Publications

A global optimization framework for SMPs and multicore clusters is presented.It exploits hierarchal and dynamic task parallelism of the Multistart method.Gradient/Hessian calculations and Newton's optimization method are parallelized.Several task distribution schemes are studied and evaluated.Our framework is applied successfully to the protein conformation problem. This paper presents an infrastructure for high performance numerical optimization on clusters of multicore systems. Building on a runtime system which implements a programming and execution environment for irregular and adaptive task-based parallelism, we extract and exploit the parallelism of a Multistart optimization strategy at multiple levels, which include second order derivative calculations for Newton-based local optimization. The runtime system can support a dynamically changing hierarchical execution graph, without any assumptions on the levels of parallelization. This enables the optimization practitioners to implement, transparently, even more complicated schemes. We discuss parallelization details and task distribution schemes for managing nested and dynamic parallelism. In addition, we apply our framework to a real-world application case that concerns the protein conformation problem. Finally, we report performance results for all the components of our system on a multicore cluster.

The purpose of this paper is to study the effect of piston position on the in-cylinder swirling flow in a simplified model of a large two-stroke marine diesel engine.

In this paper a Moving Least Squares method (MLS) for the simulation of 2D free surface flows is presented. The emphasis is on the governing equations, the boundary conditions, and the numerical implementation. The compressible viscous isothermal Navier-Stokes equations are taken as the starting point. Then a boundary condition for pressure (or density) is developed. This condition is applicable at interfaces between different media such as fluid-solid or fluid-void. The effect of surface tension is included. The equations are discretized by a moving least squares method for the spatial derivatives and a Runge-Kutta method for the time derivatives. The computational frame is Lagrangian, which means that the computational nodes are convected with the flow. The method proposed here is benchmarked using the standard lid driven cavity problem, a rotating free surface problem, and the simulation of drop oscillations. A new exact solution to the unsteady incompressible Navier-Stokes equations is introduced for the rotating free surface problem.

The central dogma of molecular biology has provided a meaningful principle for data integration in the field of genomics. In this context, integration reflects the known transitions from a chromosome to a protein sequence: transcription, intron splicing, exon assembly and translation. There is no such clear principle for integrating proteomics data, since the laws governing protein folding and interactivity are not quite understood. In our effort to bring together independent pieces of information relative to proteins in a biologically meaningful way, we assess the bias of bioinformatics resources and consequent approximations in the framework of small-scale studies. We analyse proteomics data while following both a data-driven (focus on proteins smaller than 10 kDa) and a hypothesis-driven (focus on whole bacterial proteomes) approach. These applications are potentially the source of specialized complements to classical biological ontologies. Copyright (C) 2004 John Wiley Sons, Ltd.

We develop efficient CPU kernels for multiphase compressible flows and evaluate different optimization strategies. The presented software achieves up to 48% of the peak performance on shared memory architectures, outperforming by 9-14X what is considered to be state-of-the-art. On 48-core CPUs we observe speedups of 40-45X and measure up to 360 GFLOP/s over 840 GFLOP/s of the peak.

We investigate the optimal morphologies for fast and efficient anguilliform swimmers at intermediate Reynolds numbers, by combining an evolution strategy with three-dimensional viscous vortex methods. We show that anguilliform swimmer shapes enable the trapping and subsequent acceleration of regions of fluid transported along the entire body by the midline travelling wave. A sensitivity analysis of the optimal morphological traits identifies that the width thickness in the anterior of the body and the height of the caudal fin are critical factors for both speed and efficiency. The fastest swimmer without a caudal fin, however, still retains 80 % of its speed, showing that the entire body is used to generate thrust. The optimal shapes share several features with naturally occurring morphologies, but their overall appearances differ. This demonstrates that engineered swimmers can outperform biomimetic swimmers for the criteria considered here.

The mouse retina has become a prominent model for studying angiogenesis. The easy access and well-known developmental progression have significantly propelled our ability to examine and manipulate blood vessels in vivo. Nonetheless, most studies have restricted their evaluations to the superficial plexus (an upper vascular layer in contact with the vitreous). Here we present experimental data and quantification for the developmental progression of the full retina including the intermediate and deeper plexus that sprouts from the superficial layer. We analyze the origin and advancement of vertical sprouting and present the progression of vascular perfusion within the tissue. Furthermore, we introduce the use of Minkowsky functionals to quantify remodeling in the superficial and deeper plexus. The work expands information on the retina towards a 3D structure. This is of particular interest, as recent data have demonstrated differential effects of gene deletion on the upper and deeper plexus, highlighting the concept of distinct operational pathways during sprouting angiogenesis.

In this work we present the parallel implementation of a hybrid global optimization algorithm assembled specifically to tackle a class of time consuming interatomic potential fitting problems. The resulting objec- tive function is characterized by large and varying execution times, discontinuity and lack of derivative information. The presented global optimization algorithm corresponds to an irregular, two-level execu- tion task graph where tasks are spawned dynamically. We use the OpenMP tasking model to express the inherent parallelism of the algorithm on shared-memory systems and a runtime library which imple- ments the execution environment for adaptive task-based parallelism on multicore clusters. We describe in detail the hybrid global optimization algorithm and various parallel implementation issues. The pro- posed methodology is then applied to a specific instance of the interatomic potential fitting problem for the metal titanium. Extensive numerical experiments indicate that the proposed algorithm achieves the best parallel performance. In addition, its serial implementation performs well and therefore can also be used as a general purpose optimization algorithm.

We present unprecedented, high throughput simulations of cloud cavitation collapse on 1.6 million cores of Sequoia reaching 55% of its nominal peak performance, correspond- ing to 11 PFLOP/s. The destructive power of cavitation re- duces the lifetime of energy critical systems such as internal combustion engines and hydraulic turbines, yet it has been harnessed for water purification and kidney lithotripsy. The present two-phase flow simulations enable the quantitative prediction of cavitation using 13 trillion grid points to re- solve the collapse of 15’000 bubbles. We advance by one or- der of magnitude the current state-of-the-art in terms of time to solution, and by two orders the geometrical complexity of the flow. The software successfully addresses the challenges that hinder the effective solution of complex flows on con- temporary supercomputers, such as limited memory band- width, I/O bandwidth and storage capacity. The present work redefines the frontier of high performance computing for fluid dynamics simulations.

A high order converging Poisson solver is presented, based on the Green{‘}s function solution to Poisson{‘}s equation subject to free-space boundary conditions. The high order convergence is achieved by formulating regularised integration kernels, analogous to a smoothing of the solution field. The method is extended to directly solve the derivatives of the solution to Poisson{‘}s equation. In this way differential operators such as the divergence or curl of the solution field can be solved to the same high order convergence without additional computational effort. The method, is applied and validated, however not restricted, to the equations of fluid mechanics, and can be used in many applications to solve Poisson{‘}s equation on a rectangular unbounded domain.

For over five decades, molecular dynamics (MD) simulations have helped to elucidate critical mechanisms in a broad range of physiological systems and technological innovations. MD simulations are synergetic with experiments, relying on measurements to calibrate their parameters and probing {\textquotedblleft}what if scenarios{\textquotedblright} for systems that are difficult to investigate experimentally. However, in certain systems, such as nanofluidics, the results of experiments and MD simulations differ by several orders of magnitude. This discrepancy may be attributed to the spatiotemporal scales and structural information accessible by experiments and simulations. Furthermore, MD simulations rely on parameters that are often calibrated semiempirically, while the effects of their computational implementation on their predictive capabilities have only been sporadically probed. In this work, we show that experimental and MD investigations can be consolidated through a rigorous uncertainty quantification framework. We employ a Bayesian probabilistic framework for large scale MD simulations of graphitic nanostructures in aqueous environments. We assess the uncertainties in the MD predictions for quantities of interest regarding wetting behavior and hydrophobicity. We focus on three representative systems: water wetting of graphene, the aggregation of fullerenes in aqueous solution, and the water transport across carbon nanotubes. We demonstrate that the dominant mode of calibrating MD potentials in nanoscale fluid mechanics, through single values of water contact angle on graphene, leads to large uncertainties and fallible quantitative predictions. We demonstrate that the use of additional experimental data reduces uncertainty, improves the predictive accuracy of MD models, and consolidates the results of experiments and simulations.

Jet impingement heat transfer from a round gas jet to a flat wall was investigated numerically for a ratio of 2 between the jet inlet to wall distance and the jet inlet diameter. The influence of turbulence intensity at the jet inlet and choice of turbulence model on the wall heat transfer was investigated at a jet Reynolds number of 1.66 {\texttimes} 105 and a temperature difference between jet inlet and wall of 1600 K. The focus was on the convective heat transfer contribution as thermal radiation was not included in the investigation. A considerable influence of the turbulence intensity at the jet inlet was observed in the stagnation region, where the wall heat flux increased by a factor of almost 3 when increasing the turbulence intensity from 1.5% to 10%. The choice of turbulence model also influenced the heat transfer predictions significantly, especially in the stagnation region, where differences of up to about 100% were observed. Furthermore, the variation in stagnation point heat transfer was examined for jet Reynolds numbers in the range from 1.10 {\texttimes} 105 to 6.64 {\texttimes} 105. Based on the investigations, a correlation is suggested between the stagnation point Nusselt number, the jet Reynolds number, and the turbulence intensity at the jet inlet for impinging jet flows at high jet Reynolds numbers.

We present detailed visualizations of the interactions of a normal shock wave at Mach 3, with a spherical helium bubble immersed in air,1 with an interface Atwood number of −0.76 (Figure 1). The governing 3D Euler equations for two-phase compressible flows are solved using a finite volume solver with uniform resolution. We employ the 5th order WENO reconstruction of the primitive quantities, an HLL-type numerical flux, and the 3rd order TVD Runge-Kutta time stepping scheme. The software achieves 30% of the peak performance on a Cray XE6, using 4 × 109 cells. Extended simulations reveal that the shock passage compresses the bubble and generates baroclinic vorticity on the density interface. Initial distribution of the vorticity and compressions lead to the formation of an air jet, interface roll-ups, and the formation of a long lasting vortical core.

We present a coupling of multiscale frameworks with accelerated stochastic simulation algorithms for systems of chemical reactions with disparate propensities. The algorithms regulate the propensities of the fast and slow reactions of the system, using alternating micro and macro sub-steps simulated with accelerated algorithms such as τ and R-leaping. The proposed algorithms are shown to provide significant speedups in simulations of stiff systems of chemical reactions with a trade-off in accuracy as controlled by a regulating parameter. More importantly, the error of the methods exhibits a cutoff phenomenon that allows for optimal parameter choices. Numerical experiments demonstrate that hybrid algorithms involving accelerated stochastic simulations can be, in certain cases, more accurate while faster, than their corresponding stochastic simulation algorithm counterparts.

Carbon nanotube (CNT) membranes hold the promise of extraordinary fast water transport for applications such as energy efficient filtration and molecular level drug delivery. However, experiments and computations have reported flow rate enhancements over continuum hydrodynamics that contradict each other by orders of magnitude. We perform large scale molecular dynamics simulations emulating for the first time the micrometer thick CNTs membranes used in experiments. We find transport enhancement rates that are length dependent due to entrance and exit losses but asymptote to 2 orders of magnitude over the continuum predictions. These rates are far below those reported experimentally. The results suggest that the reported superfast water transport rates cannot be attributed to interactions of water with pristine CNTs alone.

Understanding and accelerating the mechanisms of endothelial wound healing is of fundamental interest for biotechnology and of significant medical utility in repairing pathologic changes to the vasculature induced by invasive medical interventions. We report the fundamental mechanisms that determine the influence of substrate topography and flow on the efficiency of endothelial regeneration. We exposed endothelial monolayers, grown on topographically engineered substrates (gratings), to controlled levels of flow-induced shear stress. The wound healing dynamics were recorded and analyzed in various configurations, defined by the relative orientation of an inflicted wound, the topography and the flow direction. Under flow perpendicular to the wound, the speed of endothelial regeneration was significantly increased on substrates with gratings oriented in the direction of the flow when compared to flat substrates. This behavior is linked to the dynamic state of cell-to-cell adhesions in the monolayer. In particular, interactions with the substrate topography counteract Vascular Endothelial Cadherin phosphorylation induced by the flow and the wounding. This effect contributes to modulating the mechanical connection between migrating cells to an optimal level, increasing their coordination and resulting in coherent cell motility and preservation of the monolayer integrity, thus accelerating wound healing. We further demonstrate that the reduction of vascular endothelial cadherin phosphorylation, through specific inhibition of Src activity, enhances endothelial wound healing in flows over flat substrates.

The dynamics of shock-bubble interaction involve an interplay of vortex stretching, dilation, and baroclinic vorticity generation. Here, we quantify the interplay of these contributions through high resolution 3D simulations for several Mach and Atwood numbers. We present a volume rendering of density and vorticity magnitude fields of shock-bubble interaction at M = 3 and air/helium density ratio {η} = 7.25 to elucidate the evolution of the flow structures. We distinguish the vorticity growth rates due to baroclinicity, stretching, and dilatation at low and high Mach numbers as well as the late time evolution of the circulation. The results demonstrate that a number of analytical models need to be revised in order to predict the late time circulation of shock-bubble interactions at high Mach numbers. To this effect, we propose a simple model for the dependence of the circulation to Mach number and ambient to bubble density ratio for air/helium shock-bubble interactions.

Reaction-diffusion systems on the surface and the interior of complex domains are potent models of growth in living organisms. The simulation of these models requires numerical methods capable of handling large deformations and the accurate coupling of the evolution of substances in the lumen and on the surface of the deforming geometries. Here, we develop a novel computational method to handle such problems by combining a remeshed particle method with a phase field method. Remeshed particle methods are well suited to discretizing deforming geometries, while the phase field method is used to impose boundary conditions that effectuate the coupling of substances evolving in their lumen and on their surfaces. We demonstrate that this hybrid method enables for the first time the accurate coupling of reaction-diffusion on a deformable surface and its interior. The method is validated on benchmark problems and the effect of lumen diffusion to a pattern forming reaction-diffusion system on a deforming surface is discussed.