Perhaps the most pressing industrial and technological need for improvement in quantum scale modeling of molecules and molecular materials is to develop new methods and algorithms that deliver improved accuracy at reduced computational cost. Specifically, today's best methods involve computational cost that increases by a factor of 128 when the molecule size increases by a factor of 2. Even worse, their robustness is questionable under some important conditions, such as chemical bond-breaking, chemical reaction barriers, and treating transition metal systems. This proposal seeks to combine three important elements to remedy the current situation. These elements are first, improved basic methods, second, a reduced scaling implementation of the key steps of these methods, and third, exploitation of massively parallel computing power. Improved accuracy is demonstrated on model problems relevant to industry, which sets the stage for Phase II research aimed at bringing the programs and algorithms to production level, including parallel computing capabilities.
