Actual source code: taosolver.c
1: #include <petsc/private/taoimpl.h>
2: #include <petsc/private/snesimpl.h>
4: PetscBool TaoRegisterAllCalled = PETSC_FALSE;
5: PetscFunctionList TaoList = NULL;
7: PetscClassId TAO_CLASSID;
9: PetscLogEvent TAO_Solve;
10: PetscLogEvent TAO_ObjectiveEval;
11: PetscLogEvent TAO_GradientEval;
12: PetscLogEvent TAO_ObjGradEval;
13: PetscLogEvent TAO_HessianEval;
14: PetscLogEvent TAO_JacobianEval;
15: PetscLogEvent TAO_ConstraintsEval;
17: const char *TaoSubSetTypes[] = {"subvec", "mask", "matrixfree", "TaoSubSetType", "TAO_SUBSET_", NULL};
19: struct _n_TaoMonitorDrawCtx {
20: PetscViewer viewer;
21: PetscInt howoften; /* when > 0 uses iteration % howoften, when negative only final solution plotted */
22: };
24: static PetscErrorCode KSPPreSolve_TAOEW_Private(KSP ksp, Vec b, Vec x, Tao tao)
25: {
26: SNES snes_ewdummy = tao->snes_ewdummy;
28: PetscFunctionBegin;
29: if (!snes_ewdummy) PetscFunctionReturn(PETSC_SUCCESS);
30: /* populate snes_ewdummy struct values used in KSPPreSolve_SNESEW */
31: snes_ewdummy->vec_func = b;
32: snes_ewdummy->rtol = tao->gttol;
33: snes_ewdummy->iter = tao->niter;
34: PetscCall(VecNorm(b, NORM_2, &snes_ewdummy->norm));
35: PetscCall(KSPPreSolve_SNESEW(ksp, b, x, snes_ewdummy));
36: snes_ewdummy->vec_func = NULL;
37: PetscFunctionReturn(PETSC_SUCCESS);
38: }
40: static PetscErrorCode KSPPostSolve_TAOEW_Private(KSP ksp, Vec b, Vec x, Tao tao)
41: {
42: SNES snes_ewdummy = tao->snes_ewdummy;
44: PetscFunctionBegin;
45: if (!snes_ewdummy) PetscFunctionReturn(PETSC_SUCCESS);
46: PetscCall(KSPPostSolve_SNESEW(ksp, b, x, snes_ewdummy));
47: PetscFunctionReturn(PETSC_SUCCESS);
48: }
50: static PetscErrorCode TaoSetUpEW_Private(Tao tao)
51: {
52: SNESKSPEW *kctx;
53: const char *ewprefix;
55: PetscFunctionBegin;
56: if (!tao->ksp) PetscFunctionReturn(PETSC_SUCCESS);
57: if (tao->ksp_ewconv) {
58: if (!tao->snes_ewdummy) PetscCall(SNESCreate(PetscObjectComm((PetscObject)tao), &tao->snes_ewdummy));
59: tao->snes_ewdummy->ksp_ewconv = PETSC_TRUE;
60: PetscCall(KSPSetPreSolve(tao->ksp, (PetscErrorCode(*)(KSP, Vec, Vec, void *))KSPPreSolve_TAOEW_Private, tao));
61: PetscCall(KSPSetPostSolve(tao->ksp, (PetscErrorCode(*)(KSP, Vec, Vec, void *))KSPPostSolve_TAOEW_Private, tao));
63: PetscCall(KSPGetOptionsPrefix(tao->ksp, &ewprefix));
64: kctx = (SNESKSPEW *)tao->snes_ewdummy->kspconvctx;
65: PetscCall(SNESEWSetFromOptions_Private(kctx, PETSC_FALSE, PetscObjectComm((PetscObject)tao), ewprefix));
66: } else PetscCall(SNESDestroy(&tao->snes_ewdummy));
67: PetscFunctionReturn(PETSC_SUCCESS);
68: }
70: /*@
71: TaoCreate - Creates a Tao solver
73: Collective
75: Input Parameter:
76: . comm - MPI communicator
78: Output Parameter:
79: . newtao - the new `Tao` context
81: Options Database Key:
82: . -tao_type - select which method Tao should use
84: Level: beginner
86: .seealso: [](ch_tao), `Tao`, `TaoSolve()`, `TaoDestroy()`, `TAOSetFromOptions()`, `TAOSetType()`
87: @*/
88: PetscErrorCode TaoCreate(MPI_Comm comm, Tao *newtao)
89: {
90: Tao tao;
92: PetscFunctionBegin;
93: PetscAssertPointer(newtao, 2);
94: PetscCall(TaoInitializePackage());
95: PetscCall(TaoLineSearchInitializePackage());
96: PetscCall(PetscHeaderCreate(tao, TAO_CLASSID, "Tao", "Optimization solver", "Tao", comm, TaoDestroy, TaoView));
98: /* Set non-NULL defaults */
99: tao->ops->convergencetest = TaoDefaultConvergenceTest;
101: tao->max_it = 10000;
102: tao->max_funcs = -1;
103: #if defined(PETSC_USE_REAL_SINGLE)
104: tao->gatol = 1e-5;
105: tao->grtol = 1e-5;
106: tao->crtol = 1e-5;
107: tao->catol = 1e-5;
108: #else
109: tao->gatol = 1e-8;
110: tao->grtol = 1e-8;
111: tao->crtol = 1e-8;
112: tao->catol = 1e-8;
113: #endif
114: tao->gttol = 0.0;
115: tao->steptol = 0.0;
116: tao->trust0 = PETSC_INFINITY;
117: tao->fmin = PETSC_NINFINITY;
119: tao->hist_reset = PETSC_TRUE;
121: PetscCall(TaoResetStatistics(tao));
122: *newtao = tao;
123: PetscFunctionReturn(PETSC_SUCCESS);
124: }
126: /*@
127: TaoSolve - Solves an optimization problem min F(x) s.t. l <= x <= u
129: Collective
131: Input Parameter:
132: . tao - the `Tao` context
134: Level: beginner
136: Notes:
137: The user must set up the `Tao` object with calls to `TaoSetSolution()`, `TaoSetObjective()`, `TaoSetGradient()`, and (if using 2nd order method) `TaoSetHessian()`.
139: You should call `TaoGetConvergedReason()` or run with `-tao_converged_reason` to determine if the optimization algorithm actually succeeded or
140: why it failed.
142: .seealso: [](ch_tao), `Tao`, `TaoCreate()`, `TaoSetObjective()`, `TaoSetGradient()`, `TaoSetHessian()`, `TaoGetConvergedReason()`, `TaoSetUp()`
143: @*/
144: PetscErrorCode TaoSolve(Tao tao)
145: {
146: static PetscBool set = PETSC_FALSE;
148: PetscFunctionBegin;
150: PetscCall(PetscCitationsRegister("@TechReport{tao-user-ref,\n"
151: "title = {Toolkit for Advanced Optimization (TAO) Users Manual},\n"
152: "author = {Todd Munson and Jason Sarich and Stefan Wild and Steve Benson and Lois Curfman McInnes},\n"
153: "Institution = {Argonne National Laboratory},\n"
154: "Year = 2014,\n"
155: "Number = {ANL/MCS-TM-322 - Revision 3.5},\n"
156: "url = {https://www.mcs.anl.gov/research/projects/tao/}\n}\n",
157: &set));
158: tao->header_printed = PETSC_FALSE;
159: PetscCall(TaoSetUp(tao));
160: PetscCall(TaoResetStatistics(tao));
161: if (tao->linesearch) PetscCall(TaoLineSearchReset(tao->linesearch));
163: PetscCall(PetscLogEventBegin(TAO_Solve, tao, 0, 0, 0));
164: PetscTryTypeMethod(tao, solve);
165: PetscCall(PetscLogEventEnd(TAO_Solve, tao, 0, 0, 0));
167: PetscCall(VecViewFromOptions(tao->solution, (PetscObject)tao, "-tao_view_solution"));
169: tao->ntotalits += tao->niter;
171: if (tao->printreason) {
172: PetscViewer viewer = PETSC_VIEWER_STDOUT_(((PetscObject)tao)->comm);
173: PetscCall(PetscViewerASCIIAddTab(viewer, ((PetscObject)tao)->tablevel));
174: if (tao->reason > 0) {
175: PetscCall(PetscViewerASCIIPrintf(viewer, " TAO %s solve converged due to %s iterations %" PetscInt_FMT "\n", ((PetscObject)tao)->prefix ? ((PetscObject)tao)->prefix : "", TaoConvergedReasons[tao->reason], tao->niter));
176: } else {
177: PetscCall(PetscViewerASCIIPrintf(viewer, " TAO %s solve did not converge due to %s iteration %" PetscInt_FMT "\n", ((PetscObject)tao)->prefix ? ((PetscObject)tao)->prefix : "", TaoConvergedReasons[tao->reason], tao->niter));
178: }
179: PetscCall(PetscViewerASCIISubtractTab(viewer, ((PetscObject)tao)->tablevel));
180: }
181: PetscCall(TaoViewFromOptions(tao, NULL, "-tao_view"));
182: PetscFunctionReturn(PETSC_SUCCESS);
183: }
185: /*@
186: TaoSetUp - Sets up the internal data structures for the later use
187: of a Tao solver
189: Collective
191: Input Parameter:
192: . tao - the `Tao` context
194: Level: advanced
196: Note:
197: The user will not need to explicitly call `TaoSetUp()`, as it will
198: automatically be called in `TaoSolve()`. However, if the user
199: desires to call it explicitly, it should come after `TaoCreate()`
200: and any TaoSetSomething() routines, but before `TaoSolve()`.
202: .seealso: [](ch_tao), `Tao`, `TaoCreate()`, `TaoSolve()`
203: @*/
204: PetscErrorCode TaoSetUp(Tao tao)
205: {
206: PetscFunctionBegin;
208: if (tao->setupcalled) PetscFunctionReturn(PETSC_SUCCESS);
209: PetscCall(TaoSetUpEW_Private(tao));
210: PetscCheck(tao->solution, PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONGSTATE, "Must call TaoSetSolution");
211: PetscTryTypeMethod(tao, setup);
212: tao->setupcalled = PETSC_TRUE;
213: PetscFunctionReturn(PETSC_SUCCESS);
214: }
216: /*@C
217: TaoDestroy - Destroys the `Tao` context that was created with `TaoCreate()`
219: Collective
221: Input Parameter:
222: . tao - the `Tao` context
224: Level: beginner
226: .seealso: [](ch_tao), `Tao`, `TaoCreate()`, `TaoSolve()`
227: @*/
228: PetscErrorCode TaoDestroy(Tao *tao)
229: {
230: PetscFunctionBegin;
231: if (!*tao) PetscFunctionReturn(PETSC_SUCCESS);
233: if (--((PetscObject)*tao)->refct > 0) {
234: *tao = NULL;
235: PetscFunctionReturn(PETSC_SUCCESS);
236: }
238: if ((*tao)->ops->destroy) PetscCall((*((*tao))->ops->destroy)(*tao));
239: PetscCall(KSPDestroy(&(*tao)->ksp));
240: PetscCall(SNESDestroy(&(*tao)->snes_ewdummy));
241: PetscCall(TaoLineSearchDestroy(&(*tao)->linesearch));
243: if ((*tao)->ops->convergencedestroy) {
244: PetscCall((*(*tao)->ops->convergencedestroy)((*tao)->cnvP));
245: if ((*tao)->jacobian_state_inv) PetscCall(MatDestroy(&(*tao)->jacobian_state_inv));
246: }
247: PetscCall(VecDestroy(&(*tao)->solution));
248: PetscCall(VecDestroy(&(*tao)->gradient));
249: PetscCall(VecDestroy(&(*tao)->ls_res));
251: if ((*tao)->gradient_norm) {
252: PetscCall(PetscObjectDereference((PetscObject)(*tao)->gradient_norm));
253: PetscCall(VecDestroy(&(*tao)->gradient_norm_tmp));
254: }
256: PetscCall(VecDestroy(&(*tao)->XL));
257: PetscCall(VecDestroy(&(*tao)->XU));
258: PetscCall(VecDestroy(&(*tao)->IL));
259: PetscCall(VecDestroy(&(*tao)->IU));
260: PetscCall(VecDestroy(&(*tao)->DE));
261: PetscCall(VecDestroy(&(*tao)->DI));
262: PetscCall(VecDestroy(&(*tao)->constraints));
263: PetscCall(VecDestroy(&(*tao)->constraints_equality));
264: PetscCall(VecDestroy(&(*tao)->constraints_inequality));
265: PetscCall(VecDestroy(&(*tao)->stepdirection));
266: PetscCall(MatDestroy(&(*tao)->hessian_pre));
267: PetscCall(MatDestroy(&(*tao)->hessian));
268: PetscCall(MatDestroy(&(*tao)->ls_jac));
269: PetscCall(MatDestroy(&(*tao)->ls_jac_pre));
270: PetscCall(MatDestroy(&(*tao)->jacobian_pre));
271: PetscCall(MatDestroy(&(*tao)->jacobian));
272: PetscCall(MatDestroy(&(*tao)->jacobian_state_pre));
273: PetscCall(MatDestroy(&(*tao)->jacobian_state));
274: PetscCall(MatDestroy(&(*tao)->jacobian_state_inv));
275: PetscCall(MatDestroy(&(*tao)->jacobian_design));
276: PetscCall(MatDestroy(&(*tao)->jacobian_equality));
277: PetscCall(MatDestroy(&(*tao)->jacobian_equality_pre));
278: PetscCall(MatDestroy(&(*tao)->jacobian_inequality));
279: PetscCall(MatDestroy(&(*tao)->jacobian_inequality_pre));
280: PetscCall(ISDestroy(&(*tao)->state_is));
281: PetscCall(ISDestroy(&(*tao)->design_is));
282: PetscCall(VecDestroy(&(*tao)->res_weights_v));
283: PetscCall(TaoCancelMonitors(*tao));
284: if ((*tao)->hist_malloc) PetscCall(PetscFree4((*tao)->hist_obj, (*tao)->hist_resid, (*tao)->hist_cnorm, (*tao)->hist_lits));
285: if ((*tao)->res_weights_n) {
286: PetscCall(PetscFree((*tao)->res_weights_rows));
287: PetscCall(PetscFree((*tao)->res_weights_cols));
288: PetscCall(PetscFree((*tao)->res_weights_w));
289: }
290: PetscCall(PetscHeaderDestroy(tao));
291: PetscFunctionReturn(PETSC_SUCCESS);
292: }
294: /*@
295: TaoKSPSetUseEW - Sets `SNES` to use Eisenstat-Walker method {cite}`ew96`for computing relative tolerance for linear solvers.
297: Logically Collective
299: Input Parameters:
300: + tao - Tao context
301: - flag - `PETSC_TRUE` or `PETSC_FALSE`
303: Level: advanced
305: Note:
306: See `SNESKSPSetUseEW()` for customization details.
308: .seealso: [](ch_tao), `Tao`, `SNESKSPSetUseEW()`
309: @*/
310: PetscErrorCode TaoKSPSetUseEW(Tao tao, PetscBool flag)
311: {
312: PetscFunctionBegin;
315: tao->ksp_ewconv = flag;
316: PetscFunctionReturn(PETSC_SUCCESS);
317: }
319: /*@
320: TaoSetFromOptions - Sets various Tao parameters from the options database
322: Collective
324: Input Parameter:
325: . tao - the `Tao` solver context
327: Options Database Keys:
328: + -tao_type <type> - The algorithm that Tao uses (lmvm, nls, etc.)
329: . -tao_gatol <gatol> - absolute error tolerance for ||gradient||
330: . -tao_grtol <grtol> - relative error tolerance for ||gradient||
331: . -tao_gttol <gttol> - reduction of ||gradient|| relative to initial gradient
332: . -tao_max_it <max> - sets maximum number of iterations
333: . -tao_max_funcs <max> - sets maximum number of function evaluations
334: . -tao_fmin <fmin> - stop if function value reaches fmin
335: . -tao_steptol <tol> - stop if trust region radius less than <tol>
336: . -tao_trust0 <t> - initial trust region radius
337: . -tao_monitor - prints function value and residual norm at each iteration
338: . -tao_smonitor - same as tao_monitor, but truncates very small values
339: . -tao_cmonitor - prints function value, residual, and constraint norm at each iteration
340: . -tao_view_solution - prints solution vector at each iteration
341: . -tao_view_ls_residual - prints least-squares residual vector at each iteration
342: . -tao_view_stepdirection - prints step direction vector at each iteration
343: . -tao_view_gradient - prints gradient vector at each iteration
344: . -tao_draw_solution - graphically view solution vector at each iteration
345: . -tao_draw_step - graphically view step vector at each iteration
346: . -tao_draw_gradient - graphically view gradient at each iteration
347: . -tao_fd_gradient - use gradient computed with finite differences
348: . -tao_fd_hessian - use hessian computed with finite differences
349: . -tao_mf_hessian - use matrix-free hessian computed with finite differences
350: . -tao_cancelmonitors - cancels all monitors (except those set with command line)
351: . -tao_view - prints information about the Tao after solving
352: - -tao_converged_reason - prints the reason Tao stopped iterating
354: Level: beginner
356: Note:
357: To see all options, run your program with the `-help` option or consult the
358: user's manual. Should be called after `TaoCreate()` but before `TaoSolve()`
360: .seealso: [](ch_tao), `Tao`, `TaoCreate()`, `TaoSolve()`
361: @*/
362: PetscErrorCode TaoSetFromOptions(Tao tao)
363: {
364: TaoType default_type = TAOLMVM;
365: char type[256], monfilename[PETSC_MAX_PATH_LEN];
366: PetscViewer monviewer;
367: PetscBool flg, found;
368: MPI_Comm comm;
370: PetscFunctionBegin;
372: PetscCall(PetscObjectGetComm((PetscObject)tao, &comm));
374: if (((PetscObject)tao)->type_name) default_type = ((PetscObject)tao)->type_name;
376: PetscObjectOptionsBegin((PetscObject)tao);
377: /* Check for type from options */
378: PetscCall(PetscOptionsFList("-tao_type", "Tao Solver type", "TaoSetType", TaoList, default_type, type, 256, &flg));
379: if (flg) {
380: PetscCall(TaoSetType(tao, type));
381: } else if (!((PetscObject)tao)->type_name) {
382: PetscCall(TaoSetType(tao, default_type));
383: }
385: /* Tao solvers do not set the prefix, set it here if not yet done
386: We do it after SetType since solver may have been changed */
387: if (tao->linesearch) {
388: const char *prefix;
389: PetscCall(TaoLineSearchGetOptionsPrefix(tao->linesearch, &prefix));
390: if (!prefix) PetscCall(TaoLineSearchSetOptionsPrefix(tao->linesearch, ((PetscObject)(tao))->prefix));
391: }
393: PetscCall(PetscOptionsReal("-tao_catol", "Stop if constraints violations within", "TaoSetConstraintTolerances", tao->catol, &tao->catol, &flg));
394: if (flg) tao->catol_changed = PETSC_TRUE;
395: PetscCall(PetscOptionsReal("-tao_crtol", "Stop if relative constraint violations within", "TaoSetConstraintTolerances", tao->crtol, &tao->crtol, &flg));
396: if (flg) tao->crtol_changed = PETSC_TRUE;
397: PetscCall(PetscOptionsReal("-tao_gatol", "Stop if norm of gradient less than", "TaoSetTolerances", tao->gatol, &tao->gatol, &flg));
398: if (flg) tao->gatol_changed = PETSC_TRUE;
399: PetscCall(PetscOptionsReal("-tao_grtol", "Stop if norm of gradient divided by the function value is less than", "TaoSetTolerances", tao->grtol, &tao->grtol, &flg));
400: if (flg) tao->grtol_changed = PETSC_TRUE;
401: PetscCall(PetscOptionsReal("-tao_gttol", "Stop if the norm of the gradient is less than the norm of the initial gradient times tol", "TaoSetTolerances", tao->gttol, &tao->gttol, &flg));
402: if (flg) tao->gttol_changed = PETSC_TRUE;
403: PetscCall(PetscOptionsInt("-tao_max_it", "Stop if iteration number exceeds", "TaoSetMaximumIterations", tao->max_it, &tao->max_it, &flg));
404: if (flg) tao->max_it_changed = PETSC_TRUE;
405: PetscCall(PetscOptionsInt("-tao_max_funcs", "Stop if number of function evaluations exceeds", "TaoSetMaximumFunctionEvaluations", tao->max_funcs, &tao->max_funcs, &flg));
406: if (flg) tao->max_funcs_changed = PETSC_TRUE;
407: PetscCall(PetscOptionsReal("-tao_fmin", "Stop if function less than", "TaoSetFunctionLowerBound", tao->fmin, &tao->fmin, &flg));
408: if (flg) tao->fmin_changed = PETSC_TRUE;
409: PetscCall(PetscOptionsReal("-tao_steptol", "Stop if step size or trust region radius less than", "", tao->steptol, &tao->steptol, &flg));
410: if (flg) tao->steptol_changed = PETSC_TRUE;
411: PetscCall(PetscOptionsReal("-tao_trust0", "Initial trust region radius", "TaoSetTrustRegionRadius", tao->trust0, &tao->trust0, &flg));
412: if (flg) tao->trust0_changed = PETSC_TRUE;
413: PetscCall(PetscOptionsString("-tao_view_solution", "view solution vector after each evaluation", "TaoSetMonitor", "stdout", monfilename, sizeof(monfilename), &flg));
414: if (flg) {
415: PetscCall(PetscViewerASCIIOpen(comm, monfilename, &monviewer));
416: PetscCall(TaoSetMonitor(tao, TaoSolutionMonitor, monviewer, (PetscErrorCode(*)(void **))PetscViewerDestroy));
417: }
419: PetscCall(PetscOptionsBool("-tao_converged_reason", "Print reason for Tao converged", "TaoSolve", tao->printreason, &tao->printreason, NULL));
420: PetscCall(PetscOptionsString("-tao_view_gradient", "view gradient vector after each evaluation", "TaoSetMonitor", "stdout", monfilename, sizeof(monfilename), &flg));
421: if (flg) {
422: PetscCall(PetscViewerASCIIOpen(comm, monfilename, &monviewer));
423: PetscCall(TaoSetMonitor(tao, TaoGradientMonitor, monviewer, (PetscErrorCode(*)(void **))PetscViewerDestroy));
424: }
426: PetscCall(PetscOptionsString("-tao_view_stepdirection", "view step direction vector after each iteration", "TaoSetMonitor", "stdout", monfilename, sizeof(monfilename), &flg));
427: if (flg) {
428: PetscCall(PetscViewerASCIIOpen(comm, monfilename, &monviewer));
429: PetscCall(TaoSetMonitor(tao, TaoStepDirectionMonitor, monviewer, (PetscErrorCode(*)(void **))PetscViewerDestroy));
430: }
432: PetscCall(PetscOptionsString("-tao_view_residual", "view least-squares residual vector after each evaluation", "TaoSetMonitor", "stdout", monfilename, sizeof(monfilename), &flg));
433: if (flg) {
434: PetscCall(PetscViewerASCIIOpen(comm, monfilename, &monviewer));
435: PetscCall(TaoSetMonitor(tao, TaoResidualMonitor, monviewer, (PetscErrorCode(*)(void **))PetscViewerDestroy));
436: }
438: PetscCall(PetscOptionsString("-tao_monitor", "Use the default convergence monitor", "TaoSetMonitor", "stdout", monfilename, sizeof(monfilename), &flg));
439: if (flg) {
440: PetscCall(PetscViewerASCIIOpen(comm, monfilename, &monviewer));
441: PetscCall(TaoSetMonitor(tao, TaoMonitorDefault, monviewer, (PetscErrorCode(*)(void **))PetscViewerDestroy));
442: }
444: PetscCall(PetscOptionsString("-tao_gmonitor", "Use the convergence monitor with extra globalization info", "TaoSetMonitor", "stdout", monfilename, sizeof(monfilename), &flg));
445: if (flg) {
446: PetscCall(PetscViewerASCIIOpen(comm, monfilename, &monviewer));
447: PetscCall(TaoSetMonitor(tao, TaoDefaultGMonitor, monviewer, (PetscErrorCode(*)(void **))PetscViewerDestroy));
448: }
450: PetscCall(PetscOptionsString("-tao_smonitor", "Use the short convergence monitor", "TaoSetMonitor", "stdout", monfilename, sizeof(monfilename), &flg));
451: if (flg) {
452: PetscCall(PetscViewerASCIIOpen(comm, monfilename, &monviewer));
453: PetscCall(TaoSetMonitor(tao, TaoDefaultSMonitor, monviewer, (PetscErrorCode(*)(void **))PetscViewerDestroy));
454: }
456: PetscCall(PetscOptionsString("-tao_cmonitor", "Use the default convergence monitor with constraint norm", "TaoSetMonitor", "stdout", monfilename, sizeof(monfilename), &flg));
457: if (flg) {
458: PetscCall(PetscViewerASCIIOpen(comm, monfilename, &monviewer));
459: PetscCall(TaoSetMonitor(tao, TaoDefaultCMonitor, monviewer, (PetscErrorCode(*)(void **))PetscViewerDestroy));
460: }
462: flg = PETSC_FALSE;
463: PetscCall(PetscOptionsBool("-tao_cancelmonitors", "cancel all monitors and call any registered destroy routines", "TaoCancelMonitors", flg, &flg, NULL));
464: if (flg) PetscCall(TaoCancelMonitors(tao));
466: flg = PETSC_FALSE;
467: PetscCall(PetscOptionsBool("-tao_draw_solution", "Plot solution vector at each iteration", "TaoSetMonitor", flg, &flg, NULL));
468: if (flg) {
469: TaoMonitorDrawCtx drawctx;
470: PetscInt howoften = 1;
471: PetscCall(TaoMonitorDrawCtxCreate(PetscObjectComm((PetscObject)tao), NULL, NULL, PETSC_DECIDE, PETSC_DECIDE, 300, 300, howoften, &drawctx));
472: PetscCall(TaoSetMonitor(tao, TaoDrawSolutionMonitor, drawctx, (PetscErrorCode(*)(void **))TaoMonitorDrawCtxDestroy));
473: }
475: flg = PETSC_FALSE;
476: PetscCall(PetscOptionsBool("-tao_draw_step", "plots step direction at each iteration", "TaoSetMonitor", flg, &flg, NULL));
477: if (flg) PetscCall(TaoSetMonitor(tao, TaoDrawStepMonitor, NULL, NULL));
479: flg = PETSC_FALSE;
480: PetscCall(PetscOptionsBool("-tao_draw_gradient", "plots gradient at each iteration", "TaoSetMonitor", flg, &flg, NULL));
481: if (flg) {
482: TaoMonitorDrawCtx drawctx;
483: PetscInt howoften = 1;
484: PetscCall(TaoMonitorDrawCtxCreate(PetscObjectComm((PetscObject)tao), NULL, NULL, PETSC_DECIDE, PETSC_DECIDE, 300, 300, howoften, &drawctx));
485: PetscCall(TaoSetMonitor(tao, TaoDrawGradientMonitor, drawctx, (PetscErrorCode(*)(void **))TaoMonitorDrawCtxDestroy));
486: }
487: flg = PETSC_FALSE;
488: PetscCall(PetscOptionsBool("-tao_fd_gradient", "compute gradient using finite differences", "TaoDefaultComputeGradient", flg, &flg, NULL));
489: if (flg) PetscCall(TaoSetGradient(tao, NULL, TaoDefaultComputeGradient, NULL));
490: flg = PETSC_FALSE;
491: PetscCall(PetscOptionsBool("-tao_fd_hessian", "compute hessian using finite differences", "TaoDefaultComputeHessian", flg, &flg, NULL));
492: if (flg) {
493: Mat H;
495: PetscCall(MatCreate(PetscObjectComm((PetscObject)tao), &H));
496: PetscCall(MatSetType(H, MATAIJ));
497: PetscCall(TaoSetHessian(tao, H, H, TaoDefaultComputeHessian, NULL));
498: PetscCall(MatDestroy(&H));
499: }
500: flg = PETSC_FALSE;
501: PetscCall(PetscOptionsBool("-tao_mf_hessian", "compute matrix-free hessian using finite differences", "TaoDefaultComputeHessianMFFD", flg, &flg, NULL));
502: if (flg) {
503: Mat H;
505: PetscCall(MatCreate(PetscObjectComm((PetscObject)tao), &H));
506: PetscCall(TaoSetHessian(tao, H, H, TaoDefaultComputeHessianMFFD, NULL));
507: PetscCall(MatDestroy(&H));
508: }
509: PetscCall(PetscOptionsBool("-tao_recycle_history", "enable recycling/re-using information from the previous TaoSolve() call for some algorithms", "TaoSetRecycleHistory", flg, &flg, &found));
510: if (found) PetscCall(TaoSetRecycleHistory(tao, flg));
511: PetscCall(PetscOptionsEnum("-tao_subset_type", "subset type", "", TaoSubSetTypes, (PetscEnum)tao->subset_type, (PetscEnum *)&tao->subset_type, NULL));
513: if (tao->ksp) {
514: PetscCall(PetscOptionsBool("-tao_ksp_ew", "Use Eisentat-Walker linear system convergence test", "TaoKSPSetUseEW", tao->ksp_ewconv, &tao->ksp_ewconv, NULL));
515: PetscCall(TaoKSPSetUseEW(tao, tao->ksp_ewconv));
516: }
518: PetscTryTypeMethod(tao, setfromoptions, PetscOptionsObject);
520: /* process any options handlers added with PetscObjectAddOptionsHandler() */
521: PetscCall(PetscObjectProcessOptionsHandlers((PetscObject)tao, PetscOptionsObject));
522: PetscOptionsEnd();
524: if (tao->linesearch) PetscCall(TaoLineSearchSetFromOptions(tao->linesearch));
525: PetscFunctionReturn(PETSC_SUCCESS);
526: }
528: /*@C
529: TaoViewFromOptions - View a `Tao` object based on values in the options database
531: Collective
533: Input Parameters:
534: + A - the `Tao` context
535: . obj - Optional object that provides the prefix for the options database
536: - name - command line option
538: Level: intermediate
540: .seealso: [](ch_tao), `Tao`, `TaoView`, `PetscObjectViewFromOptions()`, `TaoCreate()`
541: @*/
542: PetscErrorCode TaoViewFromOptions(Tao A, PetscObject obj, const char name[])
543: {
544: PetscFunctionBegin;
546: PetscCall(PetscObjectViewFromOptions((PetscObject)A, obj, name));
547: PetscFunctionReturn(PETSC_SUCCESS);
548: }
550: /*@C
551: TaoView - Prints information about the `Tao` object
553: Collective
555: Input Parameters:
556: + tao - the `Tao` context
557: - viewer - visualization context
559: Options Database Key:
560: . -tao_view - Calls `TaoView()` at the end of `TaoSolve()`
562: Level: beginner
564: Notes:
565: The available visualization contexts include
566: + `PETSC_VIEWER_STDOUT_SELF` - standard output (default)
567: - `PETSC_VIEWER_STDOUT_WORLD` - synchronized standard
568: output where only the first processor opens
569: the file. All other processors send their
570: data to the first processor to print.
572: .seealso: [](ch_tao), `Tao`, `PetscViewerASCIIOpen()`
573: @*/
574: PetscErrorCode TaoView(Tao tao, PetscViewer viewer)
575: {
576: PetscBool isascii, isstring;
577: TaoType type;
579: PetscFunctionBegin;
581: if (!viewer) PetscCall(PetscViewerASCIIGetStdout(((PetscObject)tao)->comm, &viewer));
583: PetscCheckSameComm(tao, 1, viewer, 2);
585: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii));
586: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSTRING, &isstring));
587: if (isascii) {
588: PetscCall(PetscObjectPrintClassNamePrefixType((PetscObject)tao, viewer));
590: if (tao->ops->view) {
591: PetscCall(PetscViewerASCIIPushTab(viewer));
592: PetscUseTypeMethod(tao, view, viewer);
593: PetscCall(PetscViewerASCIIPopTab(viewer));
594: }
595: if (tao->linesearch) {
596: PetscCall(PetscViewerASCIIPushTab(viewer));
597: PetscCall(TaoLineSearchView(tao->linesearch, viewer));
598: PetscCall(PetscViewerASCIIPopTab(viewer));
599: }
600: if (tao->ksp) {
601: PetscCall(PetscViewerASCIIPushTab(viewer));
602: PetscCall(KSPView(tao->ksp, viewer));
603: PetscCall(PetscViewerASCIIPrintf(viewer, "total KSP iterations: %" PetscInt_FMT "\n", tao->ksp_tot_its));
604: PetscCall(PetscViewerASCIIPopTab(viewer));
605: }
607: PetscCall(PetscViewerASCIIPushTab(viewer));
609: if (tao->XL || tao->XU) PetscCall(PetscViewerASCIIPrintf(viewer, "Active Set subset type: %s\n", TaoSubSetTypes[tao->subset_type]));
611: PetscCall(PetscViewerASCIIPrintf(viewer, "convergence tolerances: gatol=%g,", (double)tao->gatol));
612: PetscCall(PetscViewerASCIIPrintf(viewer, " steptol=%g,", (double)tao->steptol));
613: PetscCall(PetscViewerASCIIPrintf(viewer, " gttol=%g\n", (double)tao->gttol));
614: PetscCall(PetscViewerASCIIPrintf(viewer, "Residual in Function/Gradient:=%g\n", (double)tao->residual));
616: if (tao->constrained) {
617: PetscCall(PetscViewerASCIIPrintf(viewer, "convergence tolerances:"));
618: PetscCall(PetscViewerASCIIPrintf(viewer, " catol=%g,", (double)tao->catol));
619: PetscCall(PetscViewerASCIIPrintf(viewer, " crtol=%g\n", (double)tao->crtol));
620: PetscCall(PetscViewerASCIIPrintf(viewer, "Residual in Constraints:=%g\n", (double)tao->cnorm));
621: }
623: if (tao->trust < tao->steptol) {
624: PetscCall(PetscViewerASCIIPrintf(viewer, "convergence tolerances: steptol=%g\n", (double)tao->steptol));
625: PetscCall(PetscViewerASCIIPrintf(viewer, "Final trust region radius:=%g\n", (double)tao->trust));
626: }
628: if (tao->fmin > -1.e25) PetscCall(PetscViewerASCIIPrintf(viewer, "convergence tolerances: function minimum=%g\n", (double)tao->fmin));
629: PetscCall(PetscViewerASCIIPrintf(viewer, "Objective value=%g\n", (double)tao->fc));
631: PetscCall(PetscViewerASCIIPrintf(viewer, "total number of iterations=%" PetscInt_FMT ", ", tao->niter));
632: PetscCall(PetscViewerASCIIPrintf(viewer, " (max: %" PetscInt_FMT ")\n", tao->max_it));
634: if (tao->nfuncs > 0) {
635: PetscCall(PetscViewerASCIIPrintf(viewer, "total number of function evaluations=%" PetscInt_FMT ",", tao->nfuncs));
636: PetscCall(PetscViewerASCIIPrintf(viewer, " max: %" PetscInt_FMT "\n", tao->max_funcs));
637: }
638: if (tao->ngrads > 0) {
639: PetscCall(PetscViewerASCIIPrintf(viewer, "total number of gradient evaluations=%" PetscInt_FMT ",", tao->ngrads));
640: PetscCall(PetscViewerASCIIPrintf(viewer, " max: %" PetscInt_FMT "\n", tao->max_funcs));
641: }
642: if (tao->nfuncgrads > 0) {
643: PetscCall(PetscViewerASCIIPrintf(viewer, "total number of function/gradient evaluations=%" PetscInt_FMT ",", tao->nfuncgrads));
644: PetscCall(PetscViewerASCIIPrintf(viewer, " (max: %" PetscInt_FMT ")\n", tao->max_funcs));
645: }
646: if (tao->nhess > 0) PetscCall(PetscViewerASCIIPrintf(viewer, "total number of Hessian evaluations=%" PetscInt_FMT "\n", tao->nhess));
647: if (tao->nconstraints > 0) PetscCall(PetscViewerASCIIPrintf(viewer, "total number of constraint function evaluations=%" PetscInt_FMT "\n", tao->nconstraints));
648: if (tao->njac > 0) PetscCall(PetscViewerASCIIPrintf(viewer, "total number of Jacobian evaluations=%" PetscInt_FMT "\n", tao->njac));
650: if (tao->reason > 0) {
651: PetscCall(PetscViewerASCIIPrintf(viewer, "Solution converged: "));
652: switch (tao->reason) {
653: case TAO_CONVERGED_GATOL:
654: PetscCall(PetscViewerASCIIPrintf(viewer, " ||g(X)|| <= gatol\n"));
655: break;
656: case TAO_CONVERGED_GRTOL:
657: PetscCall(PetscViewerASCIIPrintf(viewer, " ||g(X)||/|f(X)| <= grtol\n"));
658: break;
659: case TAO_CONVERGED_GTTOL:
660: PetscCall(PetscViewerASCIIPrintf(viewer, " ||g(X)||/||g(X0)|| <= gttol\n"));
661: break;
662: case TAO_CONVERGED_STEPTOL:
663: PetscCall(PetscViewerASCIIPrintf(viewer, " Steptol -- step size small\n"));
664: break;
665: case TAO_CONVERGED_MINF:
666: PetscCall(PetscViewerASCIIPrintf(viewer, " Minf -- f < fmin\n"));
667: break;
668: case TAO_CONVERGED_USER:
669: PetscCall(PetscViewerASCIIPrintf(viewer, " User Terminated\n"));
670: break;
671: default:
672: PetscCall(PetscViewerASCIIPrintf(viewer, "\n"));
673: break;
674: }
675: } else {
676: PetscCall(PetscViewerASCIIPrintf(viewer, "Solver terminated: %d", tao->reason));
677: switch (tao->reason) {
678: case TAO_DIVERGED_MAXITS:
679: PetscCall(PetscViewerASCIIPrintf(viewer, " Maximum Iterations\n"));
680: break;
681: case TAO_DIVERGED_NAN:
682: PetscCall(PetscViewerASCIIPrintf(viewer, " NAN or Inf encountered\n"));
683: break;
684: case TAO_DIVERGED_MAXFCN:
685: PetscCall(PetscViewerASCIIPrintf(viewer, " Maximum Function Evaluations\n"));
686: break;
687: case TAO_DIVERGED_LS_FAILURE:
688: PetscCall(PetscViewerASCIIPrintf(viewer, " Line Search Failure\n"));
689: break;
690: case TAO_DIVERGED_TR_REDUCTION:
691: PetscCall(PetscViewerASCIIPrintf(viewer, " Trust Region too small\n"));
692: break;
693: case TAO_DIVERGED_USER:
694: PetscCall(PetscViewerASCIIPrintf(viewer, " User Terminated\n"));
695: break;
696: default:
697: PetscCall(PetscViewerASCIIPrintf(viewer, "\n"));
698: break;
699: }
700: }
701: PetscCall(PetscViewerASCIIPopTab(viewer));
702: } else if (isstring) {
703: PetscCall(TaoGetType(tao, &type));
704: PetscCall(PetscViewerStringSPrintf(viewer, " %-3.3s", type));
705: }
706: PetscFunctionReturn(PETSC_SUCCESS);
707: }
709: /*@
710: TaoSetRecycleHistory - Sets the boolean flag to enable/disable re-using
711: iterate information from the previous `TaoSolve()`. This feature is disabled by
712: default.
714: Logically Collective
716: Input Parameters:
717: + tao - the `Tao` context
718: - recycle - boolean flag
720: Options Database Key:
721: . -tao_recycle_history <true,false> - reuse the history
723: Level: intermediate
725: Notes:
726: For conjugate gradient methods (`TAOBNCG`), this re-uses the latest search direction
727: from the previous `TaoSolve()` call when computing the first search direction in a
728: new solution. By default, CG methods set the first search direction to the
729: negative gradient.
731: For quasi-Newton family of methods (`TAOBQNLS`, `TAOBQNKLS`, `TAOBQNKTR`, `TAOBQNKTL`), this re-uses
732: the accumulated quasi-Newton Hessian approximation from the previous `TaoSolve()`
733: call. By default, QN family of methods reset the initial Hessian approximation to
734: the identity matrix.
736: For any other algorithm, this setting has no effect.
738: .seealso: [](ch_tao), `Tao`, `TaoGetRecycleHistory()`, `TAOBNCG`, `TAOBQNLS`, `TAOBQNKLS`, `TAOBQNKTR`, `TAOBQNKTL`
739: @*/
740: PetscErrorCode TaoSetRecycleHistory(Tao tao, PetscBool recycle)
741: {
742: PetscFunctionBegin;
745: tao->recycle = recycle;
746: PetscFunctionReturn(PETSC_SUCCESS);
747: }
749: /*@
750: TaoGetRecycleHistory - Retrieve the boolean flag for re-using iterate information
751: from the previous `TaoSolve()`. This feature is disabled by default.
753: Logically Collective
755: Input Parameter:
756: . tao - the `Tao` context
758: Output Parameter:
759: . recycle - boolean flag
761: Level: intermediate
763: .seealso: [](ch_tao), `Tao`, `TaoSetRecycleHistory()`, `TAOBNCG`, `TAOBQNLS`, `TAOBQNKLS`, `TAOBQNKTR`, `TAOBQNKTL`
764: @*/
765: PetscErrorCode TaoGetRecycleHistory(Tao tao, PetscBool *recycle)
766: {
767: PetscFunctionBegin;
769: PetscAssertPointer(recycle, 2);
770: *recycle = tao->recycle;
771: PetscFunctionReturn(PETSC_SUCCESS);
772: }
774: /*@
775: TaoSetTolerances - Sets parameters used in `TaoSolve()` convergence tests
777: Logically Collective
779: Input Parameters:
780: + tao - the `Tao` context
781: . gatol - stop if norm of gradient is less than this
782: . grtol - stop if relative norm of gradient is less than this
783: - gttol - stop if norm of gradient is reduced by this factor
785: Options Database Keys:
786: + -tao_gatol <gatol> - Sets gatol
787: . -tao_grtol <grtol> - Sets grtol
788: - -tao_gttol <gttol> - Sets gttol
790: Stopping Criteria\:
791: .vb
792: ||g(X)|| <= gatol
793: ||g(X)|| / |f(X)| <= grtol
794: ||g(X)|| / ||g(X0)|| <= gttol
795: .ve
797: Level: beginner
799: Note:
800: Use `PETSC_DEFAULT` to leave one or more tolerances unchanged.
802: .seealso: [](ch_tao), `Tao`, `TaoConvergedReason`, `TaoGetTolerances()`
803: @*/
804: PetscErrorCode TaoSetTolerances(Tao tao, PetscReal gatol, PetscReal grtol, PetscReal gttol)
805: {
806: PetscFunctionBegin;
812: if (gatol != (PetscReal)PETSC_DEFAULT) {
813: if (gatol < 0) {
814: PetscCall(PetscInfo(tao, "Tried to set negative gatol -- ignored.\n"));
815: } else {
816: tao->gatol = PetscMax(0, gatol);
817: tao->gatol_changed = PETSC_TRUE;
818: }
819: }
821: if (grtol != (PetscReal)PETSC_DEFAULT) {
822: if (grtol < 0) {
823: PetscCall(PetscInfo(tao, "Tried to set negative grtol -- ignored.\n"));
824: } else {
825: tao->grtol = PetscMax(0, grtol);
826: tao->grtol_changed = PETSC_TRUE;
827: }
828: }
830: if (gttol != (PetscReal)PETSC_DEFAULT) {
831: if (gttol < 0) {
832: PetscCall(PetscInfo(tao, "Tried to set negative gttol -- ignored.\n"));
833: } else {
834: tao->gttol = PetscMax(0, gttol);
835: tao->gttol_changed = PETSC_TRUE;
836: }
837: }
838: PetscFunctionReturn(PETSC_SUCCESS);
839: }
841: /*@
842: TaoSetConstraintTolerances - Sets constraint tolerance parameters used in `TaoSolve()` convergence tests
844: Logically Collective
846: Input Parameters:
847: + tao - the `Tao` context
848: . catol - absolute constraint tolerance, constraint norm must be less than `catol` for used for gatol convergence criteria
849: - crtol - relative constraint tolerance, constraint norm must be less than `crtol` for used for gatol, gttol convergence criteria
851: Options Database Keys:
852: + -tao_catol <catol> - Sets catol
853: - -tao_crtol <crtol> - Sets crtol
855: Level: intermediate
857: Notes:
858: Use `PETSC_DEFAULT` to leave any tolerance unchanged.
860: .seealso: [](ch_tao), `Tao`, `TaoConvergedReason`, `TaoGetTolerances()`, `TaoGetConstraintTolerances()`, `TaoSetTolerances()`
861: @*/
862: PetscErrorCode TaoSetConstraintTolerances(Tao tao, PetscReal catol, PetscReal crtol)
863: {
864: PetscFunctionBegin;
869: if (catol != (PetscReal)PETSC_DEFAULT) {
870: if (catol < 0) {
871: PetscCall(PetscInfo(tao, "Tried to set negative catol -- ignored.\n"));
872: } else {
873: tao->catol = PetscMax(0, catol);
874: tao->catol_changed = PETSC_TRUE;
875: }
876: }
878: if (crtol != (PetscReal)PETSC_DEFAULT) {
879: if (crtol < 0) {
880: PetscCall(PetscInfo(tao, "Tried to set negative crtol -- ignored.\n"));
881: } else {
882: tao->crtol = PetscMax(0, crtol);
883: tao->crtol_changed = PETSC_TRUE;
884: }
885: }
886: PetscFunctionReturn(PETSC_SUCCESS);
887: }
889: /*@
890: TaoGetConstraintTolerances - Gets constraint tolerance parameters used in `TaoSolve()` convergence tests
892: Not Collective
894: Input Parameter:
895: . tao - the `Tao` context
897: Output Parameters:
898: + catol - absolute constraint tolerance, constraint norm must be less than `catol` for used for gatol convergence criteria
899: - crtol - relative constraint tolerance, constraint norm must be less than `crtol` for used for gatol, gttol convergence criteria
901: Level: intermediate
903: .seealso: [](ch_tao), `Tao`, `TaoConvergedReasons`,`TaoGetTolerances()`, `TaoSetTolerances()`, `TaoSetConstraintTolerances()`
904: @*/
905: PetscErrorCode TaoGetConstraintTolerances(Tao tao, PetscReal *catol, PetscReal *crtol)
906: {
907: PetscFunctionBegin;
909: if (catol) *catol = tao->catol;
910: if (crtol) *crtol = tao->crtol;
911: PetscFunctionReturn(PETSC_SUCCESS);
912: }
914: /*@
915: TaoSetFunctionLowerBound - Sets a bound on the solution objective value.
916: When an approximate solution with an objective value below this number
917: has been found, the solver will terminate.
919: Logically Collective
921: Input Parameters:
922: + tao - the Tao solver context
923: - fmin - the tolerance
925: Options Database Key:
926: . -tao_fmin <fmin> - sets the minimum function value
928: Level: intermediate
930: .seealso: [](ch_tao), `Tao`, `TaoConvergedReason`, `TaoSetTolerances()`
931: @*/
932: PetscErrorCode TaoSetFunctionLowerBound(Tao tao, PetscReal fmin)
933: {
934: PetscFunctionBegin;
937: tao->fmin = fmin;
938: tao->fmin_changed = PETSC_TRUE;
939: PetscFunctionReturn(PETSC_SUCCESS);
940: }
942: /*@
943: TaoGetFunctionLowerBound - Gets the bound on the solution objective value.
944: When an approximate solution with an objective value below this number
945: has been found, the solver will terminate.
947: Not Collective
949: Input Parameter:
950: . tao - the `Tao` solver context
952: Output Parameter:
953: . fmin - the minimum function value
955: Level: intermediate
957: .seealso: [](ch_tao), `Tao`, `TaoConvergedReason`, `TaoSetFunctionLowerBound()`
958: @*/
959: PetscErrorCode TaoGetFunctionLowerBound(Tao tao, PetscReal *fmin)
960: {
961: PetscFunctionBegin;
963: PetscAssertPointer(fmin, 2);
964: *fmin = tao->fmin;
965: PetscFunctionReturn(PETSC_SUCCESS);
966: }
968: /*@
969: TaoSetMaximumFunctionEvaluations - Sets a maximum number of function evaluations allowed for a `TaoSolve()`.
971: Logically Collective
973: Input Parameters:
974: + tao - the `Tao` solver context
975: - nfcn - the maximum number of function evaluations (>=0)
977: Options Database Key:
978: . -tao_max_funcs <nfcn> - sets the maximum number of function evaluations
980: Level: intermediate
982: .seealso: [](ch_tao), `Tao`, `TaoSetTolerances()`, `TaoSetMaximumIterations()`
983: @*/
984: PetscErrorCode TaoSetMaximumFunctionEvaluations(Tao tao, PetscInt nfcn)
985: {
986: PetscFunctionBegin;
989: if (nfcn >= 0) {
990: tao->max_funcs = PetscMax(0, nfcn);
991: } else {
992: tao->max_funcs = -1;
993: }
994: tao->max_funcs_changed = PETSC_TRUE;
995: PetscFunctionReturn(PETSC_SUCCESS);
996: }
998: /*@
999: TaoGetMaximumFunctionEvaluations - Gets a maximum number of function evaluations allowed for a `TaoSolve()`
1001: Logically Collective
1003: Input Parameter:
1004: . tao - the `Tao` solver context
1006: Output Parameter:
1007: . nfcn - the maximum number of function evaluations
1009: Level: intermediate
1011: .seealso: [](ch_tao), `Tao`, `TaoSetMaximumFunctionEvaluations()`, `TaoGetMaximumIterations()`
1012: @*/
1013: PetscErrorCode TaoGetMaximumFunctionEvaluations(Tao tao, PetscInt *nfcn)
1014: {
1015: PetscFunctionBegin;
1017: PetscAssertPointer(nfcn, 2);
1018: *nfcn = tao->max_funcs;
1019: PetscFunctionReturn(PETSC_SUCCESS);
1020: }
1022: /*@
1023: TaoGetCurrentFunctionEvaluations - Get current number of function evaluations used by a `Tao` object
1025: Not Collective
1027: Input Parameter:
1028: . tao - the `Tao` solver context
1030: Output Parameter:
1031: . nfuncs - the current number of function evaluations (maximum between gradient and function evaluations)
1033: Level: intermediate
1035: .seealso: [](ch_tao), `Tao`, `TaoSetMaximumFunctionEvaluations()`, `TaoGetMaximumFunctionEvaluations()`, `TaoGetMaximumIterations()`
1036: @*/
1037: PetscErrorCode TaoGetCurrentFunctionEvaluations(Tao tao, PetscInt *nfuncs)
1038: {
1039: PetscFunctionBegin;
1041: PetscAssertPointer(nfuncs, 2);
1042: *nfuncs = PetscMax(tao->nfuncs, tao->nfuncgrads);
1043: PetscFunctionReturn(PETSC_SUCCESS);
1044: }
1046: /*@
1047: TaoSetMaximumIterations - Sets a maximum number of iterates to be used in `TaoSolve()`
1049: Logically Collective
1051: Input Parameters:
1052: + tao - the `Tao` solver context
1053: - maxits - the maximum number of iterates (>=0)
1055: Options Database Key:
1056: . -tao_max_it <its> - sets the maximum number of iterations
1058: Level: intermediate
1060: .seealso: [](ch_tao), `Tao`, `TaoSetTolerances()`, `TaoSetMaximumFunctionEvaluations()`
1061: @*/
1062: PetscErrorCode TaoSetMaximumIterations(Tao tao, PetscInt maxits)
1063: {
1064: PetscFunctionBegin;
1067: tao->max_it = PetscMax(0, maxits);
1068: tao->max_it_changed = PETSC_TRUE;
1069: PetscFunctionReturn(PETSC_SUCCESS);
1070: }
1072: /*@
1073: TaoGetMaximumIterations - Gets a maximum number of iterates that will be used
1075: Not Collective
1077: Input Parameter:
1078: . tao - the `Tao` solver context
1080: Output Parameter:
1081: . maxits - the maximum number of iterates
1083: Level: intermediate
1085: .seealso: [](ch_tao), `Tao`, `TaoSetMaximumIterations()`, `TaoGetMaximumFunctionEvaluations()`
1086: @*/
1087: PetscErrorCode TaoGetMaximumIterations(Tao tao, PetscInt *maxits)
1088: {
1089: PetscFunctionBegin;
1091: PetscAssertPointer(maxits, 2);
1092: *maxits = tao->max_it;
1093: PetscFunctionReturn(PETSC_SUCCESS);
1094: }
1096: /*@
1097: TaoSetInitialTrustRegionRadius - Sets the initial trust region radius.
1099: Logically Collective
1101: Input Parameters:
1102: + tao - a `Tao` optimization solver
1103: - radius - the trust region radius
1105: Options Database Key:
1106: . -tao_trust0 <t0> - sets initial trust region radius
1108: Level: intermediate
1110: .seealso: [](ch_tao), `Tao`, `TaoGetTrustRegionRadius()`, `TaoSetTrustRegionTolerance()`, `TAONTR`
1111: @*/
1112: PetscErrorCode TaoSetInitialTrustRegionRadius(Tao tao, PetscReal radius)
1113: {
1114: PetscFunctionBegin;
1117: tao->trust0 = PetscMax(0.0, radius);
1118: tao->trust0_changed = PETSC_TRUE;
1119: PetscFunctionReturn(PETSC_SUCCESS);
1120: }
1122: /*@
1123: TaoGetInitialTrustRegionRadius - Gets the initial trust region radius.
1125: Not Collective
1127: Input Parameter:
1128: . tao - a `Tao` optimization solver
1130: Output Parameter:
1131: . radius - the trust region radius
1133: Level: intermediate
1135: .seealso: [](ch_tao), `Tao`, `TaoSetInitialTrustRegionRadius()`, `TaoGetCurrentTrustRegionRadius()`, `TAONTR`
1136: @*/
1137: PetscErrorCode TaoGetInitialTrustRegionRadius(Tao tao, PetscReal *radius)
1138: {
1139: PetscFunctionBegin;
1141: PetscAssertPointer(radius, 2);
1142: *radius = tao->trust0;
1143: PetscFunctionReturn(PETSC_SUCCESS);
1144: }
1146: /*@
1147: TaoGetCurrentTrustRegionRadius - Gets the current trust region radius.
1149: Not Collective
1151: Input Parameter:
1152: . tao - a `Tao` optimization solver
1154: Output Parameter:
1155: . radius - the trust region radius
1157: Level: intermediate
1159: .seealso: [](ch_tao), `Tao`, `TaoSetInitialTrustRegionRadius()`, `TaoGetInitialTrustRegionRadius()`, `TAONTR`
1160: @*/
1161: PetscErrorCode TaoGetCurrentTrustRegionRadius(Tao tao, PetscReal *radius)
1162: {
1163: PetscFunctionBegin;
1165: PetscAssertPointer(radius, 2);
1166: *radius = tao->trust;
1167: PetscFunctionReturn(PETSC_SUCCESS);
1168: }
1170: /*@
1171: TaoGetTolerances - gets the current values of some tolerances used for the convergence testing of `TaoSolve()`
1173: Not Collective
1175: Input Parameter:
1176: . tao - the `Tao` context
1178: Output Parameters:
1179: + gatol - stop if norm of gradient is less than this
1180: . grtol - stop if relative norm of gradient is less than this
1181: - gttol - stop if norm of gradient is reduced by a this factor
1183: Level: intermediate
1185: Note:
1186: `NULL` can be used as an argument if not all tolerances values are needed
1188: .seealso: [](ch_tao), `Tao`, `TaoSetTolerances()`
1189: @*/
1190: PetscErrorCode TaoGetTolerances(Tao tao, PetscReal *gatol, PetscReal *grtol, PetscReal *gttol)
1191: {
1192: PetscFunctionBegin;
1194: if (gatol) *gatol = tao->gatol;
1195: if (grtol) *grtol = tao->grtol;
1196: if (gttol) *gttol = tao->gttol;
1197: PetscFunctionReturn(PETSC_SUCCESS);
1198: }
1200: /*@
1201: TaoGetKSP - Gets the linear solver used by the optimization solver.
1203: Not Collective
1205: Input Parameter:
1206: . tao - the `Tao` solver
1208: Output Parameter:
1209: . ksp - the `KSP` linear solver used in the optimization solver
1211: Level: intermediate
1213: .seealso: [](ch_tao), `Tao`, `KSP`
1214: @*/
1215: PetscErrorCode TaoGetKSP(Tao tao, KSP *ksp)
1216: {
1217: PetscFunctionBegin;
1219: PetscAssertPointer(ksp, 2);
1220: *ksp = tao->ksp;
1221: PetscFunctionReturn(PETSC_SUCCESS);
1222: }
1224: /*@
1225: TaoGetLinearSolveIterations - Gets the total number of linear iterations
1226: used by the `Tao` solver
1228: Not Collective
1230: Input Parameter:
1231: . tao - the `Tao` context
1233: Output Parameter:
1234: . lits - number of linear iterations
1236: Level: intermediate
1238: Note:
1239: This counter is reset to zero for each successive call to `TaoSolve()`
1241: .seealso: [](ch_tao), `Tao`, `TaoGetKSP()`
1242: @*/
1243: PetscErrorCode TaoGetLinearSolveIterations(Tao tao, PetscInt *lits)
1244: {
1245: PetscFunctionBegin;
1247: PetscAssertPointer(lits, 2);
1248: *lits = tao->ksp_tot_its;
1249: PetscFunctionReturn(PETSC_SUCCESS);
1250: }
1252: /*@
1253: TaoGetLineSearch - Gets the line search used by the optimization solver.
1255: Not Collective
1257: Input Parameter:
1258: . tao - the `Tao` solver
1260: Output Parameter:
1261: . ls - the line search used in the optimization solver
1263: Level: intermediate
1265: .seealso: [](ch_tao), `Tao`, `TaoLineSearch`, `TaoLineSearchType`
1266: @*/
1267: PetscErrorCode TaoGetLineSearch(Tao tao, TaoLineSearch *ls)
1268: {
1269: PetscFunctionBegin;
1271: PetscAssertPointer(ls, 2);
1272: *ls = tao->linesearch;
1273: PetscFunctionReturn(PETSC_SUCCESS);
1274: }
1276: /*@
1277: TaoAddLineSearchCounts - Adds the number of function evaluations spent
1278: in the line search to the running total.
1280: Input Parameters:
1281: . tao - the `Tao` solver
1283: Level: developer
1285: .seealso: [](ch_tao), `Tao`, `TaoGetLineSearch()`, `TaoLineSearchApply()`
1286: @*/
1287: PetscErrorCode TaoAddLineSearchCounts(Tao tao)
1288: {
1289: PetscBool flg;
1290: PetscInt nfeval, ngeval, nfgeval;
1292: PetscFunctionBegin;
1294: if (tao->linesearch) {
1295: PetscCall(TaoLineSearchIsUsingTaoRoutines(tao->linesearch, &flg));
1296: if (!flg) {
1297: PetscCall(TaoLineSearchGetNumberFunctionEvaluations(tao->linesearch, &nfeval, &ngeval, &nfgeval));
1298: tao->nfuncs += nfeval;
1299: tao->ngrads += ngeval;
1300: tao->nfuncgrads += nfgeval;
1301: }
1302: }
1303: PetscFunctionReturn(PETSC_SUCCESS);
1304: }
1306: /*@
1307: TaoGetSolution - Returns the vector with the current solution from the `Tao` object
1309: Not Collective
1311: Input Parameter:
1312: . tao - the `Tao` context
1314: Output Parameter:
1315: . X - the current solution
1317: Level: intermediate
1319: Note:
1320: The returned vector will be the same object that was passed into `TaoSetSolution()`
1322: .seealso: [](ch_tao), `Tao`, `TaoSetSolution()`, `TaoSolve()`
1323: @*/
1324: PetscErrorCode TaoGetSolution(Tao tao, Vec *X)
1325: {
1326: PetscFunctionBegin;
1328: PetscAssertPointer(X, 2);
1329: *X = tao->solution;
1330: PetscFunctionReturn(PETSC_SUCCESS);
1331: }
1333: /*@
1334: TaoResetStatistics - Initialize the statistics collected by the `Tao` object.
1335: These statistics include the iteration number, residual norms, and convergence status.
1336: This routine gets called before solving each optimization problem.
1338: Collective
1340: Input Parameter:
1341: . tao - the `Tao` context
1343: Level: developer
1345: .seealso: [](ch_tao), `Tao`, `TaoCreate()`, `TaoSolve()`
1346: @*/
1347: PetscErrorCode TaoResetStatistics(Tao tao)
1348: {
1349: PetscFunctionBegin;
1351: tao->niter = 0;
1352: tao->nfuncs = 0;
1353: tao->nfuncgrads = 0;
1354: tao->ngrads = 0;
1355: tao->nhess = 0;
1356: tao->njac = 0;
1357: tao->nconstraints = 0;
1358: tao->ksp_its = 0;
1359: tao->ksp_tot_its = 0;
1360: tao->reason = TAO_CONTINUE_ITERATING;
1361: tao->residual = 0.0;
1362: tao->cnorm = 0.0;
1363: tao->step = 0.0;
1364: tao->lsflag = PETSC_FALSE;
1365: if (tao->hist_reset) tao->hist_len = 0;
1366: PetscFunctionReturn(PETSC_SUCCESS);
1367: }
1369: /*@C
1370: TaoSetUpdate - Sets the general-purpose update function called
1371: at the beginning of every iteration of the optimization algorithm. Called after the new solution and the gradient
1372: is determined, but before the Hessian is computed (if applicable).
1374: Logically Collective
1376: Input Parameters:
1377: + tao - The `Tao` solver context
1378: - func - The function
1380: Calling sequence of `func`:
1381: + tao - the optimizer context
1382: - ctx - The current step of the iteration
1384: Level: advanced
1386: .seealso: [](ch_tao), `Tao`, `TaoSolve()`
1387: @*/
1388: PetscErrorCode TaoSetUpdate(Tao tao, PetscErrorCode (*func)(Tao, PetscInt, void *), void *ctx)
1389: {
1390: PetscFunctionBegin;
1392: tao->ops->update = func;
1393: tao->user_update = ctx;
1394: PetscFunctionReturn(PETSC_SUCCESS);
1395: }
1397: /*@C
1398: TaoSetConvergenceTest - Sets the function that is to be used to test
1399: for convergence o fthe iterative minimization solution. The new convergence
1400: testing routine will replace Tao's default convergence test.
1402: Logically Collective
1404: Input Parameters:
1405: + tao - the `Tao` object
1406: . conv - the routine to test for convergence
1407: - ctx - [optional] context for private data for the convergence routine
1408: (may be `NULL`)
1410: Calling sequence of `conv`:
1411: + tao - the `Tao` object
1412: - ctx - [optional] convergence context
1414: Level: advanced
1416: Note:
1417: The new convergence testing routine should call `TaoSetConvergedReason()`.
1419: .seealso: [](ch_tao), `Tao`, `TaoSolve()`, `TaoSetConvergedReason()`, `TaoGetSolutionStatus()`, `TaoGetTolerances()`, `TaoSetMonitor`
1420: @*/
1421: PetscErrorCode TaoSetConvergenceTest(Tao tao, PetscErrorCode (*conv)(Tao, void *), void *ctx)
1422: {
1423: PetscFunctionBegin;
1425: tao->ops->convergencetest = conv;
1426: tao->cnvP = ctx;
1427: PetscFunctionReturn(PETSC_SUCCESS);
1428: }
1430: /*@C
1431: TaoSetMonitor - Sets an additional function that is to be used at every
1432: iteration of the solver to display the iteration's
1433: progress.
1435: Logically Collective
1437: Input Parameters:
1438: + tao - the `Tao` solver context
1439: . func - monitoring routine
1440: . ctx - [optional] user-defined context for private data for the monitor routine (may be `NULL`)
1441: - dest - [optional] function to destroy the context when the `Tao` is destroyed
1443: Calling sequence of `func`:
1444: + tao - the `Tao` solver context
1445: - ctx - [optional] monitoring context
1447: Calling sequence of `dest`:
1448: . ctx - monitoring context
1450: Options Database Keys:
1451: + -tao_monitor - sets the default monitor `TaoMonitorDefault()`
1452: . -tao_smonitor - sets short monitor
1453: . -tao_cmonitor - same as smonitor plus constraint norm
1454: . -tao_view_solution - view solution at each iteration
1455: . -tao_view_gradient - view gradient at each iteration
1456: . -tao_view_ls_residual - view least-squares residual vector at each iteration
1457: - -tao_cancelmonitors - cancels all monitors that have been hardwired into a code by calls to TaoSetMonitor(), but does not cancel those set via the options database.
1459: Level: intermediate
1461: Notes:
1462: Several different monitoring routines may be set by calling
1463: `TaoSetMonitor()` multiple times; all will be called in the
1464: order in which they were set.
1466: Fortran Notes:
1467: Only one monitor function may be set
1469: .seealso: [](ch_tao), `Tao`, `TaoSolve()`, `TaoMonitorDefault()`, `TaoCancelMonitors()`, `TaoSetDestroyRoutine()`, `TaoView()`
1470: @*/
1471: PetscErrorCode TaoSetMonitor(Tao tao, PetscErrorCode (*func)(Tao, void *), void *ctx, PetscErrorCode (*dest)(void **))
1472: {
1473: PetscInt i;
1474: PetscBool identical;
1476: PetscFunctionBegin;
1478: PetscCheck(tao->numbermonitors < MAXTAOMONITORS, PetscObjectComm((PetscObject)tao), PETSC_ERR_SUP, "Cannot attach another monitor -- max=%d", MAXTAOMONITORS);
1480: for (i = 0; i < tao->numbermonitors; i++) {
1481: PetscCall(PetscMonitorCompare((PetscErrorCode(*)(void))func, ctx, dest, (PetscErrorCode(*)(void))tao->monitor[i], tao->monitorcontext[i], tao->monitordestroy[i], &identical));
1482: if (identical) PetscFunctionReturn(PETSC_SUCCESS);
1483: }
1484: tao->monitor[tao->numbermonitors] = func;
1485: tao->monitorcontext[tao->numbermonitors] = (void *)ctx;
1486: tao->monitordestroy[tao->numbermonitors] = dest;
1487: ++tao->numbermonitors;
1488: PetscFunctionReturn(PETSC_SUCCESS);
1489: }
1491: /*@
1492: TaoCancelMonitors - Clears all the monitor functions for a `Tao` object.
1494: Logically Collective
1496: Input Parameter:
1497: . tao - the `Tao` solver context
1499: Options Database Key:
1500: . -tao_cancelmonitors - cancels all monitors that have been hardwired
1501: into a code by calls to `TaoSetMonitor()`, but does not cancel those
1502: set via the options database
1504: Level: advanced
1506: Note:
1507: There is no way to clear one specific monitor from a `Tao` object.
1509: .seealso: [](ch_tao), `Tao`, `TaoMonitorDefault()`, `TaoSetMonitor()`
1510: @*/
1511: PetscErrorCode TaoCancelMonitors(Tao tao)
1512: {
1513: PetscInt i;
1515: PetscFunctionBegin;
1517: for (i = 0; i < tao->numbermonitors; i++) {
1518: if (tao->monitordestroy[i]) PetscCall((*tao->monitordestroy[i])(&tao->monitorcontext[i]));
1519: }
1520: tao->numbermonitors = 0;
1521: PetscFunctionReturn(PETSC_SUCCESS);
1522: }
1524: /*@
1525: TaoMonitorDefault - Default routine for monitoring progress of `TaoSolve()`
1527: Collective
1529: Input Parameters:
1530: + tao - the `Tao` context
1531: - ctx - `PetscViewer` context or `NULL`
1533: Options Database Key:
1534: . -tao_monitor - turn on default monitoring
1536: Level: advanced
1538: Note:
1539: This monitor prints the function value and gradient
1540: norm at each iteration.
1542: .seealso: [](ch_tao), `Tao`, `TaoDefaultSMonitor()`, `TaoSetMonitor()`
1543: @*/
1544: PetscErrorCode TaoMonitorDefault(Tao tao, void *ctx)
1545: {
1546: PetscInt its, tabs;
1547: PetscReal fct, gnorm;
1548: PetscViewer viewer = (PetscViewer)ctx;
1550: PetscFunctionBegin;
1553: its = tao->niter;
1554: fct = tao->fc;
1555: gnorm = tao->residual;
1556: PetscCall(PetscViewerASCIIGetTab(viewer, &tabs));
1557: PetscCall(PetscViewerASCIISetTab(viewer, ((PetscObject)tao)->tablevel));
1558: if (its == 0 && ((PetscObject)tao)->prefix && !tao->header_printed) {
1559: PetscCall(PetscViewerASCIIPrintf(viewer, " Iteration information for %s solve.\n", ((PetscObject)tao)->prefix));
1560: tao->header_printed = PETSC_TRUE;
1561: }
1562: PetscCall(PetscViewerASCIIPrintf(viewer, "%3" PetscInt_FMT " TAO,", its));
1563: PetscCall(PetscViewerASCIIPrintf(viewer, " Function value: %g,", (double)fct));
1564: if (gnorm >= PETSC_INFINITY) {
1565: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: Inf \n"));
1566: } else {
1567: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: %g \n", (double)gnorm));
1568: }
1569: PetscCall(PetscViewerASCIISetTab(viewer, tabs));
1570: PetscFunctionReturn(PETSC_SUCCESS);
1571: }
1573: /*@
1574: TaoDefaultGMonitor - Default routine for monitoring progress of `TaoSolve()` with extra detail on the globalization method.
1576: Collective
1578: Input Parameters:
1579: + tao - the `Tao` context
1580: - ctx - `PetscViewer` context or `NULL`
1582: Options Database Key:
1583: . -tao_gmonitor - turn on monitoring with globalization information
1585: Level: advanced
1587: Note:
1588: This monitor prints the function value and gradient norm at each
1589: iteration, as well as the step size and trust radius. Note that the
1590: step size and trust radius may be the same for some algorithms.
1592: .seealso: [](ch_tao), `Tao`, `TaoDefaultSMonitor()`, `TaoSetMonitor()`
1593: @*/
1594: PetscErrorCode TaoDefaultGMonitor(Tao tao, void *ctx)
1595: {
1596: PetscInt its, tabs;
1597: PetscReal fct, gnorm, stp, tr;
1598: PetscViewer viewer = (PetscViewer)ctx;
1600: PetscFunctionBegin;
1603: its = tao->niter;
1604: fct = tao->fc;
1605: gnorm = tao->residual;
1606: stp = tao->step;
1607: tr = tao->trust;
1608: PetscCall(PetscViewerASCIIGetTab(viewer, &tabs));
1609: PetscCall(PetscViewerASCIISetTab(viewer, ((PetscObject)tao)->tablevel));
1610: if (its == 0 && ((PetscObject)tao)->prefix && !tao->header_printed) {
1611: PetscCall(PetscViewerASCIIPrintf(viewer, " Iteration information for %s solve.\n", ((PetscObject)tao)->prefix));
1612: tao->header_printed = PETSC_TRUE;
1613: }
1614: PetscCall(PetscViewerASCIIPrintf(viewer, "%3" PetscInt_FMT " TAO,", its));
1615: PetscCall(PetscViewerASCIIPrintf(viewer, " Function value: %g,", (double)fct));
1616: if (gnorm >= PETSC_INFINITY) {
1617: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: Inf,"));
1618: } else {
1619: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: %g,", (double)gnorm));
1620: }
1621: PetscCall(PetscViewerASCIIPrintf(viewer, " Step: %g, Trust: %g\n", (double)stp, (double)tr));
1622: PetscCall(PetscViewerASCIISetTab(viewer, tabs));
1623: PetscFunctionReturn(PETSC_SUCCESS);
1624: }
1626: /*@
1627: TaoDefaultSMonitor - Default routine for monitoring progress of `TaoSolve()`
1629: Collective
1631: Input Parameters:
1632: + tao - the `Tao` context
1633: - ctx - `PetscViewer` context of type `PETSCVIEWERASCII`
1635: Options Database Key:
1636: . -tao_smonitor - turn on default short monitoring
1638: Level: advanced
1640: Note:
1641: Same as `TaoMonitorDefault()` except
1642: it prints fewer digits of the residual as the residual gets smaller.
1643: This is because the later digits are meaningless and are often
1644: different on different machines; by using this routine different
1645: machines will usually generate the same output.
1647: .seealso: [](ch_tao), `Tao`, `TaoMonitorDefault()`, `TaoSetMonitor()`
1648: @*/
1649: PetscErrorCode TaoDefaultSMonitor(Tao tao, void *ctx)
1650: {
1651: PetscInt its, tabs;
1652: PetscReal fct, gnorm;
1653: PetscViewer viewer = (PetscViewer)ctx;
1655: PetscFunctionBegin;
1658: its = tao->niter;
1659: fct = tao->fc;
1660: gnorm = tao->residual;
1661: PetscCall(PetscViewerASCIIGetTab(viewer, &tabs));
1662: PetscCall(PetscViewerASCIISetTab(viewer, ((PetscObject)tao)->tablevel));
1663: PetscCall(PetscViewerASCIIPrintf(viewer, "iter = %3" PetscInt_FMT ",", its));
1664: PetscCall(PetscViewerASCIIPrintf(viewer, " Function value %g,", (double)fct));
1665: if (gnorm >= PETSC_INFINITY) {
1666: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: Inf \n"));
1667: } else if (gnorm > 1.e-6) {
1668: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: %g \n", (double)gnorm));
1669: } else if (gnorm > 1.e-11) {
1670: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: < 1.0e-6 \n"));
1671: } else {
1672: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: < 1.0e-11 \n"));
1673: }
1674: PetscCall(PetscViewerASCIISetTab(viewer, tabs));
1675: PetscFunctionReturn(PETSC_SUCCESS);
1676: }
1678: /*@
1679: TaoDefaultCMonitor - same as `TaoMonitorDefault()` except
1680: it prints the norm of the constraint function.
1682: Collective
1684: Input Parameters:
1685: + tao - the `Tao` context
1686: - ctx - `PetscViewer` context or `NULL`
1688: Options Database Key:
1689: . -tao_cmonitor - monitor the constraints
1691: Level: advanced
1693: .seealso: [](ch_tao), `Tao`, `TaoMonitorDefault()`, `TaoSetMonitor()`
1694: @*/
1695: PetscErrorCode TaoDefaultCMonitor(Tao tao, void *ctx)
1696: {
1697: PetscInt its, tabs;
1698: PetscReal fct, gnorm;
1699: PetscViewer viewer = (PetscViewer)ctx;
1701: PetscFunctionBegin;
1704: its = tao->niter;
1705: fct = tao->fc;
1706: gnorm = tao->residual;
1707: PetscCall(PetscViewerASCIIGetTab(viewer, &tabs));
1708: PetscCall(PetscViewerASCIISetTab(viewer, ((PetscObject)tao)->tablevel));
1709: PetscCall(PetscViewerASCIIPrintf(viewer, "iter = %" PetscInt_FMT ",", its));
1710: PetscCall(PetscViewerASCIIPrintf(viewer, " Function value: %g,", (double)fct));
1711: PetscCall(PetscViewerASCIIPrintf(viewer, " Residual: %g ", (double)gnorm));
1712: PetscCall(PetscViewerASCIIPrintf(viewer, " Constraint: %g \n", (double)tao->cnorm));
1713: PetscCall(PetscViewerASCIISetTab(viewer, tabs));
1714: PetscFunctionReturn(PETSC_SUCCESS);
1715: }
1717: /*@C
1718: TaoSolutionMonitor - Views the solution at each iteration of `TaoSolve()`
1720: Collective
1722: Input Parameters:
1723: + tao - the `Tao` context
1724: - ctx - `PetscViewer` context or `NULL`
1726: Options Database Key:
1727: . -tao_view_solution - view the solution
1729: Level: advanced
1731: .seealso: [](ch_tao), `Tao`, `TaoDefaultSMonitor()`, `TaoSetMonitor()`
1732: @*/
1733: PetscErrorCode TaoSolutionMonitor(Tao tao, void *ctx)
1734: {
1735: PetscViewer viewer = (PetscViewer)ctx;
1737: PetscFunctionBegin;
1740: PetscCall(VecView(tao->solution, viewer));
1741: PetscFunctionReturn(PETSC_SUCCESS);
1742: }
1744: /*@C
1745: TaoGradientMonitor - Views the gradient at each iteration of `TaoSolve()`
1747: Collective
1749: Input Parameters:
1750: + tao - the `Tao` context
1751: - ctx - `PetscViewer` context or `NULL`
1753: Options Database Key:
1754: . -tao_view_gradient - view the gradient at each iteration
1756: Level: advanced
1758: .seealso: [](ch_tao), `Tao`, `TaoDefaultSMonitor()`, `TaoSetMonitor()`
1759: @*/
1760: PetscErrorCode TaoGradientMonitor(Tao tao, void *ctx)
1761: {
1762: PetscViewer viewer = (PetscViewer)ctx;
1764: PetscFunctionBegin;
1767: PetscCall(VecView(tao->gradient, viewer));
1768: PetscFunctionReturn(PETSC_SUCCESS);
1769: }
1771: /*@C
1772: TaoStepDirectionMonitor - Views the step-direction at each iteration of `TaoSolve()`
1774: Collective
1776: Input Parameters:
1777: + tao - the `Tao` context
1778: - ctx - `PetscViewer` context or `NULL`
1780: Options Database Key:
1781: . -tao_view_stepdirection - view the step direction vector at each iteration
1783: Level: advanced
1785: .seealso: [](ch_tao), `Tao`, `TaoDefaultSMonitor()`, `TaoSetMonitor()`
1786: @*/
1787: PetscErrorCode TaoStepDirectionMonitor(Tao tao, void *ctx)
1788: {
1789: PetscViewer viewer = (PetscViewer)ctx;
1791: PetscFunctionBegin;
1794: PetscCall(VecView(tao->stepdirection, viewer));
1795: PetscFunctionReturn(PETSC_SUCCESS);
1796: }
1798: /*@C
1799: TaoDrawSolutionMonitor - Plots the solution at each iteration of `TaoSolve()`
1801: Collective
1803: Input Parameters:
1804: + tao - the `Tao` context
1805: - ctx - `TaoMonitorDraw` context
1807: Options Database Key:
1808: . -tao_draw_solution - draw the solution at each iteration
1810: Level: advanced
1812: .seealso: [](ch_tao), `Tao`, `TaoSolutionMonitor()`, `TaoSetMonitor()`, `TaoDrawGradientMonitor`, `TaoMonitorDraw`
1813: @*/
1814: PetscErrorCode TaoDrawSolutionMonitor(Tao tao, void *ctx)
1815: {
1816: TaoMonitorDrawCtx ictx = (TaoMonitorDrawCtx)ctx;
1818: PetscFunctionBegin;
1820: if (!(((ictx->howoften > 0) && (!(tao->niter % ictx->howoften))) || ((ictx->howoften == -1) && tao->reason))) PetscFunctionReturn(PETSC_SUCCESS);
1821: PetscCall(VecView(tao->solution, ictx->viewer));
1822: PetscFunctionReturn(PETSC_SUCCESS);
1823: }
1825: /*@C
1826: TaoDrawGradientMonitor - Plots the gradient at each iteration of `TaoSolve()`
1828: Collective
1830: Input Parameters:
1831: + tao - the `Tao` context
1832: - ctx - `PetscViewer` context
1834: Options Database Key:
1835: . -tao_draw_gradient - draw the gradient at each iteration
1837: Level: advanced
1839: .seealso: [](ch_tao), `Tao`, `TaoGradientMonitor()`, `TaoSetMonitor()`, `TaoDrawSolutionMonitor`
1840: @*/
1841: PetscErrorCode TaoDrawGradientMonitor(Tao tao, void *ctx)
1842: {
1843: TaoMonitorDrawCtx ictx = (TaoMonitorDrawCtx)ctx;
1845: PetscFunctionBegin;
1847: if (!(((ictx->howoften > 0) && (!(tao->niter % ictx->howoften))) || ((ictx->howoften == -1) && tao->reason))) PetscFunctionReturn(PETSC_SUCCESS);
1848: PetscCall(VecView(tao->gradient, ictx->viewer));
1849: PetscFunctionReturn(PETSC_SUCCESS);
1850: }
1852: /*@C
1853: TaoDrawStepMonitor - Plots the step direction at each iteration of `TaoSolve()`
1855: Collective
1857: Input Parameters:
1858: + tao - the `Tao` context
1859: - ctx - the `PetscViewer` context
1861: Options Database Key:
1862: . -tao_draw_step - draw the step direction at each iteration
1864: Level: advanced
1866: .seealso: [](ch_tao), `Tao`, `TaoSetMonitor()`, `TaoDrawSolutionMonitor`
1867: @*/
1868: PetscErrorCode TaoDrawStepMonitor(Tao tao, void *ctx)
1869: {
1870: PetscViewer viewer = (PetscViewer)ctx;
1872: PetscFunctionBegin;
1875: PetscCall(VecView(tao->stepdirection, viewer));
1876: PetscFunctionReturn(PETSC_SUCCESS);
1877: }
1879: /*@C
1880: TaoResidualMonitor - Views the least-squares residual at each iteration of `TaoSolve()`
1882: Collective
1884: Input Parameters:
1885: + tao - the `Tao` context
1886: - ctx - the `PetscViewer` context or `NULL`
1888: Options Database Key:
1889: . -tao_view_ls_residual - view the residual at each iteration
1891: Level: advanced
1893: .seealso: [](ch_tao), `Tao`, `TaoDefaultSMonitor()`, `TaoSetMonitor()`
1894: @*/
1895: PetscErrorCode TaoResidualMonitor(Tao tao, void *ctx)
1896: {
1897: PetscViewer viewer = (PetscViewer)ctx;
1899: PetscFunctionBegin;
1902: PetscCall(VecView(tao->ls_res, viewer));
1903: PetscFunctionReturn(PETSC_SUCCESS);
1904: }
1906: /*@
1907: TaoDefaultConvergenceTest - Determines whether the solver should continue iterating
1908: or terminate.
1910: Collective
1912: Input Parameters:
1913: + tao - the `Tao` context
1914: - dummy - unused dummy context
1916: Level: developer
1918: Notes:
1919: This routine checks the residual in the optimality conditions, the
1920: relative residual in the optimity conditions, the number of function
1921: evaluations, and the function value to test convergence. Some
1922: solvers may use different convergence routines.
1924: .seealso: [](ch_tao), `Tao`, `TaoSetTolerances()`, `TaoGetConvergedReason()`, `TaoSetConvergedReason()`
1925: @*/
1926: PetscErrorCode TaoDefaultConvergenceTest(Tao tao, void *dummy)
1927: {
1928: PetscInt niter = tao->niter, nfuncs = PetscMax(tao->nfuncs, tao->nfuncgrads);
1929: PetscInt max_funcs = tao->max_funcs;
1930: PetscReal gnorm = tao->residual, gnorm0 = tao->gnorm0;
1931: PetscReal f = tao->fc, steptol = tao->steptol, trradius = tao->step;
1932: PetscReal gatol = tao->gatol, grtol = tao->grtol, gttol = tao->gttol;
1933: PetscReal catol = tao->catol, crtol = tao->crtol;
1934: PetscReal fmin = tao->fmin, cnorm = tao->cnorm;
1935: TaoConvergedReason reason = tao->reason;
1937: PetscFunctionBegin;
1939: if (reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(PETSC_SUCCESS);
1941: if (PetscIsInfOrNanReal(f)) {
1942: PetscCall(PetscInfo(tao, "Failed to converged, function value is Inf or NaN\n"));
1943: reason = TAO_DIVERGED_NAN;
1944: } else if (f <= fmin && cnorm <= catol) {
1945: PetscCall(PetscInfo(tao, "Converged due to function value %g < minimum function value %g\n", (double)f, (double)fmin));
1946: reason = TAO_CONVERGED_MINF;
1947: } else if (gnorm <= gatol && cnorm <= catol) {
1948: PetscCall(PetscInfo(tao, "Converged due to residual norm ||g(X)||=%g < %g\n", (double)gnorm, (double)gatol));
1949: reason = TAO_CONVERGED_GATOL;
1950: } else if (f != 0 && PetscAbsReal(gnorm / f) <= grtol && cnorm <= crtol) {
1951: PetscCall(PetscInfo(tao, "Converged due to residual ||g(X)||/|f(X)| =%g < %g\n", (double)(gnorm / f), (double)grtol));
1952: reason = TAO_CONVERGED_GRTOL;
1953: } else if (gnorm0 != 0 && ((gttol == 0 && gnorm == 0) || gnorm / gnorm0 < gttol) && cnorm <= crtol) {
1954: PetscCall(PetscInfo(tao, "Converged due to relative residual norm ||g(X)||/||g(X0)|| = %g < %g\n", (double)(gnorm / gnorm0), (double)gttol));
1955: reason = TAO_CONVERGED_GTTOL;
1956: } else if (max_funcs >= 0 && nfuncs > max_funcs) {
1957: PetscCall(PetscInfo(tao, "Exceeded maximum number of function evaluations: %" PetscInt_FMT " > %" PetscInt_FMT "\n", nfuncs, max_funcs));
1958: reason = TAO_DIVERGED_MAXFCN;
1959: } else if (tao->lsflag != 0) {
1960: PetscCall(PetscInfo(tao, "Tao Line Search failure.\n"));
1961: reason = TAO_DIVERGED_LS_FAILURE;
1962: } else if (trradius < steptol && niter > 0) {
1963: PetscCall(PetscInfo(tao, "Trust region/step size too small: %g < %g\n", (double)trradius, (double)steptol));
1964: reason = TAO_CONVERGED_STEPTOL;
1965: } else if (niter >= tao->max_it) {
1966: PetscCall(PetscInfo(tao, "Exceeded maximum number of iterations: %" PetscInt_FMT " > %" PetscInt_FMT "\n", niter, tao->max_it));
1967: reason = TAO_DIVERGED_MAXITS;
1968: } else {
1969: reason = TAO_CONTINUE_ITERATING;
1970: }
1971: tao->reason = reason;
1972: PetscFunctionReturn(PETSC_SUCCESS);
1973: }
1975: /*@C
1976: TaoSetOptionsPrefix - Sets the prefix used for searching for all
1977: Tao options in the database.
1979: Logically Collective
1981: Input Parameters:
1982: + tao - the `Tao` context
1983: - p - the prefix string to prepend to all Tao option requests
1985: Level: advanced
1987: Notes:
1988: A hyphen (-) must NOT be given at the beginning of the prefix name.
1989: The first character of all runtime options is AUTOMATICALLY the hyphen.
1991: For example, to distinguish between the runtime options for two
1992: different Tao solvers, one could call
1993: .vb
1994: TaoSetOptionsPrefix(tao1,"sys1_")
1995: TaoSetOptionsPrefix(tao2,"sys2_")
1996: .ve
1998: This would enable use of different options for each system, such as
1999: .vb
2000: -sys1_tao_method blmvm -sys1_tao_grtol 1.e-3
2001: -sys2_tao_method lmvm -sys2_tao_grtol 1.e-4
2002: .ve
2004: .seealso: [](ch_tao), `Tao`, `TaoSetFromOptions()`, `TaoAppendOptionsPrefix()`, `TaoGetOptionsPrefix()`
2005: @*/
2006: PetscErrorCode TaoSetOptionsPrefix(Tao tao, const char p[])
2007: {
2008: PetscFunctionBegin;
2010: PetscCall(PetscObjectSetOptionsPrefix((PetscObject)tao, p));
2011: if (tao->linesearch) PetscCall(TaoLineSearchSetOptionsPrefix(tao->linesearch, p));
2012: if (tao->ksp) PetscCall(KSPSetOptionsPrefix(tao->ksp, p));
2013: PetscFunctionReturn(PETSC_SUCCESS);
2014: }
2016: /*@C
2017: TaoAppendOptionsPrefix - Appends to the prefix used for searching for all Tao options in the database.
2019: Logically Collective
2021: Input Parameters:
2022: + tao - the `Tao` solver context
2023: - p - the prefix string to prepend to all `Tao` option requests
2025: Level: advanced
2027: Note:
2028: A hyphen (-) must NOT be given at the beginning of the prefix name.
2029: The first character of all runtime options is automatically the hyphen.
2031: .seealso: [](ch_tao), `Tao`, `TaoSetFromOptions()`, `TaoSetOptionsPrefix()`, `TaoGetOptionsPrefix()`
2032: @*/
2033: PetscErrorCode TaoAppendOptionsPrefix(Tao tao, const char p[])
2034: {
2035: PetscFunctionBegin;
2037: PetscCall(PetscObjectAppendOptionsPrefix((PetscObject)tao, p));
2038: if (tao->linesearch) PetscCall(PetscObjectAppendOptionsPrefix((PetscObject)tao->linesearch, p));
2039: if (tao->ksp) PetscCall(KSPAppendOptionsPrefix(tao->ksp, p));
2040: PetscFunctionReturn(PETSC_SUCCESS);
2041: }
2043: /*@C
2044: TaoGetOptionsPrefix - Gets the prefix used for searching for all
2045: Tao options in the database
2047: Not Collective
2049: Input Parameter:
2050: . tao - the `Tao` context
2052: Output Parameter:
2053: . p - pointer to the prefix string used is returned
2055: Fortran Notes:
2056: Pass in a string 'prefix' of sufficient length to hold the prefix.
2058: Level: advanced
2060: .seealso: [](ch_tao), `Tao`, `TaoSetFromOptions()`, `TaoSetOptionsPrefix()`, `TaoAppendOptionsPrefix()`
2061: @*/
2062: PetscErrorCode TaoGetOptionsPrefix(Tao tao, const char *p[])
2063: {
2064: PetscFunctionBegin;
2066: PetscCall(PetscObjectGetOptionsPrefix((PetscObject)tao, p));
2067: PetscFunctionReturn(PETSC_SUCCESS);
2068: }
2070: /*@C
2071: TaoSetType - Sets the `TaoType` for the minimization solver.
2073: Collective
2075: Input Parameters:
2076: + tao - the `Tao` solver context
2077: - type - a known method
2079: Options Database Key:
2080: . -tao_type <type> - Sets the method; use -help for a list
2081: of available methods (for instance, "-tao_type lmvm" or "-tao_type tron")
2083: Level: intermediate
2085: .seealso: [](ch_tao), `Tao`, `TaoCreate()`, `TaoGetType()`, `TaoType`
2086: @*/
2087: PetscErrorCode TaoSetType(Tao tao, TaoType type)
2088: {
2089: PetscErrorCode (*create_xxx)(Tao);
2090: PetscBool issame;
2092: PetscFunctionBegin;
2095: PetscCall(PetscObjectTypeCompare((PetscObject)tao, type, &issame));
2096: if (issame) PetscFunctionReturn(PETSC_SUCCESS);
2098: PetscCall(PetscFunctionListFind(TaoList, type, (void (**)(void)) & create_xxx));
2099: PetscCheck(create_xxx, PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_UNKNOWN_TYPE, "Unable to find requested Tao type %s", type);
2101: /* Destroy the existing solver information */
2102: PetscTryTypeMethod(tao, destroy);
2103: PetscCall(KSPDestroy(&tao->ksp));
2104: PetscCall(TaoLineSearchDestroy(&tao->linesearch));
2105: tao->ops->setup = NULL;
2106: tao->ops->solve = NULL;
2107: tao->ops->view = NULL;
2108: tao->ops->setfromoptions = NULL;
2109: tao->ops->destroy = NULL;
2111: tao->setupcalled = PETSC_FALSE;
2113: PetscCall((*create_xxx)(tao));
2114: PetscCall(PetscObjectChangeTypeName((PetscObject)tao, type));
2115: PetscFunctionReturn(PETSC_SUCCESS);
2116: }
2118: /*@C
2119: TaoRegister - Adds a method to the Tao package for minimization.
2121: Not Collective
2123: Input Parameters:
2124: + sname - name of a new user-defined solver
2125: - func - routine to Create method context
2127: Example Usage:
2128: .vb
2129: TaoRegister("my_solver", MySolverCreate);
2130: .ve
2132: Then, your solver can be chosen with the procedural interface via
2133: $ TaoSetType(tao, "my_solver")
2134: or at runtime via the option
2135: $ -tao_type my_solver
2137: Level: advanced
2139: Note:
2140: `TaoRegister()` may be called multiple times to add several user-defined solvers.
2142: .seealso: [](ch_tao), `Tao`, `TaoSetType()`, `TaoRegisterAll()`, `TaoRegisterDestroy()`
2143: @*/
2144: PetscErrorCode TaoRegister(const char sname[], PetscErrorCode (*func)(Tao))
2145: {
2146: PetscFunctionBegin;
2147: PetscCall(TaoInitializePackage());
2148: PetscCall(PetscFunctionListAdd(&TaoList, sname, (void (*)(void))func));
2149: PetscFunctionReturn(PETSC_SUCCESS);
2150: }
2152: /*@C
2153: TaoRegisterDestroy - Frees the list of minimization solvers that were
2154: registered by `TaoRegister()`.
2156: Not Collective
2158: Level: advanced
2160: .seealso: [](ch_tao), `Tao`, `TaoRegisterAll()`, `TaoRegister()`
2161: @*/
2162: PetscErrorCode TaoRegisterDestroy(void)
2163: {
2164: PetscFunctionBegin;
2165: PetscCall(PetscFunctionListDestroy(&TaoList));
2166: TaoRegisterAllCalled = PETSC_FALSE;
2167: PetscFunctionReturn(PETSC_SUCCESS);
2168: }
2170: /*@
2171: TaoGetIterationNumber - Gets the number of `TaoSolve()` iterations completed
2172: at this time.
2174: Not Collective
2176: Input Parameter:
2177: . tao - the `Tao` context
2179: Output Parameter:
2180: . iter - iteration number
2182: Notes:
2183: For example, during the computation of iteration 2 this would return 1.
2185: Level: intermediate
2187: .seealso: [](ch_tao), `Tao`, `TaoGetLinearSolveIterations()`, `TaoGetResidualNorm()`, `TaoGetObjective()`
2188: @*/
2189: PetscErrorCode TaoGetIterationNumber(Tao tao, PetscInt *iter)
2190: {
2191: PetscFunctionBegin;
2193: PetscAssertPointer(iter, 2);
2194: *iter = tao->niter;
2195: PetscFunctionReturn(PETSC_SUCCESS);
2196: }
2198: /*@
2199: TaoGetResidualNorm - Gets the current value of the norm of the residual (gradient)
2200: at this time.
2202: Not Collective
2204: Input Parameter:
2205: . tao - the `Tao` context
2207: Output Parameter:
2208: . value - the current value
2210: Level: intermediate
2212: Developer Notes:
2213: This is the 2-norm of the residual, we cannot use `TaoGetGradientNorm()` because that has
2214: a different meaning. For some reason `Tao` sometimes calls the gradient the residual.
2216: .seealso: [](ch_tao), `Tao`, `TaoGetLinearSolveIterations()`, `TaoGetIterationNumber()`, `TaoGetObjective()`
2217: @*/
2218: PetscErrorCode TaoGetResidualNorm(Tao tao, PetscReal *value)
2219: {
2220: PetscFunctionBegin;
2222: PetscAssertPointer(value, 2);
2223: *value = tao->residual;
2224: PetscFunctionReturn(PETSC_SUCCESS);
2225: }
2227: /*@
2228: TaoSetIterationNumber - Sets the current iteration number.
2230: Logically Collective
2232: Input Parameters:
2233: + tao - the `Tao` context
2234: - iter - iteration number
2236: Level: developer
2238: .seealso: [](ch_tao), `Tao`, `TaoGetLinearSolveIterations()`
2239: @*/
2240: PetscErrorCode TaoSetIterationNumber(Tao tao, PetscInt iter)
2241: {
2242: PetscFunctionBegin;
2245: PetscCall(PetscObjectSAWsTakeAccess((PetscObject)tao));
2246: tao->niter = iter;
2247: PetscCall(PetscObjectSAWsGrantAccess((PetscObject)tao));
2248: PetscFunctionReturn(PETSC_SUCCESS);
2249: }
2251: /*@
2252: TaoGetTotalIterationNumber - Gets the total number of `TaoSolve()` iterations
2253: completed. This number keeps accumulating if multiple solves
2254: are called with the `Tao` object.
2256: Not Collective
2258: Input Parameter:
2259: . tao - the `Tao` context
2261: Output Parameter:
2262: . iter - number of iterations
2264: Level: intermediate
2266: Note:
2267: The total iteration count is updated after each solve, if there is a current
2268: `TaoSolve()` in progress then those iterations are not included in the count
2270: .seealso: [](ch_tao), `Tao`, `TaoGetLinearSolveIterations()`
2271: @*/
2272: PetscErrorCode TaoGetTotalIterationNumber(Tao tao, PetscInt *iter)
2273: {
2274: PetscFunctionBegin;
2276: PetscAssertPointer(iter, 2);
2277: *iter = tao->ntotalits;
2278: PetscFunctionReturn(PETSC_SUCCESS);
2279: }
2281: /*@
2282: TaoSetTotalIterationNumber - Sets the current total iteration number.
2284: Logically Collective
2286: Input Parameters:
2287: + tao - the `Tao` context
2288: - iter - the iteration number
2290: Level: developer
2292: .seealso: [](ch_tao), `Tao`, `TaoGetLinearSolveIterations()`
2293: @*/
2294: PetscErrorCode TaoSetTotalIterationNumber(Tao tao, PetscInt iter)
2295: {
2296: PetscFunctionBegin;
2299: PetscCall(PetscObjectSAWsTakeAccess((PetscObject)tao));
2300: tao->ntotalits = iter;
2301: PetscCall(PetscObjectSAWsGrantAccess((PetscObject)tao));
2302: PetscFunctionReturn(PETSC_SUCCESS);
2303: }
2305: /*@
2306: TaoSetConvergedReason - Sets the termination flag on a `Tao` object
2308: Logically Collective
2310: Input Parameters:
2311: + tao - the `Tao` context
2312: - reason - the `TaoConvergedReason`
2314: Level: intermediate
2316: .seealso: [](ch_tao), `Tao`, `TaoConvergedReason`
2317: @*/
2318: PetscErrorCode TaoSetConvergedReason(Tao tao, TaoConvergedReason reason)
2319: {
2320: PetscFunctionBegin;
2323: tao->reason = reason;
2324: PetscFunctionReturn(PETSC_SUCCESS);
2325: }
2327: /*@
2328: TaoGetConvergedReason - Gets the reason the `TaoSolve()` was stopped.
2330: Not Collective
2332: Input Parameter:
2333: . tao - the `Tao` solver context
2335: Output Parameter:
2336: . reason - value of `TaoConvergedReason`
2338: Level: intermediate
2340: .seealso: [](ch_tao), `Tao`, `TaoConvergedReason`, `TaoSetConvergenceTest()`, `TaoSetTolerances()`
2341: @*/
2342: PetscErrorCode TaoGetConvergedReason(Tao tao, TaoConvergedReason *reason)
2343: {
2344: PetscFunctionBegin;
2346: PetscAssertPointer(reason, 2);
2347: *reason = tao->reason;
2348: PetscFunctionReturn(PETSC_SUCCESS);
2349: }
2351: /*@
2352: TaoGetSolutionStatus - Get the current iterate, objective value,
2353: residual, infeasibility, and termination from a `Tao` object
2355: Not Collective
2357: Input Parameter:
2358: . tao - the `Tao` context
2360: Output Parameters:
2361: + its - the current iterate number (>=0)
2362: . f - the current function value
2363: . gnorm - the square of the gradient norm, duality gap, or other measure indicating distance from optimality.
2364: . cnorm - the infeasibility of the current solution with regard to the constraints.
2365: . xdiff - the step length or trust region radius of the most recent iterate.
2366: - reason - The termination reason, which can equal `TAO_CONTINUE_ITERATING`
2368: Level: intermediate
2370: Notes:
2371: Tao returns the values set by the solvers in the routine `TaoMonitor()`.
2373: If any of the output arguments are set to `NULL`, no corresponding value will be returned.
2375: .seealso: [](ch_tao), `TaoMonitor()`, `TaoGetConvergedReason()`
2376: @*/
2377: PetscErrorCode TaoGetSolutionStatus(Tao tao, PetscInt *its, PetscReal *f, PetscReal *gnorm, PetscReal *cnorm, PetscReal *xdiff, TaoConvergedReason *reason)
2378: {
2379: PetscFunctionBegin;
2381: if (its) *its = tao->niter;
2382: if (f) *f = tao->fc;
2383: if (gnorm) *gnorm = tao->residual;
2384: if (cnorm) *cnorm = tao->cnorm;
2385: if (reason) *reason = tao->reason;
2386: if (xdiff) *xdiff = tao->step;
2387: PetscFunctionReturn(PETSC_SUCCESS);
2388: }
2390: /*@C
2391: TaoGetType - Gets the current `TaoType` being used in the `Tao` object
2393: Not Collective
2395: Input Parameter:
2396: . tao - the `Tao` solver context
2398: Output Parameter:
2399: . type - the `TaoType`
2401: Level: intermediate
2403: .seealso: [](ch_tao), `Tao`, `TaoType`, `TaoSetType()`
2404: @*/
2405: PetscErrorCode TaoGetType(Tao tao, TaoType *type)
2406: {
2407: PetscFunctionBegin;
2409: PetscAssertPointer(type, 2);
2410: *type = ((PetscObject)tao)->type_name;
2411: PetscFunctionReturn(PETSC_SUCCESS);
2412: }
2414: /*@C
2415: TaoMonitor - Monitor the solver and the current solution. This
2416: routine will record the iteration number and residual statistics,
2417: and call any monitors specified by the user.
2419: Input Parameters:
2420: + tao - the `Tao` context
2421: . its - the current iterate number (>=0)
2422: . f - the current objective function value
2423: . res - the gradient norm, square root of the duality gap, or other measure indicating distance from optimality. This measure will be recorded and
2424: used for some termination tests.
2425: . cnorm - the infeasibility of the current solution with regard to the constraints.
2426: - steplength - multiple of the step direction added to the previous iterate.
2428: Options Database Key:
2429: . -tao_monitor - Use the default monitor, which prints statistics to standard output
2431: Level: developer
2433: .seealso: [](ch_tao), `Tao`, `TaoGetConvergedReason()`, `TaoMonitorDefault()`, `TaoSetMonitor()`
2434: @*/
2435: PetscErrorCode TaoMonitor(Tao tao, PetscInt its, PetscReal f, PetscReal res, PetscReal cnorm, PetscReal steplength)
2436: {
2437: PetscInt i;
2439: PetscFunctionBegin;
2441: tao->fc = f;
2442: tao->residual = res;
2443: tao->cnorm = cnorm;
2444: tao->step = steplength;
2445: if (!its) {
2446: tao->cnorm0 = cnorm;
2447: tao->gnorm0 = res;
2448: }
2449: PetscCall(VecLockReadPush(tao->solution));
2450: for (i = 0; i < tao->numbermonitors; i++) PetscCall((*tao->monitor[i])(tao, tao->monitorcontext[i]));
2451: PetscCall(VecLockReadPop(tao->solution));
2452: PetscFunctionReturn(PETSC_SUCCESS);
2453: }
2455: /*@
2456: TaoSetConvergenceHistory - Sets the array used to hold the convergence history.
2458: Logically Collective
2460: Input Parameters:
2461: + tao - the `Tao` solver context
2462: . obj - array to hold objective value history
2463: . resid - array to hold residual history
2464: . cnorm - array to hold constraint violation history
2465: . lits - integer array holds the number of linear iterations for each Tao iteration
2466: . na - size of `obj`, `resid`, and `cnorm`
2467: - reset - `PETSC_TRUE` indicates each new minimization resets the history counter to zero,
2468: else it continues storing new values for new minimizations after the old ones
2470: Level: intermediate
2472: Notes:
2473: If set, `Tao` will fill the given arrays with the indicated
2474: information at each iteration. If 'obj','resid','cnorm','lits' are
2475: *all* `NULL` then space (using size `na`, or 1000 if na is `PETSC_DECIDE` or
2476: `PETSC_DEFAULT`) is allocated for the history.
2477: If not all are `NULL`, then only the non-`NULL` information categories
2478: will be stored, the others will be ignored.
2480: Any convergence information after iteration number 'na' will not be stored.
2482: This routine is useful, e.g., when running a code for purposes
2483: of accurate performance monitoring, when no I/O should be done
2484: during the section of code that is being timed.
2486: .seealso: [](ch_tao), `TaoGetConvergenceHistory()`
2487: @*/
2488: PetscErrorCode TaoSetConvergenceHistory(Tao tao, PetscReal obj[], PetscReal resid[], PetscReal cnorm[], PetscInt lits[], PetscInt na, PetscBool reset)
2489: {
2490: PetscFunctionBegin;
2492: if (obj) PetscAssertPointer(obj, 2);
2493: if (resid) PetscAssertPointer(resid, 3);
2494: if (cnorm) PetscAssertPointer(cnorm, 4);
2495: if (lits) PetscAssertPointer(lits, 5);
2497: if (na == PETSC_DECIDE || na == PETSC_DEFAULT) na = 1000;
2498: if (!obj && !resid && !cnorm && !lits) {
2499: PetscCall(PetscCalloc4(na, &obj, na, &resid, na, &cnorm, na, &lits));
2500: tao->hist_malloc = PETSC_TRUE;
2501: }
2503: tao->hist_obj = obj;
2504: tao->hist_resid = resid;
2505: tao->hist_cnorm = cnorm;
2506: tao->hist_lits = lits;
2507: tao->hist_max = na;
2508: tao->hist_reset = reset;
2509: tao->hist_len = 0;
2510: PetscFunctionReturn(PETSC_SUCCESS);
2511: }
2513: /*@C
2514: TaoGetConvergenceHistory - Gets the arrays used that hold the convergence history.
2516: Collective
2518: Input Parameter:
2519: . tao - the `Tao` context
2521: Output Parameters:
2522: + obj - array used to hold objective value history
2523: . resid - array used to hold residual history
2524: . cnorm - array used to hold constraint violation history
2525: . lits - integer array used to hold linear solver iteration count
2526: - nhist - size of `obj`, `resid`, `cnorm`, and `lits`
2528: Level: advanced
2530: Notes:
2531: This routine must be preceded by calls to `TaoSetConvergenceHistory()`
2532: and `TaoSolve()`, otherwise it returns useless information.
2534: This routine is useful, e.g., when running a code for purposes
2535: of accurate performance monitoring, when no I/O should be done
2536: during the section of code that is being timed.
2538: Fortran Notes:
2539: The calling sequence is
2540: .vb
2541: call TaoGetConvergenceHistory(Tao tao, PetscInt nhist, PetscErrorCode ierr)
2542: .ve
2544: .seealso: [](ch_tao), `Tao`, `TaoSolve()`, `TaoSetConvergenceHistory()`
2545: @*/
2546: PetscErrorCode TaoGetConvergenceHistory(Tao tao, PetscReal **obj, PetscReal **resid, PetscReal **cnorm, PetscInt **lits, PetscInt *nhist)
2547: {
2548: PetscFunctionBegin;
2550: if (obj) *obj = tao->hist_obj;
2551: if (cnorm) *cnorm = tao->hist_cnorm;
2552: if (resid) *resid = tao->hist_resid;
2553: if (lits) *lits = tao->hist_lits;
2554: if (nhist) *nhist = tao->hist_len;
2555: PetscFunctionReturn(PETSC_SUCCESS);
2556: }
2558: /*@
2559: TaoSetApplicationContext - Sets the optional user-defined context for a `Tao` solver.
2561: Logically Collective
2563: Input Parameters:
2564: + tao - the `Tao` context
2565: - usrP - optional user context
2567: Level: intermediate
2569: .seealso: [](ch_tao), `Tao`, `TaoGetApplicationContext()`
2570: @*/
2571: PetscErrorCode TaoSetApplicationContext(Tao tao, void *usrP)
2572: {
2573: PetscFunctionBegin;
2575: tao->user = usrP;
2576: PetscFunctionReturn(PETSC_SUCCESS);
2577: }
2579: /*@
2580: TaoGetApplicationContext - Gets the user-defined context for a `Tao` solver
2582: Not Collective
2584: Input Parameter:
2585: . tao - the `Tao` context
2587: Output Parameter:
2588: . usrP - user context
2590: Level: intermediate
2592: .seealso: [](ch_tao), `Tao`, `TaoSetApplicationContext()`
2593: @*/
2594: PetscErrorCode TaoGetApplicationContext(Tao tao, void *usrP)
2595: {
2596: PetscFunctionBegin;
2598: PetscAssertPointer(usrP, 2);
2599: *(void **)usrP = tao->user;
2600: PetscFunctionReturn(PETSC_SUCCESS);
2601: }
2603: /*@
2604: TaoSetGradientNorm - Sets the matrix used to define the norm that measures the size of the gradient.
2606: Collective
2608: Input Parameters:
2609: + tao - the `Tao` context
2610: - M - matrix that defines the norm
2612: Level: beginner
2614: .seealso: [](ch_tao), `Tao`, `TaoGetGradientNorm()`, `TaoGradientNorm()`
2615: @*/
2616: PetscErrorCode TaoSetGradientNorm(Tao tao, Mat M)
2617: {
2618: PetscFunctionBegin;
2621: PetscCall(PetscObjectReference((PetscObject)M));
2622: PetscCall(MatDestroy(&tao->gradient_norm));
2623: PetscCall(VecDestroy(&tao->gradient_norm_tmp));
2624: tao->gradient_norm = M;
2625: PetscCall(MatCreateVecs(M, NULL, &tao->gradient_norm_tmp));
2626: PetscFunctionReturn(PETSC_SUCCESS);
2627: }
2629: /*@
2630: TaoGetGradientNorm - Returns the matrix used to define the norm used for measuring the size of the gradient.
2632: Not Collective
2634: Input Parameter:
2635: . tao - the `Tao` context
2637: Output Parameter:
2638: . M - gradient norm
2640: Level: beginner
2642: .seealso: [](ch_tao), `Tao`, `TaoSetGradientNorm()`, `TaoGradientNorm()`
2643: @*/
2644: PetscErrorCode TaoGetGradientNorm(Tao tao, Mat *M)
2645: {
2646: PetscFunctionBegin;
2648: PetscAssertPointer(M, 2);
2649: *M = tao->gradient_norm;
2650: PetscFunctionReturn(PETSC_SUCCESS);
2651: }
2653: /*@C
2654: TaoGradientNorm - Compute the norm using the `NormType`, the user has selected
2656: Collective
2658: Input Parameters:
2659: + tao - the `Tao` context
2660: . gradient - the gradient to be computed
2661: - type - the norm type
2663: Output Parameter:
2664: . gnorm - the gradient norm
2666: Level: advanced
2668: .seealso: [](ch_tao), `Tao`, `TaoSetGradientNorm()`, `TaoGetGradientNorm()`
2669: @*/
2670: PetscErrorCode TaoGradientNorm(Tao tao, Vec gradient, NormType type, PetscReal *gnorm)
2671: {
2672: PetscFunctionBegin;
2676: PetscAssertPointer(gnorm, 4);
2677: if (tao->gradient_norm) {
2678: PetscScalar gnorms;
2680: PetscCheck(type == NORM_2, PetscObjectComm((PetscObject)gradient), PETSC_ERR_ARG_WRONG, "Norm type must be NORM_2 if an inner product for the gradient norm is set.");
2681: PetscCall(MatMult(tao->gradient_norm, gradient, tao->gradient_norm_tmp));
2682: PetscCall(VecDot(gradient, tao->gradient_norm_tmp, &gnorms));
2683: *gnorm = PetscRealPart(PetscSqrtScalar(gnorms));
2684: } else {
2685: PetscCall(VecNorm(gradient, type, gnorm));
2686: }
2687: PetscFunctionReturn(PETSC_SUCCESS);
2688: }
2690: /*@C
2691: TaoMonitorDrawCtxCreate - Creates the monitor context for `TaoMonitorDrawSolution()`
2693: Collective
2695: Input Parameters:
2696: + comm - the communicator to share the context
2697: . host - the name of the X Windows host that will display the monitor
2698: . label - the label to put at the top of the display window
2699: . x - the horizontal coordinate of the lower left corner of the window to open
2700: . y - the vertical coordinate of the lower left corner of the window to open
2701: . m - the width of the window
2702: . n - the height of the window
2703: - howoften - how many `Tao` iterations between displaying the monitor information
2705: Output Parameter:
2706: . ctx - the monitor context
2708: Options Database Key:
2709: . -tao_draw_solution_initial - show initial guess as well as current solution
2711: Level: intermediate
2713: .seealso: [](ch_tao), `Tao`, `TaoMonitorSet()`, `TaoMonitorDefault()`, `VecView()`, `TaoMonitorDrawCtx()`
2714: @*/
2715: PetscErrorCode TaoMonitorDrawCtxCreate(MPI_Comm comm, const char host[], const char label[], int x, int y, int m, int n, PetscInt howoften, TaoMonitorDrawCtx *ctx)
2716: {
2717: PetscFunctionBegin;
2718: PetscCall(PetscNew(ctx));
2719: PetscCall(PetscViewerDrawOpen(comm, host, label, x, y, m, n, &(*ctx)->viewer));
2720: PetscCall(PetscViewerSetFromOptions((*ctx)->viewer));
2721: (*ctx)->howoften = howoften;
2722: PetscFunctionReturn(PETSC_SUCCESS);
2723: }
2725: /*@C
2726: TaoMonitorDrawCtxDestroy - Destroys the monitor context for `TaoMonitorDrawSolution()`
2728: Collective
2730: Input Parameter:
2731: . ictx - the monitor context
2733: Level: intermediate
2735: .seealso: [](ch_tao), `Tao`, `TaoMonitorSet()`, `TaoMonitorDefault()`, `VecView()`, `TaoMonitorDrawSolution()`
2736: @*/
2737: PetscErrorCode TaoMonitorDrawCtxDestroy(TaoMonitorDrawCtx *ictx)
2738: {
2739: PetscFunctionBegin;
2740: PetscCall(PetscViewerDestroy(&(*ictx)->viewer));
2741: PetscCall(PetscFree(*ictx));
2742: PetscFunctionReturn(PETSC_SUCCESS);
2743: }