# Restoring Chaos Using Deep Reinforcement Learning

arXiv: https://arxiv.org/abs/1912.00947 · HTML: https://mlsanity.com/paper/1912.00947/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 27 Nov 2019.

## Built on by (most cited first, 5 of 5)

- [Hybrid analysis and modeling, eclecticism, and multifidelity computing toward digital twin revolution](https://mlsanity.com/paper/2103.14629/) (2021)
- [Symmetry reduction for deep reinforcement learning active control of chaotic spatiotemporal dynamics](https://mlsanity.com/paper/2104.05437/) (2021)
- [Multi-fidelity reinforcement learning framework for shape optimization](https://mlsanity.com/paper/2202.11170/) (2022)
- [Multi-fidelity Reinforcement Learning Control for Complex Dynamical Systems](https://mlsanity.com/paper/2504.05588/) (2025)
- [Hyperfastrl: Hypernetwork-based reinforcement learning for unified control of parametric chaotic PDEs](https://mlsanity.com/paper/2604.06497/) (2026)

## Abstract (arXiv)

A catastrophic bifurcation in non-linear dynamical systems, called crisis, often leads to their convergence to an undesirable non-chaotic state after some initial chaotic transients. Preventing such behavior has proved to be quite challenging. We demonstrate that deep Reinforcement Learning (RL) is able to restore chaos in a transiently-chaotic regime of the Lorenz system of equations. Without requiring any a priori knowledge of the underlying dynamics of the governing equations, the RL agent discovers an effective perturbation strategy for sustaining the chaotic trajectory. We analyze the agent's autonomous control-decisions, and identify and implement a simple control-law that successfully restores chaos in the Lorenz system. Our results demonstrate the utility of using deep RL for controlling the occurrence of catastrophes and extreme-events in non-linear dynamical systems.