Dakota Reference Manual
Version 6.4
LargeScale Engineering Optimization and Uncertainty Analysis

Experimental capability to generate a random field representation. from data, from simulation runs, or from a covariance matrix. The representation may then be sampled for use as a random field input to another simulation. THIS IS AN EXPERIMENTAL CAPABILITY.
Alias: none
Argument(s): none
Required/Optional  Description of Group  Dakota Keyword  Dakota Keyword Description  

Optional  build_source  Specify how the random field will be built: from a data file, from simulation runs, or from a covariance matrix. THIS IS AN EXPERIMENTAL CAPABILITY.  
Optional  expansion_form  Specify the form of the expansion to be used in the random field representation. THIS IS AN EXPERIMENTAL CAPABILITY.  
Optional  expansion_bases  Specify the number of basis functions to be used in the random field representation. THIS IS AN EXPERIMENTAL CAPABILITY.  
Optional  truncation_tolerance  Specify a percent of the response variance that should be captured with the random field representation. THIS IS AN EXPERIMENTAL CAPABILITY.  
Required  propagation_model_pointer  Pointer to the model that will accept realizations of the random field and use them for subsequent analysis. Typcially, this model will take the random field as inputs, e.g. a random field defining a pressure boundary or temperature boundary condition over a structure. THIS IS AN EXPERIMENTAL CAPABILITY. 
Capability to generate a random field representation from data, from simulation runs, or from a covariance matrix. The random field may then be sampled for use as a random field input to another simulation. THIS IS AN EXPERIMENTAL CAPABILITY UNDER ACTIVE DEVELOPMENT.
Default Behavior There are three main sections of the random_field
model. The first section tells Dakota what data to use to build the random field. This is specified with build_source
. The source of data to build the random field may be a file with data (where the N rows of data correspond to N samples of the random field and the M columns correspond to field values), or it may be a simulation that generates field data, or it may be specified given a mesh and a covariance matrix governing how the field varies over the mesh. In the case of using a simulation to generate field data, the simulation is defined with dace_method_pointer
. In the case of using a mesh and a covariance, the form of the covariance is defined with analytic_covariance
.
The next section of the random fiel model specifies the form of the expansion, expansion_form
. This can be either a KarhunenLoeve expansion or a Principal components analysis. These are very similar: both involve the eigenvalues of the covariance matrix of the field data. The only difference is in the treatment of the estimation of the coefficients of the eigenvector basis functions. In the PCA case, we have developed an approach which makes the coefficients explicit functions of the uncertain variables used to generate the random field. The specification of the random field can also include the number of bases to retain or a truncation tolerance, which defines the percent variance that the expansion should capture.
The final section of the random field model allows the user to specify a pointer to a model over which the random field will be propagated, meaning the model which will be driven with the random field input. This part of the specification is optional: one can build a random field but not use it in a downstream model.
As stated above, this is an emerging capability. The syntax currently looks like the following:
random_field build_source rf_data_file  dace_method_pointer  analytic_covariance expansion_form karhunen_loeve  principal_components expansion_bases truncation_tolerance propagation_model_pointer