Research
Our group works on developing new theoretical and computational algorithms to understand challenging chemical and physical systems. Our work includes the development of new theories to simulate strongly correlated systems, efficient implementation of these theories and simulation of challenging gas and condensed phase systems. Below is a summary of current projects in the group. If you are interested in joining our team, please send an email to [email protected]
Current projects
1. Auxiliary field quantum Monte Carlo (AFQMC)
AFQMC has emerged as a highly accurate method for treating the electronic structure of both molecules and solids. In our group, we are working to transform AFQMC from a promising method into a broadly applicable computational framework for chemistry. We have recently shown that, with accurate trial states, AFQMC can surpass CCSD(T), long regarded as the gold standard for main-group chemistry, in accuracy. Initial tests further indicate that AFQMC holds strong potential for systems containing transition-metal clusters.
We are now developing AFQMC into a comprehensive computational tool by introducing efficient and accurate algorithmic variants, analytic gradients, local-correlation extensions, and excited-state formalisms.
Despite its long history, several fundamental aspects of AFQMC remain poorly understood: the precise nature of its bias, the limits under which this bias vanishes, and the statistical distribution being sampled. To address these questions, we are developing analytic techniques that probe the stochastic drift and diffusion of AFQMC's random walkers.
2. Efficient basis for electronic structure
Gaussian-type orbitals (GTOs) are the most widely used basis functions in quantum chemistry. They adapt to the molecular geometry and usually achieve rapid convergence toward the basis set limit with relatively few functions. However, for large systems they become inefficient because two-electron integrals scale as four-index tensors.
Plane waves, more common in periodic systems, avoid this issue: the two-electron integrals can be expressed as two-index quantities, reducing asymptotic scaling. Yet plane waves are often less efficient for smaller systems since they are not adapted to the molecular geometry.
Our current work focuses on combining the advantages of both approaches: basis functions that (i) adapt to the input geometry, allowing fewer functions, and (ii) retain a "grid-like" structure so that two-electron integrals are two-index quantities, as in plane waves. These ideas connect naturally to emerging tensor factorization techniques such as tensor hypercontraction (THC).
3. Correlated wavefunctions for condensed phase systems
It has long been recognized that 3d transition-metal materials host exotic phases of matter arising from the delicate interplay between electron correlation and electron delocalization. More recently, attention has shifted to 5d transition-metal systems, where strong spin–orbit coupling adds a new layer of complexity and provides a rich playground for designing materials with tailored properties.
One striking example is Ba₄Ir₃O₁₀ which was recently synthesize by Gang Cao's group. Ba₄Ir₃O₁₀ appears to exhibit an unusual spin-liquid ground state. The material shows strong nearest-neighbor exchange coupling J, yet its Néel temperature is anomalously low, a classic signature of frustration. The microscopic origin of this spin frustration, however, remains unknown.
Our group is developing methodologies to bring correlated wavefunction techniques that are traditionally used to understand spin interactions in molecular systems into the study of Iridates. Our goal is to uncover the mechanisms behind their frustrated magnetism.
4. Ab initio elementary reaction networks aided by machine learning
Chemical transformations can be broken down into constituent steps, giving rise to elementary reaction mechanisms. These mechanisms are powerful tools for predicting the fate of chemical reactions and, in principle, can be formulated without empirical parameters fitted to experiment.
However, constructing such mechanisms by hand is extremely challenging and time-consuming. Heuristic rules derived from large datasets and more recently machine learning (ML) approaches, have automated many steps that were once performed manually. While this has accelerated mechanism generation, it has also introduced empirical parameters, reducing first-principles rigor.
The goal of this project is to combine the strengths of both approaches: using ML methods to provide a first pass at mechanism construction, followed by high-accuracy ab initio quantum chemistry and state-of-the-art statistical mechanical theories (e.g., transition state theory) to deliver reliable, parameter-free energies and reaction rates. Our ultimate aim is to generate elementary mechanisms in a fully automated, ab initio manner, with little to no user intervention.
5. Proton coupled electron transfer in Cobaltocene
Peters lab has recently introduced a cobaltocene-tethered Bronsted base dimethylanilinium mediators that allows the reduction of several substrates including nitrogen to ammonia in electrochemical setting. It has several advantages including, it prevents electrode poisoning, these mediators can reduce an organic substrate via a concerted PCET (proton coupled electron transfer) requiring lower onset potentials and, the lower onset potential can suppress the competing hydrogen evolving reaction (HER).
While preliminary DFT model studies have been undertaken to understand the nature of the rate-limiting transition state(s) in this PCET-mediated electrocatalysis, pointing to a degree of concertedness, we seek to better understand the intimate pathway by which the electron and the proton are transferred, and how these transfers are gated in space and time (i.e., concerted vs stepwise, through bond vs through solvent). This work will hopefully provide insights into improved PCET-mediator design and ultimately more efficient electrocatalytic systems.