LASSP

[ Up ] [ Materials Science ] [ Solving QCD ] [ Teaching Physics ] [ LASSP ] [ Chemistry ] [ Physics Training ] [ Mathematics ]

LASSP: Condensed-Matter Physics

The Laboratory of Atomic and Solid-State Physics (LASSP) proposes to stretch the computational limits of the hardware provided by Intel and the networking and software environment provided by our colleagues within the program. We will develop a usable software working environment for science and engineering research for Windows NT, using commercial software when available and developing our own tools when necessary. We will provide Web information about the recommended packages and Web distribution of those tools we develop. Scientific progress in diverse fields will be supported, including quantum Monte Carlo, realistic simulations of high-energy MBE growth, electromigration-induced void failure of aluminum interconnects, exotic new phases of matter, large-scale hydrodynamics, and universal features of noise in magnetic tapes.

LASSP has played a leadership role in scientific computing. Supported by a substantial $3M equipment grant from IBM in 1990, we have developed LASSPTools, a set of Unix utilities for numerical analysis and graphics, which are freely available for a variety of platforms on the Web. Seven years later, these tools are becoming widely used in science and engineering departments.

In the next decade, advances in communications and processor speeds will (again) transform scientific computation. Our making effective use of the massive computational power promised in the next years will demand much higher bandwidth connections, much higher performance, and much more data storage capacity than has been envisioned in the past. We intend to implement now the computational environment that will be widespread in the coming ten years. To do this, we plan to:

Create a network of NT workstations conveniently configured for use by individuals, as a local cluster, and as part of the Scalable Distributed Computation Resource;

Stress these workstations to their limits with various supercomputer-scale simulations:

	floating-point molecular-dynamics simulations which take months but fit in cache
	bit-intensive simulations demanding random access to RAM up to gigabytes in size, but which nonetheless execute in a few hours or days
	large-scale floating-bottlenecked hydrodynamic and electronic-structure calculations, some of which can be coarsely distributed and others of which are feasible only in an SMP environment

Locate and develop necessary tools for scientific productivity, and make them available on the Web.

As we did with LASSPTools, we will develop sophisticated numerical simulation software for understanding a wide variety of materials and physical phenomena. As an example, in an avalanche in Sethna’s simulation of Barkhausen noise in magnetic storage devices (magnetic tapes), simulations of up to one billion domains (1000³) have been possible using clever algorithms. It would be invaluable to us to have significantly more memory: 2 GB would allow us to run systems of size 2500³. Sethna is also modeling electromigration void failure in aluminum interconnects, presumably of more direct interest to Intel.

A recent cover of Nature featured an illustration of the electronic structure of Teter and Elser’s proposed negative-curvature graphite tube lattice analogue of the Buckyball. This illustration was generated using LASSPTools software. LASSPTools is most useful for active monitoring of large-scale simulations, when one needs rapid graphical animations of large amounts of data. The simultaneous demands of simulating and graphics place huge demands on the desktop workstations. We believe we can meet these demands with the new Intel Architecture machines we have requested.

Participants

James P. Sethna, Professor, Department of Physics

Web Links

http://www/tech2000/LASSPStatus.htm
http://www/tech2000/PhysicsStatus.htm
http://www.physics.cornell.edu/sethna/teaching/683/NTSimulations/NTSimulations.html

Last modified on: 07/30/99