.. _method-sampling-d_optimal:

"""""""""
d_optimal
"""""""""


Generate a D-optimal sampling design


.. toctree::
   :hidden:
   :maxdepth: 1

   method-sampling-d_optimal-candidate_designs
   method-sampling-d_optimal-leja_oversample_ratio


**Specification**

- *Alias:* None

- *Arguments:* None

- *Default:* off


**Child Keywords:**

+-------------------------+--------------------+---------------------------+-----------------------------------------------+
| Required/Optional       | Description of     | Dakota Keyword            | Dakota Keyword Description                    |
|                         | Group              |                           |                                               |
+=========================+====================+===========================+===============================================+
| Optional (Choose One)   | Design Strategy    | `candidate_designs`__     | Number of candidate sampling designs from     |
|                         |                    |                           | which to select the most D-optimal            |
|                         |                    +---------------------------+-----------------------------------------------+
|                         |                    | `leja_oversample_ratio`__ | Oversampling ratio for generating candidate   |
|                         |                    |                           | point set                                     |
+-------------------------+--------------------+---------------------------+-----------------------------------------------+

.. __: method-sampling-d_optimal-candidate_designs.html
__ method-sampling-d_optimal-leja_oversample_ratio.html



**Description**


This option will generate a sampling design that is approximately
determinant-optimal (D-optimal) by downselecting from a set of
candidate sample points.

*Default Behavior*

If not specified, a standard sampling design (MC or LHS) will be
generated.  When ``d_optimal`` is specified, 100 candidate designs will
be generated and the most D-optimal will be selected.

*Usage Tips*

D-optimal designs are only supported for
:ref:`variables:uncertain:auv`.  The default candidate-based
D-optimal strategy works for all submethods except incremental LHS (
``lhs`` with ``refinement_samples``).  The Leja sampling option only works
for continuous variables, and when used with LHS designs, the
candidates point set will be Latin, but the final design will not be.




**Examples**



.. code-block::

    method
      sampling
        sample_type random
        samples = 20
        d_optimal