[RMCProfile-users] 来自"朱敏" <minzhu at mail.sim.ac.cn>的邮件

Wed Jul 15 11:46:14 BST 2015

Dear all,

Thanks to Giles and Nick for responding to this query. This message sparked some interesting discussion and as a result I thought it would be a good idea to post a couple of general points about modelling glassy/amorphous materials with RMCProfile.

With crystalline materials the crystal structure provides an obvious starting point for modelling but clearly this is not the case for amorphous materials. 

The simplest way to generate a starting point is to distribute your atoms at random in your box and hope that the data and the constraints that you apply will drive the configuration to a physically and chemically reasonable structure. This purely random approach often means that your starting model will have atoms that are far too close together. You should be able to avoid this using the utility program 'dwbuild' which is included with RMCProfile wherein you specify a minimum closest approach for pairs of atoms. A random starting point will also cause problems with potentials, for the reasons already discussed by Giles and Nick. A lot of care needs to be taken with this method but it can provide useful information, especially if many different random starting models and their subsequent refined structures are taken and analysed as an ensemble.

Another approach is to start from a crystal structure (if one exists) for a material with the same/similar stoichiometry and allow RMCProfile to disorder it. The problem with this is that the final model is likely to be heavily biased towards the connectivity and structural features in the crystal. Nevertheless this method can produce interesting results, particularly if the crystal structure proves entirely unable to model the amorphous structure (implying that the connectivity is significantly different).

Probably the most robust method is to use a modelling technique such as molecular dynamics to produce a feasible starting model, and use RMCProfile to refine it. Obviously this is more labour intensive but my feeling is it avoids the bigger issues with the other methods.

Whatever route is used to generate a starting model the application of appropriate constraints is almost certainly going to be necessary to produce a good result, but this should be done with care to avoid simply producing a model that obeys the constraints and ignores the data!

I'd be interested to hear other opinions on this subject as I certainly cannot claim to be an expert, but hopefully this post provides some useful information to people who would like to model amorphous materials.

Best wishes,
Helen

-----Original Message-----
From: rmcprofile-users [mailto:rmcprofile-users-bounces at rmcprofile.org] On Behalf Of Nicholas Funnell
Sent: 13 July 2015 11:30
To: <rmcprofile-users at rmcprofile.org>
Subject: Re: [RMCProfile-users] 来自"朱敏" <minzhu at mail.sim.ac.cn>的邮件
Importance: High

Dear Min Zhu,

I think the short bonds may simply be due to a combination of not allowing the refinement to run long enough and the use of potentials on a highly-disordered structure. I can see in your starting model that you have many bond distances shorter than 1 A, prior to refinement. It could be that RMCProfile just needs a bit longer to push all of these atoms apart. 

I can also see that you are using potentials - what this will do is search for the atom pairs declared by each '>STRETCH' instruction within +/- 10% (in your case) of the distance you would like the pair to be restrained to. This will happen at the start of the RMC refinement and a .bonds file will be generated, containing all the bonded pairs - these will remain 'bonded' throughout the refinement. Similarly no atoms will become 'bonded' over the course of refinement - the .bonds file is a permanent connectivity list. Given that lots of your atom-atom distances are unphysical, it could be the case that the potentials are causing the refinement to get stuck - unable to move the atoms as freely as they need to. 

You'll want to revisit your current potentials, as you have some that will be fighting against each other - for example your second and third entries, concerning the Sb-Te distances. Search distances of 2.91 +/- 10% and 2.85 +/- 10% will identify duplicate atom pairs because of the overlap in distance.  

I would suggest removing the potentials (and deleting any existing .bonds files) and using RMCProfile to push the atoms apart more first (run for longer), then perhaps using potentials much later in a subsequent refinement to tidy up the structure a bit? Alternatively, you could use 'dwbuild' to generate your random starting model, using stricter closest approach constraints when placing the atoms initially.

I should note I have little experience modelling glassy materials, and these can be very tricky to deal with. I think Helen is going to post a follow-up message to this with some tips for dealing with amorphous materials in general.

Best wishes,
Nick

On 9 Jun 2015, at 10:20, 朱敏 <minzhu at mail.sim.ac.cn> wrote:

>  Hello, everyone!
> 
>       I use the "Random" command to build a random structure (Ge2Sb2Te5, Te 3000 atoms, Sb 1200 atoms, Ge 1200 atoms, atomic number density  0.0315 A). But I find a problem: After RMC simulation, the obtained structure contain bonds smaller than 1 A. But minimum-distances were set to 2.90 (Te-Te),  2.85 (Te-Sb), 2.60 (Te-Ge), 2.60 (Sb-Sb), 2.65 (Sb-Ge), 2.45 (Ge-Ge) Angstrom. And many atoms are not bonding with other atoms, as shown below (Ge red, Sb green, Te blue). What's more, the partial F(Q)'s are also not right.  The wrong bonds, such as Ge-Ge bonds and Ge-Sb bonds, are too much. 
> Can anybody help me to find the reasons? All the starting file has been attached.
> 
>          Thank you very much.
> 
>           Kind regards, 
>           Min Zhu
> 
> 
> 
> 
> 
> <Ge2Sb2Te5.png><RMC Simulation.png><Partical 
> F(Q).png><Ge2Sb2Te5.zip>______________________________________________
> _
> rmcprofile-users mailing list
> rmcprofile-users at rmcprofile.org
> http://lists.rmcprofile.org/mailman/listinfo/rmcprofile-users

----------------------------------------------------------------------
Nicholas Funnell
Post-Doctoral Research Associate (Goodwin Group) Inorganic Chemistry Laboratory University of Oxford
Phone: 01865 272698

_______________________________________________
rmcprofile-users mailing list
rmcprofile-users at rmcprofile.org
http://lists.rmcprofile.org/mailman/listinfo/rmcprofile-users