Abstract
In the past, there was a clear consensus on which methods had to be taught to drive progress in the atomistic description of matter. A rough sketch is given of the methods that are in current use to produce quantitative, atomistic descriptions of structures and processes for systems ranging from molecules to crystals. From the point of view of a computational scientist who has been actively involved in their implementation on numerous generations of computers, features common to all of them becomes clear. This then becomes the foundation for a new focused effort of teaching and training to drive progress in the coming decades to address shortcomings in existing methods and to meet the long-promised objective of materials design and engineering.
Acknowledgement
I am grateful to the anonymous referees who made a number of detailed comments that contributed to make the paper clearer and some of the history more accurate and complete.