On Oct 28, 2009, at 10:48 AM, Ranjit Chacko wrote:
This is the method I use to update the field on each time step:
def step(self):
self.t += self.dt
self.phi.updateOld()
residual=1
while residual>1.0e-1:
residual=self.equation.sweep(var=self.phi, dt=self.dt,
solver=self.solver)
Is this the right way to use sweep?
How did you pick "residual>1.0e-1" ? How do you know your solution is
converged at that value? I suspect you're getting dramatically
different answers because your solution is not converged.
self.phi.setValue(self.phi +
self.xsi*GaussianNoiseVariable(mesh=self.mesh, mean=0.0, variance=1))
I don't think it's harmful, but this is not how I would implement
Langevin noise.
I would do:
self.noise = self.xsi*GaussianNoiseVariable(mesh=self.mesh, mean=0.0,
variance=1)
self.equation = (TransientTerm() + ConvectionTerm() == DiffusionTerm()
+ ... + self.noise)
def step(self):
self.t += self.dt
self.phi.updateOld()
self.noise.scramble() # update the noise like this
residual=1
while residual>1.0e-1:
residual=self.equation.sweep(var=self.phi, dt=self.dt,
solver=self.solver)
It's not critical at this juncture, but the variance of Langevin term
should be correlated with both cell size and time step. This is
illustrated in the GaussianNoiseVariable doc string.
