283633 New Predictive and Efficient Computational Tools for Studying Catalysis: From Transition Metal Surface Chemistry to Enzyme Engineering

Sunday, October 28, 2012
Hall B (Convention Center )
Heather J. Kulik, Chemistry, Stanford University, Stanford, CA

Atomistic simulation provides valuable mechanistic insight into catalytic activity in systems ranging from the smallest molecular catalysts to complex surface phenomena in materials science and enzyme activity in the biological sciences. For this reason, rational, computational design is a promising pathway to the development of catalysts for applications in energy science, pharmaceuticals, and nanoscale electronics. Nevertheless, the effectiveness of a computational design strategy is dependent upon both the predictive accuracy and computational cost of the methods employed.  Quantum mechanical methods, in particular, are impressively accurate: they directly incorporate the key physics needed to describe phenomena from bond rearrangement to charge transfer.  However, computational cost constrains the most accurate QM methods to only a handful of atoms.   

Density functional theory (DFT) is a widely-employed computational approach for determining the electronic structure of molecules and solids, but standard approximations in DFT cause it to suffer from a critical error known as self-interaction. This self-interaction error makes electrons over-delocalize and gives rise to erroneous descriptions of geometry and energetics, particularly for transition metals. I have developed and extended Hubbard-augmented DFT approaches that ameliorate self-interaction error and shown that these so-called DFT+U approaches reduce errors in the ordering of electronic states of small molecules and paradigmatic catalytic reactions by over an order of magnitude. Importantly, with little additional overhead, the U term is calculated self-consistently on each system considered and is thus not a fitting parameter. The efficient scaling of DFT+U and its extensions have permitted us to make new predictions that were later validated by experiment about systems ranging from self-assembled monolayers on transition-metal surfaces to the reactivity of metalloenzymes that are components of natural product biosynthesis.

In addition to accuracy, a critical concern is the system size that can be attacked by computational methods.  As part of the first successful effort to dramatically speed up QM methods, I have developed and extended tailored algorithms that harness the massively parallel nature of graphical processing units (GPUs).  These GPU-accelerated QM methods permit us to carry out unprecedented studies of proteins fully quantum mechanically, where less accurate force field methods were previously the only option.  Using these new approaches I've quantified how quantum mechanical descriptions of proteins differ from those obtained with force fields and determined that QM methods excel in describing the disordered features of proteins that are likely to be key in mediating catalysis. I've also identified that quantum mechanical effects at enzyme active sites are much larger scale than usually considered through standard QM/MM approaches.  A combination of both more accurate (DFT+U) and more efficient (GPU-accelerated) methods paves the way for predictive accuracy on large-scale, realistic systems relevant for applications in catalysis ranging from surface science to enzyme engineering.

Extended Abstract: File Not Uploaded
See more of this Session: Meet the Faculty Candidate Poster Session
See more of this Group/Topical: Education Division