1240 lines
35 KiB
C++
1240 lines
35 KiB
C++
// clang-format off
|
|
/* ----------------------------------------------------------------------
|
|
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
|
|
https://www.lammps.org/, Sandia National Laboratories
|
|
LAMMPS development team: developers@lammps.org
|
|
|
|
Copyright (2003) Sandia Corporation. Under the terms of Contract
|
|
DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
|
|
certain rights in this software. This software is distributed under
|
|
the GNU General Public License.
|
|
|
|
See the README file in the top-level LAMMPS directory.
|
|
------------------------------------------------------------------------- */
|
|
|
|
/* ----------------------------------------------------------------------
|
|
Contributing authors: Amit Kumar and Michael Bybee (UIUC)
|
|
Pieter in 't Veld (BASF), code restructuring
|
|
Dave Heine (Corning), polydispersity
|
|
------------------------------------------------------------------------- */
|
|
|
|
#include "pair_lubricateU_poly.h"
|
|
|
|
#include "atom.h"
|
|
#include "comm.h"
|
|
#include "domain.h"
|
|
#include "error.h"
|
|
#include "fix.h"
|
|
#include "fix_wall.h"
|
|
#include "force.h"
|
|
#include "input.h"
|
|
#include "math_const.h"
|
|
#include "memory.h"
|
|
#include "modify.h"
|
|
#include "neigh_list.h"
|
|
#include "neighbor.h"
|
|
#include "variable.h"
|
|
|
|
#include <cmath>
|
|
#include <cstring>
|
|
|
|
using namespace LAMMPS_NS;
|
|
using namespace MathConst;
|
|
|
|
#define TOL 1E-3 // tolerance for conjugate gradient
|
|
|
|
// same as fix_wall.cpp
|
|
|
|
enum{EDGE,CONSTANT,VARIABLE};
|
|
|
|
|
|
/* ---------------------------------------------------------------------- */
|
|
|
|
PairLubricateUPoly::PairLubricateUPoly(LAMMPS *lmp) :
|
|
PairLubricateU(lmp) {}
|
|
|
|
/* ----------------------------------------------------------------------
|
|
It first has to solve for the velocity of the particles such that
|
|
the net force on the particles is zero. NOTE: it has to be the last
|
|
type of pair interaction specified in the input file. Also, it
|
|
assumes that no other types of interactions, like k-space, is
|
|
present. As already mentioned, the net force on the particles after
|
|
this pair interaction would be identically zero.
|
|
---------------------------------------------------------------------- */
|
|
|
|
void PairLubricateUPoly::compute(int eflag, int vflag)
|
|
{
|
|
int i,j;
|
|
|
|
int nlocal = atom->nlocal;
|
|
int nghost = atom->nghost;
|
|
int nall = nlocal + nghost;
|
|
|
|
double **x = atom->x;
|
|
double **f = atom->f;
|
|
double **torque = atom->torque;
|
|
|
|
ev_init(eflag,vflag);
|
|
|
|
// grow per-atom arrays if necessary
|
|
// need to be atom->nmax in length
|
|
|
|
if (atom->nmax > nmax) {
|
|
memory->destroy(fl);
|
|
memory->destroy(Tl);
|
|
memory->destroy(xl);
|
|
nmax = atom->nmax;
|
|
memory->create(fl,nmax,3,"pair:fl");
|
|
memory->create(Tl,nmax,3,"pair:Tl");
|
|
memory->create(xl,nmax,3,"pair:xl");
|
|
}
|
|
|
|
if (6*list->inum > cgmax) {
|
|
memory->sfree(bcg);
|
|
memory->sfree(xcg);
|
|
memory->sfree(rcg);
|
|
memory->sfree(rcg1);
|
|
memory->sfree(pcg);
|
|
memory->sfree(RU);
|
|
cgmax = 6*list->inum;
|
|
memory->create(bcg,cgmax,"pair:bcg");
|
|
memory->create(xcg,cgmax,"pair:bcg");
|
|
memory->create(rcg,cgmax,"pair:bcg");
|
|
memory->create(rcg1,cgmax,"pair:bcg");
|
|
memory->create(pcg,cgmax,"pair:bcg");
|
|
memory->create(RU,cgmax,"pair:bcg");
|
|
}
|
|
|
|
// Added to implement midpoint integration scheme
|
|
// Save force, torque found so far. Also save the positions
|
|
|
|
for (i=0;i<nlocal+nghost;i++) {
|
|
for (j=0;j<3;j++) {
|
|
fl[i][j] = f[i][j];
|
|
Tl[i][j] = torque[i][j];
|
|
xl[i][j] = x[i][j];
|
|
}
|
|
}
|
|
|
|
// Stage one of Midpoint method
|
|
// Solve for velocities based on initial positions
|
|
|
|
iterate(atom->x,1);
|
|
|
|
// Find positions at half the timestep and store in xl
|
|
|
|
intermediates(nall,xl);
|
|
|
|
// Store back the saved forces and torques in original arrays
|
|
|
|
for (i=0;i<nlocal+nghost;i++) {
|
|
for (j=0;j<3;j++) {
|
|
f[i][j] = fl[i][j];
|
|
torque[i][j] = Tl[i][j];
|
|
}
|
|
}
|
|
|
|
// Stage two: This will give the final velocities
|
|
|
|
iterate(xl,2);
|
|
}
|
|
|
|
/* ------------------------------------------------------------------------
|
|
Stage one of midpoint method
|
|
------------------------------------------------------------------------- */
|
|
|
|
void PairLubricateUPoly::iterate(double **x, int stage)
|
|
{
|
|
int i,j,ii;
|
|
|
|
int inum = list->inum;
|
|
int *ilist = list->ilist;
|
|
int newton_pair = force->newton_pair;
|
|
|
|
double alpha,beta;
|
|
double normi,error,normig;
|
|
double send[2],recv[2],rcg_dot_rcg;
|
|
|
|
double **v = atom->v;
|
|
double **f = atom->f;
|
|
double **omega = atom->omega;
|
|
double **torque = atom->torque;
|
|
|
|
// First compute R_FE*E
|
|
|
|
compute_RE(x);
|
|
|
|
// Reverse communication of forces and torques to
|
|
// accumulate the net force on each of the particles
|
|
|
|
if (newton_pair) comm->reverse_comm();
|
|
|
|
// CONJUGATE GRADIENT
|
|
// Find the right hand side= -ve of all forces/torques
|
|
// b = 6*Npart in overall size
|
|
|
|
for (ii = 0; ii < inum; ii++) {
|
|
i = ilist[ii];
|
|
for (j = 0; j < 3; j++) {
|
|
bcg[6*ii+j] = -f[i][j];
|
|
bcg[6*ii+j+3] = -torque[i][j];
|
|
}
|
|
}
|
|
|
|
// Start solving the equation : F^H = -F^P -F^B - F^H_{Ef}
|
|
// Store initial guess for velocity and angular-velocities/angular momentum
|
|
// NOTE velocities and angular velocities are assumed relative to the fluid
|
|
|
|
for (ii=0;ii<inum;ii++)
|
|
for (j=0;j<3;j++) {
|
|
xcg[6*ii+j] = 0.0;
|
|
xcg[6*ii+j+3] = 0.0;
|
|
}
|
|
|
|
// Copy initial guess to the global arrays to be acted upon by R_{FU}
|
|
// and returned by f and torque arrays
|
|
|
|
copy_vec_uo(inum,xcg,v,omega);
|
|
|
|
// set velocities for ghost particles
|
|
|
|
comm->forward_comm(this);
|
|
|
|
// Find initial residual
|
|
|
|
compute_RU(x);
|
|
|
|
// reverse communication of forces and torques
|
|
|
|
if (newton_pair) comm->reverse_comm();
|
|
|
|
copy_uo_vec(inum,f,torque,RU);
|
|
|
|
for (i=0;i<6*inum;i++)
|
|
rcg[i] = bcg[i] - RU[i];
|
|
|
|
// Set initial conjugate direction
|
|
|
|
for (i=0;i<6*inum;i++)
|
|
pcg[i] = rcg[i];
|
|
|
|
// Find initial norm of the residual or norm of the RHS (either is fine)
|
|
|
|
normi = dot_vec_vec(6*inum,bcg,bcg);
|
|
|
|
MPI_Allreduce(&normi,&normig,1,MPI_DOUBLE,MPI_SUM,world);
|
|
|
|
// Loop until convergence
|
|
|
|
do {
|
|
// find R*p
|
|
|
|
copy_vec_uo(inum,pcg,v,omega);
|
|
|
|
// set velocities for ghost particles
|
|
|
|
comm->forward_comm(this);
|
|
|
|
compute_RU(x);
|
|
|
|
// reverse communication of forces and torques
|
|
|
|
if (newton_pair) comm->reverse_comm();
|
|
|
|
copy_uo_vec(inum,f,torque,RU);
|
|
|
|
// Find alpha
|
|
|
|
send[0] = dot_vec_vec(6*inum,rcg,rcg);
|
|
send[1] = dot_vec_vec(6*inum,RU,pcg);
|
|
|
|
MPI_Allreduce(send,recv,2,MPI_DOUBLE,MPI_SUM,world);
|
|
|
|
alpha = recv[0]/recv[1];
|
|
rcg_dot_rcg = recv[0];
|
|
|
|
// Find new x
|
|
|
|
for (i=0;i<6*inum;i++)
|
|
xcg[i] = xcg[i] + alpha*pcg[i];
|
|
|
|
// find new residual
|
|
|
|
for (i=0;i<6*inum;i++)
|
|
rcg1[i] = rcg[i] - alpha*RU[i];
|
|
|
|
// find beta
|
|
|
|
send[0] = dot_vec_vec(6*inum,rcg1,rcg1);
|
|
|
|
MPI_Allreduce(send,recv,1,MPI_DOUBLE,MPI_SUM,world);
|
|
|
|
beta = recv[0]/rcg_dot_rcg;
|
|
|
|
// Find new conjugate direction
|
|
|
|
for (i=0;i<6*inum;i++)
|
|
pcg[i] = rcg1[i] + beta*pcg[i];
|
|
|
|
for (i=0;i<6*inum;i++)
|
|
rcg[i] = rcg1[i];
|
|
|
|
// Find relative error
|
|
|
|
error = sqrt(recv[0]/normig);
|
|
|
|
} while (error > TOL);
|
|
|
|
|
|
// update the final converged velocities in respective arrays
|
|
|
|
copy_vec_uo(inum,xcg,v,omega);
|
|
|
|
// set velocities for ghost particles
|
|
|
|
comm->forward_comm(this);
|
|
|
|
// compute the viscosity/pressure
|
|
|
|
if (evflag && stage == 2) compute_Fh(x);
|
|
|
|
// find actual particle's velocities from relative velocities
|
|
// only non-zero component of fluid's vel : vx=gdot*y and wz=-gdot/2
|
|
|
|
for (ii=0;ii<inum;ii++) {
|
|
i = ilist[ii];
|
|
v[i][0] = v[i][0] + gdot*x[i][1];
|
|
omega[i][2] = omega[i][2] - gdot/2.0;
|
|
}
|
|
}
|
|
|
|
/* ------------------------------------------------------------------------
|
|
This function computes the final hydrodynamic force once the
|
|
velocities have converged.
|
|
------------------------------------------------------------------------- */
|
|
|
|
void PairLubricateUPoly::compute_Fh(double **x)
|
|
{
|
|
int i,j,ii,jj,inum,jnum,itype,jtype;
|
|
int *ilist,*jlist,*numneigh,**firstneigh;
|
|
|
|
int *type = atom->type;
|
|
int nlocal = atom->nlocal;
|
|
int nghost = atom->nghost;
|
|
int newton_pair = force->newton_pair;
|
|
|
|
double xtmp,ytmp,ztmp,delx,dely,delz,fx,fy,fz;
|
|
double rsq,r,h_sep,radi,radj;
|
|
double vr1,vr2,vr3,vnnr,vn1,vn2,vn3;
|
|
double vt1,vt2,vt3;
|
|
double vi[3],vj[3],wi[3],wj[3],xl[3],jl[3],pre[2];
|
|
|
|
double **v = atom->v;
|
|
double **f = atom->f;
|
|
double **omega = atom->omega;
|
|
double **torque = atom->torque;
|
|
double *radius = atom->radius;
|
|
|
|
double beta[2][5];
|
|
double vxmu2f = force->vxmu2f;
|
|
double a_sq = 0.0;
|
|
double a_sh = 0.0;
|
|
|
|
inum = list->inum;
|
|
ilist = list->ilist;
|
|
numneigh = list->numneigh;
|
|
firstneigh = list->firstneigh;
|
|
|
|
beta[0][0] = beta[1][0] = beta[1][4] = 0.0;
|
|
|
|
// This section of code adjusts R0/RT0/RS0 if necessary due to changes
|
|
// in the volume fraction as a result of fix deform or moving walls
|
|
|
|
double dims[3], wallcoord;
|
|
if (flagVF) // Flag for volume fraction corrections
|
|
if (flagdeform || flagwall == 2) { // Possible changes in volume fraction
|
|
if (flagdeform && !flagwall)
|
|
for (j = 0; j < 3; j++)
|
|
dims[j] = domain->prd[j];
|
|
else if (flagwall == 2 || (flagdeform && flagwall == 1)) {
|
|
double wallhi[3], walllo[3];
|
|
for (int j = 0; j < 3; j++) {
|
|
wallhi[j] = domain->prd[j];
|
|
walllo[j] = 0;
|
|
}
|
|
for (int m = 0; m < wallfix->nwall; m++) {
|
|
int dim = wallfix->wallwhich[m] / 2;
|
|
int side = wallfix->wallwhich[m] % 2;
|
|
if (wallfix->xstyle[m] == VARIABLE) {
|
|
wallcoord = input->variable->compute_equal(wallfix->xindex[m]);
|
|
}
|
|
else wallcoord = wallfix->coord0[m];
|
|
if (side == 0) walllo[dim] = wallcoord;
|
|
else wallhi[dim] = wallcoord;
|
|
}
|
|
for (int j = 0; j < 3; j++)
|
|
dims[j] = wallhi[j] - walllo[j];
|
|
}
|
|
double vol_T = dims[0]*dims[1]*dims[2];
|
|
double vol_f = vol_P/vol_T;
|
|
if (flaglog == 0) {
|
|
//R0 = 6*MY_PI*mu*(1.0 + 2.16*vol_f);
|
|
//RT0 = 8*MY_PI*mu;
|
|
RS0 = 20.0/3.0*MY_PI*mu*(1.0 + 3.33*vol_f + 2.80*vol_f*vol_f);
|
|
} else {
|
|
//R0 = 6*MY_PI*mu*(1.0 + 2.725*vol_f - 6.583*vol_f*vol_f);
|
|
//RT0 = 8*MY_PI*mu*(1.0 + 0.749*vol_f - 2.469*vol_f*vol_f);
|
|
RS0 = 20.0/3.0*MY_PI*mu*(1.0 + 3.64*vol_f - 6.95*vol_f*vol_f);
|
|
}
|
|
}
|
|
|
|
// end of R0 adjustment code
|
|
// Set force to zero which is the final value after this pair interaction
|
|
|
|
for (i=0;i<nlocal+nghost;i++)
|
|
for (j=0;j<3;j++) {
|
|
f[i][j] = 0.0;
|
|
torque[i][j] = 0.0;
|
|
}
|
|
|
|
// reverse communication of forces and torques
|
|
|
|
if (newton_pair) comm->reverse_comm(); // not really needed
|
|
|
|
// Find additional contribution from the stresslets
|
|
|
|
for (ii = 0; ii < inum; ii++) {
|
|
i = ilist[ii];
|
|
xtmp = x[i][0];
|
|
ytmp = x[i][1];
|
|
ztmp = x[i][2];
|
|
itype = type[i];
|
|
radi = radius[i];
|
|
jlist = firstneigh[i];
|
|
jnum = numneigh[i];
|
|
pre[1] = 8.0*(pre[0] = MY_PI*mu*radi)*radi*radi; // BROKEN?? Should be "+"??
|
|
pre[0] *= 6.0;
|
|
|
|
// Find the contribution to stress from isotropic RS0
|
|
// Set pseudo force to obtain the required contribution
|
|
// need to set delx and fy only
|
|
|
|
fx = 0.0; delx = radi;
|
|
fy = vxmu2f*RS0*pow(radi,3.0)*gdot/2.0/radi; dely = 0.0;
|
|
fz = 0.0; delz = 0.0;
|
|
if (evflag)
|
|
ev_tally_xyz(i,i,nlocal,newton_pair,0.0,0.0,-fx,-fy,-fz,delx,dely,delz);
|
|
|
|
// Find angular velocity
|
|
|
|
wi[0] = omega[i][0];
|
|
wi[1] = omega[i][1];
|
|
wi[2] = omega[i][2];
|
|
|
|
for (jj = 0; jj < jnum; jj++) {
|
|
j = jlist[jj];
|
|
j &= NEIGHMASK;
|
|
|
|
delx = xtmp - x[j][0];
|
|
dely = ytmp - x[j][1];
|
|
delz = ztmp - x[j][2];
|
|
rsq = delx*delx + dely*dely + delz*delz;
|
|
jtype = type[j];
|
|
radj = radius[j];
|
|
|
|
if (rsq < cutsq[itype][jtype]) {
|
|
r = sqrt(rsq);
|
|
|
|
// Use omega directly if it exists, else angmom
|
|
// angular momentum = I*omega = 2/5 * M*R^2 * omega
|
|
|
|
wj[0] = omega[j][0];
|
|
wj[1] = omega[j][1];
|
|
wj[2] = omega[j][2];
|
|
|
|
// loc of the point of closest approach on particle i from its cente
|
|
// POC for j is in opposite direction as for i
|
|
|
|
xl[0] = -delx/r*radi;
|
|
xl[1] = -dely/r*radi;
|
|
xl[2] = -delz/r*radi;
|
|
jl[0] = delx/r*radj;
|
|
jl[1] = dely/r*radj;
|
|
jl[2] = delz/r*radj;
|
|
|
|
h_sep = r - radi-radj;
|
|
|
|
// velocity at the point of closest approach on both particles
|
|
// v = v + omega_cross_xl
|
|
|
|
// particle i
|
|
|
|
vi[0] = v[i][0] + (wi[1]*xl[2] - wi[2]*xl[1]);
|
|
vi[1] = v[i][1] + (wi[2]*xl[0] - wi[0]*xl[2]);
|
|
vi[2] = v[i][2] + (wi[0]*xl[1] - wi[1]*xl[0]);
|
|
|
|
// particle j
|
|
|
|
vj[0] = v[j][0] + (wj[1]*jl[2] - wj[2]*jl[1]);
|
|
vj[1] = v[j][1] + (wj[2]*jl[0] - wj[0]*jl[2]);
|
|
vj[2] = v[j][2] + (wj[0]*jl[1] - wj[1]*jl[0]);
|
|
|
|
|
|
// Relative velocity at the point of closest approach
|
|
// include contribution from Einf of the fluid
|
|
|
|
vr1 = vi[0] - vj[0] -
|
|
2.0*(Ef[0][0]*xl[0] + Ef[0][1]*xl[1] + Ef[0][2]*xl[2]);
|
|
vr2 = vi[1] - vj[1] -
|
|
2.0*(Ef[1][0]*xl[0] + Ef[1][1]*xl[1] + Ef[1][2]*xl[2]);
|
|
vr3 = vi[2] - vj[2] -
|
|
2.0*(Ef[2][0]*xl[0] + Ef[2][1]*xl[1] + Ef[2][2]*xl[2]);
|
|
|
|
// Normal component (vr.n)n
|
|
|
|
vnnr = (vr1*delx + vr2*dely + vr3*delz)/r;
|
|
vn1 = vnnr*delx/r;
|
|
vn2 = vnnr*dely/r;
|
|
vn3 = vnnr*delz/r;
|
|
|
|
// Tangential component vr - (vr.n)n
|
|
|
|
vt1 = vr1 - vn1;
|
|
vt2 = vr2 - vn2;
|
|
vt3 = vr3 - vn3;
|
|
|
|
// Find the scalar resistances a_sq, a_sh and a_pu
|
|
|
|
// If less than the minimum gap use the minimum gap instead
|
|
|
|
if (r < cut_inner[itype][jtype])
|
|
h_sep = cut_inner[itype][jtype] - radi-radj;
|
|
|
|
// Scale h_sep by radi
|
|
|
|
h_sep = h_sep/radi;
|
|
beta[0][1] = radj/radi;
|
|
beta[1][1] = 1.0 + beta[0][1];
|
|
|
|
/*beta0 = radj/radi;
|
|
beta1 = 1.0 + beta0;*/
|
|
|
|
// Scalar resistances
|
|
|
|
if (flaglog) {
|
|
beta[0][2] = beta[0][1]*beta[0][1];
|
|
beta[0][3] = beta[0][2]*beta[0][1];
|
|
beta[0][4] = beta[0][3]*beta[0][1];
|
|
beta[1][2] = beta[1][1]*beta[1][1];
|
|
beta[1][3] = beta[1][2]*beta[1][1];
|
|
double log_h_sep_beta13 = log(1.0/h_sep)/beta[1][3];
|
|
double h_sep_beta11 = h_sep/beta[1][1];
|
|
|
|
a_sq = pre[0]*(beta[0][2]/beta[1][2]/h_sep
|
|
+((0.2+1.4*beta[0][1]+0.2*beta[0][2])
|
|
+(1.0+18.0*(beta[0][1]+beta[0][3])-29.0*beta[0][2]
|
|
+beta[0][4])*h_sep_beta11/21.0)*log_h_sep_beta13);
|
|
|
|
a_sh = pre[0]*((8.0*(beta[0][1]+beta[0][3])+4.0*beta[0][2])/15.0
|
|
+(64.0-180.0*(beta[0][1]+beta[0][3])+232.0*beta[0][2]
|
|
+64.0*beta[0][4])*h_sep_beta11/375.0)*log_h_sep_beta13;
|
|
|
|
/*a_sq = beta0*beta0/beta1/beta1/h_sep
|
|
+(1.0+7.0*beta0+beta0*beta0)/5.0/pow(beta1,3)*log(1.0/h_sep);
|
|
a_sq += (1.0+18.0*beta0-29.0*beta0*beta0+18.0*pow(beta0,3)
|
|
+pow(beta0,4))/21.0/pow(beta1,4)*h_sep*log(1.0/h_sep);
|
|
a_sq *= 6.0*MY_PI*mu*radi;
|
|
|
|
a_sh = 4.0*beta0*(2.0+beta0
|
|
+2.0*beta0*beta0)/15.0/pow(beta1,3)*log(1.0/h_sep);
|
|
a_sh += 4.0*(16.0-45.0*beta0+58.0*beta0*beta0-45.0*pow(beta0,3)
|
|
+16.0*pow(beta0,4))/375.0/pow(beta1,4)*h_sep*log(1.0/h_sep);
|
|
a_sh *= 6.0*MY_PI*mu*radi;*/
|
|
} else {
|
|
//a_sq = 6.0*MY_PI*mu*radi*(beta0*beta0/beta1/beta1/h_sep);
|
|
a_sq = pre[0]*(beta[0][1]*beta[0][1]/(beta[1][1]*beta[1][1]*h_sep));
|
|
}
|
|
|
|
// Find force due to squeeze type motion
|
|
|
|
fx = a_sq*vn1;
|
|
fy = a_sq*vn2;
|
|
fz = a_sq*vn3;
|
|
|
|
// Find force due to all shear kind of motions
|
|
|
|
if (flaglog) {
|
|
fx = fx + a_sh*vt1;
|
|
fy = fy + a_sh*vt2;
|
|
fz = fz + a_sh*vt3;
|
|
}
|
|
|
|
// Scale forces to obtain in appropriate units
|
|
|
|
fx = vxmu2f*fx;
|
|
fy = vxmu2f*fy;
|
|
fz = vxmu2f*fz;
|
|
|
|
// set j = nlocal so that only I gets tallied
|
|
|
|
if (evflag) ev_tally_xyz(i,nlocal,nlocal,0,
|
|
0.0,0.0,-fx,-fy,-fz,delx,dely,delz);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/* ----------------------------------------------------------------------
|
|
computes R_FU * U
|
|
---------------------------------------------------------------------- */
|
|
|
|
void PairLubricateUPoly::compute_RU(double **x)
|
|
{
|
|
int i,j,ii,jj,inum,jnum,itype,jtype;
|
|
int *ilist,*jlist,*numneigh,**firstneigh;
|
|
|
|
int *type = atom->type;
|
|
int nlocal = atom->nlocal;
|
|
int nghost = atom->nghost;
|
|
|
|
double xtmp,ytmp,ztmp,delx,dely,delz,fx,fy,fz,tx,ty,tz;
|
|
double rsq,r,radi,radj,h_sep;
|
|
double vr1,vr2,vr3,vnnr,vn1,vn2,vn3;
|
|
double vt1,vt2,vt3,wdotn,wt1,wt2,wt3;
|
|
double vi[3],vj[3],wi[3],wj[3],xl[3],jl[3],pre[2];
|
|
|
|
double **v = atom->v;
|
|
double **f = atom->f;
|
|
double **omega = atom->omega;
|
|
double **torque = atom->torque;
|
|
double *radius = atom->radius;
|
|
|
|
double beta[2][5];
|
|
double vxmu2f = force->vxmu2f;
|
|
double a_sq = 0.0;
|
|
double a_sh = 0.0;
|
|
double a_pu = 0.0;
|
|
|
|
inum = list->inum;
|
|
ilist = list->ilist;
|
|
numneigh = list->numneigh;
|
|
firstneigh = list->firstneigh;
|
|
|
|
beta[0][0] = beta[1][0] = beta[1][4] = 0.0;
|
|
|
|
// This section of code adjusts R0/RT0/RS0 if necessary due to changes
|
|
// in the volume fraction as a result of fix deform or moving walls
|
|
|
|
double dims[3], wallcoord;
|
|
if (flagVF) // Flag for volume fraction corrections
|
|
if (flagdeform || flagwall == 2) { // Possible changes in volume fraction
|
|
if (flagdeform && !flagwall)
|
|
for (j = 0; j < 3; j++)
|
|
dims[j] = domain->prd[j];
|
|
else if (flagwall == 2 || (flagdeform && flagwall == 1)) {
|
|
double wallhi[3], walllo[3];
|
|
for (j = 0; j < 3; j++) {
|
|
wallhi[j] = domain->prd[j];
|
|
walllo[j] = 0;
|
|
}
|
|
for (int m = 0; m < wallfix->nwall; m++) {
|
|
int dim = wallfix->wallwhich[m] / 2;
|
|
int side = wallfix->wallwhich[m] % 2;
|
|
if (wallfix->xstyle[m] == VARIABLE) {
|
|
wallcoord = input->variable->compute_equal(wallfix->xindex[m]);
|
|
}
|
|
else wallcoord = wallfix->coord0[m];
|
|
if (side == 0) walllo[dim] = wallcoord;
|
|
else wallhi[dim] = wallcoord;
|
|
}
|
|
for (j = 0; j < 3; j++)
|
|
dims[j] = wallhi[j] - walllo[j];
|
|
}
|
|
double vol_T = dims[0]*dims[1]*dims[2];
|
|
double vol_f = vol_P/vol_T;
|
|
if (flaglog == 0) {
|
|
R0 = 6*MY_PI*mu*(1.0 + 2.16*vol_f);
|
|
RT0 = 8*MY_PI*mu;
|
|
// RS0 = 20.0/3.0*MY_PI*mu*(1.0 + 3.33*vol_f + 2.80*vol_f*vol_f);
|
|
} else {
|
|
R0 = 6*MY_PI*mu*(1.0 + 2.725*vol_f - 6.583*vol_f*vol_f);
|
|
RT0 = 8*MY_PI*mu*(1.0 + 0.749*vol_f - 2.469*vol_f*vol_f);
|
|
// RS0 = 20.0/3.0*MY_PI*mu*(1.0 + 3.64*vol_f - 6.95*vol_f*vol_f);
|
|
}
|
|
}
|
|
|
|
// end of R0 adjustment code
|
|
|
|
// Initialize f to zero
|
|
|
|
for (i=0;i<nlocal+nghost;i++)
|
|
for (j=0;j<3;j++) {
|
|
f[i][j] = 0.0;
|
|
torque[i][j] = 0.0;
|
|
}
|
|
|
|
for (ii = 0; ii < inum; ii++) {
|
|
i = ilist[ii];
|
|
xtmp = x[i][0];
|
|
ytmp = x[i][1];
|
|
ztmp = x[i][2];
|
|
itype = type[i];
|
|
radi = radius[i];
|
|
jlist = firstneigh[i];
|
|
jnum = numneigh[i];
|
|
pre[1] = 8.0*(pre[0] = MY_PI*mu*radi)*radi*radi;
|
|
pre[0] *= 6.0;
|
|
|
|
// Find angular velocity
|
|
|
|
wi[0] = omega[i][0];
|
|
wi[1] = omega[i][1];
|
|
wi[2] = omega[i][2];
|
|
|
|
// Contribution due to the isotropic terms
|
|
|
|
f[i][0] += -vxmu2f*R0*radi*v[i][0];
|
|
f[i][1] += -vxmu2f*R0*radi*v[i][1];
|
|
f[i][2] += -vxmu2f*R0*radi*v[i][2];
|
|
|
|
const double radi3 = radi*radi*radi;
|
|
torque[i][0] += -vxmu2f*RT0*radi3*wi[0];
|
|
torque[i][1] += -vxmu2f*RT0*radi3*wi[1];
|
|
torque[i][2] += -vxmu2f*RT0*radi3*wi[2];
|
|
|
|
if (!flagHI) continue;
|
|
|
|
for (jj = 0; jj < jnum; jj++) {
|
|
j = jlist[jj];
|
|
j &= NEIGHMASK;
|
|
|
|
delx = xtmp - x[j][0];
|
|
dely = ytmp - x[j][1];
|
|
delz = ztmp - x[j][2];
|
|
rsq = delx*delx + dely*dely + delz*delz;
|
|
jtype = type[j];
|
|
radj = radius[j];
|
|
|
|
if (rsq < cutsq[itype][jtype]) {
|
|
r = sqrt(rsq);
|
|
|
|
wj[0] = omega[j][0];
|
|
wj[1] = omega[j][1];
|
|
wj[2] = omega[j][2];
|
|
|
|
// loc of the point of closest approach on particle i from its center
|
|
|
|
xl[0] = -delx/r*radi;
|
|
xl[1] = -dely/r*radi;
|
|
xl[2] = -delz/r*radi;
|
|
jl[0] = delx/r*radj;
|
|
jl[1] = dely/r*radj;
|
|
jl[2] = delz/r*radj;
|
|
|
|
// velocity at the point of closest approach on both particles
|
|
// v = v + omega_cross_xl
|
|
|
|
// particle i
|
|
|
|
vi[0] = v[i][0] + (wi[1]*xl[2] - wi[2]*xl[1]);
|
|
vi[1] = v[i][1] + (wi[2]*xl[0] - wi[0]*xl[2]);
|
|
vi[2] = v[i][2] + (wi[0]*xl[1] - wi[1]*xl[0]);
|
|
|
|
// particle j
|
|
|
|
vj[0] = v[j][0] + (wj[1]*jl[2] - wj[2]*jl[1]);
|
|
vj[1] = v[j][1] + (wj[2]*jl[0] - wj[0]*jl[2]);
|
|
vj[2] = v[j][2] + (wj[0]*jl[1] - wj[1]*jl[0]);
|
|
|
|
// Find the scalar resistances a_sq and a_sh
|
|
|
|
h_sep = r - radi-radj;
|
|
|
|
// If less than the minimum gap use the minimum gap instead
|
|
|
|
if (r < cut_inner[itype][jtype])
|
|
h_sep = cut_inner[itype][jtype] - radi-radj;
|
|
|
|
// Scale h_sep by radi
|
|
|
|
h_sep = h_sep/radi;
|
|
beta[0][1] = radj/radi;
|
|
beta[1][1] = 1.0 + beta[0][1];
|
|
|
|
// Scalar resistances
|
|
|
|
if (flaglog) {
|
|
beta[0][2] = beta[0][1]*beta[0][1];
|
|
beta[0][3] = beta[0][2]*beta[0][1];
|
|
beta[0][4] = beta[0][3]*beta[0][1];
|
|
beta[1][2] = beta[1][1]*beta[1][1];
|
|
beta[1][3] = beta[1][2]*beta[1][1];
|
|
double log_h_sep = log(1.0/h_sep);
|
|
double log_h_sep_beta13 = log(1.0/h_sep)/beta[1][3];
|
|
double h_sep_beta11 = h_sep/beta[1][1];
|
|
|
|
a_sq = pre[0]*(beta[0][2]/beta[1][2]/h_sep
|
|
+((0.2+1.4*beta[0][1]+0.2*beta[0][2])
|
|
+(1.0+18.0*(beta[0][1]+beta[0][3])-29.0*beta[0][2]
|
|
+beta[0][4])*h_sep_beta11/21.0)*log_h_sep_beta13);
|
|
|
|
a_sh = pre[0]*((8.0*(beta[0][1]+beta[0][3])+4.0*beta[0][2])/15.0
|
|
+(64.0-180.0*(beta[0][1]+beta[0][3])+232.0*beta[0][2]
|
|
+64.0*beta[0][4])*h_sep_beta11/375.0)*log_h_sep_beta13;
|
|
|
|
// old invalid eq for pumping term
|
|
// changed 29Jul16 from eq 9.25 -> 9.27 in Kim and Karilla
|
|
// a_pu = pre[1]*((0.4*beta[0][1]+0.1*beta[0][2])*beta[1][1]
|
|
// +(0.128-0.132*beta[0][1]+0.332*beta[0][2]
|
|
// +0.172*beta[0][3])*h_sep)*log_h_sep_beta13;
|
|
a_pu = pre[1]*(0.4*beta[0][1]*beta[1][1]
|
|
+(0.128+0.096*beta[0][1]+0.528*beta[0][2])*beta[1][2]*h_sep)
|
|
*log_h_sep;
|
|
|
|
/*//a_sq = 6*MY_PI*mu*radi*(1.0/4.0/h_sep + 9.0/40.0*log(1/h_sep));
|
|
a_sq = beta0*beta0/beta1/beta1/h_sep
|
|
+(1.0+7.0*beta0+beta0*beta0)/5.0/pow(beta1,3)*log(1.0/h_sep);
|
|
a_sq += (1.0+18.0*beta0-29.0*beta0*beta0+18.0*pow(beta0,3)
|
|
+pow(beta0,4))/21.0/pow(beta1,4)*h_sep*log(1.0/h_sep);
|
|
a_sq *= 6.0*MY_PI*mu*radi;
|
|
|
|
a_sh = 4.0*beta0*(2.0+beta0
|
|
+2.0*beta0*beta0)/15.0/pow(beta1,3)*log(1.0/h_sep);
|
|
a_sh += 4.0*(16.0-45.0*beta0+58.0*beta0*beta0-45.0*pow(beta0,3)
|
|
+16.0*pow(beta0,4))/375.0/pow(beta1,4)*h_sep*log(1.0/h_sep);
|
|
a_sh *= 6.0*MY_PI*mu*radi;
|
|
|
|
a_pu = beta0*(4.0+beta0)/10.0/beta1/beta1*log(1.0/h_sep);
|
|
a_pu += (32.0-33.0*beta0+83.0*beta0*beta0
|
|
+43.0*pow(beta0,3))/250.0/pow(beta1,3)*h_sep*log(1.0/h_sep);
|
|
a_pu *= 8.0*MY_PI*mu*pow(radi,3);*/
|
|
} else
|
|
a_sq = pre[0]*(beta[0][1]*beta[0][1]/(beta[1][1]*beta[1][1]*h_sep));
|
|
|
|
// Relative velocity at the point of closest approach
|
|
|
|
vr1 = vi[0] - vj[0];
|
|
vr2 = vi[1] - vj[1];
|
|
vr3 = vi[2] - vj[2];
|
|
|
|
// Normal component (vr.n)n
|
|
|
|
vnnr = (vr1*delx + vr2*dely + vr3*delz)/r;
|
|
vn1 = vnnr*delx/r;
|
|
vn2 = vnnr*dely/r;
|
|
vn3 = vnnr*delz/r;
|
|
|
|
// Tangential component vr - (vr.n)n
|
|
|
|
vt1 = vr1 - vn1;
|
|
vt2 = vr2 - vn2;
|
|
vt3 = vr3 - vn3;
|
|
|
|
// Find force due to squeeze type motion
|
|
|
|
fx = a_sq*vn1;
|
|
fy = a_sq*vn2;
|
|
fz = a_sq*vn3;
|
|
|
|
// Find force due to all shear kind of motions
|
|
|
|
if (flaglog) {
|
|
fx = fx + a_sh*vt1;
|
|
fy = fy + a_sh*vt2;
|
|
fz = fz + a_sh*vt3;
|
|
}
|
|
|
|
// Scale forces to obtain in appropriate units
|
|
|
|
fx = vxmu2f*fx;
|
|
fy = vxmu2f*fy;
|
|
fz = vxmu2f*fz;
|
|
|
|
// Add to the total force
|
|
|
|
f[i][0] -= fx;
|
|
f[i][1] -= fy;
|
|
f[i][2] -= fz;
|
|
|
|
// Find torque due to this force
|
|
|
|
if (flaglog) {
|
|
tx = xl[1]*fz - xl[2]*fy;
|
|
ty = xl[2]*fx - xl[0]*fz;
|
|
tz = xl[0]*fy - xl[1]*fx;
|
|
|
|
// Why a scale factor ?
|
|
|
|
torque[i][0] -= vxmu2f*tx;
|
|
torque[i][1] -= vxmu2f*ty;
|
|
torque[i][2] -= vxmu2f*tz;
|
|
|
|
// Torque due to a_pu
|
|
|
|
wdotn = ((wi[0]-wj[0])*delx +
|
|
(wi[1]-wj[1])*dely + (wi[2]-wj[2])*delz)/r;
|
|
wt1 = (wi[0]-wj[0]) - wdotn*delx/r;
|
|
wt2 = (wi[1]-wj[1]) - wdotn*dely/r;
|
|
wt3 = (wi[2]-wj[2]) - wdotn*delz/r;
|
|
|
|
tx = a_pu*wt1;
|
|
ty = a_pu*wt2;
|
|
tz = a_pu*wt3;
|
|
|
|
// add to total
|
|
|
|
torque[i][0] -= vxmu2f*tx;
|
|
torque[i][1] -= vxmu2f*ty;
|
|
torque[i][2] -= vxmu2f*tz;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/* ----------------------------------------------------------------------
|
|
This computes R_{FE}*E , where E is the rate of strain of tensor which is
|
|
known apriori, as it depends only on the known fluid velocity.
|
|
So, this part of the hydrodynamic interaction can be pre computed and
|
|
transferred to the RHS
|
|
---------------------------------------------------------------------- */
|
|
|
|
void PairLubricateUPoly::compute_RE(double **x)
|
|
{
|
|
int i,j,ii,jj,inum,jnum,itype,jtype;
|
|
int *ilist,*jlist,*numneigh,**firstneigh;
|
|
|
|
int *type = atom->type;
|
|
|
|
double xtmp,ytmp,ztmp,delx,dely,delz,fx,fy,fz,tx,ty,tz;
|
|
double rsq,r,h_sep,radi,radj;
|
|
//double beta0,beta1,lhsep;
|
|
double vr1,vr2,vr3,vnnr,vn1,vn2,vn3;
|
|
double vt1,vt2,vt3;
|
|
double xl[3],pre[2];
|
|
|
|
double **f = atom->f;
|
|
double **torque = atom->torque;
|
|
double *radius = atom->radius;
|
|
|
|
double beta[2][5];
|
|
double vxmu2f = force->vxmu2f;
|
|
double a_sq = 0.0;
|
|
double a_sh = 0.0;
|
|
|
|
if (!flagHI) return;
|
|
|
|
inum = list->inum;
|
|
ilist = list->ilist;
|
|
numneigh = list->numneigh;
|
|
firstneigh = list->firstneigh;
|
|
|
|
beta[0][0] = beta[1][0] = beta[1][4] = 0.0;
|
|
|
|
for (ii = 0; ii < inum; ii++) {
|
|
i = ilist[ii];
|
|
xtmp = x[i][0];
|
|
ytmp = x[i][1];
|
|
ztmp = x[i][2];
|
|
itype = type[i];
|
|
radi = radius[i];
|
|
jlist = firstneigh[i];
|
|
jnum = numneigh[i];
|
|
pre[1] = 8.0*(pre[0] = MY_PI*mu*radi)*radi*radi;
|
|
pre[0] *= 6.0;
|
|
|
|
// No contribution from isotropic terms due to E
|
|
for (jj = 0; jj < jnum; jj++) {
|
|
j = jlist[jj];
|
|
j &= NEIGHMASK;
|
|
|
|
delx = xtmp - x[j][0];
|
|
dely = ytmp - x[j][1];
|
|
delz = ztmp - x[j][2];
|
|
rsq = delx*delx + dely*dely + delz*delz;
|
|
jtype = type[j];
|
|
radj = radius[j];
|
|
|
|
if (rsq < cutsq[itype][jtype]) {
|
|
r = sqrt(rsq);
|
|
|
|
// loc of the point of closest approach on particle i from its center
|
|
|
|
xl[0] = -delx/r*radi;
|
|
xl[1] = -dely/r*radi;
|
|
xl[2] = -delz/r*radi;
|
|
|
|
// Find the scalar resistances a_sq and a_sh
|
|
|
|
h_sep = r - radi-radj;
|
|
|
|
// If less than the minimum gap use the minimum gap instead
|
|
|
|
if (r < cut_inner[itype][jtype])
|
|
h_sep = cut_inner[itype][jtype] - radi-radj;
|
|
|
|
// Scale h_sep by radi
|
|
|
|
h_sep = h_sep/radi;
|
|
beta[0][1] = radj/radi;
|
|
beta[1][1] = 1.0 + beta[0][1];
|
|
|
|
/*beta0 = radj/radi;
|
|
beta1 = 1.0 + beta0;
|
|
lhsep = log(1.0/h_sep);*/
|
|
|
|
// Scalar resistance for Squeeze type motions
|
|
|
|
|
|
if (flaglog) {
|
|
beta[0][2] = beta[0][1]*beta[0][1];
|
|
beta[0][3] = beta[0][2]*beta[0][1];
|
|
beta[0][4] = beta[0][3]*beta[0][1];
|
|
beta[1][2] = beta[1][1]*beta[1][1];
|
|
beta[1][3] = beta[1][2]*beta[1][1];
|
|
double log_h_sep_beta13 = log(1.0/h_sep)/beta[1][3];
|
|
double h_sep_beta11 = h_sep/beta[1][1];
|
|
|
|
a_sq = pre[0]*(beta[0][2]/beta[1][2]/h_sep
|
|
+((0.2+1.4*beta[0][1]+0.2*beta[0][2])
|
|
+(1.0+18.0*(beta[0][1]+beta[0][3])-29.0*beta[0][2]
|
|
+beta[0][4])*h_sep_beta11/21.0)*log_h_sep_beta13);
|
|
|
|
a_sh = pre[0]*((8.0*(beta[0][1]+beta[0][3])+4.0*beta[0][2])/15.0
|
|
+(64.0-180.0*(beta[0][1]+beta[0][3])+232.0*beta[0][2]
|
|
+64.0*beta[0][4])*h_sep_beta11/375.0)*log_h_sep_beta13;
|
|
} else
|
|
a_sq = pre[0]*(beta[0][1]*beta[0][1]/(beta[1][1]*beta[1][1]*h_sep));
|
|
|
|
// Relative velocity at the point of closest approach due to Ef only
|
|
|
|
vr1 = -2.0*(Ef[0][0]*xl[0] + Ef[0][1]*xl[1] + Ef[0][2]*xl[2]);
|
|
vr2 = -2.0*(Ef[1][0]*xl[0] + Ef[1][1]*xl[1] + Ef[1][2]*xl[2]);
|
|
vr3 = -2.0*(Ef[2][0]*xl[0] + Ef[2][1]*xl[1] + Ef[2][2]*xl[2]);
|
|
|
|
// Normal component (vr.n)n
|
|
|
|
vnnr = (vr1*delx + vr2*dely + vr3*delz)/r;
|
|
vn1 = vnnr*delx/r;
|
|
vn2 = vnnr*dely/r;
|
|
vn3 = vnnr*delz/r;
|
|
|
|
// Tangential component vr - (vr.n)n
|
|
|
|
vt1 = vr1 - vn1;
|
|
vt2 = vr2 - vn2;
|
|
vt3 = vr3 - vn3;
|
|
|
|
// Find force due to squeeze type motion
|
|
|
|
fx = a_sq*vn1;
|
|
fy = a_sq*vn2;
|
|
fz = a_sq*vn3;
|
|
|
|
// Find force due to all shear kind of motions
|
|
|
|
if (flaglog) {
|
|
fx = fx + a_sh*vt1;
|
|
fy = fy + a_sh*vt2;
|
|
fz = fz + a_sh*vt3;
|
|
}
|
|
|
|
// Scale forces to obtain in appropriate units
|
|
|
|
fx = vxmu2f*fx;
|
|
fy = vxmu2f*fy;
|
|
fz = vxmu2f*fz;
|
|
|
|
// Add to the total force
|
|
|
|
f[i][0] -= fx;
|
|
f[i][1] -= fy;
|
|
f[i][2] -= fz;
|
|
|
|
// Find torque due to this force
|
|
|
|
if (flaglog) {
|
|
tx = xl[1]*fz - xl[2]*fy;
|
|
ty = xl[2]*fx - xl[0]*fz;
|
|
tz = xl[0]*fy - xl[1]*fx;
|
|
|
|
// Why a scale factor ?
|
|
|
|
torque[i][0] -= vxmu2f*tx;
|
|
torque[i][1] -= vxmu2f*ty;
|
|
torque[i][2] -= vxmu2f*tz;
|
|
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/*-----------------------------------------------------------------------
|
|
global settings
|
|
------------------------------------------------------------------------- */
|
|
|
|
void PairLubricateUPoly::settings(int narg, char **arg)
|
|
{
|
|
if (narg < 5 || narg > 7) error->all(FLERR,"Illegal pair_style command");
|
|
|
|
mu = utils::numeric(FLERR,arg[0],false,lmp);
|
|
flaglog = utils::inumeric(FLERR,arg[1],false,lmp);
|
|
cut_inner_global = utils::numeric(FLERR,arg[2],false,lmp);
|
|
cut_global = utils::numeric(FLERR,arg[3],false,lmp);
|
|
gdot = utils::numeric(FLERR,arg[4],false,lmp);
|
|
|
|
flagHI = flagVF = 1;
|
|
if (narg >= 6) flagHI = utils::inumeric(FLERR,arg[5],false,lmp);
|
|
if (narg == 7) flagVF = utils::inumeric(FLERR,arg[6],false,lmp);
|
|
|
|
if (flaglog == 1 && flagHI == 0) {
|
|
error->warning(FLERR,"Cannot include log terms without 1/r terms; "
|
|
"setting flagHI to 1");
|
|
flagHI = 1;
|
|
}
|
|
|
|
// reset cutoffs that have been explicitly set
|
|
|
|
if (allocated) {
|
|
int i,j;
|
|
for (i = 1; i <= atom->ntypes; i++)
|
|
for (j = i; j <= atom->ntypes; j++)
|
|
if (setflag[i][j]) {
|
|
cut_inner[i][j] = cut_inner_global;
|
|
cut[i][j] = cut_global;
|
|
}
|
|
}
|
|
|
|
// Store the rate of strain tensor
|
|
|
|
Ef[0][0] = 0.0;
|
|
Ef[0][1] = 0.5*gdot;
|
|
Ef[0][2] = 0.0;
|
|
Ef[1][0] = 0.5*gdot;
|
|
Ef[1][1] = 0.0;
|
|
Ef[1][2] = 0.0;
|
|
Ef[2][0] = 0.0;
|
|
Ef[2][1] = 0.0;
|
|
Ef[2][2] = 0.0;
|
|
}
|
|
|
|
/* ----------------------------------------------------------------------
|
|
init specific to this pair style
|
|
------------------------------------------------------------------------- */
|
|
|
|
void PairLubricateUPoly::init_style()
|
|
{
|
|
if (force->newton_pair == 1)
|
|
error->all(FLERR,"Pair lubricateU/poly requires newton pair off");
|
|
if (comm->ghost_velocity == 0)
|
|
error->all(FLERR,
|
|
"Pair lubricateU/poly requires ghost atoms store velocity");
|
|
if (!atom->sphere_flag)
|
|
error->all(FLERR,"Pair lubricate/poly requires atom style sphere");
|
|
|
|
// ensure all particles are finite-size
|
|
// for pair hybrid, should limit test to types using the pair style
|
|
|
|
double *radius = atom->radius;
|
|
int nlocal = atom->nlocal;
|
|
|
|
for (int i = 0; i < nlocal; i++)
|
|
if (radius[i] == 0.0)
|
|
error->one(FLERR,"Pair lubricate/poly requires extended particles");
|
|
|
|
// Set the isotropic constants depending on the volume fraction
|
|
|
|
// Find the total volume
|
|
// check for fix deform, if exists it must use "remap v"
|
|
// If box will change volume, set appropriate flag so that volume
|
|
// and v.f. corrections are re-calculated at every step.
|
|
//
|
|
// If available volume is different from box volume
|
|
// due to walls, set volume appropriately; if walls will
|
|
// move, set appropriate flag so that volume and v.f. corrections
|
|
// are re-calculated at every step.
|
|
|
|
flagdeform = flagwall = 0;
|
|
for (int i = 0; i < modify->nfix; i++) {
|
|
if (strcmp(modify->fix[i]->style,"deform") == 0)
|
|
flagdeform = 1;
|
|
else if (strstr(modify->fix[i]->style,"wall") != nullptr) {
|
|
if (flagwall)
|
|
error->all(FLERR,
|
|
"Cannot use multiple fix wall commands with "
|
|
"pair lubricateU");
|
|
flagwall = 1; // Walls exist
|
|
wallfix = dynamic_cast<FixWall *>(modify->fix[i]);
|
|
if (wallfix->xflag) flagwall = 2; // Moving walls exist
|
|
}
|
|
}
|
|
|
|
// set the isotropic constants depending on the volume fraction
|
|
// vol_T = total volumeshearing = flagdeform = flagwall = 0;
|
|
|
|
double vol_T, wallcoord;
|
|
if (!flagwall) vol_T = domain->xprd*domain->yprd*domain->zprd;
|
|
else {
|
|
double wallhi[3], walllo[3];
|
|
for (int j = 0; j < 3; j++) {
|
|
wallhi[j] = domain->prd[j];
|
|
walllo[j] = 0;
|
|
}
|
|
for (int m = 0; m < wallfix->nwall; m++) {
|
|
int dim = wallfix->wallwhich[m] / 2;
|
|
int side = wallfix->wallwhich[m] % 2;
|
|
if (wallfix->xstyle[m] == VARIABLE) {
|
|
wallfix->xindex[m] = input->variable->find(wallfix->xstr[m]);
|
|
//Since fix->wall->init happens after pair->init_style
|
|
wallcoord = input->variable->compute_equal(wallfix->xindex[m]);
|
|
}
|
|
|
|
else wallcoord = wallfix->coord0[m];
|
|
|
|
if (side == 0) walllo[dim] = wallcoord;
|
|
else wallhi[dim] = wallcoord;
|
|
}
|
|
vol_T = (wallhi[0] - walllo[0]) * (wallhi[1] - walllo[1]) *
|
|
(wallhi[2] - walllo[2]);
|
|
}
|
|
|
|
// Assuming monodisperse spheres, find the volume of the particles
|
|
|
|
double volP = 0.0;
|
|
for (int i = 0; i < nlocal; i++)
|
|
volP += (4.0/3.0)*MY_PI*pow(atom->radius[i],3.0);
|
|
MPI_Allreduce(&volP,&vol_P,1,MPI_DOUBLE,MPI_SUM,world);
|
|
|
|
double vol_f = vol_P/vol_T;
|
|
|
|
//DRH volume fraction needs to be defined manually
|
|
// if excluded volume regions are present
|
|
// vol_f=0.5;
|
|
|
|
if (!flagVF) vol_f = 0;
|
|
|
|
if (!comm->me) {
|
|
if (logfile)
|
|
fprintf(logfile, "lubricateU: vol_f = %g, vol_p = %g, vol_T = %g\n",
|
|
vol_f,vol_P,vol_T);
|
|
if (screen)
|
|
fprintf(screen, "lubricateU: vol_f = %g, vol_p = %g, vol_T = %g\n",
|
|
vol_f,vol_P,vol_T);
|
|
}
|
|
|
|
// Set the isotropic constant
|
|
|
|
if (flaglog == 0) {
|
|
R0 = 6*MY_PI*mu*(1.0 + 2.16*vol_f);
|
|
RT0 = 8*MY_PI*mu; // Not needed actually
|
|
RS0 = 20.0/3.0*MY_PI*mu*(1.0 + 3.33*vol_f + 2.80*vol_f*vol_f);
|
|
} else {
|
|
R0 = 6*MY_PI*mu*(1.0 + 2.725*vol_f - 6.583*vol_f*vol_f);
|
|
RT0 = 8*MY_PI*mu*(1.0 + 0.749*vol_f - 2.469*vol_f*vol_f);
|
|
RS0 = 20.0/3.0*MY_PI*mu*(1.0 + 3.64*vol_f - 6.95*vol_f*vol_f);
|
|
}
|
|
|
|
neighbor->add_request(this, NeighConst::REQ_FULL);
|
|
}
|