InTheLoop | 09.22.2014
CS Staff Contribute to Successful National Lab Day in Washington
Seventeen DOE national labs pooled their expertise and accomplishments to stage National Lab Day in the Rayburn Senate Office Building on Tuesday, Sept. 16. The event focused on labs-as-a-system, rather than highlighting individual labs, and drew 15 members of Congress and more than 100 staff members
The HPC display organization was co-led by David Skinner of NERSC, along with Kim Cupps of LLNL and Becky Verastegui of ORNL. Jon Bashor of Computing Sciences helped collect and organize scientific visualizations and write descriptions for an interactive display of accomplishment in areas such as basic science, clean energy, transportation, health and national security.
CRD researchers Burlen Loring, Zarija Lukic, David Trebotich, Daniela Ushizima and members of the Center for Computational Sciences and Engineering contributed their work to the display, which also included the Materials Project hosted by NERSC. »Read more.
John Shalf Gives Talk at San Francisco High Performance Computing Meetup
In his role as NERSC’s chief technology officer, John Shalf gave a talk on “Converging Interconnect Requirements for HPC and Warehouse Scale Computing" at San Francisco High Performance Computing Meetup. The Sept. 17 meeting was held at GeekdomSF in downtown San Francisco.
The group, which describes itself at “High Performance People getting together to discuss topics in High Performance Computing (HPC),” is interested in topics ranging from using the latest web and cloud to make HPC more accessible to efficient scaling simulation workflows beyond 50,000 cores. Members are mostly from industry, including Nvidia, Yahoo!, Cisco, Facebook and various start ups.
In his talk, Shalf discussed the emerging trends towards convergence of cloud and HPC interconnect requirements due to both market forces and increasingly aligned performance requirements. »Read more
This Week's CS Seminars
Modeling and Simulation of Multiphysical Processes for Advanced Manufacturing of New Multifunctional Materials
Wednesday, Sept. 24th, 3:30 – 4:30 p.m., 939 Evans Hall, UC Berkeley
Tarek Zohdi, University of California, Berkeley
Recently, several manufacturing applications have arisen that involve the dynamic of response of particulate systems in the presence of strong electromagnetic fields. In many cases, there is significant multifield coupling, which requires methods that can capture the unique and essential physics of these systems. In this presentation, Zohdi will discuss the modeling and simulation of three such applications: Electromagnetic composites, charged particulate jet sprays and droplets, and rapid, energy-efficient, sintering of materials comprised of heterogeneous powders. He will also touch upon simulation of selective laser-assisted sintering and ablation. »More.
Matrix Computations and Scientific Computing Seminar (UCB): On Self-convexity
Wednesday, Sept. 24, 12:10 – 1 p.m., 380 Soda Hall, UC Berkeley
Gregorio Malajovich, Universidade Federal do Rio de Janeiro
Numerical analysis has a rich story of interplay between two definitions of the condition number: analytic and geometric. The analytic condition number is the norm of the derivative of the solution with respect to the coefficients. The geometric condition number is the reciprocal distance to the set of ill-posed instances. For instance, the Eckart-Young theorem says that for affine equation solving or least squares, those definitions are the same. Self-convexity is an attempt to encompass some of the convex geometric properties of condition numbers in general. It was motivated by an attack to Smale's 17-th problem, which deals with systems of polynomial equations. However, the results obtained so far are still limited to affine systems. The objective of this talk is to present the basic ideas, results and motivations. Much of this subject is still wide open, so there may be more questions than theorems.
Julia: A Programming Language for Technical Computing
Thursday, Sept. 25, 2014, 10 – 11:00 a.m., Bldg. 50B, Room 4205
Arch D. Robison, Intel Corp.
This seminar is an introduction to the Julia language, a new programming language for technical computing that combines high-level convenience with high performance in a single language. On the convenience side, Julia features a concise syntax, dynamic typing with optional type declarations, exception handling, and can be used interactively. On the performance side, Julia exploits type-inference, code specialization, and a just-in-time compiler. Furthermore, it employs well known Fortran/C libraries for linear algebra, signal processing, random-number generation, and string processing, and can easily latch onto other Fortran/C libraries. Julia directly supports metaprogramming, which has been powerful, but often awkward, in other languages. Julia supports distributed-memory parallelism, and support for shared-memory parallelism is being explored. Julia is open source with 275 contributors as of Sept. 4.
Variational framework for spectral approximations in Kohn-Sham density functional theory and an all-electron calculation using Spectral Gauss Quadratures
Thursday, Sept. 25, 11 a.m. – 12 noon, Bldg. 50B, Room 4205
Xin (Cindy) Wang, Mechanical Engineering, California Institute of Technology
Kohn-Sham density functional theory (KSDFT) is widely used to study the electronic structure of materials. These nonlinear equations are typically solved numerically by fixed point iteration (self-consistent field method) that yields an eigenvalue problem associated with the linearized Hamiltonian. Further, approximate methods that exploit spectral theory – the density matrix expansion methods – are increasingly used to study the linearized problem because of computational efficiency. The talk will consists of two sections. The first section focuses on understanding the relation between spectral approximations of the linearized problem to the original nonlinear problem. We reformulate KSDFT as a nested variational problem that enables spectral approximations. By introducing a new approxi- mation – the binning approximation – we show convergence with respect to both these binning approximation and with spatial discretization. The second section focuses on performing an all-electron KSDFT calculation using a type of spectral approximation, spectral-Gauss quadrature method.
Taking Ownership of Information from Sensors
Thursday, Sept. 25, 12 noon – 1 p.m., Bldg. 50F, Room 1647
Dani Ushizima, Visualization Group, Lawrence Berkeley National Laboratory
We invite you to take ownership of information gathered by sensors. This talk will describe some of the speaker's work on applying algorithms using R-statistical framework to perform data mining in order to get models ready for prediction. There is an enormous potential for analytics that handles sensor data, such as those collected by phone apps and other data logging systems. We will illustrate potential applications of this investigation, including processing information from BIDS identification card usage to data facility log files in order to better understand flows and improve resource utilization.