Student Posters Repository

Electron correlations in chemical systems give rise to a wide range of interesting physical properties. Although traditional mean-field quantum chemical algorithms can reliably calculate ground state observables, finite temperature and spectral properties are only accessible with explicit inclusion of electron correlations. Diagrammatic Monte Carlo (DiagMC), which expands the physical observable in terms of connected Feynman diagrams and samples the resulting series stochastically, is a powerful technique for studying electron correlations and does not suffer from numerical sign problem which worsens with increasing system size. Recent developments in DiagMC algorithms have greatly improved their numerical efficiency. In this poster, we aim to introduce our DiagMC implementation for multi-orbital systems, and discuss the difficulties and potentials when it is applied to realistic molecular systems.

Understanding the processes that drive galaxy formation and shape the observed properties of galaxies is an important and challenging frontier problem of modern astrophysics. The role of cosmic rays (CRs), high-energy particles accelerated in supernovae, in the evolution of galaxies has recently begun to receive significant attention due to the realization that they can efficiently accelerate galactic winds and heat the interstellar medium. Simulations of the interactions between gas, stars, and CRs in a galaxy help probe these phenomena in a way that is otherwise difficult. Recent improvements in computational power and algorithm efficiencies have made it possible to move beyond purely hydrodynamic simulations of the galactic environment, to magnetohydrodynamic (MHD) simulations. Modeling CR transport through the galaxy requires simplifying assumptions and remains an open research question. The central objective of our research is to significantly increase the realism of these simulations. We will conduct three-dimensional coupled cosmic-ray MHD simulations of a galactic disk in order to identify key CR physics that could influence the evolution of a galaxy.

Performing incompressible multiphase fluid dynamics simulations requires immense computational power. However, these simulations are vital for gaining a deeper understanding of many interesting scientific processes. Improving the efficiency of fuel injection systems for instance, is crucial to the advancement of many engineering devices. For example, most engines rely on atomization, the breakup of fuel into droplets, to perform combustion. Fuel atomization is a widely studied phenomena within the field of fluid dynamics. However, current simulations that use direct numerical simulations are limited to lower Reynolds and Weber numbers than common fuel injection operating conditions due to the associated computational costs. The computation required for such simulations uses high performance computing resources with more than 10,000 processors running constantly for multiple weeks. The approximation of surface tension forces play an important role in the numerical calculation of these multiphase flow problems. This research explores novel numerical methods for computing the surface tension force and implementing these methods into highly parallel frameworks.

Observing planet formation in detail is hard, since planets are small, dark and only form far away from earth. The formation of smaller objects like asteroids poses an even bigger challenge. We try to shed light on the gasless stage of planet formation (when there is so much dust in a given region, that gas influence becomes neglectable) using a GPU based Smooth Particle Hydrodynamics (SPH) simulation. While SPH is commonly used for fluid simulation like gases and liquids, at it's core we find a method for solving partial differential equations that can be extended to simulate solids, deformable objects and granular materials. The flexible CUDA implementation allows for test runs at interactive speeds even on a laptop computer, as well as slower high accuracy simulations on workstations and supercomputing nodes that feature more powerful GPUs. While test runs show promising results, we are currently working to increase the supported particle number as well the stability of the used algorithms. Along with accelerating the code further, that will allow us to simulate larger time frames and explore object formation in the outer solar system. We hope to extend the model in the future to also handle gas-solid-interaction.

We study the early growth of supermassive black holes using constrained realization cosmological simulations. The technical part is twofold: the initial condition generator and the hydrodynamic solver. We use an initial condition generator parallelized in MPI to generate constrained Gaussian realizations to feed in the hydrodynamic solver. The hydrodynamic solver we use for the cosmological simulations is a smooth particle hydrodynamics (SPH) based code also parallelized with MPI. 

Structure-based models (SBM) allow milliseconds effective timescale simulations of large biomolecular assemblies.  In an SBM, the prevalent potential energy minima are defined to correspond to native structures. Here, we have developed an all-atom SBM in which non-hydrogen atoms and diffuse ions (K⁺, Cl-, Mg²⁺) electrostatics are described explicitly. The effective desolvation potentials for ion-ion and ion-RNA interactions are introduced such that solvation is treated implicitly. Excluded volume term and long range Coulomb interaction are also included in the forcefield. To calibrate the model, we simulate RNA Helix 44 and compare various observable quantities with explicit-solvent simulations. Using an iterative refinement protocol, the parameters are optimized to reproduce the potential of mean force obtained from two-microsecond explicit-solvent simulations, until the parameters converge iteratively. This transferable forcefield for ion-ion and ion-RNA interactions are then used in SBM of large assemblies. In an initial application of this model, we show how diffuse ions and the excess ion atmosphere can modulate large-scale conformation dynamics in the ribosome. This opens the door to investigate the role of ions in a wide range of RNA-protein biomolecular assemblies.

Studying, understanding and predicting the effects of plasma turbulence in the Scrape-Off layer of a plasma device is of highest importance for future fusion power plants. Our goal is to study turbulence with the gyrokinetic code GENE. In order to apply GENE successfully in the Scrape-Off layer multiple adaptions to the code are necessary. 

The magnetic geometry in the Scrape-Off layer of a fusion device is numerically challenging. In order to meet these challenges new data structures and algorithms need to be developed and implemented. We present the software layout for a new version of the GENE code which has the potential to run at exascale and simulate turbulence in the Scrape-Off layer of fusion power plants in the future.

White matter refers to the myelin that covers axons in the brain and helps regulate conduction velocity in the transmission of neural signals. It plays a vital role in neural communication and maintaining neural synchrony across brain areas; when disrupted it can lead to neurological disorders and cognitive deficits. In this research, I study the role of myelin in shaping human brain dynamics through computational simulation of biophysical neural network models based on anatomical connectome data. I present a computational pipeline for estimating certain network characteristics by combining neural population activity measurements such as EEG/MEG, neural modeling and an evolutionary optimization algorithm. The goal is to find the optimal parameters of the network model to replicate the brain activity measured in vivo. Thus, the physiological representation of the parameters will give insight on the white matter structure in the clinical participants’ brain networks. For optimization of my model parameters, I use a modified version of the differential evolution algorithm. Traditional machine learning approaches are not employable because my model lacks a gradient. The search space is very large due to the high dimensionality of the problem. As such, I parallelize the searches using the MPI framework (mpi4py for Python) to decrease computation time and increase the breadth of search. I have applied this approach to estimate network weights and also isolate conduction velocities along myelin tracts in the network model. Currently I am working on estimating temporal delays in interactions of brain regions. The use of this computational pipeline has the potential to be useful for understanding various neural disorders such as Alzheimer’s, and be used as a diagnostic tool.