Flexible Uncertainty Analysis of Computer Models with Alchemy

摘要

This manuscript introduces the Alchemy code package, which is intended to provide a flexible toolkit that enables the application of Reduced-order Modeling (ROM) methodology to large-scale engineering calculations. To achieve this, Alchemy treats the model code in question as a black box, with a small input-output layer that handles data exchange with the model. This approach has been employed with great success in other codes like Sandia's DAKOTA package and allows a user to realize the large computational savings of using an ROM without the needing to specialize the ROM algorithm for every new code package. Instead, the user only writes a small interface for their code (or uses a pre-existing interface included with Alchemy), while Alchemy handles the remaining tasks like parallel execution, random sampling or actual construction and analysis of the ROM.