Methods, software, and benchmarks for modeling long timescale dynamics in solid-state atomic systems

textThe timescale of chemical reactions in solid-state systems greatly exceeds what may be modeled by direct integration of Newton's equation of motion. This limitation spawned the development of many different methods such as (adaptive) kinetic Monte Carlo (A)KMC, (harmonic) transition state theory (H)TST, parallel replica dynamics (PRD), hyperdynamics (HD), and temperature accelerated dynamics. The focus of this thesis was to (1) implement many of these methods in a single open-source software package (2) develop standard benchmarks to compare their accuracy and computational cost and (3) develop new long timescale methods. The lack of a open-source package that implements long timescale methods makes it difficult to directly evaluate the quality of different approaches. It also impedes the development of new techniques. Due to these concerns we developed Eon, a program that implements several long timescale methods including PRD, HD, and AKMC as well as global optimization algorithms basin hopping, and minima hopping. Standard benchmarks to evaluate the performance of local geometry optimization; global optimization; and single-ended and double-ended saddle point searches were created. Using Eon and several other well known programs, the accuracy and performance of different algorithms was compared. Important to this work is a website where anyone may download the code to repeat any of the numerical experiments. A new method for long timescale simulations is also introduced: molecular dynamics saddle search adaptive kinetic Monte Carlo (AKMC-MDSS). AKMC-MDSS improves upon AKMC by using short high-temperature MD trajectories to locate the important low-temperature reaction mechanisms of interest. Most importantly, the use of MD enables the development of a proper stopping criterion for the AKMC simulation that ensures that the relevant reaction mechanisms at the low temperature have been found. Important to the simulation of any material is knowledge of the experimental structure. Extended x-ray absorption fine structure (EXAFS) is a technique often used to determine local atomic structure. We propose a technique to quantitatively measure the accuracy of the commonly used fitting models. This technique reveals that the fitting models interpreted nanoparticles as being significantly more ordered and of much shorter bond length than they really are.Chemistr