InTheLoop | 12.08.2014
2014 Bay Area Scientific Computing Day
The 2014 Bay Area Scientific Computing Day (BASCD) will be hosted by Stanford University on Saturday, December 13, 2014.
BASCD is an annual one-day meeting focused on fostering interactions and collaborations between researchers in the fields of scientific computing and computational science and engineering from the San Francisco Bay Area. Berkeley Lab and Stanford University take turns in organizing the event.
Those interested in scientific computing and computational science, particularly junior researchers, are encouraged to attend BASCD. »Learn more.
New Journal Shares Advances in Statistics, Mathematics of Atmospheric and Ocean Research
A new journal—Advances in Statistical Climatology, Meteorology and Oceanography (ASCMO)—gives statisticians and researchers specializing in the atmospheric and ocean sciences an outlet to publish the details of their statistical and mathematical developments, which will effectively lead to improved models and methods for these fields.
Michael Wehner, a climate scientist at the Lawrence Berkeley National Laboratory (Berkeley Lab), and Jennifer Hoeting, a professor of statistics at Colorado State University, are both founders and executive editors of the new publication. »Read more.
ASCR Leadership Computing Challenge Applications Due Feb. 3
Open to scientists from the research community in industry, academia, and national laboratories, the ALCC program allocates up to 30 percent of the computational resources at ASCR’s supercomputing facilities to that advance the DOE mission and further the goals of DOE program offices. ALCC facilities include NERSC at Lawrence Berkeley National Laboratory and the Leadership Computing Facilities at Argonne and Oak Ridge National Laboratories. Applications for the 2015 ALCC are due by 11:59 PM EST Tuesday February 3rd, 2015. Proposals undergo peer review and awards will be announced in May 2015. Allocations begin July 01, 2015. »Learn More.
New BIDS Web Site Launches
The Berkeley Institute for Data Science (BIDS) has launched a new web site where interested researchers and community members can to learn more about the people, projects, events, and collaborative opportunities at BIDS.
Site visitors can
- Learn more about BIDS fellows, senior fellows, and staff;
- Browse research projects our fellows and senior fellows are working on;
- Book office hours with fellows to discuss data science research and problems;
- View and subscribe to events calendar and news;
- Check out some of the tools and technologies fellows and senior fellows use;
- And more.
Questions or comments about our the new site (or BIDS in general) can be directed to firstname.lastname@example.org. »Visit http://bids.berkeley.edu.
This Week's CS Seminars
Hybrid Programming with MPI
Monday, Dec. 8, 2–3 p.m., NERSC OSF 943, Room 238
Dr. Pavan Balaji, Argonne National Laboratory
MPI has long been considered the de facto standard for parallel programming. Two of the primary strengths of MPI are its continuously evolving nature that allows it to absorb and incorporate the best practices in parallel computing in a standard and portable form, and its ability to interoperate with other programming models such as OpenMP or OpenACC. With the growing prominence of multicore and manycore architectures, MPI+X programming, where MPI is used together with a second programming model, is gaining a lot of interest. There has been a lot of ongoing research on improving both MPI and "X" (e.g., OpenMP). However, little attention has been paid to the "+" in "MPI+X" programming, i.e., whether these models truly interact well with each other or if changes are needed. In this talk, I'll concentrate on this missing piece and describe some shortcomings that need to be addressed.
Tensor Hypercontraction for Electronic Structure Theory
Wednesday, Dec. 10, 3:30–4:30 p.m., 939 Evans Hall, UC Berkeley
Todd Martinez, Stanford University
The solution of the electronic Schrodinger equation, often known as quantum chemistry, can be computationally challenging. It has often been thought that the problem was of exponential complexity (with respect to the size of the molecule). A number of methods have been developed to solve the problem approximately in polynomial time but the simplest wavefunction-based methods including electron correlation already require computational effort scaling with the fifth power of the molecular size. We discuss the origin of this scaling behavior and give an overview of the types of manipulations which are required by the standard methods. Then we will introduce the tensor hypercontraction method as applied to the electron repulsion integrals. Tensor hypercontraction is a tensor decomposition technique and leads to scaling reductions of up to two powers of the molecular size for widely used methods. We show that the numerical accuracy of tensor hypercontraction is well within chemical accuracy (and indeed is more accurate than many other commonly used approximations). Relationships between tensor hypercontraction and numerical quadrature are explored. We further explain how the tensor hypercontraction idea can be applied not only to the Coulomb operator but also to the molecular wavefunction.