1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
4 * This source code is part of
8 * GROningen MAchine for Chemical Simulations
11 * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
12 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
13 * Copyright (c) 2001-2004, The GROMACS development team,
14 * check out http://www.gromacs.org for more information.
16 * This program is free software; you can redistribute it and/or
17 * modify it under the terms of the GNU General Public License
18 * as published by the Free Software Foundation; either version 2
19 * of the License, or (at your option) any later version.
21 * If you want to redistribute modifications, please consider that
22 * scientific software is very special. Version control is crucial -
23 * bugs must be traceable. We will be happy to consider code for
24 * inclusion in the official distribution, but derived work must not
25 * be called official GROMACS. Details are found in the README & COPYING
26 * files - if they are missing, get the official version at www.gromacs.org.
28 * To help us fund GROMACS development, we humbly ask that you cite
29 * the papers on the package - you can find them in the top README file.
31 * For more info, check our website at http://www.gromacs.org
34 * GROwing Monsters And Cloning Shrimps
55 #include "gmx_fatal.h"
70 #include "sparsematrix.h"
74 #include "gmx_wallcycle.h"
75 #include "mtop_util.h"
88 static em_state_t *init_em_state()
97 static void print_em_start(FILE *fplog,t_commrec *cr,gmx_runtime_t *runtime,
98 gmx_wallcycle_t wcycle,
103 runtime_start(runtime);
105 sprintf(buf,"Started %s",name);
106 print_date_and_time(fplog,cr->nodeid,buf,NULL);
108 wallcycle_start(wcycle,ewcRUN);
110 static void em_time_end(FILE *fplog,t_commrec *cr,gmx_runtime_t *runtime,
111 gmx_wallcycle_t wcycle)
113 wallcycle_stop(wcycle,ewcRUN);
115 runtime_end(runtime);
118 static void sp_header(FILE *out,const char *minimizer,real ftol,int nsteps)
121 fprintf(out,"%s:\n",minimizer);
122 fprintf(out," Tolerance (Fmax) = %12.5e\n",ftol);
123 fprintf(out," Number of steps = %12d\n",nsteps);
126 static void warn_step(FILE *fp,real ftol,gmx_bool bLastStep,gmx_bool bConstrain)
130 fprintf(fp,"\nReached the maximum number of steps before reaching Fmax < %g\n",ftol);
134 fprintf(fp,"\nStepsize too small, or no change in energy.\n"
135 "Converged to machine precision,\n"
136 "but not to the requested precision Fmax < %g\n",
138 if (sizeof(real)<sizeof(double))
140 fprintf(fp,"\nDouble precision normally gives you higher accuracy.\n");
144 fprintf(fp,"You might need to increase your constraint accuracy, or turn\n"
145 "off constraints alltogether (set constraints = none in mdp file)\n");
152 static void print_converged(FILE *fp,const char *alg,real ftol,
153 gmx_large_int_t count,gmx_bool bDone,gmx_large_int_t nsteps,
154 real epot,real fmax, int nfmax, real fnorm)
156 char buf[STEPSTRSIZE];
159 fprintf(fp,"\n%s converged to Fmax < %g in %s steps\n",
160 alg,ftol,gmx_step_str(count,buf));
161 else if(count<nsteps)
162 fprintf(fp,"\n%s converged to machine precision in %s steps,\n"
163 "but did not reach the requested Fmax < %g.\n",
164 alg,gmx_step_str(count,buf),ftol);
166 fprintf(fp,"\n%s did not converge to Fmax < %g in %s steps.\n",
167 alg,ftol,gmx_step_str(count,buf));
170 fprintf(fp,"Potential Energy = %21.14e\n",epot);
171 fprintf(fp,"Maximum force = %21.14e on atom %d\n",fmax,nfmax+1);
172 fprintf(fp,"Norm of force = %21.14e\n",fnorm);
174 fprintf(fp,"Potential Energy = %14.7e\n",epot);
175 fprintf(fp,"Maximum force = %14.7e on atom %d\n",fmax,nfmax+1);
176 fprintf(fp,"Norm of force = %14.7e\n",fnorm);
180 static void get_f_norm_max(t_commrec *cr,
181 t_grpopts *opts,t_mdatoms *mdatoms,rvec *f,
182 real *fnorm,real *fmax,int *a_fmax)
185 real fmax2,fmax2_0,fam;
186 int la_max,a_max,start,end,i,m,gf;
188 /* This routine finds the largest force and returns it.
189 * On parallel machines the global max is taken.
195 start = mdatoms->start;
196 end = mdatoms->homenr + start;
197 if (mdatoms->cFREEZE) {
198 for(i=start; i<end; i++) {
199 gf = mdatoms->cFREEZE[i];
202 if (!opts->nFreeze[gf][m])
211 for(i=start; i<end; i++) {
221 if (la_max >= 0 && DOMAINDECOMP(cr)) {
222 a_max = cr->dd->gatindex[la_max];
227 snew(sum,2*cr->nnodes+1);
228 sum[2*cr->nodeid] = fmax2;
229 sum[2*cr->nodeid+1] = a_max;
230 sum[2*cr->nnodes] = fnorm2;
231 gmx_sumd(2*cr->nnodes+1,sum,cr);
232 fnorm2 = sum[2*cr->nnodes];
233 /* Determine the global maximum */
234 for(i=0; i<cr->nnodes; i++) {
235 if (sum[2*i] > fmax2) {
237 a_max = (int)(sum[2*i+1] + 0.5);
244 *fnorm = sqrt(fnorm2);
251 static void get_state_f_norm_max(t_commrec *cr,
252 t_grpopts *opts,t_mdatoms *mdatoms,
255 get_f_norm_max(cr,opts,mdatoms,ems->f,&ems->fnorm,&ems->fmax,&ems->a_fmax);
258 void init_em(FILE *fplog,const char *title,
259 t_commrec *cr,t_inputrec *ir,
260 t_state *state_global,gmx_mtop_t *top_global,
261 em_state_t *ems,gmx_localtop_t **top,
262 rvec **f,rvec **f_global,
263 t_nrnb *nrnb,rvec mu_tot,
264 t_forcerec *fr,gmx_enerdata_t **enerd,
265 t_graph **graph,t_mdatoms *mdatoms,gmx_global_stat_t *gstat,
266 gmx_vsite_t *vsite,gmx_constr_t constr,
267 int nfile,const t_filenm fnm[],
268 gmx_mdoutf_t **outf,t_mdebin **mdebin)
275 fprintf(fplog,"Initiating %s\n",title);
278 state_global->ngtc = 0;
280 /* Initiate some variables */
281 if (ir->efep != efepNO)
283 state_global->lambda = ir->init_lambda;
287 state_global->lambda = 0.0;
292 if (DOMAINDECOMP(cr))
294 *top = dd_init_local_top(top_global);
296 dd_init_local_state(cr->dd,state_global,&ems->s);
300 /* Distribute the charge groups over the nodes from the master node */
301 dd_partition_system(fplog,ir->init_step,cr,TRUE,1,
302 state_global,top_global,ir,
303 &ems->s,&ems->f,mdatoms,*top,
304 fr,vsite,NULL,constr,
306 dd_store_state(cr->dd,&ems->s);
310 snew(*f_global,top_global->natoms);
320 snew(*f,top_global->natoms);
322 /* Just copy the state */
323 ems->s = *state_global;
324 snew(ems->s.x,ems->s.nalloc);
325 snew(ems->f,ems->s.nalloc);
326 for(i=0; i<state_global->natoms; i++)
328 copy_rvec(state_global->x[i],ems->s.x[i]);
330 copy_mat(state_global->box,ems->s.box);
332 if (PAR(cr) && ir->eI != eiNM)
334 /* Initialize the particle decomposition and split the topology */
335 *top = split_system(fplog,top_global,ir,cr);
337 pd_cg_range(cr,&fr->cg0,&fr->hcg);
341 *top = gmx_mtop_generate_local_top(top_global,ir);
345 if (ir->ePBC != epbcNONE && !ir->bPeriodicMols)
347 *graph = mk_graph(fplog,&((*top)->idef),0,top_global->natoms,FALSE,FALSE);
356 pd_at_range(cr,&start,&homenr);
362 homenr = top_global->natoms;
364 atoms2md(top_global,ir,0,NULL,start,homenr,mdatoms);
365 update_mdatoms(mdatoms,state_global->lambda);
369 set_vsite_top(vsite,*top,mdatoms,cr);
375 if (ir->eConstrAlg == econtSHAKE &&
376 gmx_mtop_ftype_count(top_global,F_CONSTR) > 0)
378 gmx_fatal(FARGS,"Can not do energy minimization with %s, use %s\n",
379 econstr_names[econtSHAKE],econstr_names[econtLINCS]);
382 if (!DOMAINDECOMP(cr))
384 set_constraints(constr,*top,ir,mdatoms,cr);
387 if (!ir->bContinuation)
389 /* Constrain the starting coordinates */
391 constrain(PAR(cr) ? NULL : fplog,TRUE,TRUE,constr,&(*top)->idef,
392 ir,NULL,cr,-1,0,mdatoms,
393 ems->s.x,ems->s.x,NULL,ems->s.box,
394 ems->s.lambda,&dvdlambda,
395 NULL,NULL,nrnb,econqCoord,FALSE,0,0);
401 *gstat = global_stat_init(ir);
404 *outf = init_mdoutf(nfile,fnm,0,cr,ir,NULL);
407 init_enerdata(top_global->groups.grps[egcENER].nr,ir->n_flambda,*enerd);
411 /* Init bin for energy stuff */
412 *mdebin = init_mdebin((*outf)->fp_ene,top_global,ir,NULL);
416 calc_shifts(ems->s.box,fr->shift_vec);
419 static void finish_em(FILE *fplog,t_commrec *cr,gmx_mdoutf_t *outf,
420 gmx_runtime_t *runtime,gmx_wallcycle_t wcycle)
422 if (!(cr->duty & DUTY_PME)) {
423 /* Tell the PME only node to finish */
429 em_time_end(fplog,cr,runtime,wcycle);
432 static void swap_em_state(em_state_t *ems1,em_state_t *ems2)
441 static void copy_em_coords(em_state_t *ems,t_state *state)
445 for(i=0; (i<state->natoms); i++)
447 copy_rvec(ems->s.x[i],state->x[i]);
451 static void write_em_traj(FILE *fplog,t_commrec *cr,
453 gmx_bool bX,gmx_bool bF,const char *confout,
454 gmx_mtop_t *top_global,
455 t_inputrec *ir,gmx_large_int_t step,
457 t_state *state_global,rvec *f_global)
461 if ((bX || bF || confout != NULL) && !DOMAINDECOMP(cr))
463 copy_em_coords(state,state_global);
468 if (bX) { mdof_flags |= MDOF_X; }
469 if (bF) { mdof_flags |= MDOF_F; }
470 write_traj(fplog,cr,outf,mdof_flags,
471 top_global,step,(double)step,
472 &state->s,state_global,state->f,f_global,NULL,NULL);
474 if (confout != NULL && MASTER(cr))
476 if (ir->ePBC != epbcNONE && !ir->bPeriodicMols && DOMAINDECOMP(cr))
478 /* Make molecules whole only for confout writing */
479 do_pbc_mtop(fplog,ir->ePBC,state_global->box,top_global,
483 write_sto_conf_mtop(confout,
484 *top_global->name,top_global,
485 state_global->x,NULL,ir->ePBC,state_global->box);
489 static void do_em_step(t_commrec *cr,t_inputrec *ir,t_mdatoms *md,
490 em_state_t *ems1,real a,rvec *f,em_state_t *ems2,
491 gmx_constr_t constr,gmx_localtop_t *top,
492 t_nrnb *nrnb,gmx_wallcycle_t wcycle,
493 gmx_large_int_t count)
497 int start,end,gf,i,m;
504 if (DOMAINDECOMP(cr) && s1->ddp_count != cr->dd->ddp_count)
505 gmx_incons("state mismatch in do_em_step");
507 s2->flags = s1->flags;
509 if (s2->nalloc != s1->nalloc) {
510 s2->nalloc = s1->nalloc;
511 srenew(s2->x,s1->nalloc);
512 srenew(ems2->f, s1->nalloc);
513 if (s2->flags & (1<<estCGP))
514 srenew(s2->cg_p, s1->nalloc);
517 s2->natoms = s1->natoms;
518 s2->lambda = s1->lambda;
519 copy_mat(s1->box,s2->box);
522 end = md->start + md->homenr;
527 for(i=start; i<end; i++) {
530 for(m=0; m<DIM; m++) {
531 if (ir->opts.nFreeze[gf][m])
534 x2[i][m] = x1[i][m] + a*f[i][m];
538 if (s2->flags & (1<<estCGP)) {
539 /* Copy the CG p vector */
542 for(i=start; i<end; i++)
543 copy_rvec(x1[i],x2[i]);
546 if (DOMAINDECOMP(cr)) {
547 s2->ddp_count = s1->ddp_count;
548 if (s2->cg_gl_nalloc < s1->cg_gl_nalloc) {
549 s2->cg_gl_nalloc = s1->cg_gl_nalloc;
550 srenew(s2->cg_gl,s2->cg_gl_nalloc);
552 s2->ncg_gl = s1->ncg_gl;
553 for(i=0; i<s2->ncg_gl; i++)
554 s2->cg_gl[i] = s1->cg_gl[i];
555 s2->ddp_count_cg_gl = s1->ddp_count_cg_gl;
559 wallcycle_start(wcycle,ewcCONSTR);
561 constrain(NULL,TRUE,TRUE,constr,&top->idef,
562 ir,NULL,cr,count,0,md,
563 s1->x,s2->x,NULL,s2->box,s2->lambda,
564 &dvdlambda,NULL,NULL,nrnb,econqCoord,FALSE,0,0);
565 wallcycle_stop(wcycle,ewcCONSTR);
569 static void em_dd_partition_system(FILE *fplog,int step,t_commrec *cr,
570 gmx_mtop_t *top_global,t_inputrec *ir,
571 em_state_t *ems,gmx_localtop_t *top,
572 t_mdatoms *mdatoms,t_forcerec *fr,
573 gmx_vsite_t *vsite,gmx_constr_t constr,
574 t_nrnb *nrnb,gmx_wallcycle_t wcycle)
576 /* Repartition the domain decomposition */
577 wallcycle_start(wcycle,ewcDOMDEC);
578 dd_partition_system(fplog,step,cr,FALSE,1,
581 mdatoms,top,fr,vsite,NULL,constr,
583 dd_store_state(cr->dd,&ems->s);
584 wallcycle_stop(wcycle,ewcDOMDEC);
587 static void evaluate_energy(FILE *fplog,gmx_bool bVerbose,t_commrec *cr,
588 t_state *state_global,gmx_mtop_t *top_global,
589 em_state_t *ems,gmx_localtop_t *top,
590 t_inputrec *inputrec,
591 t_nrnb *nrnb,gmx_wallcycle_t wcycle,
592 gmx_global_stat_t gstat,
593 gmx_vsite_t *vsite,gmx_constr_t constr,
595 t_graph *graph,t_mdatoms *mdatoms,
596 t_forcerec *fr,rvec mu_tot,
597 gmx_enerdata_t *enerd,tensor vir,tensor pres,
598 gmx_large_int_t count,gmx_bool bFirst)
603 tensor force_vir,shake_vir,ekin;
604 real dvdl,prescorr,enercorr,dvdlcorr;
607 /* Set the time to the initial time, the time does not change during EM */
608 t = inputrec->init_t;
611 (DOMAINDECOMP(cr) && ems->s.ddp_count < cr->dd->ddp_count)) {
612 /* This the first state or an old state used before the last ns */
616 if (inputrec->nstlist > 0) {
618 } else if (inputrec->nstlist == -1) {
619 nabnsb = natoms_beyond_ns_buffer(inputrec,fr,&top->cgs,NULL,ems->s.x);
621 gmx_sumi(1,&nabnsb,cr);
627 construct_vsites(fplog,vsite,ems->s.x,nrnb,1,NULL,
628 top->idef.iparams,top->idef.il,
629 fr->ePBC,fr->bMolPBC,graph,cr,ems->s.box);
631 if (DOMAINDECOMP(cr)) {
633 /* Repartition the domain decomposition */
634 em_dd_partition_system(fplog,count,cr,top_global,inputrec,
635 ems,top,mdatoms,fr,vsite,constr,
640 /* Calc force & energy on new trial position */
641 /* do_force always puts the charge groups in the box and shifts again
642 * We do not unshift, so molecules are always whole in congrad.c
644 do_force(fplog,cr,inputrec,
645 count,nrnb,wcycle,top,top_global,&top_global->groups,
646 ems->s.box,ems->s.x,&ems->s.hist,
647 ems->f,force_vir,mdatoms,enerd,fcd,
648 ems->s.lambda,graph,fr,vsite,mu_tot,t,NULL,NULL,TRUE,
649 GMX_FORCE_STATECHANGED | GMX_FORCE_ALLFORCES | GMX_FORCE_VIRIAL |
650 (bNS ? GMX_FORCE_NS | GMX_FORCE_DOLR : 0));
652 /* Clear the unused shake virial and pressure */
653 clear_mat(shake_vir);
656 /* Communicate stuff when parallel */
657 if (PAR(cr) && inputrec->eI != eiNM)
659 wallcycle_start(wcycle,ewcMoveE);
661 global_stat(fplog,gstat,cr,enerd,force_vir,shake_vir,mu_tot,
662 inputrec,NULL,NULL,NULL,1,&terminate,
663 top_global,&ems->s,FALSE,
669 wallcycle_stop(wcycle,ewcMoveE);
672 /* Calculate long range corrections to pressure and energy */
673 calc_dispcorr(fplog,inputrec,fr,count,top_global->natoms,ems->s.box,ems->s.lambda,
674 pres,force_vir,&prescorr,&enercorr,&dvdlcorr);
675 enerd->term[F_DISPCORR] = enercorr;
676 enerd->term[F_EPOT] += enercorr;
677 enerd->term[F_PRES] += prescorr;
678 enerd->term[F_DVDL] += dvdlcorr;
680 ems->epot = enerd->term[F_EPOT];
683 /* Project out the constraint components of the force */
684 wallcycle_start(wcycle,ewcCONSTR);
686 constrain(NULL,FALSE,FALSE,constr,&top->idef,
687 inputrec,NULL,cr,count,0,mdatoms,
688 ems->s.x,ems->f,ems->f,ems->s.box,ems->s.lambda,&dvdl,
689 NULL,&shake_vir,nrnb,econqForceDispl,FALSE,0,0);
690 if (fr->bSepDVDL && fplog)
691 fprintf(fplog,sepdvdlformat,"Constraints",t,dvdl);
692 enerd->term[F_DHDL_CON] += dvdl;
693 m_add(force_vir,shake_vir,vir);
694 wallcycle_stop(wcycle,ewcCONSTR);
696 copy_mat(force_vir,vir);
700 enerd->term[F_PRES] =
701 calc_pres(fr->ePBC,inputrec->nwall,ems->s.box,ekin,vir,pres);
703 sum_dhdl(enerd,ems->s.lambda,inputrec);
705 if (EI_ENERGY_MINIMIZATION(inputrec->eI))
707 get_state_f_norm_max(cr,&(inputrec->opts),mdatoms,ems);
711 static double reorder_partsum(t_commrec *cr,t_grpopts *opts,t_mdatoms *mdatoms,
713 em_state_t *s_min,em_state_t *s_b)
717 int ncg,*cg_gl,*index,c,cg,i,a0,a1,a,gf,m;
719 unsigned char *grpnrFREEZE;
722 fprintf(debug,"Doing reorder_partsum\n");
727 cgs_gl = dd_charge_groups_global(cr->dd);
728 index = cgs_gl->index;
730 /* Collect fm in a global vector fmg.
731 * This conflicts with the spirit of domain decomposition,
732 * but to fully optimize this a much more complicated algorithm is required.
734 snew(fmg,mtop->natoms);
736 ncg = s_min->s.ncg_gl;
737 cg_gl = s_min->s.cg_gl;
739 for(c=0; c<ncg; c++) {
743 for(a=a0; a<a1; a++) {
744 copy_rvec(fm[i],fmg[a]);
748 gmx_sum(mtop->natoms*3,fmg[0],cr);
750 /* Now we will determine the part of the sum for the cgs in state s_b */
752 cg_gl = s_b->s.cg_gl;
756 grpnrFREEZE = mtop->groups.grpnr[egcFREEZE];
757 for(c=0; c<ncg; c++) {
761 for(a=a0; a<a1; a++) {
762 if (mdatoms->cFREEZE && grpnrFREEZE) {
765 for(m=0; m<DIM; m++) {
766 if (!opts->nFreeze[gf][m]) {
767 partsum += (fb[i][m] - fmg[a][m])*fb[i][m];
779 static real pr_beta(t_commrec *cr,t_grpopts *opts,t_mdatoms *mdatoms,
781 em_state_t *s_min,em_state_t *s_b)
787 /* This is just the classical Polak-Ribiere calculation of beta;
788 * it looks a bit complicated since we take freeze groups into account,
789 * and might have to sum it in parallel runs.
792 if (!DOMAINDECOMP(cr) ||
793 (s_min->s.ddp_count == cr->dd->ddp_count &&
794 s_b->s.ddp_count == cr->dd->ddp_count)) {
799 /* This part of code can be incorrect with DD,
800 * since the atom ordering in s_b and s_min might differ.
802 for(i=mdatoms->start; i<mdatoms->start+mdatoms->homenr; i++) {
803 if (mdatoms->cFREEZE)
804 gf = mdatoms->cFREEZE[i];
806 if (!opts->nFreeze[gf][m]) {
807 sum += (fb[i][m] - fm[i][m])*fb[i][m];
811 /* We need to reorder cgs while summing */
812 sum = reorder_partsum(cr,opts,mdatoms,mtop,s_min,s_b);
817 return sum/sqr(s_min->fnorm);
820 double do_cg(FILE *fplog,t_commrec *cr,
821 int nfile,const t_filenm fnm[],
822 const output_env_t oenv, gmx_bool bVerbose,gmx_bool bCompact,
824 gmx_vsite_t *vsite,gmx_constr_t constr,
826 t_inputrec *inputrec,
827 gmx_mtop_t *top_global,t_fcdata *fcd,
828 t_state *state_global,
830 t_nrnb *nrnb,gmx_wallcycle_t wcycle,
833 int repl_ex_nst,int repl_ex_seed,
835 real cpt_period,real max_hours,
836 const char *deviceOptions,
838 gmx_runtime_t *runtime)
840 const char *CG="Polak-Ribiere Conjugate Gradients";
842 em_state_t *s_min,*s_a,*s_b,*s_c;
844 gmx_enerdata_t *enerd;
846 gmx_global_stat_t gstat;
848 rvec *f_global,*p,*sf,*sfm;
849 double gpa,gpb,gpc,tmp,sum[2],minstep;
856 gmx_bool converged,foundlower;
858 gmx_bool do_log=FALSE,do_ene=FALSE,do_x,do_f;
860 int number_steps,neval=0,nstcg=inputrec->nstcgsteep;
862 int i,m,gf,step,nminstep;
867 s_min = init_em_state();
868 s_a = init_em_state();
869 s_b = init_em_state();
870 s_c = init_em_state();
872 /* Init em and store the local state in s_min */
873 init_em(fplog,CG,cr,inputrec,
874 state_global,top_global,s_min,&top,&f,&f_global,
875 nrnb,mu_tot,fr,&enerd,&graph,mdatoms,&gstat,vsite,constr,
876 nfile,fnm,&outf,&mdebin);
878 /* Print to log file */
879 print_em_start(fplog,cr,runtime,wcycle,CG);
881 /* Max number of steps */
882 number_steps=inputrec->nsteps;
885 sp_header(stderr,CG,inputrec->em_tol,number_steps);
887 sp_header(fplog,CG,inputrec->em_tol,number_steps);
889 /* Call the force routine and some auxiliary (neighboursearching etc.) */
890 /* do_force always puts the charge groups in the box and shifts again
891 * We do not unshift, so molecules are always whole in congrad.c
893 evaluate_energy(fplog,bVerbose,cr,
894 state_global,top_global,s_min,top,
895 inputrec,nrnb,wcycle,gstat,
896 vsite,constr,fcd,graph,mdatoms,fr,
897 mu_tot,enerd,vir,pres,-1,TRUE);
901 /* Copy stuff to the energy bin for easy printing etc. */
902 upd_mdebin(mdebin,FALSE,FALSE,(double)step,
903 mdatoms->tmass,enerd,&s_min->s,s_min->s.box,
904 NULL,NULL,vir,pres,NULL,mu_tot,constr);
906 print_ebin_header(fplog,step,step,s_min->s.lambda);
907 print_ebin(outf->fp_ene,TRUE,FALSE,FALSE,fplog,step,step,eprNORMAL,
908 TRUE,mdebin,fcd,&(top_global->groups),&(inputrec->opts));
912 /* Estimate/guess the initial stepsize */
913 stepsize = inputrec->em_stepsize/s_min->fnorm;
916 fprintf(stderr," F-max = %12.5e on atom %d\n",
917 s_min->fmax,s_min->a_fmax+1);
918 fprintf(stderr," F-Norm = %12.5e\n",
919 s_min->fnorm/sqrt(state_global->natoms));
920 fprintf(stderr,"\n");
921 /* and copy to the log file too... */
922 fprintf(fplog," F-max = %12.5e on atom %d\n",
923 s_min->fmax,s_min->a_fmax+1);
924 fprintf(fplog," F-Norm = %12.5e\n",
925 s_min->fnorm/sqrt(state_global->natoms));
928 /* Start the loop over CG steps.
929 * Each successful step is counted, and we continue until
930 * we either converge or reach the max number of steps.
933 for(step=0; (number_steps<0 || (number_steps>=0 && step<=number_steps)) && !converged;step++) {
935 /* start taking steps in a new direction
936 * First time we enter the routine, beta=0, and the direction is
937 * simply the negative gradient.
940 /* Calculate the new direction in p, and the gradient in this direction, gpa */
945 for(i=mdatoms->start; i<mdatoms->start+mdatoms->homenr; i++) {
946 if (mdatoms->cFREEZE)
947 gf = mdatoms->cFREEZE[i];
948 for(m=0; m<DIM; m++) {
949 if (!inputrec->opts.nFreeze[gf][m]) {
950 p[i][m] = sf[i][m] + beta*p[i][m];
951 gpa -= p[i][m]*sf[i][m];
952 /* f is negative gradient, thus the sign */
959 /* Sum the gradient along the line across CPUs */
963 /* Calculate the norm of the search vector */
964 get_f_norm_max(cr,&(inputrec->opts),mdatoms,p,&pnorm,NULL,NULL);
966 /* Just in case stepsize reaches zero due to numerical precision... */
968 stepsize = inputrec->em_stepsize/pnorm;
971 * Double check the value of the derivative in the search direction.
972 * If it is positive it must be due to the old information in the
973 * CG formula, so just remove that and start over with beta=0.
974 * This corresponds to a steepest descent step.
978 step--; /* Don't count this step since we are restarting */
979 continue; /* Go back to the beginning of the big for-loop */
982 /* Calculate minimum allowed stepsize, before the average (norm)
983 * relative change in coordinate is smaller than precision
986 for (i=mdatoms->start; i<mdatoms->start+mdatoms->homenr; i++) {
987 for(m=0; m<DIM; m++) {
988 tmp = fabs(s_min->s.x[i][m]);
995 /* Add up from all CPUs */
997 gmx_sumd(1,&minstep,cr);
999 minstep = GMX_REAL_EPS/sqrt(minstep/(3*state_global->natoms));
1001 if(stepsize<minstep) {
1006 /* Write coordinates if necessary */
1007 do_x = do_per_step(step,inputrec->nstxout);
1008 do_f = do_per_step(step,inputrec->nstfout);
1010 write_em_traj(fplog,cr,outf,do_x,do_f,NULL,
1011 top_global,inputrec,step,
1012 s_min,state_global,f_global);
1014 /* Take a step downhill.
1015 * In theory, we should minimize the function along this direction.
1016 * That is quite possible, but it turns out to take 5-10 function evaluations
1017 * for each line. However, we dont really need to find the exact minimum -
1018 * it is much better to start a new CG step in a modified direction as soon
1019 * as we are close to it. This will save a lot of energy evaluations.
1021 * In practice, we just try to take a single step.
1022 * If it worked (i.e. lowered the energy), we increase the stepsize but
1023 * the continue straight to the next CG step without trying to find any minimum.
1024 * If it didn't work (higher energy), there must be a minimum somewhere between
1025 * the old position and the new one.
1027 * Due to the finite numerical accuracy, it turns out that it is a good idea
1028 * to even accept a SMALL increase in energy, if the derivative is still downhill.
1029 * This leads to lower final energies in the tests I've done. / Erik
1031 s_a->epot = s_min->epot;
1033 c = a + stepsize; /* reference position along line is zero */
1035 if (DOMAINDECOMP(cr) && s_min->s.ddp_count < cr->dd->ddp_count) {
1036 em_dd_partition_system(fplog,step,cr,top_global,inputrec,
1037 s_min,top,mdatoms,fr,vsite,constr,
1041 /* Take a trial step (new coords in s_c) */
1042 do_em_step(cr,inputrec,mdatoms,s_min,c,s_min->s.cg_p,s_c,
1043 constr,top,nrnb,wcycle,-1);
1046 /* Calculate energy for the trial step */
1047 evaluate_energy(fplog,bVerbose,cr,
1048 state_global,top_global,s_c,top,
1049 inputrec,nrnb,wcycle,gstat,
1050 vsite,constr,fcd,graph,mdatoms,fr,
1051 mu_tot,enerd,vir,pres,-1,FALSE);
1053 /* Calc derivative along line */
1057 for(i=mdatoms->start; i<mdatoms->start+mdatoms->homenr; i++) {
1058 for(m=0; m<DIM; m++)
1059 gpc -= p[i][m]*sf[i][m]; /* f is negative gradient, thus the sign */
1061 /* Sum the gradient along the line across CPUs */
1063 gmx_sumd(1,&gpc,cr);
1065 /* This is the max amount of increase in energy we tolerate */
1066 tmp=sqrt(GMX_REAL_EPS)*fabs(s_a->epot);
1068 /* Accept the step if the energy is lower, or if it is not significantly higher
1069 * and the line derivative is still negative.
1071 if (s_c->epot < s_a->epot || (gpc < 0 && s_c->epot < (s_a->epot + tmp))) {
1073 /* Great, we found a better energy. Increase step for next iteration
1074 * if we are still going down, decrease it otherwise
1077 stepsize *= 1.618034; /* The golden section */
1079 stepsize *= 0.618034; /* 1/golden section */
1081 /* New energy is the same or higher. We will have to do some work
1082 * to find a smaller value in the interval. Take smaller step next time!
1085 stepsize *= 0.618034;
1091 /* OK, if we didn't find a lower value we will have to locate one now - there must
1092 * be one in the interval [a=0,c].
1093 * The same thing is valid here, though: Don't spend dozens of iterations to find
1094 * the line minimum. We try to interpolate based on the derivative at the endpoints,
1095 * and only continue until we find a lower value. In most cases this means 1-2 iterations.
1097 * I also have a safeguard for potentially really patological functions so we never
1098 * take more than 20 steps before we give up ...
1100 * If we already found a lower value we just skip this step and continue to the update.
1106 /* Select a new trial point.
1107 * If the derivatives at points a & c have different sign we interpolate to zero,
1108 * otherwise just do a bisection.
1111 b = a + gpa*(a-c)/(gpc-gpa);
1115 /* safeguard if interpolation close to machine accuracy causes errors:
1116 * never go outside the interval
1121 if (DOMAINDECOMP(cr) && s_min->s.ddp_count != cr->dd->ddp_count) {
1122 /* Reload the old state */
1123 em_dd_partition_system(fplog,-1,cr,top_global,inputrec,
1124 s_min,top,mdatoms,fr,vsite,constr,
1128 /* Take a trial step to this new point - new coords in s_b */
1129 do_em_step(cr,inputrec,mdatoms,s_min,b,s_min->s.cg_p,s_b,
1130 constr,top,nrnb,wcycle,-1);
1133 /* Calculate energy for the trial step */
1134 evaluate_energy(fplog,bVerbose,cr,
1135 state_global,top_global,s_b,top,
1136 inputrec,nrnb,wcycle,gstat,
1137 vsite,constr,fcd,graph,mdatoms,fr,
1138 mu_tot,enerd,vir,pres,-1,FALSE);
1140 /* p does not change within a step, but since the domain decomposition
1141 * might change, we have to use cg_p of s_b here.
1146 for(i=mdatoms->start; i<mdatoms->start+mdatoms->homenr; i++) {
1147 for(m=0; m<DIM; m++)
1148 gpb -= p[i][m]*sf[i][m]; /* f is negative gradient, thus the sign */
1150 /* Sum the gradient along the line across CPUs */
1152 gmx_sumd(1,&gpb,cr);
1155 fprintf(debug,"CGE: EpotA %f EpotB %f EpotC %f gpb %f\n",
1156 s_a->epot,s_b->epot,s_c->epot,gpb);
1158 epot_repl = s_b->epot;
1160 /* Keep one of the intervals based on the value of the derivative at the new point */
1162 /* Replace c endpoint with b */
1163 swap_em_state(s_b,s_c);
1167 /* Replace a endpoint with b */
1168 swap_em_state(s_b,s_a);
1174 * Stop search as soon as we find a value smaller than the endpoints.
1175 * Never run more than 20 steps, no matter what.
1178 } while ((epot_repl > s_a->epot || epot_repl > s_c->epot) &&
1181 if (fabs(epot_repl - s_min->epot) < fabs(s_min->epot)*GMX_REAL_EPS ||
1183 /* OK. We couldn't find a significantly lower energy.
1184 * If beta==0 this was steepest descent, and then we give up.
1185 * If not, set beta=0 and restart with steepest descent before quitting.
1192 /* Reset memory before giving up */
1198 /* Select min energy state of A & C, put the best in B.
1200 if (s_c->epot < s_a->epot) {
1202 fprintf(debug,"CGE: C (%f) is lower than A (%f), moving C to B\n",
1203 s_c->epot,s_a->epot);
1204 swap_em_state(s_b,s_c);
1209 fprintf(debug,"CGE: A (%f) is lower than C (%f), moving A to B\n",
1210 s_a->epot,s_c->epot);
1211 swap_em_state(s_b,s_a);
1218 fprintf(debug,"CGE: Found a lower energy %f, moving C to B\n",
1220 swap_em_state(s_b,s_c);
1225 /* new search direction */
1226 /* beta = 0 means forget all memory and restart with steepest descents. */
1227 if (nstcg && ((step % nstcg)==0))
1230 /* s_min->fnorm cannot be zero, because then we would have converged
1234 /* Polak-Ribiere update.
1235 * Change to fnorm2/fnorm2_old for Fletcher-Reeves
1237 beta = pr_beta(cr,&inputrec->opts,mdatoms,top_global,s_min,s_b);
1239 /* Limit beta to prevent oscillations */
1240 if (fabs(beta) > 5.0)
1244 /* update positions */
1245 swap_em_state(s_min,s_b);
1248 /* Print it if necessary */
1251 fprintf(stderr,"\rStep %d, Epot=%12.6e, Fnorm=%9.3e, Fmax=%9.3e (atom %d)\n",
1252 step,s_min->epot,s_min->fnorm/sqrt(state_global->natoms),
1253 s_min->fmax,s_min->a_fmax+1);
1254 /* Store the new (lower) energies */
1255 upd_mdebin(mdebin,FALSE,FALSE,(double)step,
1256 mdatoms->tmass,enerd,&s_min->s,s_min->s.box,
1257 NULL,NULL,vir,pres,NULL,mu_tot,constr);
1258 do_log = do_per_step(step,inputrec->nstlog);
1259 do_ene = do_per_step(step,inputrec->nstenergy);
1261 print_ebin_header(fplog,step,step,s_min->s.lambda);
1262 print_ebin(outf->fp_ene,do_ene,FALSE,FALSE,
1263 do_log ? fplog : NULL,step,step,eprNORMAL,
1264 TRUE,mdebin,fcd,&(top_global->groups),&(inputrec->opts));
1267 /* Stop when the maximum force lies below tolerance.
1268 * If we have reached machine precision, converged is already set to true.
1270 converged = converged || (s_min->fmax < inputrec->em_tol);
1272 } /* End of the loop */
1275 step--; /* we never took that last step in this case */
1277 if (s_min->fmax > inputrec->em_tol)
1281 warn_step(stderr,inputrec->em_tol,step-1==number_steps,FALSE);
1282 warn_step(fplog ,inputrec->em_tol,step-1==number_steps,FALSE);
1288 /* If we printed energy and/or logfile last step (which was the last step)
1289 * we don't have to do it again, but otherwise print the final values.
1292 /* Write final value to log since we didn't do anything the last step */
1293 print_ebin_header(fplog,step,step,s_min->s.lambda);
1295 if (!do_ene || !do_log) {
1296 /* Write final energy file entries */
1297 print_ebin(outf->fp_ene,!do_ene,FALSE,FALSE,
1298 !do_log ? fplog : NULL,step,step,eprNORMAL,
1299 TRUE,mdebin,fcd,&(top_global->groups),&(inputrec->opts));
1303 /* Print some stuff... */
1305 fprintf(stderr,"\nwriting lowest energy coordinates.\n");
1308 * For accurate normal mode calculation it is imperative that we
1309 * store the last conformation into the full precision binary trajectory.
1311 * However, we should only do it if we did NOT already write this step
1312 * above (which we did if do_x or do_f was true).
1314 do_x = !do_per_step(step,inputrec->nstxout);
1315 do_f = (inputrec->nstfout > 0 && !do_per_step(step,inputrec->nstfout));
1317 write_em_traj(fplog,cr,outf,do_x,do_f,ftp2fn(efSTO,nfile,fnm),
1318 top_global,inputrec,step,
1319 s_min,state_global,f_global);
1321 fnormn = s_min->fnorm/sqrt(state_global->natoms);
1324 print_converged(stderr,CG,inputrec->em_tol,step,converged,number_steps,
1325 s_min->epot,s_min->fmax,s_min->a_fmax,fnormn);
1326 print_converged(fplog,CG,inputrec->em_tol,step,converged,number_steps,
1327 s_min->epot,s_min->fmax,s_min->a_fmax,fnormn);
1329 fprintf(fplog,"\nPerformed %d energy evaluations in total.\n",neval);
1332 finish_em(fplog,cr,outf,runtime,wcycle);
1334 /* To print the actual number of steps we needed somewhere */
1335 runtime->nsteps_done = step;
1338 } /* That's all folks */
1341 double do_lbfgs(FILE *fplog,t_commrec *cr,
1342 int nfile,const t_filenm fnm[],
1343 const output_env_t oenv, gmx_bool bVerbose,gmx_bool bCompact,
1345 gmx_vsite_t *vsite,gmx_constr_t constr,
1347 t_inputrec *inputrec,
1348 gmx_mtop_t *top_global,t_fcdata *fcd,
1351 t_nrnb *nrnb,gmx_wallcycle_t wcycle,
1354 int repl_ex_nst,int repl_ex_seed,
1355 gmx_membed_t membed,
1356 real cpt_period,real max_hours,
1357 const char *deviceOptions,
1358 unsigned long Flags,
1359 gmx_runtime_t *runtime)
1361 static const char *LBFGS="Low-Memory BFGS Minimizer";
1363 gmx_localtop_t *top;
1364 gmx_enerdata_t *enerd;
1366 gmx_global_stat_t gstat;
1369 int ncorr,nmaxcorr,point,cp,neval,nminstep;
1370 double stepsize,gpa,gpb,gpc,tmp,minstep;
1371 real *rho,*alpha,*ff,*xx,*p,*s,*lastx,*lastf,**dx,**dg;
1372 real *xa,*xb,*xc,*fa,*fb,*fc,*xtmp,*ftmp;
1373 real a,b,c,maxdelta,delta;
1374 real diag,Epot0,Epot,EpotA,EpotB,EpotC;
1375 real dgdx,dgdg,sq,yr,beta;
1377 gmx_bool converged,first;
1380 gmx_bool do_log,do_ene,do_x,do_f,foundlower,*frozen;
1382 int start,end,number_steps;
1384 int i,k,m,n,nfmax,gf,step;
1389 gmx_fatal(FARGS,"Cannot do parallel L-BFGS Minimization - yet.\n");
1391 n = 3*state->natoms;
1392 nmaxcorr = inputrec->nbfgscorr;
1394 /* Allocate memory */
1395 /* Use pointers to real so we dont have to loop over both atoms and
1396 * dimensions all the time...
1397 * x/f are allocated as rvec *, so make new x0/f0 pointers-to-real
1398 * that point to the same memory.
1412 snew(alpha,nmaxcorr);
1415 for(i=0;i<nmaxcorr;i++)
1419 for(i=0;i<nmaxcorr;i++)
1426 init_em(fplog,LBFGS,cr,inputrec,
1427 state,top_global,&ems,&top,&f,&f_global,
1428 nrnb,mu_tot,fr,&enerd,&graph,mdatoms,&gstat,vsite,constr,
1429 nfile,fnm,&outf,&mdebin);
1430 /* Do_lbfgs is not completely updated like do_steep and do_cg,
1431 * so we free some memory again.
1436 xx = (real *)state->x;
1439 start = mdatoms->start;
1440 end = mdatoms->homenr + start;
1442 /* Print to log file */
1443 print_em_start(fplog,cr,runtime,wcycle,LBFGS);
1445 do_log = do_ene = do_x = do_f = TRUE;
1447 /* Max number of steps */
1448 number_steps=inputrec->nsteps;
1450 /* Create a 3*natoms index to tell whether each degree of freedom is frozen */
1452 for(i=start; i<end; i++) {
1453 if (mdatoms->cFREEZE)
1454 gf = mdatoms->cFREEZE[i];
1455 for(m=0; m<DIM; m++)
1456 frozen[3*i+m]=inputrec->opts.nFreeze[gf][m];
1459 sp_header(stderr,LBFGS,inputrec->em_tol,number_steps);
1461 sp_header(fplog,LBFGS,inputrec->em_tol,number_steps);
1464 construct_vsites(fplog,vsite,state->x,nrnb,1,NULL,
1465 top->idef.iparams,top->idef.il,
1466 fr->ePBC,fr->bMolPBC,graph,cr,state->box);
1468 /* Call the force routine and some auxiliary (neighboursearching etc.) */
1469 /* do_force always puts the charge groups in the box and shifts again
1470 * We do not unshift, so molecules are always whole
1475 evaluate_energy(fplog,bVerbose,cr,
1476 state,top_global,&ems,top,
1477 inputrec,nrnb,wcycle,gstat,
1478 vsite,constr,fcd,graph,mdatoms,fr,
1479 mu_tot,enerd,vir,pres,-1,TRUE);
1483 /* Copy stuff to the energy bin for easy printing etc. */
1484 upd_mdebin(mdebin,FALSE,FALSE,(double)step,
1485 mdatoms->tmass,enerd,state,state->box,
1486 NULL,NULL,vir,pres,NULL,mu_tot,constr);
1488 print_ebin_header(fplog,step,step,state->lambda);
1489 print_ebin(outf->fp_ene,TRUE,FALSE,FALSE,fplog,step,step,eprNORMAL,
1490 TRUE,mdebin,fcd,&(top_global->groups),&(inputrec->opts));
1494 /* This is the starting energy */
1495 Epot = enerd->term[F_EPOT];
1501 /* Set the initial step.
1502 * since it will be multiplied by the non-normalized search direction
1503 * vector (force vector the first time), we scale it by the
1504 * norm of the force.
1508 fprintf(stderr,"Using %d BFGS correction steps.\n\n",nmaxcorr);
1509 fprintf(stderr," F-max = %12.5e on atom %d\n",fmax,nfmax+1);
1510 fprintf(stderr," F-Norm = %12.5e\n",fnorm/sqrt(state->natoms));
1511 fprintf(stderr,"\n");
1512 /* and copy to the log file too... */
1513 fprintf(fplog,"Using %d BFGS correction steps.\n\n",nmaxcorr);
1514 fprintf(fplog," F-max = %12.5e on atom %d\n",fmax,nfmax+1);
1515 fprintf(fplog," F-Norm = %12.5e\n",fnorm/sqrt(state->natoms));
1516 fprintf(fplog,"\n");
1522 dx[point][i] = ff[i]; /* Initial search direction */
1526 stepsize = 1.0/fnorm;
1529 /* Start the loop over BFGS steps.
1530 * Each successful step is counted, and we continue until
1531 * we either converge or reach the max number of steps.
1536 /* Set the gradient from the force */
1538 for(step=0; (number_steps<0 || (number_steps>=0 && step<=number_steps)) && !converged; step++) {
1540 /* Write coordinates if necessary */
1541 do_x = do_per_step(step,inputrec->nstxout);
1542 do_f = do_per_step(step,inputrec->nstfout);
1544 write_traj(fplog,cr,outf,MDOF_X | MDOF_F,
1545 top_global,step,(real)step,state,state,f,f,NULL,NULL);
1547 /* Do the linesearching in the direction dx[point][0..(n-1)] */
1549 /* pointer to current direction - point=0 first time here */
1552 /* calculate line gradient */
1553 for(gpa=0,i=0;i<n;i++)
1556 /* Calculate minimum allowed stepsize, before the average (norm)
1557 * relative change in coordinate is smaller than precision
1559 for(minstep=0,i=0;i<n;i++) {
1566 minstep = GMX_REAL_EPS/sqrt(minstep/n);
1568 if(stepsize<minstep) {
1573 /* Store old forces and coordinates */
1585 /* Take a step downhill.
1586 * In theory, we should minimize the function along this direction.
1587 * That is quite possible, but it turns out to take 5-10 function evaluations
1588 * for each line. However, we dont really need to find the exact minimum -
1589 * it is much better to start a new BFGS step in a modified direction as soon
1590 * as we are close to it. This will save a lot of energy evaluations.
1592 * In practice, we just try to take a single step.
1593 * If it worked (i.e. lowered the energy), we increase the stepsize but
1594 * the continue straight to the next BFGS step without trying to find any minimum.
1595 * If it didn't work (higher energy), there must be a minimum somewhere between
1596 * the old position and the new one.
1598 * Due to the finite numerical accuracy, it turns out that it is a good idea
1599 * to even accept a SMALL increase in energy, if the derivative is still downhill.
1600 * This leads to lower final energies in the tests I've done. / Erik
1605 c = a + stepsize; /* reference position along line is zero */
1607 /* Check stepsize first. We do not allow displacements
1608 * larger than emstep.
1618 if(maxdelta>inputrec->em_stepsize)
1620 } while(maxdelta>inputrec->em_stepsize);
1622 /* Take a trial step */
1624 xc[i] = lastx[i] + c*s[i];
1627 /* Calculate energy for the trial step */
1628 ems.s.x = (rvec *)xc;
1630 evaluate_energy(fplog,bVerbose,cr,
1631 state,top_global,&ems,top,
1632 inputrec,nrnb,wcycle,gstat,
1633 vsite,constr,fcd,graph,mdatoms,fr,
1634 mu_tot,enerd,vir,pres,step,FALSE);
1637 /* Calc derivative along line */
1638 for(gpc=0,i=0; i<n; i++) {
1639 gpc -= s[i]*fc[i]; /* f is negative gradient, thus the sign */
1641 /* Sum the gradient along the line across CPUs */
1643 gmx_sumd(1,&gpc,cr);
1645 /* This is the max amount of increase in energy we tolerate */
1646 tmp=sqrt(GMX_REAL_EPS)*fabs(EpotA);
1648 /* Accept the step if the energy is lower, or if it is not significantly higher
1649 * and the line derivative is still negative.
1651 if(EpotC<EpotA || (gpc<0 && EpotC<(EpotA+tmp))) {
1653 /* Great, we found a better energy. Increase step for next iteration
1654 * if we are still going down, decrease it otherwise
1657 stepsize *= 1.618034; /* The golden section */
1659 stepsize *= 0.618034; /* 1/golden section */
1661 /* New energy is the same or higher. We will have to do some work
1662 * to find a smaller value in the interval. Take smaller step next time!
1665 stepsize *= 0.618034;
1668 /* OK, if we didn't find a lower value we will have to locate one now - there must
1669 * be one in the interval [a=0,c].
1670 * The same thing is valid here, though: Don't spend dozens of iterations to find
1671 * the line minimum. We try to interpolate based on the derivative at the endpoints,
1672 * and only continue until we find a lower value. In most cases this means 1-2 iterations.
1674 * I also have a safeguard for potentially really patological functions so we never
1675 * take more than 20 steps before we give up ...
1677 * If we already found a lower value we just skip this step and continue to the update.
1684 /* Select a new trial point.
1685 * If the derivatives at points a & c have different sign we interpolate to zero,
1686 * otherwise just do a bisection.
1690 b = a + gpa*(a-c)/(gpc-gpa);
1694 /* safeguard if interpolation close to machine accuracy causes errors:
1695 * never go outside the interval
1700 /* Take a trial step */
1702 xb[i] = lastx[i] + b*s[i];
1705 /* Calculate energy for the trial step */
1706 ems.s.x = (rvec *)xb;
1708 evaluate_energy(fplog,bVerbose,cr,
1709 state,top_global,&ems,top,
1710 inputrec,nrnb,wcycle,gstat,
1711 vsite,constr,fcd,graph,mdatoms,fr,
1712 mu_tot,enerd,vir,pres,step,FALSE);
1717 for(gpb=0,i=0; i<n; i++)
1718 gpb -= s[i]*fb[i]; /* f is negative gradient, thus the sign */
1720 /* Sum the gradient along the line across CPUs */
1722 gmx_sumd(1,&gpb,cr);
1724 /* Keep one of the intervals based on the value of the derivative at the new point */
1726 /* Replace c endpoint with b */
1730 /* swap coord pointers b/c */
1738 /* Replace a endpoint with b */
1742 /* swap coord pointers a/b */
1752 * Stop search as soon as we find a value smaller than the endpoints,
1753 * or if the tolerance is below machine precision.
1754 * Never run more than 20 steps, no matter what.
1757 } while((EpotB>EpotA || EpotB>EpotC) && (nminstep<20));
1759 if(fabs(EpotB-Epot0)<GMX_REAL_EPS || nminstep>=20) {
1760 /* OK. We couldn't find a significantly lower energy.
1761 * If ncorr==0 this was steepest descent, and then we give up.
1762 * If not, reset memory to restart as steepest descent before quitting.
1771 /* Search in gradient direction */
1774 /* Reset stepsize */
1775 stepsize = 1.0/fnorm;
1780 /* Select min energy state of A & C, put the best in xx/ff/Epot
1811 /* Update the memory information, and calculate a new
1812 * approximation of the inverse hessian
1815 /* Have new data in Epot, xx, ff */
1820 dg[point][i]=lastf[i]-ff[i];
1821 dx[point][i]*=stepsize;
1827 dgdg+=dg[point][i]*dg[point][i];
1828 dgdx+=dg[point][i]*dx[point][i];
1833 rho[point]=1.0/dgdx;
1845 /* Recursive update. First go back over the memory points */
1846 for(k=0;k<ncorr;k++) {
1855 alpha[cp]=rho[cp]*sq;
1858 p[i] -= alpha[cp]*dg[cp][i];
1864 /* And then go forward again */
1865 for(k=0;k<ncorr;k++) {
1868 yr += p[i]*dg[cp][i];
1871 beta = alpha[cp]-beta;
1874 p[i] += beta*dx[cp][i];
1883 dx[point][i] = p[i];
1889 /* Test whether the convergence criterion is met */
1890 get_f_norm_max(cr,&(inputrec->opts),mdatoms,f,&fnorm,&fmax,&nfmax);
1892 /* Print it if necessary */
1895 fprintf(stderr,"\rStep %d, Epot=%12.6e, Fnorm=%9.3e, Fmax=%9.3e (atom %d)\n",
1896 step,Epot,fnorm/sqrt(state->natoms),fmax,nfmax+1);
1897 /* Store the new (lower) energies */
1898 upd_mdebin(mdebin,FALSE,FALSE,(double)step,
1899 mdatoms->tmass,enerd,state,state->box,
1900 NULL,NULL,vir,pres,NULL,mu_tot,constr);
1901 do_log = do_per_step(step,inputrec->nstlog);
1902 do_ene = do_per_step(step,inputrec->nstenergy);
1904 print_ebin_header(fplog,step,step,state->lambda);
1905 print_ebin(outf->fp_ene,do_ene,FALSE,FALSE,
1906 do_log ? fplog : NULL,step,step,eprNORMAL,
1907 TRUE,mdebin,fcd,&(top_global->groups),&(inputrec->opts));
1910 /* Stop when the maximum force lies below tolerance.
1911 * If we have reached machine precision, converged is already set to true.
1914 converged = converged || (fmax < inputrec->em_tol);
1916 } /* End of the loop */
1919 step--; /* we never took that last step in this case */
1921 if(fmax>inputrec->em_tol)
1925 warn_step(stderr,inputrec->em_tol,step-1==number_steps,FALSE);
1926 warn_step(fplog ,inputrec->em_tol,step-1==number_steps,FALSE);
1931 /* If we printed energy and/or logfile last step (which was the last step)
1932 * we don't have to do it again, but otherwise print the final values.
1934 if(!do_log) /* Write final value to log since we didn't do anythin last step */
1935 print_ebin_header(fplog,step,step,state->lambda);
1936 if(!do_ene || !do_log) /* Write final energy file entries */
1937 print_ebin(outf->fp_ene,!do_ene,FALSE,FALSE,
1938 !do_log ? fplog : NULL,step,step,eprNORMAL,
1939 TRUE,mdebin,fcd,&(top_global->groups),&(inputrec->opts));
1941 /* Print some stuff... */
1943 fprintf(stderr,"\nwriting lowest energy coordinates.\n");
1946 * For accurate normal mode calculation it is imperative that we
1947 * store the last conformation into the full precision binary trajectory.
1949 * However, we should only do it if we did NOT already write this step
1950 * above (which we did if do_x or do_f was true).
1952 do_x = !do_per_step(step,inputrec->nstxout);
1953 do_f = !do_per_step(step,inputrec->nstfout);
1954 write_em_traj(fplog,cr,outf,do_x,do_f,ftp2fn(efSTO,nfile,fnm),
1955 top_global,inputrec,step,
1959 print_converged(stderr,LBFGS,inputrec->em_tol,step,converged,
1960 number_steps,Epot,fmax,nfmax,fnorm/sqrt(state->natoms));
1961 print_converged(fplog,LBFGS,inputrec->em_tol,step,converged,
1962 number_steps,Epot,fmax,nfmax,fnorm/sqrt(state->natoms));
1964 fprintf(fplog,"\nPerformed %d energy evaluations in total.\n",neval);
1967 finish_em(fplog,cr,outf,runtime,wcycle);
1969 /* To print the actual number of steps we needed somewhere */
1970 runtime->nsteps_done = step;
1973 } /* That's all folks */
1976 double do_steep(FILE *fplog,t_commrec *cr,
1977 int nfile, const t_filenm fnm[],
1978 const output_env_t oenv, gmx_bool bVerbose,gmx_bool bCompact,
1980 gmx_vsite_t *vsite,gmx_constr_t constr,
1982 t_inputrec *inputrec,
1983 gmx_mtop_t *top_global,t_fcdata *fcd,
1984 t_state *state_global,
1986 t_nrnb *nrnb,gmx_wallcycle_t wcycle,
1989 int repl_ex_nst,int repl_ex_seed,
1990 gmx_membed_t membed,
1991 real cpt_period,real max_hours,
1992 const char *deviceOptions,
1993 unsigned long Flags,
1994 gmx_runtime_t *runtime)
1996 const char *SD="Steepest Descents";
1997 em_state_t *s_min,*s_try;
1999 gmx_localtop_t *top;
2000 gmx_enerdata_t *enerd;
2002 gmx_global_stat_t gstat;
2004 real stepsize,constepsize;
2005 real ustep,dvdlambda,fnormn;
2008 gmx_bool bDone,bAbort,do_x,do_f;
2013 int steps_accepted=0;
2017 s_min = init_em_state();
2018 s_try = init_em_state();
2020 /* Init em and store the local state in s_try */
2021 init_em(fplog,SD,cr,inputrec,
2022 state_global,top_global,s_try,&top,&f,&f_global,
2023 nrnb,mu_tot,fr,&enerd,&graph,mdatoms,&gstat,vsite,constr,
2024 nfile,fnm,&outf,&mdebin);
2026 /* Print to log file */
2027 print_em_start(fplog,cr,runtime,wcycle,SD);
2029 /* Set variables for stepsize (in nm). This is the largest
2030 * step that we are going to make in any direction.
2032 ustep = inputrec->em_stepsize;
2035 /* Max number of steps */
2036 nsteps = inputrec->nsteps;
2039 /* Print to the screen */
2040 sp_header(stderr,SD,inputrec->em_tol,nsteps);
2042 sp_header(fplog,SD,inputrec->em_tol,nsteps);
2044 /**** HERE STARTS THE LOOP ****
2045 * count is the counter for the number of steps
2046 * bDone will be TRUE when the minimization has converged
2047 * bAbort will be TRUE when nsteps steps have been performed or when
2048 * the stepsize becomes smaller than is reasonable for machine precision
2053 while( !bDone && !bAbort ) {
2054 bAbort = (nsteps >= 0) && (count == nsteps);
2056 /* set new coordinates, except for first step */
2058 do_em_step(cr,inputrec,mdatoms,s_min,stepsize,s_min->f,s_try,
2059 constr,top,nrnb,wcycle,count);
2062 evaluate_energy(fplog,bVerbose,cr,
2063 state_global,top_global,s_try,top,
2064 inputrec,nrnb,wcycle,gstat,
2065 vsite,constr,fcd,graph,mdatoms,fr,
2066 mu_tot,enerd,vir,pres,count,count==0);
2069 print_ebin_header(fplog,count,count,s_try->s.lambda);
2072 s_min->epot = s_try->epot + 1;
2074 /* Print it if necessary */
2077 fprintf(stderr,"Step=%5d, Dmax= %6.1e nm, Epot= %12.5e Fmax= %11.5e, atom= %d%c",
2078 count,ustep,s_try->epot,s_try->fmax,s_try->a_fmax+1,
2079 (s_try->epot < s_min->epot) ? '\n' : '\r');
2082 if (s_try->epot < s_min->epot) {
2083 /* Store the new (lower) energies */
2084 upd_mdebin(mdebin,FALSE,FALSE,(double)count,
2085 mdatoms->tmass,enerd,&s_try->s,s_try->s.box,
2086 NULL,NULL,vir,pres,NULL,mu_tot,constr);
2087 print_ebin(outf->fp_ene,TRUE,
2088 do_per_step(steps_accepted,inputrec->nstdisreout),
2089 do_per_step(steps_accepted,inputrec->nstorireout),
2090 fplog,count,count,eprNORMAL,TRUE,
2091 mdebin,fcd,&(top_global->groups),&(inputrec->opts));
2096 /* Now if the new energy is smaller than the previous...
2097 * or if this is the first step!
2098 * or if we did random steps!
2101 if ( (count==0) || (s_try->epot < s_min->epot) ) {
2104 /* Test whether the convergence criterion is met... */
2105 bDone = (s_try->fmax < inputrec->em_tol);
2107 /* Copy the arrays for force, positions and energy */
2108 /* The 'Min' array always holds the coords and forces of the minimal
2110 swap_em_state(s_min,s_try);
2114 /* Write to trn, if necessary */
2115 do_x = do_per_step(steps_accepted,inputrec->nstxout);
2116 do_f = do_per_step(steps_accepted,inputrec->nstfout);
2117 write_em_traj(fplog,cr,outf,do_x,do_f,NULL,
2118 top_global,inputrec,count,
2119 s_min,state_global,f_global);
2122 /* If energy is not smaller make the step smaller... */
2125 if (DOMAINDECOMP(cr) && s_min->s.ddp_count != cr->dd->ddp_count) {
2126 /* Reload the old state */
2127 em_dd_partition_system(fplog,count,cr,top_global,inputrec,
2128 s_min,top,mdatoms,fr,vsite,constr,
2133 /* Determine new step */
2134 stepsize = ustep/s_min->fmax;
2136 /* Check if stepsize is too small, with 1 nm as a characteristic length */
2138 if (count == nsteps || ustep < 1e-12)
2140 if (count == nsteps || ustep < 1e-6)
2145 warn_step(stderr,inputrec->em_tol,count==nsteps,constr!=NULL);
2146 warn_step(fplog ,inputrec->em_tol,count==nsteps,constr!=NULL);
2152 } /* End of the loop */
2154 /* Print some shit... */
2156 fprintf(stderr,"\nwriting lowest energy coordinates.\n");
2157 write_em_traj(fplog,cr,outf,TRUE,inputrec->nstfout,ftp2fn(efSTO,nfile,fnm),
2158 top_global,inputrec,count,
2159 s_min,state_global,f_global);
2161 fnormn = s_min->fnorm/sqrt(state_global->natoms);
2164 print_converged(stderr,SD,inputrec->em_tol,count,bDone,nsteps,
2165 s_min->epot,s_min->fmax,s_min->a_fmax,fnormn);
2166 print_converged(fplog,SD,inputrec->em_tol,count,bDone,nsteps,
2167 s_min->epot,s_min->fmax,s_min->a_fmax,fnormn);
2170 finish_em(fplog,cr,outf,runtime,wcycle);
2172 /* To print the actual number of steps we needed somewhere */
2173 inputrec->nsteps=count;
2175 runtime->nsteps_done = count;
2178 } /* That's all folks */
2181 double do_nm(FILE *fplog,t_commrec *cr,
2182 int nfile,const t_filenm fnm[],
2183 const output_env_t oenv, gmx_bool bVerbose,gmx_bool bCompact,
2185 gmx_vsite_t *vsite,gmx_constr_t constr,
2187 t_inputrec *inputrec,
2188 gmx_mtop_t *top_global,t_fcdata *fcd,
2189 t_state *state_global,
2191 t_nrnb *nrnb,gmx_wallcycle_t wcycle,
2194 int repl_ex_nst,int repl_ex_seed,
2195 gmx_membed_t membed,
2196 real cpt_period,real max_hours,
2197 const char *deviceOptions,
2198 unsigned long Flags,
2199 gmx_runtime_t *runtime)
2201 const char *NM = "Normal Mode Analysis";
2206 gmx_localtop_t *top;
2207 gmx_enerdata_t *enerd;
2209 gmx_global_stat_t gstat;
2216 gmx_bool bSparse; /* use sparse matrix storage format */
2218 gmx_sparsematrix_t * sparse_matrix = NULL;
2219 real * full_matrix = NULL;
2220 em_state_t * state_work;
2222 /* added with respect to mdrun */
2224 real der_range=10.0*sqrt(GMX_REAL_EPS);
2230 gmx_fatal(FARGS,"Constraints present with Normal Mode Analysis, this combination is not supported");
2233 state_work = init_em_state();
2235 /* Init em and store the local state in state_minimum */
2236 init_em(fplog,NM,cr,inputrec,
2237 state_global,top_global,state_work,&top,
2239 nrnb,mu_tot,fr,&enerd,&graph,mdatoms,&gstat,vsite,constr,
2240 nfile,fnm,&outf,NULL);
2242 natoms = top_global->natoms;
2250 "NOTE: This version of Gromacs has been compiled in single precision,\n"
2251 " which MIGHT not be accurate enough for normal mode analysis.\n"
2252 " Gromacs now uses sparse matrix storage, so the memory requirements\n"
2253 " are fairly modest even if you recompile in double precision.\n\n");
2257 /* Check if we can/should use sparse storage format.
2259 * Sparse format is only useful when the Hessian itself is sparse, which it
2260 * will be when we use a cutoff.
2261 * For small systems (n<1000) it is easier to always use full matrix format, though.
2263 if(EEL_FULL(fr->eeltype) || fr->rlist==0.0)
2265 fprintf(stderr,"Non-cutoff electrostatics used, forcing full Hessian format.\n");
2268 else if(top_global->natoms < 1000)
2270 fprintf(stderr,"Small system size (N=%d), using full Hessian format.\n",top_global->natoms);
2275 fprintf(stderr,"Using compressed symmetric sparse Hessian format.\n");
2279 sz = DIM*top_global->natoms;
2281 fprintf(stderr,"Allocating Hessian memory...\n\n");
2285 sparse_matrix=gmx_sparsematrix_init(sz);
2286 sparse_matrix->compressed_symmetric = TRUE;
2290 snew(full_matrix,sz*sz);
2293 /* Initial values */
2294 t = inputrec->init_t;
2295 lambda = inputrec->init_lambda;
2301 /* Write start time and temperature */
2302 print_em_start(fplog,cr,runtime,wcycle,NM);
2304 /* fudge nr of steps to nr of atoms */
2305 inputrec->nsteps = natoms*2;
2309 fprintf(stderr,"starting normal mode calculation '%s'\n%d steps.\n\n",
2310 *(top_global->name),(int)inputrec->nsteps);
2313 nnodes = cr->nnodes;
2315 /* Make evaluate_energy do a single node force calculation */
2317 evaluate_energy(fplog,bVerbose,cr,
2318 state_global,top_global,state_work,top,
2319 inputrec,nrnb,wcycle,gstat,
2320 vsite,constr,fcd,graph,mdatoms,fr,
2321 mu_tot,enerd,vir,pres,-1,TRUE);
2322 cr->nnodes = nnodes;
2324 /* if forces are not small, warn user */
2325 get_state_f_norm_max(cr,&(inputrec->opts),mdatoms,state_work);
2329 fprintf(stderr,"Maximum force:%12.5e\n",state_work->fmax);
2330 if (state_work->fmax > 1.0e-3)
2332 fprintf(stderr,"Maximum force probably not small enough to");
2333 fprintf(stderr," ensure that you are in an \nenergy well. ");
2334 fprintf(stderr,"Be aware that negative eigenvalues may occur");
2335 fprintf(stderr," when the\nresulting matrix is diagonalized.\n");
2339 /***********************************************************
2341 * Loop over all pairs in matrix
2343 * do_force called twice. Once with positive and
2344 * once with negative displacement
2346 ************************************************************/
2348 /* Steps are divided one by one over the nodes */
2349 for(atom=cr->nodeid; atom<natoms; atom+=nnodes)
2352 for (d=0; d<DIM; d++)
2354 x_min = state_work->s.x[atom][d];
2356 state_work->s.x[atom][d] = x_min - der_range;
2358 /* Make evaluate_energy do a single node force calculation */
2360 evaluate_energy(fplog,bVerbose,cr,
2361 state_global,top_global,state_work,top,
2362 inputrec,nrnb,wcycle,gstat,
2363 vsite,constr,fcd,graph,mdatoms,fr,
2364 mu_tot,enerd,vir,pres,atom*2,FALSE);
2366 for(i=0; i<natoms; i++)
2368 copy_rvec(state_work->f[i], fneg[i]);
2371 state_work->s.x[atom][d] = x_min + der_range;
2373 evaluate_energy(fplog,bVerbose,cr,
2374 state_global,top_global,state_work,top,
2375 inputrec,nrnb,wcycle,gstat,
2376 vsite,constr,fcd,graph,mdatoms,fr,
2377 mu_tot,enerd,vir,pres,atom*2+1,FALSE);
2378 cr->nnodes = nnodes;
2380 /* x is restored to original */
2381 state_work->s.x[atom][d] = x_min;
2383 for(j=0; j<natoms; j++)
2385 for (k=0; (k<DIM); k++)
2388 -(state_work->f[j][k] - fneg[j][k])/(2*der_range);
2396 #define mpi_type MPI_DOUBLE
2398 #define mpi_type MPI_FLOAT
2400 MPI_Send(dfdx[0],natoms*DIM,mpi_type,MASTERNODE(cr),cr->nodeid,
2401 cr->mpi_comm_mygroup);
2406 for(node=0; (node<nnodes && atom+node<natoms); node++)
2412 MPI_Recv(dfdx[0],natoms*DIM,mpi_type,node,node,
2413 cr->mpi_comm_mygroup,&stat);
2418 row = (atom + node)*DIM + d;
2420 for(j=0; j<natoms; j++)
2422 for(k=0; k<DIM; k++)
2428 if (col >= row && dfdx[j][k] != 0.0)
2430 gmx_sparsematrix_increment_value(sparse_matrix,
2431 row,col,dfdx[j][k]);
2436 full_matrix[row*sz+col] = dfdx[j][k];
2443 if (bVerbose && fplog)
2448 /* write progress */
2449 if (MASTER(cr) && bVerbose)
2451 fprintf(stderr,"\rFinished step %d out of %d",
2452 min(atom+nnodes,natoms),natoms);
2459 fprintf(stderr,"\n\nWriting Hessian...\n");
2460 gmx_mtxio_write(ftp2fn(efMTX,nfile,fnm),sz,sz,full_matrix,sparse_matrix);
2463 finish_em(fplog,cr,outf,runtime,wcycle);
2465 runtime->nsteps_done = natoms*2;