Understanding the fundamental aspects, and the formation mechanisms of glasses remains an outstanding unsolved problem of condensed-matter physics.

Glass-formation in simple atomic systems is generally associated with the icosahedral local order. In this study, different aspects of the liquid-glass transition are investigated in a computer simulation using a model system with predominantly icosahedral local order. We demonstrate that the dramatic slowing down of the relaxation dynamics observed in this system when approaching the glass transition from the liquid phase, and the eventual structural arrest are related to a structural transformation which is recognized as formation of a percolating cluster composed of icosahedra. It is concluded that the rapid cluster growth is inherently related to the tendency for low-dimensional cluster geometry induced by the particular form of the pair potential. A model is suggested interpreting the characteristic dynamical anomalies commonly associated with supercooled liquid dynamics, including breaking the Stokes-Einstein relation, as arising from the long-time decomposition of the relevant phase-space region. This decomposition, caused by the development of long-range clusters, is also manifested in growing cooperativity of diffusive dynamics.

Another part of this study concerns the vibrational dynamics in a glass formed by the investigated model from the liquid state. We analyze the glass dynamics by comparing it with dynamics of the $\sigma$ phase, a Frank-Kasper crystal which is morphologically related to this glass and can possibly be regarded as the crystalline ground state for this system. The structural similarity of the $\sigma$ phase with this glass is also revealed by using the wavelet analysis of pair-correlation functions. The pattern of vibrational dynamics in this glass is found to be distinctly different from that in the glass formed using the Lennard-Jones potential. It is demonstrated that these distinctions can be accounted for by considering the vibrational dynamics in the respective crystalline phases.

In order to facilitate the computational studies of extremely slow dynamics of supercooled liquids, we have developed an efficient, cache coherent and parallelizable molecular dynamics algorithm. This algorithm was implemented for a shared memory symmetric multi-processor architecture using OpenMP directives. A large number of auxiliary utilities has been created or adapted for the analysis of the data from our molecular dynamics simulations.