Participating institutions:
Scientific Computing, Institute of Information Technology , Uppsala University	Maya Neytcheva, Ali Dorostkar
Geophysics, Department of Earth Sciences , Uppsala University	Björn Lund, Peter Schmidt, Jari Toivanen
Institute of Geonics , Czech Academy of Sciences	Radim Blaheta, Ondrej Jakl, Roman Kohut, Jiri Stary, Rostislav Hrtus, Martin Hasal
Institute of Information and Communication Technologies , Bulgarian Academy of Sciences	Svetozar Margenov, Krassimir Georgiev, Ivan Lirkov
PARALUTION project	Dimitar Lukarski

Project aims and scope:

Targeting to enable very large scale scientific computations on modern many-core computer systems, the project blends modelling, discretization, solution method development and program implementation aspects, in order to achieve fast and reliable computer simulations of a large class of problems originating in Geophysics and Geomechanics. The problems to simulate are primarily related to the impact of various factors on the safety of nuclear waste repositories, mining activities and more, such as

• current glacial isostatic adjustment due to melting glaciers and its effect on volcanism and earthquakes,
• self-gravitation and compressibility,
• incompressible flow in highly heterogeneous media,
• visco-poroelastic effects,
• thermo-effects,
• geothermal energy,
• cause of instabilities due to material damage.

The simulations are based on large finite element models.

The universal computational kernel, used in the numerical simulation of all above processes is the solution of very large algebraic systems (both linear and nonlinear), which arise from the underlying finite element models. Developing robust, fast and scalable solution techniques for those is therefore of prime interest. We emphasize our understanding that the aimed numerical efficiency is achievable if the specific features of the application field are utilized in the numerical solution techniques and their program implementation, remaining still applicable to the considered class of problems. Within this project, we intend to

• enrich the mathematical models with features of particular interest for the considered application fields, which are usually not included in the available software (commercial or public domain),
• development high performance iterative solvers, tailored to the above applications, emphasizing parallel computations on multicore architectures,
• voxel based analysis of processes in microscale and their influence to macroscale
• processing of seismic and tomographic images, providing the input data
• consider optimization techniques for model calibration and identification of material properties
• incorporate varying material properties, allowing for a strong heterogeneity of the media
• apply novel programming/processing techniques
• apply operator splitting techniques as a general framework to segment complex coupled models
• use suitable data structures to enhance existing algorithms with high degree of parallelism, such as alternating direction methods
• conduct performance studies on heterogeneous computer platforms (multicore CPU with and GPU) to evaluate the matchbetween the complex numerical algorithms and the specifics of the computer architecture

The project team has a long term experience in solving problems of the above level difficulty and has been collaborating both informally and through numerous projects since 1995 (Copernicus project 940820 High performance computing in Geosciences). Now we intend to address new challenges, formulated in the projects below and to combine our efforts to address the HPC aspect in the solution methods to be developed and tested.

• Decovalex 2015 - DEvelopment of COupled models and their VALidation against EXperiments
• IT4Innovations
• [NESUS] Network for sustainable ultrascale computing, within the EU COST framework.
• [PRACE 1IP] Partnership for Advanced Computing in Europe

We perform benchmarking of the numerical methods on the following HPC resources:

• Uppmax clusters (Uppsala)
• Inhomigeneous computer platforms (CPU and GPU)
• HPC facilities (Ostrava)
• BlueGene (Sofia)