The connections between dense linear algebra and quantum chemistry methods such as Hartree-Fock, Density-Functional Theory and coupled-cluster theory will be explored from the perspective of programming models and numerical algorithms.
In the first part, we describe the well-known massively parallel quantum chemistry code NWChem and its associated programming model and runtime system (Global Arrays) and compare it to more traditional synchronous MPI-based approaches.
The apparent discrepancies between these models have been reconciled, enabling portability and performance of NWChem and Global Arrays on state of the art supercomputers.
In the second part, we consider the algorithms currently used for tensor contractions in quantum chemistry codes and demonstrate a completely new approach that is intimately connected to well-understood fast algorithms for matrix multiplication and parallel sorting.
The challenge of load-imbalance in coupled-cluster codes is completely eliminated in this new approach, which has significant implications for runtime system requirements.

This includes joint work with Jim Dinan (Argonne), Edgar Solomonik (Berkeley) and Devin Matthews (Texas).