Temperature‐Adaptive Thermal‐Photonic Metamaterials: Generalized Infrared Engineering via Anharmonic Phonon Self‐Energy Datasets and Dimensionality‐Augmented Neural Networks

Abstract All thermal‐photonic metamaterials (TPMs) inherently exhibit the fundamental phenomenon of temperature‐dependent spectral redshift in practical applications, yet designs rarely address temperature adaptability. It is established a universal framework integrating ab initio anharmonic phonon self‐energy datasets with dimensionality‐augmented neural networks, introducing temperature as a novel design dimension. This framework decodes high‐dimensional temperature‐structure‐material‐spectrum mappings by leveraging spectral shifts as “physical perturbations”. A physics‐constrained dataset enables a simple fully connected neural network (FCNN) to achieve unprecedented generalization, facilitating rapid inverse design for narrowband thermophotovoltaics (1‐5 µm, 1500 K), broadband & laser camouflage (3–14 µm, 1.06/1.55/10.6 µm, 300–1500 K) and full‐spectrum radiative coolers (1–25 µm, 300 K), with 120 times acceleration over conventional algorithms. Dual‐band TPMs demonstrate 55% emissivity suppression at 1100 K with 65 K cooling reduction, and 2% optical drift over 12 h at 1500 K, with 84% visible and 98% microwave (X‐band) transparency enabling multispectral integration. This work enables temperature‐adaptive TPMs for stealth systems, energy harvesting, and thermal management.