Implementing a customized computation for preprocessing, postprocessing or outputs

This subsection explains how to implement a new customized computation. There are three types of computations we defined in the ATOmiCS toolbox:

  1. The computation that can be directly implemented using OpenMDAO

    An example can be the interpolation component for case study 3 in the paper. In order to formulate this kind of component, the user needs to refer to OpenMDAO documentation for the explicit component.

    We briefly summarize this process in this documentation. Defining a customized explicit component requires writing a python class inherited from OpenMDAO ExplicitComponent There are four main functions inside the class:

    setup() that takes the input and outputs of the component;

    setup_partials that defines the relationship of the outputs with respect to the inputs;

    compute() that contains the computation of the outputs given the inputs;

    and compute_partials() computes the partial derivative of the outputs with respect to the inputs.

  2. The computation that combines OpenMDAO and FEniCS and outputs a scalar on the entire domain

    This kind of output is wrapped as scalar_output in ATOmiCS. The user can define this type of output in the run_file.

    output_form = ...
pde_problem.add_scalar_output(<output_name>, <output_form>, <argument_names>)

    The <output_name> is the name of the customized output; <output_form> is the form expressing the output using FEniCS UFL; <argument_names> are the name list of arguments that the user wants to take derivatives with respect to.

  3. The computation that combines OpenMDAO and FEniCS and outputs a scalar on each vertex

output_form = ...
pde_problem.add_field_output(<output_name>, <output_form>, <argument_names>)

The <output_name> is the name of the customized output; <output_form> is the form expressing the output using FEniCS UFL; <argument_names> are the name list of arguments that the user wants to take derivatives with respect to.