Big data, sparse information: Bayesian inference for large-scale models, with application to inverse modeling of Antarctic ice sheet dynamics; Omar Ghattas, Center for Computational Geosciences, The University of Texas at Austin, USA

Predictive models of complex geoscience systems often contain numerous uncertain parameters. Rapidly expanding volumes of observational data present opportunities to reduce these uncertainties via solution of inverse problems. Bayesian inference provides a systematic framework for inferring model parameters with associated uncertainties from (possibly noisy) data and any prior information. However, solution of the Bayesian inverse problem via conventional Markov chain Monte Carlo methods remains prohibitive for expensive models and high-dimensional parameters, as result from discretization of infinite dimensional problems with uncertain fields. Despite the large size of observational datasets, typically they can provide only sparse information on model parameters. Based on this property we design MCMC methods that adapt to the structure of the posterior probability and exploit an effectively-reduced parameter dimension, thereby making Bayesian inference tractable for some large-scale, high-dimensional inverse problems. We apply the methodology to an inverse problem for Antarctic ice sheet flow: given a non-Newtonian flow model of the ice sheet and free surface velocity observations, we seek to infer the sliding coefficient field at the base of the ice.

This work is joint with Tan Bui-Thanh, Tobin Isaac, James Martin, Noemi Petra, Georg Stadler, Hari Sundar, and Hongyu Zhu.

The High Level Support Team — A support unit for petaflop computing; Roman Hatzky, Max-Planck-Institut fuer Plasmaphysik, IPP-EURATOM-Association, Germany

Modern petaflop computer architectures are already so demanding that an optimal exploitation can be only performed with a special support unit. The High Level Support Team (HLST) provides support to scientists from all EFDA (European Fusion Development Agreement) Associates for the development and optimization of codes to be used on the dedicated IFERC-CSC supercomputer located in Rokkasho, Japan. The supercomputer delivers computing power of about 1.52 petaflop/s and is optimally suited for the fusion scientists’ simulation programs. The HLST consists of a core team based at the IPP Garching (Max-Planck-Institut für Plasmaphysik) and of other high level support staff provided by the Associates. The HLST members are all HPC experts with a background in developing large scientific applications and particular expertise in numerical algorithms and and visualization.

Towards the development of wavefunction-based methods for materials modeling; Ali Alavi, Department of Chemistry, Trinity College, University of Cambridge, United Kingdom

We review some recent progress in using Monte Carlo methods in combination with quantum chemical methods to solve to a high degree of accuracy the many-electron Schrodinger equation, both in the molecular context as well as periodic boundary conditions. At the heart of the methods is FCIQMC [1,2,3], which simulates an exact stochastic representation of the FCI wavefunction of a physical system expressed in a finite basis. The sampled FCI wavefunction can be used to compute the 1- and 2-electron reduced density matrix, from which a great many physical properties can be extracted. In particular, these  can be used to compute a perturbative basis-set correction [3], which very accurately accounts for the basis-set error arising from the use of finite basis sets. We have recently extended this method to periodic systems [4], and results will be shown in this regard. Finally, we will review recent progress on alleviating the time-step bottleneck of FCIQMC, opening the possibility to simulate substantially larger systems than currently possible.


[1] G.H. Booth, A.J.W. Thom and Ali Alavi, J. Chem. Phys., 131 , 054106, (2009).
[2] Deidre Cleland, G.H. Booth, and Ali Alavi, J. Chem. Phys., 132, 041103, (2010).
[3] George H. Booth, Andreas Grueneis, Georg Kresse, and Ali Alavi, Nature , 493 365-370, (2013).
[4] G.H. Booth, D.M. Cleland, Ali Alavi, David Tew, J. Chem. Phys., 137, 164112, (2012).
[5] A. Grueneis, JJ Shepherd, A. Alavi, D.P. Tew, G.H.Booth, J. Chem. Phys, 139, 084112 (2013).

The agenda of the CSCS User Meeting includes the following topics:

• 13:10 Message from the Director
• 13:30 User Lab Overview
• 13:40 Systems Overview
• 14:00 HPC Data & Visualization Services Overview
• 14:20 Training Overview

HPC & The Inventor’s Dilemma: Abandoing 20 years of Molecular Simulation Knowledge and Moving to GPUs; Eric Lindahl, SciLifeLab Stockholm, KTH, Sweden

Molecular dynamics simulations are undergoing a second revolution. Computers are suddenly fast, large, and cheap enough that it is possible to perform simulate a large number of biomolecular processes that were completely out of reach only a few years ago, for instance the opening or closing of an ion channel. However, the free lunch in terms of constantly faster processors is likely over – while computers get cheaper per processor, the improved performance now has to come from increasing parallelism, which is a huge problem for a computational algorithm such as molecular dynamics that historically has not scaled well for small problem sizes.

Second, after 20 years of work in a field programs tend to be large and complex, which makes them difficult to turn around.Here, I will describe our early efforts on high-end optimization & parallelization of a scientific code, and in particular how we decided to revisit the entire problem from scratch a few years ago. Not only did this lead to a molecular simulation implementation that is by far the fastest of any on GPUs, but surprisingly we could take many of the ideas back to modern CPUs and improve scaling considerably.