Quantum Chemistry at MCSR - PowerPoint PPT Presentation

1 / 21
About This Presentation
Title:

Quantum Chemistry at MCSR

Description:

Choose theoretical methods that provide a good balance of accuracy ... Exotic or cutting edge methods tend to be much more limited. Multi-reference, post-HF ... – PowerPoint PPT presentation

Number of Views:58
Avg rating:3.0/5.0
Slides: 22
Provided by: BrianH110
Category:

less

Transcript and Presenter's Notes

Title: Quantum Chemistry at MCSR


1
Quantum Chemistry at MCSR
  • Brian W. Hopkins
  • Mississippi Center for Supercomputing Research
  • 12 February 2009

2
What Were Doing Here
  • Talk about quantum chemistry procedures.
  • Talk about available quantum programs.
  • Discuss user needs and program options.
  • Emphasis will be on modernization.

3
QChem Anatomy of A Project
  • Choose a chemical problem youre interested in.
  • Construct a methodology for examining this
    problem.
  • Build a (crude) model system.
  • Choose theoretical methods that provide a good
    balance of accuracy and efficiency.
  • Select a program suite that can perform the
    necessary calculations.
  • Build the program on an appropriate computer.
  • Organize and run the jobs efficiently.
  • Manage data.
  • Analyze results.
  • Write papers.

4
Program Selection Basis
  • Most every program has two methods for using GTO
    basis sets
  • A standard library. ? be careful!
  • A parser. ? https//bse.pnl.gov/bse/portal
  • Sticking points
  • Plane waves.
  • STOs.
  • Pseudopotentials.

5
Program Selection Electronic Structure Methods
  • Most QC programs have base functionality
  • Single-reference HF, DFT, MP, CC and CI
  • Multi-reference HF (eg. MCSCF or CASSCF)
  • Exotic or cutting edge methods tend to be much
    more limited
  • Multi-reference, post-HF
  • Excited-state theories (EOM, TD)

6
Program Selection Other Methodological Concerns
  • 1st and 2nd Derivatives
  • Periodic boundary conditions
  • Scanning or sampling techniques
  • c.

7
Program Selection Performance
  • Generality
  • Some codes have had most development attention
    focused on adding features
  • Performance
  • Some codes have had most development attention
    focused on improving performance

8
Program Selection Ease of Use
  • Ease of use is important!
  • Being concerned about this doesnt make you lazy.
  • We must make rational decisions about the amount
    of time that ought to be invested in learning to
    work a program.
  • This time investment needs to be balanced against
    the expected benefit of using the new program.

9
Program Selection License
  • QC programs are big producing them is very labor
    intensive.
  • The authors of these programs have rights that
    must be respected.
  • Authors have the right to distribute programs as
    they see fit.
  • DO NOT install any program on a UM or MCSR
    computer in violation of its license!

10
Thats A Lot! Where Do I Start?
  • Ask your coworkers.
  • Consult online program manuals.
  • Read papers, and pay attention to the
    Computational Details or similar sections.

http//en.wikipedia.org/wiki/Quantum_chemistry_com
puter_programs
11
A Note On Program Promiscuity
  • Can you mix results from different programs?
  • Generally yes.
  • Sometimes no.
  • Things to watch out for
  • DFT flavors integration grids
  • Basis set details
  • Various program defaults (FC, c.)
  • Symmetry tolerances

12
So What Have We Got? 1.
  • Gaussian 03
  • Lots of different capabilities
  • Very easy to use
  • Limited scalability
  • Excels with optimization and surface-scanning
    techniques
  • Complete site license for Ole Miss
  • Fully functional on all MCSR systems
  • Automated PBS interface g03sub

13
So What Have We Got? 2.
  • NWChem
  • Lots of different capabilities
  • Very easy to use
  • Good scalability
  • Poor serial efficiency for some methods
  • Free academic license
  • Fully functional on sweetgum, redwood, and
    mimosa.
  • Automated PBS interface nwchemsub.

14
So What Have We Got? 3.
  • GAMESS
  • Lots of different capabilities
  • Fairly easy to use
  • Some scalability
  • Free academic license
  • Functioning on sequoia (single-node only)
  • No automated PBS interface

15
So What Have We Got? 4.
  • MOLPRO
  • Excels for MR and correlated treatments.
  • Screaming fast serial code for many methods.
  • Difficult to use
  • Some scalability
  • Commercial license for MCSR machines
  • Functioning on sequoia (single-node only)
  • No automated PBS interface

16
So What Have We Got? 5.
  • MPQC
  • Does only a few things.
  • Difficult to use
  • Screaming fast.
  • Extreme scalability
  • Free academic license
  • Functioning on sequoia (single-node only)
  • No automated PBS interface

17
So What Have We Got? 6.
  • We will support most any application you can put
    in our hands.
  • Stuff thats free
  • PSI, ACES, CPMD, CP2K, c.
  • Stuff that you buy a license for
  • Jaguar, TURBOMOL, Qchem, c.
  • We will probably not buy licenses.
  • We will not violate any software license.

18
So Ive Chosen a Program. What Now?
  • Choose a system
  • Build it!
  • Most scientific codes are distributed as source
    and have to be compiled.
  • We are almost always willing to build stuff for
    you.
  • Theres a separate seminar on building comp.
    chemistry apps scheduled for March 4th.
  • We have special queue facilities for building and
    testing new apps.

http//www.mcsr.olemiss.edu/educationsubpage.php?p
agenamemarcamp09.inc
19
After The Build
  • Develop PBS scripts for running jobs.
  • Please dont run I/O intensive QC jobs in /ptmp!
  • We have examples and can help!
  • Run jobs
  • If queues bog down, dont be afraid to ask for
    special help!

20
Data Management
  • Not usually an issue with QC work.
  • Exceptions
  • During-job scratch
  • Relocate to scratch filesystem
  • Checkpoint files
  • Keep these zipped when not in use, delete them
    ASAP
  • Core files
  • limit coredumpsize 1k

21
Data Analysis
  • Automated analysis is best!
  • All of our systems have tools to facilitate
    automated data analysis
  • C and Fortran compilers
  • Ask us about MPI!
  • Perl and Python interpreters
  • Linux utilities (eg, awk)
  • We can help build, debug, and run your analysis
    code as needed.
Write a Comment
User Comments (0)
About PowerShow.com