Litcius/Paper detail

Prediction of frontier band spin splitting in 2D perovskites via deep neural networks

Deyang Liang, Zheng Pan, Siyuan Zhang, Zhaoyang Zhang, Ruyi Song, Rundong Zhao

2025npj Computational Materials6 citationsDOIOpen Access PDF

Abstract

Two-dimensional (2D) hybrid organic-inorganic perovskites (HOIPs) have strong potential for optoelectronic applications due to their polarized photon absorption and emission properties. These macroscopic behaviors are intrinsically linked to microscopic symmetry breaking, particularly the emergence of momentum-dependent, non–centrosymmetric spin splitting in frontier electronic bands. To efficiently identify such spin-related phenomena, we combine first-principles calculations and deep learning models to explore the correlation between in-plane bond distortions and spin-orbit splitting. Our model achieves 100% accuracy in qualitatively identifying systems with observable spin splitting, and over 80% quantitative accuracy in predicting its magnitude and location, confirming that in-plane structural distortions are key descriptors of spin splitting. The trained model can be readily extended to real 2D HOIP systems and is expected to benefit experimentalists by enabling rapid screening and discovery of functional materials, especially in cases where ab initio calculations are not feasible due to computational cost.

Topics & Concepts

ObservablePhysicsSpin (aerodynamics)Statistical physicsSymmetry (geometry)Ab initioCondensed matter physicsArtificial neural networkPhotonAbsorption (acoustics)Ab initio quantum chemistry methodsKey (lock)Electronic structureElectronic band structureSeries (stratigraphy)Quantum mechanicsDeep learningCurrent (fluid)Computational physicsComputer scienceComplex systemMaterials scienceNanoelectronicsMachine Learning in Materials SciencePerovskite Materials and Applications2D Materials and Applications
Prediction of frontier band spin splitting in 2D perovskites via deep neural networks | Litcius