Project Description

Electronic structure calculations of a wide range of molecular properties are an integrated part of many branches of molecular sciences. The coupled-cluster (CC) model is the state-of-the-art wave function method, and, for smaller molecular systems, various molecular properties have been computed to an accuracy challenging experimental results. However, the application range of CC has so far been limited to small molecular systems due to their computational scaling with system size.

For this reason, density function theory (DFT) has developed into a workhorse for large-scale applications. The major drawback of DFT calculations is that they generally do not possess the accuracy and the predictive power of CC methods.

The objective of the proposed project is to extend the application range of accurate modeling techniques in chemistry and molecular sciences by making the CC methods applicable to large molecular systems and ready-to-use on the supercomputers of tomorrow. The proposed project is a continuation of previous INCITE work.

The team plans to remove bottlenecks for Hartree-Fock equation optimization and orbital localization, which is essential to implement the team’s Divide-Expand-Consolidate (DEC) algorithm. The algorithm’s framework is well established for coupled cluster models, but the overall precision is still limited by memory constraints. By removing these bottlenecks, the team hopes to address molecular systems containing up to 100,000 basis functions.

All the team’s goals involve the implementation of massively parallel algorithms which can efficiently utilize the entire Titan supercomputer when applied to large molecules. All developments will be implemented in the non-commercial LS-Dalton program package and thus become freely available to scientists in other fields.

Allocation History

Source Hours Start Date End Date