Dakota Reference Manual  Version 6.10
Explore and Predict with Confidence
 All Pages
optional_interface_pointer


Pointer to interface that provides non-nested responses

Topics

This keyword is related to the topics:

Specification

Alias: none

Argument(s): STRING

Default: no optional interface

Child Keywords:

Required/Optional Description of Group Dakota Keyword Dakota Keyword Description
Optional optional_interface_responses_pointer Pointer to responses block that defines non-nested responses

Description

optional_interface_pointer is used to specify an optional inferface (using that interface block's id_interface label) to provide non-nested responses to the nested model.

To generate the complete set of primary responses (response functions, objective functions, and calibration terms) expected by the nested model, primary responses from the optional interface may be combined (by addition) with responses derived from sub-iterator results, or they may be presented unchanged to the nested model. The way that Dakota treats primary responses from these two sources depends on the number of rows of and entries in the primary_response_mapping matrix.

To understand how, it is helpful to consider the following expression. Nested model primary responses $r$ are computed from the column vector of optional interface responses $f$, the primary_response_mapping matrix $W$, and the column vector of results from the sub-iterator $s$.

\[ \left\{r\right\} = \left\{ \begin{array}{c} f \\ \hline 0 \end{array} \right\} + \left\{ \begin{array}{c} \left[ W \right] \left\{s\right\} \\ \hline 0 \end{array} \right\} \]

The number of rows in $f$ is equal to the number of primary responses returned by the optional interface, and in $r$, the number of primary responses expected by the nested model. In this expression, $0$ represents "padding" that may be needed by either the optional interface vector or the sub-iterator vector to make them the same dimension as $r$. One or the other of these quantities may be padded, but not both. In terms of the Dakota input, this means that the number of nested model primary responses must equal the larger of these two quantities: the number of optional interface responses or the number of rows in the primary_response_mapping.

The nested model's secondary responses (nonlinear constraints) are the secondary responses from the optional interface augmented by the secondary responses constructed from sub-iterator results. Unlike primary responses from these two sources, which may be combined by adding, it is not possible presently to combine secondary responses. For this reason, the number of nested model's secondary responses must equal the number of secondary responses provided by the optional interface, plus the number provided by the sub-iterator (i.e. the number of rows in the secondary_response_mapping). Secondary responses from the optional interface are inserted first into the nested model's nonlinear constraints; those from the sub-iterator come after. If the nested model has secondary responses and an optional interface is used, it is necessary to specify a separate response block for the optional interface using the optional_interface_response_pointer. Targets and upper and lower bounds for the nonlinear constraints are taken from the nested model's response block.

Examples

The first example illustrates the use of an optional interface and primary response mapping. No secondary responses are present. Assume that the inner method returns two results (suppose its the mean and standard deviation of a single response), so that the dimensions of primary_response_mapping will be interpreted by Dakota as 1x2, just as they are written here.

The nested model (nested) expects two primary responses (outer_resp block), and the optional interface returns two (opt_resp block).

model
  id_model 'nested'
  nested
    optional_interface_pointer 'opt_intf'
      optional_interface_response_pointer 'opt_resp'
    sub_method_pointer 'inner_method'
      primary_response_mapping   1.0 2.0 
    
    responses_pointer 'outer_resp'
    variables_pointer 'outer_vars'

responses
  id_responses 'opt_resp'
  response_functions 2
    descriptors 'opt1' 'opt2'
  no_gradients
  no_hessians

responses
  id_responses 'outer_resp'
  response_functions 2
    descriptors 'out1' 'out2'
  no_gradients
  no_hessians

The first nested model response, $out1$ will be computed $ out1 = opt1 + 1.0*mean + 2.0*stddev $, and the second response $out2$ will be equal to simply $opt2$. In the matrix form introduced above, including the implicit padding:

\[ \left\{ \begin{array}{c} out1 \\ out2 \end{array} \right\} = \left\{ \begin{array}{c} opt1 \\ opt2 \end{array} \right\} + \left\{ \begin{array}{c} 1.0 \cdot mean + 2.0 \cdot stddev \\ 0 \end{array} \right\} \]

The optional_interface_response_pointer was not strictly needed. Because the number of responses returned by the optional interface and expected by the nested model is the same (and no secondary responses are present), the nested model's response specification alone would have sufficed.

If it were desired to combine $opt2$ with the sub-iterator's results and set $out1$ to $ opt1 $ with no combination, then the primary_response_mapping matrix could be changed to:

      primary_response_mapping   0.0 0.0
                                 1.0 2.0 

Secondary responses will be demonstrated in the second example.

model
  id_model 'nested'
  nested
    optional_interface_pointer 'opt_intf'
      optional_interface_response_pointer 'opt_resp'
    sub_method_pointer 'inner_method'
      primary_response_mapping   1.0 2.0 
      secondary_response_mapping 1.0 0.0
    responses_pointer 'outer_resp'
    variables_pointer 'outer_vars'

responses
  id_responses 'opt_resp'
  objective_functions 2
    descriptors 'opt1' 'opt2'
  nonlinear_inequality_constraints 1
    descriptors 'optc'
  no_gradients
  no_hessians

responses
  id_responses 'outer_resp'
  objective_functions 2
    descriptors 'out1' 'out2'
  nonlinear_inequality_constraints 2
    descriptors 'outc1' 'outc2'
  no_gradients
  no_hessians

The nested model expects a total of two constraints. Because the secondary_response_mapping has one row, one of these constraints will be supplied from the sub-iterator. The other is provided by the optional interface. Secondary responses from the optional iterator come first, and so optc is placed in outc1. outc2 takes the value computed from the sub-iterator results.