Page 122 - Read Online

Page 14 of 31 Li et al. J. Mater. Inf. 2026, 6, 10

Figure 5. Agents use material computational tools and ML models to calculate the properties of materials. (A) Computational methods
that Aitomia can use and the computing tasks it can perform [72] ; (B) GNN model for predicting Peierls barrier and potential energy
change [83] . Figure 5A is reproduced from “Aitomia: Your Intelligent Assistant for AI-Driven Atomistic and Quantum Chemical Simulations”,
arXiv:2505.08195, under CC BY 4.0 license [72] . Figure 5B is reproduced from “Rapid and Automated Alloy Design with Graph Neural
Network-Powered LLM-Driven Multi-Agent Systems”, by Alireza Ghafarollahi and Markus J. Buehler, arXiv:2410.13768, under CC
BY-NC-ND 4.0 license and the image has not been modified [83] . ML: Machine learning; GNN: graph neural networks; AI: artificial
intelligence; LLM: large language model; QM: quantum mechanics; DFT: density functional theory; ANI-1ccx-gelu: a universal interatomic
potential for calculating IR anharmonic frequencies; OMNI-P2x: a universal neural network potential for excited-state simulations;
MACE-OFF: a series of short-range transferable force fields for organic molecules; AIMent-2: the 2nd generation of atoms-in-molecules
neural network potential; AIQM: AI-enhanced quantum mechanics methods; DENS24: density functional ensembles.

efficient screening for carbon capture and battery materials using ML . The Digital Catalysis Platform
[89]
(DigCat) establishes a cloud-based global closed-loop feedback system, automating the entire workflow from
catalyst design to micro-dynamics simulation and achieving highly efficient catalyst screening . This
[90]
“prediction-evaluation-feedback” closed-loop paradigm surpasses the traditional slow screening method
based on theoretical computation, making AI agents the core driving force for high-throughput material
screening and design.

With the in-depth application of AI agents in material property calculation, computational simulations have
been used to address the hallucination and credibility issues of LLMs in the materials domain. Here,