Mathematics and Natural Sciences

Letian Dou, PhD

PROFESSOR, EMORY COLLEGE OF ARTS AND SCIENCES, CHEMISTRY

Scalable Cyclic Polymer for Universal Compatibilization of Heterogeneous Plastic Waste

Mechanical upcycling offers the lowest energy and carbon footprint among plastic recycling strategies, yet its impact is limited by the immiscibility of mixed polymer waste streams. Real post-consumer plastics contain chemically incompatible polymers, additives, and degradation products that create weak interfaces and lead to down-cycled materials during conventional melt processing. Although compatibilizers can improve interfacial adhesion, current reactive approaches alter polymer chemistry and hinder recyclability, while non-reactive systems typically require composition-specific designs that lack universality. This proposal introduces a fundamentally different approach: cyclic polymers as universal, non-reactive, topology-driven compatibilizers for heterogeneous plastic waste. By eliminating chain ends, cyclic polymers exhibit unique interfacial dynamics that physically constrain immiscible polymer phases through a topology-enabled “bundle effect,” stabilizing interfaces without covalent bonding or precise chemical matching. Using a green, additive-free photo–melt bulk polymerization platform, this work will develop scalable routes to topology-pure cyclic polymers and systematically study how ring size, side-chain architecture, and processing conditions govern compatibilization efficiency. The project integrates synthesis, processing, and performance evaluation using realistic post-consumer plastics, including PE, iPP, PET, and PS. Mechanical, morphological, and recyclability studies will establish structure–topology–property relationships and assess the recoverability of cyclic compatibilizers. Together, these efforts will position polymer topology as a new design principle for universal compatibilization and enable scalable, circular strategies for high-value mechanical upcycling of plastic waste.

Michael Heaven, PhD

PROFESSOR, EMORY COLLEGE OF ARTS AND SCIENCES, CHEMISTRY

Spectroscopy of BaNH2 for Applications to Searches of Physics Beyond the Standard Model

Complementary to the progress of high-energy physics (HEP), atomic, molecular, and optical (AMO) physics has been and continues to be at the forefront of efforts in the search for physics Beyond-the-Standard-Model (BSM). As a part of these efforts, a particular class of highly polar molecular species (MX) comprised of a heavy closed-shell metal (M = Sr, Ba, Yb) bound to a monovalent ligand (X = F, OH, NH2, CH3) shows the most promise for high-fidelity quantum state preparation and readout. BaNH2 is a molecule that combines the well-understood advantages of barium-containing compounds bound to monovalent ligands (BaF, BaOH), which are the core test subjects of 4 experimental endeavors underway across the globe in search of the permanent electric dipole moment of the electron (eEDM), and the theoretically proposed advantages of laser coolable polyatomic molecules in the search for BSM physics. As a laser cooling candidate by virtue of potentially possessing highly diagonal transitions, BaNH2 is a tempting molecule to study as the closely space doublet states of opposite parity in the |K| = 1 manifold of the ground state X ̃^2 A_1 offer the possibility of coherence times on the order of minutes, which will have a wide range of applications in not only high-precision measurements in searches for fundamental quantities such as the eEDM but also in the advancement of quantum computing and quantum information science. This work is a spectroscopic characterization and evaluation of the suitability of BaNH2 for laser cooling and experimental investigations of the eEDM.

Youngjoong Kwon, PhD

ASSISTANT PROFESSOR, EMORY COLLEGE OF ARTS AND SCIENCES, COMPUTER SCIENCE

GUM-V: A Physically-Grounded, Urban-Scale World Model for Autonomous Vehicle Simulation

Training Autonomous Vehicles (AVs) in the real world is dangerous and costly. While current Generative AI can create photorealistic videos of driving, these digital hallucinations fail as scientific simulation tools: they lack physical reality, 3D consistency, and urban logic. This interdisciplinary proposal introduces GUM-V, a simulation framework for Autonomous Vehicles bridging Computer Vision and Urban Geographic Science. We utilize our proprietary 1TB Atlanta Dataset to train a novel Urban Semantic Foundation Model, ensuring that generated urban layouts strictly adhere to real-world geospatial zoning rules. Furthermore, we implement a persistent 3D Gaussian Cache serving as a digital long-term memory, which enables the generative model to overcome the forgetting problem and thus generate 3D consistent urban video. Uniquely, this cache transforms fleeting AI images into interactable, collision-ready 3D meshes for physical autonomous vehicle simulation.

Shengpu Tang, PhD

ASSISTANT PROFESSOR, EMORY COLLEGE OF ARTS AND SCIENCES, COMPUTER SCIENCE

RIDE-RL: Robust, Interpretable Clinical Decision Support with Disentangled Dynamics-Aware Phenotyping and Reinforcement Learning

Computational phenotyping distills complex patient data from electronic health records (EHR) into interpretable disease subtypes for supporting various clinical decisions. Despite its appeal, studies have found phenotypes to correlate poorly with treatment response, limiting their utility for guiding therapy. One key reason is that clinical data is a product of patient physiology and clinician behavior, but is analyzed as a purely observational process, leading to conflation of cause and effect. Reinforcement learning (RL) is uniquely suited for this setup as it explicitly reasons about effects of actions (treatments) on an environment (patient’s health), yet contemporary approaches make use of black-box models that lack transparency. Furthermore, these methods still suffer from entanglement, contaminating representations of patient health with the very human imperfections they aim to improve upon. We propose RIDE-RL, a framework for Robust, Interpretable clinical decision support with DisEntangled dynamics-aware phenotyping and Reinforcement Learning. First, we develop a novel causal representation learning framework to disentangle the underlying physiological state from instrumental clinician behavior. Second, the disentangled state representations are transformed into dynamics-aware phenotypes to learn interpretable treatment policies. We explore several approaches including discrete state abstractions and hierarchical mixture-of-experts, as well as iterative phenotype refinement through expectation-maximization. In collaboration with Drs. Siva Bhavani (SOM) and Gloria Kwak (SON), we will validate RIDE-RL using EHR data on sepsis from multiple hospitals. These results will enable us to seek ex

Marco Tezzele, PhD

ASSISTANT PROFESSOR, EMORY COLLEGE OF ARTS AND SCIENCES, MATHEMATICS 

Scalable Operator Inference via Randomized Numerical Linear Algebra

Operator Inference (OpInf) is a non-intrusive reduced order modeling approach that incorporates physics by defining a structured polynomial form for the reduced model, subsequently learning the corresponding operators from simulated training data. While OpInf excels in low-dimensional regimes, it faces a critical scalability barrier when applied to transport-dominated phenomena, such as advection-diffusion-reaction systems, which require high-dimensional reduced spaces ($r > 100$) and cubic nonlinearities. In this regime, the combinatorial explosion of polynomial features creates a "Memory Wall", rendering standard least-squares solutions computationally intractable. The overall objective of this proposal is to formulate and validate a Randomized Numerical Linear Algebra (RandNLA)-OpInf framework that circumvents the explicit formation of high-order polynomial feature matrices. We aim to enable the inference of stable, cubic reduced-order models by utilizing structural sketching (TensorSketch) for implicit feature embedding and fast exact leverage score sampling via Hadamard identities. This approach effectively reduces "Big Data" to "Smart Data", prioritizing snapshots that capture unique nonlinear dynamics over redundant transport modes. The PI will work with a graduate student to derive theoretical error bounds, implement the algorithms in the open-source opinf software package, and validate the methodology on the transport-augmented Gray-Scott model. This project will establish a scalable foundation for next-generation digital twins in complex multi-physics applications.

Feng Zhai, PhD

ASSISTANT PROFESSOR, EMORY COLLEGE OF ARTS AND SCIENCES, CHEMISTRY

Bridging Copper(I) Homogeneous and Heterogeneous Catalysis through Cooperative Ligand Design

Ligand design is a central strategy for controlling metal nuclearity and reactivity in organometallic catalysis. While most homogeneous catalytic systems rely on well-defined mononuclear metal complexes, increasing metal nuclearity can enable cooperative effects that enhance activity, selectivity, and access to new reaction pathways. Bridging the mechanistic knowledge gap between low-nuclearity homogeneous catalysis and high-nuclearity heterogeneous catalysis through rational ligand design remains a fundamental challenge. Copper(I) catalysis offers a compelling platform to address this challenge. In particular, cyclic (alkyl)(amino)carbenes (CAACs) have emerged as powerful ligands that form exceptionally robust Cu(I) complexes and enable reactivity distinct from that supported by phosphines and conventional N-heterocyclic carbenes (NHCs). Despite these advantages, the ability of CAAC ligands to control Cu nuclearity, especially in binuclear and multinuclear complexes, remains underexplored. This project aims to establish fundamental principles governing CAAC-supported Cu nuclearity control through the development of a new semi-rigid bis(CAAC) ligand designed to favor metal–metal proximity while disfavoring intramolecular chelation. We will synthesize and structurally characterize bis(CAAC)-supported binuclear Cu(I) complexes, evaluate the impact of Cu–Cu cooperation in homogeneous catalytic reactions, and extend this ligand framework to the construction of higher-nuclearity Cu clusters. By correlating ligand architecture, metal nuclearity, and catalytic performance, this work will generate new design strategies for Cu-based catalysts and provide mechanistic insight into cooperative metal reactivity. The proposed research will advance fundamental organometallic chemistry and lay the groundwork for sustainable catalytic systems based on earth-abundant metals.