Skip to main content
Research Highlight

Machine Learning Force Fields for Neutron Scattering Data Analysis

Brief: A workflow has been developed to simulate neutron scattering spectra using complex atomistic models enabled by machine learning force fields.

Accomplishment:  To take advantage of the machine learning force fields (MLFFs) to study neutron scattering related phenomena, a preliminary workflow has been established on the edge computing resources (DGX) at Spallation Neutron Source (SNS).1,2 It has been demonstrated that the developed MLFFs can reproduce inelastic neutron scattering spectra, at a level of accuracy close to density functional theory (DFT) simulations but with a speed several orders of magnitude faster. The excellent performance is encouraging, and it shows the immediate impact Artificial Intelligence (AI) can make on the analysis of neutron data at Oak Ridge National Laboratory (ORNL) and neutron scattering facilities elsewhere. This opens tremendous opportunities for advanced neutron scattering data analysis involving complex atomistic models, for which direct first-principles/DFT simulations are impractical.

Data analysis and interpretation are a major bottleneck for the efficiency and productivity at high flux neutron scattering facilities. The structural and dynamical information, which is usually the goal of the measurements, is encoded in the neutron scattering patterns, and a direct forward solution is often difficult to come by. To this end, an atomistic model is an ideal companion to explain the underlying origin of the neutron scattering features.3,4 Such models, unfortunately, are often difficult to build and analyze. Current workflows based on DFT suffer from multiple limitations (e.g., time and length scales), which are becoming a major roadblock as more complex materials and processes are studied with neutrons. Recent development in AI and machine learning shows that it is possible to describe interatomic interactions with a trained neural network, so that a surrogate system driven by MLFFs can be used to model structure and dynamics beyond the DFT limit, with high efficiency and accuracy.5 In this project, a workflow has been developed on DGX to train and utilize MLFFs for neutron scattering data analysis. Promising preliminary data suggests that this approach has the potential to transform model-based neutron data analysis by allowing us to better address many challenging problems such as anharmonicity, disordered and heterogeneous systems, chemical processes, nuclear quantum effects, and jump diffusion over multiple length and time scales. 

Acknowledgement: This research was funded by the AI Initiative, as part of the Laboratory Directed Research and Development Program of Oak Ridge National Laboratory, managed by UT-Battelle, LLC, for the U.S. Department of Energy (DOE).  

Publications and presentations resulting from this work: Manuscript in preparation.

Contact: Yongqiang Cheng (

Team: Yongqiang Cheng, Yuanpeng Zhang, Gabriele Sala, Matthew Stone, Andrei Savici, Daniel Pajerowski


  1. DEEPMD: Zhang, L.;  Han, J.;  Wang, H.;  Car, R.; E, W., Deep Potential Molecular Dynamics: A Scalable Model with the Accuracy of Quantum Mechanics. Physical Review Letters 2018, 120 (14), 143001.
  2. NEQUIP: Batzner, S.;  Smidt, T.;  Sun, L.;  Mailoa, J.;  Kornbluth, M.;  Molinari, N.; Kozinsky, B., SE(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials. 2021.
  3. Cheng, Y. Q.;  Daemen, L. L.;  Kolesnikov, A. I.; Ramirez-Cuesta, A. J., Simulation of Inelastic Neutron Scattering Spectra Using OCLIMAX. Journal of Chemical Theory and Computation 2019, 15 (3), 1974-1982.
  4. Cheng, Y. Q.;  Kolesnikov, A. I.; Ramirez-Cuesta, A. J., Simulation of Inelastic Neutron Scattering Spectra Directly from Molecular Dynamics Trajectories. Journal of Chemical Theory and Computation 2020, 16 (12), 7702-7708.
  5. Unke, O. T.;  Chmiela, S.;  Sauceda, H. E.;  Gastegger, M.;  Poltavsky, I.;  Schütt, K. T.;  Tkatchenko, A.; Müller, K.-R., Machine Learning Force Fields. Chemical Reviews 2021, 121 (16), 10142-10186.