This project will improve the scalability of quantum Monte Carlo (QMC) methods to better exploit modern cyberinfrastructure (CI) in finding very accurate energies of chemical and material systems. The research will include a collaborative effort across multiple disciplines to develop a linear scaling QMC algorithm suitable for parallelization at the petascale. This new QMC method will allow complex materials systems to be characterized theoretically with a high degree of accuracy. Presently, density functional theory (DFT) is used for large scale material modeling despite problems caused by electron self-energy interactions and the failure to describe long-range dispersion; however, modern CI has recently made QMC calculations of real materials tractable. The proposed research will exploit the use of localized basis functions and orbitals in order to reduce the computational cost of QMC calculations and to apply QMC to larger molecular and materials systems.
The project will also integrate education, career development, and research. Through a combination of curriculum development and outreach, the integration of computational chemistry into chemistry curricula will be promoted. The outreach program partners with the PSC in order to promote the participation of underrepresented groups in the computational sciences. The plan develops a broad educational infrastructure in computational science that is suited to students ranging from K-12 through advanced undergraduates. By extending the applicability of QMC, computational scientists will be able to use CI in order to design new solar cells and other alternative energy materials, model the behavior of molecules adsorbing at interfaces, and study highly-correlated semiconductors. The algorithms will be included in a freely available QMC code, CASINO, in order to promote their usage.
Through the funding of this postdoctoral fellowship, I investigated the scaling behavior and applicability of quantum Monte Carlo (QMC) methods to small molecules with unique and useful electronic properties. Among the most challenging tasks in modern computational chemistry is understanding and calculating the properties of electrons as they interact with one another and the nuclei. In theory, if the behavior of the electrons can be known exactly, then all properties can also be known. Unfortunately, this task is impossible. Traditional approaches to the electronic structure problem focus on methods for solving for either the wave function or the charge density; however, QMC methods represent a third way of solving this problem. In this class of methods, Monte Carlo integration techniques are used to numerically solve the Schrödinger equation. Although QMC methods hold great promise, there are significant drawbacks due to the computational cost. In particular, the most common variant of QMC is diffusion Monte Carlo (DMC), which is, in principle, exact, but is computationally expensive. As part of this fellowship, I investigated the tradeoffs between scaling and accuracy in order to determine the proper basis sets, functional form for Jastrow factors, and other parameters that are required in order to perform DMC calculations. In terms of applications, I chose to focus on small molecules that are particularly challenging for most traditional methods. One molecule in particular, tetramethyleneethane, is particularly challenging; this molecule has unique disjoint diradical character and has remained a challenge for theoretical tools for decades. I was able to demonstrate the proper recipe of basis set, trial wave function, parameters, and other aspects of the QMC methodology in order to produce the most high-quality results possible. The results have a wide applicability to many other scientifically interesting molecules in the field of molecular electronics. The results were published in the Journal of the American Chemical Society (JACS), one of the premier journals in chemistry. In terms of the broader impacts of the work on this grant, I developed several computational tools. One script for handling trial wave functions was produced for the freely available CASINO QMC code and has been made freely available. In addition, I performed multiple outreach activities with the goal of increasing interest in science by underrepresented groups. In terms of career development activities, I as a mentor to graduate and undergraduate students--both on projects in which I was a direct collaborator and on projects outside of my own research.
