UC San Diego

As technology moves forward with newer hardware, new computational techniques, and new algorithms, it is important for science to keep up with these advances. Without people maintaining code to the latest standards and modifying the underlying algorithms to closely match the shifting hardware landscape, scientific software easily becomes outdated and deprecated on the latest architectures. Similarly as technology advances and acapability increases so the underlying methods should be adapted, extended or replaced in order to fully realize the scientific potential of the underlying improvements in the hardware. This research focuses on developing faster and higher accuracy free energy prediction methods as well as the development of simulation techniques necessary to improve the sampling efficiency of all atom, condensed phase, molecular dynamics simulations on exa-scale computational hardware. Specifically, it will cover efforts to develop GPU accelerated thermodynamic integration algorithms, low level code optimization on next generation CPU architectures, techniques for more efficient 3D domain decomposition and work to explore more efficient sampling techniques. These methods have been implemented as part of this work within the Amber software package. Additionally, to show the accuracy and performance of the code, comparisons against previous Amber versions as well as existing free energy prediction packages such as Schrodinger's FEP+ will be analyzed and discussed.