Crystal Fractional Graph Neural Network for Energy Prediction of High

Physics > Computational Physics
[Submitted on 26 Apr 2026]
Title:Crystal Fractional Graph Neural Network for Energy Prediction of High-Entropy Alloys
View PDF HTML (experimental)Abstract:High-entropy alloys (HEAs) have attracted growing attention for their exceptional mechanical and thermal properties arising from complex atomic configurations. In this paper, we propose crystal fractional graph neural network for predicting the energy of high-entropy alloys by explicitly integrating both local atomic environments and global compositional information. The model consists of three components: a crystal graph neural network, which employs graph attention network layers to learn local interactions among 16 on-site atoms within the crystal lattice; fractional neural network, a fully connected network that embeds the global fraction of constituent elements; and feature fusion neural network, which fuses the outputs of the two submodels to predict the total crystal energy. We train the model on a dataset of 1,049 crystal structures and validate it on 198 quaternary structures, optimizing all hyperparameters via Optuna. Our results show that our model achieves an RMSE comparable to first-principles calculations and maintains high accuracy even for low-energy configurations. However, the model exhibits limitations in handling large crystal cells, which we aim to address in future work to extend its applicability to more complex systems.
Current browse context:
physics.comp-ph
Change to browse by:
References & Citations
Loading...
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Connected Papers (What is Connected Papers?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
alphaXiv (What is alphaXiv?)
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Hugging Face (What is Huggingface?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Facts Only

* A crystal fractional graph neural network was proposed for predicting the energy of high-entropy alloys.
* The model integrates local atomic environments and global compositional information.
* The model comprises three components: a crystal graph neural network, a fractional neural network, and a feature fusion neural network.
* The crystal graph neural network uses graph attention network layers to learn local interactions among 16 on-site atoms.
* The fractional neural network embeds the global fraction of constituent elements.
* The model was trained on 1,049 crystal structures.
* The model was validated on 198 quaternary structures.
* The results show an RMSE comparable to first-principles calculations.
* The model maintains high accuracy for low-energy configurations.
* The model exhibits limitations in handling large crystal cells.

Executive Summary

A crystal fractional graph neural network is proposed to predict the energy of high-entropy alloys by integrating local atomic environments and global compositional information. The model is structured around three interconnected components: a crystal graph neural network using graph attention layers to model local interactions among 16 on-site atoms; a fractional neural network that embeds the global fraction of constituent elements; and a feature fusion neural network that combines the outputs of the submodels to predict the total crystal energy. The model was trained on a dataset of 1,049 crystal structures and validated against 198 quaternary structures. The results indicate that the model achieves a Root Mean Square Error (RMSE) comparable to first-principles calculations and maintains high accuracy, particularly for low-energy configurations. A stated limitation of the current model is its difficulty in handling large crystal cells, which the authors plan to address in future work.

Full Take

The proposed methodology addresses a critical challenge in materials science by attempting to bridge the gap between complex atomic structure and thermodynamic properties using machine learning. The novelty lies in the explicit, multi-scale integration of crystal structure data—combining fine-grained local geometry (GNN) with coarse-grained compositional data (fractional NN). This approach attempts to capture the synergistic effects of local bonding environments and overall composition that govern the energy of high-entropy alloys. The claim of achieving RMSE comparable to first-principles calculations is a strong benchmark, suggesting the learned representation is physically meaningful, but this must be scrutinized against the complexity of the input space and the limitations imposed by the crystal cell size constraint.
A key area for scrutiny is the handling of scale. The acknowledged limitation regarding large crystal cells implies that the model's predictive power might be sensitive to the size of the input system, potentially limiting its applicability in scenarios involving extended defects or large-scale simulations. Future work must not only resolve this limitation but also establish whether the achieved accuracy is robust across varied crystal cell sizes and different material classes. The true value of this work rests on whether this hybrid framework offers a generalizable, computationally efficient alternative to expensive first-principles methods, and whether the discovered pattern in the feature fusion is physically intuitive.

Sentinel — Human

Confidence

This abstract presents a methodologically sound and technically specific proposal for using deep learning to predict material properties, exhibiting the structure and domain knowledge consistent with human scientific authorship.

Signals Detected

Moderate sentence length variance; formal, dense structure typical of academic writing.

High coherence; focus is narrow and objective, lacking superfluous, non-academic flow.

Matches a standard, well-structured scientific abstract template; methodology and results are logically linked.

The claims and terminology are highly consistent with cutting-edge computational physics literature; no immediate sign of LLM confabulation or unnatural phrasing.

Human Indicators

The specific references to methodologies (GNN layers, Optuna) and the precise quantification of the training/validation datasets suggest authentic engagement with the research process.

The inclusion of a specific, articulated limitation (handling large crystal cells) points to genuine, self-aware scientific exploration rather than generic output.