Glossary

This page defines terms used throughout the Dakota documentation. Each entry links to the page that covers the topic in depth.

aleatory uncertain variables

Uncertain variables that represent inherent randomness, which more information can’t reduce. They are described by probability distributions, such as normal or uniform. See Aleatory Uncertain Variables.

analysis driver

The program Dakota runs for each evaluation to map parameters to responses. It can be a script, such as Python or bash, or a workflow built in the Dakota GUI. Set with analysis_drivers. See Connecting Your Code: The Analysis Driver and Using Next-Gen Workflow as a Dakota Analysis Driver.

black box
black-box simulation

A simulation that Dakota treats only as a mapping from variables to responses. Dakota needs nothing else about the code, however the data is exchanged. See Connecting Your Code: The Analysis Driver.

dakota_parser

A command-line tool, in the bin folder of a Dakota installation, that converts a freeform format input file to the JSON format: dakota_parser <file> --json. See JSON Format Input Files.

design variables

Variables you control, such as dimensions or operating settings, which optimization and parameter studies vary. See Design Variables.

direct interface

An interface in which Dakota calls the simulation inside its own process, with no parameters file or results file. It covers Dakota’s built-in test functions, such as rosenbrock, and codes compiled into Dakota. See direct and Direct Simulation Interface.

dprepro

A template processor included with Dakota. It fills a template file with the variable values from a parameters file, either from the command line or through the dakota.interfacing Python module. See Preprocessing with dprepro and pyprepro.

ensemble

A set of runs of the same simulation with different parameter values. Dakota chooses the values, manages the runs, and analyzes the responses. See What is Dakota?.

environment

The input-file block for settings that apply to the whole study, such as tabular data file output. It’s optional, and a study has at most one. See environment.

epistemic uncertain variables

Uncertain variables that represent lack of knowledge, which more information could reduce. They are often described by intervals rather than probability distributions. See Epistemic Uncertain Variables.

evaluation

One mapping of a set of variables values to responses, for example one run of the analysis driver. A method requests many evaluations during a study. See Interfaces.

evaluation concurrency

The (maximum) number of evaluations Dakota runs at the same time. Set with evaluation_concurrency under asynchronous. See Parallel Computing.

fork interface
system interface

Interfaces that run an analysis driver as a separate process and exchange data through a parameters file and a results file. Use fork; system is not recommended. See fork and Simulation Interfaces.

freeform format

One of the two supported syntaxes for Dakota input files, run with dakota -i file.in. It describes the same study structure as the JSON format. See Freeform Input Files.

interface

The input-file block that tells Dakota how to perform an evaluation, for example which analysis driver to run and how to exchange data with it. A study needs at least one. See interface and Interfaces.

JSON format

One of the two supported syntaxes for Dakota input files, run with dakota -json file.json. It describes the same study structure as the freeform format as nested JSON objects, which scripts and tools can easily generate and validate. See JSON Format Input Files.

library API

Python and C++ programming interfaces for building and running Dakota studies inside another application, instead of running Dakota from the command line. See Library APIs.

method

The input-file block that selects the algorithm, such as sampling, optimization, or calibration, and its settings. A study needs at least one. See method and Study Types.

model

The input-file block that connects a method to variables, an interface, and responses. It’s optional in simple studies; other model types build surrogates or nest studies. See model and Models.

outer-loop analysis

An analysis that runs a simulation many times, with Dakota choosing the inputs for each run and analyzing the results, such as optimization, uncertainty quantification, sensitivity analysis or calibration. See What is Dakota?.

parameters file

The file Dakota writes for each evaluation through a fork interface. It holds the current variables values and which response data the method needs. See Parameter File Format.

post-processing

The step of an analysis driver that runs after your code. It extracts the responses from your code’s output and writes them to the results file.

pre-processing

The step of an analysis driver that runs before your code. It puts the variable values from the parameters file into your code’s input, usually with a template file and dprepro.

pyprepro

A template processor included with Dakota. It fills in a template file with values you give it, and templates can use Python features such as loops and conditionals. dprepro uses the same template engine, and also reads Dakota’s parameters file. See Preprocessing with dprepro and pyprepro.

Python direct interface

An interface in which Dakota calls a function in your Python module and passes data in memory, with no parameters file or results file. Set with python; available only in Dakota builds that include it. See Python.

responses
quantities of interest
QoIs

The outputs Dakota analyzes, such as objective functions, calibration terms, or general response functions, and optionally their derivatives. Also the input-file block that describes them; a study needs at least one. See responses and Responses.

restart file

A binary file, dakota.rst by default, in which Dakota records each completed evaluation, so an interrupted study can continue without repeating work. See Restarting Dakota.

results file

The file the analysis driver writes for each evaluation, containing the responses Dakota requested. See Dakota Results File Data Formats.

state variables

Variables for fixed model settings, such as solver tolerances or mesh resolution, that are neither designed nor treated as uncertain. See State Variables.

study

One complete Dakota analysis, described by an input file made of environment, method, model, variables, interface, and responses blocks. See Dakota Input File.

surrogate

An inexpensive approximation of a costly simulation, such as a Gaussian process, built from evaluation data and used in place of the simulation. Defined in the model block. See Surrogate Models.

tabular data file

A text file listing the variables and responses of every evaluation, one row each. Written when you specify tabular_data. See Tabular Output Data.

template file

A copy of your code’s input file with placeholders, such as {x1}, where variable values go. dprepro fills them in for each evaluation. See Building a Black-Box Interface to a Simulation Code.

uncertain variables

Variables whose values aren’t known exactly, which uncertainty quantification methods propagate to the responses. They are either aleatory or epistemic. See Uncertain Variables.

variables

The parameters Dakota varies, with their bounds or distributions. Also the input-file block that defines them; a study needs at least one. Variables are design, uncertain, or state, and continuous or discrete. See variables and Variables.