Announcement Detail
Instructor: Gary Hu, Argonne National Laboratory
Dates: June 29 – July 1, 2026
Time: 10:00 AM – 2:00 PM CDT each day
Location: Online (meeting link provided after registration)
Registration Fee:
-
$300 – Non-students
-
$200 – Students
Software Requirements:
-
Linux or macOS preferred
-
Google Colab supported
Overview
NEML2 is a next-generation material modeling framework built around a simple but powerful principle: author a constitutive model once in Python and deploy it everywhere—on CPUs, GPUs, and as a compiled, Python-free C++ kernel suitable for production PDE solvers.
The same model definition drives every backend, serving analysts seeking turnkey constitutive updates, researchers requiring full access to model internals and sensitivities, and machine learning practitioners integrating neural networks, Gaussian processes, or other PyTorch modules directly into constitutive models.
NEML2 provides a complete workflow for model development, calibration, and deployment. Models can be calibrated using deterministic gradient-based optimization or Bayesian inference methods such as stochastic variational inference (SVI) and Hamiltonian Monte Carlo (HMC). For path-dependent constitutive models, NEML2 integrates with pyzag to automate adjoint construction and enable parallel-in-time acceleration for significantly faster calibration workflows.
When models are ready for deployment, they can be compiled into standalone device kernels using PyTorch's export and ahead-of-time compilation infrastructure, enabling efficient integration into large-scale PDE simulations without Python dependencies.
While NEML2's built-in library is strongest in solid mechanics—including elasticity, viscoplasticity, crystal plasticity, damage, and phase-field fracture—the framework's modular architecture naturally extends to chemical reactions, porous flow, precipitation modeling, and multiphysics applications.
Learning Outcomes
By the end of this workshop, participants will be able to:
-
Run models from the NEML2 catalog using Python or the command-line interface
-
Compose constitutive models from existing building blocks
-
Develop custom Model subclasses from scratch
-
Implement implicit constitutive updates
-
Drive models through realistic loading histories
-
Calibrate model parameters against experimental data
-
Compile and deploy calibrated models for CPU and GPU execution
-
Couple constitutive models with PDE solvers
-
Apply NEML2's modular architecture to new multiphysics applications
Participants will leave with both practical experience and a conceptual understanding of how NEML2 is structured, enabling them to confidently extend the framework to their own research and engineering problems.
Workshop Schedule
Day 1 — Foundations: Loading, Evaluating, and Driving Models
June 29, 2026 | 10:00 AM – 2:00 PM CDT
Session 1.1 (60 min) — Welcome and Orientation
Break (10 min)
Session 1.2 (60 min) — Running Your First Model from Python
Lunch Break (30 min)
Session 1.3 (75 min) — Parameters, Devices, and Vectorization
Break (10 min)
Session 1.4 (45 min) — CLI Tools and Cross-Referencing
Wrap-Up (5 min) — Preview of model composition, implicit models, and the TransientDriver
Day 2 — Composition, Implicit Models, and Transient Integration
June 30, 2026 | 10:00 AM – 2:00 PM CDT
Session 2.1 (60 min) — Model Composition
Break (10 min)
Session 2.2 (60 min) — Parameters Revisited
Lunch Break (30 min)
Session 2.3 (75 min) — Implicit Models and Newton Solves
Break (10 min)
Session 2.4 (45 min) — TransientDriver and the Physics Catalog Tour
Wrap-Up (5 min) — Preview of custom models, calibration, and deployment
Day 3 — Extension, Calibration, and Deployment
July 1, 2026 | 10:00 AM – 2:00 PM CDT
Session 3.1 (75 min) — Writing a Custom Model
Break (10 min)
Session 3.2 (45 min) — Composing Custom and Built-In Components
Lunch Break (30 min)
Session 3.3 (75 min) — Parameter Calibration with Autograd and pyzag
Break (10 min)
Session 3.4 (45 min) — Compilation and Deployment
Workshop Wrap-Up (10 min)
Additional Resources
NEML2 Documentation and Project Website: https://applied-material-modeling.github.io/neml2/