Abstracts

| A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P | Q | R | S | T | U | V | W | X | Y | Z |

D

Scientific computing using streaming processors

Eric Darve

Abstract:
In recent years, several companies such as NVIDIA, ATI, ClearSpeed, and IBM-Toshiba-Sony have released a new generation of processors based on a streaming architecture. These processors achieve an unprecedented level of performance by using a large number of light cores able to concurrently process data. The most spectacular success in that direction are graphics cards from NVIDIA and ATI. Even though they were originally developed for graphics applications, they are capable of many general purpose computing tasks. We have been investigating the use of these processors for scientific computing. We will outline the capabilities of these processors and the constraints they impose on the type of algorithms which can run efficiently. These processors require significantly rethinking the way algorithms and data structures are designed. In our last project, we have developed code to solve the compressible Euler equations on a GPU. We will describe some of the techniques developed for this simulation. We performed a simulation of a 3D hypersonic vehicle at Mach 5 with 1.5 million mesh nodes. The measured speed-up was 20x compared to a conventional one core processor.

F

Stability of Asynchronous Variational Integrators

Will Fong

Abstract:
The formulation of multiple-time-step integrators can provide substantial computational savings for mechanical systems with multiple time scales. However, the scope of these savings may be severely limited by the range of allowable time step choices. We have performed an exhaustive study of the linear stability of the fully asynchronous methods called AVI (asynchronous variational integrator), with two time steps, for essentially any combination of their values. In addition we have obtained approximate analytical expressions for the time step ratios that may render the scheme unstable in the case of linear equations, and verified them with extensive numerical computations. Synchronous multiple time stepping schemes such as r-RESPA show resonances when the outer step is a multiple of the effective half period of one of the fast oscillators. An elegant generalization is derived in the fully asynchronous case of AVI.

K

Discrete Variational Fluids

Eva Kanso

Abstract:
The variational principles for incompressible fluid mechanics are best expressed in a Lagrangian formalism. However, computational efficiency calls for an Eulerian treatment of fluids to avoid the numerical issues inherent to deforming meshes. An Eulerian variational treatment of discrete fluids is extremely delicate due to restriction on variations in the Eulerian formalism. In this talk, we will present our recent attempts towards developing discrete variational fluids and discuss our successes and failures. In particular, we propose a computationally-tractable discretization of the fundamental configuration space, i.e., the group of volume preserving diffeomorphisms, in terms of the group of doubly stochastic matrices. Alas, a fully-Eulerian variational formulation based on this novel discretization fails to produce an accurate fluid integrator. We then propose the use of a material representation of the fluid motion (a discrete inverse map) and discuss how this idea may be used to develop discrete variational fluids.

DMOC and Robotic Systems

Marin Kobilarov

Abstract:
The talk is about the optimal control of systems with symmetries and nonholonomic constraints in a discrete mechanics setting. Along with a standard direct approach, an alternative solution method is introduced based on higher-order conditions. Computational efficiency is addressed in terms of the choice of formulation (direct vs indirect), coordinate (free) representation, choice of initial trajectory, and trajectory refinement scheme. Applications to rigid body groups and to simple robotic models (an RC helicopter and a car-like robot) are presented.

L

A discontinuous-Galerkin-based immersed boundary method

Adrian Lew

Abstract:
A numerical method to approximate partial differential equations on meshes that do not conform to the domain boundaries is introduced. The proposed method is conceptually simple and free of user-defined parameters, and has optimal order of convergence. This is shown through numerical experiments in reaction-diffusion and elasticity problems.

---

Variational integrators and optimal control of constrained systems

Sigrid Leyendecker

Abstract:
From the variety of methods to enforce holonomic constraints in the framework of Lagrangian dynamics, the focus is on the Lagrange multiplier method and a null space method, both yielding exact constraint fulfilment. These methods can be used in conjunction with a variational integrator leading to a discrete trajectory which, besides respecting a discrete symplectic structure and conserving discrete momentum maps exactly, also shows good energy behaviour. However the two methods differ significantly in the dimension of the system of nonlinear equations, condition number of the iteration matrix during the iterative solution procedure and computational costs, whereby the discrete system resulting from the discrete null space method perform excellently in the mentioned categories. In particular, it has the minimal possible dimension making it exceptionally well-suited to be used as equality constraints in a nonlinear optimisation problem which determines the actuation forces. Together with initial and final conditions on the configuration and conjugate momentum, they serve as nonlinear equality constraints for the minimisation of a given cost functional. The algorithm yields a sequence of discrete configurations together with a sequence of actuating forces, optimally guiding the system from the initial to the desired final state.

M

Foliation Processing

Alex McKenzie, Patrick Mullen

Abstract:
We present a purely Eulerian framework for variational geometry processing of surfaces and foliations. Instead of focusing on a single isosurface as in the Level Set method (LSM), we process all isosurfaces at once (i.e., the foliation induces by the level set) to achieve a variational treatment of basic interface motions like mean curvature flow (MCF). At the core of our approach is the use of the Coarea Formula (a classical result of Geometric Measure Theory) to express area integrals over isosurfaces as volume integrals. With this formula, our treatment of MCF no longer has to rely on LSM-like finite differences, but is achieved via a true minimization of "Eulerian surface area" of the foliation-based interface. A simple Petrov-Galerkin approach to more general gradient flows will be presented. Numerical tests demonstrate an improved accuracy compared to LSM, and we hope to exploit these initial results to offer a better handling of free surface flows with surface tension, without the traditional use of smoothed Dirac functions at the interface (and the delicate trade-offs involved in their use).

N

Constructing point vortex equilibria via Brownian ratchets

Paul Newton

Abstract:
We pose the problem of how to find the positions and strengths of particles in the plane that interact via a logarithmic Hamiltonian so that the configuration moves as a rigid body (i,e. relative equilibria). If written as a dynamical system for the interparticle distances, the problem can be formulated as one in linear algebra. Solve , where is an by non-normal () `configuration matrix', and is the vector of particle strengths. We do this by performing a singular value decomposition on (i.e. calculate the eigenvalues of the covariance matrix ) for an arbitrary initial configuration, then use the smallest non-zero singular value as a ratchet, driving it to zero (guaranteeing that will have a non-empty nullspace) by allowing the particles to undergo a random walk in and keeping only those new configurations that decrease the smallest singular value until it drops below some convergence threshold. The final distribution of non-zero singular values allows us to calculate the Shannon entropy of the configuration (a measure of its disorder), its size as measured by Frobenius norm, and the `distance' between two different equilibrium configurations. Joint work with George Chamoun.

P

Discrete Mechanics and Optimal Control for the Compass Biped

David Pekarek

Abstract:
A methodology for generating locally optimal control policies for simple hybrid mechanical systems is presented and demonstrated on the compass gait biped. Principles from discrete mechanics are utilized so that optimal control policies may be generated as the solution to constrained nonlinear optimization problems. In the context of bipedal walking, this procedure provides a comparative measure of the suboptimality of existing control policies. Furthermore, the methodology can potentially be used as a tool for designing controlled trajectories that are locally optimal with respect to both continuous and discrete behaviors.

S

DEC and AVIs for Computational Electromagnetism

Ari Stern

Abstract:
In classical field theory, Maxwell's equations have an elegant, coordinate-free Lagrangian formulation in terms of spacetime differential forms. However, many methods of computational electromagnetism fail to preserve this geometric structure, resulting in spurious modes and other numerical difficulties. Using discrete exterior calculus (DEC) to discretize the classical spacetime forms, we derive a general family of variational integrators for Maxwell's equations. These methods include the well-known Yee scheme as a special case, but also generalize to irregular spacetime meshes, including a new asynchronous variational integrator (AVI) for electromagnetism. These methods preserve both the variational structure and the differential forms structure of the continuous theory, and consequently, they are multisymplectic, have discrete gauge symmetry, and conserve the associated momentum map (i.e. conservation of charge).

W

Cutoff phenomena in Markov Chain models of chaotic mixing

Matt West

Abstract:
We explore the connections between mixing in fluid flows, chaotic maps and Markov Chains. For Markov Chain mixing, the cutoff phenomenon has been much studied in recent years, whereby the system experiences a sharp transition from unmixed to mixed as time evolves. We consider chaotic maps and fluid flows in the limit of zero diffusion and provide analytical and numerical evidence to support the conjecture that cutoff also occurs in such systems.