Capturing Short-Range Order in High-Entropy Alloys with Machine Learning Potentials

High-entropy alloys (HEAs) are revolutionizing the field of materials science by combining multiple metallic elements in nearly equal proportions. This seemingly random distribution of atoms leads to extraordinary mechanical properties such as high fracture resistance, exceptional toughness, and resistance to deformation. However, the microscopic structural features that govern these properties are not entirely random — they are influenced by chemical short-range order (SRO), a tendency for certain atomic arrangements to occur more frequently than others.

Machine Learning Potentials for High-Entropy Alloys

Machine learning potentials provide new insights into chemical short-range order in CrCoNi high-entropy alloys.

A recent study by Yifan Cao, Killian Sheriff, and Rodrigo Freitas, published in npj Computational Materials, introduces a systematic approach to capturing SRO in high-entropy alloys using machine learning interatomic potentials (MLPs). Their research focuses on the CrCoNi alloy, a benchmark system in the study of HEAs, which has already demonstrated remarkable damage tolerance even at cryogenic temperatures.

Why Short-Range Order Matters

Chemical short-range order influences a range of fundamental properties in metallic alloys. It impacts:

  • Dislocation mobility, which governs how metals deform under stress.
  • Grain boundary stability, affecting strength and durability.
  • Stacking-fault energy, a crucial parameter for predicting deformation mechanisms such as twinning and slip.
  • Phase stability, determining whether a material favors face-centered cubic (fcc) or hexagonal close-packed (hcp) crystal structures.

Traditional computational methods such as density-functional theory (DFT) struggle to capture SRO at realistic length scales due to computational limits. This is where machine learning potentials become powerful — they enable simulations with thousands of atoms while preserving the quantum-level accuracy of DFT.

The Machine Learning Approach

The authors employ Moment Tensor Potentials (MTPs), a class of MLPs known for their ability to systematically improve in accuracy. By carefully designing training datasets that capture the complexity of atomic motifs in CrCoNi, the researchers develop potentials that not only predict energies and forces but also reproduce subtle short-range ordering effects.

Their systematic training strategy progresses from random chemical configurations to more physically informed datasets derived from DFT Monte Carlo simulations. This leads to models (such as their TS-2 and TS-3 datasets) capable of stabilizing both the crystal and liquid phases of CrCoNi — a major breakthrough in simulating these complex systems.

Key Findings

The study demonstrates that:

  • Energy accuracy alone is insufficient — models must be tested on their ability to reproduce material properties such as stacking-fault energy and phase stability.
  • Machine learning potentials trained with targeted chemical motifs significantly outperform those trained on random data alone.
  • Large-scale simulations (up to ~10,000 atoms) are required to fully capture the length scales of SRO, beyond the reach of traditional DFT.
  • SRO strongly influences temperature-dependent stacking-fault energies and fcc–hcp phase stability, providing new insights into mechanical behavior.

Broader Implications

The ability to capture short-range order using machine learning opens the door to rational materials design. By better understanding the interplay between atomic structure and mechanical properties, scientists can engineer high-performance alloys for aerospace, energy, and quantum technologies. Moreover, the design principles outlined in this work are generalizable to a wide variety of metallic systems, including ordered compounds and defect-rich materials.

Conclusion

This research highlights a new paradigm: accurate modeling of high-entropy alloys requires more than just raw computational accuracy — it requires physically meaningful training data that reflects the chemical complexity of real materials. Machine learning potentials, when designed with this principle in mind, provide unprecedented fidelity in simulating the properties of HEAs.

Footnote: This blog article was prepared with the assistance of AI technologies.

Sponsored by PWmat (Lonxun Quantum) – a leading developer of GPU-accelerated materials simulation software for cutting-edge quantum, energy, and semiconductor research. Learn more about our solutions at: https://www.pwmat.com/en

📘 Download our latest company brochure to explore our software features, capabilities, and success stories: PWmat PDF Brochure

🎁 Interested in trying our software? Fill out our quick online form to request a free trial and receive additional information tailored to your R&D needs: Request a Free Trial and Info

📞 Phone: +86 400-618-6006
📧 Email: support@pwmat.com

Original research source: npj Computational Materials

Hashtags: #MaterialsScience #HighEntropyAlloys #MachineLearning #QuantumMaterials #ComputationalMaterials #AlloyDesign #Nanotechnology #PWmat

Comments

Post a Comment

Popular posts from this blog

AI Tools for Chemistry: The ‘Death’ of DFT or the Beginning of a New Computational Era?

Image
For decades, Density Functional Theory (DFT) has been the workhorse of quantum chemistry and materials modeling. From predicting electronic structures to optimizing molecular geometries, DFT has powered countless breakthroughs in fields as diverse as catalysis, materials design, and condensed matter physics. But recent announcements in the scientific community have caused ripples: is DFT “dead” — or is this simply the dawn of a new computational era powered by artificial intelligence ? The Shock of Declaring a Field “Dead” The provocative statement came during the EuChemS Inorganic Chemistry Conference , where theoretical chemist Markus Reiher announced the “death of DFT.” His claim wasn’t about obsolescence in a literal sense, but about a paradigm shift: AI models can now deliver DFT-level accuracy at a fraction of the cost , fundamentally changing how chemists approach molecular modeling. This echoes previous moments in scientific history where disruptive technologi...
Read more

Revolutionize Your Materials R&D with PWmat

Image
GPU-Accelerated Simulation Software for Cutting-Edge Research Are you working in quantum materials , semiconductors , or energy materials ? Then it’s time to supercharge your research with PWmat by Lonxun Quantum – a leader in GPU-accelerated first-principles simulation software designed to meet the evolving demands of advanced materials science. PWmat offers a powerful suite of high-performance computing tools specifically optimized for modern NVIDIA GPUs , helping researchers: ✅ Simulate complex materials at quantum scale ✅ Accelerate DFT calculations ✅ Reduce time-to-result drastically ✅ Optimize large-scale simulations with intuitive workflows Whether you're in academia, industry, or a national lab , PWmat is trusted by institutions and scientists across the globe pushing the boundaries of what’s possible in computational materials science. 🎁 Clai...
Read more

Quantum Chemistry Meets AI: A New Era for Molecular Machine Learning

Image
In a powerful leap forward for computational chemistry and materials design, researchers from Carnegie Mellon University have developed a new kind of molecular machine learning model—one that goes beyond simplistic molecular graphs and integrates fundamental principles of quantum chemistry. Their work, published in Nature Machine Intelligence , could radically enhance our ability to predict chemical behavior, optimize catalysts, and discover new drugs, all while using less data and gaining more scientific insight. The Problem: Traditional Molecular Representations Fall Short Molecular machine learning (ML) models underpin key processes in drug discovery, materials science, and catalysis. However, most existing models use simplified representations such as SMILES strings, 2D graphs, or coordinate matrices. These techniques often lack the quantum-chemical information that governs how molecules truly behave—particularly electron interactions that define reactivity, stability, and...
Read more