Page MenuHomec4science

fix_srd.cpp
No OneTemporary

File Metadata

Created
Mon, Jun 17, 19:25

fix_srd.cpp

/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
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.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing authors: Jeremy Lechman (SNL), Pieter in 't Veld (BASF)
------------------------------------------------------------------------- */
#include "math.h"
#include "string.h"
#include "stdlib.h"
#include "fix_srd.h"
#include "atom.h"
#include "atom_vec.h"
#include "group.h"
#include "update.h"
#include "force.h"
#include "pair.h"
#include "domain.h"
#include "neighbor.h"
#include "comm.h"
#include "modify.h"
#include "fix_deform.h"
#include "fix_wall_srd.h"
#include "random_mars.h"
#include "random_park.h"
#include "memory.h"
#include "error.h"
using namespace LAMMPS_NS;
enum{SLIP,NOSLIP};
enum{SPHERE,ELLIPSOID,WALL};
enum{INSIDE_ERROR,INSIDE_WARN,INSIDE_IGNORE};
enum{BIG_MOVE,SRD_MOVE,SRD_ROTATE};
enum{CUBIC_ERROR,CUBIC_WARN};
enum{SHIFT_NO,SHIFT_YES,SHIFT_POSSIBLE};
enum{NO_REMAP,X_REMAP,V_REMAP}; // same as fix_deform.cpp
#define INERTIA 0.4
#define ATOMPERBIN 10
#define BIG 1.0e20
#define VBINSIZE 5
#define MIN(A,B) ((A) < (B)) ? (A) : (B)
#define MAX(A,B) ((A) > (B)) ? (A) : (B)
//#define SRD_DEBUG 1
//#define SRD_DEBUG_ATOMID 58
//#define SRD_DEBUG_TIMESTEP 449
/* ---------------------------------------------------------------------- */
FixSRD::FixSRD(LAMMPS *lmp, int narg, char **arg) : Fix(lmp, narg, arg)
{
if (narg < 8) error->all("Illegal fix srd command");
restart_pbc = 1;
vector_flag = 1;
size_vector = 12;
global_freq = 1;
extvector = 0;
nevery = atoi(arg[3]);
bigexist = 1;
if (strcmp(arg[4],"NULL") == 0) bigexist = 0;
else biggroup = group->find(arg[4]);
temperature_srd = atof(arg[5]);
gridsrd = atof(arg[6]);
int seed = atoi(arg[7]);
// parse options
lamdaflag = 0;
collidestyle = NOSLIP;
overlap = 0;
insideflag = INSIDE_ERROR;
exactflag = 1;
radfactor = 1.0;
maxbounceallow = 0;
gridsearch = gridsrd;
cubicflag = CUBIC_ERROR;
cubictol = 0.01;
shiftuser = SHIFT_NO;
shiftseed = 0;
streamflag = 1;
int iarg = 8;
while (iarg < narg) {
if (strcmp(arg[iarg],"lamda") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
lamda = atof(arg[iarg+1]);
lamdaflag = 1;
iarg += 2;
} else if (strcmp(arg[iarg],"collision") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
if (strcmp(arg[iarg+1],"slip") == 0) collidestyle = SLIP;
else if (strcmp(arg[iarg+1],"noslip") == 0) collidestyle = NOSLIP;
else error->all("Illegal fix srd command");
iarg += 2;
} else if (strcmp(arg[iarg],"overlap") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
if (strcmp(arg[iarg+1],"yes") == 0) overlap = 1;
else if (strcmp(arg[iarg+1],"no") == 0) overlap = 0;
else error->all("Illegal fix srd command");
iarg += 2;
} else if (strcmp(arg[iarg],"inside") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
if (strcmp(arg[iarg+1],"error") == 0) insideflag = INSIDE_ERROR;
else if (strcmp(arg[iarg+1],"warn") == 0) insideflag = INSIDE_WARN;
else if (strcmp(arg[iarg+1],"ignore") == 0) insideflag = INSIDE_IGNORE;
else error->all("Illegal fix srd command");
iarg += 2;
} else if (strcmp(arg[iarg],"exact") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
if (strcmp(arg[iarg+1],"yes") == 0) exactflag = 1;
else if (strcmp(arg[iarg+1],"no") == 0) exactflag = 0;
else error->all("Illegal fix srd command");
iarg += 2;
} else if (strcmp(arg[iarg],"radius") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
radfactor = atof(arg[iarg+1]);
iarg += 2;
} else if (strcmp(arg[iarg],"bounce") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
maxbounceallow = atoi(arg[iarg+1]);
iarg += 2;
} else if (strcmp(arg[iarg],"search") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
gridsearch = atof(arg[iarg+1]);
iarg += 2;
} else if (strcmp(arg[iarg],"cubic") == 0) {
if (iarg+3 > narg) error->all("Illegal fix srd command");
if (strcmp(arg[iarg+1],"error") == 0) cubicflag = CUBIC_ERROR;
else if (strcmp(arg[iarg+1],"warn") == 0) cubicflag = CUBIC_WARN;
else error->all("Illegal fix srd command");
cubictol = atof(arg[iarg+2]);
iarg += 3;
} else if (strcmp(arg[iarg],"shift") == 0) {
if (iarg+3 > narg) error->all("Illegal fix srd command");
else if (strcmp(arg[iarg+1],"no") == 0) shiftuser = SHIFT_NO;
else if (strcmp(arg[iarg+1],"yes") == 0) shiftuser = SHIFT_YES;
else if (strcmp(arg[iarg+1],"possible") == 0) shiftuser = SHIFT_POSSIBLE;
else error->all("Illegal fix srd command");
shiftseed = atoi(arg[iarg+2]);
iarg += 3;
} else if (strcmp(arg[iarg],"stream") == 0) {
if (iarg+2 > narg) error->all("Illegal fix srd command");
if (strcmp(arg[iarg+1],"yes") == 0) streamflag = 1;
else if (strcmp(arg[iarg+1],"no") == 0) streamflag = 0;
else error->all("Illegal fix srd command");
iarg += 2;
} else error->all("Illegal fix srd command");
}
// error check
if (nevery <= 0) error->all("Illegal fix srd command");
if (bigexist && biggroup < 0) error->all("Could not find fix srd group ID");
if (gridsrd <= 0.0) error->all("Illegal fix srd command");
if (temperature_srd <= 0.0) error->all("Illegal fix srd command");
if (seed <= 0) error->all("Illegal fix srd command");
if (radfactor <= 0.0) error->all("Illegal fix srd command");
if (maxbounceallow < 0) error->all("Illegal fix srd command");
if (lamdaflag && lamda <= 0.0) error->all("Illegal fix srd command");
if (gridsearch <= 0.0) error->all("Illegal fix srd command");
if (cubictol < 0.0 || cubictol > 1.0) error->all("Illegal fix srd command");
if ((shiftuser == SHIFT_YES || shiftuser == SHIFT_POSSIBLE) &&
shiftseed <= 0) error->all("Illegal fix srd command");
// initialize Marsaglia RNG with processor-unique seed
me = comm->me;
nprocs = comm->nprocs;
random = new RanMars(lmp,seed + me);
// if requested, initialize shift RNG, same on every proc
if (shiftuser == SHIFT_YES || shiftuser == SHIFT_POSSIBLE)
randomshift = new RanPark(lmp,shiftseed);
else randomshift = NULL;
// initialize data structs and flags
if (bigexist) biggroupbit = group->bitmask[biggroup];
else biggroupbit = 0;
nmax = 0;
binhead = NULL;
maxbin1 = 0;
binnext = NULL;
maxbuf = 0;
sbuf1 = sbuf2 = rbuf1 = rbuf2 = NULL;
shifts[0].maxvbin = shifts[1].maxvbin = 0;
shifts[0].vbin = shifts[1].vbin = NULL;
shifts[0].maxbinsq = shifts[1].maxbinsq = 0;
for (int ishift = 0; ishift < 2; ishift++)
for (int iswap = 0; iswap < 6; iswap++)
shifts[ishift].bcomm[iswap].sendlist =
shifts[ishift].bcomm[iswap].recvlist = NULL;
maxbin2 = 0;
nbinbig = NULL;
binbig = NULL;
binsrd = NULL;
nstencil = maxstencil = 0;
stencil = NULL;
maxbig = 0;
biglist = NULL;
stats_flag = 1;
for (int i = 0; i < size_vector; i++) stats_all[i] = 0.0;
initflag = 0;
srd_bin_temp = 0.0;
srd_bin_count = 0;
// fix parameters
if (collidestyle == SLIP) comm_reverse = 3;
else comm_reverse = 6;
force_reneighbor = 1;
}
/* ---------------------------------------------------------------------- */
FixSRD::~FixSRD()
{
delete random;
delete randomshift;
memory->destroy(binhead);
memory->destroy(binnext);
memory->destroy(sbuf1);
memory->destroy(sbuf2);
memory->destroy(rbuf1);
memory->destroy(rbuf2);
memory->sfree(shifts[0].vbin);
memory->sfree(shifts[1].vbin);
for (int ishift = 0; ishift < 2; ishift++)
for (int iswap = 0; iswap < 6; iswap++) {
memory->destroy(shifts[ishift].bcomm[iswap].sendlist);
memory->destroy(shifts[ishift].bcomm[iswap].recvlist);
}
memory->destroy(nbinbig);
memory->destroy(binbig);
memory->destroy(binsrd);
memory->destroy(stencil);
memory->sfree(biglist);
}
/* ---------------------------------------------------------------------- */
int FixSRD::setmask()
{
int mask = 0;
mask |= PRE_NEIGHBOR;
mask |= POST_FORCE;
return mask;
}
/* ---------------------------------------------------------------------- */
void FixSRD::init()
{
// error checks
if (force->newton_pair == 0) error->all("Fix srd requires newton pair on");
if (bigexist && comm->ghost_velocity == 0)
error->all("Fix srd requires ghost atoms store velocity");
if (bigexist && !atom->sphere_flag && !atom->ellipsoid_flag)
error->all("Fix SRD requires atom style sphere or ellipsoid");
if (bigexist && atom->angmom_flag && atom->omega_flag)
error->all("Fix SRD cannot have both atom attributes angmom and omega");
if (bigexist && collidestyle == NOSLIP && !atom->torque_flag)
error->all("Fix SRD no-slip requires atom attribute torque");
if (initflag && update->dt != dt_big)
error->all("Cannot change timestep once fix srd is setup");
// orthogonal vs triclinic simulation box
// could be static or shearing box
triclinic = domain->triclinic;
// wallexist = 1 if SRD wall(s) are defined
wallexist = 0;
for (int m = 0; m < modify->nfix; m++) {
if (strcmp(modify->fix[m]->style,"wall/srd") == 0) {
if (wallexist) error->all("Cannot use fix wall/srd more than once");
wallexist = 1;
wallfix = (FixWallSRD *) modify->fix[m];
nwall = wallfix->nwall;
wallvarflag = wallfix->varflag;
wallwhich = wallfix->wallwhich;
xwall = wallfix->xwall;
xwallhold = wallfix->xwallhold;
vwall = wallfix->vwall;
fwall = wallfix->fwall;
walltrigger = 0.5 * neighbor->skin;
if (wallfix->overlap && overlap == 0 && me == 0)
error->warning("Fix SRD walls overlap but fix srd overlap not set");
}
}
// set change_flags if box size or shape changes
change_size = change_shape = 0;
if (domain->nonperiodic == 2) change_size = 1;
for (int i = 0; i < modify->nfix; i++)
if (modify->fix[i]->box_change) {
if (modify->fix[i]->box_change_size) change_size = 1;
if (modify->fix[i]->box_change_shape) change_shape = 1;
if (strcmp(modify->fix[i]->style,"deform") == 0) {
FixDeform *deform = (FixDeform *) modify->fix[i];
if (deform->box_change_shape && deform->remapflag != V_REMAP)
error->all("Using fix srd with inconsistent "
"fix deform remap option");
}
}
// parameterize based on current box volume
dimension = domain->dimension;
parameterize();
// limit initial SRD velocities if necessary
double **v = atom->v;
int *mask = atom->mask;
int nlocal = atom->nlocal;
double vsq;
nrescale = 0;
for (int i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
vsq = v[i][0]*v[i][0] + v[i][1]*v[i][1] + v[i][2]*v[i][2];
if (vsq > vmaxsq) nrescale++;
}
int all;
MPI_Allreduce(&nrescale,&all,1,MPI_INT,MPI_SUM,world);
if (me == 0) {
if (screen)
fprintf(screen," # of rescaled SRD velocities = %d\n",all);
if (logfile)
fprintf(logfile," # of rescaled SRD velocities = %d\n",all);
}
velocity_stats(igroup);
if (bigexist) velocity_stats(biggroup);
// zero per-run stats
bouncemaxnum = 0;
bouncemax = 0;
reneighcount = 0;
initflag = 1;
next_reneighbor = -1;
}
/* ---------------------------------------------------------------------- */
void FixSRD::setup(int vflag)
{
setup_bounds();
if (dist_srd_reneigh < nevery*dt_big*vmax && me == 0)
error->warning("Fix srd SRD moves may trigger frequent reneighboring");
// setup search bins and search stencil based on these distances
if (bigexist || wallexist) {
setup_search_bins();
setup_search_stencil();
} else nbins2 = 0;
// perform first bining of SRD and big particles and walls
// set reneighflag to turn off SRD rotation
// don't do SRD rotation in setup, only during timestepping
reneighflag = BIG_MOVE;
pre_neighbor();
}
/* ----------------------------------------------------------------------
assign SRD particles to bins
assign big particles to all bins they overlap
------------------------------------------------------------------------- */
void FixSRD::pre_neighbor()
{
int i,j,m,ix,iy,iz,jx,jy,jz,ibin,jbin,lo,hi;
double rsq,cutbinsq;
double xlamda[3];
// grow SRD per-atom bin arrays if necessary
if (atom->nlocal > nmax) {
nmax = atom->nmax;
memory->destroy(binsrd);
memory->destroy(binnext);
memory->create(binsrd,nmax,"fix/srd:binsrd");
memory->create(binnext,nmax,"fix/srd:binnext");
}
// setup and grow BIG info list if necessary
// set index ptrs to BIG particles and to WALLS
// big_static() adds static properties to info list
if (bigexist || wallexist) {
if (bigexist) {
if (biggroup == atom->firstgroup) nbig = atom->nfirst + atom->nghost;
else {
int *mask = atom->mask;
int nlocal = atom->nlocal;
nbig = atom->nghost;
for (i = 0; i < nlocal; i++)
if (mask[i] & biggroupbit) nbig++;
}
} else nbig = 0;
int ninfo = nbig;
if (wallexist) ninfo += nwall;
if (ninfo > maxbig) {
maxbig = ninfo;
memory->destroy(biglist);
biglist = (Big *) memory->smalloc(maxbig*sizeof(Big),"fix/srd:biglist");
}
if (bigexist) {
int *mask = atom->mask;
int nlocal = atom->nlocal;
if (biggroup == atom->firstgroup) nlocal = atom->nfirst;
nbig = 0;
for (i = 0; i < nlocal; i++)
if (mask[i] & biggroupbit) biglist[nbig++].index = i;
int nall = atom->nlocal + atom->nghost;
for (i = atom->nlocal; i < nall; i++)
if (mask[i] & biggroupbit) biglist[nbig++].index = i;
big_static();
}
if (wallexist) {
for (m = 0; m < nwall; m++) {
biglist[nbig+m].index = m;
biglist[nbig+m].type = WALL;
}
wallfix->wall_params(1);
}
}
// if simulation box size changes, reset velocity bins
// if big particles exist, reset search bins if box size or shape changes,
// b/c finite-size particles will overlap different bins as the box tilts
if (change_size) setup_bounds();
if (change_size) setup_velocity_bins();
if ((change_size || change_shape) && (bigexist || wallexist)) {
setup_search_bins();
setup_search_stencil();
}
// map each owned & ghost big particle to search bins it overlaps
// zero out bin counters for big particles
// if firstgroup is defined, only loop over first and ghost particles
// for each big particle: loop over stencil to find overlap bins
int *mask = atom->mask;
double **x = atom->x;
int nlocal = atom->nlocal;
int nall = nlocal + atom->nghost;
int nfirst = nlocal;
if (bigexist && biggroup == atom->firstgroup) nfirst = atom->nfirst;
if (bigexist || wallexist)
for (i = 0; i < nbins2; i++)
nbinbig[i] = 0;
if (bigexist) {
i = nbig = 0;
while (i < nall) {
if (mask[i] & biggroupbit) {
ix = static_cast<int> ((x[i][0]-xblo2)*bininv2x);
iy = static_cast<int> ((x[i][1]-yblo2)*bininv2y);
iz = static_cast<int> ((x[i][2]-zblo2)*bininv2z);
ibin = iz*nbin2y*nbin2x + iy*nbin2x + ix;
if (ix < 0 || ix >= nbin2x || iy < 0 || iy >= nbin2y ||
iz < 0 || iz >= nbin2z)
error->one("Fix SRD: bad search bin assignment");
cutbinsq = biglist[nbig].cutbinsq;
for (j = 0; j < nstencil; j++) {
jx = ix + stencil[j][0];
jy = iy + stencil[j][1];
jz = iz + stencil[j][2];
if (jx < 0 || jx >= nbin2x || jy < 0 || jy >= nbin2y ||
jz < 0 || jz >= nbin2z) {
printf("Big particle %d %d %g %g %g\n",
atom->tag[i],i,x[i][0],x[i][1],x[i][2]);
printf("Bin indices: %d %d %d, %d %d %d, %d %d %d\n",
ix,iy,iz,jx,jy,jz,nbin2x,nbin2y,nbin2z);
error->one("Fix SRD: bad stencil bin for big particle");
}
rsq = point_bin_distance(x[i],jx,jy,jz);
if (rsq < cutbinsq) {
jbin = ibin + stencil[j][3];
if (nbinbig[jbin] == ATOMPERBIN)
error->one("Fix SRD: too many big particles in bin");
binbig[jbin][nbinbig[jbin]++] = nbig;
}
}
nbig++;
}
i++;
if (i == nfirst) i = nlocal;
}
}
// map each wall to search bins it covers, up to non-periodic boundary
// if wall moves, add walltrigger to its position
// this insures it is added to all search bins it may move into
// may not overlap any of my search bins
if (wallexist) {
double delta = 0.0;
if (wallvarflag) delta = walltrigger;
for (m = 0; m < nwall; m++) {
int dim = wallwhich[m] / 2;
int side = wallwhich[m] % 2;
if (dim == 0) {
if (side == 0) {
hi = static_cast<int> ((xwall[m]+delta-xblo2)*bininv2x);
if (hi < 0) continue;
if (hi >= nbin2x) error->all("Fix SRD: bad search bin assignment");
lo = 0;
} else {
lo = static_cast<int> ((xwall[m]-delta-xblo2)*bininv2x);
if (lo >= nbin2x) continue;
if (lo < 0) error->all("Fix SRD: bad search bin assignment");
hi = nbin2x-1;
}
for (ix = lo; ix <= hi; ix++)
for (iy = 0; iy < nbin2y; iy++)
for (iz = 0; iz < nbin2z; iz++) {
ibin = iz*nbin2y*nbin2x + iy*nbin2x + ix;
if (nbinbig[ibin] == ATOMPERBIN)
error->all("Fix SRD: too many walls in bin");
binbig[ibin][nbinbig[ibin]++] = nbig+m;
}
} else if (dim == 1) {
if (side == 0) {
hi = static_cast<int> ((xwall[m]+delta-yblo2)*bininv2y);
if (hi < 0) continue;
if (hi >= nbin2y) error->all("Fix SRD: bad search bin assignment");
lo = 0;
} else {
lo = static_cast<int> ((xwall[m]-delta-yblo2)*bininv2y);
if (lo >= nbin2y) continue;
if (lo < 0) error->all("Fix SRD: bad search bin assignment");
hi = nbin2y-1;
}
for (iy = lo; iy <= hi; iy++)
for (ix = 0; ix < nbin2x; ix++)
for (iz = 0; iz < nbin2z; iz++) {
ibin = iz*nbin2y*nbin2x + iy*nbin2x + ix;
if (nbinbig[ibin] == ATOMPERBIN)
error->all("Fix SRD: too many walls in bin");
binbig[ibin][nbinbig[ibin]++] = nbig+m;
}
} else if (dim == 2) {
if (side == 0) {
hi = static_cast<int> ((xwall[m]+delta-zblo2)*bininv2z);
if (hi < 0) continue;
if (hi >= nbin2z) error->all("Fix SRD: bad search bin assignment");
lo = 0;
} else {
lo = static_cast<int> ((xwall[m]-delta-zblo2)*bininv2z);
if (lo >= nbin2z) continue;
if (lo < 0) error->all("Fix SRD: bad search bin assignment");
hi = nbin2z-1;
}
for (iz = lo; iz < hi; iz++)
for (ix = 0; ix < nbin2x; ix++)
for (iy = 0; iy < nbin2y; iy++) {
ibin = iz*nbin2y*nbin2x + iy*nbin2x + ix;
if (nbinbig[ibin] == ATOMPERBIN)
error->all("Fix SRD: too many walls in bin");
binbig[ibin][nbinbig[ibin]++] = nbig+m;
}
}
}
}
// rotate SRD velocities on SRD timestep
// done now since all SRDs are currently inside my sub-domain
if (reneighflag == SRD_ROTATE) reset_velocities();
// log stats if reneighboring occurred b/c SRDs moved too far
if (reneighflag == SRD_MOVE) reneighcount++;
reneighflag = BIG_MOVE;
}
/* ----------------------------------------------------------------------
advect SRD particles and detect collisions between SRD and BIG particles
when collision occurs, change x,v of SRD, force,torque of BIG particle
------------------------------------------------------------------------- */
void FixSRD::post_force(int vflag)
{
int i,m,ix,iy,iz;
double xlamda[3];
// zero per-timestep stats
stats_flag = 0;
ncheck = ncollide = nbounce = ninside = nrescale = 0;
// zero ghost forces & torques on BIG particles
double **f = atom->f;
double **torque = atom->torque;
int nlocal = atom->nlocal;
int nall = nlocal + atom->nghost;
if (bigexist == 0) nall = 0;
for (i = nlocal; i < nall; i++)
f[i][0] = f[i][1] = f[i][2] = 0.0;
if (collidestyle == NOSLIP)
for (i = nlocal; i < nall; i++)
torque[i][0] = torque[i][1] = torque[i][2] = 0.0;
// advect SRD particles
// assign to search bins if big particles or walls exist
int *mask = atom->mask;
double **x = atom->x;
double **v = atom->v;
if (bigexist || wallexist) {
for (i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
x[i][0] += dt_big*v[i][0];
x[i][1] += dt_big*v[i][1];
x[i][2] += dt_big*v[i][2];
ix = static_cast<int> ((x[i][0]-xblo2)*bininv2x);
iy = static_cast<int> ((x[i][1]-yblo2)*bininv2y);
iz = static_cast<int> ((x[i][2]-zblo2)*bininv2z);
binsrd[i] = iz*nbin2y*nbin2x + iy*nbin2x + ix;
if (ix < 0 || ix >= nbin2x || iy < 0 || iy >= nbin2y ||
iz < 0 || iz >= nbin2z) {
if (screen) {
fprintf(screen,"SRD particle %d on step " BIGINT_FORMAT "\n",
atom->tag[i],update->ntimestep);
fprintf(screen,"v = %g %g %g\n",v[i][0],v[i][1],v[i][2]);
fprintf(screen,"x = %g %g %g\n",x[i][0],x[i][1],x[i][2]);
fprintf(screen,"ix,iy,iz nx,ny,nz = %d %d %d %d %d %d\n",
ix,iy,iz,nbin2x,nbin2y,nbin2z);
}
error->one("Fix SRD: bad bin assignment for SRD advection");
}
}
} else {
for (i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
x[i][0] += dt_big*v[i][0];
x[i][1] += dt_big*v[i][1];
x[i][2] += dt_big*v[i][2];
}
}
// detect collision of SRDs with BIG particles or walls
if (bigexist || wallexist) {
if (bigexist && (collidestyle == NOSLIP || any_ellipsoids)) big_dynamic();
if (wallexist) wallfix->wall_params(0);
if (overlap) collisions_multi();
else collisions_single();
}
// reverse communicate forces & torques on BIG particles
if (bigexist) {
flocal = f;
tlocal = torque;
comm->reverse_comm_fix(this);
}
// if any SRD particle has moved too far, trigger reneigh on next step
// for triclinic, perform check in lamda units
int flag = 0;
if (triclinic) domain->x2lamda(nlocal);
for (i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
if (x[i][0] < srdlo_reneigh[0] || x[i][0] > srdhi_reneigh[0] ||
x[i][1] < srdlo_reneigh[1] || x[i][1] > srdhi_reneigh[1] ||
x[i][2] < srdlo_reneigh[2] || x[i][2] > srdhi_reneigh[2]) flag = 1;
}
if (triclinic) domain->lamda2x(nlocal);
int flagall;
MPI_Allreduce(&flag,&flagall,1,MPI_INT,MPI_SUM,world);
if (flagall) {
next_reneighbor = update->ntimestep + 1;
reneighflag = SRD_MOVE;
}
// if wall has moved too far, trigger reneigh on next step
// analagous to neighbor check for big particle moving 1/2 of skin distance
if (wallexist) {
for (m = 0; m < nwall; m++)
if (fabs(xwall[m]-xwallhold[m]) > walltrigger)
next_reneighbor = update->ntimestep + 1;
}
// if next timestep is SRD timestep, trigger reneigh
if ((update->ntimestep+1) % nevery == 0) {
next_reneighbor = update->ntimestep + 1;
reneighflag = SRD_ROTATE;
}
}
/* ----------------------------------------------------------------------
reset SRD velocities
may perform random shifting by 1/2 bin in each dimension
called at pre-neighbor stage when all SRDs are now inside my sub-domain
for triclinic, will set mean velocity to box deformation velocity
------------------------------------------------------------------------- */
void FixSRD::reset_velocities()
{
int i,j,n,ix,iy,iz,ibin,axis,sign,irandom;
double u[3],vave[3];
double vx,vy,vz,vsq;
double *vold,*vnew,*xlamda;
double vstream[3];
// if requested, perform a dynamic shift
if (shiftflag) {
double *boxlo;
if (triclinic == 0) boxlo = domain->boxlo;
else boxlo = domain->boxlo_lamda;
shifts[1].corner[0] = boxlo[0] - binsize1x*randomshift->uniform();
shifts[1].corner[1] = boxlo[1] - binsize1y*randomshift->uniform();
if (dimension == 3)
shifts[1].corner[2] = boxlo[2] - binsize1z*randomshift->uniform();
else shifts[1].corner[2] = boxlo[2];
setup_velocity_shift(1,1);
}
double *corner = shifts[shiftflag].corner;
int *binlo = shifts[shiftflag].binlo;
int *binhi = shifts[shiftflag].binhi;
int nbins = shifts[shiftflag].nbins;
int nbinx = shifts[shiftflag].nbinx;
int nbiny = shifts[shiftflag].nbiny;
BinAve *vbin = shifts[shiftflag].vbin;
// binhead = 1st SRD particle in each bin
// binnext = index of next particle in bin
// bin assignment is done in lamda units for triclinic
int *mask = atom->mask;
double **x = atom->x;
double **v = atom->v;
int nlocal = atom->nlocal;
if (triclinic) domain->x2lamda(nlocal);
for (i = 0; i < nbins; i++) binhead[i] = -1;
for (i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
ix = static_cast<int> ((x[i][0]-corner[0])*bininv1x);
ix = MAX(ix,binlo[0]);
ix = MIN(ix,binhi[0]);
iy = static_cast<int> ((x[i][1]-corner[1])*bininv1y);
iy = MAX(iy,binlo[1]);
iy = MIN(iy,binhi[1]);
iz = static_cast<int> ((x[i][2]-corner[2])*bininv1z);
iz = MAX(iz,binlo[2]);
iz = MIN(iz,binhi[2]);
ibin = (iz-binlo[2])*nbiny*nbinx + (iy-binlo[1])*nbinx + (ix-binlo[0]);
binnext[i] = binhead[ibin];
binhead[ibin] = i;
}
if (triclinic) domain->lamda2x(nlocal);
if (triclinic && streamflag) {
double *h_rate = domain->h_rate;
double *h_ratelo = domain->h_ratelo;
for (i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
xlamda = x[i];
v[i][0] -= h_rate[0]*xlamda[0] + h_rate[5]*xlamda[1] +
h_rate[4]*xlamda[2] + h_ratelo[0];
v[i][1] -= h_rate[1]*xlamda[1] + h_rate[3]*xlamda[2] + h_ratelo[1];
v[i][2] -= h_rate[2]*xlamda[2] + h_ratelo[2];
}
}
// for each bin I have particles contributing to:
// compute ave v and v^2 of particles in that bin
// if I own the bin, set its random value, else set to 0.0
for (i = 0; i < nbins; i++) {
n = 0;
vave[0] = vave[1] = vave[2] = 0.0;
for (j = binhead[i]; j >= 0; j = binnext[j]) {
vx = v[j][0];
vy = v[j][1];
vz = v[j][2];
vave[0] += vx;
vave[1] += vy;
vave[2] += vz;
n++;
}
vbin[i].vave[0] = vave[0];
vbin[i].vave[1] = vave[1];
vbin[i].vave[2] = vave[2];
vbin[i].n = n;
if (vbin[i].owner) vbin[i].random = random->uniform();
else vbin[i].random = 0.0;
}
// communicate bin info for bins which more than 1 proc contribute to
if (shifts[shiftflag].commflag) vbin_comm(shiftflag);
// for each bin I have particles contributing to:
// reassign particle velocity by rotation around a random axis
// accumulate T_srd for each bin I own
// for triclinic, replace mean velocity with stream velocity
srd_bin_temp = 0.0;
srd_bin_count = 0;
if (dimension == 2) axis = 2;
for (i = 0; i < nbins; i++) {
n = vbin[i].n;
if (n == 0) continue;
vold = vbin[i].vave;
vold[0] /= n;
vold[1] /= n;
vold[2] /= n;
vnew = vold;
irandom = static_cast<int> (6.0*vbin[i].random);
sign = irandom % 2;
if (dimension == 3) axis = irandom / 2;
vsq = 0.0;
for (j = binhead[i]; j >= 0; j = binnext[j]) {
if (axis == 0) {
u[0] = v[j][0]-vold[0];
u[1] = sign ? v[j][2]-vold[2] : vold[2]-v[j][2];
u[2] = sign ? vold[1]-v[j][1] : v[j][1]-vold[1];
} else if (axis == 1) {
u[1] = v[j][1]-vold[1];
u[0] = sign ? v[j][2]-vold[2] : vold[2]-v[j][2];
u[2] = sign ? vold[0]-v[j][0] : v[j][0]-vold[0];
} else {
u[2] = v[j][2]-vold[2];
u[1] = sign ? v[j][0]-vold[0] : vold[0]-v[j][0];
u[0] = sign ? vold[1]-v[j][1] : v[j][1]-vold[1];
}
vsq += u[0]*u[0] + u[1]*u[1] + u[2]*u[2];
v[j][0] = u[0] + vnew[0];
v[j][1] = u[1] + vnew[1];
v[j][2] = u[2] + vnew[2];
}
// sum partial contribution of my particles to T even if I don't own bin
// but only count bin if I own it, so that bin is counted exactly once
if (n > 1) {
srd_bin_temp += vsq / (n-1);
if (vbin[i].owner) srd_bin_count++;
}
}
srd_bin_temp *= force->mvv2e * mass_srd / (dimension * force->boltz);
if (triclinic && streamflag) {
double *h_rate = domain->h_rate;
double *h_ratelo = domain->h_ratelo;
for (i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
xlamda = x[i];
v[i][0] += h_rate[0]*xlamda[0] + h_rate[5]*xlamda[1] +
h_rate[4]*xlamda[2] + h_ratelo[0];
v[i][1] += h_rate[1]*xlamda[1] + h_rate[3]*xlamda[2] + h_ratelo[1];
v[i][2] += h_rate[2]*xlamda[2] + h_ratelo[2];
}
}
// rescale any too-large velocities
for (i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
vsq = v[i][0]*v[i][0] + v[i][1]*v[i][1] + v[i][2]*v[i][2];
if (vsq > vmaxsq) nrescale++;
}
}
/* ----------------------------------------------------------------------
communicate summed particle info for bins that overlap 1 or more procs
------------------------------------------------------------------------- */
void FixSRD::vbin_comm(int ishift)
{
BinComm *bcomm1,*bcomm2;
MPI_Request request1,request2;
MPI_Status status;
// send/recv bins in both directions in each dimension
// don't send if nsend = 0
// due to static bins aliging with proc boundary
// due to dynamic bins across non-periodic global boundary
// copy to self if sendproc = me
// MPI send to another proc if sendproc != me
// don't recv if nrecv = 0
// copy from self if recvproc = me
// MPI recv from another proc if recvproc != me
BinAve *vbin = shifts[ishift].vbin;
int *procgrid = comm->procgrid;
int iswap = 0;
for (int idim = 0; idim < dimension; idim++) {
bcomm1 = &shifts[ishift].bcomm[iswap++];
bcomm2 = &shifts[ishift].bcomm[iswap++];
if (procgrid[idim] == 1) {
if (bcomm1->nsend)
vbin_pack(vbin,bcomm1->nsend,bcomm1->sendlist,sbuf1);
if (bcomm2->nsend)
vbin_pack(vbin,bcomm2->nsend,bcomm2->sendlist,sbuf2);
if (bcomm1->nrecv)
vbin_unpack(sbuf1,vbin,bcomm1->nrecv,bcomm1->recvlist);
if (bcomm2->nrecv)
vbin_unpack(sbuf2,vbin,bcomm2->nrecv,bcomm2->recvlist);
} else {
if (bcomm1->nrecv)
MPI_Irecv(rbuf1,bcomm1->nrecv*VBINSIZE,MPI_DOUBLE,bcomm1->recvproc,0,
world,&request1);
if (bcomm2->nrecv)
MPI_Irecv(rbuf2,bcomm2->nrecv*VBINSIZE,MPI_DOUBLE,bcomm2->recvproc,0,
world,&request2);
if (bcomm1->nsend) {
vbin_pack(vbin,bcomm1->nsend,bcomm1->sendlist,sbuf1);
MPI_Send(sbuf1,bcomm1->nsend*VBINSIZE,MPI_DOUBLE,
bcomm1->sendproc,0,world);
}
if (bcomm2->nsend) {
vbin_pack(vbin,bcomm2->nsend,bcomm2->sendlist,sbuf2);
MPI_Send(sbuf2,bcomm2->nsend*VBINSIZE,MPI_DOUBLE,
bcomm2->sendproc,0,world);
}
if (bcomm1->nrecv) {
MPI_Wait(&request1,&status);
vbin_unpack(rbuf1,vbin,bcomm1->nrecv,bcomm1->recvlist);
}
if (bcomm2->nrecv) {
MPI_Wait(&request2,&status);
vbin_unpack(rbuf2,vbin,bcomm2->nrecv,bcomm2->recvlist);
}
}
}
}
/* ----------------------------------------------------------------------
pack velocity bin data into a message buffer for sending
------------------------------------------------------------------------- */
void FixSRD::vbin_pack(BinAve *vbin, int n, int *list, double *buf)
{
int j;
int m = 0;
for (int i = 0; i < n; i++) {
j = list[i];
buf[m++] = vbin[j].n;
buf[m++] = vbin[j].vave[0];
buf[m++] = vbin[j].vave[1];
buf[m++] = vbin[j].vave[2];
buf[m++] = vbin[j].random;
}
}
/* ----------------------------------------------------------------------
unpack velocity bin data from a message buffer and sum values to my bins
------------------------------------------------------------------------- */
void FixSRD::vbin_unpack(double *buf, BinAve *vbin, int n, int *list)
{
int j;
int m = 0;
for (int i = 0; i < n; i++) {
j = list[i];
vbin[j].n += static_cast<int> (buf[m++]);
vbin[j].vave[0] += buf[m++];
vbin[j].vave[1] += buf[m++];
vbin[j].vave[2] += buf[m++];
vbin[j].random += buf[m++];
}
}
/* ----------------------------------------------------------------------
detect all collisions between SRD and BIG particles or WALLS
assume SRD can be inside at most one BIG particle or WALL at a time
unoverlap SRDs for each collision
------------------------------------------------------------------------- */
void FixSRD::collisions_single()
{
int i,j,k,m,type,mbig,ibin,ibounce,inside,collide_flag;
double dt,t_remain;
double norm[3],xscoll[3],xbcoll[3],vsnew[3];
Big *big;
// outer loop over SRD particles
// inner loop over BIG particles or WALLS that overlap SRD particle bin
// if overlap between SRD and BIG particle or wall:
// for exact, compute collision pt in time
// for inexact, push SRD to surf of BIG particle or WALL
// update x,v of SRD and f,torque on BIG particle
// re-bin SRD particle after collision
// iterate until the SRD particle has no overlaps with BIG particles or WALLS
double **x = atom->x;
double **v = atom->v;
double **f = atom->f;
double **torque = atom->torque;
int *mask = atom->mask;
int nlocal = atom->nlocal;
for (i = 0; i < nlocal; i++) {
if (!(mask[i] & groupbit)) continue;
ibin = binsrd[i];
if (nbinbig[ibin] == 0) continue;
ibounce = 0;
collide_flag = 1;
dt = dt_big;
while (collide_flag) {
mbig = nbinbig[ibin];
if (ibounce == 0) ncheck += mbig;
collide_flag = 0;
for (m = 0; m < mbig; m++) {
k = binbig[ibin][m];
big = &biglist[k];
j = big->index;
type = big->type;
if (type == SPHERE) inside = inside_sphere(x[i],x[j],big);
else if (type == ELLIPSOID) inside = inside_ellipsoid(x[i],x[j],big);
else inside = inside_wall(x[i],j);
if (inside) {
if (exactflag) {
if (type == SPHERE)
t_remain = collision_sphere_exact(x[i],x[j],v[i],v[j],big,
xscoll,xbcoll,norm);
else if (type == ELLIPSOID)
t_remain = collision_ellipsoid_exact(x[i],x[j],v[i],v[j],big,
xscoll,xbcoll,norm);
else
t_remain = collision_wall_exact(x[i],j,v[i],xscoll,xbcoll,norm);
} else {
t_remain = 0.5*dt;
if (type == SPHERE)
collision_sphere_inexact(x[i],x[j],big,xscoll,xbcoll,norm);
else if (type == ELLIPSOID)
collision_ellipsoid_inexact(x[i],x[j],big,xscoll,xbcoll,norm);
else
collision_wall_inexact(x[i],j,xscoll,xbcoll,norm);
}
#ifdef SRD_DEBUG
if (update->ntimestep == SRD_DEBUG_TIMESTEP &&
atom->tag[i] == SRD_DEBUG_ATOMID)
print_collision(i,j,ibounce,t_remain,dt,xscoll,xbcoll,norm,type);
#endif
if (t_remain > dt) {
ninside++;
if (insideflag == INSIDE_ERROR || insideflag == INSIDE_WARN) {
char str[128];
if (type != WALL)
sprintf(str,
"SRD particle %d started "
"inside big particle %d on step " BIGINT_FORMAT
" bounce %d\n",
atom->tag[i],atom->tag[j],update->ntimestep,ibounce+1);
else
sprintf(str,
"SRD particle %d started "
"inside big particle %d on step " BIGINT_FORMAT
" bounce %d\n",
atom->tag[i],atom->tag[j],update->ntimestep,ibounce+1);
if (insideflag == INSIDE_ERROR) error->one(str);
error->warning(str);
}
break;
}
if (collidestyle == SLIP) {
if (type == SPHERE)
slip_sphere(v[i],v[j],norm,vsnew);
else if (type == ELLIPSOID)
slip_ellipsoid(v[i],v[j],x[j],big,xscoll,norm,vsnew);
else
slip_wall(v[i],j,norm,vsnew);
} else {
if (type != WALL)
noslip(v[i],v[j],x[j],big,xscoll,norm,vsnew);
else
noslip_wall(v[i],j,xscoll,norm,vsnew);
}
if (dimension == 2) vsnew[2] = 0.0;
// check on rescaling of vsnew
double vsq = vsnew[0]*vsnew[0] + vsnew[1]*vsnew[1] +
vsnew[2]*vsnew[2];
if (vsq > vmaxsq) nrescale++;
// update BIG particle and WALL and SRD
// BIG particle is not torqued if sphere and SLIP collision
if (collidestyle == SLIP && type == SPHERE)
force_torque(v[i],vsnew,xscoll,xbcoll,f[j],NULL);
else if (type != WALL)
force_torque(v[i],vsnew,xscoll,xbcoll,f[j],torque[j]);
else if (type == WALL)
force_wall(v[i],vsnew,j);
ibin = binsrd[i] = update_srd(i,t_remain,xscoll,vsnew,x[i],v[i]);
if (ibounce == 0) ncollide++;
ibounce++;
if (ibounce < maxbounceallow || maxbounceallow == 0)
collide_flag = 1;
dt = t_remain;
break;
}
}
}
nbounce += ibounce;
if (maxbounceallow && ibounce >= maxbounceallow) bouncemaxnum++;
if (ibounce > bouncemax) bouncemax = ibounce;
}
}
/* ----------------------------------------------------------------------
detect all collisions between SRD and big particles
an SRD can be inside more than one big particle at a time
requires finding which big particle SRD collided with first
unoverlap SRDs for each collision
------------------------------------------------------------------------- */
void FixSRD::collisions_multi()
{
int i,j,k,m,type,mbig,ibin,ibounce,inside,jfirst,typefirst;
double dt,t_remain,t_first;
double norm[3],xscoll[3],xbcoll[3],vsnew[3];
double normfirst[3],xscollfirst[3],xbcollfirst[3];
Big *big;
// outer loop over SRD particles
// inner loop over BIG particles or WALLS that overlap SRD particle bin
// loop over all BIG and WALLS to find which one SRD collided with first
// if overlap between SRD and BIG particle or wall:
// compute collision pt in time
// update x,v of SRD and f,torque on BIG particle
// re-bin SRD particle after collision
// iterate until the SRD particle has no overlaps with BIG particles or WALLS
double **x = atom->x;
double **v = atom->v;
double **f = atom->f;
double **torque = atom->torque;
int *mask = atom->mask;
int nlocal = atom->nlocal;
for (i = 0; i < nlocal; i++) {
if (!(mask[i] & groupbit)) continue;
ibin = binsrd[i];
if (nbinbig[ibin] == 0) continue;
ibounce = 0;
dt = dt_big;
while (1) {
mbig = nbinbig[ibin];
if (ibounce == 0) ncheck += mbig;
t_first = 0.0;
for (m = 0; m < mbig; m++) {
k = binbig[ibin][m];
big = &biglist[k];
j = big->index;
type = big->type;
if (type == SPHERE) inside = inside_sphere(x[i],x[j],big);
else if (type == ELLIPSOID) inside = inside_ellipsoid(x[i],x[j],big);
else inside = inside_wall(x[i],j);
if (inside) {
if (type == SPHERE)
t_remain = collision_sphere_exact(x[i],x[j],v[i],v[j],big,
xscoll,xbcoll,norm);
else if (type == ELLIPSOID)
t_remain = collision_ellipsoid_exact(x[i],x[j],v[i],v[j],big,
xscoll,xbcoll,norm);
else
t_remain = collision_wall_exact(x[i],j,v[i],xscoll,xbcoll,norm);
#ifdef SRD_DEBUG
if (update->ntimestep == SRD_DEBUG_TIMESTEP &&
atom->tag[i] == SRD_DEBUG_ATOMID)
print_collision(i,j,ibounce,t_remain,dt,xscoll,xbcoll,norm,type);
#endif
if (t_remain > dt || t_remain < 0.0) {
ninside++;
if (insideflag == INSIDE_ERROR || insideflag == INSIDE_WARN) {
char str[128];
sprintf(str,
"SRD particle %d started "
"inside big particle %d on step " BIGINT_FORMAT
" bounce %d\n",
atom->tag[i],atom->tag[j],update->ntimestep,ibounce+1);
if (insideflag == INSIDE_ERROR) error->one(str);
error->warning(str);
}
t_first = 0.0;
break;
}
if (t_remain > t_first) {
t_first = t_remain;
jfirst = j;
typefirst = type;
xscollfirst[0] = xscoll[0];
xscollfirst[1] = xscoll[1];
xscollfirst[2] = xscoll[2];
xbcollfirst[0] = xbcoll[0];
xbcollfirst[1] = xbcoll[1];
xbcollfirst[2] = xbcoll[2];
normfirst[0] = norm[0];
normfirst[1] = norm[1];
normfirst[2] = norm[2];
}
}
}
if (t_first == 0.0) break;
j = jfirst;
type = typefirst;
xscoll[0] = xscollfirst[0];
xscoll[1] = xscollfirst[1];
xscoll[2] = xscollfirst[2];
xbcoll[0] = xbcollfirst[0];
xbcoll[1] = xbcollfirst[1];
xbcoll[2] = xbcollfirst[2];
norm[0] = normfirst[0];
norm[1] = normfirst[1];
norm[2] = normfirst[2];
if (collidestyle == SLIP) {
if (type == SPHERE)
slip_sphere(v[i],v[j],norm,vsnew);
else if (type == ELLIPSOID)
slip_ellipsoid(v[i],v[j],x[j],big,xscoll,norm,vsnew);
else
slip_wall(v[i],j,norm,vsnew);
} else {
if (type != WALL)
noslip(v[i],v[j],x[j],big,xscoll,norm,vsnew);
else
noslip_wall(v[i],j,xscoll,norm,vsnew);
}
if (dimension == 2) vsnew[2] = 0.0;
// check on rescaling of vsnew
double vsq = vsnew[0]*vsnew[0] + vsnew[1]*vsnew[1] + vsnew[2]*vsnew[2];
if (vsq > vmaxsq) nrescale++;
// update BIG particle and WALL and SRD
// BIG particle is not torqued if sphere and SLIP collision
if (collidestyle == SLIP && type == SPHERE)
force_torque(v[i],vsnew,xscoll,xbcoll,f[j],NULL);
else if (type != WALL)
force_torque(v[i],vsnew,xscoll,xbcoll,f[j],torque[j]);
else if (type == WALL)
force_wall(v[i],vsnew,j);
ibin = binsrd[i] = update_srd(i,t_first,xscoll,vsnew,x[i],v[i]);
if (ibounce == 0) ncollide++;
ibounce++;
if (ibounce == maxbounceallow) break;
dt = t_first;
}
nbounce += ibounce;
if (maxbounceallow && ibounce >= maxbounceallow) bouncemaxnum++;
if (ibounce > bouncemax) bouncemax = ibounce;
}
}
/* ----------------------------------------------------------------------
check if SRD particle S is inside spherical big particle B
------------------------------------------------------------------------- */
int FixSRD::inside_sphere(double *xs, double *xb, Big *big)
{
double dx,dy,dz;
dx = xs[0] - xb[0];
dy = xs[1] - xb[1];
dz = xs[2] - xb[2];
if (dx*dx + dy*dy + dz*dz <= big->radsq) return 1;
return 0;
}
/* ----------------------------------------------------------------------
check if SRD particle S is inside ellipsoidal big particle B
------------------------------------------------------------------------- */
int FixSRD::inside_ellipsoid(double *xs, double *xb, Big *big)
{
double x,y,z;
double *ex = big->ex;
double *ey = big->ey;
double *ez = big->ez;
double xs_xb[3];
xs_xb[0] = xs[0] - xb[0];
xs_xb[1] = xs[1] - xb[1];
xs_xb[2] = xs[2] - xb[2];
x = xs_xb[0]*ex[0] + xs_xb[1]*ex[1] + xs_xb[2]*ex[2];
y = xs_xb[0]*ey[0] + xs_xb[1]*ey[1] + xs_xb[2]*ey[2];
z = xs_xb[0]*ez[0] + xs_xb[1]*ez[1] + xs_xb[2]*ez[2];
if (x*x*big->aradsqinv + y*y*big->bradsqinv + z*z*big->cradsqinv <= 1.0)
return 1;
return 0;
}
/* ----------------------------------------------------------------------
check if SRD particle S is inside wall IWALL
------------------------------------------------------------------------- */
int FixSRD::inside_wall(double *xs, int iwall)
{
int dim = wallwhich[iwall] / 2;
int side = wallwhich[iwall] % 2;
if (side == 0 && xs[dim] < xwall[iwall]) return 1;
if (side && xs[dim] > xwall[iwall]) return 1;
return 0;
}
/* ----------------------------------------------------------------------
collision of SRD particle S with surface of spherical big particle B
exact because compute time of collision
dt = time previous to now at which collision occurs
xscoll = collision pt = position of SRD at time of collision
xbcoll = position of big particle at time of collision
norm = surface normal of collision pt at time of collision
------------------------------------------------------------------------- */
double FixSRD::collision_sphere_exact(double *xs, double *xb,
double *vs, double *vb, Big *big,
double *xscoll, double *xbcoll,
double *norm)
{
double vs_dot_vs,vb_dot_vb,vs_dot_vb;
double vs_dot_xb,vb_dot_xs,vs_dot_xs,vb_dot_xb;
double xs_dot_xs,xb_dot_xb,xs_dot_xb;
double a,b,c,scale,dt;
vs_dot_vs = vs[0]*vs[0] + vs[1]*vs[1] + vs[2]*vs[2];
vb_dot_vb = vb[0]*vb[0] + vb[1]*vb[1] + vb[2]*vb[2];
vs_dot_vb = vs[0]*vb[0] + vs[1]*vb[1] + vs[2]*vb[2];
vs_dot_xb = vs[0]*xb[0] + vs[1]*xb[1] + vs[2]*xb[2];
vb_dot_xs = vb[0]*xs[0] + vb[1]*xs[1] + vb[2]*xs[2];
vs_dot_xs = vs[0]*xs[0] + vs[1]*xs[1] + vs[2]*xs[2];
vb_dot_xb = vb[0]*xb[0] + vb[1]*xb[1] + vb[2]*xb[2];
xs_dot_xs = xs[0]*xs[0] + xs[1]*xs[1] + xs[2]*xs[2];
xb_dot_xb = xb[0]*xb[0] + xb[1]*xb[1] + xb[2]*xb[2];
xs_dot_xb = xs[0]*xb[0] + xs[1]*xb[1] + xs[2]*xb[2];
a = vs_dot_vs + vb_dot_vb - 2.0*vs_dot_vb;
b = 2.0 * (vs_dot_xb + vb_dot_xs - vs_dot_xs - vb_dot_xb);
c = xs_dot_xs + xb_dot_xb - 2.0*xs_dot_xb - big->radsq;
dt = (-b + sqrt(b*b - 4.0*a*c)) / (2.0*a);
xscoll[0] = xs[0] - dt*vs[0];
xscoll[1] = xs[1] - dt*vs[1];
xscoll[2] = xs[2] - dt*vs[2];
xbcoll[0] = xb[0] - dt*vb[0];
xbcoll[1] = xb[1] - dt*vb[1];
xbcoll[2] = xb[2] - dt*vb[2];
norm[0] = xscoll[0] - xbcoll[0];
norm[1] = xscoll[1] - xbcoll[1];
norm[2] = xscoll[2] - xbcoll[2];
scale = 1.0/sqrt(norm[0]*norm[0] + norm[1]*norm[1] + norm[2]*norm[2]);
norm[0] *= scale;
norm[1] *= scale;
norm[2] *= scale;
return dt;
}
/* ----------------------------------------------------------------------
collision of SRD particle S with surface of spherical big particle B
inexact because just push SRD to surface of big particle at end of step
time of collision = end of step
xscoll = collision pt = position of SRD at time of collision
xbcoll = xb = position of big particle at time of collision
norm = surface normal of collision pt at time of collision
------------------------------------------------------------------------- */
void FixSRD::collision_sphere_inexact(double *xs, double *xb,
Big *big,
double *xscoll, double *xbcoll,
double *norm)
{
double scale;
norm[0] = xs[0] - xb[0];
norm[1] = xs[1] - xb[1];
norm[2] = xs[2] - xb[2];
scale = 1.0/sqrt(norm[0]*norm[0] + norm[1]*norm[1] + norm[2]*norm[2]);
norm[0] *= scale;
norm[1] *= scale;
norm[2] *= scale;
xscoll[0] = xb[0] + big->radius*norm[0];
xscoll[1] = xb[1] + big->radius*norm[1];
xscoll[2] = xb[2] + big->radius*norm[2];
xbcoll[0] = xb[0];
xbcoll[1] = xb[1];
xbcoll[2] = xb[2];
}
/* ----------------------------------------------------------------------
collision of SRD particle S with surface of ellipsoidal big particle B
exact because compute time of collision
dt = time previous to now at which collision occurs
xscoll = collision pt = position of SRD at time of collision
xbcoll = position of big particle at time of collision
norm = surface normal of collision pt at time of collision
------------------------------------------------------------------------- */
double FixSRD::collision_ellipsoid_exact(double *xs, double *xb,
double *vs, double *vb, Big *big,
double *xscoll, double *xbcoll,
double *norm)
{
double vs_vb[3],xs_xb[3],omega_ex[3],omega_ey[3],omega_ez[3];
double excoll[3],eycoll[3],ezcoll[3],delta[3],xbody[3],nbody[3];
double ax,bx,cx,ay,by,cy,az,bz,cz;
double a,b,c,dt,scale;
double *omega = big->omega;
double *ex = big->ex;
double *ey = big->ey;
double *ez = big->ez;
vs_vb[0] = vs[0]-vb[0]; vs_vb[1] = vs[1]-vb[1]; vs_vb[2] = vs[2]-vb[2];
xs_xb[0] = xs[0]-xb[0]; xs_xb[1] = xs[1]-xb[1]; xs_xb[2] = xs[2]-xb[2];
omega_ex[0] = omega[1]*ex[2] - omega[2]*ex[1];
omega_ex[1] = omega[2]*ex[0] - omega[0]*ex[2];
omega_ex[2] = omega[0]*ex[1] - omega[1]*ex[0];
omega_ey[0] = omega[1]*ey[2] - omega[2]*ey[1];
omega_ey[1] = omega[2]*ey[0] - omega[0]*ey[2];
omega_ey[2] = omega[0]*ey[1] - omega[1]*ey[0];
omega_ez[0] = omega[1]*ez[2] - omega[2]*ez[1];
omega_ez[1] = omega[2]*ez[0] - omega[0]*ez[2];
omega_ez[2] = omega[0]*ez[1] - omega[1]*ez[0];
ax = vs_vb[0]*omega_ex[0] + vs_vb[1]*omega_ex[1] + vs_vb[2]*omega_ex[2];
bx = -(vs_vb[0]*ex[0] + vs_vb[1]*ex[1] + vs_vb[2]*ex[2]);
bx -= xs_xb[0]*omega_ex[0] + xs_xb[1]*omega_ex[1] + xs_xb[2]*omega_ex[2];
cx = xs_xb[0]*ex[0] + xs_xb[1]*ex[1] + xs_xb[2]*ex[2];
ay = vs_vb[0]*omega_ey[0] + vs_vb[1]*omega_ey[1] + vs_vb[2]*omega_ey[2];
by = -(vs_vb[0]*ey[0] + vs_vb[1]*ey[1] + vs_vb[2]*ey[2]);
by -= xs_xb[0]*omega_ey[0] + xs_xb[1]*omega_ey[1] + xs_xb[2]*omega_ey[2];
cy = xs_xb[0]*ey[0] + xs_xb[1]*ey[1] + xs_xb[2]*ey[2];
az = vs_vb[0]*omega_ez[0] + vs_vb[1]*omega_ez[1] + vs_vb[2]*omega_ez[2];
bz = -(vs_vb[0]*ez[0] + vs_vb[1]*ez[1] + vs_vb[2]*ez[2]);
bz -= xs_xb[0]*omega_ez[0] + xs_xb[1]*omega_ez[1] + xs_xb[2]*omega_ez[2];
cz = xs_xb[0]*ez[0] + xs_xb[1]*ez[1] + xs_xb[2]*ez[2];
a = (bx*bx + 2.0*ax*cx)*big->aradsqinv +
(by*by + 2.0*ay*cy)*big->bradsqinv +
(bz*bz + 2.0*az*cz)*big->cradsqinv;
b = 2.0 * (bx*cx*big->aradsqinv + by*cy*big->bradsqinv +
bz*cz*big->cradsqinv);
c = cx*cx*big->aradsqinv + cy*cy*big->bradsqinv +
cz*cz*big->cradsqinv - 1.0;
dt = (-b + sqrt(b*b - 4.0*a*c)) / (2.0*a);
xscoll[0] = xs[0] - dt*vs[0];
xscoll[1] = xs[1] - dt*vs[1];
xscoll[2] = xs[2] - dt*vs[2];
xbcoll[0] = xb[0] - dt*vb[0];
xbcoll[1] = xb[1] - dt*vb[1];
xbcoll[2] = xb[2] - dt*vb[2];
// calculate normal to ellipsoid at collision pt
// Excoll,Eycoll,Ezcoll = orientation of ellipsoid at collision time
// nbody = normal in body frame of ellipsoid (Excoll,Eycoll,Ezcoll)
// norm = normal in space frame
// only worry about normalizing final norm vector
excoll[0] = ex[0] - dt * (omega[1]*ex[2] - omega[2]*ex[1]);
excoll[1] = ex[1] - dt * (omega[2]*ex[0] - omega[0]*ex[2]);
excoll[2] = ex[2] - dt * (omega[0]*ex[1] - omega[1]*ex[0]);
eycoll[0] = ey[0] - dt * (omega[1]*ey[2] - omega[2]*ey[1]);
eycoll[1] = ey[1] - dt * (omega[2]*ey[0] - omega[0]*ey[2]);
eycoll[2] = ey[2] - dt * (omega[0]*ey[1] - omega[1]*ey[0]);
ezcoll[0] = ez[0] - dt * (omega[1]*ez[2] - omega[2]*ez[1]);
ezcoll[1] = ez[1] - dt * (omega[2]*ez[0] - omega[0]*ez[2]);
ezcoll[2] = ez[2] - dt * (omega[0]*ez[1] - omega[1]*ez[0]);
delta[0] = xscoll[0] - xbcoll[0];
delta[1] = xscoll[1] - xbcoll[1];
delta[2] = xscoll[2] - xbcoll[2];
xbody[0] = delta[0]*excoll[0] + delta[1]*excoll[1] + delta[2]*excoll[2];
xbody[1] = delta[0]*eycoll[0] + delta[1]*eycoll[1] + delta[2]*eycoll[2];
xbody[2] = delta[0]*ezcoll[0] + delta[1]*ezcoll[1] + delta[2]*ezcoll[2];
nbody[0] = xbody[0]*big->aradsqinv;
nbody[1] = xbody[1]*big->bradsqinv;
nbody[2] = xbody[2]*big->cradsqinv;
norm[0] = excoll[0]*nbody[0] + eycoll[0]*nbody[1] + ezcoll[0]*nbody[2];
norm[1] = excoll[1]*nbody[0] + eycoll[1]*nbody[1] + ezcoll[1]*nbody[2];
norm[2] = excoll[2]*nbody[0] + eycoll[2]*nbody[1] + ezcoll[2]*nbody[2];
scale = 1.0/sqrt(norm[0]*norm[0] + norm[1]*norm[1] + norm[2]*norm[2]);
norm[0] *= scale;
norm[1] *= scale;
norm[2] *= scale;
return dt;
}
/* ----------------------------------------------------------------------
collision of SRD particle S with surface of ellipsoidal big particle B
inexact because just push SRD to surface of big particle at end of step
------------------------------------------------------------------------- */
void FixSRD::collision_ellipsoid_inexact(double *xs, double *xb,
Big *big,
double *xscoll, double *xbcoll,
double *norm)
{
double xs_xb[3],delta[3],xbody[3],nbody[3];
double x,y,z,scale;
double *ex = big->ex;
double *ey = big->ey;
double *ez = big->ez;
xs_xb[0] = xs[0] - xb[0];
xs_xb[1] = xs[1] - xb[1];
xs_xb[2] = xs[2] - xb[2];
x = xs_xb[0]*ex[0] + xs_xb[1]*ex[1] + xs_xb[2]*ex[2];
y = xs_xb[0]*ey[0] + xs_xb[1]*ey[1] + xs_xb[2]*ey[2];
z = xs_xb[0]*ez[0] + xs_xb[1]*ez[1] + xs_xb[2]*ez[2];
scale = 1.0/sqrt(x*x*big->aradsqinv + y*y*big->bradsqinv +
z*z*big->cradsqinv);
x *= scale;
y *= scale;
z *= scale;
xscoll[0] = x*ex[0] + y*ey[0] + z*ez[0] + xb[0];
xscoll[1] = x*ex[1] + y*ey[1] + z*ez[1] + xb[1];
xscoll[2] = x*ex[2] + y*ey[2] + z*ez[2] + xb[2];
xbcoll[0] = xb[0];
xbcoll[1] = xb[1];
xbcoll[2] = xb[2];
// calculate normal to ellipsoid at collision pt
// nbody = normal in body frame of ellipsoid
// norm = normal in space frame
// only worry about normalizing final norm vector
delta[0] = xscoll[0] - xbcoll[0];
delta[1] = xscoll[1] - xbcoll[1];
delta[2] = xscoll[2] - xbcoll[2];
xbody[0] = delta[0]*ex[0] + delta[1]*ex[1] + delta[2]*ex[2];
xbody[1] = delta[0]*ey[0] + delta[1]*ey[1] + delta[2]*ey[2];
xbody[2] = delta[0]*ez[0] + delta[1]*ez[1] + delta[2]*ez[2];
nbody[0] = xbody[0]*big->aradsqinv;
nbody[1] = xbody[1]*big->bradsqinv;
nbody[2] = xbody[2]*big->cradsqinv;
norm[0] = ex[0]*nbody[0] + ey[0]*nbody[1] + ez[0]*nbody[2];
norm[1] = ex[1]*nbody[0] + ey[1]*nbody[1] + ez[1]*nbody[2];
norm[2] = ex[2]*nbody[0] + ey[2]*nbody[1] + ez[2]*nbody[2];
scale = 1.0/sqrt(norm[0]*norm[0] + norm[1]*norm[1] + norm[2]*norm[2]);
norm[0] *= scale;
norm[1] *= scale;
norm[2] *= scale;
}
/* ----------------------------------------------------------------------
collision of SRD particle S with wall IWALL
exact because compute time of collision
dt = time previous to now at which collision occurs
xscoll = collision pt = position of SRD at time of collision
norm = surface normal of collision pt at time of collision
------------------------------------------------------------------------- */
double FixSRD::collision_wall_exact(double *xs, int iwall, double *vs,
double *xscoll, double *xbcoll,
double *norm)
{
int dim = wallwhich[iwall] / 2;
double dt = (xs[dim] - xwall[iwall]) / (vs[dim] - vwall[iwall]);
xscoll[0] = xs[0] - dt*vs[0];
xscoll[1] = xs[1] - dt*vs[1];
xscoll[2] = xs[2] - dt*vs[2];
xbcoll[0] = xbcoll[1] = xbcoll[2] = 0.0;
xbcoll[dim] = xwall[iwall] - dt*vwall[iwall];
int side = wallwhich[iwall] % 2;
norm[0] = norm[1] = norm[2] = 0.0;
if (side == 0) norm[dim] = 1.0;
else norm[dim] = -1.0;
return dt;
}
/* ----------------------------------------------------------------------
collision of SRD particle S with wall IWALL
inexact because just push SRD to surface of wall at end of step
time of collision = end of step
xscoll = collision pt = position of SRD at time of collision
norm = surface normal of collision pt at time of collision
------------------------------------------------------------------------- */
void FixSRD::collision_wall_inexact(double *xs, int iwall, double *xscoll,
double *xbcoll, double *norm)
{
int dim = wallwhich[iwall] / 2;
xscoll[0] = xs[0];
xscoll[1] = xs[1];
xscoll[2] = xs[2];
xscoll[dim] = xwall[iwall];
xbcoll[0] = xbcoll[1] = xbcoll[2] = 0.0;
xbcoll[dim] = xwall[iwall];
int side = wallwhich[iwall] % 2;
norm[0] = norm[1] = norm[2] = 0.0;
if (side == 0) norm[dim] = 1.0;
else norm[dim] = -1.0;
}
/* ----------------------------------------------------------------------
SLIP collision with sphere
vs = velocity of SRD, vb = velocity of BIG
norm = unit normal from surface of BIG at collision pt
v of BIG particle in direction of surf normal is added to v of SRD
return vsnew of SRD
------------------------------------------------------------------------- */
void FixSRD::slip_sphere(double *vs, double *vb, double *norm, double *vsnew)
{
double r1,r2,vnmag,vs_dot_n,vsurf_dot_n;
double tangent[3];
while (1) {
r1 = sigma * random->gaussian();
r2 = sigma * random->gaussian();
vnmag = sqrt(r1*r1 + r2*r2);
if (vnmag*vnmag <= vmaxsq) break;
}
vs_dot_n = vs[0]*norm[0] + vs[1]*norm[1] + vs[2]*norm[2];
tangent[0] = vs[0] - vs_dot_n*norm[0];
tangent[1] = vs[1] - vs_dot_n*norm[1];
tangent[2] = vs[2] - vs_dot_n*norm[2];
// vsurf = velocity of collision pt = translation/rotation of BIG particle
// for sphere, only vb (translation) can contribute in normal direction
vsurf_dot_n = vb[0]*norm[0] + vb[1]*norm[1] + vb[2]*norm[2];
vsnew[0] = (vnmag+vsurf_dot_n)*norm[0] + tangent[0];
vsnew[1] = (vnmag+vsurf_dot_n)*norm[1] + tangent[1];
vsnew[2] = (vnmag+vsurf_dot_n)*norm[2] + tangent[2];
}
/* ----------------------------------------------------------------------
SLIP collision with ellipsoid
vs = velocity of SRD, vb = velocity of BIG
xb = position of BIG, omega = rotation of BIG
xsurf = collision pt on surf of BIG
norm = unit normal from surface of BIG at collision pt
v of BIG particle in direction of surf normal is added to v of SRD
includes component due to rotation of ellipsoid
return vsnew of SRD
------------------------------------------------------------------------- */
void FixSRD::slip_ellipsoid(double *vs, double *vb, double *xb, Big *big,
double *xsurf, double *norm, double *vsnew)
{
double r1,r2,vnmag,vs_dot_n,vsurf_dot_n;
double tangent[3],vsurf[3];
double *omega = big->omega;
while (1) {
r1 = sigma * random->gaussian();
r2 = sigma * random->gaussian();
vnmag = sqrt(r1*r1 + r2*r2);
if (vnmag*vnmag <= vmaxsq) break;
}
vs_dot_n = vs[0]*norm[0] + vs[1]*norm[1] + vs[2]*norm[2];
tangent[0] = vs[0] - vs_dot_n*norm[0];
tangent[1] = vs[1] - vs_dot_n*norm[1];
tangent[2] = vs[2] - vs_dot_n*norm[2];
// vsurf = velocity of collision pt = translation/rotation of BIG particle
vsurf[0] = vb[0] + omega[1]*(xsurf[2]-xb[2]) - omega[2]*(xsurf[1]-xb[1]);
vsurf[1] = vb[1] + omega[2]*(xsurf[0]-xb[0]) - omega[0]*(xsurf[2]-xb[2]);
vsurf[2] = vb[2] + omega[0]*(xsurf[1]-xb[1]) - omega[1]*(xsurf[0]-xb[0]);
vsurf_dot_n = vsurf[0]*norm[0] + vsurf[1]*norm[1] + vsurf[2]*norm[2];
vsnew[0] = (vnmag+vsurf_dot_n)*norm[0] + tangent[0];
vsnew[1] = (vnmag+vsurf_dot_n)*norm[1] + tangent[1];
vsnew[2] = (vnmag+vsurf_dot_n)*norm[2] + tangent[2];
}
/* ----------------------------------------------------------------------
SLIP collision with wall IWALL
vs = velocity of SRD
norm = unit normal from WALL at collision pt
v of WALL in direction of surf normal is added to v of SRD
return vsnew of SRD
------------------------------------------------------------------------- */
void FixSRD::slip_wall(double *vs, int iwall, double *norm, double *vsnew)
{
double vs_dot_n,scale,r1,r2,vnmag,vtmag1,vtmag2;
double tangent1[3],tangent2[3];
vs_dot_n = vs[0]*norm[0] + vs[1]*norm[1] + vs[2]*norm[2];
tangent1[0] = vs[0] - vs_dot_n*norm[0];
tangent1[1] = vs[1] - vs_dot_n*norm[1];
tangent1[2] = vs[2] - vs_dot_n*norm[2];
scale = 1.0/sqrt(tangent1[0]*tangent1[0] + tangent1[1]*tangent1[1] +
tangent1[2]*tangent1[2]);
tangent1[0] *= scale;
tangent1[1] *= scale;
tangent1[2] *= scale;
tangent2[0] = norm[1]*tangent1[2] - norm[2]*tangent1[1];
tangent2[1] = norm[2]*tangent1[0] - norm[0]*tangent1[2];
tangent2[2] = norm[0]*tangent1[1] - norm[1]*tangent1[0];
while (1) {
r1 = sigma * random->gaussian();
r2 = sigma * random->gaussian();
vnmag = sqrt(r1*r1 + r2*r2);
vtmag1 = sigma * random->gaussian();
vtmag2 = sigma * random->gaussian();
if (vnmag*vnmag + vtmag1*vtmag1 + vtmag2*vtmag2 <= vmaxsq) break;
}
vsnew[0] = vnmag*norm[0] + vtmag1*tangent1[0] + vtmag2*tangent2[0];
vsnew[1] = vnmag*norm[1] + vtmag1*tangent1[1] + vtmag2*tangent2[1];
vsnew[2] = vnmag*norm[2] + vtmag1*tangent1[2] + vtmag2*tangent2[2];
// add in velocity of collision pt = velocity of wall
int dim = wallwhich[iwall] / 2;
vsnew[dim] += vwall[iwall];
}
/* ----------------------------------------------------------------------
NO-SLIP collision with sphere or ellipsoid
vs = velocity of SRD, vb = velocity of BIG
xb = position of BIG, omega = rotation of BIG
xsurf = collision pt on surf of BIG
norm = unit normal from surface of BIG at collision pt
v of collision pt is added to v of SRD
includes component due to rotation of BIG
return vsnew of SRD
------------------------------------------------------------------------- */
void FixSRD::noslip(double *vs, double *vb, double *xb, Big *big,
double *xsurf, double *norm, double *vsnew)
{
double vs_dot_n,scale,r1,r2,vnmag,vtmag1,vtmag2;
double tangent1[3],tangent2[3];
double *omega = big->omega;
vs_dot_n = vs[0]*norm[0] + vs[1]*norm[1] + vs[2]*norm[2];
tangent1[0] = vs[0] - vs_dot_n*norm[0];
tangent1[1] = vs[1] - vs_dot_n*norm[1];
tangent1[2] = vs[2] - vs_dot_n*norm[2];
scale = 1.0/sqrt(tangent1[0]*tangent1[0] + tangent1[1]*tangent1[1] +
tangent1[2]*tangent1[2]);
tangent1[0] *= scale;
tangent1[1] *= scale;
tangent1[2] *= scale;
tangent2[0] = norm[1]*tangent1[2] - norm[2]*tangent1[1];
tangent2[1] = norm[2]*tangent1[0] - norm[0]*tangent1[2];
tangent2[2] = norm[0]*tangent1[1] - norm[1]*tangent1[0];
while (1) {
r1 = sigma * random->gaussian();
r2 = sigma * random->gaussian();
vnmag = sqrt(r1*r1 + r2*r2);
vtmag1 = sigma * random->gaussian();
vtmag2 = sigma * random->gaussian();
if (vnmag*vnmag + vtmag1*vtmag1 + vtmag2*vtmag2 <= vmaxsq) break;
}
vsnew[0] = vnmag*norm[0] + vtmag1*tangent1[0] + vtmag2*tangent2[0];
vsnew[1] = vnmag*norm[1] + vtmag1*tangent1[1] + vtmag2*tangent2[1];
vsnew[2] = vnmag*norm[2] + vtmag1*tangent1[2] + vtmag2*tangent2[2];
// add in velocity of collision pt = translation/rotation of BIG particle
vsnew[0] += vb[0] + omega[1]*(xsurf[2]-xb[2]) - omega[2]*(xsurf[1]-xb[1]);
vsnew[1] += vb[1] + omega[2]*(xsurf[0]-xb[0]) - omega[0]*(xsurf[2]-xb[2]);
vsnew[2] += vb[2] + omega[0]*(xsurf[1]-xb[1]) - omega[1]*(xsurf[0]-xb[0]);
}
/* ----------------------------------------------------------------------
NO-SLIP collision with wall IWALL
vs = velocity of SRD
xsurf = collision pt on surf of WALL
norm = unit normal from WALL at collision pt
v of collision pt is added to v of SRD
return vsnew of SRD
------------------------------------------------------------------------- */
void FixSRD::noslip_wall(double *vs, int iwall,
double *xsurf, double *norm, double *vsnew)
{
double vs_dot_n,scale,r1,r2,vnmag,vtmag1,vtmag2;
double tangent1[3],tangent2[3];
vs_dot_n = vs[0]*norm[0] + vs[1]*norm[1] + vs[2]*norm[2];
tangent1[0] = vs[0] - vs_dot_n*norm[0];
tangent1[1] = vs[1] - vs_dot_n*norm[1];
tangent1[2] = vs[2] - vs_dot_n*norm[2];
scale = 1.0/sqrt(tangent1[0]*tangent1[0] + tangent1[1]*tangent1[1] +
tangent1[2]*tangent1[2]);
tangent1[0] *= scale;
tangent1[1] *= scale;
tangent1[2] *= scale;
tangent2[0] = norm[1]*tangent1[2] - norm[2]*tangent1[1];
tangent2[1] = norm[2]*tangent1[0] - norm[0]*tangent1[2];
tangent2[2] = norm[0]*tangent1[1] - norm[1]*tangent1[0];
while (1) {
r1 = sigma * random->gaussian();
r2 = sigma * random->gaussian();
vnmag = sqrt(r1*r1 + r2*r2);
vtmag1 = sigma * random->gaussian();
vtmag2 = sigma * random->gaussian();
if (vnmag*vnmag + vtmag1*vtmag1 + vtmag2*vtmag2 <= vmaxsq) break;
}
vsnew[0] = vnmag*norm[0] + vtmag1*tangent1[0] + vtmag2*tangent2[0];
vsnew[1] = vnmag*norm[1] + vtmag1*tangent1[1] + vtmag2*tangent2[1];
vsnew[2] = vnmag*norm[2] + vtmag1*tangent1[2] + vtmag2*tangent2[2];
// add in velocity of collision pt = velocity of wall
int dim = wallwhich[iwall] / 2;
vsnew[dim] += vwall[iwall];
}
/* ----------------------------------------------------------------------
impart force and torque to BIG particle
force on BIG particle = -dp/dt of SRD particle
torque on BIG particle = r cross (-dp/dt)
------------------------------------------------------------------------- */
void FixSRD::force_torque(double *vsold, double *vsnew,
double *xs, double *xb,
double *fb, double *tb)
{
double dpdt[3],xs_xb[3];
double factor = mass_srd / dt_big / force->ftm2v;
dpdt[0] = factor * (vsnew[0] - vsold[0]);
dpdt[1] = factor * (vsnew[1] - vsold[1]);
dpdt[2] = factor * (vsnew[2] - vsold[2]);
fb[0] -= dpdt[0];
fb[1] -= dpdt[1];
fb[2] -= dpdt[2];
// no torque if SLIP collision and BIG is a sphere
if (tb) {
xs_xb[0] = xs[0]-xb[0];
xs_xb[1] = xs[1]-xb[1];
xs_xb[2] = xs[2]-xb[2];
tb[0] -= xs_xb[1]*dpdt[2] - xs_xb[2]*dpdt[1];
tb[1] -= xs_xb[2]*dpdt[0] - xs_xb[0]*dpdt[2];
tb[2] -= xs_xb[0]*dpdt[1] - xs_xb[1]*dpdt[0];
}
}
/* ----------------------------------------------------------------------
impart force to WALL
force on WALL = -dp/dt of SRD particle
------------------------------------------------------------------------- */
void FixSRD::force_wall(double *vsold, double *vsnew, int iwall)
{
double dpdt[3],xs_xb[3];
double factor = mass_srd / dt_big / force->ftm2v;
dpdt[0] = factor * (vsnew[0] - vsold[0]);
dpdt[1] = factor * (vsnew[1] - vsold[1]);
dpdt[2] = factor * (vsnew[2] - vsold[2]);
fwall[iwall][0] -= dpdt[0];
fwall[iwall][1] -= dpdt[1];
fwall[iwall][2] -= dpdt[2];
}
/* ----------------------------------------------------------------------
update SRD particle position & velocity & search bin assignment
check if SRD moved outside of valid region
if so, may overlap off-processor BIG particle
------------------------------------------------------------------------- */
int FixSRD::update_srd(int i, double dt, double *xscoll, double *vsnew,
double *xs, double *vs)
{
int ix,iy,iz;
vs[0] = vsnew[0];
vs[1] = vsnew[1];
vs[2] = vsnew[2];
xs[0] = xscoll[0] + dt*vsnew[0];
xs[1] = xscoll[1] + dt*vsnew[1];
xs[2] = xscoll[2] + dt*vsnew[2];
if (triclinic) domain->x2lamda(xs,xs);
if (xs[0] < srdlo[0] || xs[0] > srdhi[0] ||
xs[1] < srdlo[1] || xs[1] > srdhi[1] ||
xs[2] < srdlo[2] || xs[2] > srdhi[2]) {
if (screen) {
error->warning("Fix srd particle moved outside valid domain");
fprintf(screen," particle %d on proc %d at timestep " BIGINT_FORMAT,
atom->tag[i],me,update->ntimestep);
fprintf(screen," xnew %g %g %g\n",xs[0],xs[1],xs[2]);
fprintf(screen," srdlo/hi x %g %g\n",srdlo[0],srdhi[0]);
fprintf(screen," srdlo/hi y %g %g\n",srdlo[1],srdhi[1]);
fprintf(screen," srdlo/hi z %g %g\n",srdlo[2],srdhi[2]);
}
}
if (triclinic) domain->lamda2x(xs,xs);
ix = static_cast<int> ((xs[0]-xblo2)*bininv2x);
iy = static_cast<int> ((xs[1]-yblo2)*bininv2y);
iz = static_cast<int> ((xs[2]-zblo2)*bininv2z);
return iz*nbin2y*nbin2x + iy*nbin2x + ix;
}
/* ----------------------------------------------------------------------
setup all SRD parameters with big particles
------------------------------------------------------------------------- */
void FixSRD::parameterize()
{
double PI = 4.0*atan(1.0);
// timesteps
dt_big = update->dt;
dt_srd = nevery * update->dt;
// maxbigdiam,minbigdiam = max/min diameter of any big particle
// big particle must either have radius > 0 or shape > 0 defined
// apply radfactor at end
double *radius = atom->radius;
double **shape = atom->shape;
int *mask = atom->mask;
int nlocal = atom->nlocal;
any_ellipsoids = 0;
maxbigdiam = 0.0;
minbigdiam = BIG;
for (int i = 0; i < nlocal; i++)
if (mask[i] & biggroupbit) {
if (radius && radius[i] > 0.0) {
maxbigdiam = MAX(maxbigdiam,2.0*radius[i]);
minbigdiam = MIN(minbigdiam,2.0*radius[i]);
} else if (shape && shape[i][0] > 0.0) {
maxbigdiam = MAX(maxbigdiam,2.0*shape[i][0]);
maxbigdiam = MAX(maxbigdiam,2.0*shape[i][1]);
maxbigdiam = MAX(maxbigdiam,2.0*shape[i][2]);
minbigdiam = MIN(minbigdiam,2.0*shape[i][0]);
minbigdiam = MIN(minbigdiam,2.0*shape[i][1]);
minbigdiam = MIN(minbigdiam,2.0*shape[i][2]);
if (shape[i][0] != shape[i][1] || shape[i][0] != shape[i][2])
any_ellipsoids = 1;
} else
error->one("Big particle in fix srd cannot be point particle");
}
double tmp = maxbigdiam;
MPI_Allreduce(&tmp,&maxbigdiam,1,MPI_DOUBLE,MPI_MAX,world);
tmp = minbigdiam;
MPI_Allreduce(&tmp,&minbigdiam,1,MPI_DOUBLE,MPI_MIN,world);
maxbigdiam *= radfactor;
minbigdiam *= radfactor;
int itmp = any_ellipsoids;
MPI_Allreduce(&itmp,&any_ellipsoids,1,MPI_INT,MPI_MAX,world);
// big particles are only torqued if are ellipsoids or NOSLIP collisions
if (any_ellipsoids == 0 && collidestyle == SLIP) torqueflag = 0;
else torqueflag = 1;
// mass of SRD particles, require monodispersity
double *rmass = atom->rmass;
double *mass = atom->mass;
int *type = atom->type;
int flag = 0;
mass_srd = 0.0;
for (int i = 0; i < nlocal; i++)
if (mask[i] & groupbit) {
if (rmass) {
if (mass_srd == 0.0) {
mass_srd = rmass[i];
printf("MASS SRD %g\n",rmass[i]);
} else if (rmass[i] != mass_srd) flag = 1;
} else {
if (mass_srd == 0.0) mass_srd = mass[type[i]];
else if (mass[type[i]] != mass_srd) flag = 1;
}
}
int flagall;
MPI_Allreduce(&flag,&flagall,1,MPI_INT,MPI_MAX,world);
if (flagall)
error->all("Fix srd requires SRD particles all have same mass");
// set temperature and lamda of SRD particles from each other
// lamda = dt_srd * sqrt(boltz * temperature_srd / mass_srd);
if (lamdaflag == 0)
lamda = dt_srd * sqrt(force->boltz*temperature_srd/mass_srd/force->mvv2e);
else
temperature_srd = force->mvv2e *
(lamda/dt_srd)*(lamda/dt_srd) * mass_srd/force->boltz;
printf("AAA %g %g\n",mass_srd,lamda);
// vmax = maximum velocity of an SRD particle
// dmax = maximum distance an SRD can move = 4*lamda = vmax * dt_srd
sigma = lamda/dt_srd;
dmax = 4.0*lamda;
vmax = dmax/dt_srd;
vmaxsq = vmax*vmax;
// volbig = total volume of all big particles
// apply radfactor to reduce final volume
double volbig = 0.0;
if (dimension == 3) {
for (int i = 0; i < nlocal; i++)
if (mask[i] & biggroupbit) {
if (radius && radius[i] > 0.0)
volbig += 4.0/3.0*PI*radius[i]*radius[i]*radius[i];
else if (shape && shape[i][0] > 0.0)
volbig += 4.0/3.0*PI * shape[i][0]*shape[i][1]*shape[i][2];
}
} else {
for (int i = 0; i < nlocal; i++)
if (mask[i] & biggroupbit) {
if (radius && radius[i] > 0.0)
volbig += PI*radius[i]*radius[i];
else if (shape && shape[i][0] > 0.0)
volbig += PI*shape[i][0]*shape[i][1];
}
}
tmp = volbig;
MPI_Allreduce(&tmp,&volbig,1,MPI_DOUBLE,MPI_SUM,world);
if (dimension == 3) volbig *= radfactor*radfactor*radfactor;
else volbig *= radfactor*radfactor;
// particle counts
bigint mbig = 0;
if (bigexist) mbig = group->count(biggroup);
bigint nsrd = group->count(igroup);
// mass_big = total mass of all big particles
mass_big = 0.0;
for (int i = 0; i < nlocal; i++)
if (mask[i] & biggroupbit) mass_big += rmass[i];
tmp = mass_big;
MPI_Allreduce(&tmp,&mass_big,1,MPI_DOUBLE,MPI_SUM,world);
// mass density ratio = big / SRD
double density_big = 0.0;
if (bigexist) density_big = mass_big / volbig;
double volsrd,density_srd;
if (dimension == 3) {
volsrd = (domain->xprd * domain->yprd * domain->zprd) - volbig;
density_srd = nsrd * mass_srd /
(domain->xprd*domain->yprd*domain->zprd - volbig);
} else {
volsrd = (domain->xprd * domain->yprd) - volbig;
density_srd = nsrd * mass_srd /
(domain->xprd*domain->yprd - volbig);
}
double mdratio = density_big/density_srd;
// create grid for binning/rotating SRD particles from gridsrd
setup_velocity_bins();
// binsize3 = binsize1 in box units (not lamda units for triclinic)
double binsize3x = binsize1x;
double binsize3y = binsize1y;
double binsize3z = binsize1z;
if (triclinic) {
binsize3x = binsize1x * domain->xprd;
binsize3y = binsize1y * domain->yprd;
binsize3z = binsize1z * domain->zprd;
}
// srd_per_grid = # of SRD particles per SRD grid cell
double ncell;
if (dimension == 3)
ncell = volsrd / (binsize3x*binsize3y*binsize3z);
else
ncell = volsrd / (binsize3x*binsize3y);
srd_per_cell = (double) nsrd / ncell;
// kinematic viscosity of SRD fluid
// output in cm^2/sec units, converted by xxt2kmu
double viscosity;
if (dimension == 3)
viscosity = gridsrd*gridsrd/(18.0*dt_srd) *
(1.0-(1.0-exp(-srd_per_cell))/srd_per_cell) +
(force->boltz*temperature_srd*dt_srd/(4.0*mass_srd*force->mvv2e)) *
((srd_per_cell+2.0)/(srd_per_cell-1.0));
else
viscosity =
(force->boltz*temperature_srd*dt_srd/(2.0*mass_srd*force->mvv2e)) *
(srd_per_cell/(srd_per_cell-1.0 + exp(-srd_per_cell)) - 1.0) +
(gridsrd*gridsrd)/(12.0*dt_srd) *
((srd_per_cell-1.0 + exp(-srd_per_cell))/srd_per_cell);
viscosity *= force->xxt2kmu;
// print SRD parameters
if (me == 0) {
if (screen) {
fprintf(screen,"SRD info:\n");
fprintf(screen,
" SRD/big particles = " BIGINT_FORMAT " " BIGINT_FORMAT "\n",
nsrd,mbig);
fprintf(screen," big particle diameter max/min = %g %g\n",
maxbigdiam,minbigdiam);
fprintf(screen," SRD temperature & lamda = %g %g\n",
temperature_srd,lamda);
fprintf(screen," SRD max distance & max velocity = %g %g\n",dmax,vmax);
fprintf(screen," SRD grid counts: %d %d %d\n",nbin1x,nbin1y,nbin1z);
fprintf(screen," SRD grid size: request, actual (xyz) = %g, %g %g %g\n",
gridsrd,binsize3x,binsize3y,binsize3z);
fprintf(screen," SRD per actual grid cell = %g\n",srd_per_cell);
fprintf(screen," SRD viscosity = %g\n",viscosity);
fprintf(screen," big/SRD mass density ratio = %g\n",mdratio);
}
if (logfile) {
fprintf(logfile,"SRD info:\n");
fprintf(logfile,
" SRD/big particles = " BIGINT_FORMAT " " BIGINT_FORMAT "\n",
nsrd,mbig);
fprintf(logfile," big particle diameter max/min = %g %g\n",
maxbigdiam,minbigdiam);
fprintf(logfile," SRD temperature & lamda = %g %g\n",
temperature_srd,lamda);
fprintf(logfile," SRD max distance & max velocity = %g %g\n",dmax,vmax);
fprintf(logfile," SRD grid counts: %d %d %d\n",nbin1x,nbin1y,nbin1z);
fprintf(logfile," SRD grid size: request, actual (xyz) = %g, %g %g %g\n",
gridsrd,binsize3x,binsize3y,binsize3z);
fprintf(logfile," SRD per actual grid cell = %g\n",srd_per_cell);
fprintf(logfile," SRD viscosity = %g\n",viscosity);
fprintf(logfile," big/SRD mass density ratio = %g\n",mdratio);
}
}
// error if less than 1 SRD bin per processor in some dim
if (nbin1x < comm->procgrid[0] || nbin1y < comm->procgrid[1] ||
nbin1z < comm->procgrid[2])
error->all("Fewer SRD bins than processors in some dimension");
// check if SRD bins are within tolerance for shape and size
int tolflag = 0;
if (binsize3y/binsize3x > 1.0+cubictol ||
binsize3x/binsize3y > 1.0+cubictol) tolflag = 1;
if (dimension == 3) {
if (binsize3z/binsize3x > 1.0+cubictol ||
binsize3x/binsize3z > 1.0+cubictol) tolflag = 1;
}
if (tolflag) {
if (cubicflag == CUBIC_ERROR)
error->all("SRD bins for fix srd are not cubic enough");
if (me == 0)
error->warning("SRD bins for fix srd are not cubic enough");
}
tolflag = 0;
if (binsize3x/gridsrd > 1.0+cubictol || gridsrd/binsize3x > 1.0+cubictol)
tolflag = 1;
if (binsize3y/gridsrd > 1.0+cubictol || gridsrd/binsize3y > 1.0+cubictol)
tolflag = 1;
if (dimension == 3) {
if (binsize3z/gridsrd > 1.0+cubictol || gridsrd/binsize3z > 1.0+cubictol)
tolflag = 1;
}
if (tolflag) {
if (cubicflag == CUBIC_ERROR)
error->all("SRD bin size for fix srd differs from user request");
if (me == 0)
error->warning("SRD bin size for fix srd differs from user request");
}
// error if lamda < 0.6 of SRD grid size and no shifting allowed
// turn on shifting in this case if allowed
double maxgridsrd = MAX(binsize3x,binsize3y);
if (dimension == 3) maxgridsrd = MAX(maxgridsrd,binsize3z);
shiftflag = 0;
if (lamda < 0.6*maxgridsrd && shiftuser == SHIFT_NO)
error->all("Fix srd lamda must be >= 0.6 of SRD grid size");
else if (lamda < 0.6*maxgridsrd && shiftuser == SHIFT_POSSIBLE) {
shiftflag = 1;
if (me == 0)
error->warning("SRD bin shifting turned on due to small lamda");
} else if (shiftuser == SHIFT_YES) shiftflag = 1;
// warnings
if (bigexist && maxgridsrd > 0.25 * minbigdiam && me == 0)
error->warning("Fix srd grid size > 1/4 of big particle diameter");
if (viscosity < 0.0 && me == 0)
error->warning("Fix srd viscosity < 0.0 due to low SRD density");
if (bigexist && dt_big*vmax > minbigdiam && me == 0)
error->warning("Fix srd particles may move > big particle diameter");
if (wallexist && collidestyle == NOSLIP && shiftflag == 1)
error->warning("Fix srd no-slip wall collisions with bin shifting");
}
/* ----------------------------------------------------------------------
set static parameters of each big particle, owned and ghost
called each reneighboring
use radfactor in distance parameters
------------------------------------------------------------------------- */
void FixSRD::big_static()
{
int i;
double rad,arad,brad,crad;
double *radius = atom->radius;
double **shape = atom->shape;
int *type = atom->type;
double skinhalf = 0.5 * neighbor->skin;
for (int k = 0; k < nbig; k++) {
i = biglist[k].index;
if (radius && radius[i] > 0.0) {
biglist[k].type = SPHERE;
rad = radfactor*radius[i];
biglist[k].radius = rad;
biglist[k].radsq = rad*rad;
biglist[k].cutbinsq = (rad+skinhalf) * (rad+skinhalf);
} else if (shape && shape[i][0] > 0.0) {
biglist[k].type = ELLIPSOID;
arad = radfactor*shape[i][0];
brad = radfactor*shape[i][1];
crad = radfactor*shape[i][2];
biglist[k].aradsqinv = 1.0/(arad*arad);
biglist[k].bradsqinv = 1.0/(brad*brad);
biglist[k].cradsqinv = 1.0/(crad*crad);
rad = MAX(arad,brad);
rad = MAX(rad,crad);
biglist[k].cutbinsq = (rad+skinhalf) * (rad+skinhalf);
}
}
}
/* ----------------------------------------------------------------------
set dynamics parameters of each big particle, owned and ghost
for ELLIPSOID, need current omega and ex,ey,ez
for SPHERE, need current omega from atom->omega or atom->angmom
called each timestep
------------------------------------------------------------------------- */
void FixSRD::big_dynamic()
{
int i;
double **omega = atom->omega;
double **angmom = atom->angmom;
double **quat = atom->quat;
double **shape = atom->shape;
double *rmass = atom->rmass;
for (int k = 0; k < nbig; k++) {
i = biglist[k].index;
// sphere with omega
// set omega from atom->omega directly
if (biglist[k].type == SPHERE) {
biglist[k].omega[0] = omega[i][0];
biglist[k].omega[1] = omega[i][1];
biglist[k].omega[2] = omega[i][2];
// ellipsoid with angmom
// calculate ex,ey,ez and omega from quaternion and angmom
} else if (biglist[k].type == ELLIPSOID) {
exyz_from_q(quat[i],biglist[k].ex,biglist[k].ey,biglist[k].ez);
omega_from_mq(angmom[i],biglist[k].ex,biglist[k].ey,biglist[k].ez,
rmass[i],shape[i],biglist[k].omega);
}
}
}
/* ----------------------------------------------------------------------
set bounds for big and SRD particle movement
called at setup() and when box size changes (but not shape)
------------------------------------------------------------------------- */
void FixSRD::setup_bounds()
{
// triclinic scale factors
// convert a real distance (perpendicular to box face) to a lamda distance
double length0,length1,length2;
if (triclinic) {
double *h_inv = domain->h_inv;
length0 = sqrt(h_inv[0]*h_inv[0] + h_inv[5]*h_inv[5] + h_inv[4]*h_inv[4]);
length1 = sqrt(h_inv[1]*h_inv[1] + h_inv[3]*h_inv[3]);
length2 = h_inv[2];
}
// collision object = CO = big particle or wall
// big particles can be owned or ghost or unknown, walls are all owned
// dist_ghost = distance from sub-domain (SD) that
// owned/ghost CO may move to before reneigh,
// used to bound search bins in setup_search_bins()
// dist_srd = distance from SD at which SRD could collide with unknown CO,
// used to error check bounds of SRD movement after collisions via srdlo/hi
// dist_srd_reneigh = distance from SD at which an SRD should trigger
// a reneigh, b/c next SRD move might overlap with unknown CO,
// used for SRD triggering of reneighboring via srdlo/hi_reneigh
// onemove = max distance an SRD can move in one step
// if big_particles (and possibly walls):
// dist_ghost = cut + 1/2 skin due to moving away before reneigh
// dist_srd = cut - 1/2 skin - 1/2 diam due to ghost CO moving towards
// dist_reneigh = dist_srd - onemove
// if walls and no big particles:
// dist_ghost = 0.0, since not used
// if no big particles or walls:
// dist_ghost and dist_srd = 0.0, since not used since no search bins
// dist_srd_reneigh = subsize - onemove =
// max distance to move without being lost during comm->exchange()
// subsize = perp distance between sub-domain faces (orthog or triclinic)
double cut = MAX(neighbor->cutneighmax,comm->cutghostuser);
double onemove = dt_big*vmax;
if (bigexist) {
dist_ghost = cut + 0.5*neighbor->skin;
dist_srd = cut - 0.5*neighbor->skin - 0.5*maxbigdiam;
dist_srd_reneigh = dist_srd - onemove;
} else if (wallexist) {
dist_ghost = 4*onemove;
dist_srd = 4*onemove;
dist_srd_reneigh = 4*onemove - onemove;
} else {
dist_ghost = dist_srd = 0.0;
double subsize;
if (triclinic == 0) {
subsize = domain->prd[0]/comm->procgrid[0];
subsize = MIN(subsize,domain->prd[1]/comm->procgrid[1]);
if (dimension == 3)
subsize = MIN(subsize,domain->prd[2]/comm->procgrid[2]);
} else {
subsize = 1.0/comm->procgrid[0]/length0;
subsize = MIN(subsize,1.0/comm->procgrid[1]/length1);
if (dimension == 3)
subsize = MIN(subsize,1.0/comm->procgrid[2]/length2);
}
dist_srd_reneigh = subsize - onemove;
}
// lo/hi = bbox on this proc which SRD particles must stay inside
// lo/hi reneigh = bbox on this proc outside of which SRDs trigger a reneigh
// for triclinic, these bbox are in lamda units
if (triclinic == 0) {
srdlo[0] = domain->sublo[0] - dist_srd;
srdhi[0] = domain->subhi[0] + dist_srd;
srdlo[1] = domain->sublo[1] - dist_srd;
srdhi[1] = domain->subhi[1] + dist_srd;
srdlo[2] = domain->sublo[2] - dist_srd;
srdhi[2] = domain->subhi[2] + dist_srd;
srdlo_reneigh[0] = domain->sublo[0] - dist_srd_reneigh;
srdhi_reneigh[0] = domain->subhi[0] + dist_srd_reneigh;
srdlo_reneigh[1] = domain->sublo[1] - dist_srd_reneigh;
srdhi_reneigh[1] = domain->subhi[1] + dist_srd_reneigh;
srdlo_reneigh[2] = domain->sublo[2] - dist_srd_reneigh;
srdhi_reneigh[2] = domain->subhi[2] + dist_srd_reneigh;
} else {
srdlo[0] = domain->sublo_lamda[0] - dist_srd*length0;
srdhi[0] = domain->subhi_lamda[0] + dist_srd*length0;
srdlo[1] = domain->sublo_lamda[1] - dist_srd*length1;
srdhi[1] = domain->subhi_lamda[1] + dist_srd*length1;
srdlo[2] = domain->sublo_lamda[2] - dist_srd*length2;
srdhi[2] = domain->subhi_lamda[2] + dist_srd*length2;
srdlo_reneigh[0] = domain->sublo_lamda[0] - dist_srd_reneigh*length0;
srdhi_reneigh[0] = domain->subhi_lamda[0] + dist_srd_reneigh*length0;
srdlo_reneigh[1] = domain->sublo_lamda[1] - dist_srd_reneigh*length1;
srdhi_reneigh[1] = domain->subhi_lamda[1] + dist_srd_reneigh*length1;
srdlo_reneigh[2] = domain->sublo_lamda[2] - dist_srd_reneigh*length2;
srdhi_reneigh[2] = domain->subhi_lamda[2] + dist_srd_reneigh*length2;
}
}
/* ----------------------------------------------------------------------
setup bins used for binning SRD particles for velocity reset
gridsrd = desired bin size
also setup bin shifting parameters
also setup comm of bins that straddle processor boundaries
called at beginning of each run
called every reneighbor if box size changes, but not if box shape changes
------------------------------------------------------------------------- */
void FixSRD::setup_velocity_bins()
{
// require integer # of bins across global domain
nbin1x = static_cast<int> (domain->xprd/gridsrd + 0.5);
nbin1y = static_cast<int> (domain->yprd/gridsrd + 0.5);
nbin1z = static_cast<int> (domain->zprd/gridsrd + 0.5);
if (dimension == 2) nbin1z = 1;
if (nbin1x == 0) nbin1x = 1;
if (nbin1y == 0) nbin1y = 1;
if (nbin1z == 0) nbin1z = 1;
if (triclinic == 0) {
binsize1x = domain->xprd / nbin1x;
binsize1y = domain->yprd / nbin1y;
binsize1z = domain->zprd / nbin1z;
bininv1x = 1.0/binsize1x;
bininv1y = 1.0/binsize1y;
bininv1z = 1.0/binsize1z;
} else {
binsize1x = 1.0 / nbin1x;
binsize1y = 1.0 / nbin1y;
binsize1z = 1.0 / nbin1z;
bininv1x = nbin1x;
bininv1y = nbin1y;
bininv1z = nbin1z;
}
nbins1 = nbin1x*nbin1y*nbin1z;
// setup two shifts, 0 = no shift, 1 = shift
// initialize no shift case since static
// shift case is dynamic, has to be initialized each time shift occurs
// setup_velocity_shift allocates memory for vbin and sendlist/recvlist
double *boxlo;
if (triclinic == 0) boxlo = domain->boxlo;
else boxlo = domain->boxlo_lamda;
shifts[0].corner[0] = boxlo[0];
shifts[0].corner[1] = boxlo[1];
shifts[0].corner[2] = boxlo[2];
setup_velocity_shift(0,0);
shifts[1].corner[0] = boxlo[0];
shifts[1].corner[1] = boxlo[1];
shifts[1].corner[2] = boxlo[2];
setup_velocity_shift(1,0);
// allocate binhead based on max # of bins in either shift
int max = shifts[0].nbins;
max = MAX(max,shifts[1].nbins);
if (max > maxbin1) {
memory->destroy(binhead);
maxbin1 = max;
memory->create(binhead,max,"fix/srd:binhead");
}
// allocate sbuf,rbuf based on biggest bin message
max = 0;
for (int ishift = 0; ishift < 2; ishift++)
for (int iswap = 0; iswap < 2*dimension; iswap++) {
max = MAX(max,shifts[ishift].bcomm[iswap].nsend);
max = MAX(max,shifts[ishift].bcomm[iswap].nrecv);
}
if (max > maxbuf) {
memory->destroy(sbuf1);
memory->destroy(sbuf2);
memory->destroy(rbuf1);
memory->destroy(rbuf2);
maxbuf = max;
memory->create(sbuf1,max*VBINSIZE,"fix/srd:sbuf");
memory->create(sbuf2,max*VBINSIZE,"fix/srd:sbuf");
memory->create(rbuf1,max*VBINSIZE,"fix/srd:rbuf");
memory->create(rbuf2,max*VBINSIZE,"fix/srd:rbuf");
}
// commflag = 1 if any comm required due to bins overlapping proc boundaries
shifts[0].commflag = 0;
if (nbin1x % comm->procgrid[0]) shifts[0].commflag = 1;
if (nbin1y % comm->procgrid[1]) shifts[0].commflag = 1;
if (nbin1z % comm->procgrid[2]) shifts[0].commflag = 1;
shifts[1].commflag = 1;
}
/* ----------------------------------------------------------------------
setup velocity shift parameters
set binlo[]/binhi[] and nbins,nbinx,nbiny,nbinz for this proc
set bcomm[6] params based on bin overlaps with proc boundaries
no comm of bins across non-periodic global boundaries
set vbin owner flags for bins I am owner of
ishift = 0, dynamic = 0:
set all settings since are static
allocate and set bcomm params and vbins
do not comm bins that align with proc boundaries
ishift = 1, dynamic = 0:
set max bounds on bin counts and message sizes
allocate and set bcomm params and vbins based on max bounds
other settings will later change dynamically
ishift = 1, dynamic = 1:
set actual bin bounds and counts for specific shift
set bcomm params and vbins (already allocated)
called by setup_velocity_bins() and reset_velocities()
------------------------------------------------------------------------- */
void FixSRD::setup_velocity_shift(int ishift, int dynamic)
{
int i,j,k,m,id,nsend;
int *sendlist;
BinComm *first,*second;
BinAve *vbin;
double *sublo,*subhi;
if (triclinic == 0) {
sublo = domain->sublo;
subhi = domain->subhi;
} else {
sublo = domain->sublo_lamda;
subhi = domain->subhi_lamda;
}
int *binlo = shifts[ishift].binlo;
int *binhi = shifts[ishift].binhi;
double *corner = shifts[ishift].corner;
int *procgrid = comm->procgrid;
int *myloc = comm->myloc;
binlo[0] = static_cast<int> ((sublo[0]-corner[0])*bininv1x);
binlo[1] = static_cast<int> ((sublo[1]-corner[1])*bininv1y);
binlo[2] = static_cast<int> ((sublo[2]-corner[2])*bininv1z);
if (dimension == 2) shifts[ishift].binlo[2] = 0;
binhi[0] = static_cast<int> ((subhi[0]-corner[0])*bininv1x);
binhi[1] = static_cast<int> ((subhi[1]-corner[1])*bininv1y);
binhi[2] = static_cast<int> ((subhi[2]-corner[2])*bininv1z);
if (dimension == 2) shifts[ishift].binhi[2] = 0;
if (ishift == 0) {
if (myloc[0]*nbin1x % procgrid[0] == 0)
binlo[0] = myloc[0]*nbin1x/procgrid[0];
if (myloc[1]*nbin1y % procgrid[1] == 0)
binlo[1] = myloc[1]*nbin1y/procgrid[1];
if (myloc[2]*nbin1z % procgrid[2] == 0)
binlo[2] = myloc[2]*nbin1z/procgrid[2];
if ((myloc[0]+1)*nbin1x % procgrid[0] == 0)
binhi[0] = (myloc[0]+1)*nbin1x/procgrid[0] - 1;
if ((myloc[1]+1)*nbin1y % procgrid[1] == 0)
binhi[1] = (myloc[1]+1)*nbin1y/procgrid[1] - 1;
if ((myloc[2]+1)*nbin1z % procgrid[2] == 0)
binhi[2] = (myloc[2]+1)*nbin1z/procgrid[2] - 1;
}
int nbinx = binhi[0] - binlo[0] + 1;
int nbiny = binhi[1] - binlo[1] + 1;
int nbinz = binhi[2] - binlo[2] + 1;
// allow for one extra bin if shifting will occur
if (ishift == 1 && dynamic == 0) {
nbinx++;
nbiny++;
if (dimension == 3) nbinz++;
}
int nbins = nbinx*nbiny*nbinz;
int nbinxy = nbinx*nbiny;
int nbinsq = nbinx*nbiny;
nbinsq = MAX(nbiny*nbinz,nbinsq);
nbinsq = MAX(nbinx*nbinz,nbinsq);
shifts[ishift].nbins = nbins;
shifts[ishift].nbinx = nbinx;
shifts[ishift].nbiny = nbiny;
shifts[ishift].nbinz = nbinz;
int reallocflag = 0;
if (dynamic == 0 && nbinsq > shifts[ishift].maxbinsq) {
shifts[ishift].maxbinsq = nbinsq;
reallocflag = 1;
}
// bcomm neighbors
// first = send in lo direction, recv from hi direction
// second = send in hi direction, recv from lo direction
if (dynamic == 0) {
shifts[ishift].bcomm[0].sendproc = comm->procneigh[0][0];
shifts[ishift].bcomm[0].recvproc = comm->procneigh[0][1];
shifts[ishift].bcomm[1].sendproc = comm->procneigh[0][1];
shifts[ishift].bcomm[1].recvproc = comm->procneigh[0][0];
shifts[ishift].bcomm[2].sendproc = comm->procneigh[1][0];
shifts[ishift].bcomm[2].recvproc = comm->procneigh[1][1];
shifts[ishift].bcomm[3].sendproc = comm->procneigh[1][1];
shifts[ishift].bcomm[3].recvproc = comm->procneigh[1][0];
shifts[ishift].bcomm[4].sendproc = comm->procneigh[2][0];
shifts[ishift].bcomm[4].recvproc = comm->procneigh[2][1];
shifts[ishift].bcomm[5].sendproc = comm->procneigh[2][1];
shifts[ishift].bcomm[5].recvproc = comm->procneigh[2][0];
}
// set nsend,nrecv and sendlist,recvlist for each swap in x,y,z
// set nsend,nrecv = 0 if static bins align with proc boundary
// or to prevent dynamic bin swapping across non-periodic global boundary
// allocate sendlist,recvlist only for dynamic = 0
first = &shifts[ishift].bcomm[0];
second = &shifts[ishift].bcomm[1];
first->nsend = first->nrecv = second->nsend = second->nrecv = nbiny*nbinz;
if (ishift == 0) {
if (myloc[0]*nbin1x % procgrid[0] == 0)
first->nsend = second->nrecv = 0;
if ((myloc[0]+1)*nbin1x % procgrid[0] == 0)
second->nsend = first->nrecv = 0;
} else if (dynamic == 0 && domain->xperiodic == 0) {
if (myloc[0] == 0) first->nsend = second->nrecv = 0;
if (myloc[0] == procgrid[0]-1) second->nsend = first->nrecv = 0;
}
if (reallocflag) {
memory->destroy(first->sendlist);
memory->destroy(first->recvlist);
memory->destroy(second->sendlist);
memory->destroy(second->recvlist);
memory->create(first->sendlist,nbinsq,"fix/srd:sendlist");
memory->create(first->recvlist,nbinsq,"fix/srd:sendlist");
memory->create(second->sendlist,nbinsq,"fix/srd:sendlist");
memory->create(second->recvlist,nbinsq,"fix/srd:sendlist");
}
m = 0;
i = 0;
for (j = 0; j < nbiny; j++)
for (k = 0; k < nbinz; k++) {
id = k*nbinxy + j*nbinx + i;
first->sendlist[m] = second->recvlist[m] = id;
m++;
}
m = 0;
i = nbinx-1;
for (j = 0; j < nbiny; j++)
for (k = 0; k < nbinz; k++) {
id = k*nbinxy + j*nbinx + i;
second->sendlist[m] = first->recvlist[m] = id;
m++;
}
first = &shifts[ishift].bcomm[2];
second = &shifts[ishift].bcomm[3];
first->nsend = first->nrecv = second->nsend = second->nrecv = nbinx*nbinz;
if (ishift == 0) {
if (myloc[1]*nbin1y % procgrid[1] == 0)
first->nsend = second->nrecv = 0;
if ((myloc[1]+1)*nbin1y % procgrid[1] == 0)
second->nsend = first->nrecv = 0;
} else if (dynamic == 0 && domain->yperiodic == 0) {
if (myloc[1] == 0) first->nsend = second->nrecv = 0;
if (myloc[1] == procgrid[1]-1) second->nsend = first->nrecv = 0;
}
if (reallocflag) {
memory->destroy(first->sendlist);
memory->destroy(first->recvlist);
memory->destroy(second->sendlist);
memory->destroy(second->recvlist);
memory->create(first->sendlist,nbinsq,"fix/srd:sendlist");
memory->create(first->recvlist,nbinsq,"fix/srd:sendlist");
memory->create(second->sendlist,nbinsq,"fix/srd:sendlist");
memory->create(second->recvlist,nbinsq,"fix/srd:sendlist");
}
m = 0;
j = 0;
for (i = 0; i < nbinx; i++)
for (k = 0; k < nbinz; k++) {
id = k*nbinxy + j*nbinx + i;
first->sendlist[m] = second->recvlist[m] = id;
m++;
}
m = 0;
j = nbiny-1;
for (i = 0; i < nbinx; i++)
for (k = 0; k < nbinz; k++) {
id = k*nbinxy + j*nbinx + i;
second->sendlist[m] = first->recvlist[m] = id;
m++;
}
if (dimension == 3) {
first = &shifts[ishift].bcomm[4];
second = &shifts[ishift].bcomm[5];
first->nsend = first->nrecv = second->nsend = second->nrecv = nbinx*nbiny;
if (ishift == 0) {
if (myloc[2]*nbin1z % procgrid[2] == 0)
first->nsend = second->nrecv = 0;
if ((myloc[2]+1)*nbin1z % procgrid[2] == 0)
second->nsend = first->nrecv = 0;
} else if (dynamic == 0 && domain->zperiodic == 0) {
if (myloc[2] == 0) first->nsend = second->nrecv = 0;
if (myloc[2] == procgrid[2]-1) second->nsend = first->nrecv = 0;
}
if (reallocflag) {
memory->destroy(first->sendlist);
memory->destroy(first->recvlist);
memory->destroy(second->sendlist);
memory->destroy(second->recvlist);
memory->create(first->sendlist,nbinx*nbiny,"fix/srd:sendlist");
memory->create(first->recvlist,nbinx*nbiny,"fix/srd:sendlist");
memory->create(second->sendlist,nbinx*nbiny,"fix/srd:sendlist");
memory->create(second->recvlist,nbinx*nbiny,"fix/srd:sendlist");
}
m = 0;
k = 0;
for (i = 0; i < nbinx; i++)
for (j = 0; j < nbiny; j++) {
id = k*nbinxy + j*nbinx + i;
first->sendlist[m] = second->recvlist[m] = id;
m++;
}
m = 0;
k = nbinz-1;
for (i = 0; i < nbinx; i++)
for (j = 0; j < nbiny; j++) {
id = k*nbinxy + j*nbinx + i;
second->sendlist[m] = first->recvlist[m] = id;
m++;
}
}
// allocate vbins, only for dynamic = 0
if (dynamic == 0 && nbins > shifts[ishift].maxvbin) {
memory->destroy(shifts[ishift].vbin);
shifts[ishift].maxvbin = nbins;
shifts[ishift].vbin = (BinAve *)
memory->smalloc(nbins*sizeof(BinAve),"fix/srd:vbin");
}
// for vbins I own, set owner = 1
// if bin never sent to anyone, I own it
// if bin sent to lower numbered proc, I do not own it
// if bin sent to self, I do not own it on even swap (avoids double counting)
vbin = shifts[ishift].vbin;
for (i = 0; i < nbins; i++) vbin[i].owner = 1;
for (int iswap = 0; iswap < 2*dimension; iswap++) {
if (shifts[ishift].bcomm[iswap].sendproc > me) continue;
if (shifts[ishift].bcomm[iswap].sendproc == me && iswap % 2 == 0) continue;
nsend = shifts[ishift].bcomm[iswap].nsend;
sendlist = shifts[ishift].bcomm[iswap].sendlist;
for (m = 0; m < nsend; m++) vbin[sendlist[m]].owner = 0;
}
// if triclinic and streamflag:
// set xctr (in lamda units) for all nbins so can compute bin vstream
if (triclinic && streamflag) {
m = 0;
for (k = 0; k < nbinz; k++)
for (j = 0; j < nbiny; j++)
for (i = 0; i < nbinx; i++) {
vbin[m].xctr[0] = corner[0] + (i+binlo[0]+0.5)/nbin1x;
vbin[m].xctr[1] = corner[1] + (j+binlo[1]+0.5)/nbin1y;
vbin[m].xctr[2] = corner[2] + (k+binlo[2]+0.5)/nbin1z;
m++;
}
}
}
/* ----------------------------------------------------------------------
setup bins used for big and SRD particle searching
gridsearch = desired bin size
bins are orthogonal whether simulation box is orthogonal or triclinic
for orthogonal boxes, called at each setup since cutghost may change
for triclinic boxes, called at every reneigh, since tilt may change
sets the following:
nbin2 xyz = # of bins in each dim
binsize2 and bininv2 xyz = size of bins in each dim
xyz blo2 = origin of bins
------------------------------------------------------------------------- */
void FixSRD::setup_search_bins()
{
// subboxlo/hi = real space bbox which
// owned/ghost big particles or walls can be in
// start with bounding box for my sub-domain, add dist_ghost
// for triclinic, need to:
// a) convert dist_ghost to lamda space via length012
// b) lo/hi = sub-domain big particle bbox in lamda space
// c) convert lo/hi to real space bounding box via domain->bbox()
// similar to neighbor::setup_bins() and comm::cutghost[] calculation
double subboxlo[3],subboxhi[3];
if (triclinic == 0) {
subboxlo[0] = domain->sublo[0] - dist_ghost;
subboxlo[1] = domain->sublo[1] - dist_ghost;
subboxlo[2] = domain->sublo[2] - dist_ghost;
subboxhi[0] = domain->subhi[0] + dist_ghost;
subboxhi[1] = domain->subhi[1] + dist_ghost;
subboxhi[2] = domain->subhi[2] + dist_ghost;
} else {
double *h_inv = domain->h_inv;
double length0,length1,length2;
length0 = sqrt(h_inv[0]*h_inv[0] + h_inv[5]*h_inv[5] + h_inv[4]*h_inv[4]);
length1 = sqrt(h_inv[1]*h_inv[1] + h_inv[3]*h_inv[3]);
length2 = h_inv[2];
double lo[3],hi[3];
lo[0] = domain->sublo_lamda[0] - dist_ghost*length0;
lo[1] = domain->sublo_lamda[1] - dist_ghost*length1;
lo[2] = domain->sublo_lamda[2] - dist_ghost*length2;
hi[0] = domain->subhi_lamda[0] + dist_ghost*length0;
hi[1] = domain->subhi_lamda[1] + dist_ghost*length1;
hi[2] = domain->subhi_lamda[2] + dist_ghost*length2;
domain->bbox(lo,hi,subboxlo,subboxhi);
}
// require integer # of bins for that volume
nbin2x = static_cast<int> ((subboxhi[0] - subboxlo[0]) / gridsearch);
nbin2y = static_cast<int> ((subboxhi[1] - subboxlo[1]) / gridsearch);
nbin2z = static_cast<int> ((subboxhi[2] - subboxlo[2]) / gridsearch);
if (dimension == 2) nbin2z = 1;
if (nbin2x == 0) nbin2x = 1;
if (nbin2y == 0) nbin2y = 1;
if (nbin2z == 0) nbin2z = 1;
binsize2x = (subboxhi[0] - subboxlo[0]) / nbin2x;
binsize2y = (subboxhi[1] - subboxlo[1]) / nbin2y;
binsize2z = (subboxhi[2] - subboxlo[2]) / nbin2z;
bininv2x = 1.0/binsize2x;
bininv2y = 1.0/binsize2y;
bininv2z = 1.0/binsize2z;
// add bins on either end due to extent of big particles
// radmax = max distance from central bin that biggest particle overlaps
// includes skin movement
// nx,ny,nz = max # of bins to search away from central bin
double radmax = 0.5*maxbigdiam + 0.5*neighbor->skin;
int nx = static_cast<int> (radmax/binsize2x) + 1;
int ny = static_cast<int> (radmax/binsize2y) + 1;
int nz = static_cast<int> (radmax/binsize2z) + 1;
if (dimension == 2) nz = 0;
nbin2x += 2*nx;
nbin2y += 2*ny;
nbin2z += 2*nz;
xblo2 = subboxlo[0] - nx*binsize2x;
yblo2 = subboxlo[1] - ny*binsize2y;
zblo2 = subboxlo[2] - nz*binsize2z;
if (dimension == 2) zblo2 = domain->boxlo[2];
// allocate bins
// first deallocate previous bins if necessary
nbins2 = nbin2x*nbin2y*nbin2z;
if (nbins2 > maxbin2) {
memory->destroy(nbinbig);
memory->destroy(binbig);
maxbin2 = nbins2;
memory->create(nbinbig,nbins2,"fix/srd:nbinbig");
memory->create(binbig,nbins2,ATOMPERBIN,"fix/srd:binbig");
}
}
/* ----------------------------------------------------------------------
compute stencil of bin offsets for a big particle overlapping search bins
------------------------------------------------------------------------- */
void FixSRD::setup_search_stencil()
{
// radmax = max distance from central bin that any big particle overlaps
// includes skin movement
// nx,ny,nz = max # of bins to search away from central bin
double radmax = 0.5*maxbigdiam + 0.5*neighbor->skin;
double radsq = radmax*radmax;
int nx = static_cast<int> (radmax/binsize2x) + 1;
int ny = static_cast<int> (radmax/binsize2y) + 1;
int nz = static_cast<int> (radmax/binsize2z) + 1;
if (dimension == 2) nz = 0;
// allocate stencil array
// deallocate previous stencil if necessary
int max = (2*nx+1) * (2*ny+1) * (2*nz+1);
if (max > maxstencil) {
memory->destroy(stencil);
maxstencil = max;
memory->create(stencil,max,4,"fix/srd:stencil");
}
// loop over all bins
// add bin to stencil:
// if distance of closest corners of bin and central bin is within cutoff
nstencil = 0;
for (int k = -nz; k <= nz; k++)
for (int j = -ny; j <= ny; j++)
for (int i = -nx; i <= nx; i++)
if (bin_bin_distance(i,j,k) < radsq) {
stencil[nstencil][0] = i;
stencil[nstencil][1] = j;
stencil[nstencil][2] = k;
stencil[nstencil][3] = k*nbin2y*nbin2x + j*nbin2x + i;
nstencil++;
}
}
/* ----------------------------------------------------------------------
compute closest squared distance between point x and bin ibin
------------------------------------------------------------------------- */
double FixSRD::point_bin_distance(double *x, int i, int j, int k)
{
double delx,dely,delz;
double xlo = xblo2 + i*binsize2x;
double xhi = xlo + binsize2x;
double ylo = yblo2 + j*binsize2y;
double yhi = ylo + binsize2y;
double zlo = zblo2 + k*binsize2z;
double zhi = zlo + binsize2z;
if (x[0] < xlo) delx = xlo - x[0];
else if (x[0] > xhi) delx = x[0] - xhi;
else delx = 0.0;
if (x[1] < ylo) dely = ylo - x[1];
else if (x[1] > yhi) dely = x[1] - yhi;
else dely = 0.0;
if (x[2] < zlo) delz = zlo - x[2];
else if (x[2] > zhi) delz = x[2] - zhi;
else delz = 0.0;
return (delx*delx + dely*dely + delz*delz);
}
/* ----------------------------------------------------------------------
compute closest squared distance between 2 bins
central bin (0,0,0) and bin (i,j,k)
------------------------------------------------------------------------- */
double FixSRD::bin_bin_distance(int i, int j, int k)
{
double delx,dely,delz;
if (i > 0) delx = (i-1)*binsize2x;
else if (i == 0) delx = 0.0;
else delx = (i+1)*binsize2x;
if (j > 0) dely = (j-1)*binsize2y;
else if (j == 0) dely = 0.0;
else dely = (j+1)*binsize2y;
if (k > 0) delz = (k-1)*binsize2z;
else if (k == 0) delz = 0.0;
else delz = (k+1)*binsize2z;
return (delx*delx + dely*dely + delz*delz);
}
/* ----------------------------------------------------------------------
compute space-frame ex,ey,ez from current quaternion q
ex,ey,ez = space-frame coords of 1st,2nd,3rd principal axis
operation is ex = q' d q = Q d, where d is (1,0,0) = 1st axis in body frame
------------------------------------------------------------------------- */
void FixSRD::exyz_from_q(double *q, double *ex, double *ey, double *ez)
{
ex[0] = q[0]*q[0] + q[1]*q[1] - q[2]*q[2] - q[3]*q[3];
ex[1] = 2.0 * (q[1]*q[2] + q[0]*q[3]);
ex[2] = 2.0 * (q[1]*q[3] - q[0]*q[2]);
ey[0] = 2.0 * (q[1]*q[2] - q[0]*q[3]);
ey[1] = q[0]*q[0] - q[1]*q[1] + q[2]*q[2] - q[3]*q[3];
ey[2] = 2.0 * (q[2]*q[3] + q[0]*q[1]);
ez[0] = 2.0 * (q[1]*q[3] + q[0]*q[2]);
ez[1] = 2.0 * (q[2]*q[3] - q[0]*q[1]);
ez[2] = q[0]*q[0] - q[1]*q[1] - q[2]*q[2] + q[3]*q[3];
}
/* ----------------------------------------------------------------------
compute omega from angular momentum
w = omega = angular velocity in space frame
wbody = angular velocity in body frame
set wbody component to 0.0 if inertia component is 0.0
otherwise body can spin easily around that axis
project space-frame angular momentum onto body axes
and divide by principal moments
------------------------------------------------------------------------- */
void FixSRD::omega_from_mq(double *m, double *ex, double *ey, double *ez,
double mass, double *shape, double *w)
{
double inertia[3],wbody[3];
inertia[0] = 0.2*mass * (shape[1]*shape[1]+shape[2]*shape[2]);
inertia[1] = 0.2*mass * (shape[0]*shape[0]+shape[2]*shape[2]);
inertia[2] = 0.2*mass * (shape[0]*shape[0]+shape[1]*shape[1]);
wbody[0] = (m[0]*ex[0] + m[1]*ex[1] + m[2]*ex[2]) / inertia[0];
wbody[1] = (m[0]*ey[0] + m[1]*ey[1] + m[2]*ey[2]) / inertia[1];
wbody[2] = (m[0]*ez[0] + m[1]*ez[1] + m[2]*ez[2]) / inertia[2];
w[0] = wbody[0]*ex[0] + wbody[1]*ey[0] + wbody[2]*ez[0];
w[1] = wbody[0]*ex[1] + wbody[1]*ey[1] + wbody[2]*ez[1];
w[2] = wbody[0]*ex[2] + wbody[1]*ey[2] + wbody[2]*ez[2];
}
/* ---------------------------------------------------------------------- */
int FixSRD::pack_reverse_comm(int n, int first, double *buf)
{
int i,m,last;
m = 0;
last = first + n;
if (torqueflag) {
for (i = first; i < last; i++) {
buf[m++] = flocal[i][0];
buf[m++] = flocal[i][1];
buf[m++] = flocal[i][2];
buf[m++] = tlocal[i][0];
buf[m++] = tlocal[i][1];
buf[m++] = tlocal[i][2];
}
} else {
for (i = first; i < last; i++) {
buf[m++] = flocal[i][0];
buf[m++] = flocal[i][1];
buf[m++] = flocal[i][2];
}
}
return comm_reverse;
}
/* ---------------------------------------------------------------------- */
void FixSRD::unpack_reverse_comm(int n, int *list, double *buf)
{
int i,j,m;
m = 0;
if (torqueflag) {
for (i = 0; i < n; i++) {
j = list[i];
flocal[j][0] += buf[m++];
flocal[j][1] += buf[m++];
flocal[j][2] += buf[m++];
tlocal[j][0] += buf[m++];
tlocal[j][1] += buf[m++];
tlocal[j][2] += buf[m++];
}
} else {
for (i = 0; i < n; i++) {
j = list[i];
flocal[j][0] += buf[m++];
flocal[j][1] += buf[m++];
flocal[j][2] += buf[m++];
}
}
}
/* ----------------------------------------------------------------------
SRD stats
------------------------------------------------------------------------- */
double FixSRD::compute_vector(int n)
{
// only sum across procs one time
if (stats_flag == 0) {
stats[0] = ncheck;
stats[1] = ncollide;
stats[2] = nbounce;
stats[3] = ninside;
stats[4] = nrescale;
stats[5] = nbins2;
stats[6] = nbins1;
stats[7] = srd_bin_count;
stats[8] = srd_bin_temp;
stats[9] = bouncemaxnum;
stats[10] = bouncemax;
stats[11] = reneighcount;
MPI_Allreduce(stats,stats_all,10,MPI_DOUBLE,MPI_SUM,world);
MPI_Allreduce(&stats[10],&stats_all[10],1,MPI_DOUBLE,MPI_MAX,world);
if (stats_all[7] != 0.0) stats_all[8] /= stats_all[7];
stats_all[6] /= nprocs;
stats_flag = 1;
}
return stats_all[n];
}
/* ---------------------------------------------------------------------- */
void FixSRD::velocity_stats(int groupnum)
{
int bitmask = group->bitmask[groupnum];
double **v = atom->v;
int *mask = atom->mask;
int nlocal = atom->nlocal;
double vone;
double vave = 0.0;
double vmax = 0.0;
for (int i = 0; i < nlocal; i++)
if (mask[i] & bitmask) {
vone = sqrt(v[i][0]*v[i][0] + v[i][1]*v[i][1] + v[i][2]*v[i][2]);
vave += vone;
if (vone > vmax) vmax = vone;
}
double all;
MPI_Allreduce(&vave,&all,1,MPI_DOUBLE,MPI_SUM,world);
double count = group->count(groupnum);
if (count != 0.0) vave = all/count;
else vave = 0.0;
MPI_Allreduce(&vmax,&all,1,MPI_DOUBLE,MPI_MAX,world);
vmax = all;
if (me == 0) {
if (screen)
fprintf(screen," ave/max %s velocity = %g %g\n",
group->names[groupnum],vave,vmax);
if (logfile)
fprintf(logfile," ave/max %s velocity = %g %g\n",
group->names[groupnum],vave,vmax);
}
}
/* ---------------------------------------------------------------------- */
double FixSRD::memory_usage()
{
double bytes = 0.0;
bytes += (shifts[0].nbins + shifts[1].nbins) * sizeof(BinAve);
bytes += nmax * sizeof(int);
if (bigexist) {
bytes += nbins2 * sizeof(int);
bytes += nbins2*ATOMPERBIN * sizeof(int);
}
bytes += nmax * sizeof(int);
return bytes;
}
/* ----------------------------------------------------------------------
useful debugging functions
------------------------------------------------------------------------- */
double FixSRD::distance(int i, int j)
{
double dx = atom->x[i][0] - atom->x[j][0];
double dy = atom->x[i][1] - atom->x[j][1];
double dz = atom->x[i][2] - atom->x[j][2];
return sqrt(dx*dx + dy*dy + dz*dz);
}
/* ---------------------------------------------------------------------- */
void FixSRD::print_collision(int i, int j, int ibounce,
double t_remain, double dt,
double *xscoll, double *xbcoll, double *norm,
int type)
{
double xsstart[3],xbstart[3];
double **x = atom->x;
double **v = atom->v;
if (type != WALL) {
printf("COLLISION between SRD %d and BIG %d\n",atom->tag[i],atom->tag[j]);
printf(" bounce # = %d\n",ibounce+1);
printf(" local indices: %d %d\n",i,j);
printf(" timestep = %g\n",dt);
printf(" time remaining post-collision = %g\n",t_remain);
xsstart[0] = x[i][0] - dt*v[i][0];
xsstart[1] = x[i][1] - dt*v[i][1];
xsstart[2] = x[i][2] - dt*v[i][2];
xbstart[0] = x[j][0] - dt*v[j][0];
xbstart[1] = x[j][1] - dt*v[j][1];
xbstart[2] = x[j][2] - dt*v[j][2];
printf(" SRD start position = %g %g %g\n",
xsstart[0],xsstart[1],xsstart[2]);
printf(" BIG start position = %g %g %g\n",
xbstart[0],xbstart[1],xbstart[2]);
printf(" SRD coll position = %g %g %g\n",
xscoll[0],xscoll[1],xscoll[2]);
printf(" BIG coll position = %g %g %g\n",
xbcoll[0],xbcoll[1],xbcoll[2]);
printf(" SRD end position = %g %g %g\n",x[i][0],x[i][1],x[i][2]);
printf(" BIG end position = %g %g %g\n",x[j][0],x[j][1],x[j][2]);
printf(" SRD vel = %g %g %g\n",v[i][0],v[i][1],v[i][2]);
printf(" BIG vel = %g %g %g\n",v[j][0],v[j][1],v[j][2]);
printf(" surf norm = %g %g %g\n",norm[0],norm[1],norm[2]);
double rstart = sqrt((xsstart[0]-xbstart[0])*(xsstart[0]-xbstart[0]) +
(xsstart[1]-xbstart[1])*(xsstart[1]-xbstart[1]) +
(xsstart[2]-xbstart[2])*(xsstart[2]-xbstart[2]));
double rcoll = sqrt((xscoll[0]-xbcoll[0])*(xscoll[0]-xbcoll[0]) +
(xscoll[1]-xbcoll[1])*(xscoll[1]-xbcoll[1]) +
(xscoll[2]-xbcoll[2])*(xscoll[2]-xbcoll[2]));
double rend = sqrt((x[i][0]-x[j][0])*(x[i][0]-x[j][0]) +
(x[i][1]-x[j][1])*(x[i][1]-x[j][1]) +
(x[i][2]-x[j][2])*(x[i][2]-x[j][2]));
printf(" separation at start = %g\n",rstart);
printf(" separation at coll = %g\n",rcoll);
printf(" separation at end = %g\n",rend);
} else {
int dim = wallwhich[j] / 2;
printf("COLLISION between SRD %d and WALL %d\n",atom->tag[i],j);
printf(" bounce # = %d\n",ibounce+1);
printf(" local indices: %d %d\n",i,j);
printf(" timestep = %g\n",dt);
printf(" time remaining post-collision = %g\n",t_remain);
xsstart[0] = x[i][0] - dt*v[i][0];
xsstart[1] = x[i][1] - dt*v[i][1];
xsstart[2] = x[i][2] - dt*v[i][2];
xbstart[0] = xbstart[1] = xbstart[2] = 0.0;
xbstart[dim] = xwall[j] - dt*vwall[j];
printf(" SRD start position = %g %g %g\n",
xsstart[0],xsstart[1],xsstart[2]);
printf(" WALL start position = %g\n",xbstart[dim]);
printf(" SRD coll position = %g %g %g\n",
xscoll[0],xscoll[1],xscoll[2]);
printf(" WALL coll position = %g\n",xbcoll[dim]);
printf(" SRD end position = %g %g %g\n",x[i][0],x[i][1],x[i][2]);
printf(" WALL end position = %g\n",xwall[j]);
printf(" SRD vel = %g %g %g\n",v[i][0],v[i][1],v[i][2]);
printf(" WALL vel = %g\n",vwall[j]);
printf(" surf norm = %g %g %g\n",norm[0],norm[1],norm[2]);
double rstart = xsstart[dim]-xbstart[dim];
double rcoll = xscoll[dim]-xbcoll[dim];
double rend = x[dim][0]-xwall[j];
printf(" separation at start = %g\n",rstart);
printf(" separation at coll = %g\n",rcoll);
printf(" separation at end = %g\n",rend);
}
}

Event Timeline