void do_force(FILE *fplog,t_commrec *cr, t_inputrec *inputrec, int step,t_nrnb *nrnb,gmx_wallcycle_t wcycle, gmx_localtop_t *top, gmx_groups_t *groups, matrix box,rvec x[],history_t *hist, rvec f[],rvec buf[], tensor vir_force, t_mdatoms *mdatoms, gmx_enerdata_t *enerd,t_fcdata *fcd, real lambda,t_graph *graph, t_forcerec *fr,gmx_vsite_t *vsite,rvec mu_tot, real t,FILE *field,gmx_edsam_t ed, int flags) { static rvec box_size; int cg0,cg1,i,j; int start,homenr; static double mu[2*DIM]; rvec mu_tot_AB[2]; bool bSepDVDL,bStateChanged,bNS,bFillGrid,bCalcCGCM,bBS,bDoForces; matrix boxs; real e,v,dvdl; t_pbc pbc; float cycles_ppdpme,cycles_pme,cycles_force; start = mdatoms->start; homenr = mdatoms->homenr; bSepDVDL = (fr->bSepDVDL && do_per_step(step,inputrec->nstlog)); clear_mat(vir_force); if (PARTDECOMP(cr)) { pd_cg_range(cr,&cg0,&cg1); } else { cg0 = 0; if (DOMAINDECOMP(cr)) cg1 = cr->dd->ncg_tot; else cg1 = top->cgs.nr; if (fr->n_tpi > 0) cg1--; } bStateChanged = (flags & GMX_FORCE_STATECHANGED); bNS = (flags & GMX_FORCE_NS); bFillGrid = (bNS && bStateChanged); bCalcCGCM = (bFillGrid && !DOMAINDECOMP(cr)); bDoForces = (flags & GMX_FORCE_FORCES); if (bStateChanged) { update_forcerec(fplog,fr,box); /* Calculate total (local) dipole moment in a temporary common array. * This makes it possible to sum them over nodes faster. */ calc_mu(start,homenr, x,mdatoms->chargeA,mdatoms->chargeB,mdatoms->nChargePerturbed, mu,mu+DIM); } if (fr->ePBC != epbcNONE) { /* Compute shift vectors every step, * because of pressure coupling or box deformation! */ if (DYNAMIC_BOX(*inputrec) && bStateChanged) calc_shifts(box,fr->shift_vec); if (bCalcCGCM) { put_charge_groups_in_box(fplog,cg0,cg1,fr->ePBC,box, &(top->cgs),x,fr->cg_cm); inc_nrnb(nrnb,eNR_CGCM,homenr); inc_nrnb(nrnb,eNR_RESETX,cg1-cg0); } else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph) { unshift_self(graph,box,x); } } else if (bCalcCGCM) { calc_cgcm(fplog,cg0,cg1,&(top->cgs),x,fr->cg_cm); inc_nrnb(nrnb,eNR_CGCM,homenr); } if (bCalcCGCM) { if (PAR(cr)) { move_cgcm(fplog,cr,fr->cg_cm); } if (gmx_debug_at) pr_rvecs(debug,0,"cgcm",fr->cg_cm,top->cgs.nr); } #ifdef GMX_MPI if (!(cr->duty & DUTY_PME)) { /* Send particle coordinates to the pme nodes. * Since this is only implemented for domain decomposition * and domain decomposition does not use the graph, * we do not need to worry about shifting. */ wallcycle_start(wcycle,ewcPP_PMESENDX); GMX_MPE_LOG(ev_send_coordinates_start); bBS = (inputrec->nwall == 2); if (bBS) { copy_mat(box,boxs); svmul(inputrec->wall_ewald_zfac,boxs[ZZ],boxs[ZZ]); } gmx_pme_send_x(cr,bBS ? boxs : box,x,mdatoms->nChargePerturbed,lambda); GMX_MPE_LOG(ev_send_coordinates_finish); wallcycle_stop(wcycle,ewcPP_PMESENDX); } #endif /* GMX_MPI */ /* Communicate coordinates and sum dipole if necessary */ if (PAR(cr)) { wallcycle_start(wcycle,ewcMOVEX); if (DOMAINDECOMP(cr)) { dd_move_x(cr->dd,box,x,buf); } else { move_x(fplog,cr,GMX_LEFT,GMX_RIGHT,x,nrnb); } /* When we don't need the total dipole we sum it in global_stat */ if (NEED_MUTOT(*inputrec)) gmx_sumd(2*DIM,mu,cr); wallcycle_stop(wcycle,ewcMOVEX); } for(i=0; i<2; i++) for(j=0;j<DIM;j++) mu_tot_AB[i][j] = mu[i*DIM + j]; if (fr->efep == efepNO) copy_rvec(mu_tot_AB[0],mu_tot); else for(j=0; j<DIM; j++) mu_tot[j] = (1.0 - lambda)*mu_tot_AB[0][j] + lambda*mu_tot_AB[1][j]; /* Reset energies */ reset_energies(&(inputrec->opts),fr,bNS,enerd,MASTER(cr)); if (bNS) { wallcycle_start(wcycle,ewcNS); if (graph && bStateChanged) /* Calculate intramolecular shift vectors to make molecules whole */ mk_mshift(fplog,graph,fr->ePBC,box,x); /* Reset long range forces if necessary */ if (fr->bTwinRange) { clear_rvecs(fr->f_twin_n,fr->f_twin); clear_rvecs(SHIFTS,fr->fshift_twin); } /* Do the actual neighbour searching and if twin range electrostatics * also do the calculation of long range forces and energies. */ dvdl = 0; ns(fplog,fr,x,f,box,groups,&(inputrec->opts),top,mdatoms, cr,nrnb,step,lambda,&dvdl,&enerd->grpp,bFillGrid,bDoForces); if (bSepDVDL) fprintf(fplog,sepdvdlformat,"LR non-bonded",0,dvdl); enerd->dvdl_lr = dvdl; enerd->term[F_DVDL] += dvdl; wallcycle_stop(wcycle,ewcNS); } if (DOMAINDECOMP(cr)) { if (!(cr->duty & DUTY_PME)) { wallcycle_start(wcycle,ewcPPDURINGPME); dd_force_flop_start(cr->dd,nrnb); } } /* Start the force cycle counter. * This counter is stopped in do_forcelow_level. * No parallel communication should occur while this counter is running, * since that will interfere with the dynamic load balancing. */ wallcycle_start(wcycle,ewcFORCE); if (bDoForces) { /* Reset PME/Ewald forces if necessary */ if (fr->bF_NoVirSum) { GMX_BARRIER(cr->mpi_comm_mygroup); if (fr->bDomDec) clear_rvecs(fr->f_novirsum_n,fr->f_novirsum); else clear_rvecs(homenr,fr->f_novirsum+start); GMX_BARRIER(cr->mpi_comm_mygroup); } /* Copy long range forces into normal buffers */ if (fr->bTwinRange) { for(i=0; i<fr->f_twin_n; i++) copy_rvec(fr->f_twin[i],f[i]); for(i=0; i<SHIFTS; i++) copy_rvec(fr->fshift_twin[i],fr->fshift[i]); } else { if (DOMAINDECOMP(cr)) clear_rvecs(cr->dd->nat_tot,f); else clear_rvecs(mdatoms->nr,f); clear_rvecs(SHIFTS,fr->fshift); } clear_rvec(fr->vir_diag_posres); GMX_BARRIER(cr->mpi_comm_mygroup); } if (inputrec->ePull == epullCONSTRAINT) clear_pull_forces(inputrec->pull); /* update QMMMrec, if necessary */ if(fr->bQMMM) update_QMMMrec(cr,fr,x,mdatoms,box,top); if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0) { /* Position restraints always require full pbc */ set_pbc(&pbc,inputrec->ePBC,box); v = posres(top->idef.il[F_POSRES].nr,top->idef.il[F_POSRES].iatoms, top->idef.iparams_posres, (const rvec*)x,fr->f_novirsum,fr->vir_diag_posres, inputrec->ePBC==epbcNONE ? NULL : &pbc,lambda,&dvdl, fr->rc_scaling,fr->ePBC,fr->posres_com,fr->posres_comB); if (bSepDVDL) { fprintf(fplog,sepdvdlformat, interaction_function[F_POSRES].longname,v,dvdl); } enerd->term[F_POSRES] += v; enerd->term[F_DVDL] += dvdl; inc_nrnb(nrnb,eNR_POSRES,top->idef.il[F_POSRES].nr/2); } /* Compute the bonded and non-bonded forces */ do_force_lowlevel(fplog,step,fr,inputrec,&(top->idef), cr,nrnb,wcycle,mdatoms,&(inputrec->opts), x,hist,f,enerd,fcd,box,lambda,graph,&(top->excls),mu_tot_AB, flags,&cycles_force); GMX_BARRIER(cr->mpi_comm_mygroup); if (ed) { do_flood(fplog,cr,x,f,ed,box,step); } if (DOMAINDECOMP(cr)) { dd_force_flop_stop(cr->dd,nrnb); if (wcycle) dd_cycles_add(cr->dd,cycles_force,ddCyclF); } if (bDoForces) { /* Compute forces due to electric field */ calc_f_el(MASTER(cr) ? field : NULL, start,homenr,mdatoms->chargeA,x,f,inputrec->ex,inputrec->et,t); /* When using PME/Ewald we compute the long range virial there. * otherwise we do it based on long range forces from twin range * cut-off based calculation (or not at all). */ /* Communicate the forces */ if (PAR(cr)) { wallcycle_start(wcycle,ewcMOVEF); if (DOMAINDECOMP(cr)) { dd_move_f(cr->dd,f,buf,fr->fshift); /* Position restraint do not introduce inter-cg forces */ if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl) dd_move_f(cr->dd,fr->f_novirsum,buf,NULL); } else { move_f(fplog,cr,GMX_LEFT,GMX_RIGHT,f,buf,nrnb); } wallcycle_stop(wcycle,ewcMOVEF); } } if (bDoForces) { if (vsite) { wallcycle_start(wcycle,ewcVSITESPREAD); spread_vsite_f(fplog,vsite,x,f,fr->fshift,nrnb, &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr); wallcycle_stop(wcycle,ewcVSITESPREAD); } /* Calculation of the virial must be done after vsites! */ calc_virial(fplog,mdatoms->start,mdatoms->homenr,x,f, vir_force,graph,box,nrnb,fr,inputrec->ePBC); } if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F) { /* Calculate the center of mass forces, this requires communication, * which is why pull_potential is called close to other communication. * The virial contribution is calculated directly, * which is why we call pull_potential after calc_virial. */ set_pbc(&pbc,inputrec->ePBC,box); dvdl = 0; enerd->term[F_COM_PULL] = pull_potential(inputrec->ePull,inputrec->pull,mdatoms,&pbc, cr,t,lambda,x,f,vir_force,&dvdl); if (bSepDVDL) fprintf(fplog,sepdvdlformat,"Com pull",enerd->term[F_COM_PULL],dvdl); enerd->term[F_DVDL] += dvdl; } if (!(cr->duty & DUTY_PME)) { cycles_ppdpme = wallcycle_stop(wcycle,ewcPPDURINGPME); dd_cycles_add(cr->dd,cycles_ppdpme,ddCyclPPduringPME); } #ifdef GMX_MPI if (PAR(cr) && !(cr->duty & DUTY_PME)) { /* In case of node-splitting, the PP nodes receive the long-range * forces, virial and energy from the PME nodes here. */ wallcycle_start(wcycle,ewcPP_PMEWAITRECVF); dvdl = 0; gmx_pme_receive_f(cr,fr->f_novirsum,fr->vir_el_recip,&e,&dvdl, &cycles_pme); if (bSepDVDL) fprintf(fplog,sepdvdlformat,"PME mesh",e,dvdl); enerd->term[F_COUL_RECIP] += e; enerd->term[F_DVDL] += dvdl; if (wcycle) dd_cycles_add(cr->dd,cycles_pme,ddCyclPME); wallcycle_stop(wcycle,ewcPP_PMEWAITRECVF); } #endif if (bDoForces && fr->bF_NoVirSum) { if (vsite) { /* Spread the mesh force on virtual sites to the other particles... * This is parallellized. MPI communication is performed * if the constructing atoms aren't local. */ wallcycle_start(wcycle,ewcVSITESPREAD); spread_vsite_f(fplog,vsite,x,fr->f_novirsum,NULL,nrnb, &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr); wallcycle_stop(wcycle,ewcVSITESPREAD); } /* Now add the forces, this is local */ if (fr->bDomDec) { sum_forces(0,fr->f_novirsum_n,f,fr->f_novirsum); } else { sum_forces(start,start+homenr,f,fr->f_novirsum); } if (EEL_FULL(fr->eeltype)) { /* Add the mesh contribution to the virial */ m_add(vir_force,fr->vir_el_recip,vir_force); } if (debug) pr_rvecs(debug,0,"vir_force",vir_force,DIM); } /* Sum the potential energy terms from group contributions */ sum_epot(&(inputrec->opts),enerd); if (fr->print_force >= 0 && bDoForces) print_large_forces(stderr,mdatoms,cr,step,fr->print_force,x,f); }
double do_tpi(FILE *fplog, t_commrec *cr, int nfile, const t_filenm fnm[], const output_env_t oenv, gmx_bool bVerbose, gmx_bool gmx_unused bCompact, int gmx_unused nstglobalcomm, gmx_vsite_t gmx_unused *vsite, gmx_constr_t gmx_unused constr, int gmx_unused stepout, t_inputrec *inputrec, gmx_mtop_t *top_global, t_fcdata *fcd, t_state *state, t_mdatoms *mdatoms, t_nrnb *nrnb, gmx_wallcycle_t wcycle, gmx_edsam_t gmx_unused ed, t_forcerec *fr, int gmx_unused repl_ex_nst, int gmx_unused repl_ex_nex, int gmx_unused repl_ex_seed, gmx_membed_t gmx_unused membed, real gmx_unused cpt_period, real gmx_unused max_hours, const char gmx_unused *deviceOptions, int gmx_unused imdport, unsigned long gmx_unused Flags, gmx_walltime_accounting_t walltime_accounting) { const char *TPI = "Test Particle Insertion"; gmx_localtop_t *top; gmx_groups_t *groups; gmx_enerdata_t *enerd; rvec *f; real lambda, t, temp, beta, drmax, epot; double embU, sum_embU, *sum_UgembU, V, V_all, VembU_all; t_trxstatus *status; t_trxframe rerun_fr; gmx_bool bDispCorr, bCharge, bRFExcl, bNotLastFrame, bStateChanged, bNS; tensor force_vir, shake_vir, vir, pres; int cg_tp, a_tp0, a_tp1, ngid, gid_tp, nener, e; rvec *x_mol; rvec mu_tot, x_init, dx, x_tp; int nnodes, frame; gmx_int64_t frame_step_prev, frame_step; gmx_int64_t nsteps, stepblocksize = 0, step; gmx_int64_t rnd_count_stride, rnd_count; gmx_int64_t seed; double rnd[4]; int i, start, end; FILE *fp_tpi = NULL; char *ptr, *dump_pdb, **leg, str[STRLEN], str2[STRLEN]; double dbl, dump_ener; gmx_bool bCavity; int nat_cavity = 0, d; real *mass_cavity = NULL, mass_tot; int nbin; double invbinw, *bin, refvolshift, logV, bUlogV; real dvdl, prescorr, enercorr, dvdlcorr; gmx_bool bEnergyOutOfBounds; const char *tpid_leg[2] = {"direct", "reweighted"}; /* Since there is no upper limit to the insertion energies, * we need to set an upper limit for the distribution output. */ real bU_bin_limit = 50; real bU_logV_bin_limit = bU_bin_limit + 10; nnodes = cr->nnodes; top = gmx_mtop_generate_local_top(top_global, inputrec); groups = &top_global->groups; bCavity = (inputrec->eI == eiTPIC); if (bCavity) { ptr = getenv("GMX_TPIC_MASSES"); if (ptr == NULL) { nat_cavity = 1; } else { /* Read (multiple) masses from env var GMX_TPIC_MASSES, * The center of mass of the last atoms is then used for TPIC. */ nat_cavity = 0; while (sscanf(ptr, "%lf%n", &dbl, &i) > 0) { srenew(mass_cavity, nat_cavity+1); mass_cavity[nat_cavity] = dbl; fprintf(fplog, "mass[%d] = %f\n", nat_cavity+1, mass_cavity[nat_cavity]); nat_cavity++; ptr += i; } if (nat_cavity == 0) { gmx_fatal(FARGS, "Found %d masses in GMX_TPIC_MASSES", nat_cavity); } } } /* init_em(fplog,TPI,inputrec,&lambda,nrnb,mu_tot, state->box,fr,mdatoms,top,cr,nfile,fnm,NULL,NULL);*/ /* We never need full pbc for TPI */ fr->ePBC = epbcXYZ; /* Determine the temperature for the Boltzmann weighting */ temp = inputrec->opts.ref_t[0]; if (fplog) { for (i = 1; (i < inputrec->opts.ngtc); i++) { if (inputrec->opts.ref_t[i] != temp) { fprintf(fplog, "\nWARNING: The temperatures of the different temperature coupling groups are not identical\n\n"); fprintf(stderr, "\nWARNING: The temperatures of the different temperature coupling groups are not identical\n\n"); } } fprintf(fplog, "\n The temperature for test particle insertion is %.3f K\n\n", temp); } beta = 1.0/(BOLTZ*temp); /* Number of insertions per frame */ nsteps = inputrec->nsteps; /* Use the same neighborlist with more insertions points * in a sphere of radius drmax around the initial point */ /* This should be a proper mdp parameter */ drmax = inputrec->rtpi; /* An environment variable can be set to dump all configurations * to pdb with an insertion energy <= this value. */ dump_pdb = getenv("GMX_TPI_DUMP"); dump_ener = 0; if (dump_pdb) { sscanf(dump_pdb, "%lf", &dump_ener); } atoms2md(top_global, inputrec, 0, NULL, top_global->natoms, mdatoms); update_mdatoms(mdatoms, inputrec->fepvals->init_lambda); snew(enerd, 1); init_enerdata(groups->grps[egcENER].nr, inputrec->fepvals->n_lambda, enerd); snew(f, top_global->natoms); /* Print to log file */ walltime_accounting_start(walltime_accounting); wallcycle_start(wcycle, ewcRUN); print_start(fplog, cr, walltime_accounting, "Test Particle Insertion"); /* The last charge group is the group to be inserted */ cg_tp = top->cgs.nr - 1; a_tp0 = top->cgs.index[cg_tp]; a_tp1 = top->cgs.index[cg_tp+1]; if (debug) { fprintf(debug, "TPI cg %d, atoms %d-%d\n", cg_tp, a_tp0, a_tp1); } if (a_tp1 - a_tp0 > 1 && (inputrec->rlist < inputrec->rcoulomb || inputrec->rlist < inputrec->rvdw)) { gmx_fatal(FARGS, "Can not do TPI for multi-atom molecule with a twin-range cut-off"); } snew(x_mol, a_tp1-a_tp0); bDispCorr = (inputrec->eDispCorr != edispcNO); bCharge = FALSE; for (i = a_tp0; i < a_tp1; i++) { /* Copy the coordinates of the molecule to be insterted */ copy_rvec(state->x[i], x_mol[i-a_tp0]); /* Check if we need to print electrostatic energies */ bCharge |= (mdatoms->chargeA[i] != 0 || (mdatoms->chargeB && mdatoms->chargeB[i] != 0)); } bRFExcl = (bCharge && EEL_RF(fr->eeltype) && fr->eeltype != eelRF_NEC); calc_cgcm(fplog, cg_tp, cg_tp+1, &(top->cgs), state->x, fr->cg_cm); if (bCavity) { if (norm(fr->cg_cm[cg_tp]) > 0.5*inputrec->rlist && fplog) { fprintf(fplog, "WARNING: Your TPI molecule is not centered at 0,0,0\n"); fprintf(stderr, "WARNING: Your TPI molecule is not centered at 0,0,0\n"); } } else { /* Center the molecule to be inserted at zero */ for (i = 0; i < a_tp1-a_tp0; i++) { rvec_dec(x_mol[i], fr->cg_cm[cg_tp]); } } if (fplog) { fprintf(fplog, "\nWill insert %d atoms %s partial charges\n", a_tp1-a_tp0, bCharge ? "with" : "without"); fprintf(fplog, "\nWill insert %d times in each frame of %s\n", (int)nsteps, opt2fn("-rerun", nfile, fnm)); } if (!bCavity) { if (inputrec->nstlist > 1) { if (drmax == 0 && a_tp1-a_tp0 == 1) { gmx_fatal(FARGS, "Re-using the neighborlist %d times for insertions of a single atom in a sphere of radius %f does not make sense", inputrec->nstlist, drmax); } if (fplog) { fprintf(fplog, "Will use the same neighborlist for %d insertions in a sphere of radius %f\n", inputrec->nstlist, drmax); } } } else { if (fplog) { fprintf(fplog, "Will insert randomly in a sphere of radius %f around the center of the cavity\n", drmax); } } ngid = groups->grps[egcENER].nr; gid_tp = GET_CGINFO_GID(fr->cginfo[cg_tp]); nener = 1 + ngid; if (bDispCorr) { nener += 1; } if (bCharge) { nener += ngid; if (bRFExcl) { nener += 1; } if (EEL_FULL(fr->eeltype)) { nener += 1; } } snew(sum_UgembU, nener); /* Copy the random seed set by the user */ seed = inputrec->ld_seed; /* We use the frame step number as one random counter. * The second counter use the insertion (step) count. But we * need multiple random numbers per insertion. This number is * not fixed, since we generate random locations in a sphere * by putting locations in a cube and some of these fail. * A count of 20 is already extremely unlikely, so 10000 is * a safe margin for random numbers per insertion. */ rnd_count_stride = 10000; if (MASTER(cr)) { fp_tpi = xvgropen(opt2fn("-tpi", nfile, fnm), "TPI energies", "Time (ps)", "(kJ mol\\S-1\\N) / (nm\\S3\\N)", oenv); xvgr_subtitle(fp_tpi, "f. are averages over one frame", oenv); snew(leg, 4+nener); e = 0; sprintf(str, "-kT log(<Ve\\S-\\betaU\\N>/<V>)"); leg[e++] = strdup(str); sprintf(str, "f. -kT log<e\\S-\\betaU\\N>"); leg[e++] = strdup(str); sprintf(str, "f. <e\\S-\\betaU\\N>"); leg[e++] = strdup(str); sprintf(str, "f. V"); leg[e++] = strdup(str); sprintf(str, "f. <Ue\\S-\\betaU\\N>"); leg[e++] = strdup(str); for (i = 0; i < ngid; i++) { sprintf(str, "f. <U\\sVdW %s\\Ne\\S-\\betaU\\N>", *(groups->grpname[groups->grps[egcENER].nm_ind[i]])); leg[e++] = strdup(str); } if (bDispCorr) { sprintf(str, "f. <U\\sdisp c\\Ne\\S-\\betaU\\N>"); leg[e++] = strdup(str); } if (bCharge) { for (i = 0; i < ngid; i++) { sprintf(str, "f. <U\\sCoul %s\\Ne\\S-\\betaU\\N>", *(groups->grpname[groups->grps[egcENER].nm_ind[i]])); leg[e++] = strdup(str); } if (bRFExcl) { sprintf(str, "f. <U\\sRF excl\\Ne\\S-\\betaU\\N>"); leg[e++] = strdup(str); } if (EEL_FULL(fr->eeltype)) { sprintf(str, "f. <U\\sCoul recip\\Ne\\S-\\betaU\\N>"); leg[e++] = strdup(str); } } xvgr_legend(fp_tpi, 4+nener, (const char**)leg, oenv); for (i = 0; i < 4+nener; i++) { sfree(leg[i]); } sfree(leg); } clear_rvec(x_init); V_all = 0; VembU_all = 0; invbinw = 10; nbin = 10; snew(bin, nbin); /* Avoid frame step numbers <= -1 */ frame_step_prev = -1; bNotLastFrame = read_first_frame(oenv, &status, opt2fn("-rerun", nfile, fnm), &rerun_fr, TRX_NEED_X); frame = 0; if (rerun_fr.natoms - (bCavity ? nat_cavity : 0) != mdatoms->nr - (a_tp1 - a_tp0)) { gmx_fatal(FARGS, "Number of atoms in trajectory (%d)%s " "is not equal the number in the run input file (%d) " "minus the number of atoms to insert (%d)\n", rerun_fr.natoms, bCavity ? " minus one" : "", mdatoms->nr, a_tp1-a_tp0); } refvolshift = log(det(rerun_fr.box)); switch (inputrec->eI) { case eiTPI: stepblocksize = inputrec->nstlist; break; case eiTPIC: stepblocksize = 1; break; default: gmx_fatal(FARGS, "Unknown integrator %s", ei_names[inputrec->eI]); } #ifdef GMX_SIMD /* Make sure we don't detect SIMD overflow generated before this point */ gmx_simd_check_and_reset_overflow(); #endif while (bNotLastFrame) { frame_step = rerun_fr.step; if (frame_step <= frame_step_prev) { /* We don't have step number in the trajectory file, * or we have constant or decreasing step numbers. * Ensure we have increasing step numbers, since we use * the step numbers as a counter for random numbers. */ frame_step = frame_step_prev + 1; } frame_step_prev = frame_step; lambda = rerun_fr.lambda; t = rerun_fr.time; sum_embU = 0; for (e = 0; e < nener; e++) { sum_UgembU[e] = 0; } /* Copy the coordinates from the input trajectory */ for (i = 0; i < rerun_fr.natoms; i++) { copy_rvec(rerun_fr.x[i], state->x[i]); } copy_mat(rerun_fr.box, state->box); V = det(state->box); logV = log(V); bStateChanged = TRUE; bNS = TRUE; step = cr->nodeid*stepblocksize; while (step < nsteps) { /* Initialize the second counter for random numbers using * the insertion step index. This ensures that we get * the same random numbers independently of how many * MPI ranks we use. Also for the same seed, we get * the same initial random sequence for different nsteps. */ rnd_count = step*rnd_count_stride; if (!bCavity) { /* Random insertion in the whole volume */ bNS = (step % inputrec->nstlist == 0); if (bNS) { /* Generate a random position in the box */ gmx_rng_cycle_2uniform(frame_step, rnd_count++, seed, RND_SEED_TPI, rnd); gmx_rng_cycle_2uniform(frame_step, rnd_count++, seed, RND_SEED_TPI, rnd+2); for (d = 0; d < DIM; d++) { x_init[d] = rnd[d]*state->box[d][d]; } } if (inputrec->nstlist == 1) { copy_rvec(x_init, x_tp); } else { /* Generate coordinates within |dx|=drmax of x_init */ do { gmx_rng_cycle_2uniform(frame_step, rnd_count++, seed, RND_SEED_TPI, rnd); gmx_rng_cycle_2uniform(frame_step, rnd_count++, seed, RND_SEED_TPI, rnd+2); for (d = 0; d < DIM; d++) { dx[d] = (2*rnd[d] - 1)*drmax; } } while (norm2(dx) > drmax*drmax); rvec_add(x_init, dx, x_tp); } } else { /* Random insertion around a cavity location * given by the last coordinate of the trajectory. */ if (step == 0) { if (nat_cavity == 1) { /* Copy the location of the cavity */ copy_rvec(rerun_fr.x[rerun_fr.natoms-1], x_init); } else { /* Determine the center of mass of the last molecule */ clear_rvec(x_init); mass_tot = 0; for (i = 0; i < nat_cavity; i++) { for (d = 0; d < DIM; d++) { x_init[d] += mass_cavity[i]*rerun_fr.x[rerun_fr.natoms-nat_cavity+i][d]; } mass_tot += mass_cavity[i]; } for (d = 0; d < DIM; d++) { x_init[d] /= mass_tot; } } } /* Generate coordinates within |dx|=drmax of x_init */ do { gmx_rng_cycle_2uniform(frame_step, rnd_count++, seed, RND_SEED_TPI, rnd); gmx_rng_cycle_2uniform(frame_step, rnd_count++, seed, RND_SEED_TPI, rnd+2); for (d = 0; d < DIM; d++) { dx[d] = (2*rnd[d] - 1)*drmax; } } while (norm2(dx) > drmax*drmax); rvec_add(x_init, dx, x_tp); } if (a_tp1 - a_tp0 == 1) { /* Insert a single atom, just copy the insertion location */ copy_rvec(x_tp, state->x[a_tp0]); } else { /* Copy the coordinates from the top file */ for (i = a_tp0; i < a_tp1; i++) { copy_rvec(x_mol[i-a_tp0], state->x[i]); } /* Rotate the molecule randomly */ gmx_rng_cycle_2uniform(frame_step, rnd_count++, seed, RND_SEED_TPI, rnd); gmx_rng_cycle_2uniform(frame_step, rnd_count++, seed, RND_SEED_TPI, rnd+2); rotate_conf(a_tp1-a_tp0, state->x+a_tp0, NULL, 2*M_PI*rnd[0], 2*M_PI*rnd[1], 2*M_PI*rnd[2]); /* Shift to the insertion location */ for (i = a_tp0; i < a_tp1; i++) { rvec_inc(state->x[i], x_tp); } } /* Clear some matrix variables */ clear_mat(force_vir); clear_mat(shake_vir); clear_mat(vir); clear_mat(pres); /* Set the charge group center of mass of the test particle */ copy_rvec(x_init, fr->cg_cm[top->cgs.nr-1]); /* Calc energy (no forces) on new positions. * Since we only need the intermolecular energy * and the RF exclusion terms of the inserted molecule occur * within a single charge group we can pass NULL for the graph. * This also avoids shifts that would move charge groups * out of the box. * * Some checks above ensure than we can not have * twin-range interactions together with nstlist > 1, * therefore we do not need to remember the LR energies. */ /* Make do_force do a single node force calculation */ cr->nnodes = 1; do_force(fplog, cr, inputrec, step, nrnb, wcycle, top, &top_global->groups, state->box, state->x, &state->hist, f, force_vir, mdatoms, enerd, fcd, state->lambda, NULL, fr, NULL, mu_tot, t, NULL, NULL, FALSE, GMX_FORCE_NONBONDED | GMX_FORCE_ENERGY | (bNS ? GMX_FORCE_DYNAMICBOX | GMX_FORCE_NS | GMX_FORCE_DO_LR : 0) | (bStateChanged ? GMX_FORCE_STATECHANGED : 0)); cr->nnodes = nnodes; bStateChanged = FALSE; bNS = FALSE; /* Calculate long range corrections to pressure and energy */ calc_dispcorr(fplog, inputrec, fr, step, top_global->natoms, state->box, lambda, pres, vir, &prescorr, &enercorr, &dvdlcorr); /* figure out how to rearrange the next 4 lines MRS 8/4/2009 */ enerd->term[F_DISPCORR] = enercorr; enerd->term[F_EPOT] += enercorr; enerd->term[F_PRES] += prescorr; enerd->term[F_DVDL_VDW] += dvdlcorr; epot = enerd->term[F_EPOT]; bEnergyOutOfBounds = FALSE; #ifdef GMX_SIMD_X86_SSE2_OR_HIGHER /* With SSE the energy can overflow, check for this */ if (gmx_mm_check_and_reset_overflow()) { if (debug) { fprintf(debug, "Found an SSE overflow, assuming the energy is out of bounds\n"); } bEnergyOutOfBounds = TRUE; } #endif /* If the compiler doesn't optimize this check away * we catch the NAN energies. * The epot>GMX_REAL_MAX check catches inf values, * which should nicely result in embU=0 through the exp below, * but it does not hurt to check anyhow. */ /* Non-bonded Interaction usually diverge at r=0. * With tabulated interaction functions the first few entries * should be capped in a consistent fashion between * repulsion, dispersion and Coulomb to avoid accidental * negative values in the total energy. * The table generation code in tables.c does this. * With user tbales the user should take care of this. */ if (epot != epot || epot > GMX_REAL_MAX) { bEnergyOutOfBounds = TRUE; } if (bEnergyOutOfBounds) { if (debug) { fprintf(debug, "\n time %.3f, step %d: non-finite energy %f, using exp(-bU)=0\n", t, (int)step, epot); } embU = 0; } else { embU = exp(-beta*epot); sum_embU += embU; /* Determine the weighted energy contributions of each energy group */ e = 0; sum_UgembU[e++] += epot*embU; if (fr->bBHAM) { for (i = 0; i < ngid; i++) { sum_UgembU[e++] += (enerd->grpp.ener[egBHAMSR][GID(i, gid_tp, ngid)] + enerd->grpp.ener[egBHAMLR][GID(i, gid_tp, ngid)])*embU; } } else { for (i = 0; i < ngid; i++) { sum_UgembU[e++] += (enerd->grpp.ener[egLJSR][GID(i, gid_tp, ngid)] + enerd->grpp.ener[egLJLR][GID(i, gid_tp, ngid)])*embU; } } if (bDispCorr) { sum_UgembU[e++] += enerd->term[F_DISPCORR]*embU; } if (bCharge) { for (i = 0; i < ngid; i++) { sum_UgembU[e++] += (enerd->grpp.ener[egCOULSR][GID(i, gid_tp, ngid)] + enerd->grpp.ener[egCOULLR][GID(i, gid_tp, ngid)])*embU; } if (bRFExcl) { sum_UgembU[e++] += enerd->term[F_RF_EXCL]*embU; } if (EEL_FULL(fr->eeltype)) { sum_UgembU[e++] += enerd->term[F_COUL_RECIP]*embU; } } } if (embU == 0 || beta*epot > bU_bin_limit) { bin[0]++; } else { i = (int)((bU_logV_bin_limit - (beta*epot - logV + refvolshift))*invbinw + 0.5); if (i < 0) { i = 0; } if (i >= nbin) { realloc_bins(&bin, &nbin, i+10); } bin[i]++; } if (debug) { fprintf(debug, "TPI %7d %12.5e %12.5f %12.5f %12.5f\n", (int)step, epot, x_tp[XX], x_tp[YY], x_tp[ZZ]); } if (dump_pdb && epot <= dump_ener) { sprintf(str, "t%g_step%d.pdb", t, (int)step); sprintf(str2, "t: %f step %d ener: %f", t, (int)step, epot); write_sto_conf_mtop(str, str2, top_global, state->x, state->v, inputrec->ePBC, state->box); } step++; if ((step/stepblocksize) % cr->nnodes != cr->nodeid) { /* Skip all steps assigned to the other MPI ranks */ step += (cr->nnodes - 1)*stepblocksize; } } if (PAR(cr)) { /* When running in parallel sum the energies over the processes */ gmx_sumd(1, &sum_embU, cr); gmx_sumd(nener, sum_UgembU, cr); } frame++; V_all += V; VembU_all += V*sum_embU/nsteps; if (fp_tpi) { if (bVerbose || frame%10 == 0 || frame < 10) { fprintf(stderr, "mu %10.3e <mu> %10.3e\n", -log(sum_embU/nsteps)/beta, -log(VembU_all/V_all)/beta); } fprintf(fp_tpi, "%10.3f %12.5e %12.5e %12.5e %12.5e", t, VembU_all == 0 ? 20/beta : -log(VembU_all/V_all)/beta, sum_embU == 0 ? 20/beta : -log(sum_embU/nsteps)/beta, sum_embU/nsteps, V); for (e = 0; e < nener; e++) { fprintf(fp_tpi, " %12.5e", sum_UgembU[e]/nsteps); } fprintf(fp_tpi, "\n"); fflush(fp_tpi); } bNotLastFrame = read_next_frame(oenv, status, &rerun_fr); } /* End of the loop */ walltime_accounting_end(walltime_accounting); close_trj(status); if (fp_tpi != NULL) { gmx_fio_fclose(fp_tpi); } if (fplog != NULL) { fprintf(fplog, "\n"); fprintf(fplog, " <V> = %12.5e nm^3\n", V_all/frame); fprintf(fplog, " <mu> = %12.5e kJ/mol\n", -log(VembU_all/V_all)/beta); } /* Write the Boltzmann factor histogram */ if (PAR(cr)) { /* When running in parallel sum the bins over the processes */ i = nbin; global_max(cr, &i); realloc_bins(&bin, &nbin, i); gmx_sumd(nbin, bin, cr); } if (MASTER(cr)) { fp_tpi = xvgropen(opt2fn("-tpid", nfile, fnm), "TPI energy distribution", "\\betaU - log(V/<V>)", "count", oenv); sprintf(str, "number \\betaU > %g: %9.3e", bU_bin_limit, bin[0]); xvgr_subtitle(fp_tpi, str, oenv); xvgr_legend(fp_tpi, 2, (const char **)tpid_leg, oenv); for (i = nbin-1; i > 0; i--) { bUlogV = -i/invbinw + bU_logV_bin_limit - refvolshift + log(V_all/frame); fprintf(fp_tpi, "%6.2f %10d %12.5e\n", bUlogV, (int)(bin[i]+0.5), bin[i]*exp(-bUlogV)*V_all/VembU_all); } gmx_fio_fclose(fp_tpi); } sfree(bin); sfree(sum_UgembU); walltime_accounting_set_nsteps_done(walltime_accounting, frame*inputrec->nsteps); return 0; }
void do_force(FILE *log,t_commrec *cr,t_commrec *mcr, t_parm *parm,t_nsborder *nsb,tensor vir_part,tensor pme_vir, int step,t_nrnb *nrnb,t_topology *top,t_groups *grps, rvec x[],rvec v[],rvec f[],rvec buf[], t_mdatoms *mdatoms,real ener[],t_fcdata *fcd,bool bVerbose, real lambda,t_graph *graph, bool bNS,bool bNBFonly,t_forcerec *fr, rvec mu_tot, bool bGatherOnly) { static rvec box_size; static real dvdl_lr = 0; int cg0,cg1,i,j; int start,homenr; static real mu_and_q[DIM+1]; real qsum; start = START(nsb); homenr = HOMENR(nsb); cg0 = CG0(nsb); cg1 = CG1(nsb); update_forcerec(log,fr,parm->box); /* Calculate total (local) dipole moment in a temporary common array. * This makes it possible to sum them over nodes faster. */ calc_mu_and_q(nsb,x,mdatoms->chargeT,mu_and_q,mu_and_q+DIM); if (fr->ePBC != epbcNONE) { /* Compute shift vectors every step, because of pressure coupling! */ if (parm->ir.epc != epcNO) calc_shifts(parm->box,box_size,fr->shift_vec); if (bNS) { put_charge_groups_in_box(log,cg0,cg1,parm->box,box_size, &(top->blocks[ebCGS]),x,fr->cg_cm); inc_nrnb(nrnb,eNR_RESETX,homenr); } else if (parm->ir.eI==eiSteep || parm->ir.eI==eiCG) unshift_self(graph,parm->box,x); } else if (bNS) calc_cgcm(log,cg0,cg1,&(top->blocks[ebCGS]),x,fr->cg_cm); if (bNS) { inc_nrnb(nrnb,eNR_CGCM,cg1-cg0); if (PAR(cr)) move_cgcm(log,cr,fr->cg_cm,nsb->workload); if (debug) pr_rvecs(debug,0,"cgcm",fr->cg_cm,nsb->cgtotal); } /* Communicate coordinates and sum dipole and net charge if necessary */ if (PAR(cr)) { move_x(log,cr->left,cr->right,x,nsb,nrnb); gmx_sum(DIM+1,mu_and_q,cr); } for(i=0;i<DIM;i++) mu_tot[i]=mu_and_q[i]; qsum=mu_and_q[DIM]; /* Reset energies */ reset_energies(&(parm->ir.opts),grps,fr,bNS,ener); if (bNS) { if (fr->ePBC != epbcNONE) /* Calculate intramolecular shift vectors to make molecules whole */ mk_mshift(log,graph,parm->box,x); /* Reset long range forces if necessary */ if (fr->bTwinRange) { clear_rvecs(nsb->natoms,fr->f_twin); clear_rvecs(SHIFTS,fr->fshift_twin); } /* Do the actual neighbour searching and if twin range electrostatics * also do the calculation of long range forces and energies. */ dvdl_lr = 0; ns(log,fr,x,f,parm->box,grps,&(parm->ir.opts),top,mdatoms, cr,nrnb,nsb,step,lambda,&dvdl_lr); } /* Reset PME/Ewald forces if necessary */ if (EEL_LR(fr->eeltype)) clear_rvecs(homenr,fr->f_pme+start); /* Copy long range forces into normal buffers */ if (fr->bTwinRange) { for(i=0; i<nsb->natoms; i++) copy_rvec(fr->f_twin[i],f[i]); for(i=0; i<SHIFTS; i++) copy_rvec(fr->fshift_twin[i],fr->fshift[i]); } else { clear_rvecs(nsb->natoms,f); clear_rvecs(SHIFTS,fr->fshift); } /* Compute the forces */ force(log,step,fr,&(parm->ir),&(top->idef),nsb,cr,mcr,nrnb,grps,mdatoms, top->atoms.grps[egcENER].nr,&(parm->ir.opts), x,f,ener,fcd,bVerbose,parm->box,lambda,graph,&(top->atoms.excl), bNBFonly,pme_vir,mu_tot,qsum,bGatherOnly); /* Take long range contribution to free energy into account */ ener[F_DVDL] += dvdl_lr; #ifdef DEBUG if (bNS) print_nrnb(log,nrnb); #endif /* The short-range virial from surrounding boxes */ clear_mat(vir_part); calc_vir(log,SHIFTS,fr->shift_vec,fr->fshift,vir_part); inc_nrnb(nrnb,eNR_VIRIAL,SHIFTS); if (debug) pr_rvecs(debug,0,"vir_shifts",vir_part,DIM); /* Compute forces due to electric field */ calc_f_el(start,homenr,mdatoms->chargeT,f,parm->ir.ex); /* When using PME/Ewald we compute the long range virial (pme_vir) there. * otherwise we do it based on long range forces from twin range * cut-off based calculation (or not at all). */ /* Communicate the forces */ if (PAR(cr)) move_f(log,cr->left,cr->right,f,buf,nsb,nrnb); }