Next:
9.1 Problem Structure
Up:
Parallel Computing Works
Previous:
8.3.6 Conclusion
9 Loosely Synchronous Problems
9.1 Problem Structure
Geomorphology by Micromechanical Simulations
Plasma Particle-in-Cell Simulation of anElectron Beam Plasma Instability
9.3.1 Introduction
9.3.2 GCPIC Algorithm
9.3.3 Electron Beam Plasma Instability
Performance Results for One-DimensionalElectrostatic Code
9.3.5 One-Dimensional Electromagnetic Code
9.3.6 Dynamic Load Balancing
9.3.7 Summary
9.4 Computational Electromagnetics
LU Factorization of Sparse, Unsymmetric Jacobian Matrices
9.5.1 Introduction
9.5.2 Design Overview
9.5.3 Reduced-Communication Pivoting
9.5.4 New Data Distributions
9.5.5 Performance Versus Scattering
9.5.6 Performance
Order 13040 Example
Order 2500 Example
9.5.7 Conclusions
Concurrent DASSL Applied to Dynamic Distillation Column Simulation
9.6.1 Introduction
9.6.2 Mathematical Formulation
9.6.3 proto-Cdyn - Simulation Layer
Template Structure
Problem Preformulation
9.6.4 Concurrent Formulation
Overview
Single Integration Step
The Integration Computations
Single Residuals
Jacobian Computation
Exploitation of Latency
The LU Factorization
Forward- and Back-solving Steps
Residual Communication
9.6.5 Chemical Engineering Example
9.6.6 Conclusions
9.7 Adaptive Multigrid
9.7.1 Introduction
9.7.2 The Basic Algorithm
9.7.3 The Adaptive Algorithm
9.7.4 The Concurrent Algorithm
9.7.5 Summary
9.8 Munkres Algorithm for Assignment
9.8.1 Introduction
9.8.2 The Sequential Algorithm
9.8.3 The Concurrent Algorithm
Optimization Methods for Neural Nets:Automatic Parameter Tuning and FasterConvergence
9.9.1 Deficiencies of Steepest Descent
9.9.2 The ``Bold Driver'' Network
The Broyden-Fletcher-Goldfarb-Shanno One-StepMemoryless Quasi-Newton Method
9.9.4 Parallel Optimization
9.9.5 Experiment: the Dichotomy Problem
9.9.6 Experiment: Time Series Prediction
9.9.7 Summary
Guy Robinson
Wed Mar 1 10:19:35 EST 1995