The availability of big data and the computational resources that can crunch them transformed machine learning into a very useful tool for science and industry. On the other hand, accurate and robust quantum chemistry methods transformed computational simulations into a predictive tool for chemists. However, most quantum chemistry software can not efficiently make use of the massively-parallel computing architectures available today and we do not have accurate, diverse and large data sets to make use of deep learning algorithms effectively. Additionally, the relative lack of input/output standardization and APIs makes it difficult to curate data based on a workflow that depends on various different software packages. With exascale resources on the horizon, we need to address these challenges by developing scalable and sustainable software stacks and by engaging community standards as developed by MolSSI. In this talk, I will focus on three projects, SIPs, QTC, and VALENCE that are designed to address these challenges with exascale resources in mind. SIPs is a parallel sparse eigensolver available through the open-source SLEPc library and can be integrated into quantum chemistry packages through the ELSI interface. QTC is a Python workflow code that glues various quantum chemistry packages with thermochemistry codes to generate accurate thermochemical data sets in a high-throughput mode. VALENCE is a highly scalable orthogonalization-free quantum chemistry code based on variational subspace valence bond method. VALENCE is integrated with SIMINT library for vectorized electron repulsion integrals using CCA standard ordering. Main functions of VALENCE can be accessed through a simple API, and inputs can be automatically generated by using a tool based on Open-Babel. I will discuss how modern software engineering tools (version control, auto-documentation, continuous integration) and productivity tools (Jupyter notebooks, integrated development environments, containers) help develop and maintain these codes and improve reproducibility of scientific workflows.

Facing exascale and big data challenges with modular software, resilient workflows and productivity tools

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.