accuracy

ripening phenomena, dislocations, grainboundary topology
Post Reply
zhubq
Posts: 84
Joined: Mon Jun 22, 2009 7:33 pm

accuracy

Post by zhubq » Wed Jan 06, 2010 9:58 pm

Hi,
It seems if the driving force is too high compared with interface energy, both the time step and cell size should be decreased. If only diminish time step but keep cell size as before, the accuracy decreases.
For example, when mobility=1, driv_force=1, velocity=0.9 (90% accurate); mobility=1,driv_force=100, velocity=80 (80% accurate). This will lead to nonlinear relationship between velocity and driving force in a simulation where driving force varies greatly.
How to deal with this problem, decrease time step, cell size and/or increase interface thickness? it is noted that the change of the cell size and thickness will affect the domain size and upper bound of curvature.

Bernd
Posts: 1155
Joined: Mon Jun 23, 2008 9:29 pm

Re: accuracy

Post by Bernd » Wed Jan 06, 2010 11:22 pm

Dear zhubq,

high driving forces lead to a deformation of the interface profile. A higher resolution (i.e. smaller deltaX) will increase the interface stabilisation term and such reduce errors coming from that.

A badly resolved interface profile (small interface thickness in term of number of cells) will reduce accuracy and lead to a smaller velocity (probably the reason for only 90% of the expected velocity in case of the smaller driving force).
Which interface thickness did you use?

Probably you need to do both! Inceasing resolution as well as the interface thickness should help, but will considerably increase the simulation time, if a given microstructure with a given size has to be properly resolved. I think, there is no other way than to try to find out the best combination of the two parameters!

Maybe, a future MICRESS version will be able to separate curvature from stabilisation, and thus allow increasing stabilisation without changing the interfacial energy...

Please tell us about your findings!

Bernd

zhubq
Posts: 84
Joined: Mon Jun 22, 2009 7:33 pm

Re: accuracy

Post by zhubq » Thu Jan 07, 2010 7:00 pm

Hi, Bernd.
I usually use 5*del_x as the interface thickness. del_x=0.1um. It is fine when i use normal interface energy value. I met unstable problem just when i use 1/100 of the interface energy to simulate recrystallization problem. But I am thinking that the driving force of transformation is quite high compared with recrystallization, i.e. the ratio of driving force to interface energy is quite high which leads to interface instability. So I do not know the del_x and interface thickness people usually use in diffusion coupled problem.

Post Reply