Email: ccsong@ucdavis.edu
Related websites:
July 2021 - Present: Assistant Professor, UC Davis, Davis, CA
Sep 2018 - May 2021: Postdoctoral researcher, UC Berkeley and Lawrence Berkeley National Lab, Berkeley, CA
Sep 2012 - June 2018: Ph.D. in Chemistry, Stanford University, Stanford, CA
Aug 2008 - June 2012: B.S. in Chemistry, Tsinghua University, Beijing, China
Theoretical understanding of photochemistry has been challenging, because quantum mechanics is required for describing the qualitative change of electronic degrees of freedom. The desired electronic structure methods would satisfy the following requirements: dynamic correlation is needed for accurate predictions of barriers; static correlation is needed for bond breaking in photoisomerization; the topology of conical intersections (i.e. degeneracy is lifted by two degrees of freedom) is required for describing non-radiative transitions. Multi-state multi-reference perturbation theory (MSPT2) can fulfill all of the above requirements, and is often considered as a gold standard for studying photochemistry, but is limited to small molecules due to its high computational cost. With the help of a series of general tools that I developed over the years (described below), I have been able to make MSPT2 faster by at least an order of magnitude, thus enabling applications that were previously inaccessible.
Different electronic structure methods often share similar structure and patterns during their development. For example, most correlation methods can be formulated as tensor contractions. By recognizing and identifying these common patterns, one can then use diagrams to represent the structure in an intuitive way, and to facilitate the development of an automatic and systematic procedure for generating optimized codes. Within this context, I have developed the Automated code engine (ACE) for generating GPU integral kernels based on decision graphs, as well as Automatic differentiation tools for developing analytical gradients based on circuit diagrams from quantum computing. These tools can greatly save effort when implementing new methods.
The fourth order electron repulsion integral (ERI) tensor is a major source of complexity in electronic structure theory. Fortunately, because large molecules often contain multiple instances of the same atoms, bonds, and functional groups, we expect ERI to be sparse and compressible, i.e. the information contained is much more compact compared to the total number of elements. By adopting ideas from signal processing, tensor hyper-contraction shows that ERI can be accurately reconstructed from a small set of sampling points. This allows ERI to be approximated with product of second order matrices, leading to great savings in both memory and operational cost.
Multi-reference effects, also called strong or static electron correlation, are important for describing bond breaking and conical intersections in photochemistry. However, representation of multi-reference effects in the canonical orbitals are often very complicated. Inspired by the properties of Kronecker sum, I showed that multi-reference perturbation theory can be reformulated in an exact but much simpler way if represented in "dressed orbitals" instead of the commonly used canonical orbitals. I'm interested in studying whether this idea could be extended to other multi-reference approaches beyond perturbation theory.
Graphical processing units (GPUs) are specialized hardware for graphical renderings. Because video games and graphics renderings rely on simulations of mechanics and optics based on physical principles, it is not surprising that GPUs can also be used for general scientific simulations. Due to its fundamental differences from traditional processors, GPUs are best suited for executing a large group of similar tasks with minimal communication, and require completely different code design strategies. All of my methods have been optimized for GPUs, and it is now possible to use a single desktop workstation for calculations that used to require supercomputers. Such experience also prepares me for adapting quantum chemistry methods to other types of specialized novel hardware that may come out in the future.