Actual source code: blmvm.c
1: #include <petsctaolinesearch.h>
2: #include <../src/tao/unconstrained/impls/lmvm/lmvm.h>
3: #include <../src/tao/bound/impls/blmvm/blmvm.h>
5: /*------------------------------------------------------------*/
6: static PetscErrorCode TaoSolve_BLMVM(Tao tao)
7: {
8: PetscErrorCode ierr;
9: TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data;
10: TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING;
11: PetscReal f, fold, gdx, gnorm, gnorm2;
12: PetscReal stepsize = 1.0,delta;
15: /* Project initial point onto bounds */
16: TaoComputeVariableBounds(tao);
17: VecMedian(tao->XL,tao->solution,tao->XU,tao->solution);
18: TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);
21: /* Check convergence criteria */
22: TaoComputeObjectiveAndGradient(tao, tao->solution,&f,blmP->unprojected_gradient);
23: VecBoundGradientProjection(blmP->unprojected_gradient,tao->solution, tao->XL,tao->XU,tao->gradient);
25: TaoGradientNorm(tao, tao->gradient,NORM_2,&gnorm);
26: if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PetscObjectComm((PetscObject)tao),PETSC_ERR_USER, "User provided compute function generated Inf or NaN");
28: tao->reason = TAO_CONTINUE_ITERATING;
29: TaoLogConvergenceHistory(tao,f,gnorm,0.0,tao->ksp_its);
30: TaoMonitor(tao,tao->niter,f,gnorm,0.0,stepsize);
31: (*tao->ops->convergencetest)(tao,tao->cnvP);
32: if (tao->reason != TAO_CONTINUE_ITERATING) return(0);
34: /* Set counter for gradient/reset steps */
35: if (!blmP->recycle) {
36: blmP->grad = 0;
37: blmP->reset = 0;
38: MatLMVMReset(blmP->M, PETSC_FALSE);
39: }
41: /* Have not converged; continue with Newton method */
42: while (tao->reason == TAO_CONTINUE_ITERATING) {
43: /* Call general purpose update function */
44: if (tao->ops->update) {
45: (*tao->ops->update)(tao, tao->niter, tao->user_update);
46: }
47: /* Compute direction */
48: gnorm2 = gnorm*gnorm;
49: if (gnorm2 == 0.0) gnorm2 = PETSC_MACHINE_EPSILON;
50: if (f == 0.0) {
51: delta = 2.0 / gnorm2;
52: } else {
53: delta = 2.0 * PetscAbsScalar(f) / gnorm2;
54: }
55: MatLMVMSymBroydenSetDelta(blmP->M, delta);
56: MatLMVMUpdate(blmP->M, tao->solution, tao->gradient);
57: MatSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);
58: VecBoundGradientProjection(tao->stepdirection,tao->solution,tao->XL,tao->XU,tao->gradient);
60: /* Check for success (descent direction) */
61: VecDot(blmP->unprojected_gradient, tao->gradient, &gdx);
62: if (gdx <= 0) {
63: /* Step is not descent or solve was not successful
64: Use steepest descent direction (scaled) */
65: ++blmP->grad;
67: MatLMVMReset(blmP->M, PETSC_FALSE);
68: MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);
69: MatSolve(blmP->M,blmP->unprojected_gradient, tao->stepdirection);
70: }
71: VecScale(tao->stepdirection,-1.0);
73: /* Perform the linesearch */
74: fold = f;
75: VecCopy(tao->solution, blmP->Xold);
76: VecCopy(blmP->unprojected_gradient, blmP->Gold);
77: TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);
78: TaoLineSearchApply(tao->linesearch, tao->solution, &f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);
79: TaoAddLineSearchCounts(tao);
81: if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) {
82: /* Linesearch failed
83: Reset factors and use scaled (projected) gradient step */
84: ++blmP->reset;
86: f = fold;
87: VecCopy(blmP->Xold, tao->solution);
88: VecCopy(blmP->Gold, blmP->unprojected_gradient);
90: MatLMVMReset(blmP->M, PETSC_FALSE);
91: MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);
92: MatSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);
93: VecScale(tao->stepdirection, -1.0);
95: /* This may be incorrect; linesearch has values for stepmax and stepmin
96: that should be reset. */
97: TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);
98: TaoLineSearchApply(tao->linesearch,tao->solution,&f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);
99: TaoAddLineSearchCounts(tao);
101: if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) {
102: tao->reason = TAO_DIVERGED_LS_FAILURE;
103: break;
104: }
105: }
107: /* Check for converged */
108: VecBoundGradientProjection(blmP->unprojected_gradient, tao->solution, tao->XL, tao->XU, tao->gradient);
109: TaoGradientNorm(tao, tao->gradient, NORM_2, &gnorm);
110: if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PetscObjectComm((PetscObject)tao),PETSC_ERR_USER, "User provided compute function generated Not-a-Number");
111: tao->niter++;
112: TaoLogConvergenceHistory(tao,f,gnorm,0.0,tao->ksp_its);
113: TaoMonitor(tao,tao->niter,f,gnorm,0.0,stepsize);
114: (*tao->ops->convergencetest)(tao,tao->cnvP);
115: }
116: return(0);
117: }
119: static PetscErrorCode TaoSetup_BLMVM(Tao tao)
120: {
121: TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data;
125: /* Existence of tao->solution checked in TaoSetup() */
126: VecDuplicate(tao->solution,&blmP->Xold);
127: VecDuplicate(tao->solution,&blmP->Gold);
128: VecDuplicate(tao->solution, &blmP->unprojected_gradient);
130: if (!tao->stepdirection) {
131: VecDuplicate(tao->solution, &tao->stepdirection);
132: }
133: if (!tao->gradient) {
134: VecDuplicate(tao->solution,&tao->gradient);
135: }
136: if (!tao->XL) {
137: VecDuplicate(tao->solution,&tao->XL);
138: VecSet(tao->XL,PETSC_NINFINITY);
139: }
140: if (!tao->XU) {
141: VecDuplicate(tao->solution,&tao->XU);
142: VecSet(tao->XU,PETSC_INFINITY);
143: }
144: /* Allocate matrix for the limited memory approximation */
145: MatLMVMAllocate(blmP->M,tao->solution,blmP->unprojected_gradient);
147: /* If the user has set a matrix to solve as the initial H0, set the options prefix here, and set up the KSP */
148: if (blmP->H0) {
149: MatLMVMSetJ0(blmP->M, blmP->H0);
150: }
151: return(0);
152: }
154: /* ---------------------------------------------------------- */
155: static PetscErrorCode TaoDestroy_BLMVM(Tao tao)
156: {
157: TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data;
161: if (tao->setupcalled) {
162: VecDestroy(&blmP->unprojected_gradient);
163: VecDestroy(&blmP->Xold);
164: VecDestroy(&blmP->Gold);
165: }
166: MatDestroy(&blmP->M);
167: if (blmP->H0) {
168: PetscObjectDereference((PetscObject)blmP->H0);
169: }
170: PetscFree(tao->data);
171: return(0);
172: }
174: /*------------------------------------------------------------*/
175: static PetscErrorCode TaoSetFromOptions_BLMVM(PetscOptionItems* PetscOptionsObject,Tao tao)
176: {
177: TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data;
179: PetscBool is_spd;
182: PetscOptionsHead(PetscOptionsObject,"Limited-memory variable-metric method for bound constrained optimization");
183: PetscOptionsBool("-tao_blmvm_recycle","enable recycling of the BFGS matrix between subsequent TaoSolve() calls","",blmP->recycle,&blmP->recycle,NULL);
184: PetscOptionsTail();
185: TaoLineSearchSetFromOptions(tao->linesearch);
186: MatSetFromOptions(blmP->M);
187: MatGetOption(blmP->M, MAT_SPD, &is_spd);
188: if (!is_spd) SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_INCOMP, "LMVM matrix must be symmetric positive-definite");
189: return(0);
190: }
193: /*------------------------------------------------------------*/
194: static PetscErrorCode TaoView_BLMVM(Tao tao, PetscViewer viewer)
195: {
196: TAO_BLMVM *lmP = (TAO_BLMVM *)tao->data;
197: PetscBool isascii;
201: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii);
202: if (isascii) {
203: PetscViewerASCIIPrintf(viewer, "Gradient steps: %D\n", lmP->grad);
204: PetscViewerPushFormat(viewer, PETSC_VIEWER_ASCII_INFO);
205: MatView(lmP->M, viewer);
206: PetscViewerPopFormat(viewer);
207: }
208: return(0);
209: }
211: static PetscErrorCode TaoComputeDual_BLMVM(Tao tao, Vec DXL, Vec DXU)
212: {
213: TAO_BLMVM *blm = (TAO_BLMVM *) tao->data;
220: if (!tao->gradient || !blm->unprojected_gradient) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Dual variables don't exist yet or no longer exist.\n");
222: VecCopy(tao->gradient,DXL);
223: VecAXPY(DXL,-1.0,blm->unprojected_gradient);
224: VecSet(DXU,0.0);
225: VecPointwiseMax(DXL,DXL,DXU);
227: VecCopy(blm->unprojected_gradient,DXU);
228: VecAXPY(DXU,-1.0,tao->gradient);
229: VecAXPY(DXU,1.0,DXL);
230: return(0);
231: }
233: /* ---------------------------------------------------------- */
234: /*MC
235: TAOBLMVM - Bounded limited memory variable metric is a quasi-Newton method
236: for nonlinear minimization with bound constraints. It is an extension
237: of TAOLMVM
239: Options Database Keys:
240: . -tao_lmm_recycle - enable recycling of LMVM information between subsequent TaoSolve calls
242: Level: beginner
243: M*/
244: PETSC_EXTERN PetscErrorCode TaoCreate_BLMVM(Tao tao)
245: {
246: TAO_BLMVM *blmP;
247: const char *morethuente_type = TAOLINESEARCHMT;
251: tao->ops->setup = TaoSetup_BLMVM;
252: tao->ops->solve = TaoSolve_BLMVM;
253: tao->ops->view = TaoView_BLMVM;
254: tao->ops->setfromoptions = TaoSetFromOptions_BLMVM;
255: tao->ops->destroy = TaoDestroy_BLMVM;
256: tao->ops->computedual = TaoComputeDual_BLMVM;
258: PetscNewLog(tao,&blmP);
259: blmP->H0 = NULL;
260: blmP->recycle = PETSC_FALSE;
261: tao->data = (void*)blmP;
263: /* Override default settings (unless already changed) */
264: if (!tao->max_it_changed) tao->max_it = 2000;
265: if (!tao->max_funcs_changed) tao->max_funcs = 4000;
267: TaoLineSearchCreate(((PetscObject)tao)->comm, &tao->linesearch);
268: PetscObjectIncrementTabLevel((PetscObject)tao->linesearch, (PetscObject)tao, 1);
269: TaoLineSearchSetType(tao->linesearch, morethuente_type);
270: TaoLineSearchUseTaoRoutines(tao->linesearch,tao);
271: TaoLineSearchSetOptionsPrefix(tao->linesearch,tao->hdr.prefix);
273: KSPInitializePackage();
274: MatCreate(((PetscObject)tao)->comm, &blmP->M);
275: MatSetType(blmP->M, MATLMVMBFGS);
276: PetscObjectIncrementTabLevel((PetscObject)blmP->M, (PetscObject)tao, 1);
277: MatSetOptionsPrefix(blmP->M, "tao_blmvm_");
278: return(0);
279: }
281: /*@
282: TaoLMVMRecycle - Enable/disable recycling of the QN history between subsequent TaoSolve calls.
284: Input Parameters:
285: + tao - the Tao solver context
286: - flg - Boolean flag for recycling (PETSC_TRUE or PETSC_FALSE)
288: Level: intermediate
289: @*/
290: PetscErrorCode TaoLMVMRecycle(Tao tao, PetscBool flg)
291: {
292: TAO_LMVM *lmP;
293: TAO_BLMVM *blmP;
294: PetscBool is_lmvm, is_blmvm;
298: PetscObjectTypeCompare((PetscObject)tao,TAOLMVM,&is_lmvm);
299: PetscObjectTypeCompare((PetscObject)tao,TAOBLMVM,&is_blmvm);
300: if (is_lmvm) {
301: lmP = (TAO_LMVM *)tao->data;
302: lmP->recycle = flg;
303: } else if (is_blmvm) {
304: blmP = (TAO_BLMVM *)tao->data;
305: blmP->recycle = flg;
306: }
307: return(0);
308: }
310: /*@
311: TaoLMVMSetH0 - Set the initial Hessian for the QN approximation
313: Input Parameters:
314: + tao - the Tao solver context
315: - H0 - Mat object for the initial Hessian
317: Level: advanced
319: .seealso: TaoLMVMGetH0(), TaoLMVMGetH0KSP()
320: @*/
321: PetscErrorCode TaoLMVMSetH0(Tao tao, Mat H0)
322: {
323: TAO_LMVM *lmP;
324: TAO_BLMVM *blmP;
325: PetscBool is_lmvm, is_blmvm;
329: PetscObjectTypeCompare((PetscObject)tao,TAOLMVM,&is_lmvm);
330: PetscObjectTypeCompare((PetscObject)tao,TAOBLMVM,&is_blmvm);
331: if (is_lmvm) {
332: lmP = (TAO_LMVM *)tao->data;
333: PetscObjectReference((PetscObject)H0);
334: lmP->H0 = H0;
335: } else if (is_blmvm) {
336: blmP = (TAO_BLMVM *)tao->data;
337: PetscObjectReference((PetscObject)H0);
338: blmP->H0 = H0;
339: }
340: return(0);
341: }
343: /*@
344: TaoLMVMGetH0 - Get the matrix object for the QN initial Hessian
346: Input Parameters:
347: . tao - the Tao solver context
349: Output Parameters:
350: . H0 - Mat object for the initial Hessian
352: Level: advanced
354: .seealso: TaoLMVMSetH0(), TaoLMVMGetH0KSP()
355: @*/
356: PetscErrorCode TaoLMVMGetH0(Tao tao, Mat *H0)
357: {
358: TAO_LMVM *lmP;
359: TAO_BLMVM *blmP;
360: PetscBool is_lmvm, is_blmvm;
361: Mat M;
365: PetscObjectTypeCompare((PetscObject)tao,TAOLMVM,&is_lmvm);
366: PetscObjectTypeCompare((PetscObject)tao,TAOBLMVM,&is_blmvm);
367: if (is_lmvm) {
368: lmP = (TAO_LMVM *)tao->data;
369: M = lmP->M;
370: } else if (is_blmvm) {
371: blmP = (TAO_BLMVM *)tao->data;
372: M = blmP->M;
373: } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONG, "This routine applies to TAO_LMVM and TAO_BLMVM.");
374: MatLMVMGetJ0(M, H0);
375: return(0);
376: }
378: /*@
379: TaoLMVMGetH0KSP - Get the iterative solver for applying the inverse of the QN initial Hessian
381: Input Parameters:
382: . tao - the Tao solver context
384: Output Parameters:
385: . ksp - KSP solver context for the initial Hessian
387: Level: advanced
389: .seealso: TaoLMVMGetH0(), TaoLMVMGetH0KSP()
390: @*/
391: PetscErrorCode TaoLMVMGetH0KSP(Tao tao, KSP *ksp)
392: {
393: TAO_LMVM *lmP;
394: TAO_BLMVM *blmP;
395: PetscBool is_lmvm, is_blmvm;
396: Mat M;
400: PetscObjectTypeCompare((PetscObject)tao,TAOLMVM,&is_lmvm);
401: PetscObjectTypeCompare((PetscObject)tao,TAOBLMVM,&is_blmvm);
402: if (is_lmvm) {
403: lmP = (TAO_LMVM *)tao->data;
404: M = lmP->M;
405: } else if (is_blmvm) {
406: blmP = (TAO_BLMVM *)tao->data;
407: M = blmP->M;
408: } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONG, "This routine applies to TAO_LMVM and TAO_BLMVM.");
409: MatLMVMGetJ0KSP(M, ksp);
410: return(0);
411: }