search

materialsinnovation / pymks

Materials Knowledge System in Python
https://pymks.readthedocs.io/
Other
115 stars 77 forks source link

correct make_cahn_hilliard initial microstructure values #137

Closed davidbrough1 closed 9 years ago

davidbrough1 commented 9 years ago

The initial microstructure values generated by the make_cahn_hilliard function come from an uniform distribution. They should come from a normal distribution with a small variance to more accurately represent the physical phenomenon.

wd15 commented 9 years ago

I think it needs to be a uniform distribution to seed the model properly. The "make_cahn_hilliard" function is to create the data to calibrate the model, not to represent the physical phenomena correctly. The MKS model should be seeded with evenly distributed values over the grid to capture as many differences as possible between values. You may be right that when we are testing the model that we then need to use normal distributions, but using enough uniform distributions to calibrate the model should account for any particular initial condition.

It may also be possible to use normal distributions to calibrate the model but then we need to max and min at 1 and -1 and also vary the initial location of the distribution and the width of the distribution. That doesn't seem to make sense when a uniform distribution can account for all kinds of initial conditions.