Geometry and Meshing Support for Shape Optimization

A new ITAPS/TOPS/ SLAC collaboration focuses on optimizing the shape of accelerator cavities for the International Linear Collider (ILC). At various optimization stages, the code a) generates a cavity geometry based on a set of design parameters, b) projects a mesh onto that geometry and smooths it to improve quality, c) computes the derivative of surface vertex positions with respect to design parameters, and d) passes these results to the function evaluation (an electro-magnetic analysis) through the mesh interface. These capabilities use implementations of the geometry, mesh, and relations interfaces, along with Mesquite mesh improvement services, and are packaged in the DDRIV code library and executable driver.

We will continue construction and application of a higher-level shape optimization component along two specific fronts. First, we will continue the collaboration with SLAC to apply shape optimization to accelerator cavities used in the International Linear Collider (ILC). Second, we propose generalizing and enhancing this capability, so that this work can be leveraged in future optimization efforts.

Since function evaluations in the optimization for SciDAC applications are highly parallelized, driving them as part of an optimization loop requires all software components in the optimization process be parallel as well. Furthermore, these components should be automated such that no manual intervention is required. DDRIV will be parallelized to facilitate automation of the optimization loop. We will investigate whether to use the same partition for DDRIV functions as that used by the application or a different partition. In the longer term, methods for treating larger parameter variations, leading to topological changes to the underlying geometric model, will be investigated. This may require the introduction of re-meshing and/or local mesh topological changes to the process to improve mesh quality. This research will use the dynamic mesh services, parallel mesh generation, and parallel deforming mesh capabilities discussed in this proposal. DDRIV will be released as open source early in this process, and we will seek other applications of this technology.