xref: /petsc/src/tao/bound/impls/blmvm/blmvm.c (revision 76be6f4ff3bd4e251c19fc00ebbebfd58b6e7589)
1 #include <petsctaolinesearch.h>      /*I "petsctaolinesearch.h" I*/
2 #include <../src/tao/unconstrained/impls/lmvm/lmvm.h>
3 #include <../src/tao/bound/impls/blmvm/blmvm.h>
4 
5 /*------------------------------------------------------------*/
6 static PetscErrorCode TaoSolve_BLMVM(Tao tao)
7 {
8   TAO_BLMVM                    *blmP = (TAO_BLMVM *)tao->data;
9   TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING;
10   PetscReal                    f, fold, gdx, gnorm, gnorm2;
11   PetscReal                    stepsize = 1.0,delta;
12 
13   PetscFunctionBegin;
14   /*  Project initial point onto bounds */
15   PetscCall(TaoComputeVariableBounds(tao));
16   PetscCall(VecMedian(tao->XL,tao->solution,tao->XU,tao->solution));
17   PetscCall(TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU));
18 
19   /* Check convergence criteria */
20   PetscCall(TaoComputeObjectiveAndGradient(tao, tao->solution,&f,blmP->unprojected_gradient));
21   PetscCall(VecBoundGradientProjection(blmP->unprojected_gradient,tao->solution, tao->XL,tao->XU,tao->gradient));
22 
23   PetscCall(TaoGradientNorm(tao, tao->gradient,NORM_2,&gnorm));
24   PetscCheck(!PetscIsInfOrNanReal(f) && !PetscIsInfOrNanReal(gnorm),PetscObjectComm((PetscObject)tao),PETSC_ERR_USER, "User provided compute function generated Inf or NaN");
25 
26   tao->reason = TAO_CONTINUE_ITERATING;
27   PetscCall(TaoLogConvergenceHistory(tao,f,gnorm,0.0,tao->ksp_its));
28   PetscCall(TaoMonitor(tao,tao->niter,f,gnorm,0.0,stepsize));
29   PetscCall((*tao->ops->convergencetest)(tao,tao->cnvP));
30   if (tao->reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0);
31 
32   /* Set counter for gradient/reset steps */
33   if (!blmP->recycle) {
34     blmP->grad = 0;
35     blmP->reset = 0;
36     PetscCall(MatLMVMReset(blmP->M, PETSC_FALSE));
37   }
38 
39   /* Have not converged; continue with Newton method */
40   while (tao->reason == TAO_CONTINUE_ITERATING) {
41     /* Call general purpose update function */
42     if (tao->ops->update) {
43       PetscCall((*tao->ops->update)(tao, tao->niter, tao->user_update));
44     }
45     /* Compute direction */
46     gnorm2 = gnorm*gnorm;
47     if (gnorm2 == 0.0) gnorm2 = PETSC_MACHINE_EPSILON;
48     if (f == 0.0) {
49       delta = 2.0 / gnorm2;
50     } else {
51       delta = 2.0 * PetscAbsScalar(f) / gnorm2;
52     }
53     PetscCall(MatLMVMSymBroydenSetDelta(blmP->M, delta));
54     PetscCall(MatLMVMUpdate(blmP->M, tao->solution, tao->gradient));
55     PetscCall(MatSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection));
56     PetscCall(VecBoundGradientProjection(tao->stepdirection,tao->solution,tao->XL,tao->XU,tao->gradient));
57 
58     /* Check for success (descent direction) */
59     PetscCall(VecDot(blmP->unprojected_gradient, tao->gradient, &gdx));
60     if (gdx <= 0) {
61       /* Step is not descent or solve was not successful
62          Use steepest descent direction (scaled) */
63       ++blmP->grad;
64 
65       PetscCall(MatLMVMReset(blmP->M, PETSC_FALSE));
66       PetscCall(MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient));
67       PetscCall(MatSolve(blmP->M,blmP->unprojected_gradient, tao->stepdirection));
68     }
69     PetscCall(VecScale(tao->stepdirection,-1.0));
70 
71     /* Perform the linesearch */
72     fold = f;
73     PetscCall(VecCopy(tao->solution, blmP->Xold));
74     PetscCall(VecCopy(blmP->unprojected_gradient, blmP->Gold));
75     PetscCall(TaoLineSearchSetInitialStepLength(tao->linesearch,1.0));
76     PetscCall(TaoLineSearchApply(tao->linesearch, tao->solution, &f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status));
77     PetscCall(TaoAddLineSearchCounts(tao));
78 
79     if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) {
80       /* Linesearch failed
81          Reset factors and use scaled (projected) gradient step */
82       ++blmP->reset;
83 
84       f = fold;
85       PetscCall(VecCopy(blmP->Xold, tao->solution));
86       PetscCall(VecCopy(blmP->Gold, blmP->unprojected_gradient));
87 
88       PetscCall(MatLMVMReset(blmP->M, PETSC_FALSE));
89       PetscCall(MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient));
90       PetscCall(MatSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection));
91       PetscCall(VecScale(tao->stepdirection, -1.0));
92 
93       /* This may be incorrect; linesearch has values for stepmax and stepmin
94          that should be reset. */
95       PetscCall(TaoLineSearchSetInitialStepLength(tao->linesearch,1.0));
96       PetscCall(TaoLineSearchApply(tao->linesearch,tao->solution,&f, blmP->unprojected_gradient, tao->stepdirection,  &stepsize, &ls_status));
97       PetscCall(TaoAddLineSearchCounts(tao));
98 
99       if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) {
100         tao->reason = TAO_DIVERGED_LS_FAILURE;
101         break;
102       }
103     }
104 
105     /* Check for converged */
106     PetscCall(VecBoundGradientProjection(blmP->unprojected_gradient, tao->solution, tao->XL, tao->XU, tao->gradient));
107     PetscCall(TaoGradientNorm(tao, tao->gradient, NORM_2, &gnorm));
108     PetscCheck(!PetscIsInfOrNanReal(f) && !PetscIsInfOrNanReal(gnorm),PetscObjectComm((PetscObject)tao),PETSC_ERR_USER, "User provided compute function generated Not-a-Number");
109     tao->niter++;
110     PetscCall(TaoLogConvergenceHistory(tao,f,gnorm,0.0,tao->ksp_its));
111     PetscCall(TaoMonitor(tao,tao->niter,f,gnorm,0.0,stepsize));
112     PetscCall((*tao->ops->convergencetest)(tao,tao->cnvP));
113   }
114   PetscFunctionReturn(0);
115 }
116 
117 static PetscErrorCode TaoSetup_BLMVM(Tao tao)
118 {
119   TAO_BLMVM      *blmP = (TAO_BLMVM *)tao->data;
120 
121   PetscFunctionBegin;
122   /* Existence of tao->solution checked in TaoSetup() */
123   PetscCall(VecDuplicate(tao->solution,&blmP->Xold));
124   PetscCall(VecDuplicate(tao->solution,&blmP->Gold));
125   PetscCall(VecDuplicate(tao->solution, &blmP->unprojected_gradient));
126   if (!tao->stepdirection) {
127     PetscCall(VecDuplicate(tao->solution, &tao->stepdirection));
128   }
129   if (!tao->gradient) {
130     PetscCall(VecDuplicate(tao->solution,&tao->gradient));
131   }
132   /* Allocate matrix for the limited memory approximation */
133   PetscCall(MatLMVMAllocate(blmP->M,tao->solution,blmP->unprojected_gradient));
134 
135   /* If the user has set a matrix to solve as the initial H0, set the options prefix here, and set up the KSP */
136   if (blmP->H0) {
137     PetscCall(MatLMVMSetJ0(blmP->M, blmP->H0));
138   }
139   PetscFunctionReturn(0);
140 }
141 
142 /* ---------------------------------------------------------- */
143 static PetscErrorCode TaoDestroy_BLMVM(Tao tao)
144 {
145   TAO_BLMVM      *blmP = (TAO_BLMVM *)tao->data;
146 
147   PetscFunctionBegin;
148   if (tao->setupcalled) {
149     PetscCall(VecDestroy(&blmP->unprojected_gradient));
150     PetscCall(VecDestroy(&blmP->Xold));
151     PetscCall(VecDestroy(&blmP->Gold));
152   }
153   PetscCall(MatDestroy(&blmP->M));
154   if (blmP->H0) {
155     PetscObjectDereference((PetscObject)blmP->H0);
156   }
157   PetscCall(PetscFree(tao->data));
158   PetscFunctionReturn(0);
159 }
160 
161 /*------------------------------------------------------------*/
162 static PetscErrorCode TaoSetFromOptions_BLMVM(PetscOptionItems* PetscOptionsObject,Tao tao)
163 {
164   TAO_BLMVM      *blmP = (TAO_BLMVM *)tao->data;
165   PetscBool      is_spd;
166 
167   PetscFunctionBegin;
168   PetscOptionsHeadBegin(PetscOptionsObject,"Limited-memory variable-metric method for bound constrained optimization");
169   PetscCall(PetscOptionsBool("-tao_blmvm_recycle","enable recycling of the BFGS matrix between subsequent TaoSolve() calls","",blmP->recycle,&blmP->recycle,NULL));
170   PetscOptionsHeadEnd();
171   PetscCall(MatSetOptionsPrefix(blmP->M, ((PetscObject)tao)->prefix));
172   PetscCall(MatAppendOptionsPrefix(blmP->M, "tao_blmvm_"));
173   PetscCall(MatSetFromOptions(blmP->M));
174   PetscCall(MatGetOption(blmP->M, MAT_SPD, &is_spd));
175   PetscCheck(is_spd,PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_INCOMP, "LMVM matrix must be symmetric positive-definite");
176   PetscFunctionReturn(0);
177 }
178 
179 /*------------------------------------------------------------*/
180 static PetscErrorCode TaoView_BLMVM(Tao tao, PetscViewer viewer)
181 {
182   TAO_BLMVM      *lmP = (TAO_BLMVM *)tao->data;
183   PetscBool      isascii;
184 
185   PetscFunctionBegin;
186   PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii));
187   if (isascii) {
188     PetscCall(PetscViewerASCIIPrintf(viewer, "Gradient steps: %" PetscInt_FMT "\n", lmP->grad));
189     PetscCall(PetscViewerPushFormat(viewer, PETSC_VIEWER_ASCII_INFO));
190     PetscCall(MatView(lmP->M, viewer));
191     PetscCall(PetscViewerPopFormat(viewer));
192   }
193   PetscFunctionReturn(0);
194 }
195 
196 static PetscErrorCode TaoComputeDual_BLMVM(Tao tao, Vec DXL, Vec DXU)
197 {
198   TAO_BLMVM      *blm = (TAO_BLMVM *) tao->data;
199 
200   PetscFunctionBegin;
201   PetscValidHeaderSpecific(tao,TAO_CLASSID,1);
202   PetscValidHeaderSpecific(DXL,VEC_CLASSID,2);
203   PetscValidHeaderSpecific(DXU,VEC_CLASSID,3);
204   PetscCheck(tao->gradient && blm->unprojected_gradient,PETSC_COMM_SELF,PETSC_ERR_ORDER,"Dual variables don't exist yet or no longer exist.");
205 
206   PetscCall(VecCopy(tao->gradient,DXL));
207   PetscCall(VecAXPY(DXL,-1.0,blm->unprojected_gradient));
208   PetscCall(VecSet(DXU,0.0));
209   PetscCall(VecPointwiseMax(DXL,DXL,DXU));
210 
211   PetscCall(VecCopy(blm->unprojected_gradient,DXU));
212   PetscCall(VecAXPY(DXU,-1.0,tao->gradient));
213   PetscCall(VecAXPY(DXU,1.0,DXL));
214   PetscFunctionReturn(0);
215 }
216 
217 /* ---------------------------------------------------------- */
218 /*MC
219   TAOBLMVM - Bounded limited memory variable metric is a quasi-Newton method
220          for nonlinear minimization with bound constraints. It is an extension
221          of TAOLMVM
222 
223   Options Database Keys:
224 .     -tao_lmm_recycle - enable recycling of LMVM information between subsequent TaoSolve calls
225 
226   Level: beginner
227 M*/
228 PETSC_EXTERN PetscErrorCode TaoCreate_BLMVM(Tao tao)
229 {
230   TAO_BLMVM      *blmP;
231   const char     *morethuente_type = TAOLINESEARCHMT;
232 
233   PetscFunctionBegin;
234   tao->ops->setup = TaoSetup_BLMVM;
235   tao->ops->solve = TaoSolve_BLMVM;
236   tao->ops->view = TaoView_BLMVM;
237   tao->ops->setfromoptions = TaoSetFromOptions_BLMVM;
238   tao->ops->destroy = TaoDestroy_BLMVM;
239   tao->ops->computedual = TaoComputeDual_BLMVM;
240 
241   PetscCall(PetscNewLog(tao,&blmP));
242   blmP->H0 = NULL;
243   blmP->recycle = PETSC_FALSE;
244   tao->data = (void*)blmP;
245 
246   /* Override default settings (unless already changed) */
247   if (!tao->max_it_changed) tao->max_it = 2000;
248   if (!tao->max_funcs_changed) tao->max_funcs = 4000;
249 
250   PetscCall(TaoLineSearchCreate(((PetscObject)tao)->comm, &tao->linesearch));
251   PetscCall(PetscObjectIncrementTabLevel((PetscObject)tao->linesearch, (PetscObject)tao, 1));
252   PetscCall(TaoLineSearchSetType(tao->linesearch, morethuente_type));
253   PetscCall(TaoLineSearchUseTaoRoutines(tao->linesearch,tao));
254 
255   PetscCall(KSPInitializePackage());
256   PetscCall(MatCreate(((PetscObject)tao)->comm, &blmP->M));
257   PetscCall(MatSetType(blmP->M, MATLMVMBFGS));
258   PetscCall(PetscObjectIncrementTabLevel((PetscObject)blmP->M, (PetscObject)tao, 1));
259   PetscFunctionReturn(0);
260 }
261 
262 /*@
263   TaoLMVMRecycle - Enable/disable recycling of the QN history between subsequent TaoSolve calls.
264 
265   Input Parameters:
266 +  tao  - the Tao solver context
267 -  flg - Boolean flag for recycling (PETSC_TRUE or PETSC_FALSE)
268 
269   Level: intermediate
270 @*/
271 PetscErrorCode TaoLMVMRecycle(Tao tao, PetscBool flg)
272 {
273   TAO_LMVM       *lmP;
274   TAO_BLMVM      *blmP;
275   PetscBool      is_lmvm, is_blmvm;
276 
277   PetscFunctionBegin;
278   PetscCall(PetscObjectTypeCompare((PetscObject)tao,TAOLMVM,&is_lmvm));
279   PetscCall(PetscObjectTypeCompare((PetscObject)tao,TAOBLMVM,&is_blmvm));
280   if (is_lmvm) {
281     lmP = (TAO_LMVM *)tao->data;
282     lmP->recycle = flg;
283   } else if (is_blmvm) {
284     blmP = (TAO_BLMVM *)tao->data;
285     blmP->recycle = flg;
286   }
287   PetscFunctionReturn(0);
288 }
289 
290 /*@
291   TaoLMVMSetH0 - Set the initial Hessian for the QN approximation
292 
293   Input Parameters:
294 +  tao  - the Tao solver context
295 -  H0 - Mat object for the initial Hessian
296 
297   Level: advanced
298 
299 .seealso: `TaoLMVMGetH0()`, `TaoLMVMGetH0KSP()`
300 @*/
301 PetscErrorCode TaoLMVMSetH0(Tao tao, Mat H0)
302 {
303   TAO_LMVM       *lmP;
304   TAO_BLMVM      *blmP;
305   PetscBool      is_lmvm, is_blmvm;
306 
307   PetscFunctionBegin;
308   PetscCall(PetscObjectTypeCompare((PetscObject)tao,TAOLMVM,&is_lmvm));
309   PetscCall(PetscObjectTypeCompare((PetscObject)tao,TAOBLMVM,&is_blmvm));
310   if (is_lmvm) {
311     lmP = (TAO_LMVM *)tao->data;
312     PetscCall(PetscObjectReference((PetscObject)H0));
313     lmP->H0 = H0;
314   } else if (is_blmvm) {
315     blmP = (TAO_BLMVM *)tao->data;
316     PetscCall(PetscObjectReference((PetscObject)H0));
317     blmP->H0 = H0;
318   }
319   PetscFunctionReturn(0);
320 }
321 
322 /*@
323   TaoLMVMGetH0 - Get the matrix object for the QN initial Hessian
324 
325   Input Parameters:
326 .  tao  - the Tao solver context
327 
328   Output Parameters:
329 .  H0 - Mat object for the initial Hessian
330 
331   Level: advanced
332 
333 .seealso: `TaoLMVMSetH0()`, `TaoLMVMGetH0KSP()`
334 @*/
335 PetscErrorCode TaoLMVMGetH0(Tao tao, Mat *H0)
336 {
337   TAO_LMVM       *lmP;
338   TAO_BLMVM      *blmP;
339   PetscBool      is_lmvm, is_blmvm;
340   Mat            M;
341 
342   PetscFunctionBegin;
343   PetscCall(PetscObjectTypeCompare((PetscObject)tao,TAOLMVM,&is_lmvm));
344   PetscCall(PetscObjectTypeCompare((PetscObject)tao,TAOBLMVM,&is_blmvm));
345   if (is_lmvm) {
346     lmP = (TAO_LMVM *)tao->data;
347     M = lmP->M;
348   } else if (is_blmvm) {
349     blmP = (TAO_BLMVM *)tao->data;
350     M = blmP->M;
351   } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONG, "This routine applies to TAO_LMVM and TAO_BLMVM.");
352   PetscCall(MatLMVMGetJ0(M, H0));
353   PetscFunctionReturn(0);
354 }
355 
356 /*@
357   TaoLMVMGetH0KSP - Get the iterative solver for applying the inverse of the QN initial Hessian
358 
359   Input Parameters:
360 .  tao  - the Tao solver context
361 
362   Output Parameters:
363 .  ksp - KSP solver context for the initial Hessian
364 
365   Level: advanced
366 
367 .seealso: `TaoLMVMGetH0()`, `TaoLMVMGetH0KSP()`
368 @*/
369 PetscErrorCode TaoLMVMGetH0KSP(Tao tao, KSP *ksp)
370 {
371   TAO_LMVM       *lmP;
372   TAO_BLMVM      *blmP;
373   PetscBool      is_lmvm, is_blmvm;
374   Mat            M;
375 
376   PetscFunctionBegin;
377   PetscCall(PetscObjectTypeCompare((PetscObject)tao,TAOLMVM,&is_lmvm));
378   PetscCall(PetscObjectTypeCompare((PetscObject)tao,TAOBLMVM,&is_blmvm));
379   if (is_lmvm) {
380     lmP = (TAO_LMVM *)tao->data;
381     M = lmP->M;
382   } else if (is_blmvm) {
383     blmP = (TAO_BLMVM *)tao->data;
384     M = blmP->M;
385   } else SETERRQ(PetscObjectComm((PetscObject)tao), PETSC_ERR_ARG_WRONG, "This routine applies to TAO_LMVM and TAO_BLMVM.");
386   PetscCall(MatLMVMGetJ0KSP(M, ksp));
387   PetscFunctionReturn(0);
388 }
389