Correlating Structural Disorder and Inelastic Neutron Scattering Spectra through Graph-Based Voxelization Technique

Understanding structural disorder and configurational entropy is key to improving the performance of amorphous and liquid crystalline materials such as organic semiconductors. Inelastic Neutron Scattering (INS) gives highly detailed spectra that are sensitive to atomic motion, but connecting these spectra directly to molecular structures from Molecular Dynamics (MD) simulations remains difficult. We present the Voxel ML Graph Fusion framework, a new approach that uses graph-based method to link atomic structure to INS spectra. The method divides MD trajectories into small 3D regions, or voxels, and builds graph models that describe both molecular connections and how neighboring molecules interact. This captures how local structure changes with disorder. We applied the method to 56 simulated structures ranging from crystalline to amorphous and compared their INS spectra with experimental data from the VISION instrument at Oak Ridge National Laboratory. Analysis of the simulated spectra, especially the ratio of peak to baseline intensity, shows clear agreement with experiments. These results suggest that the graph features capture the real structural changes driving the spectral response and open a new way to estimate configurational entropy directly from INS data.