anneal_func(t, parameters["tau"], parameters["tau_ramp"],
parameters["t_ramp"]))
# Compute energy
# u_.assign(u_1)
Eout_0 = df.assemble(geo_map.form(E_0))
Eout_2 = df.assemble(geo_map.form(E_2))
E_before = Eout_0 + Eout_2
converged = False
while not converged:
try:
solver.solve()
converged = True
except:
info_blue("Did not converge. Chopping timestep.")
dt.chop()
info_blue("New timestep is: dt = {}".format(dt.get()))
if converged:
Eout_0 = df.assemble(geo_map.form(E_0))
Eout_2 = df.assemble(geo_map.form(E_2))
E_after = Eout_0 + Eout_2
dE = E_after - E_before
if not initial_step and dE > 1e-1 and False:
info_blue(
"Converged, but energy increased. Chopping timestep.")
info_blue("E_before = {}".format(E_before))
info_blue("E_after = {}".format(E_after))
info_blue("dE = {}".format(dE))
info_blue("dt = {}".format(dt.get()))
t+dt.get(),
parameters["tau"],
parameters["tau_ramp"],
parameters["t_ramp"]))
# Compute energy
u_.assign(u_1)
Eout_0 = df.assemble(geo_map.form(E_0))
Eout_2 = df.assemble(geo_map.form(E_2))
E_before = Eout_0 + Eout_2
try:
solver.solve()
converged = True
except:
info_blue("Did not converge. Chopping timestep.")
dt.chop()
info_blue("New timestep is: dt = {}".format(dt.get()))
Eout_0 = df.assemble(geo_map.form(E_0))
Eout_2 = df.assemble(geo_map.form(E_2))
E_after = Eout_0 + Eout_2
dE = E_after - E_before
if not initial_step and dE > 0.0:
dt.chop()
converged = False
initial_step = False
# Update time with final dt value
t += dt.get()
u_1.vector()[:] += np.random.normal(
0, initial_noise/(1+0.02*t),
len(u_1.vector().get_local())) # Annealing noise
solver.solve()
if tstep % 1 == 0:
# ts.dump(tstep)
ts.dump(t)
E_0 = (2*nu_**2 - 2 * geo_map.gab[i, j]*psi_.dx(i)*psi_.dx(j)
+ w(psi_, tau))
E_nonK = df.assemble(geo_map.form(E_0))
E_K = df.assemble(geo_map.form((h**2/12)*(
2*(4*nuhat_**2 + 4*H*nuhat_*nu_ - 5*K*nu_**2)
- 2 * (2*H*nuhat_ - 2*K*gab[i, j]*psi_.dx(i)*psi_.dx(j))
+ (tau/2)*K*psi_**2 + (1/4)*K*psi_**4)))
# Double checked for correctness
E_tot = E_nonK+E_K
grad_mu = ts.get_function("abs_grad_mu")
grad_mu_max = mpi_max(grad_mu.vector().get_local())
# Assigning timestep size according to grad_mu_max:
dt.assign(max(0.01, 0.5/grad_mu_max))
info_blue("t = {}".format(t))
info_blue("dt = {}".format(float(dt.values())))
info_blue("noise = {}".format(initial_noise))
ts.dump_stats(t, [grad_mu_max, float(
dt.values()), float(h.values()), E_nonK, E_K, E_tot], "data")
if tstep % parameters["checkpoint_intv"] == 0 or t >= T:
save_checkpoint(tstep, t, geo_map.ref_mesh,
u_, u_1, ts.folder, parameters)
