Actual source code: gpcglinesearch.c

petsc-3.7.7 2017-09-25
Report Typos and Errors
  1: #include <petsc/private/taolinesearchimpl.h>
  2: #include <../src/tao/linesearch/impls/gpcglinesearch/gpcglinesearch.h>

  4: /* ---------------------------------------------------------- */

  8: static PetscErrorCode TaoLineSearchDestroy_GPCG(TaoLineSearch ls)
  9: {
 10:   PetscErrorCode     ierr;
 11:   TaoLineSearch_GPCG *ctx = (TaoLineSearch_GPCG *)ls->data;

 14:   VecDestroy(&ctx->W1);
 15:   VecDestroy(&ctx->W2);
 16:   VecDestroy(&ctx->Gold);
 17:   VecDestroy(&ctx->x);
 18:   PetscFree(ls->data);
 19:   return(0);
 20: }


 23: /*------------------------------------------------------------*/
 26: static PetscErrorCode TaoLineSearchView_GPCG(TaoLineSearch ls, PetscViewer viewer)
 27: {
 28:   PetscBool      isascii;

 32:   PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&isascii);
 33:   if (isascii) {
 34:     PetscViewerASCIIPrintf(viewer," GPCG Line search");
 35:   }
 36:   return(0);
 37: }

 39: /*------------------------------------------------------------*/
 42: static PetscErrorCode TaoLineSearchApply_GPCG(TaoLineSearch ls, Vec x, PetscReal *f, Vec g, Vec s)
 43: {
 44:   TaoLineSearch_GPCG *neP = (TaoLineSearch_GPCG *)ls->data;
 45:   PetscErrorCode     ierr;
 46:   PetscInt           i;
 47:   PetscBool          g_computed=PETSC_FALSE; /* to prevent extra gradient computation */
 48:   PetscReal          d1,finit,actred,prered,rho, gdx;

 51:   /* ls->stepmin - lower bound for step */
 52:   /* ls->stepmax - upper bound for step */
 53:   /* ls->rtol     - relative tolerance for an acceptable step */
 54:   /* ls->ftol     - tolerance for sufficient decrease condition */
 55:   /* ls->gtol     - tolerance for curvature condition */
 56:   /* ls->nfeval   - number of function evaluations */
 57:   /* ls->nfeval   - number of function/gradient evaluations */
 58:   /* ls->max_funcs  - maximum number of function evaluations */

 60:   ls->reason = TAOLINESEARCH_CONTINUE_ITERATING;
 61:   ls->step = ls->initstep;
 62:   if (!neP->W2) {
 63:     VecDuplicate(x,&neP->W2);
 64:     VecDuplicate(x,&neP->W1);
 65:     VecDuplicate(x,&neP->Gold);
 66:     neP->x = x;
 67:     PetscObjectReference((PetscObject)neP->x);
 68:   } else if (x != neP->x) {
 69:     VecDestroy(&neP->x);
 70:     VecDestroy(&neP->W1);
 71:     VecDestroy(&neP->W2);
 72:     VecDestroy(&neP->Gold);
 73:     VecDuplicate(x,&neP->W1);
 74:     VecDuplicate(x,&neP->W2);
 75:     VecDuplicate(x,&neP->Gold);
 76:     PetscObjectDereference((PetscObject)neP->x);
 77:     neP->x = x;
 78:     PetscObjectReference((PetscObject)neP->x);
 79:   }

 81:   VecDot(g,s,&gdx);
 82:    if (gdx > 0) {
 83:      PetscInfo1(ls,"Line search error: search direction is not descent direction. dot(g,s) = %g\n",(double)gdx);
 84:     ls->reason = TAOLINESEARCH_FAILED_ASCENT;
 85:     return(0);
 86:   }
 87:   VecCopy(x,neP->W2);
 88:   VecCopy(g,neP->Gold);
 89:   if (ls->bounded) {
 90:     /* Compute the smallest steplength that will make one nonbinding variable  equal the bound */
 91:     VecStepBoundInfo(x,s,ls->lower,ls->upper,&rho,&actred,&d1);
 92:     ls->step = PetscMin(ls->step,d1);
 93:   }
 94:   rho=0; actred=0;

 96:   if (ls->step < 0) {
 97:     PetscInfo1(ls,"Line search error: initial step parameter %g< 0\n",(double)ls->step);
 98:     ls->reason = TAOLINESEARCH_HALTED_OTHER;
 99:     return(0);
100:   }

102:   /* Initialization */
103:   finit = *f;
104:   for (i=0; i< ls->max_funcs; i++) {
105:     /* Force the step to be within the bounds */
106:     ls->step = PetscMax(ls->step,ls->stepmin);
107:     ls->step = PetscMin(ls->step,ls->stepmax);

109:     VecCopy(x,neP->W2);
110:     VecAXPY(neP->W2,ls->step,s);
111:     if (ls->bounded) {
112:       /* Make sure new vector is numerically within bounds */
113:       VecMedian(neP->W2,ls->lower,ls->upper,neP->W2);
114:     }

116:     /* Gradient is not needed here.  Unless there is a separate
117:        gradient routine, compute it here anyway to prevent recomputing at
118:        the end of the line search */
119:     if (ls->hasobjective) {
120:       TaoLineSearchComputeObjective(ls,neP->W2,f);
121:       g_computed=PETSC_FALSE;
122:     } else if (ls->usegts){
123:       TaoLineSearchComputeObjectiveAndGTS(ls,neP->W2,f,&gdx);
124:       g_computed=PETSC_FALSE;
125:     } else {
126:       TaoLineSearchComputeObjectiveAndGradient(ls,neP->W2,f,g);
127:       g_computed=PETSC_TRUE;
128:     }

130:     if (0 == i) {
131:         ls->f_fullstep = *f;
132:     }

134:     actred = *f - finit;
135:     VecCopy(neP->W2,neP->W1);
136:     VecAXPY(neP->W1,-1.0,x);    /* W1 = W2 - X */
137:     VecDot(neP->W1,neP->Gold,&prered);

139:     if (fabs(prered)<1.0e-100) prered=1.0e-12;
140:     rho = actred/prered;

142:     /*
143:        If sufficient progress has been obtained, accept the
144:        point.  Otherwise, backtrack.
145:     */

147:     if (actred > 0) {
148:       PetscInfo(ls,"Step resulted in ascent, rejecting.\n");
149:       ls->step = (ls->step)/2;
150:     } else if (rho > ls->ftol){
151:       break;
152:     } else{
153:       ls->step = (ls->step)/2;
154:     }

156:     /* Convergence testing */

158:     if (ls->step <= ls->stepmin || ls->step >= ls->stepmax) {
159:       ls->reason = TAOLINESEARCH_HALTED_OTHER;
160:       PetscInfo(ls,"Rounding errors may prevent further progress.  May not be a step satisfying\n");
161:       PetscInfo(ls,"sufficient decrease and curvature conditions. Tolerances may be too small.\n");
162:      break;
163:     }
164:     if (ls->step == ls->stepmax) {
165:       PetscInfo1(ls,"Step is at the upper bound, stepmax (%g)\n",(double)ls->stepmax);
166:       ls->reason = TAOLINESEARCH_HALTED_UPPERBOUND;
167:       break;
168:     }
169:     if (ls->step == ls->stepmin) {
170:       PetscInfo1(ls,"Step is at the lower bound, stepmin (%g)\n",(double)ls->stepmin);
171:       ls->reason = TAOLINESEARCH_HALTED_LOWERBOUND;
172:       break;
173:     }
174:     if ((ls->nfeval+ls->nfgeval) >= ls->max_funcs) {
175:       PetscInfo2(ls,"Number of line search function evals (%D) > maximum (%D)\n",ls->nfeval+ls->nfgeval,ls->max_funcs);
176:       ls->reason = TAOLINESEARCH_HALTED_MAXFCN;
177:       break;
178:     }
179:     if ((neP->bracket) && (ls->stepmax - ls->stepmin <= ls->rtol*ls->stepmax)){
180:       PetscInfo1(ls,"Relative width of interval of uncertainty is at most rtol (%g)\n",(double)ls->rtol);
181:       ls->reason = TAOLINESEARCH_HALTED_RTOL;
182:       break;
183:     }
184:   }
185:   PetscInfo2(ls,"%D function evals in line search, step = %g\n",ls->nfeval+ls->nfgeval,(double)ls->step);
186:   /* set new solution vector and compute gradient if necessary */
187:   VecCopy(neP->W2, x);
188:   if (ls->reason == TAOLINESEARCH_CONTINUE_ITERATING) {
189:     ls->reason = TAOLINESEARCH_SUCCESS;
190:   }
191:   if (!g_computed) {
192:     TaoLineSearchComputeGradient(ls,x,g);
193:   }
194:   return(0);
195: }

197: /* ---------------------------------------------------------- */
200: PETSC_EXTERN PetscErrorCode TaoLineSearchCreate_GPCG(TaoLineSearch ls)
201: {
202:   PetscErrorCode     ierr;
203:   TaoLineSearch_GPCG *neP;

206:   ls->ftol                = 0.05;
207:   ls->rtol                = 0.0;
208:   ls->gtol                = 0.0;
209:   ls->stepmin             = 1.0e-20;
210:   ls->stepmax             = 1.0e+20;
211:   ls->nfeval              = 0;
212:   ls->max_funcs           = 30;
213:   ls->step                = 1.0;

215:   PetscNewLog(ls,&neP);
216:   neP->bracket            = 0;
217:   neP->infoc              = 1;
218:   ls->data = (void*)neP;

220:   ls->ops->setup = 0;
221:   ls->ops->reset = 0;
222:   ls->ops->apply=TaoLineSearchApply_GPCG;
223:   ls->ops->view =TaoLineSearchView_GPCG;
224:   ls->ops->destroy=TaoLineSearchDestroy_GPCG;
225:   ls->ops->setfromoptions=0;
226:   return(0);
227: }