BES Greenbook Presentation

1
BES Greenbook Presentation

• Theresa L. Windus
• Pacific Northwest National Laboratory

2
The Punch Line
Bigger
Better
More Realistic
3
Highlights of Research Supported by BES
4
Materials Science

• Types of systems (examples)
• Quantum nanostructures such as wires, dots,
  films, tubes and boxes properties vs. size
• Semiconductors and insulators band gaps, laser
  effects
• Metal clusters pressure effects, crack
  propagation
• Alloys such as with transition metals -
  impurities
• Surface phenomena Chemical Vapor Deposition
  (CVD), surface reconstruction, chemi- and
  physi-sorption
• Ceramics synthesis, defects, irradiation

5
Nanostructures

• Tailor materials at the nanoscale for desired
  structure/function properties
• Materials with enhanced physical, mechanical,
  optical, electrical, tribological, or catalytic
  properties
• Materials with the ability to self assemble, self
  repair, sense and respond to the environment
• Long-term, high-risk, interagency activity -- a
  unique instance of common scientific and
  technological frontiers
• Combines expertise in materials sciences,
  chemistry, physics, biology, engineering, and
  computation
• Expected are technological developments to rival
  the impact of the transistor

Richard Smalley http//cnst.rice.edu/pics.html
6
Materials Properties
G. Malcom Stocks http//www. ornl.gov/ORNLReview/v
30n3-4/develop.htm

• Superconductivity
• Band gaps
• Local and non-local Density Approximations

• Magnetic Properties
• Local Density Approximation
• O(N) Locally Selfconsistent Multiple Scattering
  (LSMS)

7
Shapes in Metal Alloys

• Sizes and shapes of precipitates is needed for
  understanding of strengthening mechanisms in
  metal alloys.
• Linear Expansion in Geometric Object, LEGO
  method basically a cluster expansion
• Scan many different alloys in a relatively quick
  time
• Based on first-principles calculations

Alex Zunger http//www.sst.nrel.gov/topics/new_mat
.html
8
Materials Defects

• Surface Reconstruction
• Chemisorption
• Physisorption
• Chemical Vapor Deposition
• STM modelling
• Corrosion

Alex Zunger http//www.sst.nrel.gov/research/defec
t.html
9
Types of Algorithms

• Density Functional based on local orbitals
  Local Density Approximation (LDA) or non-local
  (NLDA) methods
• Scale roughly as N3 or N4 where N is the number
  of local orbitals (lower for tight-binding
  methods)
• Bottlenecks for scalability tend to be either
  matrix inversion or eigenvalue problems
• CPU, memory and disk intensive

10
Types of Algorithms (II)

• Density Functional with Planewaves LDA and NLDA
• Approximately NeNaNb of k points where Ne is
  the number of electrons, Na is the number of
  atoms, and Nb is the number of basis functions
  (planewaves)
• Bottleneck for scalability is 3-D Fast Fourier
  Transform O(NeNb(logNb))
• CPU and memory intensive

11
Types of Algorithms (III)

• Molecular Dynamics, Monte Carlo, or
  Car-Parrinello
• Usually bound by the DFT method (with additional
  force calculation)
• Update usually causes additional problems for
  communication (especially latency)
• Memory intensive
• Lots of disk (TB)

12
Chemical Sciences

• Types of systems (examples)
• Quantum nanostructures such as wires, dots,
  films, tubes and boxes properties vs. size
• Flames kinetic effects, turbulence
• Heavy element systems thermodynamics, kinetics,
  excited state properties
• Excited states photochemistry, optical
  properties, and radiation

13
Flame Chemistry
Jackie Chen http//www.ca.sandia.gov/CRF/staff/Che
n.html

• Laminar and Turbulent flow
• Autoignition
• Diffusion Effects
• Structure and Propagation
• Chemical Reactions

14
Heavy Element Chemistry

• Waste Tank Remediation
• Relativistic Effects
• Highly Accurate Thermochemistry
• Excited State Properties
• Solvation Properties

15
Types of Algorithms

• Direct Numerical Simulation (DNS)
• How much physics and chemistry? Navier-Stokes,
  energy equations, velocity, time steps, amount of
  chemistry involved
• Also depends on the number of grid points (mesh
  size)
• Bottlenecks are communication and disk latency
  and bandwidth need TB of local disk

16
Types of Algorithms (II)

• Molecular Mechanics/Molecular Dynamics O(N)
• Bottlenecks for scalability are communication
  latency and disk I/O
• Load balancing
• Eigensolvers O(N3)
• Bottlenecks for scalability are communication
  bandwidth and latency
• Alternate algorithms (second order methods)
• Many body methods O(N5) to O(N!)
• CPU, memory and I/O intensive
• Bottlenecks for scalability are communication
  bandwidth and memory (depends on the algorithm)

17
Types of Algorithms (III)
18
Balanced System
Robert Harrison and Jeff Nichols Pacific
Northwest National Laboratory

• memoryM/F - the ratio of bytes of memory to
  flops/sec of computing
• diskM/F the ratio of bytes of disk to flops/sec
  of computing
• memoryB/F the ratio of bandwidth between memory
  and processor in bytes/sec to flops/sec of
  computing
• diskB/F the ratio of bandwidth between disk and
  processor in bytes/sec to flops/sec of computing
• netB/F the ratio of network bandwidth (with
  latency) in bytes/sec to flops/sec of computing

19
Geosciences
Garrison Sposito http//esd.lbl.gov/sposito 2.5
million-step Monte Carlo simulation shows that
Sodium ions (Na) in the interlayer of
montmorillonite are forming outer-sphere
complexes.

• Surface properties of clays and minerals
• Colloidal behavior
• Use of same methods as in materials sciences
• Transport processes in porous media
• Dependent on grid size and chemistry involved

20
Other Computational Needs

• Extra long batch queues
• Very low-latency communication system (switch)
• Large network bandwidth from NERSC to remote
  sites (especially National Labs)
• Large number of files
• Reliable C compilers
• Good parallel debuggers
• New algorithms
• Data visualization of very large data sets with
  synchronous data reduction