/*
 * This file is part of the GROMACS molecular simulation package.
 *
 * Copyright 1991- The GROMACS Authors
 * and the project initiators Erik Lindahl, Berk Hess and David van der Spoel.
 * Consult the AUTHORS/COPYING files and https://www.gromacs.org for details.
 *
 * GROMACS is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public License
 * as published by the Free Software Foundation; either version 2.1
 * of the License, or (at your option) any later version.
 *
 * GROMACS is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with GROMACS; if not, see
 * https://www.gnu.org/licenses, or write to the Free Software Foundation,
 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
 *
 * If you want to redistribute modifications to GROMACS, please
 * consider that scientific software is very special. Version
 * control is crucial - bugs must be traceable. We will be happy to
 * consider code for inclusion in the official distribution, but
 * derived work must not be called official GROMACS. Details are found
 * in the README & COPYING files - if they are missing, get the
 * official version at https://www.gromacs.org.
 *
 * To help us fund GROMACS development, we humbly ask that you cite
 * the research papers on the package. Check out https://www.gromacs.org.
 */

/*! \internal \file
 *
 * \brief Implements the replica exchange routines.
 *
 * \author David van der Spoel <david.vanderspoel@icm.uu.se>
 * \author Mark Abraham <mark.j.abraham@gmail.com>
 * \ingroup module_mdrun
 */
#include "gmxpre.h"

#include "gromacs/utility/enumerationhelpers.h"
#include "replicaexchange.h"

#include "config.h"

#include <cmath>

#include <random>

#include "gromacs/domdec/collect.h"
#include "gromacs/gmxlib/network.h"
#include "gromacs/math/units.h"
#include "gromacs/math/vec.h"
#include "gromacs/mdrunutility/multisim.h"
#include "gromacs/mdtypes/commrec.h"
#include "gromacs/mdtypes/enerdata.h"
#include "gromacs/mdtypes/inputrec.h"
#include "gromacs/mdtypes/md_enums.h"
#include "gromacs/mdtypes/state.h"
#include "gromacs/random/threefry.h"
#include "gromacs/random/uniformintdistribution.h"
#include "gromacs/random/uniformrealdistribution.h"
#include "gromacs/utility/enumerationhelpers.h"
#include "gromacs/utility/fatalerror.h"
#include "gromacs/utility/pleasecite.h"
#include "gromacs/utility/smalloc.h"

//! Helps cut off probability values.
constexpr int c_probabilityCutoff = 100;

/* we don't bother evaluating if events are more rare than exp(-100) = 3.7x10^-44 */

//! Rank in the multisimulation
#define MSRANK(ms, nodeid) (nodeid)

//! Enum for replica exchange flavours
enum class ReplicaExchangeType : int
{
    Temperature,
    Lambda,
    EndSingle,
    TemperatureLambda,
    Count
};
/*! \brief Strings describing replica exchange flavours.
 *
 *  end_single_marker merely notes the end of single variable replica
 *  exchange. All types higher than it are multiple replica exchange
 *  methods.
 *
 * Eventually, should add 'pressure', 'temperature and pressure',
 *  'lambda_and_pressure', 'temperature_lambda_pressure'?; Let's wait
 *  until we feel better about the pressure control methods giving
 *  exact ensembles.  Right now, we assume constant pressure */
static const char* enumValueToString(ReplicaExchangeType enumValue)
{
    constexpr gmx::EnumerationArray<ReplicaExchangeType, const char*> replicateExchangeTypeNames = {
        "temperature", "lambda", "end_single_marker", "temperature and lambda"
    };
    return replicateExchangeTypeNames[enumValue];
}

//! Working data for replica exchange.
struct gmx_repl_ex
{
    //! Replica ID
    int repl;
    //! Total number of replica
    int nrepl;
    //! Temperature
    real temp;
    //! Replica exchange type from ReplicaExchangeType enum
    ReplicaExchangeType type;
    //! Quantity, e.g. temperature or lambda; first index is ere, second index is replica ID
    gmx::EnumerationArray<ReplicaExchangeType, real*> q;
    //! Use constant pressure and temperature
    gmx_bool bNPT;
    //! Replica pressures
    real* pres;
    //! Replica indices
    int* ind;
    //! Used for keeping track of all the replica swaps
    int* allswaps;
    //! Replica exchange interval (number of steps)
    int nst;
    //! Number of exchanges per interval
    int nex;
    //! Random seed
    int seed;
    //! Number of even and odd replica change attempts
    int nattempt[2];
    //! Sum of probabilities
    real* prob_sum;
    //! Number of moves between replicas i and j
    int** nmoves;
    //! i-th element of the array is the number of exchanges between replica i-1 and i
    int* nexchange;

    /*! \brief Helper arrays for replica exchange; allocated here
     * so they don't have to be allocated each time */
    //! \{
    int*      destinations;
    int**     cyclic;
    int**     order;
    int*      tmpswap;
    gmx_bool* incycle;
    gmx_bool* bEx;
    //! \}

    //! Helper arrays to hold the quantities that are exchanged.
    //! \{
    real*  prob;
    real*  Epot;
    real  *Ekin_store; //jawang Kinetic Store
    real  *Ekin; //jawang     Kinetic Term;
    real  *Epot_store; //jawang Potential Store for Epot+pot and Epot+wat
    real  *Epot_store2; //jawang Potential Store for Ewat
    real*  beta;
    real*  Vol;
    real** de;
    //! \}
};

// TODO We should add Doxygen here some time.
//! \cond

static gmx_bool repl_quantity(const gmx_multisim_t* ms, struct gmx_repl_ex* re, ReplicaExchangeType ere, real q)
{
    real*    qall;
    gmx_bool bDiff;
    int      s;

    snew(qall, ms->numSimulations_);
    qall[re->repl] = q;
    gmx_sum_sim(ms->numSimulations_, qall, ms);

    bDiff = FALSE;
    for (s = 1; s < ms->numSimulations_; s++)
    {
        if (fabs(qall[s] - qall[0]) > 1e-5) /* Each quantity normally has 2-4 significant digits */
        {
            bDiff = TRUE;
        }
    }

    if (bDiff)
    {
        /* Set the replica exchange type and quantities */
        re->type = ere;

        snew(re->q[ere], re->nrepl);
        for (s = 0; s < ms->numSimulations_; s++)
        {
            re->q[ere][s] = qall[s];
        }
    }
    sfree(qall);
    return bDiff;
}

gmx_repl_ex_t init_replica_exchange(FILE*                            fplog,
                                    const gmx_multisim_t*            ms,
                                    int                              numAtomsInSystem,
                                    const t_inputrec*                ir,
                                    const ReplicaExchangeParameters& replExParams)
{
    real                pres;
    int                 i, j;
    struct gmx_repl_ex* re;
    gmx_bool            bTemp;
    gmx_bool            bLambda = FALSE;

    fprintf(fplog, "\nInitializing Replica Exchange\n");

    if (!isMultiSim(ms) || ms->numSimulations_ == 1)
    {
        gmx_fatal(FARGS,
                  "Nothing to exchange with only one replica, maybe you forgot to set the "
                  "-multidir option of mdrun?");
    }
    if (replExParams.numExchanges < 0)
    {
        gmx_fatal(FARGS, "Replica exchange number of exchanges needs to be positive");
    }

    if (!EI_DYNAMICS(ir->eI))
    {
        gmx_fatal(FARGS, "Replica exchange is only supported by dynamical simulations");
        /* Note that PAR(cr) is defined by cr->nnodes > 1, which is
         * distinct from isMultiSim(ms). A multi-simulation only runs
         * with real MPI parallelism, but this does not imply PAR(cr)
         * is true!
         *
         * Since we are using a dynamical integrator, the only
         * decomposition is DD, so PAR(cr) and haveDDAtomOrdering(*cr) are
         * synonymous. The only way for cr->nnodes > 1 to be true is
         * if we are using DD. */
    }

    snew(re, 1);

    re->repl  = ms->simulationIndex_;
    re->nrepl = ms->numSimulations_;

    fprintf(fplog, "Repl  There are %d replicas:\n", re->nrepl);

    /* We only check that the number of atoms in the systms match.
     * This, of course, do not guarantee that the systems are the same,
     * but it does guarantee that we can perform replica exchange.
     */
    check_multi_int(fplog, ms, numAtomsInSystem, "the number of atoms", FALSE);
    check_multi_int(fplog, ms, static_cast<int>(ir->eI), "the integrator", FALSE);
    check_multi_int64(fplog, ms, ir->init_step + ir->nsteps, "init_step+nsteps", FALSE);
    const int nst = replExParams.exchangeInterval;
    check_multi_int64(
            fplog, ms, (ir->init_step + nst - 1) / nst, "first exchange step: init_step/-replex", FALSE);
    check_multi_int(fplog, ms, static_cast<int>(ir->etc), "the temperature coupling", FALSE);
    check_multi_int(fplog, ms, ir->opts.ngtc, "the number of temperature coupling groups", FALSE);
    check_multi_int(fplog, ms, static_cast<int>(ir->epc), "the pressure coupling", FALSE);
    check_multi_int(fplog, ms, static_cast<int>(ir->efep), "free energy", FALSE);
    check_multi_int(fplog, ms, ir->fepvals->n_lambda, "number of lambda states", FALSE);

    re->temp = ir->opts.ref_t[0];
    for (i = 1; (i < ir->opts.ngtc); i++)
    {
        if (ir->opts.ref_t[i] != re->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");
        }
    }

    re->type = ReplicaExchangeType::Count;
    bTemp    = repl_quantity(ms, re, ReplicaExchangeType::Temperature, re->temp);
    if (ir->efep != FreeEnergyPerturbationType::No)
    {
        bLambda = repl_quantity(
                ms, re, ReplicaExchangeType::Lambda, static_cast<real>(ir->fepvals->init_fep_state));
    }
    if (re->type == ReplicaExchangeType::Count) /* nothing was assigned */
    {
        gmx_fatal(FARGS,
                  "The properties of the %d systems are all the same, there is nothing to exchange",
                  re->nrepl);
    }
    if (bLambda && bTemp)
    {
        re->type = ReplicaExchangeType::TemperatureLambda;
    }

    if (bTemp)
    {
        please_cite(fplog, "Sugita1999a");
        if (ir->epc != PressureCoupling::No)
        {
            re->bNPT = TRUE;
            fprintf(fplog, "Repl  Using Constant Pressure REMD.\n");
            please_cite(fplog, "Okabe2001a");
        }
        if (ir->etc == TemperatureCoupling::Berendsen)
        {
            gmx_fatal(FARGS,
                      "REMD with the %s thermostat does not produce correct potential energy "
                      "distributions, consider using the %s thermostat instead",
                      enumValueToString(ir->etc),
                      enumValueToString(TemperatureCoupling::VRescale));
        }
    }
    if (bLambda)
    {
        if (ir->fepvals->delta_lambda != 0) /* check this? */
        {
            gmx_fatal(FARGS, "delta_lambda is not zero");
        }
    }
    if (re->bNPT)
    {
        snew(re->pres, re->nrepl);
        if (ir->epct == PressureCouplingType::SurfaceTension)
        {
            pres = ir->ref_p[ZZ][ZZ];
        }
        else
        {
            pres = 0;
            j    = 0;
            for (i = 0; i < DIM; i++)
            {
                if (ir->compress[i][i] != 0)
                {
                    pres += ir->ref_p[i][i];
                    j++;
                }
            }
            pres /= j;
        }
        re->pres[re->repl] = pres;
        gmx_sum_sim(re->nrepl, re->pres, ms);
    }

    /* Make an index for increasing replica order */
    /* only makes sense if one or the other is varying, not both!
       if both are varying, we trust the order the person gave. */
    snew(re->ind, re->nrepl);
    for (i = 0; i < re->nrepl; i++)
    {
        re->ind[i] = i;
    }

    if (re->type < ReplicaExchangeType::EndSingle)
    {

        for (i = 0; i < re->nrepl; i++)
        {
            for (j = i + 1; j < re->nrepl; j++)
            {
                if (re->q[re->type][re->ind[j]] < re->q[re->type][re->ind[i]])
                {
                    /* Unordered replicas are supposed to work, but there
                     * is still an issues somewhere.
                     * Note that at this point still re->ind[i]=i.
                     */
                    gmx_fatal(FARGS,
                              "Replicas with indices %d < %d have %ss %g > %g, please order your "
                              "replicas on increasing %s",
                              i,
                              j,
                              enumValueToString(re->type),
                              re->q[re->type][i],
                              re->q[re->type][j],
                              enumValueToString(re->type));
                }
                else if (re->q[re->type][re->ind[j]] == re->q[re->type][re->ind[i]])
                {
                    gmx_fatal(FARGS, "Two replicas have identical %ss", enumValueToString(re->type));
                }
            }
        }
    }

    /* keep track of all the swaps, starting with the initial placement. */
    snew(re->allswaps, re->nrepl);
    for (i = 0; i < re->nrepl; i++)
    {
        re->allswaps[i] = re->ind[i];
    }

    switch (re->type)
    {
        case ReplicaExchangeType::Temperature:
            fprintf(fplog, "\nReplica exchange in temperature\n");
            for (i = 0; i < re->nrepl; i++)
            {
                fprintf(fplog, " %5.1f", re->q[re->type][re->ind[i]]);
            }
            fprintf(fplog, "\n");
            break;
        case ReplicaExchangeType::Lambda:
            fprintf(fplog, "\nReplica exchange in lambda\n");
            for (i = 0; i < re->nrepl; i++)
            {
                fprintf(fplog, " %3d", static_cast<int>(re->q[re->type][re->ind[i]]));
            }
            fprintf(fplog, "\n");
            break;
        case ReplicaExchangeType::TemperatureLambda:
            fprintf(fplog, "\nReplica exchange in temperature and lambda state\n");
            for (i = 0; i < re->nrepl; i++)
            {
                fprintf(fplog, " %5.1f", re->q[ReplicaExchangeType::Temperature][re->ind[i]]);
            }
            fprintf(fplog, "\n");
            for (i = 0; i < re->nrepl; i++)
            {
                fprintf(fplog, " %5d", static_cast<int>(re->q[ReplicaExchangeType::Lambda][re->ind[i]]));
            }
            fprintf(fplog, "\n");
            break;
        default: gmx_incons("Unknown replica exchange quantity");
    }
    if (re->bNPT)
    {
        fprintf(fplog, "\nRepl  p");
        for (i = 0; i < re->nrepl; i++)
        {
            fprintf(fplog, " %5.2f", re->pres[re->ind[i]]);
        }

        for (i = 0; i < re->nrepl; i++)
        {
            if ((i > 0) && (re->pres[re->ind[i]] < re->pres[re->ind[i - 1]]))
            {
                fprintf(fplog,
                        "\nWARNING: The reference pressures decrease with increasing "
                        "temperatures\n\n");
                fprintf(stderr,
                        "\nWARNING: The reference pressures decrease with increasing "
                        "temperatures\n\n");
            }
        }
    }
    re->nst = nst;
    if (replExParams.randomSeed == -1)
    {
        if (isMasterSim(ms))
        {
            re->seed = static_cast<int>(gmx::makeRandomSeed());
        }
        else
        {
            re->seed = 0;
        }
        gmx_sumi_sim(1, &(re->seed), ms);
    }
    else
    {
        re->seed = replExParams.randomSeed;
    }
    fprintf(fplog, "\nReplica exchange interval: %d\n", re->nst);
    fprintf(fplog, "\nReplica random seed: %d\n", re->seed);

    re->nattempt[0] = 0;
    re->nattempt[1] = 0;

    snew(re->prob_sum, re->nrepl);
    snew(re->nexchange, re->nrepl);
    snew(re->nmoves, re->nrepl);
    for (i = 0; i < re->nrepl; i++)
    {
        snew(re->nmoves[i], re->nrepl);
    }
    fprintf(fplog, "Replica exchange information below: ex and x = exchange, pr = probability\n");

    /* generate space for the helper functions so we don't have to snew each time */

    snew(re->destinations, re->nrepl);
    snew(re->incycle, re->nrepl);
    snew(re->tmpswap, re->nrepl);
    snew(re->cyclic, re->nrepl);
    snew(re->order, re->nrepl);
    for (i = 0; i < re->nrepl; i++)
    {
        snew(re->cyclic[i], re->nrepl + 1);
        snew(re->order[i], re->nrepl);
    }
    /* allocate space for the functions storing the data for the replicas */
    /* not all of these arrays needed in all cases, but they don't take
       up much space, since the max size is nrepl**2 */
    snew(re->prob, re->nrepl);
    snew(re->bEx, re->nrepl);
    snew(re->beta, re->nrepl);
    snew(re->Vol, re->nrepl);
    snew(re->Epot, re->nrepl);
    snew(re->Epot_store, re->nrepl); //jawang
    snew(re->Epot_store2, re->nrepl); //jawang
    snew(re->Ekin_store, re->nrepl);//jawang
    snew(re->Ekin, re->nrepl); //jawang
    snew(re->de, re->nrepl);
    for (i = 0; i < re->nrepl; i++)
    {
        snew(re->de[i], re->nrepl);
    }
    re->nex = replExParams.numExchanges;
    return re;
}

static void exchange_reals(const gmx_multisim_t gmx_unused* ms, int gmx_unused b, real* v, int n)
{
    real* buf;
    int   i;

    if (v)
    {
        snew(buf, n);
#if GMX_MPI
        /*
           MPI_Sendrecv(v,  n*sizeof(real),MPI_BYTE,MSRANK(ms,b),0,
           buf,n*sizeof(real),MPI_BYTE,MSRANK(ms,b),0,
           ms->mastersComm_,MPI_STATUS_IGNORE);
         */
        {
            MPI_Request mpi_req;

            MPI_Isend(v, n * sizeof(real), MPI_BYTE, MSRANK(ms, b), 0, ms->mastersComm_, &mpi_req);
            MPI_Recv(buf, n * sizeof(real), MPI_BYTE, MSRANK(ms, b), 0, ms->mastersComm_, MPI_STATUS_IGNORE);
            MPI_Wait(&mpi_req, MPI_STATUS_IGNORE);
        }
#endif
        for (i = 0; i < n; i++)
        {
            v[i] = buf[i];
        }
        sfree(buf);
    }
}


static void exchange_doubles(const gmx_multisim_t gmx_unused* ms, int gmx_unused b, double* v, int n)
{
    double* buf;
    int     i;

    if (v)
    {
        snew(buf, n);
#if GMX_MPI
        /*
           MPI_Sendrecv(v,  n*sizeof(double),MPI_BYTE,MSRANK(ms,b),0,
           buf,n*sizeof(double),MPI_BYTE,MSRANK(ms,b),0,
           ms->mastersComm_,MPI_STATUS_IGNORE);
         */
        {
            MPI_Request mpi_req;

            MPI_Isend(v, n * sizeof(double), MPI_BYTE, MSRANK(ms, b), 0, ms->mastersComm_, &mpi_req);
            MPI_Recv(buf, n * sizeof(double), MPI_BYTE, MSRANK(ms, b), 0, ms->mastersComm_, MPI_STATUS_IGNORE);
            MPI_Wait(&mpi_req, MPI_STATUS_IGNORE);
        }
#endif
        for (i = 0; i < n; i++)
        {
            v[i] = buf[i];
        }
        sfree(buf);
    }
}

static void exchange_rvecs(const gmx_multisim_t gmx_unused* ms, int gmx_unused b, rvec* v, int n)
{
    rvec* buf;
    int   i;

    if (v)
    {
        snew(buf, n);
#if GMX_MPI
        /*
           MPI_Sendrecv(v[0],  n*sizeof(rvec),MPI_BYTE,MSRANK(ms,b),0,
           buf[0],n*sizeof(rvec),MPI_BYTE,MSRANK(ms,b),0,
           ms->mastersComm_,MPI_STATUS_IGNORE);
         */
        {
            MPI_Request mpi_req;

            MPI_Isend(v[0], n * sizeof(rvec), MPI_BYTE, MSRANK(ms, b), 0, ms->mastersComm_, &mpi_req);
            MPI_Recv(buf[0], n * sizeof(rvec), MPI_BYTE, MSRANK(ms, b), 0, ms->mastersComm_, MPI_STATUS_IGNORE);
            MPI_Wait(&mpi_req, MPI_STATUS_IGNORE);
        }
#endif
        for (i = 0; i < n; i++)
        {
            copy_rvec(buf[i], v[i]);
        }
        sfree(buf);
    }
}

static void exchange_state(const gmx_multisim_t* ms, int b, t_state* state)
{
    /* When t_state changes, this code should be updated. */
    int ngtc, nnhpres;
    ngtc    = state->ngtc * state->nhchainlength;
    nnhpres = state->nnhpres * state->nhchainlength;
    exchange_rvecs(ms, b, state->box, DIM);
    exchange_rvecs(ms, b, state->box_rel, DIM);
    exchange_rvecs(ms, b, state->boxv, DIM);
    exchange_reals(ms, b, &(state->veta), 1);
    exchange_reals(ms, b, &(state->vol0), 1);
    exchange_rvecs(ms, b, state->svir_prev, DIM);
    exchange_rvecs(ms, b, state->fvir_prev, DIM);
    exchange_rvecs(ms, b, state->pres_prev, DIM);
    exchange_doubles(ms, b, state->nosehoover_xi.data(), ngtc);
    exchange_doubles(ms, b, state->nosehoover_vxi.data(), ngtc);
    exchange_doubles(ms, b, state->nhpres_xi.data(), nnhpres);
    exchange_doubles(ms, b, state->nhpres_vxi.data(), nnhpres);
    exchange_doubles(ms, b, state->therm_integral.data(), state->ngtc);
    exchange_doubles(ms, b, &state->baros_integral, 1);
    exchange_rvecs(ms, b, state->x.rvec_array(), state->natoms);
    exchange_rvecs(ms, b, state->v.rvec_array(), state->natoms);
}

static void copy_state_serial(const t_state* src, t_state* dest)
{
    if (dest != src)
    {
        /* Currently the local state is always a pointer to the global
         * in serial, so we should never end up here.
         * TODO: Implement a (trivial) t_state copy once converted to C++.
         */
        GMX_RELEASE_ASSERT(false, "State copying is currently not implemented in replica exchange");
    }
}

static void scale_velocities(gmx::ArrayRef<gmx::RVec> velocities, real fac)
{
    for (auto& v : velocities)
    {
        v *= fac;
    }
}

static void print_transition_matrix(FILE* fplog, int n, int** nmoves, const int* nattempt)
{
    int   i, j, ntot;
    float Tprint;

    ntot = nattempt[0] + nattempt[1];
    fprintf(fplog, "\n");
    fprintf(fplog, "Repl");
    for (i = 0; i < n; i++)
    {
        fprintf(fplog, "    "); /* put the title closer to the center */
    }
    fprintf(fplog, "Empirical Transition Matrix\n");

    fprintf(fplog, "Repl");
    for (i = 0; i < n; i++)
    {
        fprintf(fplog, "%8d", (i + 1));
    }
    fprintf(fplog, "\n");

    for (i = 0; i < n; i++)
    {
        fprintf(fplog, "Repl");
        for (j = 0; j < n; j++)
        {
            Tprint = 0.0;
            if (nmoves[i][j] > 0)
            {
                Tprint = nmoves[i][j] / (2.0 * ntot);
            }
            fprintf(fplog, "%8.4f", Tprint);
        }
        fprintf(fplog, "%3d\n", i);
    }
}

static void print_ind(FILE* fplog, const char* leg, int n, int* ind, const gmx_bool* bEx)
{
    int i;

    fprintf(fplog, "Repl %2s %2d", leg, ind[0]);
    for (i = 1; i < n; i++)
    {
        fprintf(fplog, " %c %2d", (bEx != nullptr && bEx[i]) ? 'x' : ' ', ind[i]);
    }
    fprintf(fplog, "\n");
}

static void print_allswitchind(FILE* fplog, int n, int* pind, int* allswaps, int* tmpswap)
{
    int i;

    for (i = 0; i < n; i++)
    {
        tmpswap[i] = allswaps[i];
    }
    for (i = 0; i < n; i++)
    {
        allswaps[i] = tmpswap[pind[i]];
    }

    fprintf(fplog, "\nAccepted Exchanges:   ");
    for (i = 0; i < n; i++)
    {
        fprintf(fplog, "%d ", pind[i]);
    }
    fprintf(fplog, "\n");

    /* the "Order After Exchange" is the state label corresponding to the configuration that
       started in state listed in order, i.e.

       3 0 1 2

       means that the:
       configuration starting in simulation 3 is now in simulation 0,
       configuration starting in simulation 0 is now in simulation 1,
       configuration starting in simulation 1 is now in simulation 2,
       configuration starting in simulation 2 is now in simulation 3
     */
    fprintf(fplog, "Order After Exchange: ");
    for (i = 0; i < n; i++)
    {
        fprintf(fplog, "%d ", allswaps[i]);
    }
    fprintf(fplog, "\n\n");
}

static void print_prob(FILE* fplog, const char* leg, int n, real* prob)
{
    int  i;
    char buf[8];

    fprintf(fplog, "Repl %2s ", leg);
    for (i = 1; i < n; i++)
    {
        if (prob[i] >= 0)
        {
            sprintf(buf, "%4.2f", prob[i]);
            fprintf(fplog, "  %3s", buf[0] == '1' ? "1.0" : buf + 1);
        }
        else
        {
            fprintf(fplog, "     ");
        }
    }
    fprintf(fplog, "\n");
}

static void print_count(FILE* fplog, const char* leg, int n, int* count)
{
    int i;

    fprintf(fplog, "Repl %2s ", leg);
    for (i = 1; i < n; i++)
    {
        fprintf(fplog, " %4d", count[i]);
    }
    fprintf(fplog, "\n");
}

static real calc_delta(FILE* fplog, gmx_bool bPrint, struct gmx_repl_ex* re, int a, int b, int ap, int bp)
{

    real   ediff, dpV, delta = 0;
    real*  Epot = re->Epot;
    real*  Vol  = re->Vol;
    real** de   = re->de;
    real*  beta = re->beta;
    real  *Epot_store = re->Epot_store; //jawang
    real  *Epot_store2= re->Epot_store2;//jawang

    /* Two cases; we are permuted and not.  In all cases, setting ap = a and bp = b will reduce
       to the non permuted case */

    switch (re->type)
    {
        case ReplicaExchangeType::Temperature:
            /*
             * Okabe et. al. Chem. Phys. Lett. 335 (2001) 435-439
             */
//            ediff = Epot[b] - Epot[a];
            ediff = Epot_store[b] - Epot_store[a];  //jawang
            delta = -(beta[bp] - beta[ap]) * ediff;
            break;
        case ReplicaExchangeType::Lambda:
            /* two cases:  when we are permuted, and not.  */
            /* non-permuted:
               ediff =  E_new - E_old
                     =  [H_b(x_a) + H_a(x_b)] - [H_b(x_b) + H_a(x_a)]
                     =  [H_b(x_a) - H_a(x_a)] + [H_a(x_b) - H_b(x_b)]
                     =  de[b][a] + de[a][b] */

            /* permuted:
               ediff =  E_new - E_old
                     =  [H_bp(x_a) + H_ap(x_b)] - [H_bp(x_b) + H_ap(x_a)]
                     =  [H_bp(x_a) - H_ap(x_a)] + [H_ap(x_b) - H_bp(x_b)]
                     =  [H_bp(x_a) - H_a(x_a) + H_a(x_a) - H_ap(x_a)] + [H_ap(x_b) - H_b(x_b) + H_b(x_b) - H_bp(x_b)]
                     =  [H_bp(x_a) - H_a(x_a)] - [H_ap(x_a) - H_a(x_a)] + [H_ap(x_b) - H_b(x_b)] - H_bp(x_b) - H_b(x_b)]
                     =  (de[bp][a] - de[ap][a]) + (de[ap][b] - de[bp][b])    */
            /* but, in the current code implementation, we flip configurations, not indices . . .
               So let's examine that.
                     =  [H_b(x_ap) - H_a(x_a)] - [H_a(x_ap) - H_a(x_a)] + [H_a(x_bp) - H_b(x_b)] - H_b(x_bp) - H_b(x_b)]
                     =  [H_b(x_ap) - H_a(x_ap)]  + [H_a(x_bp) - H_b(x_pb)]
                     = (de[b][ap] - de[a][ap]) + (de[a][bp] - de[b][bp]
                     So, if we exchange b<=> bp and a<=> ap, we return to the same result.
                     So the simple solution is to flip the
                     position of perturbed and original indices in the tests.
             */

            ediff = (de[bp][a] - de[ap][a]) + (de[ap][b] - de[bp][b]);
            delta = ediff * beta[a]; /* assume all same temperature in this case */
            break;
        case ReplicaExchangeType::TemperatureLambda:
            /* not permuted:  */
            /* delta =  reduced E_new - reduced E_old
                     =  [beta_b H_b(x_a) + beta_a H_a(x_b)] - [beta_b H_b(x_b) + beta_a H_a(x_a)]
                     =  [beta_b H_b(x_a) - beta_a H_a(x_a)] + [beta_a H_a(x_b) - beta_b H_b(x_b)]
                     =  [beta_b dH_b(x_a) + beta_b H_a(x_a) - beta_a H_a(x_a)] +
                        [beta_a dH_a(x_b) + beta_a H_b(x_b) - beta_b H_b(x_b)]
                     =  [beta_b dH_b(x_a) + [beta_a dH_a(x_b) +
                        beta_b (H_a(x_a) - H_b(x_b)]) - beta_a (H_a(x_a) - H_b(x_b))
                     =  beta_b dH_b(x_a) + beta_a dH_a(x_b) - (beta_b - beta_a)(H_b(x_b) - H_a(x_a) */
            /* delta = beta[b]*de[b][a] + beta[a]*de[a][b] - (beta[b] - beta[a])*(Epot[b] - Epot[a]; */
            /* permuted (big breath!) */
            /*   delta =  reduced E_new - reduced E_old
                     =  [beta_bp H_bp(x_a) + beta_ap H_ap(x_b)] - [beta_bp H_bp(x_b) + beta_ap H_ap(x_a)]
                     =  [beta_bp H_bp(x_a) - beta_ap H_ap(x_a)] + [beta_ap H_ap(x_b) - beta_bp H_bp(x_b)]
                     =  [beta_bp H_bp(x_a) - beta_ap H_ap(x_a)] + [beta_ap H_ap(x_b) - beta_bp H_bp(x_b)]
                        - beta_pb H_a(x_a) + beta_ap H_a(x_a) + beta_pb H_a(x_a) - beta_ap H_a(x_a)
                        - beta_ap H_b(x_b) + beta_bp H_b(x_b) + beta_ap H_b(x_b) - beta_bp H_b(x_b)
                     =  [(beta_bp H_bp(x_a) - beta_bp H_a(x_a)) - (beta_ap H_ap(x_a) - beta_ap H_a(x_a))] +
                        [(beta_ap H_ap(x_b)  - beta_ap H_b(x_b)) - (beta_bp H_bp(x_b) - beta_bp H_b(x_b))]
             + beta_pb H_a(x_a) - beta_ap H_a(x_a) + beta_ap H_b(x_b) - beta_bp H_b(x_b)
                     =  [beta_bp (H_bp(x_a) - H_a(x_a)) - beta_ap (H_ap(x_a) - H_a(x_a))] +
                        [beta_ap (H_ap(x_b) - H_b(x_b)) - beta_bp (H_bp(x_b) - H_b(x_b))]
             + beta_pb (H_a(x_a) - H_b(x_b))  - beta_ap (H_a(x_a) - H_b(x_b))
                     =  ([beta_bp de[bp][a] - beta_ap de[ap][a]) + beta_ap de[ap][b]  - beta_bp de[bp][b])
             + (beta_pb-beta_ap)(H_a(x_a) - H_b(x_b))  */
            delta = beta[bp] * (de[bp][a] - de[bp][b]) + beta[ap] * (de[ap][b] - de[ap][a])
                    - (beta[bp] - beta[ap]) * (Epot[b] - Epot[a]);
            break;
        default: gmx_incons("Unknown replica exchange quantity");
    }
    if (bPrint)
    {
        fprintf(fplog, "Repl %d <-> %d  dE_term = %10.3e (kT)\n", a, b, delta);
    }
    if (re->bNPT)
    {
        /* revist the calculation for 5.0.  Might be some improvements. */
        dpV = (beta[ap] * re->pres[ap] - beta[bp] * re->pres[bp]) * (Vol[b] - Vol[a]) / gmx::c_presfac;
        if (bPrint)
        {
            fprintf(fplog, "  dpV = %10.3e  d = %10.3e\n", dpV, delta + dpV);
        }
        delta += dpV;
    }
    return delta;
}

static void test_for_replica_exchange(FILE*                 fplog,
                                      const gmx_multisim_t* ms,
                                      struct gmx_repl_ex*   re,
                                      const gmx_enerdata_t* enerd,
                                      real                  vol,
                                      int64_t               step,
                                      real                  time,
                                      real                  *nscale)//jawang
{
    int                                m, i, j, a, b, ap, bp, i0, i1, tmp;
    real                               delta = 0;
    gmx_bool                           bPrint, bMultiEx;
    gmx_bool*                          bEx      = re->bEx;
    real*                              prob     = re->prob;
    int*                               pind     = re->destinations; /* permuted index */
    gmx_bool                           bEpot    = FALSE;
    gmx_bool                           bDLambda = FALSE;
    gmx_bool                           bVol     = FALSE;
    gmx::ThreeFry2x64<64>              rng(re->seed, gmx::RandomDomain::ReplicaExchange);
    gmx::UniformRealDistribution<real> uniformRealDist;
    gmx::UniformIntDistribution<int>   uniformNreplDist(0, re->nrepl - 1);

    bMultiEx = (re->nex > 1); /* multiple exchanges at each state */
    fprintf(fplog, "Replica exchange at step %" PRId64 " time %.5f\n", step, time);

    if (re->bNPT)
    {
        for (i = 0; i < re->nrepl; i++)
        {
            re->Vol[i] = 0;
        }
        bVol              = TRUE;
        re->Vol[re->repl] = vol;
    }
    if ((re->type == ReplicaExchangeType::Temperature || re->type == ReplicaExchangeType::TemperatureLambda))
    {
        for (i = 0; i < re->nrepl; i++)
        {
            re->Epot[i] = 0;
            re->Ekin[i] = 0;  //jawang
            re->Epot_store[i] = 0;//jawang
            re->Epot_store2[i] = 0;  //jawang
            re->Ekin_store[i] = 0;  //jawang
        }
        bEpot              = TRUE;
        re->Epot[re->repl] = enerd->term[F_EPOT];
        re->Ekin[re->repl] = enerd->term[F_EKIN]; //jawang
//added by jawang for test which group energy we need
//enerd->grpp.ener[0][0].equil."Coul-SR-Protein-Protein"
//enerd->grpp.ener[0][1].equil."Coul-SR-Protein-SOL"
//enerd->grpp.ener[0][3].equil."Coul-SR-SOL-SOL"
//enerd->grpp.ener[1][0].equil."LJ-SR-Protein-Protein"
//enerd->grpp.ener[1][1].equil."LJ-SR-Protein-SOL"
//enerd->grpp.ener[1][3].equil."LJ-SR-SOL-SOL"
//enerd->grpp.ener[6][0].equil."Coul-14-Protein-Protein"
//enerd->grpp.ener[7][0].equil."LJ-14-Protein-Protein"
//enerd->term[F_DISPCORR] equil."Disper.-corr." belong to the water
//enerd->term[F_COUL_RECIP].equil"Coul.-recip." belong to the water
//enerd->term[F_EKIN] contans the kinate energy
// because the bond angle potention in TIP3P is zero 
//        re->Epot_store[re->repl] = enerd->term[F_EPOT]-enerd->grpp.ener[0][3]-enerd->grpp.ener[1][3]-enerd->term[F_COUL_RECIP]-enerd->term[F_DISPCORR]; //jawang only the sum of pro-pro and pro-wat interaction,because the TIP3P no bond anlge interaction
//         fprintf (fplog,"test xbzhang coulombsr[3]=%f, LJSR[3]=%f, CoulombSR[1]=%f, LJSR[1]=%f,term[F_COUL_RECIP]=%f,term[F_DISPCORR]=%f,term[F_EPOT]=%f\n", enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR][3],enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR][3],enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR][1],enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR][1],enerd->term[F_COUL_RECIP],enerd->term[F_DISPCORR],enerd->term[F_EPOT]);
         re->Epot_store[re->repl] = enerd->term[F_EPOT]-enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR][3]-enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR][3]-enerd->term[F_COUL_RECIP]-enerd->term[F_DISPCORR]-0.5*(enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR][1]+enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR][1]);//only the sum of pro-pro and 0.5*pro-wat interaction
//        re->Epot_store2[re->repl] = enerd->grpp.ener[0][3]+enerd->grpp.ener[1][3]+enerd->term[F_COUL_RECIP]+enerd->term[F_DISPCORR];//the water-water interaction contain the coul-recip and dispcorr interaction for all pro-pro and pro-wat
        re->Epot_store2[re->repl] = enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR][3]+enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR][3]+enerd->term[F_COUL_RECIP]+enerd->term[F_DISPCORR]+0.5*(enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR][1]+enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR][1]);//only the sum of pro-pro and 0.5*pro-wat interaction
        re->Ekin_store[re->repl] = enerd->term[F_EKIN]; //jawang

        /* temperatures of different states*/
        for (i = 0; i < re->nrepl; i++)
        {
            re->beta[i] = 1.0 / (re->q[ReplicaExchangeType::Temperature][i] * gmx::c_boltz);
        }
    }
    else
    {
        for (i = 0; i < re->nrepl; i++)
        {
            re->beta[i] = 1.0 / (re->temp * gmx::c_boltz); /* we have a single temperature */
        }
    }
    if (re->type == ReplicaExchangeType::Lambda || re->type == ReplicaExchangeType::TemperatureLambda)
    {
        bDLambda = TRUE;
        /* lambda differences. */
        /* de[i][j] is the energy of the jth simulation in the ith Hamiltonian
           minus the energy of the jth simulation in the jth Hamiltonian */
        for (i = 0; i < re->nrepl; i++)
        {
            for (j = 0; j < re->nrepl; j++)
            {
                re->de[i][j] = 0;
            }
        }
        for (i = 0; i < re->nrepl; i++)
        {
            re->de[i][re->repl] =
                    enerd->foreignLambdaTerms.deltaH(re->q[ReplicaExchangeType::Lambda][i]);
        }
    }

    /* now actually do the communication */
     gmx_sum_sim(re->nrepl, re->Ekin_store, ms); //jawang
     gmx_sum_sim(re->nrepl, re->Epot_store2, ms); //jawang
     gmx_sum_sim(re->nrepl, re->Ekin, ms); //jawang
     gmx_sum_sim(re->nrepl, re->Epot_store, ms); //jawang
    if (bVol)
    {
        gmx_sum_sim(re->nrepl, re->Vol, ms);
    }
    if (bEpot)
    {
        gmx_sum_sim(re->nrepl, re->Epot, ms);
    }
    if (bDLambda)
    {
        for (i = 0; i < re->nrepl; i++)
        {
            gmx_sum_sim(re->nrepl, re->de[i], ms);
        }
    }

    /* make a duplicate set of indices for shuffling */
    for (i = 0; i < re->nrepl; i++)
    {
        pind[i] = re->ind[i];
    }

    rng.restart(step, 0);

    if (bMultiEx)
    {
        /* multiple random switch exchange */
        int nself = 0;


        for (i = 0; i < re->nex + nself; i++)
        {
            // For now this is superfluous, but just in case we ever add more
            // calls in different branches it is safer to always reset the distribution.
            uniformNreplDist.reset();

            /* randomly select a pair  */
            /* in theory, could reduce this by identifying only which switches had a nonneglibible
               probability of occurring (log p > -100) and only operate on those switches */
            /* find out which state it is from, and what label that state currently has. Likely
               more work that useful. */
            i0 = uniformNreplDist(rng);
            i1 = uniformNreplDist(rng);
            if (i0 == i1)
            {
                nself++;
                continue; /* self-exchange, back up and do it again */
            }

            a  = re->ind[i0]; /* what are the indices of these states? */
            b  = re->ind[i1];
            ap = pind[i0];
            bp = pind[i1];

            bPrint = FALSE; /* too noisy */
            /* calculate the energy difference */
            /* if the code changes to flip the STATES, rather than the configurations,
               use the commented version of the code */
            /* delta = calc_delta(fplog,bPrint,re,a,b,ap,bp); */
            delta = calc_delta(fplog, bPrint, re, ap, bp, a, b);

            /* we actually only use the first space in the prob and bEx array,
               since there are actually many switches between pairs. */

            if (delta <= 0)
            {
                /* accepted */
                prob[0] = 1;
                bEx[0]  = TRUE;
            }
            else
            {
                if (delta > c_probabilityCutoff)
                {
                    prob[0] = 0;
                }
                else
                {
                    prob[0] = exp(-delta);
                }
                // roll a number to determine if accepted. For now it is superfluous to
                // reset, but just in case we ever add more calls in different branches
                // it is safer to always reset the distribution.
                uniformRealDist.reset();
                bEx[0] = uniformRealDist(rng) < prob[0];
            }
            re->prob_sum[0] += prob[0];

            if (bEx[0])
            {
                /* swap the states */
                tmp      = pind[i0];
                pind[i0] = pind[i1];
                pind[i1] = tmp;
            }
        }
        re->nattempt[0]++; /* keep track of total permutation trials here */
        print_allswitchind(fplog, re->nrepl, pind, re->allswaps, re->tmpswap);
    }
    else
    {
        /* standard nearest neighbor replica exchange */

        m = (step / re->nst) % 2;
        for (i = 1; i < re->nrepl; i++)
        {
            a = re->ind[i - 1];
            b = re->ind[i];

            bPrint = (re->repl == a || re->repl == b);
            if (i % 2 == m)
            {
                delta = calc_delta(fplog, bPrint, re, a, b, a, b);
                if (delta <= 0)
                {
                    /* accepted */
                    prob[i] = 1;
                    bEx[i]  = TRUE;
                }
                else
                {
                    if (delta > c_probabilityCutoff)
                    {
                        prob[i] = 0;
                    }
                    else
                    {
                        prob[i] = exp(-delta);
                    }
                    // roll a number to determine if accepted. For now it is superfluous to
                    // reset, but just in case we ever add more calls in different branches
                    // it is safer to always reset the distribution.
                    uniformRealDist.reset();
                    bEx[i] = uniformRealDist(rng) < prob[i];
                }
                re->prob_sum[i] += prob[i];

           
//                fprintf (fplog,"test xbzhang delta=%f, prob=%f\n", delta,prob[i]);
//                fprintf (fplog,"test xbzhang Epota %f Ekina %f ,Epot_store2 %f Ekinb %f Epotb %f\n", re->Epot[a],re->Ekin[a],re->Epot_store2[b],re->Ekin[b],re->Epot[b]);
//                fprintf (fplog,"test xbzhang Epotstore,re->repl= %d is %f\n", re->repl, re->Epot_store[a]);
//                fprintf (fplog,"test xbzhang Epotstore2,re->repl= %d is %f\n", re->repl, re->Epot_store2[a]);

                if (bEx[i])
                {
                if(a == re->repl) //jawang
                  {
                     *nscale = (re->Epot_store2[a]+re->Ekin[a]-re->Epot_store2[b])/re->Ekin[b];//jawang
                    fprintf (fplog,"Epota %f Ekina %f ,Epot_store2 %f Ekinb %f Epotb %f\n", re->Epot[a],re->Ekin[a],re->Epot_store2[b],re->Ekin[b],re->Epot[b]);
                    fprintf (fplog,"Epotstore,re->repl= %d is %f\n", re->repl, re->Epot_store[a]);
                    fprintf (fplog,"Epotstore2,re->repl= %d is %f\n", re->repl, re->Epot_store2[a]);
                     
                  }
                if(b == re->repl) //jawang
                  {
                     *nscale = (re->Epot_store2[b]+re->Ekin[b]-re->Epot_store2[a])/re->Ekin[a];//jawang
                    fprintf (fplog,"Epotstore,re->repl= %d is %f\n", re->repl, re->Epot_store[b]);
                    fprintf (fplog,"Epotstore2,re->repl= %d is %f\n", re->repl, re->Epot_store2[b]);
                  }
               if((re->Epot_store2[b]+re->Ekin[b]-re->Epot_store2[a]) <=0 || (re->Epot_store2[a]+re->Ekin[a]-re->Epot_store2[b]) <=0)//FY
                {
                  prob[i] = -1;
                  bEx[i]=FALSE;
               }
                    /* swap these two */
                    tmp         = pind[i - 1];
                    pind[i - 1] = pind[i];
                    pind[i]     = tmp;
                    re->nexchange[i]++; /* statistics for back compatibility */
                }
            }
            else
            {
                prob[i] = -1;
                bEx[i]  = FALSE;
            }
        }
        /* print some statistics */
        print_ind(fplog, "ex", re->nrepl, re->ind, bEx);
        print_prob(fplog, "pr", re->nrepl, prob);
        fprintf(fplog, "\n");
        re->nattempt[m]++;
    }

    /* record which moves were made and accepted */
    for (i = 0; i < re->nrepl; i++)
    {
        re->nmoves[re->ind[i]][pind[i]] += 1;
        re->nmoves[pind[i]][re->ind[i]] += 1;
    }
    fflush(fplog); /* make sure we can see what the last exchange was */
}

static void cyclic_decomposition(const int* destinations, int** cyclic, gmx_bool* incycle, const int nrepl, int* nswap)
{

    int i, j, c, p;
    int maxlen = 1;
    for (i = 0; i < nrepl; i++)
    {
        incycle[i] = FALSE;
    }
    for (i = 0; i < nrepl; i++) /* one cycle for each replica */
    {
        if (incycle[i])
        {
            cyclic[i][0] = -1;
            continue;
        }
        cyclic[i][0] = i;
        incycle[i]   = TRUE;
        c            = 1;
        p            = i;
        for (j = 0; j < nrepl; j++) /* potentially all cycles are part, but we will break first */
        {
            p = destinations[p]; /* start permuting */
            if (p == i)
            {
                cyclic[i][c] = -1;
                if (c > maxlen)
                {
                    maxlen = c;
                }
                break; /* we've reached the original element, the cycle is complete, and we marked the end. */
            }
            else
            {
                cyclic[i][c] = p; /* each permutation gives a new member of the cycle */
                incycle[p]   = TRUE;
                c++;
            }
        }
    }
    *nswap = maxlen - 1;

    if (debug)
    {
        for (i = 0; i < nrepl; i++)
        {
            fprintf(debug, "Cycle %d:", i);
            for (j = 0; j < nrepl; j++)
            {
                if (cyclic[i][j] < 0)
                {
                    break;
                }
                fprintf(debug, "%2d", cyclic[i][j]);
            }
            fprintf(debug, "\n");
        }
        fflush(debug);
    }
}

static void compute_exchange_order(int** cyclic, int** order, const int nrepl, const int maxswap)
{
    int i, j;

    for (j = 0; j < maxswap; j++)
    {
        for (i = 0; i < nrepl; i++)
        {
            if (cyclic[i][j + 1] >= 0)
            {
                order[cyclic[i][j + 1]][j] = cyclic[i][j];
                order[cyclic[i][j]][j]     = cyclic[i][j + 1];
            }
        }
        for (i = 0; i < nrepl; i++)
        {
            if (order[i][j] < 0)
            {
                order[i][j] = i; /* if it's not exchanging, it should stay this round*/
            }
        }
    }

    if (debug)
    {
        fprintf(debug, "Replica Exchange Order\n");
        for (i = 0; i < nrepl; i++)
        {
            fprintf(debug, "Replica %d:", i);
            for (j = 0; j < maxswap; j++)
            {
                if (order[i][j] < 0)
                {
                    break;
                }
                fprintf(debug, "%2d", order[i][j]);
            }
            fprintf(debug, "\n");
        }
        fflush(debug);
    }
}

static void prepare_to_do_exchange(struct gmx_repl_ex* re, const int replica_id, int* maxswap, gmx_bool* bThisReplicaExchanged)
{
    int i, j;
    /* Hold the cyclic decomposition of the (multiple) replica
     * exchange. */
    gmx_bool bAnyReplicaExchanged = FALSE;
    *bThisReplicaExchanged        = FALSE;

    for (i = 0; i < re->nrepl; i++)
    {
        if (re->destinations[i] != re->ind[i])
        {
            /* only mark as exchanged if the index has been shuffled */
            bAnyReplicaExchanged = TRUE;
            break;
        }
    }
    if (bAnyReplicaExchanged)
    {
        /* reinitialize the placeholder arrays */
        for (i = 0; i < re->nrepl; i++)
        {
            for (j = 0; j < re->nrepl; j++)
            {
                re->cyclic[i][j] = -1;
                re->order[i][j]  = -1;
            }
        }

        /* Identify the cyclic decomposition of the permutation (very
         * fast if neighbor replica exchange). */
        cyclic_decomposition(re->destinations, re->cyclic, re->incycle, re->nrepl, maxswap);

        /* Now translate the decomposition into a replica exchange
         * order at each step. */
        compute_exchange_order(re->cyclic, re->order, re->nrepl, *maxswap);

        /* Did this replica do any exchange at any point? */
        for (j = 0; j < *maxswap; j++)
        {
            if (replica_id != re->order[replica_id][j])
            {
                *bThisReplicaExchanged = TRUE;
                break;
            }
        }
    }
}

gmx_bool replica_exchange(FILE*                 fplog,
                          const t_commrec*      cr,
                          const gmx_multisim_t* ms,
                          struct gmx_repl_ex*   re,
                          t_state*              state,
                          const gmx_enerdata_t* enerd,
                          t_state*              state_local,
                          int64_t               step,
                          real                  time)
{
    int j;
    int replica_id = 0;
    int exchange_partner;
    int maxswap = 0;
    /* Number of rounds of exchanges needed to deal with any multiple
     * exchanges. */
    /* Where each replica ends up after the exchange attempt(s). */
    /* The order in which multiple exchanges will occur. */
    gmx_bool bThisReplicaExchanged = FALSE;
    real nscale=1.0; //jawang

    if (MASTER(cr))
    {
        replica_id = re->repl;
        test_for_replica_exchange(fplog, ms, re, enerd, det(state_local->box), step, time,&nscale);
        prepare_to_do_exchange(re, replica_id, &maxswap, &bThisReplicaExchanged);
    }
    /* Do intra-simulation broadcast so all processors belonging to
     * each simulation know whether they need to participate in
     * collecting the state. Otherwise, they might as well get on with
     * the next thing to do. */
    if (haveDDAtomOrdering(*cr))
    {
#if GMX_MPI
        MPI_Bcast(&bThisReplicaExchanged, sizeof(gmx_bool), MPI_BYTE, MASTERRANK(cr), cr->mpi_comm_mygroup);
#endif
    }

    if (bThisReplicaExchanged)
    {
        /* Exchange the states */
        /* Collect the global state on the master node */
        if (haveDDAtomOrdering(*cr))
        {
            dd_collect_state(cr->dd, state_local, state);
        }
        else
        {
            copy_state_serial(state_local, state);
        }

        if (MASTER(cr))
        {
            /* There will be only one swap cycle with standard replica
             * exchange, but there may be multiple swap cycles if we
             * allow multiple swaps. */

            for (j = 0; j < maxswap; j++)
            {
                exchange_partner = re->order[replica_id][j];

                if (exchange_partner != replica_id)
                {
                    /* Exchange the global states between the master nodes */
                    if (debug)
                    {
                        fprintf(debug, "Exchanging %d with %d\n", replica_id, exchange_partner);
                    }
                    exchange_state(ms, exchange_partner, state);
                }
            }
            /* For temperature-type replica exchange, we need to scale
             * the velocities. */
            if (re->type == ReplicaExchangeType::Temperature || re->type == ReplicaExchangeType::TemperatureLambda)
            {
             fprintf (fplog,"SCALE:re->repl= %d step=%lld nscale= %f\n", re->repl, step, nscale); //jawang
             scale_velocities(state->v, sqrt(nscale)); //jawang                 
/*                scale_velocities(
                        state->v,
                        std::sqrt(re->q[ReplicaExchangeType::Temperature][replica_id]
                                  / re->q[ReplicaExchangeType::Temperature][re->destinations[replica_id]]));*/
            }
        }

        /* With domain decomposition the global state is distributed later */
        if (!haveDDAtomOrdering(*cr))
        {
            /* Copy the global state to the local state data structure */
            copy_state_serial(state, state_local);
        }
    }

    return bThisReplicaExchanged;
}

void print_replica_exchange_statistics(FILE* fplog, struct gmx_repl_ex* re)
{
    int i;

    fprintf(fplog, "\nReplica exchange statistics\n");

    if (re->nex == 0)
    {
        fprintf(fplog,
                "Repl  %d attempts, %d odd, %d even\n",
                re->nattempt[0] + re->nattempt[1],
                re->nattempt[1],
                re->nattempt[0]);

        fprintf(fplog, "Repl  average probabilities:\n");
        for (i = 1; i < re->nrepl; i++)
        {
            if (re->nattempt[i % 2] == 0)
            {
                re->prob[i] = 0;
            }
            else
            {
                re->prob[i] = re->prob_sum[i] / re->nattempt[i % 2];
            }
        }
        print_ind(fplog, "", re->nrepl, re->ind, nullptr);
        print_prob(fplog, "", re->nrepl, re->prob);

        fprintf(fplog, "Repl  number of exchanges:\n");
        print_ind(fplog, "", re->nrepl, re->ind, nullptr);
        print_count(fplog, "", re->nrepl, re->nexchange);

        fprintf(fplog, "Repl  average number of exchanges:\n");
        for (i = 1; i < re->nrepl; i++)
        {
            if (re->nattempt[i % 2] == 0)
            {
                re->prob[i] = 0;
            }
            else
            {
                re->prob[i] = (static_cast<real>(re->nexchange[i])) / re->nattempt[i % 2];
            }
        }
        print_ind(fplog, "", re->nrepl, re->ind, nullptr);
        print_prob(fplog, "", re->nrepl, re->prob);

        fprintf(fplog, "\n");
    }
    /* print the transition matrix */
    print_transition_matrix(fplog, re->nrepl, re->nmoves, re->nattempt);
}

//! \endcond
