Practicum Abstracts

Validation studies require choosing the appropriate validation metric for comparing the computational evidence with physical experiments. Validation metrics can provide helpful insight into how similar or how different the validation comparison is. But this does not guarantee that all metrics will lead to the same conclusion. This project aimed to find three fundamentally different validation metrics that would be applied as a validation study from different perspectives that lead to a collection of evidence informing a decision. This project aimed to provide a reproducible open source code to perform this study on a variety of datasets, one of which is of a medical device (my dissertation research). There was an additional focus on the type of datasets particularly those with small sample limitations.

Sandia National Laboratories have created a computational modeling workflow for Verification & Validation (V&V), Uncertainty Quantification (UQ), and Credibility Processes. I was trained on state-of-the-art processes for V&V, UQ, and sensitivity analysis (SA) for hypersonic analysis workflows. As well as with the automatically-run and recurring validation tests on HPC resources which was part of a broader verification and validation test suite by looking into parallelization of UQ. This effort is to quantify and compare sources of uncertainty in coupled aerothermal analysis.

Part of the power of quantum computing is that quantum states exist in exponentially high-dimensional space even with only a few qubits. However, if the qubits are not fully independent from one another, the qubits cannot "cover" as much of the space as they otherwise could, and computational power is thus lost. We examined recent results in quantifying this potential dimension reduction from the perspectives of quantum shadow tomography and overlapping qubits.

Ion-containing polymers (ionomers) show promise as safer and cheaper alternatives to conventional battery electrolytes. However, strong electrostatic interactions between ionomers and diffusing ions lead to ionic aggregation and slow ion dynamics, which is unfavorable for battery electrolytes. Understanding the formation of ionic aggregates and dynamics at the molecular level is essential to create and design viable ionomer materials. Coarse-grain (CG) molecular dynamics simulations can provide valuable microscopic insights into these systems that experiments cannot, while also accessing diffusion time scales. In this work, we build on previous CG simulations by incorporating explicit polarization (via Drude oscillators) into an ionomer melt. With polarization, the ions can adjust their charge structure to the local environment, which contributes to both the structural and dynamic properties in these ionomer systems.

Using Sandia's Trilinos & Kokkos, MPI for communication, ParMETIS for partitioning, and Exodus-II API for mesh I/O, implement partitioning of the mesh (domain decomposition) and assemble matrices to solve the desired PDE. Only solution provided in time was Heat Equation. End result was far from desired due to a combination of issues ranging from misunderstandings due to lack of whiteboard discussions due to being remote, delays due to conditions in Sandia's "No-Fee Agreement" policy and remote-work restrictions, a plethora of build issues, documentation issues and affiliated bugs, etc. Originally slated to run on multiple GPUs, but due to build issues and time constraints, was unable to build on target architecture. Nothing against Ryan Viertel as a supervisor, given how busy the development team was.

We train neural networks to predict the output of Kohn-Sham DFT calculations directly from descriptors of local atomic environments on a Cartesian grid. Previous efforts have focused solely on training such surrogates for single systems at a set thermodynamic state. We explore whether current and new novel architectures can be trained in a more transferable fashion, so that they may predict DFT outputs for unseen materials or thermodynamic states.

Learning to predict future events is useful for many applications. In this project we explored how predictive structure can be learned using on-chip plasticity in neuromorphic hardware. Inspired by the neuroscience of the hippocampus, we developed an algorithm based using spike-timing dependent plasticity and a memory consolidation phase to simultaneously make predictions about future stimuli while learning the predictive model, in an online fashion.

A vertical Lagrangian coordinate has been used in global atmosphere models for nearly two decades and has several advantages over other discretizations, including reducing the dimensionality of the physical problem. As the Lagrangian surfaces deform over time, it is necessary to accurately and conservatively remap them back to fixed Eulerian surfaces. A popular choice of remapping algorithm is the piecewise-parabolic method, a modified version of which is used in the dynamical core of the atmosphere component of the Energy Exascale Earth System Model. However, this version of the remapping algorithm creates unwanted noise at the model top and planetary surface for several standard test cases. We explored four alternative modifications to the algorithm and showed that the most accurate of these eliminates this noise.

We explored implementations of cutting-edge theoretical randomized algorithms with the example application of finding row and column permutations of sparse matrices which induce blocks. These algorithms can be used to accelerate sparse matrix operations. The time savings of the accelerated operation must offset the time it takes to run the algorithm, so approximate algorithms which run very quickly are highly desirable in this setting. There are applications to problems like graph partitioning.

One of the biggest problems affecting energy yields from inertial confinement fusion is that of interface mixing. In order to better understand the plasma physics occurring in this phenomenon, I worked with 1D kinetic simulations and set up an uncertainty quantification analysis on the ionization models that feed into the simulations. This allows us to determine how much the uncertainty in the ionization model propagates into the uncertainty in the overall simulation. In turn, this uncertainty quantification can be used with synthetic observation tools to explore how much the uncertainty affects experimental observations, and if experiments are sensitive enough to distinguish between ionization models.

The goal is to compute frame fields for hex meshing. The majority of frame fields have bad singular graphs which fail to be hex meshed. By modifying the singular graphs we aim to fix these frame fields so that they can be hex meshed.

We propose to utilize the time to investigate how local surface strains influence the evolution of the frictional properties of two-dimensional (2D) materials while simultaneously examining the change in reactivity induced by surface strain and applied loads; this will give insight into properties such as reduction and oxidation potentials. As a 2D material is deformed, changes in the local bond-order should render specific locations more susceptible to tribochemical reactions. The nature of tribochemical reactions is still vague; thus we will employ a coordinated set of density functional theory (DFT) and molecular dynamics (MD) simulations using high-performance computing to systematically investigate the reactions to infer their origins in detail. As a continuation from the previous practicum, we will begin to investigate the reactivity of strained states with select molecules by creating a model to compare with experimental results. Solid-state DFT is slated to be implemented to simulate further the system of interest along with determining the accuracy of the MD simulations. This DFT approach also serves as a tool for simulations when there are no MD force fields available for future problems that we want to investigate. Learning software to perform DFT calculations on solid-state systems will add to my knowledge of running DFT calculations on large periodic systems.

Tribology is the study of friction and wear at interacting surfaces. The impact of friction and wear on the global economy is directly related to energy loss in systems and how that energy may be recovered to reduce fuel consumption. Holmberg, K. et al., Friction 2017. 5, 263-264 indicated that by utilizing new surfaces, materials, and lubrication technologies for friction reduction and wear protection in vehicles and other machinery worldwide, a savings of 1.4% of the GDP annually and 8.7% of total energy consumption could be realized long term. Molybdenum disulfide (MoS2) has been an integral part of material technologies used to control friction and wear in applications. MoS2 is a transition metal dichalcogenide that is useful in solid lubrication, heterostructures, catalysis, and 2D-based electronic devices. The lamellar structure of MoS2 is similar to that of graphene, mica, and hexagonal boron nitride, with weak interactions between the lamella leading to exceptional frictional properties. This leads to the need to understand the interactions at the atomic scale, and uncovering the physics of atomic-scale friction and its influence on energetics during sliding is of value in a fundamental understanding. We investigate how local surface strains influence the evolution of the frictional properties of MoS2 while simultaneously examining the change in reactivity induced by surface strain and applied loads; this will give insight into properties such as reduction and oxidation potentials. As a 2D material is deformed, changes in the local bond-order should render specific locations more susceptible to tribochemical reactions. The nature of tribochemical reactions is still vague thus we used atomistic simulation to construct models that accurately predict results from tribological experiments. Establishing relationships between the properties of MoS2 and frictional performance aids in developing more robust experimental studies and helps to better understand friction, wear, and lubrication. Sandia National Laboratories is a multi-mission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC., a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

The project entailed creating a robust algorithm capable of creating a series of holograms that served as optical traps for atoms. By creating a series of holograms, one could potentially move atoms into certain arrangements and measure quantum effects between the atoms. The holograms that were created needed to have certain properties to ensure atoms were properly trapped, including in transition between certain end states. Furthermore, to accelerate the rate at which holograms were created, GPUs were to be used to assist in hologram generation.

Implemented finite element in angle schemes to discretize the angular portion of the charged particle transport equation. Investigated unit sphere mappings to two-dimensional coordinate systems to determine best transformations for meshing.

Two different methods for computation of release isentropes for aluminum were compared. This allowed verification of the accuracy of a new method for computation of isentropes using quantum molecular dynamics that requires fewer calculations. These isentropes are crucial for modeling shock experiments, especially those that involve transfer of shock across different materials. The high computational cost of quantum molecular dynamics simulations demands development of the most efficient schemes possible.

Multi-Agent System Simulations are costly to produce because of the amount of data required and the length of observation required to make any meaningful simulations. Massively Multiplayer Online Games (MMOG) have recently become a useful source of data in creating and validating Multi-Agent Models. I used an MMOG to study real-world phenomena of interest: the conditions that gave rise to Large Scale Conflicts (LSC's), with the intent of producing a Multi-Agent Model of virtual LSC that would help predict real-world LSC.

We continue to study algorithms for the subgraph isomorphism problem implemented for massively multithreaded shared memory architectures such as the Cray XMT. Fast algorithms for the subgraph isomorphism problem will have important implications in many fields including biology, data mining, and satellite imaging.

We study algorithms for the subgraph isomorphism problem implemented for massively multithreaded shared memory architectures such as the Cray XMT. Fast algorithms for the subgraph isomorphism problem will have important implications in Biology. Recently the availability of genomic data has increased drastically due to the use of high throughput methods which have made sequencing relatively easy. Automated techniques are needed in order to glean useful information from this wealth of data. Many of these techniques rely on solving an underlying subgraph isomorphism problem and will thus benefit from fast highly parallel algorithms. By implementing these algorithms on a multithreaded shared memory platform, we minimize the communication latency due to the need for random access into the large graph which is inherent to these types of problems.

I evaluated a new version of a two temperature model that was being implemented in the LAMMPS Molecular Dynamics Code. I first tested the model of on the annealing of a Frenkel Defect in a Lennard-Jones system. Then I modeled the annealing a heat deposit event in a Lennard-Jones Crystal, a Binary Lennard-Jones Crystal, and a alpha-Quartz crystal with various attached electronic subsystems.

The Loki-infect model is an agent based infectious disease model developed at Sandia NL and used to study community-level mitigation strategies in the event of an influenza pandemic. At this point in time, all such models (both agent-based and deterministic) use a simplified disease manifestation based mostly on epidemiological data and various assumptions regarding immeasurable quantities such as infectivity. The most popularly-used disease manifestation model was derived in a "top-down" manner, using data from a study where influenza patients and their contacts were followed, and the statistics of who got sick, how long they were sick, and how frequently they went to the hospital were used to derive the model. This model can accurately represent the observed statistics on average, but is not ideal for agent-based modeling studies. To solve this problem, we aimed to derive a completely new, "bottom-up" disease manifestation that begins with a model of viral proliferation in individuals, and uses the calculated viral load and basic physical principles to derive symptom intensity, infectiousness, death, etc. of individual agents. More specifically, our project had two elements: 1) improving the implementation of the agent-based model and 2) providing a biophysical basis for disease manifestation in individual agents.

Title: Investigating an explicit multiscale time integrator The focus of my project was to investigate the stability properties of an explicit multiscale time integration scheme, in which information obtained from fine scale computations are used to enhance the accuracy of quantities of interest on coarse grids. The method leads to substantial savings for problems that require fine spatial discretization for stress gradient or failure modeling but do not need the associated small time steps dictated by stability criteria.

My project involved coming up with the input data necessary to run a large-scale airline optimization model. Every month, airlines provide their operating data to the government, who then makes it available to the general public. This data includes details about every scheduled domestic flight by all the main carriers, as well as a 10% survey of individual itineraries by passengers. Clearly, these datasets are extremely large in size and difficult to manipulate effectively, and effective tools need to be developed to aggregate the data in order to extract key relationships among carriers, aircraft usage, flights flown, and passenger routings.

The project I was involved with dealt with the development of new stable and efficient algorithms for solving partial differential equations arising in magnetohydrodynamics (MHD), the science of electrically conducting fluids. MHD arises in a number of application areas important to the Department of Energy, including fusion energy research, nuclear weapons stewardship and simulation, geodynamics, and astrophysics. My group was particularly interested in the development of stabilized higher-order finite element methods for the simulation of MHD applications.

Currently in development, SPPARKS is a parallel implementation of the kinetic Monte Carlo (KMC) algorithm that will span the gap between atomistic (e.g., molecular dynamics) and continuum simulation approaches. Parallelizing the KMC algorithm makes it possible to maintain the stochastic nature of the modeled phenomenon while achieving the necessary size and/or time scales for large simulations. Concurrent with the development of the SPPARKS framework is the creation of a suite of materials science applications, enabling simulation of phenomena such as surface deposition and growth, grain growth, and others. My project was to implement an application for simulating particle migration within the parallel KMC simulation framework. The motivating use of the application is the simulation of biased random motion of aluminum surface atoms, which models electromigration-induced void formation in embedded current-carrying interconnects in integrated circuits.

A vector finite element method implementation in MATLAB was developed for application to microwave waveguide calculations. The use of divergence-free vector basis functions for the transverse field components ensures that all nonphysical modes appear with zero frequency. The method is capable of handling general nonsymmetric transversally anisotropic materials. The use of arbitrarily high order nodal and vector elements is enabled through a Mathematica routine that generates MATLAB code to compute the elemental matrices.

We are refining a new computational paradigm for the detection of genetic variation, using deoxyribozymes -- catalytic single-stranded DNA sequences. Deoxyribozymes can be designed to combinatorially detect the presence or absence of particular nucleotide sequences in vitro, and thus can be used for genetic subtyping. The challenges include identifying informative sequences from the samples of interest, creating optimally functional deoxyribozymes, and interpreting the signals produced by the deoxyribozyme-catalyzed reactions.

The purpose of this project is to better understand resveratrol, its stilbene derivatives and their potential targets using various quantitative methods. Resveratrol is a naturally occurring polyphenol whose anti-oxidant properties have been linked to help prevent or slow the progression of many illnesses including cancer, cardiovascular disease, and Alzheimer's disease. Understanding how resveratrol interacts in various pathways is critical to better understanding how resveratrol can be used in developing drug therapies. Relating a molecule's structure to its function is one of the larger problems in biology/chemistry, and there are many different approaches that can be taken in order to study it. The quantitative structure-activity relationship (QSAR) is a mathematical way to correlate the molecular structure with biochemical activity. Using Support Vector Machines (SVMs) to interpret this relationship has become a powerful tool due to the ability to interpret the non-linear correlations. The SVMs transforms the data to a feature space in which the model becomes linear. Using a subset of the Resveratrol Analogue Library, various techniques including the use of Support Vector Machines and Kernel Molecular Dynamics were used to elucidate the molecular structure and classification between active and inactive drug.

The Xyce electrical circuit simulation tool (developed by Sandia National Laboratories) has been extended to create a large-scale, parallel, biocircuit simulator. This required the development of individual circuit elements representing metabolic and genetic substrates and an accurate description of the coupling between elements as demonstrated in the simulation of the tryptophan biosynthesis pathway. A challenge in this project is determining simulation parameters that produce results consistent with experimental data. To this end, we are interested in computational approaches that can accurately estimate multiple parameters for biological network simulations. We focused on the simulation of the Escherichia coli K-12 central metabolic system. Our previous work with this system included the construction of a hybrid whole-cell model, which incorporates both stoichiometric and gene regulatory data in one circuit. The central metabolism model was derived from supplemental data provided in Covert and Palsson (2002). Our initial circuit model did not include empirically-determined reaction rate information in the simulation. Using Michaelis-Menten kinetics and the King-Altman method, we reformulated the metabolic subcircuits to incorporate reaction rates reported in the BRENDA database (http://www.brenda.uni-koeln.de/). In order to estimate the parameters for which no empirical rate information is available, we coupled Xyce to DAKOTA, an optimization toolkit also developed by Sandia National Laboratories. Using DAKOTA and empirical data, we examined various computational approaches for the estimation of multiple rate parameters for the E. coli central metabolic system.

We implemented a new adaptive intermolecular reactive empirical bond order (AIREBO) potential into the LAMMPS software package. This new potential allows us to simulate the interactions within hydrocarbon molecules and has additional terms to treat long range interactions between molecules. We intend to use this potential to calculate the thermal conductivity of a wide variety of single and multi-walled carbon nanotubes. Using the computational resources available to the national labs, we can explore new aspects of this problem with greater detail and confidence.

In this project, we are working on identification of the time and location of several small "micro-releases" of pathogen based on the identification of infected individuals at medical institutions. Large releases of pathogen can be more easily detected by sensors placed at a relatively small number of key locations. However, distributed micro-releases can evade sensor detection since the amount of contagion is smaller at each location and the targets can be much more varied. The first indication of the pathogen will be patients at clinics or hospitals showing sufficiently obvious symptoms for positive diagnosis. Timely identification and characterization of the threat of these micro-releases is essential for the containment of the disease propagation. Since detection is difficult and the dynamics of the disease propagation is complex, obtaining an accurate model is very challenging and computationally intensive, requiring the solution of many interesting computational science problems. The general approach towards this problem of inferring the micro-release characterization from the patient data is to formulate a Bayesian inverse problem which can provide probabilistic descriptions to identify the most likely timings and locations of the onset of pathogen. With these probabilistic descriptions, appropriate responses can be shaped to most effectively mitigate the spread of the pathogens. In order to solve the Bayesian inverse problem, a forward disease propagation problem must be solved numerous times to provide the information necessary to solve this inverse problem accurately. I also spent some of my summer working on a combinatorial optimization problem with Erik Boman.

Poorly understood biomembrane phase transitions are required for important biological processes such as cell fusing and budding. Molecular simulation gives insight into these phase transitions that adds substantially to the knowledge gained through theory and experiment(1). Recent improvements in force field development(2,3) for coarse grained molecular dynamics (CGMD) simulation of biomolecular systems has enabled computer experiments of unprecedented size and timescale. We use atomistic molecular dynamics (MD) simulations with their standard biomolecular force fields (i.e. CHARMM, AMBER) to parameterize CGMD force fields through a least-squares force-matching procedure. The resulting CGMD simulation capability enables a 50-fold or greater enhancement in simulation speed (over atomistic simulation), with only a modest loss of accuracy. With this improved CGMD simulation capability we will investigate transitions between the lamellar and hexagonal phases of biomembranes consisting of phosphatidylcholine (PC) and phosphatidylethanolamine (PE) lipid mixtures. 1. Marrink, SJ, Mark, AE, Biophys. J., 87, 3894 (2004). 2. Izvekov, S, Parrinello, M, Burnham, CJ, Voth, GA, J. Chem. Phys., 120, 10896 (2004). 3. Izvekov, S, Voth, GA, J. Phys. Chem. B, in press.

First, a nonlinear beam model was adopted. In 2-D this model couples vibrations by including stretching modes. This model was examined to determine the resonance behavior of the beam. Additionally, it also resulted in static deflection predictions. Following the theoretical work, experiments were performed where the electrical properties of the CNT were monitored while the CNT was deflected both statically and dynamically.

WHAM, weighted histogram analysis method, is an intuitive and surprisingly accurate methodology for extracting work information from molecular dynamics. One advantage of this methodology is its foundation in umbrella sampling computations and ability to combine several simulations without worrying about overlap issues. This method, along with several other established protocols, was applied to driving solvated ions through a silicon dioxide pore.

The project involved verification and validation of the MHD portion of the radiation magnetohydrodynamics code ALEGRA. In addition work was included on the study of experiment and theory of early-time axial variations that are observed in Z-pinches.

Using the LAMMPS molecular dynamics software package, the gating mechanism of Synechococcus RuBisCO and a subset of mutations were examined with the intention of comparing the results to previous simulations performed using an implicit solvent model. Additionally, information about the specificity factor was related to the work profiles of individual mutant molecules in the hopes that this relationship could be used as a potential predictor of theoretical mutant specificity.

The project goal is to develop a finite element framework that accurately models impact-induced damage in armor-grade ceramics. This framework accounts for inherent variability in material strength of brittle ceramics that is directly related to random distributions of microcracks in the material. These microcracks, which are on the order of grain size, are too small to be modeled explicitly in a field-scale simulation where the finite elements contain thousands of grains. Although the material is statistically homogeneous on this scale, observed material response is not spatially homogeneous. The material fails into cube-shaped fragments. In order to simulate this behavior, computational tools are being developed that create realizations of a material in which the strength of each finite element is allowed to vary in a manner consistent with Weibull theory so that small finite elements have a higher average strength than the large elements. This aspect of the work hypothetically eliminates mesh dependencies in existing finite element failure models without requiring users to supply microscale properties such as flaw densities. In addition to being spatially variable, the fragmentation process is often high-rate for these armor-grade ceramics. This second characteristic is being simulated by introducing material particles in failed zones that replace the role of integration points in traditional finite elements. Unlike integration points, which are tied to the mesh topology, these material points can move independently from the mesh and therefore support massively large deformations without requiring the finite element mesh to significantly distort. This new finite element framework, having a combination of material particle and spatial variability methods, should accurately predict debris-formation and penetration depths into metal-clad ceramic armor.

Sandia is currently developing the new sierra based, explicit, non-linear mechanics, finite-element code PRESTO. Currently, it is being used for component qualification of materials subjected to intense x-ray environments and impact. However, there remains the question of model validity of the currently available models in PRESTO for modeling these problems. This summer, then, was spent modeling a simple gas-gun impact problem with PRESTO using a variety of models to test he results of PRESTO with corresponding experimental results.

My practicum experience at Sandia was unique. Instead of working on just one project with one collaborator, I actually worked on four projects with four different researchers. The commonality of these projects was that we were using different molecular modeling techniques to look at the physical properties and molecular structure of different chemical species. This opportunity to work on various projects allowed me to learn a great deal about the simulation work that occurs at Sandia. The focus of three of these projects was on the molecular modeling of triglycerides and vegetable oils (which are mixtures of triglycerides), and was directly relevant to my thesis work. The first project I studied under the guidance of Dr. John Curro and my thesis advisor, Professor Juan de Pablo. This work was very similar to the studies we have been performing at the University of Wisconsin but on a much larger scale. The goal of my thesis work is to model and understand the behavior of vegetable oils which can be used as environmentally friendly lubricating oils. At low temperatures, vegetable oils tend to gel or crystallize. It is important to understand this behavior and why this phenomenon occurs. Using LAMMPS, a parallel molecular dynamics (MD) simulation code developed at Sandia, and the computer resources of Sandia, we were able to perform MD simulations of vegetable oils on a much larger scale than we could have performed at Wisconsin. The second project, performed under the guidance of Dr. John Curro, looked at an alternative way to model triglycerides. One important physical property used to describe lubricants is viscosity. However, calculating viscosity using MD simulations takes a great deal of computer time (on the order of months). We have developed a coarse grain model of the triglycerides based on soft spheres that can be used instead of an atomistic model. This developed technique has significantly reduced the amount of simulation time to calculate viscosity. The third project studied the molecular structuring of triglycerides. This project, performed under the guidance of Dr. John Curro, Dr. Laura Frink and Dr. Amalie Frischknecht, focused on using polymeric density functional theory (DFT) to determine what kinds of different structures are formed by triglycerides. Previous work at Sandia had focused on using these methods to look at molecular structuring of polymers near surfaces as well as lipid bilayers in bulk solution. Since triglycerides are lipids, we believe, and have observed in our MD simulations, that they form molecular structures. However, these simulations were performed on a small number of molecules so that absolute conclusions were not possible. By using DFT methods we can perform simulations on a much larger length scale and allow conclusions about triglyceride structuring. The fourth project, which was performed under the guidance of Dr. Marcus Martin, stemmed from the Second Industrial Fluid

Molecular dynamics simulations were used to study a model system that approximates polyelectrolytes, including double stranded DNA and sodium poly(styrene sulfonate), tethered on apposing surfaces. The effect of the surface density of the grafted polymer, the chain length, and the gap width on the structure and wall pressure were studied. Results are compared to previous results for a single grafted chain on a single wall, experimental measurements, and scaling theory results. The origin of high lubrication between these apposing walls was investigated by examining the distribution of counterions and monomers. Results are compared to experimental systems where polyelectrolyte layers have been observed to offer a high degree of lubrication.

Using molecular dynamics (LAMMPS) we analyze the electrophoretic mobility of a large spherical solute in a charge cylindrical nanopore. Factors influencing the electrophoretic mobility of the solute (mass, charge, radius) were investigated, particularly for deviations from continuum flow behavior. The fluid was composed of cationic counter ions and non-polar monatomic solvent molecules. The cylindrical surface of the pore wall was represented by a stochastic scattering boundary condition.

The project is an ongoing effort at the interface of genetics and information theory. The exact process of how ribosomal subunits bind to mRNA preceeding the translation process is not well understood, and one theory is that these ribosomal subunits act like something of an Error Correcting Code (ECC) Decoder. To evaluate this theory, one must first determine if an ECC indeed exists on the region of mRNA upstream of the start codon or not. This determination is the goal of the project.

The project involved the study of interface tracking methods for use in hydanmic applications on unstructured grids.

Rubisco (ribulose 1,5-biphosphate carboxylase/oxygenase) is a key enzyme in photosynthesis (catalyzing the fixation of carbon dioxide in the biosphere), and in photorespiration (oxidizing ribulose 1,5-biphosphate (RuBP) with molecular oxygen). The specificity factor of Rubisco, defined as the ratio of apparent first order rate constants for carboxylation and oxygenation is of utmost interest for many species. In an attempt to understand the molecular basis of Rubisco's specificity, we investigate the structure of the enzyme via molecular simulation techniques.

The ability of two simple constitutive models to represent the behavior of a highly porous concrete material under large strains was studied. These constitutive models are part of the Alegra Shock Solid Dynamics code developed at Sandia National Laboratory. One model is a Mie-Gruneisen equation of state with pressure dependent porosity (P-alpha). The second is the soil crushable foam model (SCF). This project was part of a larger effort to improve simulation capabilities for modeling buried structures subject to blast loading. The constitutive models were calibrated for a material specific to one such event, and then large-scale simulations of the event were examined with respect to differences in material model formulation and calibration. This study led to recommendations for improving the constitutive models and targeted additional areas of concern in the large scale simulations.

AUTOGEN is a software package for automatic planning of robot welding for the shipbuilding industry. The geometric modeling, inverse kinematics, and inverse dynamics required make solving this problem a true research endeavor.

Three-dimensional visualization of microscope images provides the microscope user with a sense of depth when handling microsized objects. Before recent advances in graphics hardware, the real-time three-dimensional visualization of microscope image sequences was not possible. In this project I demonstrated a method for the three-dimensional visualization of microscope image sequences in real-time. First a series of microscope images are acquired. After some preprocessing, the images are volume rendered using the ray-casting technique. The volume rendering is performed by the graphics hardware using the 3d texture mapping extension to OpenGL. The three-dimensional rendering can be updated at a rate of 12 frames per second.

The project is an ongoing effort at the interface of genetics and information theory. The exact process of how ribosomal subunits bind to mRNA preceeding the translation process is not well understood, and one theory is that these ribosomal subunits act like something of an Error Correcting Code (ECC) Decoder. To evaluate this theory, one must first determine if an ECC indeed exists on the region of mRNA upstream of the start codon or not. This determination is the goal of the project.

In many transport applications there are physical constraints imposed on a solution field. In chemical transport applications it is often the case that the solution field must be nonnegative. However, this issue is not addressed by many numerical techniques used in simulating these processes. For example, in simulating chemically reacting processes one is often concerned with the spatial concentration of a chemical species in some region of space. It does not make physical sense for the concentration to be less than zero, and relevant constitutive relations & reaction rates often reflect this (involving square-roots, logarithms, etc.). However, in standard numerical techniques negative solution values are admissible and often obtained, leading to problems. This is especially true of high-order methods required to obtain very accurate solutions. The practicum research was involved in analyzing standard techniques for these problems and deriving conditions under which they will produce nonnegative solutions. Numerical experiments were conducted to verify the validity of the methods.

The purpose of my summer project was to evaluate the performance of slip models used with continuum-based simulations to predict MEMS gas flows. Their use is motivated by the rapid simulation times compared to using Direct Simulation Monte Carlo (DSMC); however, with the increased speed comes a loss of accuracy. My project was to focus on the slip models currently proposed in literature, understand when and where they lose accuracy, and determine their range of applicability for systems design.

Trilinos is a C++ library for large-scale linear algebra problems that has been under development at Sandia for some time. My involvement with the Trilinos project was to create a C++ class (and supporting classes) that could completely define a linear problem (based on certain user specifications) for use with any of the linear solvers in the Trilinos package.

Project Summary: In mobile manipulation, most effort has been concentrated on solving the motion planning and control problems for existing manipulators, few have addressed the design issue of manipulator itself. The first step towards a design is a performance index suitable for mobile manipulation. In this project, a new performance index called the Workspace Dexterity Volume is proposed. This index characterizes the orientability of a manipulator within its workspace. The computational method has been explored. It has been demonstrated that this tool can be used both as a quantitative performance analysis tool and as a qualitative tool for workspace dexterity visualization in complex environment that can include multiple obstacles.

Extended duration respiratory support with membrane oxygenators often results in poor outcomes. Patient morbidity, mortality and device failure are frequently related to imperfect device properties such as sluggish flow paths, membranes with limited biological compatibility, and inadequate fiber arrangements. Although performance improvement is possible, current oxygenator design processes rely heavily upon prototype manufacture and testing, resulting in limited investigation of the possible design space. The advent of inexpensive computing power coupled with advanced computational fluid dynamics (CFD) and optimization software suggests a more thorough evaluation of the design space is possible given accurate and robust physical models. Existing CFD models of mass transfer can provide some spatial data required for optimization but require determination of correlative constants prior to use, resulting in a loss of model generality. The elimination of empirical correlations from the CFD mass transfer model would allow the investigation of novel oxygenator designs with little or no prior testing. One approach to eliminating empirical mass transfer correlations from oxygenator CFD models involves the use of small-scale fiber simulations. Simulation results involving single fibers and multiple fibers arranged in various configurations may reveal significant performance differences. Sound application of the models could allow prediction of device performance without testing, resulting in a shortened design cycle. In addition, many novel oxygenator designs utilize active mixing to minimize the mass transfer boundary layer thickness. Direct simulation of the coupled fluid-solid problem could allow unconventional mixing regimes to be investigated prior to

The goal of this project is to study the diffusion of molecules in nanoporous materials (such as elastic polymers) under the action of a finite chemical potential gradient. The method used is a massively parallel implementation of Dual Control Volume grand canonical molecular dynamics.

The MPSalsa project has developed a parallel finite element code to simulate fluid flow, heat transfer, and chemical reactions. My work entailed examining how the MPSalsa code interfaces the gas phase finite element problem with surface problems through the boundary conditions.

CUERVO is a finite element code for nonlinear scalar transport problems. Several finite element codes in the department have the same fundamental structure. The objective of the project was to parallelize CUERVO using MPI and find any problem areas that might prevent parallelization of more complicated codes.

An electromagnetic scattering code using the upwind-leap-frog scheme was proted to two massively parallel computer platforms and evaluated.

This project was a Cooperative Research and Development Agreement (CRADA) between Sandia, Los Alamos, and Amoco Oil Company. Sandia's responsibility included the development of a parallel version of a multiphase flow code which had originally been written for a Cray Supercomputer. The parallel version was required to run on both massively parallel computers such as the N-cube and the Paragaon, as well as a collection of workstations running in parallel.

This project is aimed at gaining insight into attainable solder configurations by using computer modelling in conjunction with pre-existing experimental techniques. 'Parallel Re-implementation of Evolver' The ultimate goal of "Parallel Re-implementation of Evolver" is rewriting Evolver to take advantage of massively parallel computer architectures. "Geometry of Grain Disappearance in Thin Polycrystalline Films" uses Evolver in novel ways to investigate the critical size-to-thickness ratio for grains (crystallites) in polycrystaline films.

This project focuses on the optimization of output measures from reliability models which have uncertainty in the inputs. The optimization problem is how to find the best system structure (configuration topology) and parameter values (i.e. component failure rates) to meet the specified goals of the system. (i.e. minimize cost, maximize system Mean Time Between Failures (MTBF), etc.) My approach to this type of problem is to discretize part of the space, turn the problem into a cominatorial one, and use new probabilistic based combinatorial optimization techniques such as simulated annealing or genetic algorithms as the solution method.