World’s Most Sensitive Dark Matter Detector Completes Search
The Large Underground Xenon (LUX) dark matter experiment, which operates beneath a mile of rock at the Sanford Underground Research Facility in the Black Hills of South Dakota, has completed its search for the missing matter of the universe. LUX’s sensitivity far exceeded the original expectations of the experiment, collaboration scientists said, but yielded no trace of a dark matter particle. LUX’s extreme sensitivity makes the team confident that if dark matter particles had interacted with the LUX’s xenon target, the detector would almost certainly have seen them. These new limits on dark matter detection will allow scientists to eliminate many potential models for dark matter particles, offering critical guidance for the next generation of dark matter experiments.
The 20-month run of LUX represents one of the largest exposures ever collected by a dark matter experiment, the researchers said. The rapid analysis of nearly a half-million gigabytes of data was made possible with the use Brown University’s Center for Computation and Visualization (CCV) and the advanced computer simulations at NERSC, a U.S. Department of Energy (DOE) Office of Science User Facility. Berkeley Lab is also the lead DOE laboratory for LUX operations.
“I am particularly pleased with the support LUX received from NERSC in processing these data,” said Kevin Lesko, group leader of Berkeley Lab’s Dark Matter group. “The Berkeley students, post-docs and visitors working on this analysis made extensive use of the NERSC for event scanning, calibration, Monte Carlo simulations and the data-blinding scheme.”
Morozov Named BIDS Data Science Fellow
The Berkeley Instititue for Data Sciences (BIDS) has named Dmitriy Morozov a 2016 Data Science Fellow. Morozov is a research scientist in the Computational Research Division's Data Analytics & Visualization Group. His work is concerned with geometric and topological data analysis, especially with the development of efficient algorithms and software in this field.
As a Data Science Fellow, Morozov joins a distinguished cohort, many from Berkeley Lab, representative of the world-class researchers leading the data science revolution within their disciplines.
Sign up for NERSC's Aug. 22 Data Day
This summer, NERSC will host a brand new data-centric event: Data Day. The main event will take place on August 22, followed by a half-day hackathon on August 23. The goal: to bring together researchers who use, or are interested in using, NERSC systems for data-intensive work.
On August 22, a full day of talks and tutorials will get you up and running with the latest data-focused tools for scientific computing. Training sessions on machine learning, Python, Spark, the Cori Burst Buffer, and science gateways will be complemented by presentations from scientists already taking advantage of these tools in their own research. A poster session will highlight work and work-in-progress that uses NERSC resources for experimental and observational science.
Stick around for the hack day, and you’ll get a chance to put your new knowledge to work creating a custom workflow. We’ll start off the half-day hackathon with an introduction to building workflows with Python and Spark, then provide local experts to help you get coding.
The entire event will be held on site in Wang Hall, NERSC’s new landmark building at Berkeley Lab. NERSC Data Day is free, but you have to register to secure your spot. Come and learn how to power up your data. Full details are available on the NERSC website.
2016 Python Bootcamp Registration is Open
BIDS is holding its 2016 Python Bootcamp on August 22 and 23 in Evans Hall on the campus of UC Berkeley. The two-day training runs from 8:30 a.m. to 5:00 p.m. each day and is designed to introduce the basics of the Python language to those already familiar with other computing languages (e.g., C, Java, FORTRAN, Lisp). The bootcamp is open to anyone within the UC Berkeley and LBNL community who qualify (that is, who have sufficient programming knowledge). To apply, visit the Python Bootcamp web site and fill out a prequalification form. If you are accepted, a small fee to offset the costs will be collected at the time of official registration.
This Week's CS Seminars
Monday, July 25
MSRI-LBNL 2016 Summer School
NWChem: Pushing the Scientific Envelope
2-3 p.m., Wang Hall (Bldg. 59), Room 3101
Bert de Jong, Lawrence Berkeley National Laboratory
NWChem is providing researchers with the software resources they need for scientific discovery and technological innovation in computational molecular sciences. In this lecture we will discuss how he extensive, and often unique, suite of capabilities available in NWChem, from high-accuracy and plane wave methods to molecular dynamics, can be coupled together to tackle large and complex scientific problems while including some of the complex dynamical behavior of nature. We'll highlight some of the difficult and complex problems that are appearing on the horizon.
Tuesday, July 26
Interdisciplinary Instrumentation Colloquium
Multimodal, multi scale tools for neuroscience: a call to arms for physicists and engineers
12 - 1 p.m., Bldg. 50 Auditorium
Dr. Peter Ledochowitsch, Allen Institute for Brain Science
Half a century ago, Hubel and Wiesel, the Rock stars of visual neuroscience, were recording from single neurons. Today, we routinely record the activity of hundreds of cells; striving to resolve hundreds of thousands of tiny, individual, neural voices, in the near future. This ‘Moore’s law of neuroscience’ not only unlocks overwhelming opportunities for biologists and clinicians, it is also a rallying cry for physicists, engineers, and computer scientists. Modern neuroscience presents a plethora of technological challenges: it demands microscopes that see deeper into the living brain and collect more photons, faster, ideally without generating heat. It covets ever-tinier, ever-denser arrays of electrometers that can be placed into the living brain without damaging tissue, and then last there forever, needing ever-less power, and, ideally, no wires. It yearns for algorithms and computing infrastructure that allow complex analyses of TBs of data performed ‘in real time’. And yet, any one sensing modality, be it using photons, ions, sound waves, or something much more exotic, throws but a shadow of the ‘true’ state of information contained in any ‘neural device under test’. The first half of this talk will focus on the development of a novel neural interface: a high-resolution, ultra-flexible, (optionally) fully transparent micro-electrocorticographic grid (μECoG) for subdural recordings, which was developed, fabricated, and tested at the UC Berkeley & UCSF Center for Neural Engineering and Prostheses (CNEP), and later commercialized by Cortera Neurotechnologies, Inc. The second half of my talk will touch on more recent work conducted in the Research Engineering Group a the Allen Institute for Brain Science: on the development of next generation ~1000 channel neural probes, on our work on optical recording and stimulation of neural activity with single-cell precision using multi-photon digital holography, and on the multi-modal integration of these two, which forms the core of my current research endeavors.
Wednesday, July 27
MSRI-LBNL 2016 Summer School
Beyond DFT: Predicting excited-state properties of materials using Green’s function formalisms
2-3 p.m., Wang Hall (Bldg. 59), Room 3101
Felipe H. da Jornada, University of California, Berkeley
Density-functional theory methods allow one to compute the ground-state energy of an interacting system exactly in terms of a functional of the ground-state electronic density. However, since DFT is a ground-state theory, the Kohn-Sham eigenvalues of the DFT cannot be directly interpreted as the quasiparticle excitation energies, even if the exact energy functional is known. In this talk, we discuss an alternative and rigorous approach to obtain excited-state properties of materials employing many-body perturbation theory techniques using Green’s function formalisms. We introduce the concept of single-particle Green’s function, which describes the propagation of an electron or a hole in the system, and which gives information on the excitation properties of systems. We discuss how these methods are connected to the commonly used GW approach to compute quasiparticle properties in materials. We also overview some computational bottlenecks associated with this method and algorithms proposed to circumvent these problems.
Thursday, July 28
Quantifying Performance Bottlenecks of Computational Kernels on Modern Multicore Processors Using the Execution-Cache-Memory Model
12-1pm, Wang Hall (Bldg. 59), Room 4102
Dr. Gerhard Wellein, Friedrich-Alexander-University Erlangen-Nuremberg
The ECM (Execution-Cache-Memory) performance model can be seen as an extension to the well-known Roofline Model (RLM). The ECM model aims at a clear identification of all relevant single core runtime contributions and uses a simple scaling approach to derive the scalability within a multicore chip. ECM model distinguishes between runtime contributions from in-core execution and the data delay, which includes all data transfers in the cache hierarchy. This way it allows to identify the relevant performance bottlenecks and to estimate performance improvements on the single core and chip level through code optimizations. Moreover, the ECM model can quantify the impact of processor clock speed on performance, which is relevant in the context of energy efficiency. The talk introduces the ECM model, presents several case studies, and reports on first work towards extending the model beyond x86 architectures. Moreover the Kerncraft tool will be presented which aims at the automatic generation of ECM and Roofline models for stencil-like and streaming kernels.
Link of the Week: The Value of a Good Science Hack