The Relationship Between Study Steps and Solver Configurations

Most studies and study steps correspond to part of a solver configuration that includes a solver for the specific problem, as listed in Table 20-1.

•

Study

•

Solver Configurations

Table 20-1: The relationship between Common Study Steps and Solvers
Study Step	Description	Corresponds to Solver
Stationary	Generates equations without time derivatives.	Stationary Solver. A parametric continuation solver can also be created by selecting an option on the study Settings window. Also see About the Stationary Solver and About the Parametric Solver.
Time Dependent	Generates equations for transient (time-dependent) simulations.	Time-Dependent Solver. Also see About the Time-Dependent Solver.
Time Discrete	Use it for performing time-dependent analysis using the projection method.	Time Discrete Solver. Also see About the Time Discrete Solver.
Time Dependent	Use the Time Explicit Solver to solve a time-dependent problem using an explicit time-stepping scheme.	Time Explicit Solver. Also see The Time Explicit Solver Algorithms.
Eigenvalue	Generates equations formulated for computing eigenvalues and eigenfunctions.	Eigenvalue Solver. Also see The Eigenvalue Solver Algorithm.
Eigenfrequency	Similar to an Eigenvalue study step but computes eigenfrequencies instead of eigenvalues.	Eigenvalue Solver (set to transform eigenvalues to eigenfrequencies). Also see The Eigenvalue Solver Algorithm.
Frequency Domain	Generates stationary equations that are used for frequency sweeps.	It corresponds to a stationary parametric solver that is preset to linearize the equations (Stationary Solver with a Parametric attribute). By selecting the Use asymptotic waveform evaluation check box, this study step corresponds to an AWE Solver.
Time Dependent, Modal	Generates equations for time-dependent modal analysis.	Modal Solver (with Study type set to Time dependent). Also see The Modal Solver Algorithm.
Frequency Domain, Modal	Generates equations for modal analysis in the frequency domain.	Modal Solver (with Study type set to Frequency domain). Also see The Modal Solver Algorithm.

There are some study steps that do not generate equations and can only be used in combination with other study steps. These study extension steps do not correspond directly to any part of a solver configuration. Instead, they correspond to a part of the job configuration or modify the behavior of another study step.

Study Extension Steps

A Parametric Sweep is used to formulate a sequence of problems that arise when you vary some parameters in the model. The problem at a fixed parameter value is defined by the rest of the study steps in the study. It generates a Parametric Sweep (Job Configurations) node, unless the problem and parameters are such that the parametric sweep can be realized through a Stationary Solver with a Parametric node, in which case such a solver is generated in the solver configuration.


	The parametric sweep can include multiple independent parameters directly, but you can also add more than one Parametric Sweep node to create nested parametric sweeps. In the Study branch, indentations of the node names indicate that the parametric sweeps are nested.

The Optimization study step is used to solve PDE-constrained optimization problems. This study step allows direct definition of objective functions and selection of model parameters, including parameters that control the geometry, for optimization. It also provides detailed control over solvers and contributions to an optimization problem defined by an Optimization interface. This study type requires an Optimization Module license.

Advanced Study Extension Steps

Batch and Batch Sweep

A Batch study creates a job that can be run without the graphical user interface and which stores the solution on disk. It generates a Batch (Job Configurations).

A Batch Sweep is used to formulate a sequence of problems that arise when you vary some parameter in the model. Each parameter tuple generates a batch job that runs the model with the given tuple. The results are stored on file and updated into the model. It generates a Batch (Job Configurations) and a Parametric Sweep (Job Configurations). A Batch Sweep is similar to a Parametric Sweep and is useful when you want to retrieve solutions for a parametric sweep during the solution process and when the problem formulation is such that the solution for each parameter is independent of the solution of all other parameters For example, it can be useful in the following situations where you may want to inspect the partial results during a solver sweep:

•

You are basing your sweep on a table of input data and it turns out that obtaining a solution for some of those tabulated values takes an unexpectedly long time, but you do not know which values beforehand. You may still want to inspect the solution for as many parameters as possible to determine if you should terminate the solution process or start analyzing the results before the entire sweep is complete.

•

You are using a mathematical expression for a certain material property or boundary condition that turns out to give nonphysical results for some parameters.

•

You wrote an external function (in C, for example) to define a complicated material, but you did not make it foolproof for all input data and it returns bad output data for certain parameter values.

•

You are running a parametric sweep where one of the parameters is a geometric dimension, but you accidentally defined too wide of a range of dimensions. By the time you realize this, the solution has already been running for a long time and you don’t want to stop it.

If you use a batch sweep in any of these cases, each parameter can be solved for in a separate process that can be started and stopped independently. The results for the parameters that have already been solved for can be stored as an MPH-file for each parameter value, and you can open and review any number of them during the solution process.

Cluster Computing and Cluster Sweep

A Cluster Computing study is used to solve the problem on a distributed-memory computer architecture. It generates a Cluster Computing (Job Configurations) and a Batch (Job Configurations).

A Cluster Sweep is used to formulate a sequence of problems that arise when you vary some parameter in the model. The program computes the solution for each parameter on a distributed-memory computer architecture. The results are stored on file and updated into the model. It generates a Cluster Computing (Job Configurations), Batch (Job Configurations), and (if applicable) Parametric Sweep (Job Configurations).

Multigrid Level

A Multigrid Level node can be added as a subnode to other study step nodes to describe a geometric multigrid level used by the study.

Sensitivity

The Sensitivity study step specifies objective functions and controls variables with respect to which sensitivity is computed. Global scalar objective functions can be specified directly in the study step, and model parameters can be selected as control variables. In addition, the study step provides control over the sensitivity solver method and contributions to the sensitivity problem defined with a Sensitivity or Optimization interface.

Batch Sweeps and Cluster Sweeps vs. Distributed Sweeps

Batch sweeps and cluster sweeps work in a different way than distributed sweeps. A distributed sweep runs different parameters in parallel within an MPI job (different processes compute different parameters). A Batch Sweep or Cluster Sweep starts several processes in parallel and runs them, and it then afterward collects the result into the main process. In most cases, a distributed sweep is easier to work with (select the check box and start in distributed mode). A Batch Sweep or Cluster Sweep requires that you set up a number of paths, but in cases where you want robustness and possibility for individual parameter restarts, a cluster sweep is preferred over a distributed parametric sweep. See also the following section.

Batch Sweeps vs. Cluster Sweeps

In addition to a Parametric Sweep, you can also perform a Batch Sweep or a Cluster Sweep (see also the section above). The Batch Sweep is available for all COMSOL Multiphysics license types. If you have a floating network license, then you have access to an additional feature called Cluster Sweep. These two sweep types are similar, but the Cluster Sweep has additional settings for remote computations and cluster configurations. With a Cluster Sweep, you can distribute a large sweep on a (potentially large) cluster. The performance benefit of doing so can be very high because independent sweeps (sometimes called embarrassingly parallel computations) typically scale very well. If you master the batch sweep, then the step toward running a cluster sweep is not that big.