xref: /petsc/src/tao/quadratic/impls/gpcg/gpcg.c (revision bcd4bb4a4158aa96f212e9537e87b40407faf83e)
1 #include <petscksp.h>
2 #include <../src/tao/quadratic/impls/gpcg/gpcg.h> /*I "gpcg.h" I*/
3 
4 static PetscErrorCode GPCGGradProjections(Tao tao);
5 static PetscErrorCode GPCGObjectiveAndGradient(TaoLineSearch, Vec, PetscReal *, Vec, void *);
6 
7 /*------------------------------------------------------------*/
8 static PetscErrorCode TaoDestroy_GPCG(Tao tao)
9 {
10   TAO_GPCG *gpcg = (TAO_GPCG *)tao->data;
11 
12   /* Free allocated memory in GPCG structure */
13   PetscFunctionBegin;
14   PetscCall(VecDestroy(&gpcg->B));
15   PetscCall(VecDestroy(&gpcg->Work));
16   PetscCall(VecDestroy(&gpcg->X_New));
17   PetscCall(VecDestroy(&gpcg->G_New));
18   PetscCall(VecDestroy(&gpcg->DXFree));
19   PetscCall(VecDestroy(&gpcg->R));
20   PetscCall(VecDestroy(&gpcg->PG));
21   PetscCall(MatDestroy(&gpcg->Hsub));
22   PetscCall(MatDestroy(&gpcg->Hsub_pre));
23   PetscCall(ISDestroy(&gpcg->Free_Local));
24   PetscCall(KSPDestroy(&tao->ksp));
25   PetscCall(PetscFree(tao->data));
26   PetscFunctionReturn(PETSC_SUCCESS);
27 }
28 
29 /*------------------------------------------------------------*/
30 static PetscErrorCode TaoSetFromOptions_GPCG(Tao tao, PetscOptionItems PetscOptionsObject)
31 {
32   TAO_GPCG *gpcg = (TAO_GPCG *)tao->data;
33   PetscBool flg;
34 
35   PetscFunctionBegin;
36   PetscOptionsHeadBegin(PetscOptionsObject, "Gradient Projection, Conjugate Gradient method for bound constrained optimization");
37   PetscCall(PetscOptionsInt("-tao_gpcg_maxpgits", "maximum number of gradient projections per GPCG iterate", NULL, gpcg->maxgpits, &gpcg->maxgpits, &flg));
38   PetscOptionsHeadEnd();
39   PetscCall(KSPSetFromOptions(tao->ksp));
40   PetscCall(TaoLineSearchSetFromOptions(tao->linesearch));
41   PetscFunctionReturn(PETSC_SUCCESS);
42 }
43 
44 /*------------------------------------------------------------*/
45 static PetscErrorCode TaoView_GPCG(Tao tao, PetscViewer viewer)
46 {
47   TAO_GPCG *gpcg = (TAO_GPCG *)tao->data;
48   PetscBool isascii;
49 
50   PetscFunctionBegin;
51   PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii));
52   if (isascii) {
53     PetscCall(PetscViewerASCIIPrintf(viewer, "Total PG its: %" PetscInt_FMT ",", gpcg->total_gp_its));
54     PetscCall(PetscViewerASCIIPrintf(viewer, "PG tolerance: %g \n", (double)gpcg->pg_ftol));
55   }
56   PetscCall(TaoLineSearchView(tao->linesearch, viewer));
57   PetscFunctionReturn(PETSC_SUCCESS);
58 }
59 
60 /* GPCGObjectiveAndGradient()
61    Compute f=0.5 * x'Hx + b'x + c
62            g=Hx + b
63 */
64 static PetscErrorCode GPCGObjectiveAndGradient(TaoLineSearch ls, Vec X, PetscReal *f, Vec G, void *tptr)
65 {
66   Tao       tao  = (Tao)tptr;
67   TAO_GPCG *gpcg = (TAO_GPCG *)tao->data;
68   PetscReal f1, f2;
69 
70   PetscFunctionBegin;
71   PetscCall(MatMult(tao->hessian, X, G));
72   PetscCall(VecDot(G, X, &f1));
73   PetscCall(VecDot(gpcg->B, X, &f2));
74   PetscCall(VecAXPY(G, 1.0, gpcg->B));
75   *f = f1 / 2.0 + f2 + gpcg->c;
76   PetscFunctionReturn(PETSC_SUCCESS);
77 }
78 
79 /* ---------------------------------------------------------- */
80 static PetscErrorCode TaoSetup_GPCG(Tao tao)
81 {
82   TAO_GPCG *gpcg = (TAO_GPCG *)tao->data;
83 
84   PetscFunctionBegin;
85   /* Allocate some arrays */
86   if (!tao->gradient) PetscCall(VecDuplicate(tao->solution, &tao->gradient));
87   if (!tao->stepdirection) PetscCall(VecDuplicate(tao->solution, &tao->stepdirection));
88 
89   PetscCall(VecDuplicate(tao->solution, &gpcg->B));
90   PetscCall(VecDuplicate(tao->solution, &gpcg->Work));
91   PetscCall(VecDuplicate(tao->solution, &gpcg->X_New));
92   PetscCall(VecDuplicate(tao->solution, &gpcg->G_New));
93   PetscCall(VecDuplicate(tao->solution, &gpcg->DXFree));
94   PetscCall(VecDuplicate(tao->solution, &gpcg->R));
95   PetscCall(VecDuplicate(tao->solution, &gpcg->PG));
96   /*
97     if (gpcg->ksp_type == GPCG_KSP_NASH) {
98         PetscCall(KSPSetType(tao->ksp,KSPNASH));
99       } else if (gpcg->ksp_type == GPCG_KSP_STCG) {
100         PetscCall(KSPSetType(tao->ksp,KSPSTCG));
101       } else {
102         PetscCall(KSPSetType(tao->ksp,KSPGLTR));
103       }
104       if (tao->ksp->ops->setfromoptions) (*tao->ksp->ops->setfromoptions)(tao->ksp);
105 
106     }
107   */
108   PetscFunctionReturn(PETSC_SUCCESS);
109 }
110 
111 static PetscErrorCode TaoSolve_GPCG(Tao tao)
112 {
113   TAO_GPCG                    *gpcg = (TAO_GPCG *)tao->data;
114   PetscInt                     its;
115   PetscReal                    actred, f, f_new, gnorm, gdx, stepsize, xtb;
116   PetscReal                    xtHx;
117   TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING;
118 
119   PetscFunctionBegin;
120   PetscCall(TaoComputeVariableBounds(tao));
121   PetscCall(VecMedian(tao->XL, tao->solution, tao->XU, tao->solution));
122   PetscCall(TaoLineSearchSetVariableBounds(tao->linesearch, tao->XL, tao->XU));
123 
124   /* Using f = .5*x'Hx + x'b + c and g=Hx + b,  compute b,c */
125   PetscCall(TaoComputeHessian(tao, tao->solution, tao->hessian, tao->hessian_pre));
126   PetscCall(TaoComputeObjectiveAndGradient(tao, tao->solution, &f, tao->gradient));
127   PetscCall(VecCopy(tao->gradient, gpcg->B));
128   PetscCall(MatMult(tao->hessian, tao->solution, gpcg->Work));
129   PetscCall(VecDot(gpcg->Work, tao->solution, &xtHx));
130   PetscCall(VecAXPY(gpcg->B, -1.0, gpcg->Work));
131   PetscCall(VecDot(gpcg->B, tao->solution, &xtb));
132   gpcg->c = f - xtHx / 2.0 - xtb;
133   if (gpcg->Free_Local) PetscCall(ISDestroy(&gpcg->Free_Local));
134   PetscCall(VecWhichInactive(tao->XL, tao->solution, tao->gradient, tao->XU, PETSC_TRUE, &gpcg->Free_Local));
135 
136   /* Project the gradient and calculate the norm */
137   PetscCall(VecCopy(tao->gradient, gpcg->G_New));
138   PetscCall(VecBoundGradientProjection(tao->gradient, tao->solution, tao->XL, tao->XU, gpcg->PG));
139   PetscCall(VecNorm(gpcg->PG, NORM_2, &gpcg->gnorm));
140   tao->step = 1.0;
141   gpcg->f   = f;
142 
143   /* Check Stopping Condition      */
144   tao->reason = TAO_CONTINUE_ITERATING;
145   PetscCall(TaoLogConvergenceHistory(tao, f, gpcg->gnorm, 0.0, tao->ksp_its));
146   PetscCall(TaoMonitor(tao, tao->niter, f, gpcg->gnorm, 0.0, tao->step));
147   PetscUseTypeMethod(tao, convergencetest, tao->cnvP);
148 
149   while (tao->reason == TAO_CONTINUE_ITERATING) {
150     /* Call general purpose update function */
151     PetscTryTypeMethod(tao, update, tao->niter, tao->user_update);
152     tao->ksp_its = 0;
153 
154     PetscCall(GPCGGradProjections(tao));
155     PetscCall(ISGetSize(gpcg->Free_Local, &gpcg->n_free));
156 
157     f     = gpcg->f;
158     gnorm = gpcg->gnorm;
159 
160     PetscCall(KSPReset(tao->ksp));
161 
162     if (gpcg->n_free > 0) {
163       /* Create a reduced linear system */
164       PetscCall(VecDestroy(&gpcg->R));
165       PetscCall(VecDestroy(&gpcg->DXFree));
166       PetscCall(TaoVecGetSubVec(tao->gradient, gpcg->Free_Local, tao->subset_type, 0.0, &gpcg->R));
167       PetscCall(VecScale(gpcg->R, -1.0));
168       PetscCall(TaoVecGetSubVec(tao->stepdirection, gpcg->Free_Local, tao->subset_type, 0.0, &gpcg->DXFree));
169       PetscCall(VecSet(gpcg->DXFree, 0.0));
170 
171       PetscCall(TaoMatGetSubMat(tao->hessian, gpcg->Free_Local, gpcg->Work, tao->subset_type, &gpcg->Hsub));
172 
173       if (tao->hessian_pre == tao->hessian) {
174         PetscCall(MatDestroy(&gpcg->Hsub_pre));
175         PetscCall(PetscObjectReference((PetscObject)gpcg->Hsub));
176         gpcg->Hsub_pre = gpcg->Hsub;
177       } else {
178         PetscCall(TaoMatGetSubMat(tao->hessian, gpcg->Free_Local, gpcg->Work, tao->subset_type, &gpcg->Hsub_pre));
179       }
180 
181       PetscCall(KSPReset(tao->ksp));
182       PetscCall(KSPSetOperators(tao->ksp, gpcg->Hsub, gpcg->Hsub_pre));
183 
184       PetscCall(KSPSolve(tao->ksp, gpcg->R, gpcg->DXFree));
185       PetscCall(KSPGetIterationNumber(tao->ksp, &its));
186       tao->ksp_its += its;
187       tao->ksp_tot_its += its;
188       PetscCall(VecSet(tao->stepdirection, 0.0));
189       PetscCall(VecISAXPY(tao->stepdirection, gpcg->Free_Local, 1.0, gpcg->DXFree));
190 
191       PetscCall(VecDot(tao->stepdirection, tao->gradient, &gdx));
192       PetscCall(TaoLineSearchSetInitialStepLength(tao->linesearch, 1.0));
193       f_new = f;
194       PetscCall(TaoLineSearchApply(tao->linesearch, tao->solution, &f_new, tao->gradient, tao->stepdirection, &stepsize, &ls_status));
195 
196       actred = f_new - f;
197 
198       /* Evaluate the function and gradient at the new point */
199       PetscCall(VecBoundGradientProjection(tao->gradient, tao->solution, tao->XL, tao->XU, gpcg->PG));
200       PetscCall(VecNorm(gpcg->PG, NORM_2, &gnorm));
201       f = f_new;
202       PetscCall(ISDestroy(&gpcg->Free_Local));
203       PetscCall(VecWhichInactive(tao->XL, tao->solution, tao->gradient, tao->XU, PETSC_TRUE, &gpcg->Free_Local));
204     } else {
205       actred     = 0;
206       gpcg->step = 1.0;
207       /* if there were no free variables, no cg method */
208     }
209 
210     tao->niter++;
211     gpcg->f      = f;
212     gpcg->gnorm  = gnorm;
213     gpcg->actred = actred;
214     PetscCall(TaoLogConvergenceHistory(tao, f, gpcg->gnorm, 0.0, tao->ksp_its));
215     PetscCall(TaoMonitor(tao, tao->niter, f, gpcg->gnorm, 0.0, tao->step));
216     PetscUseTypeMethod(tao, convergencetest, tao->cnvP);
217     if (tao->reason != TAO_CONTINUE_ITERATING) break;
218   } /* END MAIN LOOP  */
219   PetscFunctionReturn(PETSC_SUCCESS);
220 }
221 
222 static PetscErrorCode GPCGGradProjections(Tao tao)
223 {
224   TAO_GPCG                    *gpcg = (TAO_GPCG *)tao->data;
225   PetscInt                     i;
226   PetscReal                    actred = -1.0, actred_max = 0.0, gAg, gtg = gpcg->gnorm, alpha;
227   PetscReal                    f_new, gdx, stepsize;
228   Vec                          DX = tao->stepdirection, XL = tao->XL, XU = tao->XU, Work = gpcg->Work;
229   Vec                          X = tao->solution, G = tao->gradient;
230   TaoLineSearchConvergedReason lsflag = TAOLINESEARCH_CONTINUE_ITERATING;
231 
232   /*
233      The free, active, and binding variables should be already identified
234   */
235   PetscFunctionBegin;
236   for (i = 0; i < gpcg->maxgpits; i++) {
237     if (-actred <= (gpcg->pg_ftol) * actred_max) break;
238     PetscCall(VecBoundGradientProjection(G, X, XL, XU, DX));
239     PetscCall(VecScale(DX, -1.0));
240     PetscCall(VecDot(DX, G, &gdx));
241 
242     PetscCall(MatMult(tao->hessian, DX, Work));
243     PetscCall(VecDot(DX, Work, &gAg));
244 
245     gpcg->gp_iterates++;
246     gpcg->total_gp_its++;
247 
248     gtg = -gdx;
249     if (PetscAbsReal(gAg) == 0.0) {
250       alpha = 1.0;
251     } else {
252       alpha = PetscAbsReal(gtg / gAg);
253     }
254     PetscCall(TaoLineSearchSetInitialStepLength(tao->linesearch, alpha));
255     f_new = gpcg->f;
256     PetscCall(TaoLineSearchApply(tao->linesearch, X, &f_new, G, DX, &stepsize, &lsflag));
257 
258     /* Update the iterate */
259     actred     = f_new - gpcg->f;
260     actred_max = PetscMax(actred_max, -(f_new - gpcg->f));
261     gpcg->f    = f_new;
262     PetscCall(ISDestroy(&gpcg->Free_Local));
263     PetscCall(VecWhichInactive(XL, X, tao->gradient, XU, PETSC_TRUE, &gpcg->Free_Local));
264   }
265 
266   gpcg->gnorm = gtg;
267   PetscFunctionReturn(PETSC_SUCCESS);
268 } /* End gradient projections */
269 
270 static PetscErrorCode TaoComputeDual_GPCG(Tao tao, Vec DXL, Vec DXU)
271 {
272   TAO_GPCG *gpcg = (TAO_GPCG *)tao->data;
273 
274   PetscFunctionBegin;
275   PetscCall(VecBoundGradientProjection(tao->gradient, tao->solution, tao->XL, tao->XU, gpcg->Work));
276   PetscCall(VecCopy(gpcg->Work, DXL));
277   PetscCall(VecAXPY(DXL, -1.0, tao->gradient));
278   PetscCall(VecSet(DXU, 0.0));
279   PetscCall(VecPointwiseMax(DXL, DXL, DXU));
280 
281   PetscCall(VecCopy(tao->gradient, DXU));
282   PetscCall(VecAXPY(DXU, -1.0, gpcg->Work));
283   PetscCall(VecSet(gpcg->Work, 0.0));
284   PetscCall(VecPointwiseMin(DXU, gpcg->Work, DXU));
285   PetscFunctionReturn(PETSC_SUCCESS);
286 }
287 
288 /*------------------------------------------------------------*/
289 /*MC
290   TAOGPCG - gradient projected conjugate gradient algorithm is an active-set
291         conjugate-gradient based method for bound-constrained minimization
292 
293   Options Database Keys:
294 + -tao_gpcg_maxpgits - maximum number of gradient projections for GPCG iterate
295 - -tao_subset_type - "subvec","mask","matrix-free", strategies for handling active-sets
296 
297   Level: beginner
298 M*/
299 PETSC_EXTERN PetscErrorCode TaoCreate_GPCG(Tao tao)
300 {
301   TAO_GPCG *gpcg;
302 
303   PetscFunctionBegin;
304   tao->ops->setup          = TaoSetup_GPCG;
305   tao->ops->solve          = TaoSolve_GPCG;
306   tao->ops->view           = TaoView_GPCG;
307   tao->ops->setfromoptions = TaoSetFromOptions_GPCG;
308   tao->ops->destroy        = TaoDestroy_GPCG;
309   tao->ops->computedual    = TaoComputeDual_GPCG;
310 
311   PetscCall(PetscNew(&gpcg));
312   tao->data = (void *)gpcg;
313 
314   /* Override default settings (unless already changed) */
315   PetscCall(TaoParametersInitialize(tao));
316   PetscObjectParameterSetDefault(tao, max_it, 500);
317   PetscObjectParameterSetDefault(tao, max_funcs, 100000);
318   PetscObjectParameterSetDefault(tao, gatol, PetscDefined(USE_REAL_SINGLE) ? 1e-6 : 1e-12);
319   PetscObjectParameterSetDefault(tao, grtol, PetscDefined(USE_REAL_SINGLE) ? 1e-6 : 1e-12);
320 
321   /* Initialize pointers and variables */
322   gpcg->n        = 0;
323   gpcg->maxgpits = 8;
324   gpcg->pg_ftol  = 0.1;
325 
326   gpcg->gp_iterates  = 0; /* Cumulative number */
327   gpcg->total_gp_its = 0;
328 
329   /* Initialize pointers and variables */
330   gpcg->n_bind      = 0;
331   gpcg->n_free      = 0;
332   gpcg->n_upper     = 0;
333   gpcg->n_lower     = 0;
334   gpcg->subset_type = TAO_SUBSET_MASK;
335   gpcg->Hsub        = NULL;
336   gpcg->Hsub_pre    = NULL;
337 
338   PetscCall(KSPCreate(((PetscObject)tao)->comm, &tao->ksp));
339   PetscCall(PetscObjectIncrementTabLevel((PetscObject)tao->ksp, (PetscObject)tao, 1));
340   PetscCall(KSPSetOptionsPrefix(tao->ksp, tao->hdr.prefix));
341   PetscCall(KSPSetType(tao->ksp, KSPNASH));
342 
343   PetscCall(TaoLineSearchCreate(((PetscObject)tao)->comm, &tao->linesearch));
344   PetscCall(PetscObjectIncrementTabLevel((PetscObject)tao->linesearch, (PetscObject)tao, 1));
345   PetscCall(TaoLineSearchSetType(tao->linesearch, TAOLINESEARCHGPCG));
346   PetscCall(TaoLineSearchSetObjectiveAndGradientRoutine(tao->linesearch, GPCGObjectiveAndGradient, tao));
347   PetscCall(TaoLineSearchSetOptionsPrefix(tao->linesearch, tao->hdr.prefix));
348   PetscFunctionReturn(PETSC_SUCCESS);
349 }
350