iiinova — where your digital self keeps evolving

Winfried Ripken @winfried-ripken.bsky.social · 07/07/2026

🌐 Website & Colab: ml4molsim.github.io/hamiltonian-... 📄 Paper: arxiv.org/abs/2601.22123 💻 Code: github.com/ML4MolSim/ha...

ml4molsim.github.io

Learning Hamiltonian Flow Maps

000

Winfried Ripken @winfried-ripken.bsky.social · 07/07/2026

Excited to be at ICML with @plainer.bsky.social & @gregorlied.bsky.social to present our 🌟 spotlight 🌟 paper on Hamiltonian Flow Maps for large-timestep molecular dynamics! Come say hi at our poster tomorrow or send us a message if you want to chat! 🤗 🗓️ July 8, 10:30 AM – 12:15 PM 📍 HALL A #1306

120

Winfried Ripken @winfried-ripken.bsky.social · 26/04/2026

Shout-out to my amazing co-authors @plainer.bsky.social, @gregorlied.bsky.social, @thorbenfrank.bsky.social, Oliver Unke, Stefan Chmiela, @franknoe.bsky.social, and Klaus-Robert Müller!

000

Winfried Ripken @winfried-ripken.bsky.social · 26/04/2026

Happy to present our work on Hamiltonian Flow Maps today at the AI4MAT Workshop at ICLR! Feel free to come by at 2.20pm in room 203C.

130

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

(9/n) Check out our paper and code: 🌐 Website & Colab: ml4molsim.github.io/hamiltonian-... 📄 Paper: arxiv.org/abs/2601.22123 💻 Code: github.com/ML4MolSim/ha...

ml4molsim.github.io

Learning Hamiltonian Flow Maps

110

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

(8/n) Done with a brilliant team: @plainer.bsky.social, @gregorlied.bsky.social, @thorbenfrank.bsky.social, Oliver Unke, Stefan Chmiela, @franknoe.bsky.social, and Klaus-Robert Müller.

110

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

(7/n) With this, you get a single model that provides: - Instantaneous forces (like a standard MLFF) - Stable large-timestep updates far beyond classical integrators - Training and inference cost comparable to MLFFs

100

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

(6/n) How? Inspired by recent advances in few-step generative modeling, our tailored loss function combines force matching with a consistency constraint that enforces agreement of the predicted flow across different time horizons.

100

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

(5/n) Our approach: In contrast, our method learns continuous-time, large-timestep dynamics directly from decorrelated phase-space samples without requiring expensive reference trajectories, and supports arbitrary timesteps during inference.

100

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

(4/n) Existing approaches rely on trajectory data, typically generated using another model simulated with small timesteps. This potentially introduces artifacts from the teacher and is computationally expensive. Can we learn large timesteps, without ever seeing trajectories?

100

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

(3/n) The solution: Hamiltonian Flow Maps The core idea is to model the Hamiltonian evolution directly in phase space over a finite time interval. Concretely, we aim to learn a 𝗛𝗮𝗺𝗶𝗹𝘁𝗼𝗻𝗶𝗮𝗻 𝗙𝗹𝗼𝘄 𝗠𝗮𝗽 that advances positions and momenta over an interval Δt.

100

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

(2/n) The problem: Simulations are fundamentally limited by the small timesteps required for stable numerical integration. Even with ML, most of the cost still comes from taking millions of tiny integration steps.

100

Winfried Ripken @winfried-ripken.bsky.social · 19/02/2026

Ever get tired of tiny timesteps bottlenecking your MD simulations? We show how to train a model for large-timestep Hamiltonian dynamics directly on standard MLFF datasets. 𝗡𝗼 𝗿𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝘁𝗿𝗮𝗷𝗲𝗰𝘁𝗼𝗿𝗶𝗲𝘀, 𝗻𝗼 𝘂𝗻𝗿𝗼𝗹𝗹𝗶𝗻𝗴, 𝗻𝗼 𝘁𝗲𝗮𝗰𝗵𝗲𝗿 needed! 🧵👇

1177

Winfried Ripken @winfried-ripken.bsky.social · 28/11/2025

Link to github: github.com/ML4MolSim/di... Paper on arxiv: arxiv.org/abs/2506.15378

000

Winfried Ripken @winfried-ripken.bsky.social · 28/11/2025

✨ Via a modular architecture, we enable a fair comparison of symmetries changing both attention mechanism and embedding strategies of our model ✨ We transfer the powerful DiT architecture from Computer Vision to the molecular domain, proposing two complementary graph-based conditioning strategies

100

Winfried Ripken @winfried-ripken.bsky.social · 28/11/2025

We introduce DiTMC, a new way to predict molecular conformers - the different 3D shapes molecules can flex into. ✨ We learn to predict 3D geometry from molecular structure ✨ We achieve state-of-the-art results on the GEOM benchmarks

100

Winfried Ripken @winfried-ripken.bsky.social · 28/11/2025

I’m excited to be at NeurIPS 2025 next week and present our latest paper on molecular conformer generation! Huge thanks to my co-authors @thorbenfrank.bsky.social , Gregor Lied, Klaus-Robert Müller, Oliver Unke and Stefan Chmiela for an incredible collaboration. Supported by: @bifold.berlin

171