Accelerating quasiparticle physics: a review on the synergy of machine learning and density functional theory for...

Quasiparticles such as excitons, polaritons, and plasmons are central to understanding the optical and electronic properties of modern materials, with profound implications for optoelectronics, renewable energy, and quantum information science. The theoretical investigation of these many-body phenomena heavily relies on computationally intensive methods like density functional theory (DFT) and its extensions. However, the high computational cost and poor scaling of these methods present a significant bottleneck for studying large, complex systems or performing high-throughput material screening. This review discusses the emerging and transformative role of machine learning (ML) in overcoming these challenges. We begin by providing a foundational overview of the physics of excitons, polaritons, and plasmons. We then outline the standard DFT-based approaches used to model them, highlighting their capabilities and limitations. The core of this review is dedicated to exploring the multifaceted ways in which ML is being integrated with first-principles calculations. We categorize the role of ML into several key paradigms: (i) acting as surrogate models to accelerate the prediction of quasiparticle properties, (ii) building size-transferable Hamiltonians to bridge nano and meso-scales, (iii) classifying complex phases and patterns in many-body systems, and (iv) solving the inverse problem of determining system parameters from experimental observables. By surveying recent cutting-edge research, we present case studies demonstrating how this synergy is providing unprecedented insights into quasiparticle dynamics and properties. Finally, we discuss the current challenges and present a forward-looking perspective on how the continued fusion of ML and quantum mechanics is set to redefine the landscape of materials discovery and condensed matter physics.