xref: /petsc/src/tao/unconstrained/impls/lmvm/lmvm.c (revision a7fac7c2b47dbcd8ece0cc2dfdfe0e63be1bb7b5)
1 #include <petsctaolinesearch.h>
2 #include <../src/tao/matrix/lmvmmat.h>
3 #include <../src/tao/unconstrained/impls/lmvm/lmvm.h>
4 
5 #define LMVM_BFGS                0
6 #define LMVM_SCALED_GRADIENT     1
7 #define LMVM_GRADIENT            2
8 
9 #undef __FUNCT__
10 #define __FUNCT__ "TaoSolve_LMVM"
11 static PetscErrorCode TaoSolve_LMVM(Tao tao)
12 {
13   TAO_LMVM                     *lmP = (TAO_LMVM *)tao->data;
14   PetscReal                    f, fold, gdx, gnorm;
15   PetscReal                    step = 1.0;
16   PetscReal                    delta;
17   PetscErrorCode               ierr;
18   PetscInt                     stepType;
19   TaoConvergedReason           reason = TAO_CONTINUE_ITERATING;
20   TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING;
21 
22   PetscFunctionBegin;
23 
24   if (tao->XL || tao->XU || tao->ops->computebounds) {
25     ierr = PetscPrintf(((PetscObject)tao)->comm,"WARNING: Variable bounds have been set but will be ignored by lmvm algorithm\n");CHKERRQ(ierr);
26   }
27 
28   /*  Check convergence criteria */
29   ierr = TaoComputeObjectiveAndGradient(tao, tao->solution, &f, tao->gradient);CHKERRQ(ierr);
30   ierr = VecNorm(tao->gradient,NORM_2,&gnorm);CHKERRQ(ierr);
31   if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf or NaN");
32 
33   ierr = TaoMonitor(tao, tao->niter, f, gnorm, 0.0, step, &reason);CHKERRQ(ierr);
34   if (reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0);
35 
36   /*  Set initial scaling for the function */
37   if (f != 0.0) {
38     delta = 2.0 * PetscAbsScalar(f) / (gnorm*gnorm);
39   } else {
40     delta = 2.0 / (gnorm*gnorm);
41   }
42   ierr = MatLMVMSetDelta(lmP->M,delta);CHKERRQ(ierr);
43 
44   /*  Set counter for gradient/reset steps */
45   lmP->bfgs = 0;
46   lmP->sgrad = 0;
47   lmP->grad = 0;
48 
49   /*  Have not converged; continue with Newton method */
50   while (reason == TAO_CONTINUE_ITERATING) {
51     /*  Compute direction */
52     ierr = MatLMVMUpdate(lmP->M,tao->solution,tao->gradient);CHKERRQ(ierr);
53     ierr = MatLMVMSolve(lmP->M, tao->gradient, lmP->D);CHKERRQ(ierr);
54     ++lmP->bfgs;
55 
56     /*  Check for success (descent direction) */
57     ierr = VecDot(lmP->D, tao->gradient, &gdx);CHKERRQ(ierr);
58     if ((gdx <= 0.0) || PetscIsInfOrNanReal(gdx)) {
59       /* Step is not descent or direction produced not a number
60          We can assert bfgsUpdates > 1 in this case because
61          the first solve produces the scaled gradient direction,
62          which is guaranteed to be descent
63 
64          Use steepest descent direction (scaled)
65       */
66 
67       ++lmP->grad;
68 
69       if (f != 0.0) {
70         delta = 2.0 * PetscAbsScalar(f) / (gnorm*gnorm);
71       } else {
72         delta = 2.0 / (gnorm*gnorm);
73       }
74       ierr = MatLMVMSetDelta(lmP->M, delta);CHKERRQ(ierr);
75       ierr = MatLMVMReset(lmP->M);CHKERRQ(ierr);
76       ierr = MatLMVMUpdate(lmP->M, tao->solution, tao->gradient);CHKERRQ(ierr);
77       ierr = MatLMVMSolve(lmP->M,tao->gradient, lmP->D);CHKERRQ(ierr);
78 
79       /* On a reset, the direction cannot be not a number; it is a
80          scaled gradient step.  No need to check for this condition. */
81 
82       lmP->bfgs = 1;
83       ++lmP->sgrad;
84       stepType = LMVM_SCALED_GRADIENT;
85     } else {
86       if (1 == lmP->bfgs) {
87         /*  The first BFGS direction is always the scaled gradient */
88         ++lmP->sgrad;
89         stepType = LMVM_SCALED_GRADIENT;
90       } else {
91         ++lmP->bfgs;
92         stepType = LMVM_BFGS;
93       }
94     }
95     ierr = VecScale(lmP->D, -1.0);CHKERRQ(ierr);
96 
97     /*  Perform the linesearch */
98     fold = f;
99     ierr = VecCopy(tao->solution, lmP->Xold);CHKERRQ(ierr);
100     ierr = VecCopy(tao->gradient, lmP->Gold);CHKERRQ(ierr);
101 
102     ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &f, tao->gradient, lmP->D, &step,&ls_status);CHKERRQ(ierr);
103     ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr);
104 
105     while (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER && (stepType != LMVM_GRADIENT)) {
106       /*  Linesearch failed */
107       /*  Reset factors and use scaled gradient step */
108       f = fold;
109       ierr = VecCopy(lmP->Xold, tao->solution);CHKERRQ(ierr);
110       ierr = VecCopy(lmP->Gold, tao->gradient);CHKERRQ(ierr);
111 
112       switch(stepType) {
113       case LMVM_BFGS:
114         /*  Failed to obtain acceptable iterate with BFGS step */
115         /*  Attempt to use the scaled gradient direction */
116 
117         if (f != 0.0) {
118           delta = 2.0 * PetscAbsScalar(f) / (gnorm*gnorm);
119         } else {
120           delta = 2.0 / (gnorm*gnorm);
121         }
122         ierr = MatLMVMSetDelta(lmP->M, delta);CHKERRQ(ierr);
123         ierr = MatLMVMReset(lmP->M);CHKERRQ(ierr);
124         ierr = MatLMVMUpdate(lmP->M, tao->solution, tao->gradient);CHKERRQ(ierr);
125         ierr = MatLMVMSolve(lmP->M, tao->gradient, lmP->D);CHKERRQ(ierr);
126 
127         /* On a reset, the direction cannot be not a number; it is a
128            scaled gradient step.  No need to check for this condition. */
129 
130         lmP->bfgs = 1;
131         ++lmP->sgrad;
132         stepType = LMVM_SCALED_GRADIENT;
133         break;
134 
135       case LMVM_SCALED_GRADIENT:
136         /* The scaled gradient step did not produce a new iterate;
137            attempt to use the gradient direction.
138            Need to make sure we are not using a different diagonal scaling */
139         ierr = MatLMVMSetDelta(lmP->M, 1.0);CHKERRQ(ierr);
140         ierr = MatLMVMReset(lmP->M);CHKERRQ(ierr);
141         ierr = MatLMVMUpdate(lmP->M, tao->solution, tao->gradient);CHKERRQ(ierr);
142         ierr = MatLMVMSolve(lmP->M, tao->gradient, lmP->D);CHKERRQ(ierr);
143 
144         lmP->bfgs = 1;
145         ++lmP->grad;
146         stepType = LMVM_GRADIENT;
147         break;
148       }
149       ierr = VecScale(lmP->D, -1.0);CHKERRQ(ierr);
150 
151       /*  Perform the linesearch */
152       ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &f, tao->gradient, lmP->D, &step, &ls_status);CHKERRQ(ierr);
153       ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr);
154     }
155 
156     if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) {
157       /*  Failed to find an improving point */
158       f = fold;
159       ierr = VecCopy(lmP->Xold, tao->solution);CHKERRQ(ierr);
160       ierr = VecCopy(lmP->Gold, tao->gradient);CHKERRQ(ierr);
161       step = 0.0;
162       reason = TAO_DIVERGED_LS_FAILURE;
163       tao->reason = TAO_DIVERGED_LS_FAILURE;
164     }
165     /*  Check for termination */
166     ierr = VecNorm(tao->gradient, NORM_2, &gnorm);CHKERRQ(ierr);
167     tao->niter++;
168     ierr = TaoMonitor(tao,tao->niter,f,gnorm,0.0,step,&reason);CHKERRQ(ierr);
169   }
170   PetscFunctionReturn(0);
171 }
172 
173 #undef __FUNCT__
174 #define __FUNCT__ "TaoSetUp_LMVM"
175 static PetscErrorCode TaoSetUp_LMVM(Tao tao)
176 {
177   TAO_LMVM       *lmP = (TAO_LMVM *)tao->data;
178   PetscInt       n,N;
179   PetscErrorCode ierr;
180 
181   PetscFunctionBegin;
182   /* Existence of tao->solution checked in TaoSetUp() */
183   if (!tao->gradient) {ierr = VecDuplicate(tao->solution,&tao->gradient);CHKERRQ(ierr);  }
184   if (!tao->stepdirection) {ierr = VecDuplicate(tao->solution,&tao->stepdirection);CHKERRQ(ierr);  }
185   if (!lmP->D) {ierr = VecDuplicate(tao->solution,&lmP->D);CHKERRQ(ierr);  }
186   if (!lmP->Xold) {ierr = VecDuplicate(tao->solution,&lmP->Xold);CHKERRQ(ierr);  }
187   if (!lmP->Gold) {ierr = VecDuplicate(tao->solution,&lmP->Gold);CHKERRQ(ierr);  }
188 
189   /*  Create matrix for the limited memory approximation */
190   ierr = VecGetLocalSize(tao->solution,&n);CHKERRQ(ierr);
191   ierr = VecGetSize(tao->solution,&N);CHKERRQ(ierr);
192   ierr = MatCreateLMVM(((PetscObject)tao)->comm,n,N,&lmP->M);CHKERRQ(ierr);
193   ierr = MatLMVMAllocateVectors(lmP->M,tao->solution);CHKERRQ(ierr);
194   PetscFunctionReturn(0);
195 }
196 
197 /* ---------------------------------------------------------- */
198 #undef __FUNCT__
199 #define __FUNCT__ "TaoDestroy_LMVM"
200 static PetscErrorCode TaoDestroy_LMVM(Tao tao)
201 {
202   TAO_LMVM       *lmP = (TAO_LMVM *)tao->data;
203   PetscErrorCode ierr;
204 
205   PetscFunctionBegin;
206   if (tao->setupcalled) {
207     ierr = VecDestroy(&lmP->Xold);CHKERRQ(ierr);
208     ierr = VecDestroy(&lmP->Gold);CHKERRQ(ierr);
209     ierr = VecDestroy(&lmP->D);CHKERRQ(ierr);
210     ierr = MatDestroy(&lmP->M);CHKERRQ(ierr);
211   }
212   ierr = PetscFree(tao->data);CHKERRQ(ierr);
213   PetscFunctionReturn(0);
214 }
215 
216 /*------------------------------------------------------------*/
217 #undef __FUNCT__
218 #define __FUNCT__ "TaoSetFromOptions_LMVM"
219 static PetscErrorCode TaoSetFromOptions_LMVM(PetscOptions *PetscOptionsObject,Tao tao)
220 {
221   PetscErrorCode ierr;
222 
223   PetscFunctionBegin;
224   ierr = PetscOptionsHead(PetscOptionsObject,"Limited-memory variable-metric method for unconstrained optimization");CHKERRQ(ierr);
225   ierr = TaoLineSearchSetFromOptions(tao->linesearch);CHKERRQ(ierr);
226   ierr = PetscOptionsTail();CHKERRQ(ierr);
227   PetscFunctionReturn(0);
228 }
229 
230 /*------------------------------------------------------------*/
231 #undef __FUNCT__
232 #define __FUNCT__ "TaoView_LMVM"
233 static PetscErrorCode TaoView_LMVM(Tao tao, PetscViewer viewer)
234 {
235   TAO_LMVM       *lm = (TAO_LMVM *)tao->data;
236   PetscBool      isascii;
237   PetscErrorCode ierr;
238 
239   PetscFunctionBegin;
240   ierr = PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii);CHKERRQ(ierr);
241   if (isascii) {
242     ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr);
243     ierr = PetscViewerASCIIPrintf(viewer, "BFGS steps: %D\n", lm->bfgs);CHKERRQ(ierr);
244     ierr = PetscViewerASCIIPrintf(viewer, "Scaled gradient steps: %D\n", lm->sgrad);CHKERRQ(ierr);
245     ierr = PetscViewerASCIIPrintf(viewer, "Gradient steps: %D\n", lm->grad);CHKERRQ(ierr);
246     ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr);
247   }
248   PetscFunctionReturn(0);
249 }
250 
251 /* ---------------------------------------------------------- */
252 
253 /*MC
254      TAOLMVM - Limited Memory Variable Metric method is a quasi-Newton
255      optimization solver for unconstrained minimization. It solves
256      the Newton step
257               Hkdk = - gk
258 
259      using an approximation Bk in place of Hk, where Bk is composed using
260      the BFGS update formula. A More-Thuente line search is then used
261      to computed the steplength in the dk direction
262   Options Database Keys:
263 +     -tao_lmm_vectors - number of vectors to use for approximation
264 .     -tao_lmm_scale_type - "none","scalar","broyden"
265 .     -tao_lmm_limit_type - "none","average","relative","absolute"
266 .     -tao_lmm_rescale_type - "none","scalar","gl"
267 .     -tao_lmm_limit_mu - mu limiting factor
268 .     -tao_lmm_limit_nu - nu limiting factor
269 .     -tao_lmm_delta_min - minimum delta value
270 .     -tao_lmm_delta_max - maximum delta value
271 .     -tao_lmm_broyden_phi - phi factor for Broyden scaling
272 .     -tao_lmm_scalar_alpha - alpha factor for scalar scaling
273 .     -tao_lmm_rescale_alpha - alpha factor for rescaling diagonal
274 .     -tao_lmm_rescale_beta - beta factor for rescaling diagonal
275 .     -tao_lmm_scalar_history - amount of history for scalar scaling
276 .     -tao_lmm_rescale_history - amount of history for rescaling diagonal
277 -     -tao_lmm_eps - rejection tolerance
278 
279   Level: beginner
280 M*/
281 
282 #undef __FUNCT__
283 #define __FUNCT__ "TaoCreate_LMVM"
284 PETSC_EXTERN PetscErrorCode TaoCreate_LMVM(Tao tao)
285 {
286   TAO_LMVM       *lmP;
287   const char     *morethuente_type = TAOLINESEARCHMT;
288   PetscErrorCode ierr;
289 
290   PetscFunctionBegin;
291   tao->ops->setup = TaoSetUp_LMVM;
292   tao->ops->solve = TaoSolve_LMVM;
293   tao->ops->view = TaoView_LMVM;
294   tao->ops->setfromoptions = TaoSetFromOptions_LMVM;
295   tao->ops->destroy = TaoDestroy_LMVM;
296 
297   ierr = PetscNewLog(tao,&lmP);CHKERRQ(ierr);
298   lmP->D = 0;
299   lmP->M = 0;
300   lmP->Xold = 0;
301   lmP->Gold = 0;
302 
303   tao->data = (void*)lmP;
304   /* Override default settings (unless already changed) */
305   if (!tao->max_it_changed) tao->max_it = 2000;
306   if (!tao->max_funcs_changed) tao->max_funcs = 4000;
307   if (!tao->fatol_changed) tao->fatol = 1.0e-4;
308   if (!tao->frtol_changed) tao->frtol = 1.0e-4;
309 
310   ierr = TaoLineSearchCreate(((PetscObject)tao)->comm,&tao->linesearch);CHKERRQ(ierr);
311   ierr = TaoLineSearchSetType(tao->linesearch,morethuente_type);CHKERRQ(ierr);
312   ierr = TaoLineSearchUseTaoRoutines(tao->linesearch,tao);CHKERRQ(ierr);
313   ierr = TaoLineSearchSetOptionsPrefix(tao->linesearch,tao->hdr.prefix);CHKERRQ(ierr);
314   PetscFunctionReturn(0);
315 }
316 
317