Close
All seminars (unless otherwise stated) will take place on Tuesdays at 3.00pm in Room 505 in the Mathematics Department (25 Gordon Street). See how to find us for further details. There will be tea afterwards in Mathematics Room 606. If you require any more information on the Applied seminars please contact Prof Slava Kurylev e-mail: y.kurylev AT ucl.ac.uk or tel: 020-7679-7896.
Special Applied Seminar, Tuesday 14 June 2016 in Room 505 at 4:30pm
Wiliam K Devar (The Pierre Welander Professor of Oceanography, Florida State University)
Title: Gulf Stream Separation Dynamics
Abstract:
Numerical models are plagued by poor representations of Gulf Stream separation. Rather than moving to the deep ocean at Cape Hatteras, many model Gulf Streams 'overshoot' the separation, exiting the US east coast several degrees to the north of Cape Hatteras. This is responsible for the largest bias in ocean SST found in coupled climate models. Higher model resolution often helps, but even then correct separation is not guaranteed. We examine Gulf Stream separation in a regional model with a view towards understanding the dynamical controls on separation. Our results support the convergence of the isobaths northeast of the Charleston Bump as the key element. Initial tests of a theoretical explanation based on mean flow-topography interaction support the result.
Tuesday 5 July 2016 in Chadwick B.05 LT at 3pm
Florian deVuyst (ENS-Cachan, France)
Title: New numerical solvers subject to HPC performance constraints: Lagrange-flux approach for compressible hydrodynamics and Saint-Venant equations. Related Big Data problems with CFD computations.
Abstract:
Today's multicore/manycore computing architecture allow for potential high-performance computing but numerical solvers need suitable scalability with SIMD feature and rather high arithmetic intensity for that. This new constraint make the Numerical Analysis community search for new computational approaches, possibly changes of paradigm while trying to keep expected features like high-order accuracy, stability. This talk will present an instance of evolution of solvers: starting from a family of Lagrange+remap solvers for compressible Hydrodynamics, we will achieve a performance analysis from performance models (roofline, ECM). We will reinterpret the geometrical remapping step as an algebraic convection process which has a far better SIMD feature. The resulting solver, referred to as Lagrange-flux solver is actually very general and can be applied to a broad class of convection-dominated problems. We will extend the scheme to the shallow-water equations, achieving the "well-balanced" property for the gravity source term. We will also discuss the issue of handling dry/wet zones without interface tracking, still for the sake of performance. See the video : https://youtu.be/6C3HcjtsYsE . Finally, we will discuss current and outlook research about Big Data aspects from simulations of debris deposition caused by marine flooding.