Ernesto Bonomi

A stochastic cellular automaton simulation of the non-linear diffusion equation

Abstract In this article we investigate a cellular automaton simulation of the non-linear diffusion equation. The diffusion coefficient characterizing the equation is used to locally bias the random walks of particles on a square lattice. Emphasis is placed on respecting the massively parallel nature of the automaton model, while also correctly simulating the macroscopic behavior described by the equation. The result, a highly parallel algorithm, presents an interesting possibility for parallel and dedicated machines.

3D spectral reverse time migration with no-wraparound absorbing conditions

Comparative studies of methods of reverse time migration (RTM) show that spectral methods for calculating the Laplacian impose the least stringent demands on discretization stepsize; thus with spectral methods, the grid re nements often required by other methods can be avoided. Implemented with absorbing boundary conditions, which are energy-tuned to give good absorption at the boundaries, these spectral methods can be used e ectively for migration, without su ering the problems of wraparound which have traditionally plagued them (Furumyra and Takenaka, 1995).

A stochastic cellular automaton model of non-linear diffusion and diffusion with reaction

Abstract This article presents a stochastic cellular automaton model of diffusion and diffusion with reaction. The master equations for the model are examined, and we assess the difference between the implementation in which a single particle at a time moves (asynchronous dynamics) and one implementation in which all particles move simultaneously (synchronous dynamics). Biasing locally each particles random walk, we alter the diffusion coefficients of the system. By appropriately choosing the biasing function, we can impose a desired non-linear diffusive behaviour in the model. We present an application of this model, adapted to include two diffusing species, two static species, and a chemical reaction in a prototypical simulation of carbonation in concrete.

GRIDA3—a shared resources manager for environmental data analysis and applications

GRIDA3 (Shared Resources Manager for Environmental Data Analysis and Applications) is a multidisciplinary project designed to deliver an integrated system to forge solutions to some environmental challenges such as the constant increase of polluted sites, the sustainability of natural resources usage and the forecast of extreme meteorological events. The GRIDA3 portal is mainly based on Web 2.0 technologies and EnginFrame framework. The portal, now at an advanced stage of development, provides end-users with intuitive Web-interfaces and tools that simplify job submission to the underneath computing resources. The framework manages the user authentication and authorization, then controls the action and job execution into the grid computing environment, collects the results and transforms them into an useful format on the client side. The GRIDA3 Portal framework will provide a problem-solving platform allowing, through appropriate access policies, the integration and the sharing of skills, resources and tools located at multiple sites across federated domains.

Data-parallel molecular dynamics: 1-D hard-core fluid

Abstract Molecular dynamics constitute a family of techniques for solving large classical N -body problems under a variety of physical conditions. Modern massively parallel machines allow us to approach the simulation of matter at the atomistic level, thus enlarging the scope of computer modeling. Hard-core impulsive evolutions are intrinsically asynchronous and do not parallelize easily on multicomputers or vector machines. Using the 1-D hard-core fluid as a paradigm, a data-parallel molecular dynamics algorithm with at least 16384 particles has been implemented on a Connection Machine CM-2.

Blocking probability for a multistage clos connecting network

The blocking probability is computed by assuming a thermodynamic limit when the number of stages increases above a certain value. In this limit we exhibit a set of two algebraic equations which gives the blocking probability as a function of the traffic demand. A comparison with a computer simulation of the system gives an excellent agreement.

Cellular automata — lattice gas models for PDE's

Abstract Lattice gases, one class of cellular automaton model, are increasingly under investigation as a highly parallel means of simulating given partial differential equations. We examine stochastic lattice gas simulations, turning our attention particularly toward diffusive processes: nonlinear diffusion, diffusion with reaction, diffusion with advection. We summarize our results here, directing the reader to previous articles for more details. Numerical studies are included, comparing lattice gas results against more traditional numerical methods.

Experiences with HPF for Scientific Applications

The data-parallel programming paradigm emerges as the natural choice for a large class of scientific applications. One of its most significant features consists in hiding the details of the communications and synchronizations among processors, leaving them to the compiler. The advent of the new standard data-parallel language, High Performance Fortran (HPF), promises to bring portability of data-parallel codes across different parallel architectures.

Migration of seismic data

Prospecting for oil and gas resources poses the problem of determining the geological structure of the earths crust from indirect measurements. Seismic migration is an acoustic image reconstruction technique based on the inversion of the scalar wave equation. Extensive computation is necessary before reliable information can be extracted from large sets of recorded data. In this paper a collection of “industrial” migration techniques, each giving rise to a data parallel algorithm, is outlined. Computer simulations on synthetic seismic data illustrate the problem and the approach.

Isothermal Molecular Dynamics: A Practical Study

In a molecular dynamics simulation, trapping of the phase space trajectory can occur at low energy for long time intervals. To untrap that evolution and recover the canonical ensemble description, at regular clock hits, all particle velocities are randomly redistributed according to the Maxwell statistics. The organization of this hybrid evolution mixing molecular dynamics and random number generation leads to a fully parallel and synchronous algorithm.An illustration of this approach on the one-dimensional jellium is presented. For large values of the coupling λ for which the evolution at constant energy is trapped, the hybrid method allows to reproduce the exact Boltzmann statistical description. Computer simulations show that the Nose-Hoover method does not really improve the pure molecular dynamics for the one dimensional jellium.