sandbox/bugs/tracer_noise.c
Possible bug with tracer + embedded boundaries + two phases (miscible)
There is a strange behaviour where “noise” appears on the tracer field when a solid is added to this scenario.
#include "embed.h"
#include "navier-stokes/centered.h"
#include "navier-stokes/perfs.h"
#include "tracer.h"
scalar f[];
scalar * tracers = {f};This following part was taken from two-phase-generic.h, and only the portion related to the viscosity computation on faces is included. If the entire file instead, convergence issues appear immediately (when embedded boundaries are present). This could be a separate issue or not (seems related to alpha field change from what I’ve seen).
#if 1
double mu1 = 0., mu2 = 0.;
event defaults (i = 0)
{
if (mu1 || mu2)
mu = new face vector;
}
#ifndef mu
# define mu(f) (clamp(f,0.,1.)*(mu1 - mu2) + mu2)
#endif
#define sf f
event properties (i++)
{
foreach_face() {
double ff = (sf[] + sf[-1])/2.;
if (mu1 || mu2) {
face vector muv = mu;
muv.x[] = fm.x[]*mu(ff);
}
}
}
#endif
//#include "two-phase-generic.h"The setup for the scenario is below:
int maxlevel = 5;
//Parameters
double U1 = 0.001;
double H1 = 0.000249999;
double tend = 1.5;
double f_error = 1e-2;
double u_error;
u.n[left] = dirichlet(U1*cs[]);
u.t[left] = dirichlet(0.);
p[left] = neumann(0.);
pf[left] = neumann(0.);
f[left] = 1.;
u.n[right] = neumann(0.);
u.t[right] = neumann(0.);
p[right] = dirichlet(0.);
pf[right] = dirichlet(0.);
u.n[embed] = dirichlet(0.);
u.t[embed] = dirichlet(0.);A boolean variable embSolid is used to execute the case with and without the solid:
bool embSolid = true;
int main()
{
L0 = 5.*H1;
origin (0.,-L0/2.);
N = 64;
DT = 0.01;
u_error = 0.05*U1;
mu1 = 0.0966;
mu2 = 0.0023;
run();
embSolid = false;
run();
}
event end (t=tend)
{}
event init (t = 0)
{If the solid is added, the “noise” on the tracer is present (also the statistic mgu.i for the viscosity grows considerably). Otherwise the effect is not present:
if (embSolid){
solid (cs, fs, intersection(-y+H1*2. ,+y+H1*2.));
fractions_cleanup (cs, fs, 0.0001);
}
}The adaptive mesh refinement used is below. Completely removing adaptive mesh refinement also makes the effect disappear:
event adapt (i++) {
if(embSolid)
adapt_wavelet ({cs,u,f}, (double[]){1e-3,u_error,u_error,f_error}, maxlevel);
else
adapt_wavelet ({u,f}, (double[]){u_error,u_error,f_error}, maxlevel);
}Output generation:
event output (t = 1.2){
if(embSolid)
output_ppm(f, file = "tracer_noise_1.png", n = 400, map=cool_warm, spread=3);
else
output_ppm(f, file = "tracer_noise_2.png", n = 400, map=cool_warm, spread=3);
}#Results obtained in both cases The results obtained for each setup are shown below:
