A reduced-basis element method (Q5915527)

Given a partial differential equation on a region \(\Omega\) along with boundary values, the authors seek an approximate solution as a linear combination of a set of predetermined trial functions. The region \(\Omega\) is presumed to be the image of a simple ``reference domain'' \(\mathcal B\) \textit{e.g.}, a rectangle. The authors propose choosing the trial functions as solutions of the given equation on other images of \(\mathcal B\), mapped via \(\mathcal B\) into \(\Omega\). An example is presented for Laplace's equation on a quarter of an annular ring for which a solution can be found with an error less than one percent using only eleven trial functions. The authors are also interested in the case that \(\Omega\), while not itself the image of \(\mathcal B\), can yet be decomposed into a number of nonoverlapping subdomains, each of which is an image of the reference domain \(\mathcal B\). Solutions on abutting subdomains can be joined using a Lagrange multiplier on functions defined on the common subdomain boundary. Some additional care must be taken on boundaries between abutting subdomains and also that the predetermined trial functions properly represent these boundaries in order that the error near the boundaries not become large. The authors' approach is intended for use when significant resources are expended in constructing the predetermined trial functions that are then stored for reuse. The stored trial functions can then be used repeatedly to construct many related solutions.