Linear scaling algorithms: applicability, accuracy and scalable implementation
- π€ Speaker: Vincent Sacksteder (Asia Pacific Center for Theoretical Physics)
- π Date & Time: Wednesday 10 December 2008, 10:00 - 10:40
- π Venue: TCM Seminar Room, Cavendish Laboratory
Abstract
Linear scaling algorithms are already delivering on their promise for large scale electronic structure calculations on parallel computers, but their accuracy, justification, and limits are not fully understood. Mathematically speaking these algorithms are fast approximations to the density matrix, while physically speaking they are implementations of the idea called nearsightedness. In this talk I argue that linear scaling algorithms should be useful for calculating a much wider range of matrices occurring in fields ranging from nuclear physics to engineering. I also present numerical evidence that linear scaling algorithms can be exponentially accurate even in metals as long as there is some disorder. Lastly I discuss the enormous difference between having a scalable algorithm and having an implementation which scales to the largest supercomputers, and propose a general purpose linear scaling library for calculating matrices on the largest supercomputers.
Download slides (PPT)
Series This talk is part of the Electronic Structure Discussion Group series.
Included in Lists
- All Cavendish Laboratory Seminars
- All Talks (aka the CURE list)
- CamBridgeSens
- Cambridge talks
- Centre for Health Leadership and Enterprise
- Combined TCM Seminars and TCM blackboard seminar listing
- Electronic Structure Discussion Group
- Featured lists
- Lennard-Jones Centre external
- Life Science Interface Seminars
- ME Seminar
- Neurons, Fake News, DNA and your iPhone: The Mathematics of Information
- PMRFPS's
- School of Physical Sciences
- TCM Seminar Room, Cavendish Laboratory
- Thin Film Magnetic Talks
- Trust & Technology Initiative - interesting events
- yk449
Note: Ex-directory lists are not shown.
![[Talks.cam]](/static/images/talkslogosmall.gif)


Wednesday 10 December 2008, 10:00-10:40