Networks of PDE Solvers 

A multi-physics Partial Differential Equation (PDE) problem involves different physics, on different regions of an object with possible different scales. Splitting of a domain is essential due to the physical characteristics and computation efficiency. An initial domain is decomposed into two or more sub-domains each with its own local boundaries. An interface is a boundary common to two sub-domains. 
 
We use the Interface Relaxation method, to solve a multi-physics problems. There are two steps in this method;  (a) given some boundary conditions, a PDE is solved locally in each sub-domain, and (b) we smooths out a solution or its derivative iteratively. The two steps iterate until the convergence criteria is satisfied. 

Using the Bond agent framework, we have created a network of PDE solvers. Once a domain is decomposed, each sub-domain PDE is mapped to a PDESolver agent, an interface to a PDEMediator, and a PDECoordinator agent controls the entire computation and examines the convergence. Each agent has a set of states coupled with strategies. We provide the code for the three types of agents as well as the bluprints used by an agent factory to create the agents. 

This application works together with the Pellpack (Parallel Ellpack) system. 
A solver agent is a wrapper capable to start up a PDE solver in Pellpack. 

• Source Code
• Downloadable Java Class File
• User Guide (HTML) (postscript)