Core Algorithms in High Performance Scientific Computing

Date: 
28/09/2009 - 09:00 - 02/10/2009 - 17:00

Aims:

Many scientific calculations can be cast as problems in computational linear algebra and optimisation. Powerful algorithms for such problems form a common core of most scientific software. This school will develop a solid grounding in the mathematics of these algorithms; discuss the efficient implementation in a range of standard libraries and demonstrate effective use in real, high-performance scientific applications. This school is ideal for beginning PhD students or Postdocs in all scientific disciplines moving onto projects requiring substantial computational resources.

Topics

* Core algorithms underlying most of numerical computation.
* Libraries - BLAS, LAPACK, SCALAPACK, ARPACK, and more.
* Methods for dense and sparse matrices; preconditioners.
* Krylov methods such as Lanzos, Arnoldi, and GMRES.
* Cache performance and optimization.
* Fast Fourier Transforms.
* Error analysis.

Case Studies

* Computational fluid dynamics
* Molecular dynamics
* Atomic/electronic structure optimisation

Eligibility and Fees

Funding is through the HECToR national supercomputing service. Attendance is free to HECToR users and UK academics within the remit of one of the participating research councils (EPSRC, NERC and BBSRC). Industrial participants are asked to contact us for pricing via email to CSC Director. Attendees are required to cover travel and accommodation costs. Information on accomodation at the University of Warwick is available.
Prerequisites

Please note that this course does not teach a computer programming language. A good working knowledge of a scientific programming language (either Fortran- 95/2003 or C/C++) is a pre-requisite. Students will also be assumed to be familiar with basic Linux commands, text editors, and compilation of scientific code in a command-line environment. A background in a mathematical science or engineering is essential. A basic knowledge of linear algebra, Fourier methods and eigenvalue problems will be assumed.
Location

The course will be held at the University of Warwick in the Centre for Scientific Computing (CSC). Lectures and workshops will be given by staff within the CSC and from NAG Ltd - the Numerical Algorithms Group.
Registration

Students and researchers interested in attending the school should register via the HECToR website at http://www.hector.ac.uk. For further details about the course please contact Dr David Quigley d.quigley@warwick.ac.uk, Centre for Scientific Computing, University of Warwick.