Academic Year 2025–2026 Awarded Grants
Andrew Kai: Assessments of Modern Density Functional Approximations for Fullerenes Using Turbomole on WAVE HPC
Project Overview: This project benchmarks various Density Functional Theory (DFT) methods, specifically local hybrid and range-separated local hybrid functionals, to determine their accuracy in calculating isomerization energies for fullerenes. By leveraging the WAVE HPC to perform over 2,000 complex calculations, the research seeks to identify optimal functionals that balance computational cost with precision. These findings will provide theoretical insights into improving DFT models and support a larger-scale screening of fullerene chemical space for applications in nanotechnology and medicine.
Cole Mitchell: Simulation of UV-Vis Spectra of Transition Metal Complexes with Modern-Day Density Functional Approximations
Project Overview: This research focuses on benchmarking advanced computational functionals within the Time-Dependent Density Functional Theory (TD-DFT) framework to accurately model the UV-Vis spectra of transition metal complexes. The project utilizes the WAVE HPC to handle the high memory and bandwidth requirements of these simulations, aiming to identify the optimal number of energy states needed for accurate spectral representation. Additionally, the researcher will develop an open-source Python script to automate the conversion of discrete transition data into continuous spectra, ultimately contributing to more efficient and sustainable use of high-performance computing resources.
Karina Martinez: Using Scat DNA Samples to Unlock Bird Diet Mysteries and Evaluate Restoration Success in the SF Bay Estuary
Project Overview: This project utilizes "molecular scatology" to analyze bird DNA from scat samples to evaluate the success of habitat restoration in the San Francisco Bay Estuary. By identifying the specific plant species birds consume, the researcher aims to determine if they are utilizing restored native plants or persistent non-native species. The transition to WAVE HPC is essential for this work, as current local systems crash when attempting to process the terabytes of DNA sequence data generated; the HPC will allow for more accurate plant classification through parallel processing.
Neel Mukkavilli: Interpretability of ProteinMPNN via Sparse Autoencoders
Project Overview: This research aims to interpret the inner workings of ProteinMPNN, a neural network used for protein sequence design, by applying sparse autoencoders to identify the features the model learns regarding protein stability. The project involves expanding the neural network’s hidden dimensions to translate complex data into interpretable features, such as the presence of specific amino acids. Leveraging the WAVE HPC will enable the training of these autoencoders on massive datasets—ranging from hundreds of gigabytes to terabytes—providing a novel look into message-passing neural networks.
Sean Wu: Traffic Digital Twin via Scalable Simulation and Machine Learning
Project Overview: This project focuses on the creation of a "traffic digital twin" to analyze and optimize urban mobility through scalable simulation. By utilizing the WAVE HPC to run intensive reinforcement learning algorithms, the research develops more efficient traffic signal control strategies designed to reduce congestion and improve city-wide flow. This computational approach allows for the modeling of highly complex urban environments that would be impossible to process effectively on standard desktop systems.