News & Announcements

March 20, 2012

"Argonne researchers release toolkit for solving large-scale optimization applications"

Large-scale optimization problems arise in many applications, including nuclear reactor simulation, fluid dynamics, parameter estimation, and optimal control. Researchers seeking to solve such problems on high-performance architectures have faced numerous challenges, ranging from scattered support for parallel computation and lack of reuse of linear algebra software to the reality of working with large, often poorly structured legacy codes for specific applications. The Toolkit for Advanced Optimization (TAO) developed at Argonne National Laboratory is designed to address these challenges.

The latest release, TAO 2.0, provides several new algorithms, including POUNDerS, for solving nonlinear least squares problems when no derivatives are available and function evaluations are expensive, and LCALM, for solving optimization problems with partial differential equation constraints based on a linearly-constrained augmented Lagrangian method. Also included is the capability for any of the TAO line search methods to be selected regardless of the overlying TAO algorithm. Moreover, users can create new line search algorithms that may be more suitable for their applications.

TAO is built on top of the PETSc framework to enable reuse of external tools, and several changes have been made in the new TAO release to achieve a tighter association with PETSc design principles.  Moreover, TAO no longer has separate abstract classes; rather, the PETSc objects are now used directly, making TAO applications much easier to create and maintain for users familiar with PETSc programming.

TAO is suitable for both single-processor and massively parallel architectures. Recent applications using TAO include time-dependent density functional theory for quantum chemistry, laser-induced thermotherapy in biomedical engineering, image classification in machine learning, two-Skyrmion interactions in physics, and crack formation in materials science.

Development of TAO is supported by the Office of Advanced Scientific Computing Research, Office of Science, U.S. Department of Energy. The TAO 2.0 Users Manual (ANL-MCS-TM-322, January 20, 2012), as well as information about applications, publications, and licensing the open source software, is available at the TAO website, http://mcs.anl.gov/tao.

TAO 2.0 was developed by Todd Munson, Jason Sarich, and Stefan Wilde in Argonne’s Mathematics and Computer Science Division.