This version is meant to be used with TFEL Version 5.1. This version is the first that can optionnaly use directly the TFEL libraries.

This version was released on December 15, 2025. This version was released on December, 15 2025 along with:

• Version 1.1.2
• Version 1.2.3
• Version 2.0.1
• Version 2.1.1
• Version 2.2.2
• Version 3.0.2

and inherits from all fixes from those releases.

Highlights

The MGIS/Function library

The main new feature of MGIS is the MGIS/Function[^functions] library, which has been introduced to provide standard mechanical post-processings based on the TFEL/Math and TFEL/Material libraries. MGIS/Function allows reusing the large amount of documented and verified functionalities provided by those libraries at the structural scale, including: rotation from/to material frame, conversion between stress measures, Hencky and Green-Lagrange strains, Tresca, von Mises, Hosford, Hill, Barlat equivalent stresses, principal values, Lode’s angle, etc.

[^functions] A function denote a set of values associated with the elements (nodes, quadrature points, cells) of a discretized space. Functions are often called fields in many solvers.

The following snippet computes the Cauchy stress in the global frame from the first Piola-Kirchhoff stress known in the material frame.

const auto evaluator = pk1_function | as_tensor<3>                    |
                       from_pk1_to_cauchy (F_function | as_tensor<3>) |
                       rotate_backwards (R_function | as_matrix<3,3>);
const auto ok = assign(ctx, sig | as_stensor<3>, evaluator);

The library is developed in C++-20 with high-quality requirements and strives to be constexpr friendly, memory-safe, thread-safe, exception-safe, and GPU-friendly.

The data structures of the functions depend on the targeted solver, as well as the implementation of the assign, which depends on the programming model chosen by the targeted solver (OpenMP, CUDA, SYCL, Kokkos, etc.).

For standard data structures, such as the one used by Numpy’s array, so-called views are provided allowing direct usage of the library. In this case, an implementation of the assign algorithm is provided by the parallel version of the Standard library. Depending on the compiler and implementation of the standard library, various parallel programming models are used (multithreading and GPU-offloading).

The MGIS/Function library is described on this page.

Improvements

The integrate_debug functions

The integrate_debug function can be used to generate debug files to analyse integration failures. This function is very customizable, see this page for details.

Example of usage

This snippet shows that the integrate_debug functions can be used as drop-in replacement for the integrate function.

const auto r = integrate_debug(v, b);

getDatabase and loadFromDatabase

Note

This feature requires MGIS to be compiled againt the TFEL libraries.

The mgis::getDatabase function returns a global instance of an MFrontDatabase object. See this page for more details. This instance can be populated using a library, a directory or a list of directories specified in an environment variable.

The function mgis::behaviour::loadFromDatabase relies on the previous database to load a behaviour matching the given criteria:

• the name (required) of the behaviour of the loaded,
• the material (optional)

Example of usage

The following example registers all the material knownledge available in the shared libraries found in directories listed in an environment variable and lists all the behaviours compiled with the generic interface:

import mgis
db = mgis.getDatabase()
db.analyseDirectoriesListedInEnvironmentVariable('MFRONT_GALLERY_LIBRARIES',
                                                 ignore_errors = True)
for e in db.getEntryPoints(type = 'behaviour',
                           interface = 'generic',
                           material = 'concrete'):
    print(f"- {e.library}: {e.name}")

The following snippet populates the database using all the libraries of a given directory and loads a behaviour whose name of which contains Elasticity and which is associated with the material wood.

import mgis
import mgis.behaviour as mgis_bv
db = mgis.getDatabase()
db.analyseDirectory('/home/th202608/codes/MFrontGallery/master/install/lib/',
                    ignore_errors = True)
b = mgis_bv.loadFromDatabase(name='\\w+Elasticity\\w+', material='Wood',
                             hypothesis = mgis.behaviour.Hypothesis.Tridimensional)

Python bindings

Python bindings are now generated using the pybind11 library.

New tutorials

Two tutorials show how to use the python bindings to interact with a finite-strain mechanical behaviour compiled with MFront and develop an alternative to MTest:

• case of a finite strain behaviour
• case of a small strain behaviour

Tested configurations

System

Github actions

Github actions allows to check compilation on Ubuntu (latest) gcc 13.3.0.

Continuous integration provide by the PLEIADES project

System

Acknowledgements

The authors are grateful to the many contributors to the TFEL/MFront project. This research was conducted in the framework of the PLEIADES project, which was supported financially by the CEA (Commissariat à l’Énergie Atomique et aux Énergies Alternatives), EDF (Électricité de France) and Framatome. Work on MGIS/Function was performed as part of the EURATOM OperaHPC Project co-funded by the European Union.

Issues fixed

Issue 148: [doc] better document rdt usage

For more details, see https://github.com/thelfer/MFrontGenericInterfaceSupport/issues/148.