Opportunities for CyberInfrastructure at the Cornell

Nanoscale Facility

Garnet Kin-Lic ChanDepartment of Chemistry and

Chemical BiologyCornell University

Who I amModeling at the Cornell Nanoscale FacilityOpportunities for CyberInfrastructure for

Nanoscale modeling Software Infrastructure

Random ideas about the Web

Who am I? Ab-initio Quantum Chemistry /

Electronic Structure theory

Method development for large systems Renormalization Group methods for multi-

scale phenomena Ab-initio DMRG, Canonical Transformation

Theory Correlated materials problems

Conjugated Polymers, Surface chemistry Quantum transport in nanoscale

structures

User of CCMR and CNF modeling facilities

Conjugated Polymers

Low-temperatureKondo regimein Metal complexes

Experimental facilities are not data-intensive Primary use of computing/cyberinfrastructure is

materials modeling Secondary use is for social aspects e.g.

education / communication / documentation

The Cornell Center for Materials Research and the Cornell Nanoscale Facility

Cornell Nanoscale Modeling FacilityThe National Nanostructure Infrastructure Network (http://www.nnin.org)

Computational mission of CNF: Develop modeling resources that complement and expand on the current experimental capabilities.

Development of new computing clusters.

Acquisition of commercial software packages.

Construction of new codes to address user

needs.

Expansion and distribution of localized programs to user network.

Web based access – truly remote research

Main Computation NodesCornellHarvardStanford

University of Texas, Austin University of Michigan

Courtesy Derek Stewart

Computing Power Cornell Nanoscale Facility

48 node dual processor Xeon (3.06 GHz) cluster 16 AMD 64 bit Opteron workstations

Harvard University 48 node dual processor Xeon (3.06 GHz) cluster 4 4-way 32 GB Opterons from Sun Microsystems (coming soon!)

University of Texas, Austin Access to 600 processor Xeon cluster 224 (1.3 GHz) Power4 processor cluster

Stanford and University of Michigan (resources coming soon!)

Computational Resources available across the country

Duffield Hall

A Platform for more than just computation…

Services to encourage collaboration and enhance existing tools.

Web based discussion groups that allow new users to learn from existing users.

A conduit for codes developed by localized groups to reach a larger audience (beta testing, optimizing, streamlining)

Creation of input file libraries for different programs

Develop modeling resources that complement and expand on the current experimental capabilities.

Services to encourage collaboration and enhance existing tools

1. Software infrastructure

2. Web-based initiatives

How can we use CI for the CNF’s mission?

Software infrastructure for nanoscale modeling

Multiscale Inhomogeneous

Neither periodic nor isolated

Multiple energy scales Electronic, vibrational, electromagnetic

Fundamental algorithms, but also enabling software infrastructure

Interoperable components

Interoperabilitiy Many levels of theory Each level has multiple algorithms

implemented in different packages Continuum models Force-field

Tinker, NAMD, Moldy etc. Kinetic Monte Carlo Density Functional Theory

ABINIT, SIESTA, PWSCF, DFT++ Ab-initio quantum chemistry

Gaussian, QCHEM, GAMESS, MOLPRO

Any successful multiscale method cannot adopt a monolithic approachbut must reuse components

Scientific issues: different choice of basis Computer science issues: e.g. conversion of formats

Where can we look?

Software industry CORBA, DCOM, SOAP, XML, RPC Standard interfaces e.g. BLAS, LAPACK

But - cultural challenges Scientific modeling companies are notoriously

competitive No motivation for academic scientists Deciding on an implementation is not enough: it

has to be implemented (and free).

Example 1: OpenBabel

http://openbabel.sourceforge.net/ “Open Babel is a community-driven scientific project including both

cross-platform programs and a developer library designed to support molecular modeling, chemistry, and many related areas, including interconversion of file formats and data.”

Over 70 different formats

babel [OPTIONS] [-i input-type] infile [-o output-type] outfile

But only for structural and geometric information For true multiscale modeling, require more sophisticated

conversion e.g. wavefunctions, orbitals, density matrices, potentials

Data-intensive e.g. many-particle wavefunctions

Example 2: EMSL Gaussian Basis Set Order Form (PNNL)

http://www.emsl.pnl.gov/forms/basisform.html

Standardised repository for basis sets for quantum chemistry calculations

Produces outputs for essentially every QC package

Used by 100% of quantum chemists

Issues

Where is the boundary between CyberInfrastructure and traditional materials modeling algorithm development?

The Web

Web-based frontends (“enhancing existing tools”) e.g. WebMO frontend to Gaussian / GAMESS http://www.webmo.net/ Users draw structures in browser, submit to

remote server, used successfully e.g. in Cornell CHEM765

Essentially Web-mail as opposed to mail-client Convenient, but really necessary?

The thing of the moment: Flickr, MySpace, Facebook

Set up Facebook for Nanoscale scientists “Experts” database But cultural barriers

Virtual conferences? Useful – but NSF funding?

Social networking (“collaboration”)

Wikis (“collaboration”)

Accessible web-pages, anyone can edit simply by clicking and typing

Used in my lab for announcements, electronic workbooks, research notes, paper repository

Highly recommended!Several people have made suggestions

already e.g. for material specific wikis

Issues: Global vs specific

To what extent should we create specialized sites with a restricted user community?

To what extent should we control the content? Many of the revolutionary aspects of

cyberinfrastructure revolve around a truly global community E.g. Google – one place to search, do not search by

categories as in early engines Wikipedia, one stop place for all information Should we make an official contribution to these sites

rather than set up our own?