You are here
Researchers in higher mathematics have long used supercomputers to handle the huge numbers of calculations necessary for their work. Professor Doug Arnold (Mathematics; MSI Fellow) is using Itasca to solve problems with various aspects of finite-element methods for partial differential equations. Professor Arnold has developed the finite element exterior calculus (FEEC), and FEniCS (fenicsproject.org), which is one of the most impressive open-source projects in numerical partial differential equations, has been designed so that it is well suited to implementing and testing FEEC-based algorithms. FEniCS, which allows one to implement finite elements from a high-level mathematical perspective, includes a Python/C++ problem solving environment which provides access to advanced solver systems like PETSc, uBLAS, and Trilinos, which in turn use MPI, and so are in a position to make efficient use of Itasca.
Professor Arnold and his group are studying such areas as elastrodynamics and dynamic viscoelastic computations and harmonic forms, among others. The image above shows Stokes flow through a double pipe (in Stokes flow, fluid velocities are very slow, viscosities are very large, or length-scales of the flow are very small). The flow enters through the smaller pipe and exits the larger one. Since starting work on Itasca, the Arnold group has been able to implement FEniCS and tune its performance; they have been getting good results for a number of problems.
The research group of Professor Barry Finzel (Medicinal Chemistry) is developing a centralized database and associated application suite that support searching for complex substructures in macromolecules, storing and sharing user-provided macromolecular data, and presenting search results. Post-Doctoral Associate Dr. Jeff Van Voorst presented a poster about this project at the MSI 2012 Research Exhibition on April 13. The poster was one of the finalists in the competition.
A macromolecule’s structure is represented by all the pairwise distances of the amino acids and nucleotides in the structure. Given the distance representation of a complex substructure, one can search for similar distance patterns; such searching is called distance geometry matching.
The user interface for this database is HTML-based and is generated by a Python web framework (Flask). Having a web interface allows the application to be accessible from anywhere that there is a suitable network connection to the web host. Other reasons for this centralized system include the use of complex software such as the database (e.g. mySQL), the ease of sharing data between two parties, and the relatively large size of the application’s data.
The image above depicts an example of a complex substructure involving an intermolecular interaction and search results. A) Illustrates the local interaction between a single helix of a transcriptional activator and the DNA duplex to which it binds in a known crystal structure. B) Shows the skeleton of backbone atoms (red) used for substructure geometry constraints. C) Shows an ensemble of overlaid substructures that include this motif resulting from a search of all structures in the PDB. D-F) Show three very different examples of the protein:DNA complexes in which this motif can occur. D) Homeodomain transcription factors. E) Holiday Junction replicase. F) Basic region leucine zipper BZIP) proteins.
A traditional website, Drugsite, for this work is already online. Work is ongoing to build the new web-application architecture that will allow distribution of the search process over multiple processors to reduce the search time and return results to any user in seconds.
Aerosol sprays are a major area of study because of the many industrial, agricultural, and pharmaceutical (among others) applications of such work. The research group of Sean Garrick (Mechanical Engineering; MSI Fellow) in the Computational Transport Phenomena Laboratory has a long history in the direct numerical simulation and large eddy simulation of turbulent, reacting, multiphase flows. The researchers develop physical models (and their mathematical representations) and software in-house that can be used to numerically simulate a wide-variety of physical and chemical problems.
Wanjiao Liu, a graduate student in the Garrick group, presented a poster at the MSI 2012 Research Exhibition describing her work in advanced modeling of turbulent sprays. She is studying the size distributions and breakup patterns of droplets in a spray, as these characteristics determine the spray’s performance, efficiency, or safety. Studying these droplets is challenging and the characteristics of sprays and droplets are not completely understood. Ms. Liu’s poster was one of the finalists in the poster competition.
Using various numerical modeling techniques, Ms. Liu is investigating the best way to simulate atomization and spray behavior. The Garrick group runs the highly parallel computer codes necessary for these models on Itasca.
The image shows two views of a turbulent multiphase flow simulation. A liquid column is injected from the left of the domain into gas and starts to break up. The colored contour at the top shows volume of fluid (VOF). VOF equaling 1 means the local space is occupied by liquid, while VOF equaling 0 means the local space is occupied by gas. The black-and-white image at the bottom shows the magnitude of surface tension force for the same flow. In this image, darker color indicates larger surface tension, while white color indicates zero surface tension. Surface tension force acts on the liquid-gas interface, playing a critical role in droplet formation in turbulent sprays.
The ligaments formed due to Rayleigh-Taylor instability (instability induced by two contacting fluids with different densities) are visible, beginning at the left of the image. After formation of these thin, long ligaments, Plateau-Rayleigh instability (instability induced by the effect of surface tension) takes over and small pieces of fluid fragment are pinched off from the jet. Further downstream, flow becomes turbulent. Towards the end of domain, there are coherent structures as well as small-scale droplets and ligaments.
The acoustics of ancient Greek theaters and auditoriums have long fascinated archeologists and historians. Researchers using the LCSE-MSI Visualization Laboratory (LMVL) are creating virtual-reality simulations of ancient structures to determine how variables of architecture design affected the sound, sight lines, and behaviors of speakers and listeners in those spaces. This long-term project focuses specifically on structures used for political and legal oratory from the late Archaic, Classical, and Hellenistic periods (500-100 BCE). The MSI PIs are Professor Richard Graff (Writing Studies) and Daniel Keefe (Computer Science and Engineering) and students Kyungyoon Kim, Bret Jackson, and Lauren Thorson. They are collaborating with Christopher Johnstone and Azadeh Rabbani at Pennsylvania State University.
This project uses MSI resources to achieve three main goals:
- Produce and evaluate accurate virtual reconstructions of ancient Greek sites of rhetorical performance
- Provide an account of how the physical structures influenced the behaviors of speakers and listeners who gathered in them
- Assess the suitability of the structures as venues of oral performance and group deliberation
The first completed virtual-reality simulation is a structure known as “The Thersilion” at the city of Megalopolis in the Peloponnese (southern Greek mainland). Historical records indicate that as many as 10,000 people would attend meetings in this structure. The group has developed a model for generating reliable estimates of capacity in which virtual audiences of various sizes are visualized from a top-down perspective and an immersive, first-person perspective. The image above shows this simulation on the LMVL’s Powerwall.
A poster about this research was presented by lead author Kyungyoon Kim at the 2012 MSI Research Exhibition in April 2012. It was selected as a finalist in the poster competition.
Nearest Neighbors (NN) is a fundamental operation in many areas of scientific computing, including computer vision, machine learning, robotics, and data mining. It is the backbone of applications people use every day, such as Google Images. Images tend to be high-dimensional, and as the dimensionality of the data increases, the NN task becomes computationally more difficult. This is called the “curse of dimensionality” and it affects efforts to analyze and organize high-dimensional spaces.
Graduate student and MSI researcher Anoop Cherian, who works with MSI PI Professor Nikolaos Papanikolopoulos (Computer Science and Engineering), is developing a novel NN algorithm for image data. The goal is to develop an NN algorithm that is computationally tractable at high dimensions. Other needs for this algorithm include:
- state-of-the-art performance in accuracy
- good search speed
- robustness to data distortions
- storage efficiency
The algorithm is called Multi-Regularization Sparse Coding (MRSC) and is based on sparse coding and dictionary learning. The algorithm is showing great promise in accuracy, speed of retrieval, scalability, and robustness. Because of the huge computational demands that working with millions of data points requires, MSI resources are necessary for this work. The image above shows the results of a search using the MRSC algorithm on a database of images of Notre Dame (containing 1,500 images). The first column shows the query images and the other three columns are the first three nearest neighbors.
Mr. Cherian’s poster about this research was the Grand Prize winner at MSI’s Research Exhibition in April 2012 and has been submitted for publication.