Schedule

8:30am - Opening remarks

9:00am - Daniel Foti, mechanical engineering, "Understanding large-scale turbulent flows through computational fluid dynamics"

9:40am - Jonathan Epih, chemistry, "Computational Investigation of GNMT-Catalyzed Methyl Transfer Reaction: Integrating MD, QM, and ML Approaches"

10:00am - Break

10:20am - Jordan Darling, physics, "Effective Repulsive Interaction between Janus Polymer-Grafted Nanoparticles Adhering to Lipid Vesicles"

11:00am - McKenna Lab, biology, "The Evolution & Genomic Basis of Insect Diversity"

11:20am - Pankaj Jain, finance, "Augmented Intelligence using Big Blue: Analyzing On and Off Exchange Trading"

12:00pm - Lunch

1:00pm - Ranganathan Gopalakrishnan, mechanical engineering, "Investigations of particle charging in aerosols and plasmas"

1:40pm - Amirmehdi Mirshahvalad, finance, "Using HPC for Finance Research"

2:00pm - Muhammed Dada, chemistry, "Simulating resonant inelastic x-ray scattering of some model of Ruthenium complexes using time-dependent density functional theory"

2:20pm - Break

2:40pm - Christos Kyriakopoulos, CERI, "From Simulation to Education: Benefits of HPC in Earthquake Research and Outreach"

3:20pm - Kuruvitage Chameera Chathuranga Silva, CERI, "Utilizing HPC for Geodynamic Research on Continental Rifting"

3:40pm - Benjamin Keller, physics and materials science, "What Matters Most in Galaxy Formation?"

4:30pm - Wrap up

Daniel Foti, mechanical engineering, "Understanding large-scale turbulent flows through computational fluid dynamics"

Turbulence is an important aspect of a wide range of natural and industrial flows, including atmospheric and geophysical flows, biology, aviation, power generation, and hydrodynamics.  Despite its prominence, analytical and general solutions to the chaotic yet deterministic phenomenon have eluded engineers.  However, with the advent of high-performance computing, computational fluid dynamics has played a large role in elucidating details of turbulent flows.   In this seminar, I will discuss how numerical and massively parallel computational methods, including recent developments in machine learning, have been used to simulate large-scale, high Reynolds number engineering flows.  I will focus on large-eddy simulation, a powerful tool that leverages the properties of turbulence, to understand the effects of turbulence on wind energy production and cavitation of the flow over a hydrofoil.   In the former, large-scale meandering of the wake behind a wind turbine affects the power variability of downwind turbines.  In the latter, the shedding and collapse of sheet-to-cloud cavitation can cause surface erosion.  These seemingly very different flows share many aspects in common related to turbulence and require substantial computational power to investigate with sufficient resolution. 

Jonathan Epih, chemistry, "Computational Investigation of GNMT-Catalyzed Methyl Transfer Reaction: Integrating MD, QM, and ML Approaches"

This study employs computational approaches including molecular dynamics (MD) simulation, quantum mechanics (QM), and machine learning (ML) to investigate the enzymatic reaction catalyzed by glycine N-methyltransferase (GNMT), which involves the transfer of a methyl group from S-adenosylmethionine (SAM) to glycine, producing S-adenosylhomocysteine (SAH) and sarcosine. The reaction was studied using QM calculations on models that represent the enzyme's active site. These models were derived from an X-ray crystal structure (PDB:1NBH),1 varied in size, and a converged activation free energy of approximately 9.5 kcal/mol was reported, while experimental findings suggest an activation energy of about 17.5 kcal/mol for the reaction.2

To account for protein flexibility and avoid relying solely on a single, static structure, multiple MD simulations were performed to better sample the conformational space of the protein-substrate complex. Selected MD frames were then used to conduct Residue Interaction Network (RIN) analysis, allowing us to rank key residues interacting with the substrate and cofactor. This approach provides a more comprehensive view of the interactions involved in the reaction. With more relaxed and diverse structures, our QM and ML computations provide a broader view of the reaction energetics and contribute to a better understanding of the crucial residues involved in this enzymatic process.

Pankaj Jain, finance, "Augmented Intelligence using Big Blue: Analyzing On and Off Exchange Trading"

Using high performance computing clusters, we examine 202 terabytes of compressed nansecond timestamped trades and quotes (TAQ dataset) to discover intraday and trade-by-trade factors associated with the big rise in off-exchange trading volume, information share, and price discovery from 2008 to 2022. Off-exchange trading has nearly doubled both in terms of volume and information share in the last two decades to 41.62\% and 18.67\%, respectively. In trade-by-trade analysis after informed options trades, we observe a dramatic jump in the probability from about 40\% to about 60\% that the next stock trade occurs off-exchange. We theoretically model and empirically identify informed traders’ cost-based choice of off-exchange stealth trading venues, where they can avoid high predicted quoted stock spreads (forecasted by options activity). Our results sharply contrast the conventional wisdom about off-exchange volume being largely uninformed. Informed traders use low-cost off-exchange venues around earnings announcements even more intensely, particularly in light of wider spreads in those periods. This relation persists despite headwinds from option expiration settlement rules that drive up on-exchange volume. Our findings hold across traditional and machine learning methods for identifying informed trader intensity and surge around earnings announcements. Off-exchange information has important regulatory implications as it predicts higher subsequent stock price volatility. Our results are strongest in the most recent sample from 2020-22. This pattern implies that the immediacy of off-exchange has improved sufficiently in recent years to levels acceptable to informed traders who want to strategically exploit the low cost stealth trading feature of off-exchange dark pools.

Amirmehdi Mirshahvalad, finance, "Using HPC for Finance Research"

Muhammed Dada, chemistry
Abstract:
Resonant inelastic x-ray scattering (RIXS) is an advanced two-photon spectroscopy technique, in which the
energy of an incident photon is tuned to coincide with one of the x-ray atomic transitions of the system.
RIXS is fast becoming a key instrument to probe the electronic and structural properties of samples in solid,
liquid, and gaseous form. The RIXS process is theoretically described by the Kramers-Heisenberg equation,
which requires the knowledge of excited-state transition moments and energies. Nonetheless, evaluating
excited-state transition dipole moments in large transition metal complexes with high-accuracy quantum
chemical methods is expensive and oftentimes impractical.

In this work, we demonstrate that time-dependent density functional theory (TD-DFT) method offers a
practical alternative for simulating RIXS maps in transition metal complexes. By employing a reduced-cost
approach based on the unrelaxed second-order density polarization obtained from linear-response TD-DFT
computations, along with relativistic corrections as a perturbation via the zeroth-order regular
approximation (ZORA), we successfully simulate the 3p4d RIXS maps of Ru model complexes, including
[RuIII(NH3)6]3+, [RuII(bpy)3]2+, [RuII(CN)6]4-. This method bypasses the need for expensive quadratic
response calculations or relativistic equations. These results illustrate that TDDFT-based RIXS simulations
can be applied to efficiently guide and interpret experiments being currently done at state-of-the art light-
source facilities, such as LCLS-II

Benjamin Keller, physics and materials science - "What Matters Most in Galaxy Formation?"
Abstract:
Simulations of galaxy formation offer us the ability to view the multi-billion year evolution of galaxies, but like any simulation we face a trade-off.  Do we simulate a small number of galaxies at high resolution, or a large number of galaxies at lower resolution?  In the former case, we can better capture the internal processes of galaxies, but in the latter we can better measure the statistical properties of galaxy populations.  In this talk, I will introduce you to the Memphis Galaxy Simulation group in the Department of Physics and Materials Science, and show how we are working to find a compromise to this problem with a combination of cutting-edge simulations, interpretable machine learning, and new tools for building simulation initial conditions.