High-entropy materials (HEMs) offer unprecedented opportunities for superior mechanical, thermal, and catalytic properties, but their vast chemical space makes experimental discovery resource-intensive. State-of-the-art commercial large language models (LLMs) notably fail at HEM synthesizability prediction, a critical bottleneck in materials development. We demonstrate that domain-specific fine-tuning transforms open-weight local LLMs into accurate predictors. Using a dataset of 321,083 inorganic compositions with 2,560 HEM examples, we fine-tuned three 4-bit-quantized models (gpt-oss-20b, Qwen3-14b, and DeepSeek-R1-Distill-Qwen-14b), achieving remarkable balanced accuracy of 0.957, 0.961, and 0.956, respectively. Critically, these models operate efficiently on accessible hardware (< 15GB VRAM), eliminating costly API dependencies while ensuring data privacy and consistent reproducibility. This work could open new pathways toward autonomous closed-loop discovery, where distributed local models enable rapid screening and iterative improvement through experimental feedback. Future collaborative efforts in open data sharing, particularly including negative results, would address current fragmentation in synthesis reporting and accelerate community-wide HEM discovery.

An automated rapid microkinetic simulation is invaluable in predicting the products of chemical reactions ahead of experiments. Although existing models perform well in their focused domains, a dedicated comprehensive framework for the bond exchange reaction kinetics is absent. To address this, we present an integrated algorithm for the rapid kinetic simulation of bond exchange reaction networks. We introduce a scalable matrix-based enumeration method that allows economic exploration of all plausible reaction products without resorting to reaction templates. Our model primarily uses machine learning, which achieves a mean absolute error of 4.55 kcal/mol for activation energies, to more efficiently predict reaction properties, without relying on stored chemical reaction databases or expensive electronic structure calculations. The framework was validated by successfully reproducing two reactions.

Visible light-driven photocatalysis is a sustainable approach for air purification, but full mineralization of aromatic volatile organic compounds (VOCs) such as toluene to carbon dioxide (CO2) remains elusive due to their chemical stability and weak surface affinity. Here, we report a platinum-loaded, surface-fluorinated tungsten trioxide (F-WO3/Pt) that achieves mineralization of toluene under visible-light irradiation, markedly outperforming conventional Pt/WO3, which cannot degrade toluene at all under the same condition. Mechanistic studies reveal that the surface fluorination of Pt/WO3 promotes the formation of mobile hydroxyl radicals (·OH). These radicals diffuse into the gas phase, extending the active reaction zone beyond the catalyst surface. This fluorination-driven ·OH mobility compensates for WO3’s inherently low photocatalytic oxidation (PCO) activity, low surface area, and weak toluene adsorption. Remote PCO experiments, scavenger tests, and ESR analyses collectively demonstrate the generation and active role of mobile radicals, whereas DFT calculations reveal an additional effect arising from the enhanced affinity of toluene with the fluorinated surface. The synergy between radical mobility and surface affinity establishes a new conceptual framework for the design of highly efficient photocatalysts for air pollutant degradation. These results present a scalable approach to modulating radical dynamics for enhanced visible-light-driven VOC removal and underscore the broad potential of this strategy for developing advanced materials for air purification.