Student Posters Repository

We introduce our in-house finite-volume CFD code MGLET (Multi Grid Large Eddy Turbulence), and present the results of our recent performance optimisation efforts ongoing at Technical University of Munich (TUM), Norwegian University of Science and Technology (NTNU) and Leibniz Supercomputing Centre (LRZ).

Cacti and succulents have shallow root systems but still can withstand high winds in their natural environment without being uprooted or damaged. Previous experimental and numerical studies showed the ability of grooved cylinders, which mimic cactus shapes, to decrease the drag coefficient and reduce lift force fluctuations over a range of Reynolds numbers. Most of these studies were inspired by the Saguaro cactus, which is native to Mexican and South-Western American deserts, and typically has 10-30 ribs. However, there are a number of cacti and succulents that have only three or four ribs. The aim of this project is to use computational fluid dynamics to study the aerodynamics of cylindrical structures with a low number of ribs. Preliminary 2D URANS simulations were performed and showed that drag coefficient and lift force fluctuations are strongly dependant on the orientation of the cactus-shaped cylinder with respect to the incoming flow. In the next stage scale resolved (Large Eddy and Detached Eddy) simulations will be performed on regional HPC facilities to give insight into 3D flow features around cactus-shaped cylinders with four ribs. In addition, flow statistics will be compared to previous results for the smooth cylinder and cylinders with many ribs.

Galaxies, groups of galaxies and gravitational lensing effects create anisotropies in the cosmic microwave background. These anisotropies are highly non-Gaussian. This work describes our use of the bispectrum, the Fourier space equivalent of the three point function, to study these anisotropies. The bispectrum is a computationally challenging quantity to measure as for modern experiments there are $<10^{20}$ different bispectrum combinations. In the work we present our recent measurements of the bispectrum using the Atacama Cosmology Telescope and the Planck satellite.

Climate change is most likely to impact global food supply and security. Currently, models outcome of future crop productivity is quantitatively diverse because of uncertainties in projected climate change trends and variabilities, human management practices and processes within land surface models. To study impacts of climate change and management practices on agricultural production over the 21st century, we apply the integrated biogeophysical and biogeochemical models, the Integrated Science Assessment Model (ISAM), for row crops (corn, soybean, rice, and wheat). This framework accounts for dynamic crop growth processes with the adaptation of photosynthesis, crop-specific phenology, biomass accumulation, leaf area development and the effects of temperature, light and soil water and nitrogen availability on crop photosynthesis and temperature control on crop phenology and carbon allocation. We implement optimal agricultural management practices including irrigation and nitrogen fertilizer to evaluate the crop yields with interactions with climate and CO2. The study is of importance to consider carbon-temperature-water-nitrogen interactions and helps to understand the uncertainties in the model estimated the potential productivity of food crops under climate change.

Atmospheric river (AR) impacts on hydrologic extremes are greater when an AR event follows closely on the heels of another or several occur in sequence. This study uses coastal Atmospheric River Observatory measurements and Modern-Era Retrospective analysis for Research and Applications, Version 2 reanalysis from water years (WY) 2005 - 2017 to examine 228 landfalling atmospheric river events (ARs) at Bodega Bay (BBY) in Northern California. During the winter of 2016-2017, 34 ARs hit California’s Russian River basin, many in quick succession. (An “AR event” is a period of continuous AR conditions observed at BBY by the ARO.) This provided motivation to develop a definition of “atmospheric river families,” to describe their characteristics and to assess their predictability. An AR family is identified when two AR events are separated by < 120 hours, yet often include more than two AR events in long duration AR families. Using this definition, a catalog of AR families was created for BBY. Out of the 228 AR events observed, 109 (i.e., almost 50%) initiated an AR family. Composites of AR families show general characteristics of these successive storms, which differ significantly from individual events.

Accurately modelling the dynamics of strong-field, high velocity astrophysical phenomena such as binary black hole mergers require large-scale numerical calculations on HPC systems. Conventional numerical relativity codes employ a spatial discretization scheme, which leads to considerable communication overheads at each time step and severely limits their ability to achieve high levels of parallelism possible on modern HPC architectures. A spacetime discretization scheme, with patch boundaries conforming to null or space-like surfaces can significantly reduce this overhead, leading to increased parallel efficiency. As the first step towards evolving the Einstein’s equations in full generality, we implement a spacetime discretization approach to solve the scalar wave equation on a 1+1D conformally compactified spacetime. Our approach to domain decomposition allows us to concurrently compute multiple space-time elements on a single multi-core CPU, with minimal communication between elements, paving the way for future explorations in higher dimensions.

For investigating the microscopic description of non-equilibrium steady states (NESSs), it is important to examine the properties of the NESSs by using concrete models. We discuss temperature gradients in an open and non-uniform spin--1/2 XXZ chain contacting with 2 reservoirs at different temperatures. The non-uniformness is introduced by the random $xx$--coupling given by the Gaussian distribution. The density operator evolves in time according to the Lindblad type quantum master equation. We numerically obtain the NESSs and discuss their temperature profiles. We statistically analyze the relation between the fluctuation of the $xx$--coupling and 1) the temperature difference of the spins in the boundary, 2) the smoothness of the temperature profile. We clarify that there exist crossovers from flat to smooth temperature profile, and from smooth to jumped profile.

Solar cells are devices that convert the solar energy into electrical energy. The efficiency of this conversion depends directly on material defects at the atomic scale such as point defects, dislocations and grain boundaries. These types of defects are hard to study experimentally. Alternatively, computer simulations can be applied to study these defects with atomic scale resolution. Molecular Dynamics (MD) are computer simulations that allow to study the evolution and growth of these defects. MD simulations are computationally rigorous and require High Performance Computers (HPCs) to run. In this presentation, we show how we apply HPCs to run these simulations and the results. 

We consider iterative techniques for the solution of large, sparse interior eigenvalue problems. Several spectral slicing approaches, including polynomial filtering and contour integration approaches have been incorporated into BEAST, a parallel, distributed memory eigensolver framework. This framework builds an approximate subspace for the desired eigenpairs, then uses a Rayleigh-Ritz approach to reduce the eigenproblem to a size which may be solved directly. The iterative nature of the framework may be utilized for various adaptive approaches. For example, the precision or the method used to construct the subspace may be changed between iterations for improvement of performance, robustness, or accuracy. Several such strategies with initial results are shown.

I will be talking about the small number of tools I’m working on at the Modelling and Simulation Group at the University of Rostock. What unites these projects is their multi-paradigm nature. For example we have been successfully combining discrete graph representations with continuous ODE descriptions for cellular models.

These multi-paradigm models frequently allow to bridge gaps across spatio-temporal scales, leading to new and more complex problems becoming tractable. On the other hand they pose a particular challenge to the HPC-Domain, as the workload is usually highly unbalanced and not uniform.