How information integration and data mining can reduce computational complexity in materials world

Nikolai Zarkevich, Physics, UIUC

Computational complexity of multi-scale methods based on the first-principles energetics is analyzed. Computational cost of such methods is dominated by expensive first-principles determination of structural energies and atomic forces, including structural relaxations. Information integration is an opportunity to eliminate recalculation of known data, and, consequently, to reduce the total cost of multi-scale calculations. We propose the Structural Database as a universal tool for structural data integration, and explain its conceptual design and functionality, including data mining options. To exemplify benefits provided by the Structural Database, we consider search for new ground states, construction of a phase diagram for a bulk material, and prediction of surface patterning. We show, that the Structural Database is a powerful tool for information integration and data mining, which can greatly reduce computational complexity in materials world.