Predictive computational models used for stockpile stewardship studies require sophisticated models simulated on the world’s largest computers.  These models are complex; hence advanced verification (“solving the equations right”) and validation (“solving the right equations”) methodologies are needed to assess their accuracy and predictive capability. In every major ASC simulation code, complex subsystems interact in complex ways to form cohesive computer programs that predict important physical processes.  Sophisticated, component-based software enables analysts to unit test and verify the codes even if there are major improvements and changes to the subsystems of the code. Finally, even with verified numerical algorithms (in the physics and software sense) and validated physical models, the uncertainty of the model/algorithm and its sensitivity to change must be better understood.

A priority of LANL’s mission is to validate and verify (“V & V”) the complex computer programs used to model equally complex physical processes. One of the major techniques employed in the verification and validation process is sensitivity calculation. Consequently, the aim of the code-based sensitivity analysis project is to develop methods for accurately and efficiently computing sensitivities of complex scientific simulation programs. Three projects at LANL are the principal targets for code-based sensitivity analysis over the short term: the Telluride Project, the Shavano Project, and the Marmot Project.