NSF-IGERT ComputationalNeuroscience FirstSymposiumSupercomputing InstituteResearch Scholars1999 UndergraduateSummer InternsPhysics of Supersonic CosmicFlowsFlow and Transport in PorousMediaIll-Nitride SemiconductorDevicesStructural Studies of ToxinsProduced duringStaphylococcal InfectionsSubmicron MagneticStructuresPreconditioning Large SparseMatrix ProblemsBiomolecular Interactions andEnzymatic Reactions

Visitors

Research Reports

Charles Park (left) works with Ian Tregellis (right), a graduate student inProfessor Thomas Jones' Astronomy research group.

Coordinator of the Computational Neuroscience Internship Program,Kathleen Clinton (center), speaks with the Computational Neuroscienceinterns Aaron Miller (left) and David Liebelt (right).

Volume 16 Number 1 Fall 1999

his summer, fifteen undergraduate student researchersfrom across the country served ten-week internship

appointments at the Supercomputing Institute, includingthree students in the Computational Neuroscience Program.These students were selected from a pool of 105 applicants.The students worked closely with faculty advisors on manyprojects.

TheSupercomputingInstitute SummerInternshipProgram, currentlyin its ninth year,promotesundergraduateinvolvement inongoing and newresearch in scientific computing,digital technology, and visualization inthe physical, medical, and socialsciences and engineering and in newsoftware development efforts forscientific computing and graphicssupport for such research with themain goal being to carry out usefuland interesting research. Thisprogram provides an opportunity for achallenging and enriching educationalexperience for undergraduatestudents interested in pursuinggraduate or professional educationand research in scientific computingand/or graphics.

During the summer, internsparticipated in Institute sponsoredtutorials specific to high-performance

computing. To conclude the summer, the interns presented talks open to the

Sara Firl prepares for her presentation.

entire research community. These talks allowed them to share their work and togain experience making scientific presentations. The program allowed thestudents to perform research in close collaboration with faculty investigatorsand their research groups and to discuss research with faculty members,post-doctoral associates, graduate students, and other interns with similarinterests.

Project Descriptions

Sarah Alfano, a Physics major at Kent State University, worked with Professor J.Woods Halley of the Physics Department. Their work measured helium collisionsby placing resistive chromium source mounts on one end of a box and titaniumbolometers on the opposite end. It was expected that if the graph of powerthrough the chromium source mounts had a single peak, the graph of powerthrough the bolometer would also have a single peak. However, this was not thecase. Investigations of the effects of collisions were made to see if they couldaccount for observed multiple extrema. The simulation was coded in fortran andhas been run one thousand times for one hundred particles and ten thousandtime iterations. These simulations were run to obtain an average in hopes ofsuppressing the magnitude of statistical fluctuation as far as possible. From thesimulations ran, it was discovered that the working scheme was not as straight-forward as it was hoped.

Benjamin Miles, a Biochemistry major atIndiana University, worked with ProfessorWilliam Gleason of the Laboratory Medicineand Pathology Department. Benjamin's workhelped develop a three-dimensional model forthe Vascular Endothelial Growth Factor (VEGF)165 isoform. Using the macromolecularmodeling programs insight and discover, theknown structures of VEGF were joined, andunknown components were modeled. Anenergy minimization calculation was thenperformed on the completed structuralmolecule. Once complete, procheck was usedto evaluate it. In addition to the computationalwork in developing a model of VEGF, somepreliminary experiments aimed at developing aquantitative VEGF assay were done.

Sara Firl, a Chemical Engineering major at theUniversity of Minnesota, worked with ProfessorAlon McCormick of the Chemical Engineeringand Materials Science Department. Together,they studied how molecules group or clustertogether using various tools such as Monte

Thomas Miller (left) talks with Jocelyn Rodgers (right) at theprogram's kickoff reception.

Summer 1999Tutorials

64-bit Parallel MPI andMath/NumericalLibraries for the IBMSPCode OptimizationWorkshopIntroduction toInsightII/DiscoverIntroduction toMolecular AnimationIntroduction to ParallelProgrammingIntroduction to ParallelProgramming on theIBM SP Supercomputer

Carlo simulation. Sara helped with a series offortran codes used to set up a simple lattice ofcubic unit cells with established periodicboundary conditions. The energy of the unitcell was initially calculated using aLennard-Jones potential. One random moleculewas moved, and the energy was calculatedafter the move occurred. The program ran untila reasonable equilibrium was reached. Anenergy curve and radial distribution functionwere plotted from the final structure data.Further development of this project will studythe adsorption of xylene molecules into thepores of compounds such as zeolites andclatharates.

Brent Grocholski, a Physics major at the University of Minnesota, worked withProfessor George Wilcox of the Pharmacology Department. Brent and ProfessorWilcox worked on the development of improved neuronal simulators for sensorynerves. Brent helped write a program that used the Hodgkin-Huxley model and afourth order Runge-Kutta integrator to simulate neuron action potentials. Aprogram was then written to both approximate the Jacobian for the Hodgkin-Huxley equations and perform an LU factorization used to solve a vector used inDASPK (Differential Algebraic Solver with Preconditioned Krylov methods). Thepreconditioning matrix failed to reduce the number of iterations needed to solvea time course when the full Jacobian was used, but setting the first row of theJacobian to zero reduced the number of iterations required to solve the timecourse by a significant amount.

Charles Park, a Mathematics and Physicsmajor at Yale University, worked with ProfessorThomas Jones of the Astronomy Department.Charles helped produce synthetic astronomicalobservations of magnetohydrodynamicalsimulations of the jet flow in radio galaxies.These synthetic observations are useful forstudying the jets because they allow the studyof emissions from relativistic electrons in thejet and the test of reliability of actualobservations of radio galaxies and theproperties that observational astronomers inferfrom radio telescope images. The syntheticobservations were produced by a raytracingprogram. By using standard astronomical toolsto analyze the synthetic images, results ofthree-dimensional simulations were made moreaccessible to observational astronomers. The

Introduction to ParallelProgramming on theSGI Origin 2000SupercomputerIntroduction to PerlIntroduction to thePOWER3 Architectureand Code Tuning GuideIntroduction toScientific VisualizationIntroduction to ShellProgrammingIntroduction to theSupercomputingInstituteMolecular VisualizationTools Anna Saputera awaits the start of a tutorial at the

Supercomputing Institute.

Amy Beukelman (left) discusses her research with aresearcher at the Supercomputing Institute.

synthetic images suggested many of thequantities radio astronomers had infered fromactual observations. Other values were foundto rely noticeably on the orientation of theobject with respect to the observer.

ThomasMiller, aChemistryandMathematicsmajor atTexas A&MUniversity,worked withProfessorDonaldTruhlar of theChemistry

Department. Together with graduate studentMichael Hack, they implemented and tested theArmy Ants method, a statistically improvedform of the fewest switches trajectory surfacehopping method of calculating reactiondynamics for three-atom, two-state systems. Anumber of analysis programs were written tointerpret the new output, and a Monte Carlosimulated trajectory surface hopping programwas written to optimize the code's various inputparameters. The programs were applied to thecollision of an electronically excited bromineatom with a hydrogen molecule, a problematic system with very weak couplingbetween the two potential surfaces.

Jocelyn Rodgers, a Chemistry and Physics major at Harvard University, workedwith Professor Truhlar on optimizing molecular geometries via multi-level linearcombinations of electronic structure calculations. Recently, Professor Truhlar'sgroup has developed several single-point energy methods that yield moreaccurate energies without a prohibitive increase in computer time. This projectexpanded these linear combination methods to include multilevel optimization ofmolecular geometries. A code called multilevel, written in fortran90, wasdeveloped to calculate the energy, gradient, and Hessian (a multi-dimensionalmatrix of second derivatives) for a molecule with the multilevel methods. A fewmolecules have been optimized with the methods in multilevel, and the methodshows great promise.

Christine Tratz, a Chemistry major at the University of Oklahoma, worked with

Joseph Cooley (left) and Brent Grocholski (right) await atutorial at the Institute.

Professor William Gleason (left) of the Laboratory Medicineand Pathology Department with intern Benjamin Miles (right)before Benjamin's talk.

Professor Truhlar and graduate student PattonFast on multi-coefficient correlation methods(MCCM) and multi-coefficient gaussianmethods (MCGx) for the efficient and accuratecalculation of potential energy surfaces.Investigations were made on the reliability ofthese methods for types of compounds towhich they have not previously been applied.The methods were improved by optimizingthem over a larger training set. In order tocreate a larger test set for the MCCM andMCGx methods, a number of gaussian jobswere run. Spreadsheets that provide theelectronic energies were created with theresults of the gaussian calculations. The errorsin the different methods were then compared.Coefficients for each of the methods werereoptimized and modified. The mean unsignederror in the new methods were shown to beconsiderably lower than in the previousmethods for all cases‹before and afterreoptimization.

Aaron Miller worked with Professor TimothyEbner of the Neurosurgery and PhysiologyDepartment. Aaron's work was part of theComputational Neuroscience Internship

program. Aaron and Professor Ebner performed experiments that investigatedwhether the visuomotor pursuit tracking error in monkeys would be a function oftarget width and speed. A female rhesus monkey was shown to use a two-jointmanipulandum to make visually guided arm-tracking movements in thehorizontal plane. For each successful trial, x and y position points were recorded,smoothed, and digitally differentiated to obtain speed points. Position error wascalculated as the difference between the x and y positions of the hand andtarget. Speed error was calculated as the difference between the speed of thehand and the target. Position error was not found to change significantly withchanges in target size and speed. Speed error was not found to changesignificantly with target size, but did increase significantly with increasing targetspeed. Another experiment, in which the target underwent an abrupt two-foldchange in speed, found an observable transient increase in speed error.

Karis Stenback worked with Professor Robert Miller. Karis' work was also partof the Computational Neuroscience Internship program. Together, theseresearchers worked on elucidating the mechanism by which active dendriticspikes (spike initiation) occur, observing how they propagate once they arepresent, and investigating the functional differences and attributes of activedendritic spiking versus active somatic spiking. Both an equivalent cylinder (EC)

Thomas Wilson prepares his work in one of the offices at theSupercomputing Institute.

Sekar Velu (left) and Charles Park (right) before a tutorial.

model, used to exhibit amacrine cell properties, and a real cell model, createdfrom a trace of an on-off amacrine cell obtained during experiments, were used.Using the parameters from the EC amacrine cell model in the real cell model,alpha synapses were dispersed throughout the dendritic tree in proximallocations. Multiple voltage recording sites were placed throughout the dendritictree and in the soma to elucidate where dendritic spikes were initiated. To lookat the functional differences and attributes of dendritic spiking versus somaticspiking, only the calcium area was looked at. It was found that the intracellularcalcium concentration showed a ten-fold increase when the spike mechanismwas present and showed almost no change when the spike mechanism wasabsent.

Amy Beukelman, a Chemistry and Biologymajor at the University of Minnesota,worked with Professor Christopher Cramerof the Chemistry Department. Together,these researchers studied singlet-tripletgaps and transition states. After gas-phasecalculations were performed, calculationswere carried out in solvents, as the methanolenvironment chosen more closely resemblesexperimental environments. Because someof the structures chosen were ions, it wasimportant to look at the structures insolution, where ionic forms of molecules aremore favored. Single point energycalculations of the optimized structureswere then performed using the amsolprogram and methanol as a solvent. Thegamesol program was used (with methanolas a solvent) to obtain geometries of themolecules in solvent. While the triplet wasfound to behave classically, the singlet ionrearranged in the gas phase. The singlet-triplet gap was actually quite largeindicating that intersystem crossing betweenthe two related molecules was unlikely.

Sekar Velu, a Computer Science andChemistry major at Syracuse University, worked with Professor DouglasOhlendorf of the Biochemistry, Molecular Biology, and Biophysics Department.Sekar helped build a computer program to take two undocked proteins andaccurately predict their docked structure. The project involved writing a shapecomplementarity program in C. This program utilized an earlier algorithm inorder to generate a correlation coefficent between two protein surfaces. It wasthen decided that the best approach to take in determining the structure wasthrough the use of the Fourier Transform method. Extensive tests were run on

Sara Alfano discusses her work.

Computational Neuroscience interns Aaron Miller (left) andKaris Stenback (right) await the start of presentations.

FTDock to determine the most effective parameters. The generated complexeswere then tested against the crystallographically determined structure.Electrostatics were found to play a diminutive role in complex conformation, andsolvent accessibility was found to be a much more important factor. Shapecomplementarity was found to play an important role because there appears tobe an ideal value for a specific protein-protein complex.

Joseph Cooley, a Computer Science major at the University of Minnesota,worked with Professor Linda Boland of the Physiology Department. Joseph andProfessor Boland optimized a model for ion channel gating. A model thatsimulated ion channel current flow was programmed by allowing the model toinput biological data, performing serial and parallel simulations, and normalizingsimulation data with biological data. Multiple simulations were then performedvia parameter changes, and the resultant simulation with the least error incomparison to biological data was chosen. Finally, multi-platform developmentwas done. Additional work on the model added more optimizations and features.

David Dreytser, a Chemical Engineering andChemistry major at the University ofMinnesota, worked with Professor DavidThomas of the Biochemistry, MolecularBiology, and Biophysics Department. Theyworked on computational simulations ofelectron paramagnetic resonance (EPR) onPhospholamban, a 52 amino acid integralmembrane protein of the sarcoplasmicreticulum. The three-dimensional structure ofphospholamban was extended into thecytoplasmic domain. Initially, a script waswritten to convert formats, and a spin labelwas built. This spin label was attached tovarious binding sights on phospholambanbefore molecular dynamics were performed.

Anna Saputera, a Computer Science majorat the University of Minnesota, worked withProfessor M. Germana Paterlini of the Collegeof Pharmacy. Anna and Professor Paterlinicalculated the electrostatic free energy ofinteraction of an opioid peptide as a fuction ofdistance and orientation from a lipid bilayer.Anna helped create a C++ program to rotateand transform the position of the peptidemolecule with respect to the lipid. She thenused midas for visualization of the system

before and after the coordinate transformation. The electrostatic potentialenergy of the molecule/lipid systems were calculated at different translation and

orientations and with different input parameters for delphi to check theperformance of this program. The potential energy between molecule and lipidwas then calculated as a function of distance. The profile showed an atractiveinteraction energy between the lipid and the peptide. The energy profileobtained using delphi was checked for convergence using different inputparameters.

David Liebelt from Northwestern University worked with Professor DavidRottenberg of the Neurology and Radiology Department at the VA MedicalCenter. This work was part of the Computational Neuroscience InternshipProgram. David helped create a World Wide Web interface for the Corner CubeEnvironment, which faces several hurdles before becoming commonneuroimaging software. The interface consists of html documents displayed in anhtml frameset. With JavaScript, this interface is able to store image data andviewing preferences chosen by the user. With the data and preferences stored,JavaScript generates an html form that the user can submit over the Internet.The form is processed by a cgi script written in perl. This processing involvesinterpreting the data and preferences, writing data and preference files readableby the Corner Cube Environment, running the Corner Cube Environment, anddisplaying the output to the World Wide Web browser.

Cody Zilverberg, a Computer Science major at St. John's University, workedwith Professor David Lilja of the Electrical and Computer EngineeringDepartment on performance visualization tools for java programs. Coding wasdone on the 'Find Next' component of a program called JaViz. JaViz visuallydisplays a tree of nodes where each node represents a method call in a Javaprogram. 'Find Next' contains a simple graphical user interface (GUI) that allowsusers to find the next node in the JaViz tree that matches certain criteria.Because there can be many thousands of nodes in a single tree, an algorithmneeded to be developed that would read node data from a file, examine the data,and then (for memory purposes) discard unnecessary data. This search wasimplemented by using a background thread that allowed the visualizer tocontinue functioning during execution.

Thomas Wilson, a Computer Science and Architecture major at the Universityof Minnesota, worked with Professor David Yuen of the Geology and GeophysicsDepartment on a two part project. The first part of the project consisted of aWorld Wide Web presentation focused on scientific visualizations related tonuclear waste transportation and processing at the Hanford, Washington nuclearwaste reserve. This report consists of critical reviews and improvements made tothe visualizations. The second part of the project involves Java. Tom helpedimplement a generic Java applet control panel that included useful features forthe Java based pV3 graphical user interface (GUI). The features includedmathematical operations initiated by a mouse click, writing to the standardoutput, and providing an echo area in the window. Overall, this work found Javato be a good language to use for the new version of pV3 because of security,functionality, portability, and the GUI widgets needed.

Christine Tratz (left) and MaeganHarris (right), an undergraduateresearcher in Professor Truhlar's

Chemistry group.

Cody Zilverberg (left) and DavidDreytser (right).

Summer 2000 Program The Supercomputing Institute ispleased to announce its UndergraduateInternship Program for Summer 2000.Summer appointments will be full-time,ten-week appointments. The 2000program will run from June 12 throughAugust 18, 2000. A student interested inbecoming an intern must be anundergraduate student at the time of theinternship to be eligible and must be acitizen or permanent resident of theUnited States and its possessions. All applications are judgedcompetitively based on the qualificationsof the applicant and the availability of asuitable project.

Application forms, appointmentinformation, and project lists areavailable on the World Wide Web at:

www.msi.umn.edu/general/Programs/uip/uip.htm

Application forms and project lists arealso available by contacting:

Undergraduate Internship

Coordinator University of Minnesota Supercomputing Institute 1200 Washington Avenue South Minneapolis, Minnesota 55415-1227 Phone: (612) 626-7620 Email: uip@msi.umn.edu

HOME | BULLETINS | CONTACT US | PREVIOUS ARTICLE | NEXT ARTICLE | THIS ISSUE