![]() |
Dakota Reference Manual
Version 6.16
Explore and Predict with Confidence
|
Initial step size for derivative-free optimizers
Alias: none
Argument(s): REAL
Default: 1.0
The initial_delta keyword defines the size of the first search step in derivative-free optimization methods, specifically asynch_pattern_search, coliny_cobyla, coliny_pattern_search, coliny_solis_wets, and mesh_adaptive_search. It is applied in an absolute sense to all search directions.
Default Behavior
The default value is 1.0.
Usage Tips
It is recommended that initial_delta be the approximate distance from the initial point to the solution. If this distance is unknown, it is advisable to err on the side of choosing an initial_delta that is too large or to not specify it. Relative application of initial_delta is not available unless the user scales the problem accordingly.
Three example method input blocks appear below.
For asynch_pattern_search:
method
asynch_pattern_search
initial_delta = .5
contraction_factor = 0.25
merit_function merit1_smooth
smoothing_factor = 1.0
constraint_tolerance = 1.e-6
For coliny_pattern_search:
method
coliny_pattern_search
initial_delta = .2
variable_tolerance = 1.e-4
max_iterations 100
solution_accuracy = 1.e-6
seed = 1234
max_function_evaluations = 1000
For mesh_adaptive_search
method
mesh_adaptive_search
initial_delta = 2.0
seed = 1234