Page 1 of 1
Exactly same results from separate calculations
Posted: Fri Dec 11, 2020 1:41 pm
by Yongliang Ou
Dear all,
In order to collect data statistically, I set the following parameters in the driving file under the tag "Phase interaction data".
Code: Select all
# 'DeltaG' options: default
# avg ...[] max ...[J/cm^3] smooth ...[Deg] noise ...[J/cm^3] offset ...[J/cm^3]
avg 0.90 max 200.00 noise 1.00
Then I keep all parameters the same and do the calculations many times respectively. I thought I would get dendrites at different positions and some observables would fluctuate. Surprisingly, the calculation results are all identical: all output files are exactly the same. I am wondering if it is possible to get thermodynamically random results? How should I do it?
Thanks in advance and best regards,
Yongliang Ou
Re: Exactly same results from separate calculations
Posted: Fri Dec 11, 2020 5:10 pm
by janin
The amount of noise you specified in input file is locally added to the driving force. You can monitor the effect on the driving force in the *.driv output file. The values are given in units of J/cm3, if you devide them by the local entropy of fusion ( dSf ≈ 1 J/(cm3 K) you will get the extra undercooling, i.e. the amount of your noise is approximately 1 K. Noise may be helpful to trigger the generation of side-arms in cases where the standard numerical noise is not sufficient. However you have to consider that the locally applied noise is simply averaged out over the diffuse interface if it is too small. On the other hand, the noise should not be too high, as it should not change the physics of the simulation in a significant way, especially not the arm spacing, which is not random, but determined by the local diffusion field. Hence, this noise is not the right parameter to obtain statistic variation. To obtain dendrites at different positions you may add noise to your nucleation condition instead.
By the way, I would recommend always to use an average factor of 0.5 in combination with redistribution control (atc mob_corr). This is the value with was assumed in the Micress-model specific derivitation of the thin interface correction. The increased value of 0.9 (which you sepcified) is only required in case of higly numerically instable, i.e. poorly resolved computations.