Reinforcement Learning (RL) has emerged as a transformative paradigm for the optimal design and optimization of physical systems, providing new pathways beyond conventional numerical optimization and inverse-design methods. By framing design as a sequential decision-making process, RL allows algorithms to autonomously explore high-dimensional design spaces, adapt to constraints, and uncover non-intuitive solutions. This makes RL particularly powerful in contexts where multiple objectives and nonlinear physical laws define the design challenge.

RL has demonstrated promise across a wide spectrum of domains. In materials science, it can accelerate the discovery of alloys, composites, and nanostructures with tailored electronic, thermal, or mechanical properties. In optical and photonic systems, RL enables the design of metamaterials, resonators, and lenses with precise control over light–matter interactions. In mechanical and structural systems, RL advances topology optimization, lightweight designs, and vibration-resistant mechanisms. In fluid and thermal systems, RL contributes to aerodynamic optimization, turbulence management, and efficient heat transfer solutions.

A rapidly growing frontier lies in computational simulations of complex systems, where RL functions not only as an optimizer but also as a generator of new insights. In climate and environmental models, RL can support adaptive strategies for energy and resource management. In biological and epidemiological simulations, RL can optimize intervention policies or drug delivery protocols. In social systems, RL opens exciting opportunities to model and guide collective human behavior. For instance, RL can inform urban planning, transportation networks, economic policy design, disaster response strategies, and resource allocation in smart cities. By interacting with agent-based or system-dynamics simulations, RL can test “what-if” scenarios, discover adaptive policies, and reveal emergent phenomena that are difficult to predict through conventional modeling alone.

Despite its promise, significant challenges remain. RL for design and simulation must address the computational cost of high-fidelity simulations, ensure interpretability and physical plausibility of the solutions, and balance competing objectives across scales. Integrating domain knowledge from physics, engineering, economics, and social sciences with RL is crucial to creating solutions that are both efficient and trustworthy. Hybrid methods that combine RL with physics-based modeling, reduced-order simulation, or generative design are particularly promising for this endeavor.

This Research Topic will focus on, but is not limited to, the following areas:
- General Frameworks and Methodologies: Physics-informed RL, multi-objective optimization, and hybrid simulation–learning approaches.
- Materials and Nanostructure Design: RL for optimal composition and architecture discovery.
- Social Systems and Policy Design: RL for adaptive urban planning, economic modeling, collective behavior, smart cities, disaster preparedness, and societal resilience.
- Optical, Photonic, and Acoustical Systems: RL-driven design of metamaterials, photonic devices, and waveguides.

We welcome original research articles, reviews, methodological papers, and application-driven studies. By integrating perspectives from engineering, natural sciences, and social sciences, this Research Topic aims to establish Reinforcement Learning as both a design engine for physical systems and a strategic tool for complex simulations, enabling breakthroughs that span from the atomic scale of materials to the societal scale of human systems.

Are you interested in contributing a manuscript to this Research Topic?

Reinforcement Learning for the Optimal Design of Physical Systems and Complex Simulations