Dakota Reference Manual
Version 6.12
Explore and Predict with Confidence

Variance applied to simulation responses
Alias: none
Argument(s): REALLIST
Default: no variance
The variance that is applied to simulations run by Dakota, i.e. using the analysis_drivers
command. The user may supply a single variance or a vector of variances of length equal to the number of responses. In both cases, the values provided are treated as scalar variance types. If a single variance is provided, it is applied to all responses produced by the simulation code. If a vector is provided, each variance is applied to the corresponding response output produced by the simulation code.
It is important to note that the the variance defined by this keyword differs from that defined using experiment_variance_type
. These two commands apply to userprovided calibration data, specified, for example, by calibration_data
or calibration_data_file
. However, simulation_variance
applies to those responses produced by simulation code that is run by Dakota, as described above.
Usage Tips
Currently, this keyword is only in use as part of the algorithm implemented by experimental_design. In this algorithm, two models (usually, one highfidelity and one lowfidelity) are provided to Dakota, each with their own responses
section of the input script, and each of which is allowed its own simulation_variance
. The variance specified in the responses
block belonging to the highfidelity model is applied to any new highfidelity data that is produced by Dakota running the highfidelity model. In the experimental_design
algorithm, lowfidelity model responses are used during the calibration of the model parameters, the calculation of the mutual information, and the calculation of any posterior statistics after the algorithm is complete. The simulation_variance
is applied to the lowfidelity model responses that are used in the calculation of the mutual information. See the User's and Theory Manuals for more information.
The example below shows two responses
blocks, one for the lowfidelity model and one for the highfidelity model. Both contain simulation_variance
commands that will apply to the low and highfidelity model responses, respectively.
responses, id_responses = 'lowfidelity' calibration_terms = 1 simulation_variance = 0.5 responses, id_responses = 'highfidelity' calibration_terms = 1 calibration_data_file = 'dakota_bayes_expdesign.dat' freeform num_config_variables = 1 num_experiments = 1 experiment_variance_type = 'none' simulation_variance = 1.2