Universal Foundation Model ‘NEP89’ Achieves 3-Order-of-Magnitude Computational Efficiency Boost for Atomic Simulations Across 89 Elements

July 12, 2026

(unknown, likely a preprint server like arXiv or journal) International

Overview

A new foundation model, ‘NEP89,’ based on a neuroevolution potential architecture, has been introduced, covering inorganic and organic materials across 89 elements. NEP89 achieves both empirical potential-level speed and high accuracy, notably improving computational efficiency by 3 to 4 orders of magnitude compared to existing foundation models. This dramatic increase in efficiency enables previously impossible large-scale atomic simulations, opening new avenues in materials science research.

In Depth

Key Findings

A groundbreaking foundation model, ‘NEP89,’ has been developed, based on a neuroevolution potential architecture, capable of atomic simulations for inorganic and organic materials across 89 elements. The key feature of NEP89 is its ability to achieve both computational speeds comparable to empirical potentials and high accuracy approaching first-principles calculations. It successfully boosts computational efficiency by 3 to 4 orders of magnitude compared to existing machine-learning-based foundation models, making large-scale and complex atomic simulations a practical reality.

Technical / Clinical Details

The core of NEP89 lies in its high-accuracy learning of the potential energy surface describing interatomic interactions, utilizing neuroevolutionary methods. This model efficiently extracts the relationship between atomic arrangements and energy from high-precision, quantum mechanics-based datasets, constructing a universal potential applicable to various material systems. It excels in faithfully reproducing interactions in complex alloys with multiple elements and materials exhibiting diverse bonding types, such as covalent, ionic, and van der Waals forces. The dramatic increase in computational efficiency was achieved through optimized model architecture and efficient data sampling strategies. For instance, systems previously limited to simulations of a few thousand atoms can now be simulated with hundreds of thousands of atoms or more using NEP89 within a reasonable computational timeframe. This allows for detailed atomic-level analysis of complex phenomena influencing macroscopic material properties, such as solidification processes, phase transitions, and defect behaviors.

Background & Context

Atomic simulation is an extremely powerful tool in material design and process development, but its high computational cost has been a long-standing bottleneck. Especially for complex materials containing many elements or requiring tracking long-time dynamics, high-accuracy methods like Density Functional Theory (DFT) were too computationally intensive, while empirical potentials had limitations in accuracy. The advent of high-performance machine learning interatomic potentials (MLIPs) like NEP89 resolves this trade-off, enabling both accuracy and computational efficiency, thus significantly expanding the possibilities for large-scale simulations in materials informatics. This will accelerate the exploration of new materials, property prediction, and elucidation of reaction mechanisms, contributing to innovation in various industrial sectors including aerospace, energy, semiconductors, and medicine.

Strategic Significance & Outlook

NEP89 holds the potential to become a new standard for simulations in materials science research. Future work will aim to further develop this foundation model, integrating predictive capabilities for a wider range of material properties (e.g., thermal conductivity, electrical conductivity, stress response). Furthermore, by coupling NEP89 with existing materials databases and experimental data, the construction of even more accurate and reliable material design platforms is anticipated. This technology will expand the frontier of computational materials science and represents a crucial step towards realizing AI-driven material discovery. In the long term, it is expected to provide rapid solutions to complex material design challenges and accelerate the creation of groundbreaking new materials that contribute to the sustainable development of human society.

Source: https://materialsmodeling.org/publications/2026-NEP89-Universal-neuroevolution-potential-for-inorganic-and-organic-materials-across-89-elements/

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.