NERSC Initiative for Scientific Exploration (NISE) 2009 Awards

Structure and Dynamics of Premixed Spherical Flames

John Bell, Berkeley Lab

Sponsoring NERSC Project: Interaction of Turbulence and Chemistry in Lean Premixed Laboratory Flames (incite16), Principal Investigator: John Bell, Lawrence Berkeley National Laboratory

Spherical flames at high pressure exhibit a wide range of surface morphologies including pulsating instabilities, spiral waves and target patterns on the flame surface. The goal of this proposal is to simulate high-pressure hydrogen/air flames at both lean and rich conditions using detailed kinetics and transport to capture this type of phenomena computationally. In addition, we want to capture numerically the self-similar acceleration of these types of flames. These simulations are particularly demanding because the flames at high pressure are extremely thin, which places stringent requirements on resolution. This study is a collaboration with Prof. Ed Law from Princeton University, head of the EFRC for Combustion Science.

Determining the time-varying impact of cloud-aerosol interactions during the 20th century

Tom Bettge, National Center for Atmospheric Research

Sponsoring NERSC Project: Climate Change Simulations with CCSM: Moderate and High Resolution Studies (mp9), Principal Investigator: Warren Washington, National Center for Atmospheric Research

The largest uncertainty in the forcing of climate during the 20th century from the most recent IPCC report is the indirect effect of enhanced cloud-aerosol interactions. These experiments will enable a more quantitative assessment of the potential cooling from this process and to what extent they offset the warming impacts of elevated greenhouse gases to date and how these impacts may change in a future climate.

Process studies indicate that the dominant impact of elevated atmospheric aerosols on clouds is a cooling of the earth system due to brighter low clouds and increased reflection of solar radiation. However, the role of this process in the time-evolving climate system remains uncertain. This research will use a development version of CCSM that utilizes cutting edge aerosol and cloud physics that can represent the climate impacts of cloud-aerosol interactions. These high impact experiments will enable a more quantitative assessment of the potential cooling from this process and to what extent they offset the warming impacts of elevated greenhouse to date and how these impacts may change in a future climate.

Elasticity of Two b-DNA Models

Wai-Yim Ching, University of Missouri, Kansas City

Sonsoring NERSC Project: Theoretical Studies of the Electronic Structures and Properties of Complex Ceramic Crystals and Novel Materials (mp250), Principal Investigator: Wai-Yim Ching, University of Missouri, Kansas City

We propose to study the elasticity of two b-DNA models. In the last year, we have constructed two b-DNA models each with periodicity in the axial direction (z-direction) of the double helix. The first model has 10 AT base pairs (A=Adenine, T= Thymine). The second one has 10 CG base pairs (C=Cytosine, G=Guanine). The periodicity in the z direction renders the model to have a twist angle of approximately 36, which is consistent with experimental observation. Each of these two b-DNA models has 20 charged PO4 groups. 20 Na ions near the PO4 groups are added to the model as counterions. The first model has a total of 660 atoms and the second one has 650 atoms. The structures of these two models were fully relaxed using VASP and their electronic structures were calculated using the first principles OLCAO method. Both of these two codes have been used on Franklin.

We intend to study the elastic properties of these two DNA models by systematically extending the models in the z direction and obtaining their total energies for each extension. It is known that the DNA can be stretched up to 90% (strain > 0.9). The bonding between the double-stranded DNA is due to the hydrogen (H) bonding between respective base pairs. Stretching the DNA in the axial direction strand breaks the H-bonds. We plan to create 10 stretched models for each of the 10-base pair models by incrementally extending them by 2% to 5% each. From the energy vs. strain relation, the elastic moduli of the two DNA models can be obtained. Results from such accurate ab initio simulations can be compared with available experimental data thereby gain valuable insights on the mechanical properties of DNA.

Onset of Collisionless Magnetic Reconnection in Weakly Collisional Plasmas

Paul Cassak, West Virginia University

Sponsoring NERSC Project: Onset and Evolution of Magnetic Reconnection in Three Dimensions (m866), Principal Investigator: Paul Cassak, West Virginia University

Magnetic reconnection is a fundamental plasma physics process that occurs in solar flares, the Earth's magnetosphere, and fusion devices. Understanding magnetic reconnection is important for predicting space weather and producing a safe source of renewable energy. Reconnection often begins abruptly, releasing large amounts of stored energy in a short time. Understanding the conditions under which magnetic reconnection begins is a topic of considerable importance to accomplish these goals.

There are two forms of reconnection, collisional and collisionless. It is known that the former is very slow, while the latter is very fast. It has been proposed that a transition between the two is responsible for the sudden onset typically observed in reconnection events. However, the generation of so-called secondary islands is known to make collisional reconnection faster and is expected to be ubiquitous in some applications such as solar flares. Previous studies considered the transition to collisionless reconnection without including secondary islands. We propose to include secondary islands in a simulation to investigate the onset of collisionless reconnection. We predict that secondary islands alter the collisional reconnection phase, but the full-fledged transition to collisionless reconnection remains a dramatic effect. These simulations are challenging because a single computational domain needs to contain both collisional and collisionless physics, which take place at vastly different length scales.

Minimum Free Energy Paths and Free-Energy Profiles of Protein Conformational Changes

Jhih-Wei Chu, University of California, Berkeley

Sponsoring NERSC Project: Activation and Stabilization of Enzymes in Non-aqueous Environments (m787), Principal Investigator: Jhih-Wei Chu, University of California, Berkeley

We propose to apply a reaction path optimization method that we recently developed (Brokaw, Haas, and Chu, JCTC, 5, 2009, 2050-2061) for finding minimum free energy paths and computing free-energy profiles of protein conformational changes.

This method duplicates many (10-100) replicas of a protein system to form a chain that connects two metastable conformations. Molecular dynamics (MD) simulations are performed to compute the mean forces on each replica to minimize the total free energy of a chain. As shown by Vanden-Eijden and coworkers (JPCB, 2005), the resulting minimum free energy path corresponds to a most probable path for the transition between two conformations. The free-energy profile along the optimized path can then be calculated to characterize the energetics of rare events.

Advancements in computational methodology that aim to enable this method to macromolecules have been developed (Brokaw, Haas, and Chu, JCTC, 2009) and our goal is to test and apply this approach to model large scale protein conformational change. In particular, we propose to employ the open-to-closed transition of a type II topoisomerase. Each protein system is composed of ~150,000 atoms and molecular dynamics simulations will be performed using NAMD, which scales very well for such systems to hundreds of cores. The results of MD will be interfaced with CHARMM, in which the reaction path optimization code is implemented. We plan to use 64-128 cores for the MD simulation of each replica and use 25-50 replicas along the chain. Therefore, the typical job size could be up to 1600-6400 cores. This method is implemented in CHARMM as parallel-distributed replicas using MPI. Based on the weak coupling nature of this algorithm, only a small amount of communication is required between replicas.

Theory and Applications of Diamondoid-Nanoparticle Enhanced Organometallic Surfaces

Thomas Devereaux, SLAC National Accelerator Laboratory

Sponsoring NERSC Project: Simulation of Photon Spectroscopies for Correlated Electron Systems (m772), Principal Investigator: Thomas Devereaux, Stanford Linear Accelerator Center

We propose to utilize the massively parallel computing resources at NERSC to assist with the numerical aspects of our currently funded DOE SISGR (Contract: DE-AC02-76SF00515 ) project for theory and applications of diamondoid-nanoparticle enhanced organometallic surfaces.

Diamondoids are nanoscale organic molecules that can be superimposed on the diamond lattice. The smallest diamondoid is adamantane (C10H16, Fig. 1), followed by diamantane (C14H20, Fig. 2), triamantane (C18H24, Fig. 3), tetramantane (C22H28, Fig. 4), and so on. The higher diamondoids such as tetramantane only recently have been isolated from petroleum [1] and have yet to be the subject of systematic theoretical and computational studies. This family of novel carbon nanomaterials is an ideal platform for approaching the DOE grand challenges [2] of energy flow at the nanoscale and synthesis of atomically perfect new forms of matter with applications in solar energy utilizations [3], solid-state lighting [4], and catalysis [5].

Our theoretical and numerical efforts are directed by our close collaboration with ongoing state-of-the-art experiments at Stanford University. Our experimental colleagues utilize a combination of techniques (photoemission, XAFS, and STM) to develop a unique and highly detailed picture of functional diamondoid- and diamondoid-fullerene hybrid-coated metallic surfaces.

We are interested in performing first-principles density functional theory (DFT) based calculations to understand the electronic and vibrational structure of diamondoid-coated metal surfaces and how these properties change, for example, with molecular orientation relative to the surface. In addition to pure diamondoids, we will study hybridizations of diamondoid molecules with fullerene (C60) molecules in contact with metal surfaces. Using DFT calculations we will be able to predict how to utilize hybrids and diamondoid structures as molecular tools for regulating electron flow through surfaces. Furthermore, the functionalization of higher diamondoids and the ability to attach diamondoids to metal surfaces is a newly developed chemical technique [6,7] and has allowed the development of never-before-seen surfaces. In particular, metal surfaces coated with tetramantane-thiol have excellent electron emission properties with potential applications in fields such as electron microscopy and solar power.[8]

Our proposed DFT-calculations are well suited for NERSC because the large size of the higher diamondoids and hybrids, and their proximity to a (two-dimensionally) periodic metallic supercell surface, will require massively parallel computing methods. Our proposed calculations include structural relaxation/optimization of the nuclear coordinates of diamondoid-thiol molecules above metallic surfaces which will predict the arrangement and orientation of the molecules relative to the surface for comparison with recent experiments [9]. Additionally, these calculations will allow us to compute the phonon/vibrational spectrum of the surface which can be measured using tunneling microscopy. Finally, a knowledge of the electronic and vibronic states of these unique surfaces will allow us to study the possibility of using diamondoid-thiol-metallic surfaces as organometallic superconductors [10,11].

[1] J. E Dahl et al. Science, 299, 96 (2003).
[2] Grand Science Challenges Report, "Directing Matter and Energy: Five Challenges for Science and the Imagination", http://www.sc.doe.gov/bes/reports/list.html
[3] DOE Basic Research Needs for Solar Energy Utilization http://www.sc.doe.gov/bes/reports/files/SEU_rpt.pdf
[4] DOE Basic Research Needs for Solid State Lighting http://www.sc.doe.gov/bes/reports/files/SSL_rpt.pdf
[5] DOE Basic Research Needs: Catalysis for Energy http://www.sc.doe.gov/bes/reports/files/CAT_rpt.pdf
[6] B A. Tkachenko et al., Org. Lett. 8, 1767 (2006).
[7] P. R. Schreiner et al., J. Org. Chem., 71, 6709 (2006).
[8] W. L. Yang et al., Science 316, 1460 (2007).
[9] T. M. Willey et al., J. Am. Chem. Soc., 32, 130 (2008).
[10] W. A. Little, Phys. Rev., 134, A1416 (1964).
[11] Y. Ohta et al., Physica C, 460, 121 (2007).

Understanding organic photovoltaics using ab initio many-body perturbation theory

Peter Doak, Berkeley Lab

Sponsoring NERSC Project: Theory of nanostructured materials (m387), Principal Investigator: Jeffrey Neaton, Berkeley Lab

A fundamental global challenge is to develop a scalable technology, with nontoxic abundant components, for efficiently and inexpensively harvesting solar photon energy and then converting it into convenient forms for storage and transportation. Nanostructure-based organic solar cells are now being explored expressly for this purpose. Fundamentally different from more costly silicon solar cells, these devices rely on a high density of nanoscale interfaces to separate and transport electrons and holes. Despite their considerable promise, there is little microscopic intuition or theory to guide material and device design for such systems. A primary reason is the absence of a quantitative picture of the fundamental non-equilibrium electronic structure underlying key processes in solar energy conversion---absorption, charge separation, charge transport, and charge collection.

In this work, we will use accurate ab initio many-body perturbation theory to predict optical absorption and examine charge separation in a “model” organic solar cell system, donor-acceptor “pn” molecules, to understand the mechanisms of third-generation molecular solar cells. These calculations are part of an effort supported by the Helios Solar Energy Research Center at LBNL, where nanoscale organic components are being explored for artificial photosynthesis.

Using many-body perturbation theory (MBPT) to extend DFT calculations of a donor-acceptor molecule, we have recently been able to calculate all relevant single electron addition and removal energies as well as optical excitations. This has been done using the GW approximation and a Bethe-Salpeter equation approach, respectively. We seek to extend this work beyond a proof of concept to a set of prototypical molecules that, on publishing will provide a useful guide to design for the community. Additionally, we propose to calculate the effects of static electric fields on the dissociation and evolution of the excitonic states in such systems. Using these results, we can address the relationship between donor and acceptor level alignment, exciton binding, and efficient charge separation. These are fundamental questions in organic solar cells and other excitonic photovoltaic systems that have not previously been addressed from first principles. This is a unique application of large-scale DFT and MBPT theory, only possible using parallelization across thousands of cores.

Previously we have been successful in scaling the PWSCF NSCF portion of the calculation to thousands of cores. Recent improvements in the BerkeleyGW MBPT code allow better distribution of wave functions across cores allowing greatly enhanced parallelization. It is now possible to run MBPT calculations on larger systems across thousands of cores. This will allow us to take full advantage of massively parallel computing throughout all stages of the calculation. Good scaling may be possible over greater than 10,000 cores, greatly reducing the wall-time cost of the calculations.

We will be able to explore the important effects of binding groups on the free donor-acceptor molecules. We will be able to examine theoretically for the first time the effect of a static electric field on the dissociation of excitons in a D/A molecule. We will be able to demonstrate extreme scaling of MBPT calculations. This will require significant new calculations to be run on larger numbers of cores than we anticipated would be possible at the beginning of our allocation.

Reformation of Methanol on Metal Surfaces

Hua Guo, University of New Mexico

Sponsoring NERSC Project: Quantum Dynamics of Combustion Reactions (m627), Principal Investigator: Hua Guo, University of New Mexico

This research project is exploring the reaction mechanisms for the steam reforming of methanol on metal surfaces. In particular, we are focusing on the initial steps of dissociative adsorption of both H2O and CH3OH on a new alloy (PdZn) catalyst. The research uses the plane-wave DFT code VASP, which has been installed and optimized on Franklin. Typically, the NEB (nudged elastic band) calculation to find the reaction path consists of 7-9 images; each is optimized with up to 16 cores. A typical optimization requires 48-72 hours of CPU time.

Energy Landscape of Nanoclusters

De-en Jiang, Oak Ridge National Laboratory

Sponsoring NERSC Project: Computational study for chemistry of novel separations (m697), Principal Investigator: De-en Jiang, Oak Ridge National Laboratory

We propose to explore the energy landscape of nanoclusters by employing the density functional theory-based basin-hopping technique for global-minimum search. When atoms self-assemble into nanoclusters, they usually form a structure that is the most stable. Searching for this structure would greatly help understand and explain indirect experimental data, such as from mass spectrometry. Basin-hopping is a powerful technique for finding global minima (Wales, D. J.; Doye, J. P. K., J. Phys. Chem. A, 1997, 101: 5111-5116). It involves a constant-temperature Monte-Carlo simulation; at each step, a full geometry optimization is performed. For many clusters, the empirical potentials are simply not predictive enough, thereby requiring a first-principles method. Therefore, we propose using density functional theory (such as implemented in VASP) to do fast geometry optimization, to power the basin-hopping global minimum search. This new project is expected to explore a great chemical space of nanoclusters (with 50 atoms or less) and deliver many interesting discoveries.

Studies of the Sec translocase Transmembrane Channel in Protein Synthesis

Thomas Miller, California Institute of Technology

Sponsoring NERSC Project: Sampling Diffusive Dynamics on Long Timescales, and Simulating the Coupled Dynamics of Electrons and Nuclei (m822), Principal Investigator: Thomas Miller, California Institute of Technology

A critical step in the biosynthesis of many proteins involves either translocating a cellular membrane or integrating into a cellular membrane. Both processes proceed via the Sec translocase - a ubiquitous and highly conserved transmembrane channel. Recent structural studies offer high-resolution glimpses into the translocation process, and genetic studies reveal that the Sec channel is a potential target for cancer therapeutics, but many fundamental aspects of its mechanism and regulation remain unclear. Using novel coarse-graining techniques, my research group has recently discovered that hydrophobic proteins inserted into the Sec channel can stabilize a previously unobserved open state for the channel, which is thought to play a key role in regulating the translocation vs. integration pathways. We are currently extending long-timescale trajectory sampling methods and rare-event sampling techniques to address (i) the mechanism by which targeted proteins are delivered to and positioned for the channel, (ii) the role of specific amino-acid residues and steric interactions in the regulation of targeted proteins and the preservation of the transmembrane ion gradient, and (iii) the importance of membrane fluctuations and channel flexibility in the dynamics and regulation of translocation. This work will yield a unified picture for this fundamental biological pathway, as well as new pharmaceutical strategies for cancer inhibition and treatment.

Electron-Impact Double Ionization of the Be Atom

Mitch Pindzola, Auburn University

Sponsoring NERSC Project: Computational Atomic Physics for Fusion Energy (m41), Principal Investigator: Mitch Pindzola, Auburn University

We propose to calculate the electron-impact double ionization of the Be atom using a modified version of the TDCC-3D code, which has been successfully applied to the double ionization of He [1,2] and Mg [3]. We will modify the TDCC-3D code to do domain decomposition of a (192*192*192) numerical lattice over (16*16*16) nodes using MPI while using openMP threads to partition the extensive nested do-loops over the four cores on each node.

We will test our ideas on a mixed MPI and OpenMP code. Our TDCC-3D results will be compared against Rmatrix calculations using a double pseudo-state expansion (RM-2PS). The first of their kind RM-2PS calculations will be carried out using an NSF Tera-Grid allocation on the Cray XT at NICS.

Accelerating Microbial Genomics with High-Performance Computing

Edward Rubin, DOE Joint Genome Institute

Sponsoring NERSC Project: Optimizing Genomic Data Storage for Wide Accessibility (m342), Principal Investigator: Edward Rubin, DOE Joint Genome Institute

The Integrated Microbial Genomes (IMG) system is a complex data management system that integrates the Joint Genome Institute’s microbial genome data with publicly available microbial genome data and thus provides a powerful comparative context for microbial genome analysis.

One of the more computationally demanding steps in the IMG pipeline involves comparing newly sequenced genes against existing reference genes. BLAST (Basic Local Alignment Search Tool) is used to perform this comparison. This requires comparing over 12 million genes against nearly 7 million genes. This work is typically performed on local clusters with around 200 cores and takes around two weeks to complete. Using the Franklin system can reduce turnaround time to one day or less.

The request for NISE time is to finish porting the entire IMG pipeline to Franklin and prepare a computational framework for upcoming sequencers that will quickly overwhelm the computational resources at sequencing centers like JGI.

Collisionless Dissipation in Turbulent Plasmas

Michael Shay, University of Delaware

Sponsoring NERSC Project: Kinetic Physics of Magnetic Reconnection (m733), Principal Investigator: Michael Shay, University of Delaware

The dissipation of energy in turbulence plasmas plays a fundamental role in a wide range of systems, one of which is heating the outer portions of the sun and the supersonic solar wind, which is a stream of fast-moving particles that slams into the Earth at high speeds. This energy dissipation also plays a critical role in laboratory plasmas, such as fusion plasmas which could provide clean and sustainable energy in the future. However, exactly how this energy is dissipated into heat is not well understood. We hope to make progress on this critical problem of dissipation in turbulence by simulating turbulence in these plasmas using models which include the physical basis for this heating.

Description of Proposed Research: The dissipation of energy in turbulent collisionless plasmas plays a fundamental role in a wide range of systems, and a breakthrough in our understanding would lead to great progress. But the question remains: how do turbulent collisionless plasmas dissipate their energy? In turbulent plasmas, energy cascades from larger to smaller scales and eventually dissipates into heat. Exactly how this occurs, however, in a plasma with no collisions is not well understood. Answering this question will provide significant insight into the dynamics of the solar corona, the solar wind, astrophysical plasmas such as accretion disks and tokamaks. In the solar wind and corona, a key clue to this damping is the often found preferential heating perpendicular to the mean magnetic field. Large-scale PIC hybrid simulations using kinetic ions and fluid electrons will be used to simulate turbulent plasmas, and the kinetic ion heating will be carefully studied. The hybrid version of P3D allows simulations of a significant part of the inertial range while still allowing fast timescale damping effects such as ion cyclotron resonances to occur.

We have previously run modest-sized two-dimensional simulations of an Orszag-Tang vortex, a two-dimensional equilibrium that exhibits a fast transition to a nonlinear cascade and turbulence. Consistent with observations of turbulence in space plasmas, we found heating of ions perpendicular to the mean magnetic field (Parashar et al., Physics of Plasmas, 16, 032310, 2009). However, the mechanism responsible for this heating is currently unknown. To study this dissipation mechanism responsible for the heating, it will be necessary to run larger systems with different mass ratios and larger inertial ranges. This will require larger 2D Orszag-Tang simulations. In addition, we will also simulate some 2D turbulence systems with forcing initial conditions, which allows the dissipation to reach steady state, although the simulations are more computationally expensive. Finally, we will run one large-scale three-dimensional turbulence simulation, which is critical to determine if any critical effects are missing in the 2D systems.

Health Application for Research into Multicore Architectures

Brian Van Straalen, Lawrence Berkeley National Laboratory

Sponsoring NERSC Project: Health Application for ParLab research into Multicore architectures (m1058), Principal Investigator: Brian Van Straalen, Lawrence Berkeley National Laboratory

The HealthApp is a part of the ParLab. It is an attempt to do real-time cerebral blood flow modeling to aid in the treatment of stroke. With 800,000 strokes per year in the U.S. and 150,000 deaths per year, strokes are the third leading cause of mortality. The goal is to develop rapid computational tools to help surgeons and radiologists interpret three-dimensional medical images obtained from computed tomography (CT) and magnetic resonance (MR) for surgical planning and critical decision-making to improve patient care, reduce healthcare costs and save lives. These tools will build upon the successful development of a highly parallel program (2004 Gordon Bell Prize) for virtual strength testing of bones, clinical diagnosis and assessment of osteoporosis, and orthopedic surgical planning.

The HealthApp is a fluid-structures interaction physics problem. The moving boundary fluids problem is being developed in Chombo. The solid mechanics problem is being handled in Prometheus (a highly parallel version of the Athena solid modeling package from UCB). Both codes will be augmented to run in a multicore parallel hybrid mode as part of this work.

Accommodation of H₂ into Hydrate Lattices for Hydrogen Storage Devices

Sotiris Xantheas, Pacific Northwest National Laboratory

Sponsoring NERSC Project: Chemical reactivity, solvation, and multi-component heterogeneous processes in aqueous environments (mp329), Principal Investigator: Bruce Garrett, Pacific Northwest National Laboratory

Hydrate crystal structures are water scaffolds held together by hydrogen bonds that can act as host lattices by trapping guest molecules such as H2, CO2, Cl2, Br2, SO2, H2S, and CH4 within their cages via weak van der Waals interactions. The potential importance of clathrate hydrates (i.e., the combination of the host lattice and the guest molecules) as inclusion compounds relevant to renewable energy (i.e., molecular hydrogen storage devices) has recently been emphasized.

All of the natural gases that form clathrate hydrates do so in the following three crystal structures: sI (cubic), sII (cubic), and sH (hexagonal). These 3D lattices are formed by a combination of the following building blocks (cages): (i) the pentagonal dodecahedron (D-cage), consisting of 20 water molecules forming 12 pentagonal faces; (ii) the tetrakaidecahedron (T-cage), consisting of 24 water molecules forming 12 pentagonal and two hexagonal faces; (iii) the hexakaidecahedron (H-cage), consisting of 28 water molecules forming 12 pentagonal and four hexagonal faces; (iv) the irregular dodecahedron, consisting of 20 water molecules forming three tetragonal, six pentagonal, and three hexagonal faces and (v) the icosahedron, consisting of 36 water molecules that form 12 pentagonal and eight hexagonal faces.

The unit cell of the (sI) hydrate comprises two units of the D- and six units of the T-cages. The unit cell of the (sII) hydrate comprises 16 units of the D- and eight units of the H-cages. The (sII) hydrate has been shown to meet current U.S. Department of Energy’s target densities for an on-board hydrogen storage system.

A bottleneck in the modeling of those constituent cages and corresponding 3-D hydrate networks lies with the existence of a plethora [(3/2)^N] of possible isomers based on the position of the hydrogen atoms that satisfy the Bernal-Fowler rules. We have previously [see Kirov, Fanourgakis, and Xantheas, Chem. Phys. Lett. Frontiers article 461, 180 (2008)] outlined an approach to scan all possible networks of the constituent cages (30,026 for the D-cage, 3,043,836 for the T-cage) using a discrete model and subsequently refine them using levels of electronic structure theory of increasing accuracy (DFT, MP2). We have subsequently used the identified low-energy networks of the D- and T-cages to construct periodic networks of the sI hydrate [see Yoo, Kirov and Xantheas, J. Amer. Chem. Soc. Communication to the Editor, 131, 7564 (2009)].

We propose to use high-level electronic structure calculations to study the accommodation of H2 inside the D- and T-cages and subsequently into the extended periodic hydrate lattices. Previous results obtained at the DFT level of theory are in disagreement as to the amount of H2 that can be accommodated inside those cages. We plan to refine those calculations at a more accurate level of theory [MP2 and possibly CCSD(T)]. Recent developments at the Oak Ridge National Laboratory in the NWChem suite of electronic structure programs coupled with analogous advances in adapting parallel tools (Global Arrays) suggest the efficient scaling of that code to an excess of 100,000 processors for CCSD(T) calculations enabling the study of systems as large as the D- and T-cages with even a triple zeta quality basis set.

Fatigue Fracture in the Polycrystalline Microstructures of Metallic Alloys

Michael Veilleux, Cornell University

Sponsoring NERSC Project: Explicit finite element simulation of microstructurally small fatigue crack propagation in aluminum alloy 7075-T651 (m952), Principal Investigator: Michael Veilleux, Cornell University

This research is in the “research area of code scaling to higher concurrencies”, one of the three research areas of the NERSC Initiative for Scientific Exploration. The code to be investigated is Finite Element All-Wheel Drive (FEAWD), a non-linear finite element solver. FEAWD is an academic application developed by Gerd Heber of the Cornell Fracture Group (www.cfg.cornell.edu), which is being augmented by this project’s PI to optimally model fatigue fracture in the polycrystalline microstructures of metallic alloys. The 2009 NERSC Startup Allocation for this project has focused on compiling and optimizing FEAWD on NERSC’s Franklin cluster. The code has demonstrated the ability to scale up to 512 processors on Franklin. However, an additional allocation is requested to test scaling to 1024 and 2048 processors such that production-level computations can be performed. The production-level models will have approximately 20 million non-linear equations, which must be solved iteratively, up to 10,000 iterations per simulation. In addition to the scaling tests being performed, checkpointing will be tested at the larger scales, and a production-level model will be analyzed to demonstrate all capabilities.

Sensitivity to the Carbon-Nitrogen Cycle in the Ocean

Steven Yeager, National Center for Atmospheric Research

Sponsoring NERSC Project: Climate Change Simulations with CCSM: Moderate and High Resolution Studies (mp9), Principal Investigator: Warren Washington, National Center for Atmospheric Research

Two 200-year CCSM4 integrations with the 2-degree atm, 1-degree ocean to be added to previously completed simulations to obtain statistically significant results.

Two CCSM4 integrations with the 2 degree atm, 1 deg ocean version have been completed (at another site). These were present day cases without the Carbon-Nitrogen scheme. One experiment has the overflow parameterization on, the other does not have this scheme. While comparing the Atlantic MOC (AMOC) variability in these two cases, we noticed that there were substantial differences in amplitudes and frequency of the AMOC oscillations. We require computer time to continue both these cases 200 years more (each) to get statistically significant signals.