The final exercise for today is to demonstrate an ab-initio MD run for a liquid metal. We aim to perform around 2000 timesteps on a system of 32 or 64 atoms which will be sufficient to generate a RDF. Clearly such a run will not finish this afternoon so the aim is to start the job running overnight and to analyse the results tomorrow morning. Clearly you will need to be careful to ensure your job is running correctly before you leave!
I have prepared two possible systems, and you may choose between them. The first is a 32-atom system of Al with a 2x2x2 k-point grid sampling and is a simple metal. The other is silicon, whose metallic state is a liquid. This sample has 64 atoms and uses just the gamma point.
You should first look at the MD tutorial at the CASTEP website
to understand how to set up an ab-initio MD simulation in CASTEP. Unpack the exercise files
tar xvfz /home/ccpss1/CCP5SS/Exercise_3_MD.tar.gz
cd Exercise_3_MD
This is a more open ended exercise, so I would like you to vary the condistions of the simulations in some small but controlled way, and submit a job to run overnight. The choice of exact simulation point is up to you. Please feel free to discuss the viability or impact any planned changes!
Before you submit the MD run it is essential to check that the input
files are correct. This is because your job may not be started by
the queueing system until later, when you will not be around to correct
any errors. You may use the command
castep Si64-nvt -dryrunwhich will perform an initial check of the input files. Then it makes sense to submit a short job of 10 timesteps to check that your input files are correct before submitting the main job run. You will need to change the run-time reqtested for this test job to 5 minutes or so.
Exercise: Examine the output file for the first few MD steps. Plot a graph of the number of SCF cycles taken to converge for each MD step. Can the method of Born-Oppenheimer ab-initio MD help explan what you see?
You should now be in a position to estimate how many timesteps your overnight run should be able to achieve in the time and with the number of processors allotted. This will enable you to set up the full CASTEP and job script files.
The "mdtep" program referred to in the website is available on your School VM and on the Dell. and also on the pc system, which you may use to calculate the RDFs and MSDs and a few other autocorrelation functions. This will be the work of tomorrow morning's practical.
Xcrysden can be used to display an animation of the MD run in the .md file, after conversion using the geom2xsf program. Alternatively gdis may be used to produce an animation from the .castep file.