor;
2076: ksp->monitordestroy[ksp->numbermonitors] = monitordestroy;
2077: ksp->monitorcontext[ksp->numbermonitors++] = (void *)mctx;
2078: return 0;
2079: }
2081: /*@
2082: KSPMonitorCancel - Clears all monitors for a `KSP` object.
2084: Logically Collective
2086: Input Parameters:
2087: . ksp - iterative context obtained from `KSPCreate()`
2089: Options Database Key:
2090: . -ksp_monitor_cancel - Cancels all monitors that have been hardwired into a code by calls to `KSPMonitorSet()`, but does not cancel those set via the options database.
2092: Level: intermediate
2094: .seealso: [](chapter_ksp), `KSPMonitorResidual()`, `KSPMonitorSet()`, `KSP`
2095: @*/
2096: PetscErrorCode KSPMonitorCancel(KSP ksp)
2097: {
2098: PetscInt i;
2101: for (i = 0; i < ksp->numbermonitors; i++) {
2102: if (ksp->monitordestroy[i]) (*ksp->monitordestroy[i])(&ksp->monitorcontext[i]);
2103: }
2104: ksp->numbermonitors = 0;
2105: return 0;
2106: }
2108: /*@C
2109: KSPGetMonitorContext - Gets the monitoring context, as set by `KSPMonitorSet()` for the FIRST monitor only.
2111: Not Collective
2113: Input Parameter:
2114: . ksp - iterative context obtained from `KSPCreate()`
2116: Output Parameter:
2117: . ctx - monitoring context
2119: Level: intermediate
2121: .seealso: [](chapter_ksp), `KSPMonitorResidual()`, `KSP`
2122: @*/
2123: PetscErrorCode KSPGetMonitorContext(KSP ksp, void *ctx)
2124: {
2126: *(void **)ctx = ksp->monitorcontext[0];
2127: return 0;
2128: }
2130: /*@
2131: KSPSetResidualHistory - Sets the array used to hold the residual history.
2132: If set, this array will contain the residual norms computed at each
2133: iteration of the solver.
2135: Not Collective
2137: Input Parameters:
2138: + ksp - iterative context obtained from `KSPCreate()`
2139: . a - array to hold history
2140: . na - size of a
2141: - reset - `PETSC_TRUE` indicates the history counter is reset to zero
2142: for each new linear solve
2144: Level: advanced
2146: Notes:
2147: If provided, he array is NOT freed by PETSc so the user needs to keep track of it and destroy once the `KSP` object is destroyed.
2148: If 'a' is NULL then space is allocated for the history. If 'na' `PETSC_DECIDE` or `PETSC_DEFAULT` then a
2149: default array of length 10000 is allocated.
2151: If the array is not long enough then once the iterations is longer than the array length `KSPSolve()` stops recording the history
2153: .seealso: [](chapter_ksp), `KSPGetResidualHistory()`, `KSP`
2154: @*/
2155: PetscErrorCode KSPSetResidualHistory(KSP ksp, PetscReal a[], PetscInt na, PetscBool reset)
2156: {
2159: PetscFree(ksp->res_hist_alloc);
2160: if (na != PETSC_DECIDE && na != PETSC_DEFAULT && a) {
2161: ksp->res_hist = a;
2162: ksp->res_hist_max = (size_t)na;
2163: } else {
2164: if (na != PETSC_DECIDE && na != PETSC_DEFAULT) ksp->res_hist_max = (size_t)na;
2165: else ksp->res_hist_max = 10000; /* like default ksp->max_it */
2166: PetscCalloc1(ksp->res_hist_max, &ksp->res_hist_alloc);
2168: ksp->res_hist = ksp->res_hist_alloc;
2169: }
2170: ksp->res_hist_len = 0;
2171: ksp->res_hist_reset = reset;
2172: return 0;
2173: }
2175: /*@C
2176: KSPGetResidualHistory - Gets the array used to hold the residual history and the number of residuals it contains.
2178: Not Collective
2180: Input Parameter:
2181: . ksp - iterative context obtained from `KSPCreate()`
2183: Output Parameters:
2184: + a - pointer to array to hold history (or NULL)
2185: - na - number of used entries in a (or NULL)
2187: Level: advanced
2189: Note:
2190: This array is borrowed and should not be freed by the caller.
2192: Can only be called after a `KSPSetResidualHistory()` otherwise a and na are set to zero
2194: Fortran Note:
2195: The Fortran version of this routine has a calling sequence
2196: $ call `KSPGetResidualHistory`(`KSP` ksp, integer na, integer ierr)
2197: note that you have passed a Fortran array into `KSPSetResidualHistory()` and you need
2198: to access the residual values from this Fortran array you provided. Only the na (number of
2199: residual norms currently held) is set.
2201: .seealso: [](chapter_ksp), `KSPSetResidualHistory()`, `KSP`
2202: @*/
2203: PetscErrorCode KSPGetResidualHistory(KSP ksp, const PetscReal *a[], PetscInt *na)
2204: {
2206: if (a) *a = ksp->res_hist;
2207: if (na) *na = (PetscInt)ksp->res_hist_len;
2208: return 0;
2209: }
2211: /*@
2212: KSPSetErrorHistory - Sets the array used to hold the error history. If set, this array will contain the error norms computed at each iteration of the solver.
2214: Not Collective
2216: Input Parameters:
2217: + ksp - iterative context obtained from `KSPCreate()`
2218: . a - array to hold history
2219: . na - size of a
2220: - reset - `PETSC_TRUE` indicates the history counter is reset to zero for each new linear solve
2222: Level: advanced
2224: Notes:
2225: If provided, the array is NOT freed by PETSc so the user needs to keep track of it and destroy once the `KSP` object is destroyed.
2226: If 'a' is NULL then space is allocated for the history. If 'na' PETSC_DECIDE or PETSC_DEFAULT then a default array of length 10000 is allocated.
2228: If the array is not long enough then once the iterations is longer than the array length `KSPSolve()` stops recording the history
2230: .seealso: [](chapter_ksp), `KSPGetErrorHistory()`, `KSPSetResidualHistory()`, `KSP`
2231: @*/
2232: PetscErrorCode KSPSetErrorHistory(KSP ksp, PetscReal a[], PetscInt na, PetscBool reset)
2233: {
2236: PetscFree(ksp->err_hist_alloc);
2237: if (na != PETSC_DECIDE && na != PETSC_DEFAULT && a) {
2238: ksp->err_hist = a;
2239: ksp->err_hist_max = (size_t)na;
2240: } else {
2241: if (na != PETSC_DECIDE && na != PETSC_DEFAULT) ksp->err_hist_max = (size_t)na;
2242: else ksp->err_hist_max = 10000; /* like default ksp->max_it */
2243: PetscCalloc1(ksp->err_hist_max, &ksp->err_hist_alloc);
2245: ksp->err_hist = ksp->err_hist_alloc;
2246: }
2247: ksp->err_hist_len = 0;
2248: ksp->err_hist_reset = reset;
2249: return 0;
2250: }
2252: /*@C
2253: KSPGetErrorHistory - Gets the array used to hold the error history and the number of residuals it contains.
2255: Not Collective
2257: Input Parameter:
2258: . ksp - iterative context obtained from `KSPCreate()`
2260: Output Parameters:
2261: + a - pointer to array to hold history (or NULL)
2262: - na - number of used entries in a (or NULL)
2264: Level: advanced
2266: Notes:
2267: This array is borrowed and should not be freed by the caller.
2268: Can only be called after a `KSPSetErrorHistory()` otherwise a and na are set to zero
2270: Fortran Note:
2271: The Fortran version of this routine has a calling sequence
2272: $ call KSPGetErrorHistory(KSP ksp, integer na, integer ierr)
2273: note that you have passed a Fortran array into `KSPSetErrorHistory()` and you need
2274: to access the residual values from this Fortran array you provided. Only the na (number of
2275: residual norms currently held) is set.
2277: .seealso: [](chapter_ksp), `KSPSetErrorHistory()`, `KSPGetResidualHistory()`, `KSP`
2278: @*/
2279: PetscErrorCode KSPGetErrorHistory(KSP ksp, const PetscReal *a[], PetscInt *na)
2280: {
2282: if (a) *a = ksp->err_hist;
2283: if (na) *na = (PetscInt)ksp->err_hist_len;
2284: return 0;
2285: }
2287: /*
2288: KSPComputeConvergenceRate - Compute the convergence rate for the iteration
2290: Not collective
2292: Input Parameter:
2293: . ksp - The `KSP`
2295: Output Parameters:
2296: + cr - The residual contraction rate
2297: . rRsq - The coefficient of determination, R^2, indicating the linearity of the data
2298: . ce - The error contraction rate
2299: - eRsq - The coefficient of determination, R^2, indicating the linearity of the data
2301: Note:
2302: Suppose that the residual is reduced linearly, $r_k = c^k r_0$, which means $log r_k = log r_0 + k log c$. After linear regression,
2303: the slope is $\log c$. The coefficient of determination is given by $1 - \frac{\sum_i (y_i - f(x_i))^2}{\sum_i (y_i - \bar y)}$,
2304: see also https://en.wikipedia.org/wiki/Coefficient_of_determination
2306: Level: advanced
2308: .seealso: [](chapter_ksp), `KSP`, `KSPConvergedRateView()`
2309: */
2310: PetscErrorCode KSPComputeConvergenceRate(KSP ksp, PetscReal *cr, PetscReal *rRsq, PetscReal *ce, PetscReal *eRsq)
2311: {
2312: PetscReal const *hist;
2313: PetscReal *x, *y, slope, intercept, mean = 0.0, var = 0.0, res = 0.0;
2314: PetscInt n, k;
2316: if (cr || rRsq) {
2317: KSPGetResidualHistory(ksp, &hist, &n);
2318: if (!n) {
2319: if (cr) *cr = 0.0;
2320: if (rRsq) *rRsq = -1.0;
2321: } else {
2322: PetscMalloc2(n, &x, n, &y);
2323: for (k = 0; k < n; ++k) {
2324: x[k] = k;
2325: y[k] = PetscLogReal(hist[k]);
2326: mean += y[k];
2327: }
2328: mean /= n;
2329: PetscLinearRegression(n, x, y, &slope, &intercept);
2330: for (k = 0; k < n; ++k) {
2331: res += PetscSqr(y[k] - (slope * x[k] + intercept));
2332: var += PetscSqr(y[k] - mean);
2333: }
2334: PetscFree2(x, y);
2335: if (cr) *cr = PetscExpReal(slope);
2336: if (rRsq) *rRsq = var < PETSC_MACHINE_EPSILON ? 0.0 : 1.0 - (res / var);
2337: }
2338: }
2339: if (ce || eRsq) {
2340: KSPGetErrorHistory(ksp, &hist, &n);
2341: if (!n) {
2342: if (ce) *ce = 0.0;
2343: if (eRsq) *eRsq = -1.0;
2344: } else {
2345: PetscMalloc2(n, &x, n, &y);
2346: for (k = 0; k < n; ++k) {
2347: x[k] = k;
2348: y[k] = PetscLogReal(hist[k]);
2349: mean += y[k];
2350: }
2351: mean /= n;
2352: PetscLinearRegression(n, x, y, &slope, &intercept);
2353: for (k = 0; k < n; ++k) {
2354: res += PetscSqr(y[k] - (slope * x[k] + intercept));
2355: var += PetscSqr(y[k] - mean);
2356: }
2357: PetscFree2(x, y);
2358: if (ce) *ce = PetscExpReal(slope);
2359: if (eRsq) *eRsq = var < PETSC_MACHINE_EPSILON ? 0.0 : 1.0 - (res / var);
2360: }
2361: }
2362: return 0;
2363: }
2365: /*@C
2366: KSPSetConvergenceTest - Sets the function to be used to determine convergence.
2368: Logically Collective
2370: Input Parameters:
2371: + ksp - iterative context obtained from `KSPCreate()`
2372: . converge - pointer to the function
2373: . cctx - context for private data for the convergence routine (may be null)
2374: - destroy - a routine for destroying the context (may be null)
2376: Calling sequence of converge:
2377: $ converge (`KSP` ksp, `PetscInt` it, `PetscReal` rnorm, `KSPConvergedReason` *reason,void *mctx)
2379: + ksp - iterative context obtained from `KSPCreate()`
2380: . it - iteration number
2381: . rnorm - (estimated) 2-norm of (preconditioned) residual
2382: . reason - the reason why it has converged or diverged
2383: - cctx - optional convergence context, as set by `KSPSetConvergenceTest()`
2385: Level: advanced
2387: Notes:
2388: Must be called after the `KSP` type has been set so put this after
2389: a call to `KSPSetType()`, or `KSPSetFromOptions()`.
2391: The default convergence test, `KSPConvergedDefault()`, aborts if the
2392: residual grows to more than 10000 times the initial residual.
2394: The default is a combination of relative and absolute tolerances.
2395: The residual value that is tested may be an approximation; routines
2396: that need exact values should compute them.
2398: In the default PETSc convergence test, the precise values of reason
2399: are macros such as `KSP_CONVERGED_RTOL`, which are defined in petscksp.h.
2401: .seealso: [](chapter_ksp), `KSP`, `KSPConvergedDefault()`, `KSPGetConvergenceContext()`, `KSPSetTolerances()`, `KSP`, `KSPGetConvergenceTest()`, `KSPGetAndClearConvergenceTest()`
2402: @*/
2403: PetscErrorCode KSPSetConvergenceTest(KSP ksp, PetscErrorCode (*converge)(KSP, PetscInt, PetscReal, KSPConvergedReason *, void *), void *cctx, PetscErrorCode (*destroy)(void *))
2404: {
2406: if (ksp->convergeddestroy) (*ksp->convergeddestroy)(ksp->cnvP);
2407: ksp->converged = converge;
2408: ksp->convergeddestroy = destroy;
2409: ksp->cnvP = (void *)cctx;
2410: return 0;
2411: }
2413: /*@C
2414: KSPGetConvergenceTest - Gets the function to be used to determine convergence.
2416: Logically Collective
2418: Input Parameter:
2419: . ksp - iterative context obtained from `KSPCreate()`
2421: Output Parameters:
2422: + converge - pointer to convergence test function
2423: . cctx - context for private data for the convergence routine (may be null)
2424: - destroy - a routine for destroying the context (may be null)
2426: Calling sequence of converge:
2427: $ converge (`KSP` ksp, `PetscInt` it, `PetscReal` rnorm, `KSPConvergedReason` *reason,void *mctx)
2429: + ksp - iterative context obtained from `KSPCreate()`
2430: . it - iteration number
2431: . rnorm - (estimated) 2-norm of (preconditioned) residual
2432: . reason - the reason why it has converged or diverged
2433: - cctx - optional convergence context, as set by `KSPSetConvergenceTest()`
2435: Level: advanced
2437: .seealso: [](chapter_ksp), `KSP`, `KSPConvergedDefault()`, `KSPGetConvergenceContext()`, `KSPSetTolerances()`, `KSP`, `KSPSetConvergenceTest()`, `KSPGetAndClearConvergenceTest()`
2438: @*/
2439: PetscErrorCode KSPGetConvergenceTest(KSP ksp, PetscErrorCode (**converge)(KSP, PetscInt, PetscReal, KSPConvergedReason *, void *), void **cctx, PetscErrorCode (**destroy)(void *))
2440: {
2442: if (converge) *converge = ksp->converged;
2443: if (destroy) *destroy = ksp->convergeddestroy;
2444: if (cctx) *cctx = ksp->cnvP;
2445: return 0;
2446: }
2448: /*@C
2449: KSPGetAndClearConvergenceTest - Gets the function to be used to determine convergence. Removes the current test without calling destroy on the test context
2451: Logically Collective
2453: Input Parameter:
2454: . ksp - iterative context obtained from `KSPCreate()`
2456: Output Parameters:
2457: + converge - pointer to convergence test function
2458: . cctx - context for private data for the convergence routine
2459: - destroy - a routine for destroying the context
2461: Calling sequence of converge:
2462: $ converge (`KSP` ksp, `PetscInt` it, `PetscReal` rnorm, `KSPConvergedReason` *reason,void *mctx)
2464: + ksp - iterative context obtained from `KSPCreate()`
2465: . it - iteration number
2466: . rnorm - (estimated) 2-norm of (preconditioned) residual
2467: . reason - the reason why it has converged or diverged
2468: - cctx - optional convergence context, as set by `KSPSetConvergenceTest()`
2470: Level: advanced
2472: Note:
2473: This is intended to be used to allow transferring the convergence test (and its context) to another testing object (for example another `KSP`) and then calling
2474: `KSPSetConvergenceTest()` on this original `KSP`. If you just called `KSPGetConvergenceTest()` followed by `KSPSetConvergenceTest()` the original context information
2475: would be destroyed and hence the transferred context would be invalid and trigger a crash on use
2477: .seealso: [](chapter_ksp), `KSP`, `KSPConvergedDefault()`, `KSPGetConvergenceContext()`, `KSPSetTolerances()`, `KSP`, `KSPSetConvergenceTest()`, `KSPGetConvergenceTest()`
2478: @*/
2479: PetscErrorCode KSPGetAndClearConvergenceTest(KSP ksp, PetscErrorCode (**converge)(KSP, PetscInt, PetscReal, KSPConvergedReason *, void *), void **cctx, PetscErrorCode (**destroy)(void *))
2480: {
2482: *converge = ksp->converged;
2483: *destroy = ksp->convergeddestroy;
2484: *cctx = ksp->cnvP;
2485: ksp->converged = NULL;
2486: ksp->cnvP = NULL;
2487: ksp->convergeddestroy = NULL;
2488: return 0;
2489: }
2491: /*@C
2492: KSPGetConvergenceContext - Gets the convergence context set with `KSPSetConvergenceTest()`.
2494: Not Collective
2496: Input Parameter:
2497: . ksp - iterative context obtained from `KSPCreate()`
2499: Output Parameter:
2500: . ctx - monitoring context
2502: Level: advanced
2504: .seealso: [](chapter_ksp), `KSP`, `KSPConvergedDefault()`, `KSPSetConvergenceTest()`, `KSPGetConvergenceTest()`
2505: @*/
2506: PetscErrorCode KSPGetConvergenceContext(KSP ksp, void *ctx)
2507: {
2509: *(void **)ctx = ksp->cnvP;
2510: return 0;
2511: }
2513: /*@C
2514: KSPBuildSolution - Builds the approximate solution in a vector provided.
2516: Collective
2518: Input Parameter:
2519: . ctx - iterative context obtained from `KSPCreate()`
2521: Output Parameter:
2522: Provide exactly one of
2523: + v - location to stash solution.
2524: - V - the solution is returned in this location. This vector is created
2525: internally. This vector should NOT be destroyed by the user with
2526: `VecDestroy()`.
2528: Notes:
2529: This routine can be used in one of two ways
2530: .vb
2531: KSPBuildSolution(ksp,NULL,&V);
2532: or
2533: KSPBuildSolution(ksp,v,NULL); or KSPBuildSolution(ksp,v,&v);
2534: .ve
2535: In the first case an internal vector is allocated to store the solution
2536: (the user cannot destroy this vector). In the second case the solution
2537: is generated in the vector that the user provides. Note that for certain
2538: methods, such as `KSPCG`, the second case requires a copy of the solution,
2539: while in the first case the call is essentially free since it simply
2540: returns the vector where the solution already is stored. For some methods
2541: like `KSPGMRES` this is a reasonably expensive operation and should only be
2542: used in truly needed.
2544: Level: developer
2546: .seealso: [](chapter_ksp), `KSPGetSolution()`, `KSPBuildResidual()`, `KSP`
2547: @*/
2548: PetscErrorCode KSPBuildSolution(KSP ksp, Vec v, Vec *V)
2549: {
2552: if (!V) V = &v;
2553: PetscUseTypeMethod(ksp, buildsolution, v, V);
2554: return 0;
2555: }
2557: /*@C
2558: KSPBuildResidual - Builds the residual in a vector provided.
2560: Collective
2562: Input Parameter:
2563: . ksp - iterative context obtained from `KSPCreate()`
2565: Output Parameters:
2566: + v - optional location to stash residual. If v is not provided,
2567: then a location is generated.
2568: . t - work vector. If not provided then one is generated.
2569: - V - the residual
2571: Note:
2572: Regardless of whether or not v is provided, the residual is
2573: returned in V.
2575: Level: advanced
2577: .seealso: [](chapter_ksp), `KSP`, `KSPBuildSolution()`
2578: @*/
2579: PetscErrorCode KSPBuildResidual(KSP ksp, Vec t, Vec v, Vec *V)
2580: {
2581: PetscBool flag = PETSC_FALSE;
2582: Vec w = v, tt = t;
2585: if (!w) { VecDuplicate(ksp->vec_rhs, &w); }
2586: if (!tt) {
2587: VecDuplicate(ksp->vec_sol, &tt);
2588: flag = PETSC_TRUE;
2589: }
2590: PetscUseTypeMethod(ksp, buildresidual, tt, w, V);
2591: if (flag) VecDestroy(&tt);
2592: return 0;
2593: }
2595: /*@
2596: KSPSetDiagonalScale - Tells `KSP` to symmetrically diagonally scale the system
2597: before solving. This actually CHANGES the matrix (and right hand side).
2599: Logically Collective
2601: Input Parameters:
2602: + ksp - the `KSP` context
2603: - scale - `PETSC_TRUE` or `PETSC_FALSE`
2605: Options Database Key:
2606: + -ksp_diagonal_scale -
2607: - -ksp_diagonal_scale_fix - scale the matrix back AFTER the solve
2609: Level: advanced
2611: Notes:
2612: Scales the matrix by D^(-1/2) A D^(-1/2) [D^(1/2) x ] = D^(-1/2) b
2613: where D_{ii} is 1/abs(A_{ii}) unless A_{ii} is zero and then it is 1.
2615: BE CAREFUL with this routine: it actually scales the matrix and right
2616: hand side that define the system. After the system is solved the matrix
2617: and right hand side remain scaled unless you use `KSPSetDiagonalScaleFix()`
2619: This should NOT be used within the `SNES` solves if you are using a line
2620: search.
2622: If you use this with the `PCType` `PCEISENSTAT` preconditioner than you can
2623: use the `PCEisenstatSetNoDiagonalScaling()` option, or -pc_eisenstat_no_diagonal_scaling
2624: to save some unneeded, redundant flops.
2626: .seealso: [](chapter_ksp), `KSPGetDiagonalScale()`, `KSPSetDiagonalScaleFix()`, `KSP`
2627: @*/
2628: PetscErrorCode KSPSetDiagonalScale(KSP ksp, PetscBool scale)
2629: {
2632: ksp->dscale = scale;
2633: return 0;
2634: }
2636: /*@
2637: KSPGetDiagonalScale - Checks if `KSP` solver scales the matrix and right hand side, that is if `KSPSetDiagonalScale()` has been called
2639: Not Collective
2641: Input Parameter:
2642: . ksp - the `KSP` context
2644: Output Parameter:
2645: . scale - `PETSC_TRUE` or `PETSC_FALSE`
2647: Level: intermediate
2649: .seealso: [](chapter_ksp), `KSP`, `KSPSetDiagonalScale()`, `KSPSetDiagonalScaleFix()`, `KSP`
2650: @*/
2651: PetscErrorCode KSPGetDiagonalScale(KSP ksp, PetscBool *scale)
2652: {
2655: *scale = ksp->dscale;
2656: return 0;
2657: }
2659: /*@
2660: KSPSetDiagonalScaleFix - Tells `KSP` to diagonally scale the system back after solving.
2662: Logically Collective
2664: Input Parameters:
2665: + ksp - the `KSP` context
2666: - fix - `PETSC_TRUE` to scale back after the system solve, `PETSC_FALSE` to not
2667: rescale (default)
2669: Notes:
2670: Must be called after `KSPSetDiagonalScale()`
2672: Using this will slow things down, because it rescales the matrix before and
2673: after each linear solve. This is intended mainly for testing to allow one
2674: to easily get back the original system to make sure the solution computed is
2675: accurate enough.
2677: Level: intermediate
2679: .seealso: [](chapter_ksp), `KSPGetDiagonalScale()`, `KSPSetDiagonalScale()`, `KSPGetDiagonalScaleFix()`, `KSP`
2680: @*/
2681: PetscErrorCode KSPSetDiagonalScaleFix(KSP ksp, PetscBool fix)
2682: {
2685: ksp->dscalefix = fix;
2686: return 0;
2687: }
2689: /*@
2690: KSPGetDiagonalScaleFix - Determines if `KSP` diagonally scales the system back after solving. That is `KSPSetDiagonalScaleFix()` has been called
2692: Not Collective
2694: Input Parameter:
2695: . ksp - the `KSP` context
2697: Output Parameter:
2698: . fix - `PETSC_TRUE` to scale back after the system solve, `PETSC_FALSE` to not
2699: rescale (default)
2701: Level: intermediate
2703: .seealso: [](chapter_ksp), `KSPGetDiagonalScale()`, `KSPSetDiagonalScale()`, `KSPSetDiagonalScaleFix()`, `KSP`
2704: @*/
2705: PetscErrorCode KSPGetDiagonalScaleFix(KSP ksp, PetscBool *fix)
2706: {
2709: *fix = ksp->dscalefix;
2710: return 0;
2711: }
2713: /*@C
2714: KSPSetComputeOperators - set routine to compute the linear operators
2716: Logically Collective
2718: Input Parameters:
2719: + ksp - the `KSP` context
2720: . func - function to compute the operators
2721: - ctx - optional context
2723: Calling sequence of func:
2724: $ func(KSP ksp,Mat A,Mat B,void *ctx)
2726: + ksp - the `KSP` context
2727: . A - the linear operator
2728: . B - preconditioning matrix
2729: - ctx - optional user-provided context
2731: Level: beginner
2733: Notes:
2734: The user provided func() will be called automatically at the very next call to `KSPSolve()`. It will NOT be called at future `KSPSolve()` calls
2735: unless either `KSPSetComputeOperators()` or `KSPSetOperators()` is called before that `KSPSolve()` is called. This allows the same system to be solved several times
2736: with different right hand side functions but is a confusing API since one might expect it to be called for each `KSPSolve()`
2738: To reuse the same preconditioner for the next `KSPSolve()` and not compute a new one based on the most recently computed matrix call `KSPSetReusePreconditioner()`
2740: Developer Note:
2741: Perhaps this routine and `KSPSetComputeRHS()` could be combined into a new API that makes clear when new matrices are computing without requiring call this
2742: routine to indicate when the new matrix should be computed.
2744: .seealso: [](chapter_ksp), `KSP`, `KSPSetOperators()`, `KSPSetComputeRHS()`, `DMKSPSetComputeOperators()`, `KSPSetComputeInitialGuess()`
2745: @*/
2746: PetscErrorCode KSPSetComputeOperators(KSP ksp, PetscErrorCode (*func)(KSP, Mat, Mat, void *), void *ctx)
2747: {
2748: DM dm;
2751: KSPGetDM(ksp, &dm);
2752: DMKSPSetComputeOperators(dm, func, ctx);
2753: if (ksp->setupstage == KSP_SETUP_NEWRHS) ksp->setupstage = KSP_SETUP_NEWMATRIX;
2754: return 0;
2755: }
2757: /*@C
2758: KSPSetComputeRHS - set routine to compute the right hand side of the linear system
2760: Logically Collective
2762: Input Parameters:
2763: + ksp - the `KSP` context
2764: . func - function to compute the right hand side
2765: - ctx - optional context
2767: Calling sequence of func:
2768: $ func(KSP ksp,Vec b,void *ctx)
2770: + ksp - the `KSP` context
2771: . b - right hand side of linear system
2772: - ctx - optional user-provided context
2774: Notes:
2775: The routine you provide will be called EACH you call `KSPSolve()` to prepare the new right hand side for that solve
2777: Level: beginner
2779: .seealso: [](chapter_ksp), `KSP`, `KSPSolve()`, `DMKSPSetComputeRHS()`, `KSPSetComputeOperators()`, `KSPSetOperators()`
2780: @*/
2781: PetscErrorCode KSPSetComputeRHS(KSP ksp, PetscErrorCode (*func)(KSP, Vec, void *), void *ctx)
2782: {
2783: DM dm;
2786: KSPGetDM(ksp, &dm);
2787: DMKSPSetComputeRHS(dm, func, ctx);
2788: return 0;
2789: }
2791: /*@C
2792: KSPSetComputeInitialGuess - set routine to compute the initial guess of the linear system
2794: Logically Collective
2796: Input Parameters:
2797: + ksp - the `KSP` context
2798: . func - function to compute the initial guess
2799: - ctx - optional context
2801: Calling sequence of func:
2802: $ func(KSP ksp,Vec x,void *ctx)
2804: + ksp - the `KSP` context
2805: . x - solution vector
2806: - ctx - optional user-provided context
2808: Notes:
2809: This should only be used in conjunction with `KSPSetComputeRHS()` and `KSPSetComputeOperators()`, otherwise
2810: call `KSPSetInitialGuessNonzero()` and set the initial guess values in the solution vector passed to `KSPSolve()` before calling the solver
2812: Level: beginner
2814: .seealso: [](chapter_ksp), `KSP`, `KSPSolve()`, `KSPSetComputeRHS()`, `KSPSetComputeOperators()`, `DMKSPSetComputeInitialGuess()`, `KSPSetInitialGuessNonzero()`
2815: @*/
2816: PetscErrorCode KSPSetComputeInitialGuess(KSP ksp, PetscErrorCode (*func)(KSP, Vec, void *), void *ctx)
2817: {
2818: DM dm;
2821: KSPGetDM(ksp, &dm);
2822: DMKSPSetComputeInitialGuess(dm, func, ctx);
2823: return 0;
2824: }
2826: /*@
2827: KSPSetUseExplicitTranspose - Determines the explicit transpose of the operator is formed in `KSPSolveTranspose()`. In some configurations (like GPUs) it may
2828: be explicitly formed when possible since the solve is much more efficient.
2830: Logically Collective
2832: Input Parameter:
2833: . ksp - the `KSP` context
2835: Output Parameter:
2836: . flg - `PETSC_TRUE` to transpose the system in `KSPSolveTranspose()`, `PETSC_FALSE` to not transpose (default)
2838: Level: advanced
2840: .seealso: [](chapter_ksp), `KSPSolveTranspose()`, `KSP`
2841: @*/
2842: PetscErrorCode KSPSetUseExplicitTranspose(KSP ksp, PetscBool flg)
2843: {
2846: ksp->transpose.use_explicittranspose = flg;
2847: return 0;
2848: }