1 #include <petsc/private/taolinesearchimpl.h> 2 #include <../src/tao/linesearch/impls/morethuente/morethuente.h> 3 4 /* 5 This algorithm is taken from More' and Thuente, "Line search algorithms 6 with guaranteed sufficient decrease", Argonne National Laboratory, 7 Technical Report MCS-P330-1092. 8 */ 9 10 static PetscErrorCode Tao_mcstep(TaoLineSearch ls,PetscReal *stx,PetscReal *fx,PetscReal *dx,PetscReal *sty,PetscReal *fy,PetscReal *dy,PetscReal *stp,PetscReal *fp,PetscReal *dp); 11 12 static PetscErrorCode TaoLineSearchDestroy_MT(TaoLineSearch ls) 13 { 14 TaoLineSearch_MT *mt = (TaoLineSearch_MT*)(ls->data); 15 16 PetscFunctionBegin; 17 PetscCall(PetscObjectDereference((PetscObject)mt->x)); 18 PetscCall(VecDestroy(&mt->work)); 19 PetscCall(PetscFree(ls->data)); 20 PetscFunctionReturn(0); 21 } 22 23 static PetscErrorCode TaoLineSearchSetFromOptions_MT(PetscOptionItems *PetscOptionsObject,TaoLineSearch ls) 24 { 25 PetscFunctionBegin; 26 PetscFunctionReturn(0); 27 } 28 29 static PetscErrorCode TaoLineSearchMonitor_MT(TaoLineSearch ls) 30 { 31 TaoLineSearch_MT *mt = (TaoLineSearch_MT*)ls->data; 32 33 PetscFunctionBegin; 34 PetscCall(PetscViewerASCIIPrintf(ls->viewer, "stx: %g, fx: %g, dgx: %g\n", (double)mt->stx, (double)mt->fx, (double)mt->dgx)); 35 PetscCall(PetscViewerASCIIPrintf(ls->viewer, "sty: %g, fy: %g, dgy: %g\n", (double)mt->sty, (double)mt->fy, (double)mt->dgy)); 36 PetscFunctionReturn(0); 37 } 38 39 static PetscErrorCode TaoLineSearchApply_MT(TaoLineSearch ls, Vec x, PetscReal *f, Vec g, Vec s) 40 { 41 TaoLineSearch_MT *mt = (TaoLineSearch_MT*)(ls->data); 42 PetscReal xtrapf = 4.0; 43 PetscReal finit, width, width1, dginit, fm, fxm, fym, dgm, dgxm, dgym; 44 PetscReal dgx, dgy, dg, dg2, fx, fy, stx, sty, dgtest; 45 PetscReal ftest1=0.0, ftest2=0.0; 46 PetscInt i, stage1,n1,n2,nn1,nn2; 47 PetscReal bstepmin1, bstepmin2, bstepmax, ostepmin, ostepmax; 48 PetscBool g_computed = PETSC_FALSE; /* to prevent extra gradient computation */ 49 50 PetscFunctionBegin; 51 ls->reason = TAOLINESEARCH_CONTINUE_ITERATING; 52 PetscCall(TaoLineSearchMonitor(ls, 0, *f, 0.0)); 53 /* Check work vector */ 54 if (!mt->work) { 55 PetscCall(VecDuplicate(x,&mt->work)); 56 mt->x = x; 57 PetscCall(PetscObjectReference((PetscObject)mt->x)); 58 } else if (x != mt->x) { 59 PetscCall(VecDestroy(&mt->work)); 60 PetscCall(VecDuplicate(x,&mt->work)); 61 PetscCall(PetscObjectDereference((PetscObject)mt->x)); 62 mt->x = x; 63 PetscCall(PetscObjectReference((PetscObject)mt->x)); 64 } 65 66 ostepmax = ls->stepmax; 67 ostepmin = ls->stepmin; 68 69 if (ls->bounded) { 70 /* Compute step length needed to make all variables equal a bound */ 71 /* Compute the smallest steplength that will make one nonbinding variable 72 equal the bound */ 73 PetscCall(VecGetLocalSize(ls->upper,&n1)); 74 PetscCall(VecGetLocalSize(mt->x, &n2)); 75 PetscCall(VecGetSize(ls->upper,&nn1)); 76 PetscCall(VecGetSize(mt->x,&nn2)); 77 PetscCheck(n1 == n2 && nn1 == nn2,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Variable vector not compatible with bounds vector"); 78 PetscCall(VecScale(s,-1.0)); 79 PetscCall(VecBoundGradientProjection(s,x,ls->lower,ls->upper,s)); 80 PetscCall(VecScale(s,-1.0)); 81 PetscCall(VecStepBoundInfo(x,s,ls->lower,ls->upper,&bstepmin1,&bstepmin2,&bstepmax)); 82 ls->stepmax = PetscMin(bstepmax,ls->stepmax); 83 } 84 85 PetscCall(VecDot(g,s,&dginit)); 86 if (PetscIsInfOrNanReal(dginit)) { 87 PetscCall(PetscInfo(ls,"Initial Line Search step * g is Inf or Nan (%g)\n",(double)dginit)); 88 ls->reason = TAOLINESEARCH_FAILED_INFORNAN; 89 PetscFunctionReturn(0); 90 } 91 if (dginit >= 0.0) { 92 PetscCall(PetscInfo(ls,"Initial Line Search step * g is not descent direction (%g)\n",(double)dginit)); 93 ls->reason = TAOLINESEARCH_FAILED_ASCENT; 94 PetscFunctionReturn(0); 95 } 96 97 /* Initialization */ 98 mt->bracket = 0; 99 stage1 = 1; 100 finit = *f; 101 dgtest = ls->ftol * dginit; 102 width = ls->stepmax - ls->stepmin; 103 width1 = width * 2.0; 104 PetscCall(VecCopy(x,mt->work)); 105 /* Variable dictionary: 106 stx, fx, dgx - the step, function, and derivative at the best step 107 sty, fy, dgy - the step, function, and derivative at the other endpoint 108 of the interval of uncertainty 109 step, f, dg - the step, function, and derivative at the current step */ 110 111 stx = 0.0; 112 fx = finit; 113 dgx = dginit; 114 sty = 0.0; 115 fy = finit; 116 dgy = dginit; 117 118 ls->step = ls->initstep; 119 for (i=0; i<ls->max_funcs; i++) { 120 /* Set min and max steps to correspond to the interval of uncertainty */ 121 if (mt->bracket) { 122 ls->stepmin = PetscMin(stx,sty); 123 ls->stepmax = PetscMax(stx,sty); 124 } else { 125 ls->stepmin = stx; 126 ls->stepmax = ls->step + xtrapf * (ls->step - stx); 127 } 128 129 /* Force the step to be within the bounds */ 130 ls->step = PetscMax(ls->step,ls->stepmin); 131 ls->step = PetscMin(ls->step,ls->stepmax); 132 133 /* If an unusual termination is to occur, then let step be the lowest 134 point obtained thus far */ 135 if (stx != 0 && ((mt->bracket && (ls->step <= ls->stepmin || ls->step >= ls->stepmax)) || (mt->bracket && (ls->stepmax - ls->stepmin <= ls->rtol * ls->stepmax)) || 136 (ls->nfeval + ls->nfgeval >= ls->max_funcs - 1) || mt->infoc == 0)) ls->step = stx; 137 138 PetscCall(VecWAXPY(mt->work,ls->step,s,x)); /* W = X + step*S */ 139 140 if (ls->bounded) PetscCall(VecMedian(ls->lower, mt->work, ls->upper, mt->work)); 141 if (ls->usegts) { 142 PetscCall(TaoLineSearchComputeObjectiveAndGTS(ls,mt->work,f,&dg)); 143 g_computed = PETSC_FALSE; 144 } else { 145 PetscCall(TaoLineSearchComputeObjectiveAndGradient(ls,mt->work,f,g)); 146 g_computed = PETSC_TRUE; 147 if (ls->bounded) { 148 PetscCall(VecDot(g,x,&dg)); 149 PetscCall(VecDot(g,mt->work,&dg2)); 150 dg = (dg2 - dg)/ls->step; 151 } else { 152 PetscCall(VecDot(g,s,&dg)); 153 } 154 } 155 156 /* update bracketing parameters in the MT context for printouts in monitor */ 157 mt->stx = stx; 158 mt->fx = fx; 159 mt->dgx = dgx; 160 mt->sty = sty; 161 mt->fy = fy; 162 mt->dgy = dgy; 163 PetscCall(TaoLineSearchMonitor(ls, i+1, *f, ls->step)); 164 165 if (i == 0) ls->f_fullstep = *f; 166 167 if (PetscIsInfOrNanReal(*f) || PetscIsInfOrNanReal(dg)) { 168 /* User provided compute function generated Not-a-Number, assume 169 domain violation and set function value and directional 170 derivative to infinity. */ 171 *f = PETSC_INFINITY; 172 dg = PETSC_INFINITY; 173 } 174 175 ftest1 = finit + ls->step * dgtest; 176 if (ls->bounded) ftest2 = finit + ls->step * dgtest * ls->ftol; 177 178 /* Convergence testing */ 179 if ((*f - ftest1 <= PETSC_SMALL * PetscAbsReal(finit)) && (PetscAbsReal(dg) + ls->gtol*dginit <= 0.0)) { 180 PetscCall(PetscInfo(ls, "Line search success: Sufficient decrease and directional deriv conditions hold\n")); 181 ls->reason = TAOLINESEARCH_SUCCESS; 182 break; 183 } 184 185 /* Check Armijo if beyond the first breakpoint */ 186 if (ls->bounded && *f <= ftest2 && ls->step >= bstepmin2) { 187 PetscCall(PetscInfo(ls,"Line search success: Sufficient decrease.\n")); 188 ls->reason = TAOLINESEARCH_SUCCESS; 189 break; 190 } 191 192 /* Checks for bad cases */ 193 if ((mt->bracket && (ls->step <= ls->stepmin || ls->step >= ls->stepmax)) || !mt->infoc) { 194 PetscCall(PetscInfo(ls,"Rounding errors may prevent further progress. May not be a step satisfying\nsufficient decrease and curvature conditions. Tolerances may be too small.\n")); 195 ls->reason = TAOLINESEARCH_HALTED_OTHER; 196 break; 197 } 198 if (ls->step == ls->stepmax && *f <= ftest1 && dg <= dgtest) { 199 PetscCall(PetscInfo(ls,"Step is at the upper bound, stepmax (%g)\n",(double)ls->stepmax)); 200 ls->reason = TAOLINESEARCH_HALTED_UPPERBOUND; 201 break; 202 } 203 if (ls->step == ls->stepmin && *f >= ftest1 && dg >= dgtest) { 204 PetscCall(PetscInfo(ls,"Step is at the lower bound, stepmin (%g)\n",(double)ls->stepmin)); 205 ls->reason = TAOLINESEARCH_HALTED_LOWERBOUND; 206 break; 207 } 208 if (mt->bracket && (ls->stepmax - ls->stepmin <= ls->rtol*ls->stepmax)) { 209 PetscCall(PetscInfo(ls,"Relative width of interval of uncertainty is at most rtol (%g)\n",(double)ls->rtol)); 210 ls->reason = TAOLINESEARCH_HALTED_RTOL; 211 break; 212 } 213 214 /* In the first stage, we seek a step for which the modified function 215 has a nonpositive value and nonnegative derivative */ 216 if (stage1 && *f <= ftest1 && dg >= dginit * PetscMin(ls->ftol, ls->gtol)) stage1 = 0; 217 218 /* A modified function is used to predict the step only if we 219 have not obtained a step for which the modified function has a 220 nonpositive function value and nonnegative derivative, and if a 221 lower function value has been obtained but the decrease is not 222 sufficient */ 223 224 if (stage1 && *f <= fx && *f > ftest1) { 225 fm = *f - ls->step * dgtest; /* Define modified function */ 226 fxm = fx - stx * dgtest; /* and derivatives */ 227 fym = fy - sty * dgtest; 228 dgm = dg - dgtest; 229 dgxm = dgx - dgtest; 230 dgym = dgy - dgtest; 231 232 /* if (dgxm * (ls->step - stx) >= 0.0) */ 233 /* Update the interval of uncertainty and compute the new step */ 234 PetscCall(Tao_mcstep(ls,&stx,&fxm,&dgxm,&sty,&fym,&dgym,&ls->step,&fm,&dgm)); 235 236 fx = fxm + stx * dgtest; /* Reset the function and */ 237 fy = fym + sty * dgtest; /* gradient values */ 238 dgx = dgxm + dgtest; 239 dgy = dgym + dgtest; 240 } else { 241 /* Update the interval of uncertainty and compute the new step */ 242 PetscCall(Tao_mcstep(ls,&stx,&fx,&dgx,&sty,&fy,&dgy,&ls->step,f,&dg)); 243 } 244 245 /* Force a sufficient decrease in the interval of uncertainty */ 246 if (mt->bracket) { 247 if (PetscAbsReal(sty - stx) >= 0.66 * width1) ls->step = stx + 0.5*(sty - stx); 248 width1 = width; 249 width = PetscAbsReal(sty - stx); 250 } 251 } 252 if (ls->nfeval + ls->nfgeval > ls->max_funcs) { 253 PetscCall(PetscInfo(ls,"Number of line search function evals (%" PetscInt_FMT ") > maximum (%" PetscInt_FMT ")\n",ls->nfeval+ls->nfgeval,ls->max_funcs)); 254 ls->reason = TAOLINESEARCH_HALTED_MAXFCN; 255 } 256 ls->stepmax = ostepmax; 257 ls->stepmin = ostepmin; 258 259 /* Finish computations */ 260 PetscCall(PetscInfo(ls,"%" PetscInt_FMT " function evals in line search, step = %g\n",ls->nfeval+ls->nfgeval,(double)ls->step)); 261 262 /* Set new solution vector and compute gradient if needed */ 263 PetscCall(VecCopy(mt->work,x)); 264 if (!g_computed) { 265 PetscCall(TaoLineSearchComputeGradient(ls,mt->work,g)); 266 } 267 PetscFunctionReturn(0); 268 } 269 270 /*MC 271 TAOLINESEARCHMT - Line-search type with cubic interpolation that satisfies both the sufficient decrease and 272 curvature conditions. This method can take step lengths greater than 1. 273 274 More-Thuente line-search can be selected with "-tao_ls_type more-thuente". 275 276 References: 277 . * - JORGE J. MORE AND DAVID J. THUENTE, LINE SEARCH ALGORITHMS WITH GUARANTEED SUFFICIENT DECREASE. 278 ACM Trans. Math. Software 20, no. 3 (1994): 286-307. 279 280 Level: developer 281 282 .seealso: `TaoLineSearchCreate()`, `TaoLineSearchSetType()`, `TaoLineSearchApply()` 283 284 .keywords: Tao, linesearch 285 M*/ 286 PETSC_EXTERN PetscErrorCode TaoLineSearchCreate_MT(TaoLineSearch ls) 287 { 288 TaoLineSearch_MT *ctx; 289 290 PetscFunctionBegin; 291 PetscValidHeaderSpecific(ls,TAOLINESEARCH_CLASSID,1); 292 PetscCall(PetscNewLog(ls,&ctx)); 293 ctx->bracket = 0; 294 ctx->infoc = 1; 295 ls->data = (void*)ctx; 296 ls->initstep = 1.0; 297 ls->ops->setup = NULL; 298 ls->ops->reset = NULL; 299 ls->ops->apply = TaoLineSearchApply_MT; 300 ls->ops->destroy = TaoLineSearchDestroy_MT; 301 ls->ops->setfromoptions = TaoLineSearchSetFromOptions_MT; 302 ls->ops->monitor = TaoLineSearchMonitor_MT; 303 PetscFunctionReturn(0); 304 } 305 306 /* 307 The subroutine mcstep is taken from the work of Jorge Nocedal. 308 this is a variant of More' and Thuente's routine. 309 310 subroutine mcstep 311 312 the purpose of mcstep is to compute a safeguarded step for 313 a linesearch and to update an interval of uncertainty for 314 a minimizer of the function. 315 316 the parameter stx contains the step with the least function 317 value. the parameter stp contains the current step. it is 318 assumed that the derivative at stx is negative in the 319 direction of the step. if bracket is set true then a 320 minimizer has been bracketed in an interval of uncertainty 321 with endpoints stx and sty. 322 323 the subroutine statement is 324 325 subroutine mcstep(stx,fx,dx,sty,fy,dy,stp,fp,dp,bracket, 326 stpmin,stpmax,info) 327 328 where 329 330 stx, fx, and dx are variables which specify the step, 331 the function, and the derivative at the best step obtained 332 so far. The derivative must be negative in the direction 333 of the step, that is, dx and stp-stx must have opposite 334 signs. On output these parameters are updated appropriately. 335 336 sty, fy, and dy are variables which specify the step, 337 the function, and the derivative at the other endpoint of 338 the interval of uncertainty. On output these parameters are 339 updated appropriately. 340 341 stp, fp, and dp are variables which specify the step, 342 the function, and the derivative at the current step. 343 If bracket is set true then on input stp must be 344 between stx and sty. On output stp is set to the new step. 345 346 bracket is a logical variable which specifies if a minimizer 347 has been bracketed. If the minimizer has not been bracketed 348 then on input bracket must be set false. If the minimizer 349 is bracketed then on output bracket is set true. 350 351 stpmin and stpmax are input variables which specify lower 352 and upper bounds for the step. 353 354 info is an integer output variable set as follows: 355 if info = 1,2,3,4,5, then the step has been computed 356 according to one of the five cases below. otherwise 357 info = 0, and this indicates improper input parameters. 358 359 subprograms called 360 361 fortran-supplied ... abs,max,min,sqrt 362 363 argonne national laboratory. minpack project. june 1983 364 jorge j. more', david j. thuente 365 366 */ 367 368 static PetscErrorCode Tao_mcstep(TaoLineSearch ls,PetscReal *stx,PetscReal *fx,PetscReal *dx,PetscReal *sty,PetscReal *fy,PetscReal *dy,PetscReal *stp,PetscReal *fp,PetscReal *dp) 369 { 370 TaoLineSearch_MT *mtP = (TaoLineSearch_MT *) ls->data; 371 PetscReal gamma1, p, q, r, s, sgnd, stpc, stpf, stpq, theta; 372 PetscInt bound; 373 374 PetscFunctionBegin; 375 /* Check the input parameters for errors */ 376 mtP->infoc = 0; 377 PetscCheck(!mtP->bracket || (*stp > PetscMin(*stx,*sty) && *stp < PetscMax(*stx,*sty)),PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"bad stp in bracket"); 378 PetscCheck(*dx * (*stp-*stx) < 0.0,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"dx * (stp-stx) >= 0.0"); 379 PetscCheck(ls->stepmax >= ls->stepmin,PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"stepmax > stepmin"); 380 381 /* Determine if the derivatives have opposite sign */ 382 sgnd = *dp * (*dx / PetscAbsReal(*dx)); 383 384 if (*fp > *fx) { 385 /* Case 1: a higher function value. 386 The minimum is bracketed. If the cubic step is closer 387 to stx than the quadratic step, the cubic step is taken, 388 else the average of the cubic and quadratic steps is taken. */ 389 390 mtP->infoc = 1; 391 bound = 1; 392 theta = 3 * (*fx - *fp) / (*stp - *stx) + *dx + *dp; 393 s = PetscMax(PetscAbsReal(theta),PetscAbsReal(*dx)); 394 s = PetscMax(s,PetscAbsReal(*dp)); 395 gamma1 = s*PetscSqrtScalar(PetscPowScalar(theta/s,2.0) - (*dx/s)*(*dp/s)); 396 if (*stp < *stx) gamma1 = -gamma1; 397 /* Can p be 0? Check */ 398 p = (gamma1 - *dx) + theta; 399 q = ((gamma1 - *dx) + gamma1) + *dp; 400 r = p/q; 401 stpc = *stx + r*(*stp - *stx); 402 stpq = *stx + ((*dx/((*fx-*fp)/(*stp-*stx)+*dx))*0.5) * (*stp - *stx); 403 404 if (PetscAbsReal(stpc-*stx) < PetscAbsReal(stpq-*stx)) stpf = stpc; 405 else stpf = stpc + 0.5*(stpq - stpc); 406 mtP->bracket = 1; 407 } else if (sgnd < 0.0) { 408 /* Case 2: A lower function value and derivatives of 409 opposite sign. The minimum is bracketed. If the cubic 410 step is closer to stx than the quadratic (secant) step, 411 the cubic step is taken, else the quadratic step is taken. */ 412 413 mtP->infoc = 2; 414 bound = 0; 415 theta = 3*(*fx - *fp)/(*stp - *stx) + *dx + *dp; 416 s = PetscMax(PetscAbsReal(theta),PetscAbsReal(*dx)); 417 s = PetscMax(s,PetscAbsReal(*dp)); 418 gamma1 = s*PetscSqrtScalar(PetscPowScalar(theta/s,2.0) - (*dx/s)*(*dp/s)); 419 if (*stp > *stx) gamma1 = -gamma1; 420 p = (gamma1 - *dp) + theta; 421 q = ((gamma1 - *dp) + gamma1) + *dx; 422 r = p/q; 423 stpc = *stp + r*(*stx - *stp); 424 stpq = *stp + (*dp/(*dp-*dx))*(*stx - *stp); 425 426 if (PetscAbsReal(stpc-*stp) > PetscAbsReal(stpq-*stp)) stpf = stpc; 427 else stpf = stpq; 428 mtP->bracket = 1; 429 } else if (PetscAbsReal(*dp) < PetscAbsReal(*dx)) { 430 /* Case 3: A lower function value, derivatives of the 431 same sign, and the magnitude of the derivative decreases. 432 The cubic step is only used if the cubic tends to infinity 433 in the direction of the step or if the minimum of the cubic 434 is beyond stp. Otherwise the cubic step is defined to be 435 either stepmin or stepmax. The quadratic (secant) step is also 436 computed and if the minimum is bracketed then the step 437 closest to stx is taken, else the step farthest away is taken. */ 438 439 mtP->infoc = 3; 440 bound = 1; 441 theta = 3*(*fx - *fp)/(*stp - *stx) + *dx + *dp; 442 s = PetscMax(PetscAbsReal(theta),PetscAbsReal(*dx)); 443 s = PetscMax(s,PetscAbsReal(*dp)); 444 445 /* The case gamma1 = 0 only arises if the cubic does not tend 446 to infinity in the direction of the step. */ 447 gamma1 = s*PetscSqrtScalar(PetscMax(0.0,PetscPowScalar(theta/s,2.0) - (*dx/s)*(*dp/s))); 448 if (*stp > *stx) gamma1 = -gamma1; 449 p = (gamma1 - *dp) + theta; 450 q = (gamma1 + (*dx - *dp)) + gamma1; 451 r = p/q; 452 if (r < 0.0 && gamma1 != 0.0) stpc = *stp + r*(*stx - *stp); 453 else if (*stp > *stx) stpc = ls->stepmax; 454 else stpc = ls->stepmin; 455 stpq = *stp + (*dp/(*dp-*dx)) * (*stx - *stp); 456 457 if (mtP->bracket) { 458 if (PetscAbsReal(*stp-stpc) < PetscAbsReal(*stp-stpq)) stpf = stpc; 459 else stpf = stpq; 460 } else { 461 if (PetscAbsReal(*stp-stpc) > PetscAbsReal(*stp-stpq)) stpf = stpc; 462 else stpf = stpq; 463 } 464 } else { 465 /* Case 4: A lower function value, derivatives of the 466 same sign, and the magnitude of the derivative does 467 not decrease. If the minimum is not bracketed, the step 468 is either stpmin or stpmax, else the cubic step is taken. */ 469 470 mtP->infoc = 4; 471 bound = 0; 472 if (mtP->bracket) { 473 theta = 3*(*fp - *fy)/(*sty - *stp) + *dy + *dp; 474 s = PetscMax(PetscAbsReal(theta),PetscAbsReal(*dy)); 475 s = PetscMax(s,PetscAbsReal(*dp)); 476 gamma1 = s*PetscSqrtScalar(PetscPowScalar(theta/s,2.0) - (*dy/s)*(*dp/s)); 477 if (*stp > *sty) gamma1 = -gamma1; 478 p = (gamma1 - *dp) + theta; 479 q = ((gamma1 - *dp) + gamma1) + *dy; 480 r = p/q; 481 stpc = *stp + r*(*sty - *stp); 482 stpf = stpc; 483 } else if (*stp > *stx) { 484 stpf = ls->stepmax; 485 } else { 486 stpf = ls->stepmin; 487 } 488 } 489 490 /* Update the interval of uncertainty. This update does not 491 depend on the new step or the case analysis above. */ 492 493 if (*fp > *fx) { 494 *sty = *stp; 495 *fy = *fp; 496 *dy = *dp; 497 } else { 498 if (sgnd < 0.0) { 499 *sty = *stx; 500 *fy = *fx; 501 *dy = *dx; 502 } 503 *stx = *stp; 504 *fx = *fp; 505 *dx = *dp; 506 } 507 508 /* Compute the new step and safeguard it. */ 509 stpf = PetscMin(ls->stepmax,stpf); 510 stpf = PetscMax(ls->stepmin,stpf); 511 *stp = stpf; 512 if (mtP->bracket && bound) { 513 if (*sty > *stx) *stp = PetscMin(*stx+0.66*(*sty-*stx),*stp); 514 else *stp = PetscMax(*stx+0.66*(*sty-*stx),*stp); 515 } 516 PetscFunctionReturn(0); 517 } 518