SUMMARY
Accurately modeling interactions between atoms in a computationally efficient manner remains a grand challenge. Quantum mechanical methods possess great accuracy, but they are computationally bottlenecked, thus preventing the study of systems with size and temporal ranges relevant in many scientific and technological areas. Classical molecular dynamics (MD) simulations derive atomic forces from the negative gradient of analytical functions known as interatomic potentials, which approximate the potential energy between atoms. Interatomic potentials offer orders of magnitude decreases in computational cost compared to quantum mechanical methods although their approximation of interatomic energies and forces is less accurate. The ability to perform MD simulations with the accuracy of quantum mechanics and computational ease of classical interatomic potentials will propel the entire field of materials discovery and engineering. Atomic motions, namely vibrations and modes, give rise to thermal transport in solids in the form of phonons. Classical MD provides a general method to simulate heat transfer in any material or phase of matter, but its full potential has not been realized due to the lack of potentials which accurately reproduce phonon properties. This thesis presents an open-source program, POPS, which provides a convenient framework to optimize interatomic potentials that reproduce accurate phonon properties, thus resulting in phonon optimized potentials (POPs). The POPS program utilizes high performance computing and a genetic algorithm to perform fast optimizations. POPs are shown to accurately predict thermal conductivity, thus providing a computationally efficient way to study the atomic motions giving rise to phonon transport properties. Worked examples show the reproduction harmonic and anharmonic phonon properties in simple crystals, but the approach can be applied far beyond these simple systems, thus making it a crucial step forward.