What is it about?

We use reinforcement learning as a hands-off approach to determine the pathway with the lowest energy barrier between two states. We investigate different existing reinforcement learning algorithms for this task and come up with a hybrid of the state-of-the-art algorithms which performs best at this task.

Featured Image

Why is it important?

Traditional methods to find reaction pathways usually use variations of gradient descent and might get stuck in a local minima. Using a self-learning reinforcement learning technique allows exploration of the potential energy surface and a higher chance of locating the optimal pathway with a lower energy barrier.

Perspectives

This work allowed me to apply ideas learnt in the classroom to an out-of-the-box field and learn about the challenges that come along with working on such cross-disciplinary problems.

Adittya Pal
Syddansk Universitet

Read the Original

This page is a summary of: Estimating reaction barriers with deep reinforcement learning1, Data Science, October 2024, SAGE Publications,
DOI: 10.3233/ds-240063.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page