InTheLoop | 03.16.2015
Computing Sciences Staff Presenting at SIAM CSE15 Conference
Computing Sciences staff will be presenting some three dozen research talks and posters at the 2015 SIAM Conference on Computational Science and Engineering (CSE15) being held March 14-18 in Salt Lake City. »Read the list of NERSC and CRD contributions.
Between Micro and Macro, Berkeley Lab Mathematicians Model Fluids at the Mesoscale
When it comes to boiling water—or the phenomenon of applying heat to a liquid until it transitions to a gas—is there anything left for today’s scientists to study? The surprising answer is, yes, quite a bit. How the bubbles form at a surface, how they rise up and join together, what are the surface properties, what happens if the temperature increases slowly versus quickly—while these components might be understood experimentally, the mathematical models for the process of boiling are incomplete.
The math whizzes at the Center for Computational Sciences and Engineering (CCSE) at Lawrence Berkeley National Laboratory (Berkeley Lab) are tackling this and many other similar problems head on. “Modeling fluids at the mesoscale is a relatively new area,” said John Bell, principal investigator of the project looking at these phenomena. “The idea is to incorporate important physical effects at the microscale into continuum models. We’re constructing equations that have the best of both worlds.”
Along with Alejandro Garcia from San Jose State University and Aleksandar Donev from the Courant Institute, Bell and colleagues at CCSE are at the forefront of a neglected corner of the scientific world, building mathematical models for fluids at the mesoscale. »Read more.
Former NERSC Consultant Mentors Math, Computer Science Students
Frank Hale, a former consultant in NERSC’s User Services Group (USG) who currently tutors math at Diablo Valley College (DVC) in Pleasant Hill, CA, recently brought a group of computer science enthusiasts from the college to NERSC for a tour. On February 27, Hale and 10 DVC students—primarily members of the college’s Computer Science club—visited NERSC, where they were hosted by Elizabeth Bautista; heard talks from Richard Gerber, David Skinner and Jack DeSlippe on high performance computing and how it is used at NERSC to enable a broad range of science; and took a tour of the machine room »Read more.
This Week's CS Seminars
Matrix Computations and Scientific Computing Seminar: Write-avoiding Algorithms
Wednesday, March 18, 11 a.m.–12 p.m., 380 Soda Hall, UC Berkeley Campus
Jim Demmel, UC Berkeley and Berkeley Lab
Motivated by the high cost of communication, i.e. moving data, there has been much recent work on communication-avoiding (CA) algorithms. But this work does not distinguish two kinds of communication, reads and writes, which may have very different costs. For example, in emerging Non-Volatile Memory technologies, writing is much more expensive than reading. This motivates us to identify or invent write-avoiding (WA) algorithms, that do asymptotically fewer writes than reads, attaining new lower bounds when possible. We give several examples of new lower bounds on writes and WA algorithms. For example, some but not all CA algorithms for O(n3) matrix multiplication are also WA, but Strassen cannot be WA.
Adaptive finite element methods for turbulent flow and fluid-structure interaction with applications in aerodynamics and biomedicine
Wednesday, March 18, 2–3 p.m., 891 Evans Hall, UC Berkeley Campus
Johan Hoffman, KTH Royal Institute of Technology
We present adaptive finite element methods to compute weak solutions to the Navier-Stokes equations, which are used to simulate turbulent fluid flow. The mesh is adaptively refined with respect to output functionals of the weak solutions, such as lift or drag of an airplane, and we observe convergence in such functionals as the mesh is refined. The adaptive method is based on the computation of an adjoint (or dual) problem to derive a posteriori estimates of the error in a functional, which guides the mesh refinement algorithms. We also present a monolithic method for solving fluid-structure interaction problems. The numerical methods are implemented as part of the open source software FEniCS, and we show applications of the technology including the aerodynamics of airplanes and biomechanics of the human voice and the human heart.