Noise in MICRESS

[matt_hughes]

Hi MICRESS forum

I am using MICRESS 7.2 and am confused how the noise variable. How does this add noise to deltaG and how do I do this in MICRESS.


Many thanks,
Matthew

[Bernd]

Dear Matthew,

The keyword "noise" is part of the "dG-options" which allow manipulation of the driving force for phase transformation. The user can specify random noise on all interface cells by giving a noise amplitude (in J/cm3). It is important to note that noise is added after averaging ("avg") but before cutting ("max"). Using noise on the driving force can e.g. be used to help break-up of a planar front, especially if there are no other sources of noise present (like from TC-coupling).

Bernd

[matt_hughes]

Hi Bernd,

Thank you for the reply. How exactly does the value cause noise at the interface? Does it apply a random amount of noise to the solute values at the interface specified by the number? Is the noise gaussian?

Currently I am trying to run simulations that cause dendrites to fall back based on changes in the solidification conditions. I have ran a simulation till steady state for a set of growth conditions, I then change the velocity or thermal gradient independently to induce a dendrite to fall back. The problem i have is that i think the phase field simulations are too "perfect" so there is no initial perturbation for a dendrite to fall back despite growing in a solute field that may not be stable. Even when I restart simulations and change to values as low as 1% or 2% of the original velocity or thermal gradient, no dendrites fall back. Will adding noise cause a dendrite to fall back? And if so what value do you think is needed to create the right amount of noise? If the addition of noise on the surface does not cause a dendrite to fall back is there anything that can break the "perfectness" of the dendrites?

Thanks,
Matthew

[Bernd]

Dear Matthew,

The problem you describe is typical for cases where the initial microstructure is perfectly symmetric, e.g. when starting with initial dendrites at exactly identical distances (with periodic or symmetric boundary conditions at both sides). If symmetry cannot be avoided, noise from any source can break this symmetry if it is strong enough. In my view, you have three options:

1.) Avoid symmetry of the iniial microstructure.

2.) In case of TQ-coupling, if you do not use the "global" updating option without further restriction, there is always enough noise coming from not identical linearisation data. A typical setup would be to use "global" with an additional distance criterion which limits the size of the chunks which get common linearisation data. As these chunks constantly change location upon updating, there is an efficient noise which can be controlled by the chunk size and updating interval to certain extent.

3.) Using the noise on the driving force as discussed in the former post. It is non-gaussion (equal distribution within the noise interval) and is added to the driving force (thus, it is essentially noise on the calculated phase-field increment). For your application, you need to check how much noise is required. I would guess that 1-10 J/cm3 should work. It is recommended not to use unnecessarily high values in order not to introduce numerical artefacts or systematic bias.

Bernd

[matt_hughes]

Hi Bernd,

I have tried options 1 and 3 and still am not having any luck. I tested different amounts of noise and still do not see dendrites falling back after 100 seconds. I have tried avoiding symmetry by varying distances between dendrites by 5-15 microns. Do you think trying option 2 is now the most likely way to study dendrite elimination?

Thanks,
Matthew

[Bernd]

Hi Matthew,

I am not sure whether your problem is really due to missing noise. How do you set up your simulations? Do you set initial grains at equal distances at the beginning? How do your dendrites look lile in steady state?

Bernd