Monte Carlo simulation of atomic diffusion in solids

[baizq]

Deal All,

Can we simulate atomic diffusion in solids using Monte Carlo simulation?

I have noticed a couple of tutorials talking about AKMC simulation of  Pt diffusion on Pt surface

http://quantumwise.com/publications/tutorials/item/872-adaptive-kinetic-monte-carlo-simulation-of-pt-on-pt-100
http://quantumwise.com/publications/tutorials/item/879-adaptive-kinetic-monte-carlo-simulation-of-pt-island-ripening

But my picture is a bit different. My system is a bilayer XY:Z(1nm)/XY(8nm) stack, where Z is metallic dopant to the lower 1nm XY alloy region. The upper layer (8nm) of XY is free of Z at the beginning.  We observe that, the more we increase the temperature, the more of Z diffuse to the the upper XY region. We would like to understand the temperature effect.

Is the current implementation in ATK able to tackle this simulation? Thank you in advance.

baizq

[Ulrik G. Vej-Hansen]

Dear baizq

In principle, the AKMC implementation should be able to model the diffusion process. However, the XY layer is very thick, so depending on your model (DFT etc.), the time required might be impractically large.

[baizq]

Hi Ulrik, Thank you for your answers. We could reduce the thickness top XY layer to half in the model. Also there are EAM potentials available for the combination of X, Y and Z. Would this make it possible to simulate the diffusion?

Code

# -------------------------------------------------------------
# AKMC
# -------------------------------------------------------------
saddle_search_parameters = SaddleSearchParameters(max_neb_images=5)
markov_chain = MarkovChain(bulk_configuration, TotalEnergy(bulk_configuration).evaluate())
akmc = AdaptiveKineticMonteCarlo(markov_chain,
                                 kmc_temperature=300.0*Kelvin,
                                 md_temperature=2000*Kelvin,
                                 calculator=calculator,
                                 saddle_search_parameters=saddle_search_parameters,
                                 write_searches=True,
                                 constraints=constraints,
                                 confidence=0.95)
akmc.run(50)

I have not used the AKMC solver much. There are no much explanations of this script as well. Could you please also advise me which functions/parameters should I modify for such a calculation? baizq

[Ulrik G. Vej-Hansen]

Hi baizq

We have a more general introduction to the AKMC method in this tutorial: http://quantumwise.com/publications/tutorials/item/872-adaptive-kinetic-monte-carlo-simulation-of-pt-on-pt-100 Have you consulted this one? Please do so first, and then it will be easier to explain exactly what you would like to know.

We have two more tutorials you might also want to look at, if you have not already done so: http://quantumwise.com/publications/tutorials/item/879-adaptive-kinetic-monte-carlo-simulation-of-pt-island-ripening and docs.quantumwise.com/tutorials/akmc-empirical-sige.html

Regarding the practicality of modeling the system, you really need to do a test, and see how far it gets in a time that is reasonable for you. Using EAM potentials it should be possible, but it might still take a long time to get an adequate sampling.

[Anders Blom]

Having a classical potential is essential and I think it will be tractable with EAM. Just note that classical potentials generally are not so accurate away from equillibrium.

[Kim_W]

Dear Quantumwise staffs,
      I have calculated an Adaptive Kinetic Monte Carlo (AKMC) simulation to investigate the diffusion of a Pt adatom on a Pt(100) surface according to https://docs.quantumwise.com/tutorials/akmc_pt_on_pt_100_surface/akmc_pt_on_pt_100_surface.html. However, I can not get results. Can you help me?
     

[Ulrik G. Vej-Hansen]

I just tried running your script, and could not reproduce the problem. Did you use MPI or threading to run it? If yes, on how many cores?

[Anders Blom]

Did it run out of licenses? There is an error about "checkOut" which would be the license checkout. Maybe there is a separate error message in the error log or the terminal?

[Kim_W]

No errors were detected in the job debug Information. I noticed that there is a license error in the log script: 'Internal: 455 Feature: ATK Classical' when I run it with 1 processor. I want to know how to solve it? Both 24 and 1 processors are not worked. however, I can run other jobs with DFT functions.

[Anders Blom]

Right, so do you have an ATKClassical license feature? Or maybe just too few?

[Kim_W]

Yes, of course, I have a classical license. I can calculate total energy with classical function with 24 processors.

[Anders Blom]

Sure, but it appears the AKMC calculation checks out one license per saddle search, possibly, so you run out of the 10 or however many ATKClassical licenses you have. This may depend on the version of ATK, which one are you using?

[Kim_W]

2017.02
Actually, I have 1 ATK Classical license and 39 slaves. However, my computer only have 24 cores.
The problem also exist when I set as below.

[Henry Ian]

Hello everyone,

   In fact our team have the same problem with Wang Jin, but I found only one ATK-Classic license in our license manager. I want to know  if it needs more ATK-Classic licenses to run this calculation. Thank you in advance.

[Ulrik G. Vej-Hansen]

I can confirm that this is due to an undesired behavior of the license scheme when combining ATKClassical and AKMC, meaning that the 2017.2 implementation requires more licenses than it should. The problem has been fixed in the 2017.12 release.