Undergraduate Research Summer Showcase

Follow

• 

  A docking Analysis of reactivators for sarin-inhibited acetylcholinestrase can aid in screening candidate compounds

  Ryan Pirger and Steven Gwaltney

  Poster created as part of the Center for Computational Sciences' Research Experiences for Undergraduates in Computational Methods with Applications in Materials Science and presented at the 2024 Undergraduate Research Showcase.

  Organophosphate (OP) poisoning disrupts nerve signaling by inhibiting acetylcholinesterase (AChE), an essential enzyme. Current oxime-based treatments for OP poisoning lack efficacy in severe cases due to blood-brain barrier (BBB) impermeability. A prior machine learning study identified promising AChE reactivators with BBB permeability and synthetic feasibility. Our study employed in silico docking simulations using AutoDock to evaluate the interactions between these potential reactivators and a sarin-inhibited human AChE model. The 35 compounds proposed in the earlier study, along with five known good AChE reactivators, were docked against a model of human AChE that was inhibited by the OP nerve agent sarin. While all the positive control molecules yielded good docking results, none of the newly proposed compounds did. The results of this docking analysis will inform the development of novel OP antidotes capable of reaching the brain and effectively reactivating AChE in sever poisoning scenarios. We suggest adding a docking screening to any future protocol designed to generate potential reactivators.

• 

  Continuous Data Assimilation for Two-Phase Flow

  Mary H. Graveman and Amanda E. Diegel

  Poster created as part of the Center for Computational Sciences' Research Experiences for Undergraduates in Computational Methods with Applications in Materials Science and presented at the 2024 Undergraduate Research Showcase.

  We propose a numerical approximation method modeling two-phase flow (via the Cahn-Hilliard equation) that incorporates data to achieve long-time accuracy. The underlying numerical method utilizes the Galerkin finite element method for spatial discretization and a method known as continuous data assimilation to incorporate the known data. We demonstrate the method is long-time stable and long-time accurate provided enough data measurements are incorporated into the simulation, overcoming possibly inaccurate initial conditions. Numerical experiments illustrate the effectiveness of the method on a benchmark test problem. All computations are completed in MATLAB.

• 

  Developing Machine Learned Interatomic Potentials for Hydrogen Dissociation over Molybdenum Phosphide

  Jeremy Lugo, John Michael Lane, Woodrow N. Wilson, and Neeraj Rai

  Poster created as part of the Center for Computational Sciences' Research Experiences for Undergraduates in Computational Methods with Applications in Materials Science and presented at the 2024 Undergraduate Research Showcase.

  Molybdenum phosphides are promising catalysts for biomass conversion, particularly in the hydrogenation of organic molecules. Hydrogen dissociation on the surface of such catalysts is a critical step in the process. Traditional computational methods such as density functional theory (DFT) can provide valuable insights into hydrogen-catalyst interactions but are limited by their high computational cost and inefficiency for large systems and long time scales. Machine learning has emerged as a solution to these limitations. Machine learned interatomic potentials (MLIPs), trained on ab initio molecular dynamics (AIMD) simulations, enable the development of molecular dynamics (MD) for much larger systems over much longer time scales, with significantly reduced computational cost. This allows more complex interactions and rare reactive events to be observed that would otherwise be inaccessible with traditional methods. In this study, MACE, a message passing neural network, is employed to develop MLIPs for the dissociation of hydrogen on molybdenum phosphide surfaces. Utilizing MACE results in highly accurate MD simulations with time scales on the order of 1,000 times longer than those achievable with AIMD. These MD simulations provide key insights about the system, including preferred adsorption sites and atomic interactions and serve as a bridge to studying biomass conversion. This study demonstrates the efficacy of machine learning for catalytic research, illustrating the robust and efficient predictive capabilities of MLIPs and underscoring the potential of machine learning techniques for the efficient discovery, optimization, and understanding of new catalysts.

• 

  Development of a Simulation Tool to Study Secondary Electron Emission in Copper Coated with Graphene

  George Vassilakopoulos and Eric Collins

  Poster created as part of the Center for Computational Sciences' Research Experiences for Undergraduates in Computational Methods with Applications in Materials Science and presented at the 2024 Undergraduate Research Showcase.

  Secondary electron emission (SEE) arises from the interaction of high-energy particles with metallic surface materials. The emission of electrons from copper surfaces can be problematic in a variety of applications, such as particle accelerators or microchips operating in spacecraft. Emitted electrons can build up in accelerator walls, reducing the accuracy of the experiment. SEE in microchips can flip logical bits, create unwanted heat, and reduce the device's lifespan. In recent years, graphene has been proposed as a potential moderator to prevent or mitigate these effects. Graphene can be applied in layers on top of a copper substrate to reduce the secondary electron yield (SEY): the ratio of emitted secondary electrons to total incident electrons. The focus of this investigation has been to simulate the SEE process in copper and to obtain its SEY at various energies of incident primary electrons. Monte Carlo methods, such as the continuous-slowing-down approximation and numerical integration schemes, have been used to generate and track secondary electrons within the copper. The simulation generates secondary electrons from primary electron interactions consistent with existing material models. The positions and energy losses of the secondary electrons are updated using Mott's cross-section formula and the stopping power, respectively. Electrons’ paths are then propagated until their energy falls below a prescribed threshold or the electron is emitted from the surface. Simulations have successfully reproduced SEY results for copper with and without graphene surface layers for an incident electron energy range of 60 eV to 1000 eV. The results show that the SEY of the copper with graphene layers is reduced by 20% compared to the SEY of pure copper.

• 

  New Weighting Parameters for Non-Local Means based Denoising Algorithm

  Ely Carroll and Hyeona Lim

  Poster created as part of the Center for Computational Sciences' Research Experiences for Undergraduates in Computational Methods with Applications in Materials Science and presented at the 2024 Undergraduate Research Showcase.

  Current image denoising techniques can be broadly categorized into two types: variational methods which use minimizing functionals, and filtering methods via pixel averaging. Specifically, the non-local means (NLM) filtering based image denoising algorithm involves replacing each pixel value by the weighted averages of all pixels in the entire image. The basic non-local means algorithm is predicated on manually inputting a parameter to determine the weight of each pixel. We introduce and analyze a new method of implementing the weight factors, based on small variances in the pixel values to identify noise. The new method will identify newly classified noisy pixels, then while comparing blocks of pixels for similarity it is possible to disregard noisy pixels and gain a clearer comparison of structure for weighting. The new method is numerically tested and compared to conventional NLM based methods via the peak signal to noise ratio (PSNR), and visual comparison. The results show that the new method is preferable to the current NLM methods.

• 

  Repulsive Coulomb interactions enhance Superconductivity Selectively at Density 0.5 per site

  Jeremy Padvorac and Rudolf Torsten Clay

  Poster created as part of the Center for Computational Sciences' Research Experiences for Undergraduates in Computational Methods with Applications in Materials Science and presented at the 2024 Undergraduate Research Showcase.

  The globally accepted Bardeen-Cooper-Schrieffer (BCS) theory explains the pairing mechanism in low-temperature elemental superconductors but fails to explain pairing in high critical temperature superconductors. Unconventional superconductors with higher critical temperatures require moreexotic pairing mechanisms considering electron-electron interactions, such as spin-fluctuated-mediated pairing. Many-body calculations within the Hubbard model, which includes short-range electron-electron repulsion, have shown that superconducting pair correlations are enhanced when the density of carriers is close to one-half per orbital, a characteristic density in many unconventional superconductors. However, questions related to the accuracy of previous calculations remain. In this work, we employed a more accurate numerical method, Self-Consistent Constrained Path Quantum Monte Carlo (SC-CPMC), to check the accuracy of CPMC calculations without the self-consistent optimization of the trial wavefunction. By eliminating the artificial lattice symmetry breaking previously used, we restored the intrinsic symmetry of the lattice. This study also investigated effects of a larger Hubbard U. Our results confirm that the Hubbard U significantly enhancessuperconducting pair-pair correlations for carrier densities close to 0.5 per site.

• 

  Validation and Runtime Improvements in Neural Networks for Interatomic Potentials via Automatic Fingerprint Selection

  Spencer Evans-Cole, Kip Barrett, and Doyl Dickel

  Poster created as part of the Center for Computational Sciences' Research Experiences for Undergraduates in Computational Methods with Applications in Materials Science and presented at the 2024 Undergraduate Research Showcase.

  The development of interatomic potentials via machine learning on first principles results is driven by the need to accurately and quickly analyze atomic environments. When encoding atomic environments for neural networks, a series of different two and three-body functions are used to transcribe the local environments to produce a vector of inputs, making it imperative that the local environments are encoded optimally. Given a set of atomic environments, we generate a series of different fingerprints and by analysis of the data select the most optimal ones which demonstrate a reduced validation error and computation time in our model. To select the best fingerprints for the network, we first construct a large dense matrix A, where each row is an atomic environment and each column is a fingerprint with a different set of parameters. We look to create the most optimal rank k approximation of A, by factoring it as A = CUR where C is made up of k columns of A, R is made up of k rows of A and U is square and of rank no greater than k. We employ the singular value decomposition followed by two rounds of sampling the column space of our matrix, where we utilize weights based on spectral properties and then adaptively compute leverage scores to select fingerprints that were not initially chosen and that require a low computational cost with high probability. By computing the matrices C and R we find the optimal representation of our fingerprint and training data respectively. Our optimal fingerprints achieve on average a 39% reduction in validation error with comparable training errors to naive fingerprint selections for unary and binary systems, while minimizing the computational cost for fingerprint generation.