Mimer AI Factory and ENCCS jointly organise this webinar to introduce how machine-learned interatomic potentials can support larger and more efficient simulations on modern HPC systems.
Why machine-learned interatomic potentials matter
Quantum-mechanical methods such as density functional theory (DFT) are accurate but limited to small systems and short timescales. Classical force fields are fast but often not accurate or transferable enough. Machine-learned interatomic potentials (MLIPs) could break this trade-off between accuracy and scale, and the field is now moving at remarkable speed.
So-called universal or “foundation” models (e.g. MACE-MP, UMA, MatterSim, Orb, the DPA/OpenLAM series) use tens to hundreds of millions of DFT calculations spanning the periodic table for pre-training. These models approach DFT-level accuracy at a small fraction of the cost. Researchers can apply them out-of-the-box (“zero-shot”) to almost any chemistry. Afterward, they can fine-tune them to a specific system with a small amount of targeted data. This shortens the path from question to result for both academic and industrial users.
What the webinar will cover
In this webinar, we will give a conceptual tour of this rapidly evolving landscape and what it means for everyday computational research on HPC systems. We will cover:
- what universal MLIPs are and how they differ from classical force fields and system-specific ML potentials;
- the practical pre-train → fine-tune workflow and when out-of-the-box use is (and is not) good enough;
- the new generation of batched, GPU-accelerated simulation engines (e.g. TorchSim, kUPS) built for high-throughput MLIP simulation;
- how to choose and trust a model using open benchmarks;
- and a brief outlook on where the field is heading.
This includes ML making its way into electronic-structure theory itself, with learned density functionals. We will also include a short demo on European HPC resources. Throughout, we will point to open models, datasets, benchmarks, and codes. Participants can try these on their own problems right after the webinar.
Who is the webinar for
This webinar is for:
- researchers and students in computational materials science, chemistry, and condensed-matter physics who use DFT or ab initio MD and want to reach larger systems and longer timescales without giving up accuracy;
- users of classical molecular dynamics (LAMMPS, GROMACS, ASE workflows) curious about upgrading to ML-based potentials;
- industry R&D scientists and engineers exploring AI-accelerated materials and molecular discovery;
- HPC support staff and research software engineers who want an overview of the modern MLIP software stack and what it needs from GPU systems;
- anyone curious about how the foundation-model paradigm from language and vision AI is reshaping simulation in the natural sciences.
Participants do not need prior experience with multiplet theory, density functional theory (DFT), or many-body methods. Familiarity with basic concepts from solid-state or atomic physics is sufficient. The webinar will introduce the role of data and HPC in modern electronic-structure studies at a conceptual level.
Key takeaways
By the end of this webinar, participants will:
- understand what universal machine-learned interatomic potentials are and why they deliver near-DFT accuracy at a fraction of the cost;
- know the practical pre-train → fine-tune workflow: when an off-the-shelf foundation model is sufficient, when and how to fine-tune it with a small targeted dataset;
- get an overview of the modern MLIP software stack, from ASE/LAMMPS integrations to GPU-native, batched engines such as TorchSim, and what it takes to run it efficiently on EuroHPC systems;
- critically select and validate a model using open benchmarks and understand why no single model wins on every axis;
- gain an outlook on emerging directions: long-range electrostatics, MLIPs that predict polarisation and spectra, the use of machine learning within DFT itself, and early work on AI agents that help orchestrate computational workloads.
Organisers
Mimer AI Factory & ENCCS
Speaker and moderator
- Karim Elgammal
- Yonglei Wang/Wei Li