The present volume contains a selection of papers from the research presented at the workshop in Uncertainty Quantification for High Performance Computing, held at Oak Ridge National Laboratory (ORNL), from May 2 to 4, 2012. The workshop was supported by the Computational Science and Mathematics division at ORNL, the office of Advanced Scientific Computing Research (ASCR) at the United States Department of Energy (DOE), a National Science Foundation (NFC) research centre, and the Statistical and Applied Mathematical Sciences Institute (SAMSI), a National Science Foundation (NFC) research centre. With over 40 participants, this workshop gathered together leaders in the field of uncertainty quantification for scientific simulations executed on world class high-performance computing machines.
The past decades have seen an explosion of computing power, with the top machines in the world capable of sustained peta-flops rates of calculation. In other words, these machines can complete over 1015 calculations per second. These machines have opened a door to perform scientific simulations at a scale never experienced before. For example, the resolution of climate simulations that can be performed on these machines has increased spatial resolution down to a small fraction of degrees with increased physical fidelity that has never before been simulated. Material, fusion, nuclear, chemical, and bio-medical simulations are also receiving similar boosts to their accuracy on these new high-performance computing machines. Many of the applications that are simulated on these systems represent experiments that can only be performed through simulations, such as the simulation of super-nova or fusion reactor plasma dynamics or future climate scenarios, to name but a few.
The benefits of current high-performance computing has spurned on the international race to develop exascale supercomputing systems that can complete over 1018 calculations per second. It is clear from the current and projected limits of power consumption and trends in computing that future designs of exascale supercomputing systems will require a paradigm shift in their architecture. This revolution in computer design will require an equal revolution in mathematical algorithms and theory for these exascale supercomputing systems to realize their full potential. Key drivers to developing effective mathematical methods for these systems will be achieved through algorithms and theory that expose hierarchies of parallel work while minimizing the power cost of data movement and latency in communication. The field of computational uncertainty quantification will have a unique role to play in maximizing the knowledge that can be gained through performing scientific simulations of these supercomputing systems.
The uncertainty quantification (UQ) for High Performance Computing workshop gave researchers a venue to address both theoretical and computation issues involved with UQ in high-performance computing. Specific topics of the workshop included scalable algorithms for UQ, calibration, estimation and identification, and data-driven reduced order models for UQ, and this special issue presents details of the topics that were addressed. This workshop represents one in a series of ongoing workshops fostered by ASCR to develop the computational, mathematical, and statistical tools that are needed to support scientific simulation at leadership class computing facilities.
Specifically, the papers that deal with the topics referred to earlier are:
Exploring emerging manycore architectures for uncertainty quantification through embedded stochastic Galerkin methods, Eric Phipps, Jonathan Hu and Jakob T. Ostien [Citation2].
Integrating data and compute-intensive workflows for uncertainty quantification in large-scale simulation: application to model-based hazard analysis, Shivaswamy Rohit, Abani Patra and Vipin Chaudhary [Citation5].
Proper orthogonal decompositions in multifidelity uncertainty quantification of complex simulation models, Oleg Roderick, Mihai Anitescu and Yulia Peet [Citation4].
Emulation to simulate low-resolution atmospheric data, Vishwas Rao, Rick Archibald and Katherine J. Evans [Citation3].
Efficient computation of unsteady flow in complex river systems with uncertain inputs, Nathan L. Gibson, Christopher Gifford-Miears, Arturo S. Leon and Veronika S. Vasylkivska [Citation1].
An adaptive sparse-grid iterative ensemble Kalman filter approach for parameter field estimation, Clayton G. Webster, Guannan Zhang and Max Gunzburger [Citation6].
References
- N.L. Gibson, C. Gifford-Miears, A.S. Leon, and V.S. Vasylkivska, Efficient computation of unsteady flow in complex river systems with uncertain inputs, Int. J. Comput. Math. 91(4) (2014), pp. 781–797.
- E. Phipps, J. Hu, and J.T. Ostien, Exploring emerging many core architectures for uncertainty quantification through embedded stochastic Galerkin methods, Int. J. Comput. Math. 91(4) (2014), pp. 707–729.
- V. Rao, R. Archibald, and K.J. Evans, Emulation to simulate low-resolution atmospheric data, Int. J. Comput. Math. 91(4) (2014), pp. 770–780.
- O. Roderick, M. Anitescu, and Y. Peet, Proper orthogonal decompositions in multifidelity uncertainty quantification of complex simulation models, Int. J. Comput. Math. 91(4) (2014), pp. 748–769.
- S. Rohit, A. Patra, and V. Chaudhary, Integrating data and compute-intensive workflows for uncertainty quantification in large-scale simulation: Application to model-based hazard analysis, Int. J. Comput. Math. 91(4) (2014), pp. 730–747.
- C.G. Webster, G. Zhang, and M. Gunzburger, An adaptive sparse-grid iterative ensemble Kalman filter approach for parameter field estimation, Int. J. Comput. Math. 91(4) (2014), pp. 798–817.