A Direct Error Measure for Affine Models of Nonlinear Algebraic Systems

June 2, 1999 - 12:00pm
Bldg. 90

The Newton-Raphson solution of a nonlinear system iterately linearizes the equations, then steps to the solution of the resulting affine model. When a step exceeds the predictive range of its model, the method can diverge.  The traditional response -- aggregating the equations into a cost function, and applying a minimization method -- suppresses information about how each equation model performs. Direct error measures examine the equations individually, allowing finer control over step lengths.  The seminar will develop one such measure through the geometry of simple one- and two-dimensional examples, then present results from a suite of larger systems. Airflow problems typical of building energy solvers exhibit special properties that reduce the computational cost of applying such methods.

Add event to Google Calendar

Announcement List

For notification about future seminars, contact Erin Bertiglia.

Site Access

Schedule subject to change without notice. If you are coming from off-site, please call first to verify. UC staff and guests are welcome. LBNL shuttle buses stop every few minutes at marked sidewalk locations along Bancroft and Hearst Avenues and Rockridge BART.