/* ----------------------------------------------------------------------
   LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
   https://www.lammps.org/, Sandia National Laboratories
   LAMMPS development team: developers@lammps.org

   Copyright (2003) Sandia Corporation.  Under the terms of Contract
   DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
   certain rights in this software.  This software is distributed under
   the GNU General Public License.

   See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */

#include "pair_lj_class2_coul_long.h"

#include "atom.h"
#include "comm.h"
#include "error.h"
#include "force.h"
#include "kspace.h"
#include "math_const.h"
#include "memory.h"
#include "neigh_list.h"
#include "neighbor.h"
#include "respa.h"
#include "update.h"

#include <cmath>
#include <cstring>

using namespace LAMMPS_NS;
using namespace MathConst;

static constexpr double EWALD_F = 1.12837917;
static constexpr double EWALD_P = 0.3275911;
static constexpr double A1 = 0.254829592;
static constexpr double A2 = -0.284496736;
static constexpr double A3 = 1.421413741;
static constexpr double A4 = -1.453152027;
static constexpr double A5 = 1.061405429;

/* ---------------------------------------------------------------------- */

PairLJClass2CoulLong::PairLJClass2CoulLong(LAMMPS *lmp) : Pair(lmp)
{
  ewaldflag = pppmflag = 1;
  respa_enable = 1;
  writedata = 1;
  ftable = nullptr;
  cut_respa = nullptr;
}

/* ---------------------------------------------------------------------- */

PairLJClass2CoulLong::~PairLJClass2CoulLong()
{
  if (copymode) return;

  if (allocated) {
    memory->destroy(setflag);
    memory->destroy(cutsq);

    memory->destroy(cut_lj);
    memory->destroy(cut_ljsq);
    memory->destroy(epsilon);
    memory->destroy(sigma);
    memory->destroy(lj1);
    memory->destroy(lj2);
    memory->destroy(lj3);
    memory->destroy(lj4);
    memory->destroy(offset);
  }
  if (ftable) free_tables();
}

/* ---------------------------------------------------------------------- */

void PairLJClass2CoulLong::compute(int eflag, int vflag)
{
  int i, j, ii, jj, inum, jnum, itable, itype, jtype;
  double qtmp, xtmp, ytmp, ztmp, delx, dely, delz, evdwl, ecoul, fpair;
  double fraction, table;
  double rsq, r, rinv, r2inv, r3inv, r6inv, forcecoul, forcelj;
  double grij, expm2, prefactor, t, erfc;
  double factor_coul, factor_lj;
  int *ilist, *jlist, *numneigh, **firstneigh;

  evdwl = ecoul = 0.0;
  ev_init(eflag, vflag);

  double **x = atom->x;
  double **f = atom->f;
  double *q = atom->q;
  int *type = atom->type;
  int nlocal = atom->nlocal;
  double *special_coul = force->special_coul;
  double *special_lj = force->special_lj;
  int newton_pair = force->newton_pair;
  double qqrd2e = force->qqrd2e;

  inum = list->inum;
  ilist = list->ilist;
  numneigh = list->numneigh;
  firstneigh = list->firstneigh;

  // loop over neighbors of my atoms

  for (ii = 0; ii < inum; ii++) {
    i = ilist[ii];
    qtmp = q[i];
    xtmp = x[i][0];
    ytmp = x[i][1];
    ztmp = x[i][2];
    itype = type[i];
    jlist = firstneigh[i];
    jnum = numneigh[i];

    for (jj = 0; jj < jnum; jj++) {
      j = jlist[jj];
      factor_lj = special_lj[sbmask(j)];
      factor_coul = special_coul[sbmask(j)];
      j &= NEIGHMASK;

      delx = xtmp - x[j][0];
      dely = ytmp - x[j][1];
      delz = ztmp - x[j][2];
      rsq = delx * delx + dely * dely + delz * delz;
      jtype = type[j];

      if (rsq < cutsq[itype][jtype]) {
        r2inv = 1.0 / rsq;

        if (rsq < cut_coulsq) {
          if (!ncoultablebits || rsq <= tabinnersq) {
            r = sqrt(rsq);
            grij = g_ewald * r;
            expm2 = exp(-grij * grij);
            t = 1.0 / (1.0 + EWALD_P * grij);
            erfc = t * (A1 + t * (A2 + t * (A3 + t * (A4 + t * A5)))) * expm2;
            prefactor = qqrd2e * qtmp * q[j] / r;
            forcecoul = prefactor * (erfc + EWALD_F * grij * expm2);
            if (factor_coul < 1.0) forcecoul -= (1.0 - factor_coul) * prefactor;
          } else {
            union_int_float_t rsq_lookup;
            rsq_lookup.f = rsq;
            itable = rsq_lookup.i & ncoulmask;
            itable >>= ncoulshiftbits;
            fraction = (rsq_lookup.f - rtable[itable]) * drtable[itable];
            table = ftable[itable] + fraction * dftable[itable];
            forcecoul = qtmp * q[j] * table;
            if (factor_coul < 1.0) {
              table = ctable[itable] + fraction * dctable[itable];
              prefactor = qtmp * q[j] * table;
              forcecoul -= (1.0 - factor_coul) * prefactor;
            }
          }
        } else
          forcecoul = 0.0;

        if (rsq < cut_ljsq[itype][jtype]) {
          rinv = sqrt(r2inv);
          r3inv = r2inv * rinv;
          r6inv = r3inv * r3inv;
          forcelj = r6inv * (lj1[itype][jtype] * r3inv - lj2[itype][jtype]);
        } else
          forcelj = 0.0;

        fpair = (forcecoul + factor_lj * forcelj) * r2inv;

        f[i][0] += delx * fpair;
        f[i][1] += dely * fpair;
        f[i][2] += delz * fpair;
        if (newton_pair || j < nlocal) {
          f[j][0] -= delx * fpair;
          f[j][1] -= dely * fpair;
          f[j][2] -= delz * fpair;
        }

        if (eflag) {
          if (rsq < cut_coulsq) {
            if (!ncoultablebits || rsq <= tabinnersq)
              ecoul = prefactor * erfc;
            else {
              table = etable[itable] + fraction * detable[itable];
              ecoul = qtmp * q[j] * table;
            }
            if (factor_coul < 1.0) ecoul -= (1.0 - factor_coul) * prefactor;
          } else
            ecoul = 0.0;
          if (rsq < cut_ljsq[itype][jtype]) {
            evdwl = r6inv * (lj3[itype][jtype] * r3inv - lj4[itype][jtype]) - offset[itype][jtype];
            evdwl *= factor_lj;
          } else
            evdwl = 0.0;
        }

        if (evflag) ev_tally(i, j, nlocal, newton_pair, evdwl, ecoul, fpair, delx, dely, delz);
      }
    }
  }

  if (vflag_fdotr) virial_fdotr_compute();
}

/* ---------------------------------------------------------------------- */

void PairLJClass2CoulLong::compute_inner()
{
  int i, j, ii, jj, inum, jnum, itype, jtype;
  double qtmp, xtmp, ytmp, ztmp, delx, dely, delz, fpair;
  double rsq, rinv, r2inv, r3inv, r6inv, forcecoul, forcelj, factor_coul, factor_lj;
  double rsw;
  int *ilist, *jlist, *numneigh, **firstneigh;

  double **x = atom->x;
  double **f = atom->f;
  double *q = atom->q;
  int *type = atom->type;
  int nlocal = atom->nlocal;
  double *special_coul = force->special_coul;
  double *special_lj = force->special_lj;
  int newton_pair = force->newton_pair;
  double qqrd2e = force->qqrd2e;

  inum = list->inum_inner;
  ilist = list->ilist_inner;
  numneigh = list->numneigh_inner;
  firstneigh = list->firstneigh_inner;

  double cut_out_on = cut_respa[0];
  double cut_out_off = cut_respa[1];

  double cut_out_diff = cut_out_off - cut_out_on;
  double cut_out_on_sq = cut_out_on * cut_out_on;
  double cut_out_off_sq = cut_out_off * cut_out_off;

  // loop over neighbors of my atoms

  for (ii = 0; ii < inum; ii++) {
    i = ilist[ii];
    qtmp = q[i];
    xtmp = x[i][0];
    ytmp = x[i][1];
    ztmp = x[i][2];
    itype = type[i];
    jlist = firstneigh[i];
    jnum = numneigh[i];

    for (jj = 0; jj < jnum; jj++) {
      j = jlist[jj];
      factor_lj = special_lj[sbmask(j)];
      factor_coul = special_coul[sbmask(j)];
      j &= NEIGHMASK;

      delx = xtmp - x[j][0];
      dely = ytmp - x[j][1];
      delz = ztmp - x[j][2];
      rsq = delx * delx + dely * dely + delz * delz;

      if (rsq < cut_out_off_sq) {
        r2inv = 1.0 / rsq;
        forcecoul = qqrd2e * qtmp * q[j] * sqrt(r2inv);
        if (factor_coul < 1.0) forcecoul -= (1.0 - factor_coul) * forcecoul;

        jtype = type[j];
        if (rsq < cut_ljsq[itype][jtype]) {
          rinv = sqrt(r2inv);
          r3inv = r2inv * rinv;
          r6inv = r3inv * r3inv;
          forcelj = r6inv * (lj1[itype][jtype] * r3inv - lj2[itype][jtype]);
        } else
          forcelj = 0.0;

        fpair = (forcecoul + factor_lj * forcelj) * r2inv;
        if (rsq > cut_out_on_sq) {
          rsw = (sqrt(rsq) - cut_out_on) / cut_out_diff;
          fpair *= 1.0 + rsw * rsw * (2.0 * rsw - 3.0);
        }

        f[i][0] += delx * fpair;
        f[i][1] += dely * fpair;
        f[i][2] += delz * fpair;
        if (newton_pair || j < nlocal) {
          f[j][0] -= delx * fpair;
          f[j][1] -= dely * fpair;
          f[j][2] -= delz * fpair;
        }
      }
    }
  }
}

/* ---------------------------------------------------------------------- */

void PairLJClass2CoulLong::compute_middle()
{
  int i, j, ii, jj, inum, jnum, itype, jtype;
  double qtmp, xtmp, ytmp, ztmp, delx, dely, delz, fpair;
  double rsq, rinv, r2inv, r3inv, r6inv, forcecoul, forcelj, factor_coul, factor_lj;
  double rsw;
  int *ilist, *jlist, *numneigh, **firstneigh;

  double **x = atom->x;
  double **f = atom->f;
  double *q = atom->q;
  int *type = atom->type;
  int nlocal = atom->nlocal;
  double *special_coul = force->special_coul;
  double *special_lj = force->special_lj;
  int newton_pair = force->newton_pair;
  double qqrd2e = force->qqrd2e;

  inum = list->inum_middle;
  ilist = list->ilist_middle;
  numneigh = list->numneigh_middle;
  firstneigh = list->firstneigh_middle;

  double cut_in_off = cut_respa[0];
  double cut_in_on = cut_respa[1];
  double cut_out_on = cut_respa[2];
  double cut_out_off = cut_respa[3];

  double cut_in_diff = cut_in_on - cut_in_off;
  double cut_out_diff = cut_out_off - cut_out_on;
  double cut_in_off_sq = cut_in_off * cut_in_off;
  double cut_in_on_sq = cut_in_on * cut_in_on;
  double cut_out_on_sq = cut_out_on * cut_out_on;
  double cut_out_off_sq = cut_out_off * cut_out_off;

  // loop over neighbors of my atoms

  for (ii = 0; ii < inum; ii++) {
    i = ilist[ii];
    qtmp = q[i];
    xtmp = x[i][0];
    ytmp = x[i][1];
    ztmp = x[i][2];
    itype = type[i];
    jlist = firstneigh[i];
    jnum = numneigh[i];

    for (jj = 0; jj < jnum; jj++) {
      j = jlist[jj];
      factor_lj = special_lj[sbmask(j)];
      factor_coul = special_coul[sbmask(j)];
      j &= NEIGHMASK;

      delx = xtmp - x[j][0];
      dely = ytmp - x[j][1];
      delz = ztmp - x[j][2];
      rsq = delx * delx + dely * dely + delz * delz;

      if (rsq < cut_out_off_sq && rsq > cut_in_off_sq) {
        r2inv = 1.0 / rsq;
        forcecoul = qqrd2e * qtmp * q[j] * sqrt(r2inv);
        if (factor_coul < 1.0) forcecoul -= (1.0 - factor_coul) * forcecoul;

        jtype = type[j];
        if (rsq < cut_ljsq[itype][jtype]) {
          rinv = sqrt(r2inv);
          r3inv = r2inv * rinv;
          r6inv = r3inv * r3inv;
          forcelj = r6inv * (lj1[itype][jtype] * r3inv - lj2[itype][jtype]);
        } else
          forcelj = 0.0;

        fpair = (forcecoul + factor_lj * forcelj) * r2inv;
        if (rsq < cut_in_on_sq) {
          rsw = (sqrt(rsq) - cut_in_off) / cut_in_diff;
          fpair *= rsw * rsw * (3.0 - 2.0 * rsw);
        }
        if (rsq > cut_out_on_sq) {
          rsw = (sqrt(rsq) - cut_out_on) / cut_out_diff;
          fpair *= 1.0 + rsw * rsw * (2.0 * rsw - 3.0);
        }

        f[i][0] += delx * fpair;
        f[i][1] += dely * fpair;
        f[i][2] += delz * fpair;
        if (newton_pair || j < nlocal) {
          f[j][0] -= delx * fpair;
          f[j][1] -= dely * fpair;
          f[j][2] -= delz * fpair;
        }
      }
    }
  }
}

/* ---------------------------------------------------------------------- */

void PairLJClass2CoulLong::compute_outer(int eflag, int vflag)
{
  int i, j, ii, jj, inum, jnum, itype, jtype, itable;
  double qtmp, xtmp, ytmp, ztmp, delx, dely, delz, evdwl, ecoul, fpair;
  double fraction, table;
  double r, rinv, r2inv, r3inv, r6inv, forcecoul, forcelj, factor_coul, factor_lj;
  double grij, expm2, prefactor, t, erfc;
  double rsw;
  int *ilist, *jlist, *numneigh, **firstneigh;
  double rsq;

  evdwl = ecoul = 0.0;
  ev_init(eflag, vflag);

  double **x = atom->x;
  double **f = atom->f;
  double *q = atom->q;
  int *type = atom->type;
  int nlocal = atom->nlocal;
  double *special_coul = force->special_coul;
  double *special_lj = force->special_lj;
  int newton_pair = force->newton_pair;
  double qqrd2e = force->qqrd2e;

  inum = list->inum;
  ilist = list->ilist;
  numneigh = list->numneigh;
  firstneigh = list->firstneigh;

  double cut_in_off = cut_respa[2];
  double cut_in_on = cut_respa[3];

  double cut_in_diff = cut_in_on - cut_in_off;
  double cut_in_off_sq = cut_in_off * cut_in_off;
  double cut_in_on_sq = cut_in_on * cut_in_on;

  // loop over neighbors of my atoms

  for (ii = 0; ii < inum; ii++) {
    i = ilist[ii];
    qtmp = q[i];
    xtmp = x[i][0];
    ytmp = x[i][1];
    ztmp = x[i][2];
    itype = type[i];
    jlist = firstneigh[i];
    jnum = numneigh[i];

    for (jj = 0; jj < jnum; jj++) {
      j = jlist[jj];
      factor_lj = special_lj[sbmask(j)];
      factor_coul = special_coul[sbmask(j)];
      j &= NEIGHMASK;

      delx = xtmp - x[j][0];
      dely = ytmp - x[j][1];
      delz = ztmp - x[j][2];
      rsq = delx * delx + dely * dely + delz * delz;
      jtype = type[j];

      if (rsq < cutsq[itype][jtype]) {
        r2inv = 1.0 / rsq;

        if (rsq < cut_coulsq) {
          if (!ncoultablebits || rsq <= tabinnersq) {
            r = sqrt(rsq);
            grij = g_ewald * r;
            expm2 = exp(-grij * grij);
            t = 1.0 / (1.0 + EWALD_P * grij);
            erfc = t * (A1 + t * (A2 + t * (A3 + t * (A4 + t * A5)))) * expm2;
            prefactor = qqrd2e * qtmp * q[j] / r;
            forcecoul = prefactor * (erfc + EWALD_F * grij * expm2 - 1.0);
            if (rsq > cut_in_off_sq) {
              if (rsq < cut_in_on_sq) {
                rsw = (r - cut_in_off) / cut_in_diff;
                forcecoul += prefactor * rsw * rsw * (3.0 - 2.0 * rsw);
                if (factor_coul < 1.0)
                  forcecoul -= (1.0 - factor_coul) * prefactor * rsw * rsw * (3.0 - 2.0 * rsw);
              } else {
                forcecoul += prefactor;
                if (factor_coul < 1.0) forcecoul -= (1.0 - factor_coul) * prefactor;
              }
            }
          } else {
            union_int_float_t rsq_lookup;
            rsq_lookup.f = rsq;
            itable = rsq_lookup.i & ncoulmask;
            itable >>= ncoulshiftbits;
            fraction = (rsq_lookup.f - rtable[itable]) * drtable[itable];
            table = ftable[itable] + fraction * dftable[itable];
            forcecoul = qtmp * q[j] * table;
            if (factor_coul < 1.0) {
              table = ctable[itable] + fraction * dctable[itable];
              prefactor = qtmp * q[j] * table;
              forcecoul -= (1.0 - factor_coul) * prefactor;
            }
          }
        } else
          forcecoul = 0.0;

        if (rsq < cut_ljsq[itype][jtype] && rsq > cut_in_off_sq) {
          rinv = sqrt(r2inv);
          r3inv = r2inv * rinv;
          r6inv = r3inv * r3inv;
          forcelj = r6inv * (lj1[itype][jtype] * r3inv - lj2[itype][jtype]);
          if (rsq < cut_in_on_sq) {
            rsw = (sqrt(rsq) - cut_in_off) / cut_in_diff;
            forcelj *= rsw * rsw * (3.0 - 2.0 * rsw);
          }
        } else
          forcelj = 0.0;

        fpair = (forcecoul + forcelj) * r2inv;

        f[i][0] += delx * fpair;
        f[i][1] += dely * fpair;
        f[i][2] += delz * fpair;
        if (newton_pair || j < nlocal) {
          f[j][0] -= delx * fpair;
          f[j][1] -= dely * fpair;
          f[j][2] -= delz * fpair;
        }

        if (eflag) {
          if (rsq < cut_coulsq) {
            if (!ncoultablebits || rsq <= tabinnersq) {
              ecoul = prefactor * erfc;
              if (factor_coul < 1.0) ecoul -= (1.0 - factor_coul) * prefactor;
            } else {
              table = etable[itable] + fraction * detable[itable];
              ecoul = qtmp * q[j] * table;
              if (factor_coul < 1.0) {
                table = ptable[itable] + fraction * dptable[itable];
                prefactor = qtmp * q[j] * table;
                ecoul -= (1.0 - factor_coul) * prefactor;
              }
            }
          } else
            ecoul = 0.0;

          if (rsq < cut_ljsq[itype][jtype]) {
            rinv = sqrt(r2inv);
            r3inv = r2inv * rinv;
            r6inv = r3inv * r3inv;
            evdwl = r6inv * (lj3[itype][jtype] * r3inv - lj4[itype][jtype]) - offset[itype][jtype];
            evdwl *= factor_lj;
          } else
            evdwl = 0.0;
        }

        if (vflag) {
          if (rsq < cut_coulsq) {
            if (!ncoultablebits || rsq <= tabinnersq) {
              forcecoul = prefactor * (erfc + EWALD_F * grij * expm2);
              if (factor_coul < 1.0) forcecoul -= (1.0 - factor_coul) * prefactor;
            } else {
              table = vtable[itable] + fraction * dvtable[itable];
              forcecoul = qtmp * q[j] * table;
              if (factor_coul < 1.0) {
                table = ptable[itable] + fraction * dptable[itable];
                prefactor = qtmp * q[j] * table;
                forcecoul -= (1.0 - factor_coul) * prefactor;
              }
            }
          } else
            forcecoul = 0.0;

          if (rsq <= cut_in_off_sq) {
            rinv = sqrt(r2inv);
            r3inv = r2inv * rinv;
            r6inv = r3inv * r3inv;
            forcelj = r6inv * (lj1[itype][jtype] * r3inv - lj2[itype][jtype]);
          } else if (rsq <= cut_in_on_sq) {
            rinv = sqrt(r2inv);
            r3inv = r2inv * rinv;
            r6inv = r3inv * r3inv;
            forcelj = r6inv * (lj1[itype][jtype] * r3inv - lj2[itype][jtype]);
          }
          fpair = (forcecoul + factor_lj * forcelj) * r2inv;
        }

        if (evflag) ev_tally(i, j, nlocal, newton_pair, evdwl, ecoul, fpair, delx, dely, delz);
      }
    }
  }
}

/* ----------------------------------------------------------------------
   allocate all arrays
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::allocate()
{
  allocated = 1;
  const int np1 = atom->ntypes + 1;

  memory->create(setflag, np1, np1, "pair:setflag");
  for (int i = 1; i < np1; i++)
    for (int j = i; j < np1; j++) setflag[i][j] = 0;

  memory->create(cutsq, np1, np1, "pair:cutsq");
  memory->create(cut_lj, np1, np1, "pair:cut_lj");
  memory->create(cut_ljsq, np1, np1, "pair:cut_ljsq");
  memory->create(epsilon, np1, np1, "pair:epsilon");
  memory->create(sigma, np1, np1, "pair:sigma");
  memory->create(lj1, np1, np1, "pair:lj1");
  memory->create(lj2, np1, np1, "pair:lj2");
  memory->create(lj3, np1, np1, "pair:lj3");
  memory->create(lj4, np1, np1, "pair:lj4");
  memory->create(offset, np1, np1, "pair:offset");
}

/* ----------------------------------------------------------------------
   global settings
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::settings(int narg, char **arg)
{
  if (narg < 1 || narg > 2) error->all(FLERR, "Illegal pair_style command");

  cut_lj_global = utils::numeric(FLERR, arg[0], false, lmp);
  if (narg == 1)
    cut_coul = cut_lj_global;
  else
    cut_coul = utils::numeric(FLERR, arg[1], false, lmp);

  // reset cutoffs that have been explicitly set

  if (allocated) {
    int i, j;
    for (i = 1; i <= atom->ntypes; i++)
      for (j = i; j <= atom->ntypes; j++)
        if (setflag[i][j]) cut_lj[i][j] = cut_lj_global;
  }
}

/* ----------------------------------------------------------------------
   set coeffs for one or more type pairs
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::coeff(int narg, char **arg)
{
  if (narg < 4 || narg > 5) error->all(FLERR, "Incorrect args for pair coefficients");

  if (!allocated) allocate();

  int ilo, ihi, jlo, jhi;
  utils::bounds(FLERR, arg[0], 1, atom->ntypes, ilo, ihi, error);
  utils::bounds(FLERR, arg[1], 1, atom->ntypes, jlo, jhi, error);

  double epsilon_one = utils::numeric(FLERR, arg[2], false, lmp);
  double sigma_one = utils::numeric(FLERR, arg[3], false, lmp);

  double cut_lj_one = cut_lj_global;
  if (narg == 5) cut_lj_one = utils::numeric(FLERR, arg[4], false, lmp);

  int count = 0;
  for (int i = ilo; i <= ihi; i++) {
    for (int j = MAX(jlo, i); j <= jhi; j++) {
      epsilon[i][j] = epsilon_one;
      sigma[i][j] = sigma_one;
      cut_lj[i][j] = cut_lj_one;
      setflag[i][j] = 1;
      count++;
    }
  }

  if (count == 0) error->all(FLERR, "Incorrect args for pair coefficients");
}

/* ----------------------------------------------------------------------
   init specific to this pair style
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::init_style()
{
  if (!atom->q_flag) error->all(FLERR, "Pair style lj/class2/coul/long requires atom attribute q");

  // request regular or rRESPA neighbor list

  int list_style = NeighConst::REQ_DEFAULT;

  if (update->whichflag == 1 && utils::strmatch(update->integrate_style, "^respa")) {
    auto respa = dynamic_cast<Respa *>(update->integrate);
    if (respa->level_inner >= 0) list_style = NeighConst::REQ_RESPA_INOUT;
    if (respa->level_middle >= 0) list_style = NeighConst::REQ_RESPA_ALL;
  }
  neighbor->add_request(this, list_style);

  cut_coulsq = cut_coul * cut_coul;

  // set rRESPA cutoffs

  if (utils::strmatch(update->integrate_style, "^respa") &&
      (dynamic_cast<Respa *>(update->integrate))->level_inner >= 0)
    cut_respa = (dynamic_cast<Respa *>(update->integrate))->cutoff;
  else
    cut_respa = nullptr;

  // insure use of KSpace long-range solver, set g_ewald

  if (force->kspace == nullptr) error->all(FLERR, "Pair style requires a KSpace style");
  g_ewald = force->kspace->g_ewald;

  // setup force tables
  if (ncoultablebits) init_tables(cut_coul, cut_respa);
}

/* ----------------------------------------------------------------------
   init for one type pair i,j and corresponding j,i
------------------------------------------------------------------------- */

double PairLJClass2CoulLong::init_one(int i, int j)
{
  // always mix epsilon,sigma via sixthpower rules
  // mix distance via user-defined rule

  if (setflag[i][j] == 0) {
    epsilon[i][j] = 2.0 * sqrt(epsilon[i][i] * epsilon[j][j]) * pow(sigma[i][i], 3.0) *
        pow(sigma[j][j], 3.0) / (pow(sigma[i][i], 6.0) + pow(sigma[j][j], 6.0));
    sigma[i][j] = pow((0.5 * (pow(sigma[i][i], 6.0) + pow(sigma[j][j], 6.0))), 1.0 / 6.0);
    cut_lj[i][j] = mix_distance(cut_lj[i][i], cut_lj[j][j]);
    did_mix = true;
  }

  double cut = MAX(cut_lj[i][j], cut_coul);
  cut_ljsq[i][j] = cut_lj[i][j] * cut_lj[i][j];

  lj1[i][j] = 18.0 * epsilon[i][j] * pow(sigma[i][j], 9.0);
  lj2[i][j] = 18.0 * epsilon[i][j] * pow(sigma[i][j], 6.0);
  lj3[i][j] = 2.0 * epsilon[i][j] * pow(sigma[i][j], 9.0);
  lj4[i][j] = 3.0 * epsilon[i][j] * pow(sigma[i][j], 6.0);

  if (offset_flag && (cut_lj[i][j] > 0.0)) {
    double ratio = sigma[i][j] / cut_lj[i][j];
    offset[i][j] = epsilon[i][j] * (2.0 * pow(ratio, 9.0) - 3.0 * pow(ratio, 6.0));
  } else
    offset[i][j] = 0.0;

  cut_ljsq[j][i] = cut_ljsq[i][j];
  lj1[j][i] = lj1[i][j];
  lj2[j][i] = lj2[i][j];
  lj3[j][i] = lj3[i][j];
  lj4[j][i] = lj4[i][j];
  offset[j][i] = offset[i][j];

  // check interior rRESPA cutoff

  if (cut_respa && MIN(cut_lj[i][j], cut_coul) < cut_respa[3])
    error->all(FLERR, "Pair cutoff < Respa interior cutoff");

  // compute I,J contribution to long-range tail correction
  // count total # of atoms of type I and J via Allreduce

  if (tail_flag) {
    int *type = atom->type;
    int nlocal = atom->nlocal;

    double count[2], all[2];
    count[0] = count[1] = 0.0;
    for (int k = 0; k < nlocal; k++) {
      if (type[k] == i) count[0] += 1.0;
      if (type[k] == j) count[1] += 1.0;
    }
    MPI_Allreduce(count, all, 2, MPI_DOUBLE, MPI_SUM, world);

    double sig3 = sigma[i][j] * sigma[i][j] * sigma[i][j];
    double sig6 = sig3 * sig3;
    double rc3 = cut_lj[i][j] * cut_lj[i][j] * cut_lj[i][j];
    double rc6 = rc3 * rc3;
    etail_ij =
        2.0 * MY_PI * all[0] * all[1] * epsilon[i][j] * sig6 * (sig3 - 3.0 * rc3) / (3.0 * rc6);
    ptail_ij = 2.0 * MY_PI * all[0] * all[1] * epsilon[i][j] * sig6 * (sig3 - 2.0 * rc3) / rc6;
  }

  return cut;
}

/* ----------------------------------------------------------------------
   proc 0 writes to restart file
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::write_restart(FILE *fp)
{
  write_restart_settings(fp);

  int i, j;
  for (i = 1; i <= atom->ntypes; i++)
    for (j = i; j <= atom->ntypes; j++) {
      fwrite(&setflag[i][j], sizeof(int), 1, fp);
      if (setflag[i][j]) {
        fwrite(&epsilon[i][j], sizeof(double), 1, fp);
        fwrite(&sigma[i][j], sizeof(double), 1, fp);
        fwrite(&cut_lj[i][j], sizeof(double), 1, fp);
      }
    }
}

/* ----------------------------------------------------------------------
   proc 0 reads from restart file, bcasts
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::read_restart(FILE *fp)
{
  read_restart_settings(fp);
  allocate();

  int i, j;
  int me = comm->me;
  for (i = 1; i <= atom->ntypes; i++)
    for (j = i; j <= atom->ntypes; j++) {
      if (me == 0) utils::sfread(FLERR, &setflag[i][j], sizeof(int), 1, fp, nullptr, error);
      MPI_Bcast(&setflag[i][j], 1, MPI_INT, 0, world);
      if (setflag[i][j]) {
        if (me == 0) {
          utils::sfread(FLERR, &epsilon[i][j], sizeof(double), 1, fp, nullptr, error);
          utils::sfread(FLERR, &sigma[i][j], sizeof(double), 1, fp, nullptr, error);
          utils::sfread(FLERR, &cut_lj[i][j], sizeof(double), 1, fp, nullptr, error);
        }
        MPI_Bcast(&epsilon[i][j], 1, MPI_DOUBLE, 0, world);
        MPI_Bcast(&sigma[i][j], 1, MPI_DOUBLE, 0, world);
        MPI_Bcast(&cut_lj[i][j], 1, MPI_DOUBLE, 0, world);
      }
    }
}

/* ----------------------------------------------------------------------
   proc 0 writes to restart file
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::write_restart_settings(FILE *fp)
{
  fwrite(&cut_lj_global, sizeof(double), 1, fp);
  fwrite(&cut_coul, sizeof(double), 1, fp);
  fwrite(&offset_flag, sizeof(int), 1, fp);
  fwrite(&mix_flag, sizeof(int), 1, fp);
  fwrite(&tail_flag, sizeof(int), 1, fp);
  fwrite(&ncoultablebits, sizeof(int), 1, fp);
  fwrite(&tabinner, sizeof(double), 1, fp);
}

/* ----------------------------------------------------------------------
   proc 0 reads from restart file, bcasts
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::read_restart_settings(FILE *fp)
{
  if (comm->me == 0) {
    utils::sfread(FLERR, &cut_lj_global, sizeof(double), 1, fp, nullptr, error);
    utils::sfread(FLERR, &cut_coul, sizeof(double), 1, fp, nullptr, error);
    utils::sfread(FLERR, &offset_flag, sizeof(int), 1, fp, nullptr, error);
    utils::sfread(FLERR, &mix_flag, sizeof(int), 1, fp, nullptr, error);
    utils::sfread(FLERR, &tail_flag, sizeof(int), 1, fp, nullptr, error);
    utils::sfread(FLERR, &ncoultablebits, sizeof(int), 1, fp, nullptr, error);
    utils::sfread(FLERR, &tabinner, sizeof(double), 1, fp, nullptr, error);
  }
  MPI_Bcast(&cut_lj_global, 1, MPI_DOUBLE, 0, world);
  MPI_Bcast(&cut_coul, 1, MPI_DOUBLE, 0, world);
  MPI_Bcast(&offset_flag, 1, MPI_INT, 0, world);
  MPI_Bcast(&mix_flag, 1, MPI_INT, 0, world);
  MPI_Bcast(&tail_flag, 1, MPI_INT, 0, world);
  MPI_Bcast(&ncoultablebits, 1, MPI_INT, 0, world);
  MPI_Bcast(&tabinner, 1, MPI_DOUBLE, 0, world);
}

/* ----------------------------------------------------------------------
   proc 0 writes to data file
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::write_data(FILE *fp)
{
  for (int i = 1; i <= atom->ntypes; i++) fprintf(fp, "%d %g %g\n", i, epsilon[i][i], sigma[i][i]);
}

/* ----------------------------------------------------------------------
   proc 0 writes all pairs to data file
------------------------------------------------------------------------- */

void PairLJClass2CoulLong::write_data_all(FILE *fp)
{
  for (int i = 1; i <= atom->ntypes; i++)
    for (int j = i; j <= atom->ntypes; j++)
      fprintf(fp, "%d %d %g %g %g\n", i, j, epsilon[i][j], sigma[i][j], cut_lj[i][j]);
}

/* ---------------------------------------------------------------------- */

double PairLJClass2CoulLong::single(int i, int j, int itype, int jtype, double rsq,
                                    double factor_coul, double factor_lj, double &fforce)
{
  double r2inv, r, rinv, r3inv, r6inv, grij, expm2, t, erfc, prefactor;
  double fraction, table, forcecoul, forcelj, phicoul, philj;
  int itable;

  r2inv = 1.0 / rsq;
  if (rsq < cut_coulsq) {
    if (!ncoultablebits || rsq <= tabinnersq) {
      r = sqrt(rsq);
      grij = g_ewald * r;
      expm2 = exp(-grij * grij);
      t = 1.0 / (1.0 + EWALD_P * grij);
      erfc = t * (A1 + t * (A2 + t * (A3 + t * (A4 + t * A5)))) * expm2;
      prefactor = force->qqrd2e * atom->q[i] * atom->q[j] / r;
      forcecoul = prefactor * (erfc + EWALD_F * grij * expm2);
      if (factor_coul < 1.0) forcecoul -= (1.0 - factor_coul) * prefactor;
    } else {
      union_int_float_t rsq_lookup;
      rsq_lookup.f = rsq;
      itable = rsq_lookup.i & ncoulmask;
      itable >>= ncoulshiftbits;
      fraction = (rsq_lookup.f - rtable[itable]) * drtable[itable];
      table = ftable[itable] + fraction * dftable[itable];
      forcecoul = atom->q[i] * atom->q[j] * table;
      if (factor_coul < 1.0) {
        table = ctable[itable] + fraction * dctable[itable];
        prefactor = atom->q[i] * atom->q[j] * table;
        forcecoul -= (1.0 - factor_coul) * prefactor;
      }
    }
  } else
    forcecoul = 0.0;
  if (rsq < cut_ljsq[itype][jtype]) {
    rinv = sqrt(r2inv);
    r3inv = r2inv * rinv;
    r6inv = r3inv * r3inv;
    forcelj = r6inv * (lj1[itype][jtype] * r3inv - lj2[itype][jtype]);
  } else
    forcelj = 0.0;
  fforce = (forcecoul + factor_lj * forcelj) * r2inv;

  double eng = 0.0;
  if (rsq < cut_coulsq) {
    if (!ncoultablebits || rsq <= tabinnersq)
      phicoul = prefactor * erfc;
    else {
      table = etable[itable] + fraction * detable[itable];
      phicoul = atom->q[i] * atom->q[j] * table;
    }
    if (factor_coul < 1.0) phicoul -= (1.0 - factor_coul) * prefactor;
    eng += phicoul;
  }
  if (rsq < cut_ljsq[itype][jtype]) {
    philj = r6inv * (lj3[itype][jtype] * r3inv - lj4[itype][jtype]) - offset[itype][jtype];
    eng += factor_lj * philj;
  }

  return eng;
}

/* ---------------------------------------------------------------------- */

void *PairLJClass2CoulLong::extract(const char *str, int &dim)
{
  dim = 0;
  if (strcmp(str, "cut_coul") == 0) return (void *) &cut_coul;
  dim = 2;
  if (strcmp(str, "epsilon") == 0) return (void *) epsilon;
  if (strcmp(str, "sigma") == 0) return (void *) sigma;
  return nullptr;
}