Exemple #1
0
double do_md_openmm(FILE *fplog,t_commrec *cr,int nfile,const t_filenm fnm[],
                    const output_env_t oenv, gmx_bool bVerbose,gmx_bool bCompact,
                    int nstglobalcomm,
                    gmx_vsite_t *vsite,gmx_constr_t constr,
                    int stepout,t_inputrec *ir,
                    gmx_mtop_t *top_global,
                    t_fcdata *fcd,
                    t_state *state_global,
                    t_mdatoms *mdatoms,
                    t_nrnb *nrnb,gmx_wallcycle_t wcycle,
                    gmx_edsam_t ed,t_forcerec *fr,
                    int repl_ex_nst,int repl_ex_seed,
                    real cpt_period,real max_hours,
                    const char *deviceOptions,
                    unsigned long Flags,
                    gmx_runtime_t *runtime)
{
    gmx_mdoutf_t *outf;
    gmx_large_int_t step,step_rel;
    double     run_time;
    double     t,t0,lam0;
    gmx_bool       bSimAnn,
    bFirstStep,bStateFromTPX,bLastStep,bStartingFromCpt;
    gmx_bool       bInitStep=TRUE;
    gmx_bool       do_ene,do_log, do_verbose,
    bX,bV,bF,bCPT;
    tensor     force_vir,shake_vir,total_vir,pres;
    int        i,m;
    int        mdof_flags;
    rvec       mu_tot;
    t_vcm      *vcm;
    int        nchkpt=1;
    gmx_localtop_t *top;
    t_mdebin *mdebin=NULL;
    t_state    *state=NULL;
    rvec       *f_global=NULL;
    int        n_xtc=-1;
    rvec       *x_xtc=NULL;
    gmx_enerdata_t *enerd;
    rvec       *f=NULL;
    gmx_global_stat_t gstat;
    gmx_update_t upd=NULL;
    t_graph    *graph=NULL;
    globsig_t   gs;

    gmx_groups_t *groups;
    gmx_ekindata_t *ekind, *ekind_save;
    gmx_bool        bAppend;
    int         a0,a1;
    matrix      lastbox;
    real        reset_counters=0,reset_counters_now=0;
    char        sbuf[STEPSTRSIZE],sbuf2[STEPSTRSIZE];
    int         handled_stop_condition=gmx_stop_cond_none; 

    const char *ommOptions = NULL;
    void   *openmmData;

    bAppend  = (Flags & MD_APPENDFILES);
    check_ir_old_tpx_versions(cr,fplog,ir,top_global);

    groups = &top_global->groups;

    /* Initial values */
    init_md(fplog,cr,ir,oenv,&t,&t0,&state_global->lambda,&lam0,
            nrnb,top_global,&upd,
            nfile,fnm,&outf,&mdebin,
            force_vir,shake_vir,mu_tot,&bSimAnn,&vcm,state_global,Flags);

    clear_mat(total_vir);
    clear_mat(pres);
    /* Energy terms and groups */
    snew(enerd,1);
    init_enerdata(top_global->groups.grps[egcENER].nr,ir->n_flambda,enerd);
    snew(f,top_global->natoms);

    /* Kinetic energy data */
    snew(ekind,1);
    init_ekindata(fplog,top_global,&(ir->opts),ekind);
    /* needed for iteration of constraints */
    snew(ekind_save,1);
    init_ekindata(fplog,top_global,&(ir->opts),ekind_save);
    /* Copy the cos acceleration to the groups struct */
    ekind->cosacc.cos_accel = ir->cos_accel;

    gstat = global_stat_init(ir);
    debug_gmx();

    {
        double io = compute_io(ir,top_global->natoms,groups,mdebin->ebin->nener,1);
        if ((io > 2000) && MASTER(cr))
            fprintf(stderr,
                    "\nWARNING: This run will generate roughly %.0f Mb of data\n\n",
                    io);
    }

    top = gmx_mtop_generate_local_top(top_global,ir);

    a0 = 0;
    a1 = top_global->natoms;

    state = partdec_init_local_state(cr,state_global);
    f_global = f;

    atoms2md(top_global,ir,0,NULL,a0,a1-a0,mdatoms);

    if (vsite)
    {
        set_vsite_top(vsite,top,mdatoms,cr);
    }

    if (ir->ePBC != epbcNONE && !ir->bPeriodicMols)
    {
        graph = mk_graph(fplog,&(top->idef),0,top_global->natoms,FALSE,FALSE);
    }

    update_mdatoms(mdatoms,state->lambda);

    if (deviceOptions[0]=='\0')
    {
        /* empty options, which should default to OpenMM in this build */
        ommOptions=deviceOptions;
    }
    else
    {
        if (gmx_strncasecmp(deviceOptions,"OpenMM",6)!=0)
        {
            gmx_fatal(FARGS, "This Gromacs version currently only works with OpenMM. Use -device \"OpenMM:<options>\"");
        }
        else
        {
            ommOptions=strchr(deviceOptions,':');
            if (NULL!=ommOptions)
            {
                /* Increase the pointer to skip the colon */
                ommOptions++;
            }
        }
    }

    openmmData = openmm_init(fplog, ommOptions, ir, top_global, top, mdatoms, fr, state);
    please_cite(fplog,"Friedrichs2009");

    if (MASTER(cr))
    {
        /* Update mdebin with energy history if appending to output files */
        if ( Flags & MD_APPENDFILES )
        {
            restore_energyhistory_from_state(mdebin,&state_global->enerhist);
        }
        /* Set the initial energy history in state to zero by updating once */
        update_energyhistory(&state_global->enerhist,mdebin);
    }

    if (constr)
    {
        set_constraints(constr,top,ir,mdatoms,cr);
    }

    if (!ir->bContinuation)
    {
        if (mdatoms->cFREEZE && (state->flags & (1<<estV)))
        {
            /* Set the velocities of frozen particles to zero */
            for (i=mdatoms->start; i<mdatoms->start+mdatoms->homenr; i++)
            {
                for (m=0; m<DIM; m++)
                {
                    if (ir->opts.nFreeze[mdatoms->cFREEZE[i]][m])
                    {
                        state->v[i][m] = 0;
                    }
                }
            }
        }

        if (constr)
        {
            /* Constrain the initial coordinates and velocities */
            do_constrain_first(fplog,constr,ir,mdatoms,state,f,
                               graph,cr,nrnb,fr,top,shake_vir);
        }
        if (vsite)
        {
            /* Construct the virtual sites for the initial configuration */
            construct_vsites(fplog,vsite,state->x,nrnb,ir->delta_t,NULL,
                             top->idef.iparams,top->idef.il,
                             fr->ePBC,fr->bMolPBC,graph,cr,state->box);
        }
    }

    debug_gmx();

    if (MASTER(cr))
    {
        char tbuf[20];
        fprintf(fplog,"Initial temperature: %g K\n",enerd->term[F_TEMP]);
        fprintf(stderr,"starting mdrun '%s'\n",
                *(top_global->name));
        if (ir->nsteps >= 0)
        {
            sprintf(tbuf,"%8.1f",(ir->init_step+ir->nsteps)*ir->delta_t);
        }
        else
        {
            sprintf(tbuf,"%s","infinite");
        }
        if (ir->init_step > 0)
        {
            fprintf(stderr,"%s steps, %s ps (continuing from step %s, %8.1f ps).\n",
                    gmx_step_str(ir->init_step+ir->nsteps,sbuf),tbuf,
                    gmx_step_str(ir->init_step,sbuf2),
                    ir->init_step*ir->delta_t);
        }
        else
        {
            fprintf(stderr,"%s steps, %s ps.\n",
                    gmx_step_str(ir->nsteps,sbuf),tbuf);
        }
    }

    fprintf(fplog,"\n");

    /* Set and write start time */
    runtime_start(runtime);
    print_date_and_time(fplog,cr->nodeid,"Started mdrun",runtime);
    wallcycle_start(wcycle,ewcRUN);
    if (fplog)
        fprintf(fplog,"\n");

    /* safest point to do file checkpointing is here.  More general point would be immediately before integrator call */

    debug_gmx();
    /***********************************************************
     *
     *             Loop over MD steps
     *
     ************************************************************/

    /* loop over MD steps or if rerunMD to end of input trajectory */
    bFirstStep = TRUE;
    /* Skip the first Nose-Hoover integration when we get the state from tpx */
    bStateFromTPX = !opt2bSet("-cpi",nfile,fnm);
    bInitStep = bFirstStep && bStateFromTPX;
    bStartingFromCpt = (Flags & MD_STARTFROMCPT) && bInitStep;
    bLastStep = FALSE;

    init_global_signals(&gs,cr,ir,repl_ex_nst);

    step = ir->init_step;
    step_rel = 0;

    while (!bLastStep)
    {
        wallcycle_start(wcycle,ewcSTEP);

        GMX_MPE_LOG(ev_timestep1);

        bLastStep = (step_rel == ir->nsteps);
        t = t0 + step*ir->delta_t;

        if (gs.set[eglsSTOPCOND] != 0)
        {
            bLastStep = TRUE;
        }

        do_log = do_per_step(step,ir->nstlog) || bFirstStep || bLastStep;
        do_verbose = bVerbose &&
                     (step % stepout == 0 || bFirstStep || bLastStep);

        if (MASTER(cr) && do_log)
        {
            print_ebin_header(fplog,step,t,state->lambda);
        }

        clear_mat(force_vir);
        GMX_MPE_LOG(ev_timestep2);

        /* We write a checkpoint at this MD step when:
         * either when we signalled through gs (in OpenMM NS works different),
         * or at the last step (but not when we do not want confout),
         * but never at the first step.
         */
        bCPT = ((gs.set[eglsCHKPT] ||
                 (bLastStep && (Flags & MD_CONFOUT))) &&
                step > ir->init_step );
        if (bCPT)
        {
            gs.set[eglsCHKPT] = 0;
        }

        /* Now we have the energies and forces corresponding to the
         * coordinates at time t. We must output all of this before
         * the update.
         * for RerunMD t is read from input trajectory
         */
        GMX_MPE_LOG(ev_output_start);

        mdof_flags = 0;
        if (do_per_step(step,ir->nstxout))
        {
            mdof_flags |= MDOF_X;
        }
        if (do_per_step(step,ir->nstvout))
        {
            mdof_flags |= MDOF_V;
        }
        if (do_per_step(step,ir->nstfout))
        {
            mdof_flags |= MDOF_F;
        }
        if (do_per_step(step,ir->nstxtcout))
        {
            mdof_flags |= MDOF_XTC;
        }
        if (bCPT)
        {
            mdof_flags |= MDOF_CPT;
        };
        do_ene = (do_per_step(step,ir->nstenergy) || bLastStep);

        if (mdof_flags != 0 || do_ene || do_log)
        {
            wallcycle_start(wcycle,ewcTRAJ);
            bF = (mdof_flags & MDOF_F);
            bX = (mdof_flags & (MDOF_X | MDOF_XTC | MDOF_CPT));
            bV = (mdof_flags & (MDOF_V | MDOF_CPT));

            openmm_copy_state(openmmData, state, &t, f, enerd, bX, bV, bF, do_ene);

            upd_mdebin(mdebin, FALSE,TRUE,
                       t,mdatoms->tmass,enerd,state,lastbox,
                       shake_vir,force_vir,total_vir,pres,
                       ekind,mu_tot,constr);
            print_ebin(outf->fp_ene,do_ene,FALSE,FALSE,do_log?fplog:NULL,
                       step,t,
                       eprNORMAL,bCompact,mdebin,fcd,groups,&(ir->opts));
            write_traj(fplog,cr,outf,mdof_flags,top_global,
                       step,t,state,state_global,f,f_global,&n_xtc,&x_xtc);
            if (bCPT)
            {
                nchkpt++;
                bCPT = FALSE;
            }
            debug_gmx();
            if (bLastStep && step_rel == ir->nsteps &&
                    (Flags & MD_CONFOUT) && MASTER(cr))
            {
                /* x and v have been collected in write_traj,
                 * because a checkpoint file will always be written
                 * at the last step.
                 */
                fprintf(stderr,"\nWriting final coordinates.\n");
                if (ir->ePBC != epbcNONE && !ir->bPeriodicMols)
                {
                    /* Make molecules whole only for confout writing */
                    do_pbc_mtop(fplog,ir->ePBC,state->box,top_global,state_global->x);
                }
                write_sto_conf_mtop(ftp2fn(efSTO,nfile,fnm),
                                    *top_global->name,top_global,
                                    state_global->x,state_global->v,
                                    ir->ePBC,state->box);
                debug_gmx();
            }
            wallcycle_stop(wcycle,ewcTRAJ);
        }
        GMX_MPE_LOG(ev_output_finish);


        /* Determine the wallclock run time up till now */
        run_time = gmx_gettime() - (double)runtime->real;

        /* Check whether everything is still allright */
        if (((int)gmx_get_stop_condition() > handled_stop_condition)
#ifdef GMX_THREADS
            && MASTER(cr)
#endif
            )
        {
           /* this is just make gs.sig compatible with the hack 
               of sending signals around by MPI_Reduce with together with
               other floats */
            /* NOTE: this only works for serial code. For code that allows
               MPI nodes to propagate their condition, see kernel/md.c*/
            if ( gmx_get_stop_condition() == gmx_stop_cond_next_ns )
                gs.set[eglsSTOPCOND]=1;
            if ( gmx_get_stop_condition() == gmx_stop_cond_next )
                gs.set[eglsSTOPCOND]=1;
            /* < 0 means stop at next step, > 0 means stop at next NS step */
            if (fplog)
            {
                fprintf(fplog,
                        "\n\nReceived the %s signal, stopping at the next %sstep\n\n",
                        gmx_get_signal_name(),
                        gs.sig[eglsSTOPCOND]==1 ? "NS " : "");
                fflush(fplog);
            }
            fprintf(stderr,
                    "\n\nReceived the %s signal, stopping at the next %sstep\n\n",
                    gmx_get_signal_name(),
                    gs.sig[eglsSTOPCOND]==1 ? "NS " : "");
            fflush(stderr);
            handled_stop_condition=(int)gmx_get_stop_condition();
        }
        else if (MASTER(cr) &&
                 (max_hours > 0 && run_time > max_hours*60.0*60.0*0.99) &&
                 gs.set[eglsSTOPCOND] == 0)
        {
            /* Signal to terminate the run */
            gs.set[eglsSTOPCOND] = 1;
            if (fplog)
            {
                fprintf(fplog,"\nStep %s: Run time exceeded %.3f hours, will terminate the run\n",gmx_step_str(step,sbuf),max_hours*0.99);
            }
            fprintf(stderr, "\nStep %s: Run time exceeded %.3f hours, will terminate the run\n",gmx_step_str(step,sbuf),max_hours*0.99);
        }

        /* checkpoints */
        if (MASTER(cr) && (cpt_period >= 0 &&
                           (cpt_period == 0 ||
                            run_time >= nchkpt*cpt_period*60.0)) &&
                gs.set[eglsCHKPT] == 0)
        {
            gs.set[eglsCHKPT] = 1;
        }

        /* Time for performance */
        if (((step % stepout) == 0) || bLastStep)
        {
            runtime_upd_proc(runtime);
        }

        if (do_per_step(step,ir->nstlog))
        {
            if (fflush(fplog) != 0)
            {
                gmx_fatal(FARGS,"Cannot flush logfile - maybe you are out of quota?");
            }
        }

        /* Remaining runtime */
        if (MULTIMASTER(cr) && (do_verbose || gmx_got_usr_signal() ))
        {
            print_time(stderr,runtime,step,ir,cr);
        }

        bFirstStep = FALSE;
        bInitStep = FALSE;
        bStartingFromCpt = FALSE;
        step++;
        step_rel++;

        openmm_take_one_step(openmmData);
    }
    /* End of main MD loop */
    debug_gmx();

    /* Stop the time */
    runtime_end(runtime);

    if (MASTER(cr))
    {
        if (ir->nstcalcenergy > 0) 
        {
            print_ebin(outf->fp_ene,FALSE,FALSE,FALSE,fplog,step,t,
                       eprAVER,FALSE,mdebin,fcd,groups,&(ir->opts));
        }
    }

    openmm_cleanup(fplog, openmmData);

    done_mdoutf(outf);

    debug_gmx();

    runtime->nsteps_done = step_rel;

    return 0;
}
Exemple #2
0
double do_tpi(FILE *fplog,t_commrec *cr,
              int nfile, const t_filenm fnm[],
              const output_env_t oenv, gmx_bool bVerbose,gmx_bool bCompact,
              int nstglobalcomm,
              gmx_vsite_t *vsite,gmx_constr_t constr,
              int 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 ed,
              t_forcerec *fr,
              int repl_ex_nst,int repl_ex_seed,
              real cpt_period,real max_hours,
              const char *deviceOptions,
              unsigned long Flags,
              gmx_runtime_t *runtime)
{
  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,bOurStep;
  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,nsteps,step;
  int    i,start,end;
  gmx_rng_t tpi_rand;
  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,0,top_global->natoms,mdatoms);
  update_mdatoms(mdatoms,inputrec->init_lambda);

  snew(enerd,1);
  init_enerdata(groups->grps[egcENER].nr,inputrec->n_flambda,enerd);
  snew(f,top_global->natoms);

  /* Print to log file  */
  runtime_start(runtime);
  print_date_and_time(fplog,cr->nodeid,
                      "Started Test Particle Insertion",runtime); 
  wallcycle_start(wcycle,ewcRUN);

  /* 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",
	    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);

  /* Initialize random generator */
  tpi_rand = gmx_rng_init(inputrec->ld_seed);

  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);

  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));

#if ( defined(GMX_IA32_SSE) || defined(GMX_X86_64_SSE) || defined(GMX_X86_64_SSE2) )
    /* Make sure we don't detect SSE overflow generated before this point */
    gmx_mm_check_and_reset_overflow();
#endif

    while (bNotLastFrame)
    {
        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]);
        }
        
        V = det(rerun_fr.box);
        logV = log(V);
        
        bStateChanged = TRUE;
        bNS = TRUE;
        for(step=0; step<nsteps; step++)
        {
            /* In parallel all nodes generate all random configurations.
             * In that way the result is identical to a single cpu tpi run.
             */
            if (!bCavity)
            {
                /* Random insertion in the whole volume */
                bNS = (step % inputrec->nstlist == 0);
                if (bNS)
                {
                    /* Generate a random position in the box */
                    x_init[XX] = gmx_rng_uniform_real(tpi_rand)*state->box[XX][XX];
                    x_init[YY] = gmx_rng_uniform_real(tpi_rand)*state->box[YY][YY];
                    x_init[ZZ] = gmx_rng_uniform_real(tpi_rand)*state->box[ZZ][ZZ];
                }
                if (inputrec->nstlist == 1)
                {
                    copy_rvec(x_init,x_tp);
                }
                else
                {
                    /* Generate coordinates within |dx|=drmax of x_init */
                    do
                    {
                        dx[XX] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
                        dx[YY] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
                        dx[ZZ] = (2*gmx_rng_uniform_real(tpi_rand) - 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
                {
                    dx[XX] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
                    dx[YY] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
                    dx[ZZ] = (2*gmx_rng_uniform_real(tpi_rand) - 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 */
                rotate_conf(a_tp1-a_tp0,state->x+a_tp0,NULL,
                            2*M_PI*gmx_rng_uniform_real(tpi_rand),
                            2*M_PI*gmx_rng_uniform_real(tpi_rand),
                            2*M_PI*gmx_rng_uniform_real(tpi_rand));
                /* Shift to the insertion location */
                for(i=a_tp0; i<a_tp1; i++)
                {
                    rvec_inc(state->x[i],x_tp);
                }
            }
            
            /* Check if this insertion belongs to this node */
            bOurStep = TRUE;
            if (PAR(cr))
            {
                switch (inputrec->eI)
                {
                case eiTPI:
                    bOurStep = ((step / inputrec->nstlist) % nnodes == cr->nodeid);
                    break;
                case eiTPIC:
                    bOurStep = (step % nnodes == cr->nodeid);
                    break;
                default:
                    gmx_fatal(FARGS,"Unknown integrator %s",ei_names[inputrec->eI]);
                }
            }
            if (bOurStep)
            {
                /* 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,&top_global->groups,
                         rerun_fr.box,state->x,&state->hist,
                         f,force_vir,mdatoms,enerd,fcd,
                         lambda,NULL,fr,NULL,mu_tot,t,NULL,NULL,FALSE,
                         GMX_FORCE_NONBONDED |
                         (bNS ? GMX_FORCE_NS | GMX_FORCE_DOLR : 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,rerun_fr.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] += dvdlcorr;

                epot = enerd->term[F_EPOT];
                bEnergyOutOfBounds = FALSE;
#if ( defined(GMX_IA32_SSE) || defined(GMX_X86_64_SSE) || defined(GMX_X86_64_SSE2) )
                /* 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,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(fr->adress_do_drift ||fr->adress_icor == eAdressICThermoForce ){
                    sum_UgembU[e++] += enerd->term[F_ADR_DELTU]*embU;
                }

                if (debug)
                {
                    fprintf(debug,"TPI %7d %12.5e %12.5f %12.5f %12.5f\n",
                            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,step);
                    sprintf(str2,"t: %f step %d ener: %f",t,step,epot);
                    write_sto_conf_mtop(str,str2,top_global,state->x,state->v,
                                        inputrec->ePBC,state->box);
                }
            }
        }
        
        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  */
    runtime_end(runtime);

    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);

    runtime->nsteps_done = frame*inputrec->nsteps;

    return 0;
}