OSC Enhancement via Automatic Differentiation for Sensitivity & Uncertainty Analysis

Accurate and efficient sensitivity and uncertainty analysis tools are needed for modeling and simulation codes. Automatic Differentiation (AD) is a process by which an existing computer model is automatically "enhanced" to provide a complete sensitivity (Jacobian) matrix of all code outputs with respect to all inputs in a single run of the enhanced code. Although termed "automatic", the out-of-the-box AD tools require significant skill and manual effort to properly enhance the simulation code. OKSI, teamed with the developer of the AD technology, has created — under previous MDA efforts — a simplified and powerful AD implementation process, appropriately named Automated Automatic Differentiation (A2D). This package improves the automation in several ways: 1) setting up the process, 2) dealing with the AD preprocessor conflicts, 3) establishing a powerful graphical user interface, 4)validating the sensitivity results, and 5) creating error tracing tools to pinpoint the line of code introducing the validation error. A subset of the Optical Signatures Code (OSC) was used as a testbed for the development of the A2D. As such, extensive experience was gained in working with OSC, understanding the potential pitfalls in differentiating the code, and establishing a proof of feasibility in the process. The capability was recently demonstrated with the complete suite of JANNAF low altitude plume codes. OKSI has teamed with the OSC developer, Teledyne Solutions Inc. (TSI), who will provide the guidance to assure that AD-OSC best meets the needs of the large OSC user base. TSI will also support independent verification and validation (IV&V). With the given Phase-I head start, during Phase-II OKSI / TSI will focus on producing an AD-enhanced version of OSC (AD-OSC) that encompasses the entire OSC code. AD-OSC will be able provide sensitivity/uncertainty analysis for any OSC scenario. The end product of the Phase-II will be AD-OSC ready for distribution to the user community. Under additional Optional tasks, OKSI will also implement AD-OSC into an inverse problem solution — called ASCOPE — for the extraction of model parameters when matching code predictions with sensor measurements. Under a second Option, OKSI will apply AD to the OPTISIG code and perform extended testing and validation on the AD-OSC/OPTISIG codes.