Mod:scan

From ChemWiki
Revision as of 17:23, 25 January 2009 by Mja06 (Talk | contribs) (Submitting calculations to the Departmental SCAN Cluster)

Jump to: navigation, search

Submitting calculations to the Departmental SCAN Cluster

Export from Ghemical
The Chemistry department runs a SCAN (Supercomputer at Night) system, whereby teaching computers which would otherwise only idle in the middle of the night, can be used to run more time consuming calculations than is possible interactively on a single computer whilst sitting in front of it.

One far more reliable and quantitative way of modelling a molecule is to subject it to quantum mechanical modelling using Density Functional theory. In practice, this is implemented here using a program called Gaussian 03. The procedure to submit such a job is as follows:

Creating an Input file

  • After you have optimised your sketched molecule using ChemBio3D or Gaussview, as described previously, you will have a Gaussian input file saved in your H: drive by default.
  • Typical Gaussian input
    The file will have to be edited before it can be submitted. You can do this either with Gaussview as the program, but a much simpler method is to open the file (pentahelicene.gjf in this example) using eg the Windows Wordpad editor. This is invoked simply by double clicking on the file. Remove any existing lines starting with % or # and replace them with one of the following single lines (the second example also results in the vibrational frequences and from these the entropy being computed, and hence the zero-point and free-energy corrected value, ΔG). This latter option will take significantly longer however.

# B3LYP/6-31G(d) opt
or
# B3LYP/6-31G(d) opt freq
to produce a file that looks like the one shown on the right.

  • For a molecule the size of e.g. pentahelicene, the calculation will take about 4-5 hours overnite. If for some reason, your molecule is taking longer, you can always reduce the size of the basis set to e.g. B3LYP/3-21G*, or submit the job on a Friday, when it will have the entire weekend available to it. If you want greater accuracy (but for longer computing time), try e.g. # B3LYP/cc-pVTZ opt freq.

Submitting the Input file

  • Create a new job
    You will have to login as yourself. You can submit as many jobs as you wish through this mechanism, but you must prepare an input file for each first (.gjf if you want to run Gaussian).
  • Create a project
    Select a pool
    After you are logged in you should organise your jobs by project. Create a suitable new project, then select New job. The next decision is to decide which of two possible Compchem queues to use:
    1. Compchem Lab 2 runs during the period 23.00-07.30 overnight, during which time 36 computers will be processing jobs (a concurrency of 36). If a job is not completed during the overnight period, it will be scheduled to run again (from the beginning) the next night, or all weekend if scheduled on Friday. For this reason, you should only schedule jobs that can complete in an 8 hour window, or a 48 hour window on weekends.
    2. Compchem Lab 1 runs continuously during the day and night, but only has a concurrency of 8. Jobs run at about the same speed on either system.
  • Next, select an Application (currently only Gaussian and Amber PMEMD), then the Project you have just created, and press continue.
  • Upload your input file
    You now have to find the Gaussian input file, as prepared above. You should Browse to drive H: to find this file. Add a description which will help you identify the job.
  • The Chemistry Condor Pool
    The job will be added to your list of jobs, andyou can view its status, which is either running if there is a vacant slot in the queue you submitted to, or pending if there is not. Unfortunately, you cannot find out how many jobs are in front of a pending job.
  • Viewing the outputs
    When the job has completed, click on the Job List link. This will show all available outputs. Download the program Log file (this will help you chart whether the calculation was successfull) or the Gaussian Formatted Checkpoint file onto the desktop of the computer you are using, and the file should open up Gaussview, where the molecule can be viewed and checked. You can use the latter file to e.g. plot molecular orbitals for the molecule, view vibrational modes, etc. Full details of these procedures are described in the Gaussview manuals.

Archiving the output into a digital repository

Depositing an entry in DSpace
A very recent innovation is the Institutional digital repository, a resource for permanently archiving calculations, spectra and crystal structures. You can get a flavour of this by archiving your own calculation in the SPECTRa digital repository. To the right of the Portal display is a link termed Publish. If you click on this, and the calculation is actually in a state to be published (it may for example have failed for some reason) then appropriate metadata for the calculation is collected, and the collection deposited into the repository. From here, it can be retrieved in future.

Retaining the Calculations

Do not delete any completed jobs from the submission pages until your report has been graded. You may be asked to show individual jobs (via the input, or outputs) if for example the calculation has not succeeded in the manner you expected and you would like feed back on this or any other errors.