# Using Sidre with MFEM¶

Note

The functionality described in this page is only available if Axom is configured with MFEM and if the CMake variable AXOM_ENABLE_MFEM_SIDRE_DATACOLLECTION is set to ON

The MFEMSidreDataCollection class implements MFEM’s DataCollection interface for recording simulation data. Specifically, it knows about fields (mfem::GridFunction objects) and the mesh on which these fields are defined.

The MFEMSidreDataCollection class internally organizes its data according to the Mesh Blueprint, a hierarchical schema for describing mesh data.

In this page, we first discuss Getting Started_ and how MFEM objects can be associated with the MFEMSidreDataCollection. We then explain the process and options available when Saving Data to a File. The workflow for reading saved data back in is discussed in Restarting a Simulation.

## Associating MFEM Objects with an MFEMSidreDataCollection¶

We begin to describe the data in an MFEMSidreDataCollection object by registering an MFEM mesh with an instance of the class when we create the data collection:

  // Initialize the datacollection with the mesh
// Note: all fields (added with RegisterField) must be on this mesh
axom::sidre::MFEMSidreDataCollection dc("sidre_mfem_datacoll_vis_ex", mesh);


The MFEM mesh can also be registered after construction:

dc.SetMesh(/* mfem::Mesh* */ mesh);


Note

There is a 1-1 relationship between MFEMSidreDataCollection objects and MFEM meshes. That is, multiple meshes cannot be associated with a single MFEMSidreDataCollection.

After a mesh has been registered, fields can be association with the MFEMSidreDataCollection object:

  // Note: Any number of fields can be registered
dc.RegisterField("solution", &soln);


Special kinds of fields (material-dependent fields, material set volume fraction fields, and species set values fields) require additional setup before they can be registered - see Mixed-material Fields for more informantion.

## Saving Data to a File¶

The data in an instance of the MFEMSidreDataCollection class can be saved to a file using a variety of protocols. These files can be visualized with tools like VisIt or used to restart a simulation by loading them back into an instance of the MFEMSidreDataCollection class (see Restarting a Simulation).

Current options for output protocols/formats include:

• sidre_hdf5
• sidre_conduit_json
• sidre_json
• conduit_hdf5

Note

The AXOM_USE_HDF5 CMake build option must be enabled to use HDF5-based formats.

Before saving, simulation state metadata should be updated. Currently, this metadata consists of:

• cycle, the current iteration number for the simulation
• time, the current simulation time
• time_step, the current simulation time step (sometimes called dt)

Each of these variables has a corresponding setter function:

  // Save the initial state
dc.SetCycle(0);      // Iteration counter
dc.SetTime(0.0);     // Simulation time
dc.SetTimeStep(dt);  // Time step


Note

There are also corresponding accessors for these state quantities (GetCycle, GetTime, GetTimeStep).

Once state information has been updated, the complete simulation state can be written to a file:

  // Filename and protocol, both of which are optional
dc.Save("sidre_mfem_datacoll_vis_ex", sidre_protocol);


Note

By default, saved files will be written to the current working directory. To write to/read from a different directory, use SetPrefixPath to change the directory location of where the MFEMSidreDataCollection object’s data will be written.

See the sidre_mfem_datacollection_vis example for a more thorough example of the above functionality.

## Restarting a Simulation¶

Experimental support for complete reconstruction of a simulation mesh, fields, and qfields is also provided by the MFEMSidreDataCollection class. That is, when an output file is read in using the MFEMSidreDataCollection::Load() method, the data read in will be used to reconstruct MFEM objects than can be accessed with the GetField, GetQField, and GetMesh methods.

Warning

Currently, an MFEMSidreDataCollection object must own the mesh and field data in order to completely reconstruct simulation state. Mesh data ownership can be configured with the owns_mesh_data constructor option (should be set to true), and field data ownership requires that each GridFunction object be unallocated when passed to the RegisterField() method, which performs the allocation within Sidre-owned memory. After registeration, a GridFunction object can be used normally. The same conditions apply for QuadratureFunction objects.

A complete demonstration of functionality is provided in the sidre_mfem_datacollection_restart example, which is a stripped-down example of how a simulation code might utilize the automatic reconstruction logic when loading in a datastore.

Note

The mesh/field reconstruction logic requires that the saved file was created with an MFEMSidreDataCollection object. In Mesh Blueprint terms, the following constraints are imposed on the structure of the data:

• There must be a coordinate set named coords
• There must be a topology named mesh with corresponding attributes stored in a field named mesh_material_attribute
• There must be a topology named boundary with corresponding attributes stored in a field named boundary_material_attribute

## Mixed-material Fields¶

The Mesh Blueprint provides support for mixed-material simulations through its material set construct. Material metadata (stored in mfem::GridFunction objects) can be registered with the MFEMSidreDataCollection like any other field (with the RegisterField() method), given that some additional setup is performed and the field names match a specific naming convention.

Currently, there are three kinds of special fields that can be associated with mixed-material metadata:

• Volume fraction fields in a material set (the per-element volume fraction of a given material)
• Material-dependent fields (a different set of values for each material)
• Species set fields (a different set of values for each material and for each dimension)

Material sets are defined by associating a volume fraction field name with a material set name:

  // Inform the DataCollection that volume fraction information for a material set
// called "matset" will be in fields called "volume_fraction"
dc.AssociateMaterialSet("volume_fraction", "matset");


Once this association has been made, corresponding fields can be registered. Their names must follow the format <associated volume fraction field name>_<material id>, for example:

  dc.RegisterField("volume_fraction_001", vol_frac_1);
dc.RegisterField("volume_fraction_002", vol_frac_2);


Material-dependent fields are defined by associating a field name with a material set name:

  // Inform the DataCollection of a material-dependent field "density" associated
// with a material set called "matset"
dc.AssociateMaterialDependentField("density", "matset");


Material-specific values can be registered with a name in the format <associated dependent field name>_<material id>:

  dc.RegisterField("density", density_independent);
dc.RegisterField("density_001", density_dependent_1);
dc.RegisterField("density_002", density_dependent_2);


Species sets are defined by associating a field name with a species set name and corresponding material set name:

  // Inform the DataCollection that species set data for the species set "specset"
// associated with material set "matset" will be in fields called "partial_density"
constexpr bool volume_dependent = false;
dc.AssociateSpeciesSet("partial_density", "specset", "matset", volume_dependent);


Species set values can be registered with the name in the format <associated field name>_<material id>_<component>:

  dc.RegisterField("partial_density_001_001", partial_density_1_1);
dc.RegisterField("partial_density_001_002", partial_density_1_2);
dc.RegisterField("partial_density_002_001", partial_density_2_1);
dc.RegisterField("partial_density_002_002", partial_density_2_2);