Reducing the Complexity of Electronic Structure Calculations with Sparse and Low-Rank Data Structures and Novel Screening Techniques

Drew Lewis
Seminar

Formally high polynomial scaling, in computation and memory, algorithms used in electronic structure calculations can typically be reduced to linear or near linear by exploiting the rank and element sparsities of tensors constructed in local basis sets. This talk will discuss a new tensor approximation technique called the Clustered Low-Rank (CLR) framework as well as another new screening technique for reduced scaling local density fitting and show how these can be used to achieve reduced scaling and low memory usage in closed shell Hartree-Fock calculations.