Algorithms - Steepest Descent

The OSTRICH steepest descent optimization algorithm

Initial Publication:

Modified:

The following optional group will configure the gradient-based Steepest-Descent algorithm and will be processed if [ProgramType] is set to “Steepest-Descent”.

BeginSteepDescAlg
ConvergenceVal    [conv_val]
MaxIterations     [max_iter]
EndSteepDescAlg
BeginSteepDescAlg
ConvergenceVal    1.00E-6
MaxIterations     20
EndSteepDescAlg

Figure 1: General Format (left) and Example (right) of the Steepest-Descent Group

Where BeginSteepDescAlg and EndSteepDescAlg are parsing tags that wrap a set of algorithm configuration variables.

  • ConvergenceVal: This is the algorithm convergence value – searches will halt when the relative reduction over three successive iterations is less that this value. The default value is 1E-6.
  • MaxIterations: The maximum iterations in the overall method. The default value is 20.

References

Bertsekas, D. P. 2014. Constrained optimization and Lagrange multiplier methods. Academic press.