Software Automates Nanoscale Material Analysis, Bypassing Specialist Expertise Requirements

Predicting vibrational spectra remains a significant challenge in materials science despite advances in electron microscopy techniques. Harrison A. Walker (Vanderbilt University), Thomas W. Pfeifer and Jordan A. Hachtel (Oak Ridge National Laboratory), alongside Paul M. Zeiger (University of Washington) et al., have addressed this limitation by developing PySlice, a new automated framework for calculating vibrational electron energy-loss spectra directly from material structures. This research is significant because PySlice integrates universal machine learning interatomic potentials with the Time Autocorrelation of Auxiliary Wavefunction method, removing the need for bespoke potential development for each system. Consequently, researchers can now routinely predict phonon dispersions, spectral diffraction patterns, and spectrum images, facilitating materials exploration, experimental design, and the generation of training data for machine learning applications.

This unified workflow delivers phonon dispersions, spectral diffraction patterns, and spectrum images, significantly reducing the computational burden associated with vibrational spectroscopy prediction.

The modular Python architecture also supports conventional electron microscopy simulations, establishing a versatile platform for imaging and diffraction studies. By leveraging universal machine learning interatomic potentials, PySlice circumvents the traditional bottleneck of creating bespoke potentials for each material.

These potentials, trained on extensive materials datasets, accurately predict interatomic forces across a wide range of chemical compositions. Combined with GPU acceleration and efficient numerical algorithms, PySlice enables high-throughput simulations and facilitates the systematic exploration of phonon physics in diverse materials families.

This capability is poised to accelerate materials discovery and provide valuable insights for interpreting complex experimental data. The development of PySlice promises to transform vibrational spectroscopy from a specialised undertaking into a routine predictive tool. Researchers can now computationally screen for optimal experimental designs, systematically investigate phonon behaviour across materials, and generate large datasets for training future machine learning models. The framework employs these potentials during the molecular dynamics simulations, accurately capturing anharmonic atomic motion and generating realistic trajectories.

Following electron propagation, a single temporal Fourier transform is applied to the resulting time-dependent exit wavefunction, efficiently recovering the full frequency spectrum of vibrational excitations. This process delivers phonon dispersions, spectral diffraction patterns, and spectrum images, providing comprehensive vibrational information.

The modular Python architecture of PySlice facilitates both conventional electron microscopy simulations and advanced analysis. The code supports calculations of diffuse scattering phenomena, including Kikuchi bands and thermal diffuse backgrounds, enhancing the realism of simulated data. Furthermore, PySlice enables high-throughput generation of simulated datasets suitable for training future machine learning models, accelerating materials exploration and computational screening for experimental design. The modular Python architecture supports conventional electron microscopy simulations, establishing a general-purpose platform for imaging and diffraction calculations.

PySlice facilitates investigations ranging from momentum-resolved phonon dispersions to spatially-resolved mapping of interfacial phonons, localized strain fields, and defect-induced vibrational states by enabling tunability between nanometer spatial resolution and precise momentum resolution. This development lowers the barrier to predicting vibrational spectroscopy from weeks of specialised work to a matter of hours.

Validation on transition metal dichalcogenides demonstrated accurate phonon dispersions and spectral diffraction patterns generated from a single script. Capabilities extend to simulating vibrational spectrum images at material interfaces and around defects, enabling nanoscale phonon mapping. PySlice facilitates high-throughput spectroscopy prediction, the creation of training data for machine learning models, and prediction-driven experimental design in vibrational electron energy loss spectroscopy.

The authors acknowledge that the current implementation focuses on specific materials systems and simulation parameters. The modular Python architecture, however, supports broader applications in transmission electron microscopy and scanning transmission electron microscopy, offering a versatile platform for imaging and diffraction calculations. Future work will likely focus on expanding the range of applicable materials and refining the simulation parameters to improve accuracy and computational efficiency.