Investigations reveal magnetic phases in emerging quantum materials. Quantitative interpretation of μSR experiments depends on the key identification of the muon stopping site, traditionally addressed using density functional theory (DFT)-based structural relaxation. Conventional DFT approaches are computationally demanding, time-consuming, and require extensive calculations. A new machine-learning-based, physics-informed computational framework, MEOWN (Muon Engine for Optimised Weighted Networks), combines a Polarizable Unperturbed Electrostatic Potential (P-UEP) model with machine-learning-guided optimisation and symmetry-driven relaxation to predict energetically favourable structures.

Rapid prediction of equilibrium muon sites via integrated physical modelling and machine learning

MEOWN predicts the equilibrium muon stopping site within few minutes, representing a substantial improvement over traditional density functional theory methods which previously required hours or even days to achieve comparable results. This unlocks possibilities for real-time analysis of muon spin rotation/relaxation data, something impossible with conventional computational approaches due to their intensive demands on processing power and time. Integrating physics principles such as electrostatic interactions, electronic screening, ion polarization effects, and quantum zero-point motion into an optimised workflow allowed scientists at Rajiv Gandhi Institute of Petroleum Technology to develop this machine-learning framework.

Five benchmark materials, manganese silicide, cobalt fluoride, calcium fluoride, lithium fluoride and sodium fluoride, were used by researchers at the Rajiv Gandhi Institute of Petroleum Technology to validate MEOWN. The predicted locations closely matched those obtained using established density functional theory plus muon methods confirming its reliability across different chemical compositions and ability to accurately predict dipolar fields, a measure of magnetic behaviour around the stopping site.

Although current evaluations do not yet extend beyond crystalline solids or address complex scenarios involving defects or disordered systems limiting immediate widespread application, it establishes an important foundation for broader use in more challenging material environments where traditional computational techniques struggle with both speed and accuracy.

Machine learning accelerates accurate prediction of muon locations in crystalline structures

Pinpointing precisely where implanted muons come to rest within a crystal structure is reliable interpretation’s cornerstone when probing complex materials such as those investigated for magnetism or superconductivity; this location dictates how the muon interacts with its surroundings and influences experimental results. This capability allows understanding fundamental properties that are otherwise difficult to measure directly. A collaboration between The researchers Source and Rajiv Gandhi Institute of Petroleum Technology resulted in MEOWN, a machine-learning framework rapidly predicting these stopping sites while sharply reducing calculation demands compared to conventional techniques relying on detailed modelling of electronic structure. By combining physics principles with machine learning optimisation, MEOWN circumvents lengthy calculations traditionally required by density functional theory methods whilst maintaining physically sound results validated against benchmark compounds like manganese silicide and calcium fluoride. The resulting speedup enables more efficient analysis of experimental data and opens avenues for rapid screening of potential quantum materials previously limited by processing demands; this facilitates exploration of a wider range of candidate substances in the search for novel properties.

MEOWN accurately predicted energetically favourable muon stopping sites within crystalline solids using a machine-learning framework. This is important because identifying these locations is fundamental to interpreting experiments that probe magnetic behaviour and superconductivity in complex materials. Validated against materials including manganese silicide, calcium fluoride, lithium fluoride, and sodium fluoride, MEOWN achieves reliable results with substantially reduced computational cost compared to conventional methods. The researchers note current evaluations are limited to crystalline solids but establish a foundation for applying the method to more challenging material environments.

👉 More information
🗞 Development of a Physics-Informed Neural Framework, MEOWN, for Rapid Prediction of Muon Stopping Sites in Crystalline Materials, for understanding Quantum Magnet employing Muon Spectroscopy
✍️ A. Pandey, K. Sharma, S. Ghosh, G. Roy and T. Basu
🧠 ArXiv: https://arxiv.org/abs/2609.17063

More like this