NAIS Lecturers

Lubomir Banas Department of Mathematics Heriot-Watt University l.banas@hw.ac.uk

Numerical analysis, scientific computing, mathematical modelling, nonlinear partial differential equations (including stochastic PDEs), multiscale methods with applications to multiphase flow problems, magnetohydrodynamics, electromagnetism, micromagnetism and porous media flow.

Ozgur Ergul University of Strathclyde Livingstone Tower 26 Richmond Street ozgur.ergul@strath.ac.uk

Dr Ergul’s research interests include fast and accurate algorithms for electromagnetic scattering, radiation, and transmission problems. He develops parallel implementations of the multilevel fast multipole algorithm (MLFMA) for rigorous solutions of real-life problems [1], such as scattering from airborne targets, scattering from human tissues [2], radiation from antennas and their compatibility with the environment, wireless communications, and electromagnetic transmission through artificial structures, namely, metamaterials and photonic crystals [3]. Recently, Dr Ergul developed a hierarchical partitioning strategy, which enables very efficient parallelization of MLFMA on distributed-memory architectures. The resulting implementations have been employed on moderate computers to solve real-life problems of three-dimensional objects (modeled with dense matrix equations) discretized with hundreds of millions of unknowns [1]. Full-wave solutions of these large-scale problems mean more accurate simulations of realistic scenarios involving various geometries, excitations, and ranges of frequency. Accurate simulation results can lead to new horizons in the aforementioned areas, e.g., by enabling the engineers to explore novel designs of radar systems, medical diagnosis tools, antennas, wireless systems, metamaterials and photonic crystals without their actual realizations and reducing the time/expense involved in building prototypes and carrying out laboratory tests. Dr Ergul’s current research aims to extend and improve parallel algorithms and their implementations for more realistic simulations, particularly considering novel aspects in high-performance computing and parallel computers. [1]O. Ergul and L. Gurel, Rigorous solutions of electromagnetic problems involving hundreds of millions of unknowns, IEEE Antennas Propag. Mag., vol. 53, no. 1, pp. 18-26, (2011) [2]O. Ergul, A. Arslan-Ergul, and L. Gurel, Computational study of scattering from healthy and diseased red blood cells using surface integral equations and the multilevel fast multipole algorithm, J. Biomed. Opt., vol. 15, no. 4, pp. 1-8 (2010) [3]O. Ergul, T. Malas, and L. Gurel, Analysis of dielectric photonic-crystal problems with MLFMA and Schur-complement preconditioners, J. Lightwave Technol., vol. 29, no. 6, pp. 888-897 (2011)

Sebastien Loisel

Sebastien Loisel has his PhD from McGill University. He comes to us from a postdoctoral position at the University of Geneva. Sebastien studies iterative methods such as domain decomposition, Schwarz preconditioners, and low-rank approximations in statistics.

Vijay Nagarajan

Vijay 's research interests lie in the areas of compilers, computer architecture and software engineering. His dissertation proposes an efficient and programmable runtime monitoring approach for multicores, which can be used to increase the performance and reliability of parallel programs running on such architectures. While parallel architectures are becoming ubiquitous, extracting performance from them is contingent on programmers writing parallel software. However, writing and debugging parallel software is notoriously tricky. One significant reason is because of the intricacies involved in the underlying memory consistency model. Among the various memory consistency models, the sequential consistency (SC) model in which memory operations appear to take place in the order specified by the program, is most intuitive. However, most of the processors choose not to support SC because of performance reasons. In [1], we developed an efficient and lightweight approach for guaranteeing SC at less than 2pc performance overhead, thus paving the way for programmable, yet efficient memory models for multiprocessors. [1]Changhui Lin, Vijay Nagarajan and Rajiv Gupta, "Efficient Sequential Consistency using Conditional Fences," Parallel Architectures and Compilation Techniques (PACT), September 2010, pp 295-306.

Magnus Svard University of Edinburgh Mayfield Road Edinburgh, EH9 3JZ Magnus.Svard@ed.ac.uk

Flow problems in engineering applications are computationally extremely challenging. Turbulence generation and propagation requires vast computer resources to resolve (See [2].) Highly accurate numerical schemes will reduce the computer resources needed and therefore push the boundaries of what is possible to compute. (See [1].) Another prominent feature for high-speed flows is shock waves. Numerical schemes designed for shock wave solutions are usually of low accuracy and not optimal for resolving turbulence. The primary goal of this project is to design highly accurate and robust numerical schemes, that tackle the above phenomena in a unified way. (See [3].) Although the approach is mathematical, the aim is to make the schemes as simple as possible to be attractive to engineers and crucially, to be effective on modern parallel architectures. In the next step, the schemes will be extended to the equations of Magneto-Hydrodynamics with the aim to make large-scale computations of fusion reactors. [1]M. Svard, J. Lundberg, and J. Nordstom, A computational study of vortex–airfoil interaction using high-order finite difference methods, Comp and Fluids, 39 (2010) [2]M. Shoeybi, M. Svard b, F. E. Ham and P. Moin, An adaptive implicitexplicit scheme for the DNS and LES of compressible flows on unstructured grids, J. Comp. Phys., 229, 2010 [3]M. Svard and S. Mishra, Shock capturing artificial dissipation for high-order finite difference schemes, J. Sci. Comp., 39, (2009)