From left, Sandia computer scientists Brian Adams, Jim Stewart, and John Siirola look over the results of a Dakota optimization study.
Sandia’s Dakota software delivers advanced parametric analysis techniques enabling design exploration, model calibration, risk analysis, and quantification of margins and uncertainty with computational models. Dakota v5.4 was released November 15, 2013, and deployed on supported Sandia computing systems.
The release includes new capabilities for uncertainty quantification (UQ), optimization, and calibration. UQ highlights include a new PoFDarts method for reliability/failure calculations as well as several stochastic expansion improvements—multiple-fidelity, compressive sensing, orthogonal least interpolation, and extensions to higher dimension. Optimization and calibration highlights include a new direct search mixed integer optimization method and extended Bayesian calibration methods.
Also included in Dakota v5.4 is a new capability to export surrogate models, improvements to parallel scheduling, improved testing across release platforms, an initial rollout of a new reference manual format, and more.
Dakota is publicly available under an open source license; is used broadly by academic, government, and corporate institutions; and plays a critical role in DOE Advanced Simulation and Computing, Office of Science, and Nuclear Energy programs.