Accurate Crystal Structure Prediction of New 2D Hybrid Organic Inorganic Perovskites

TL;DR

This work introduces a MACE-based machine-learned interatomic potential (MLIP) for lead-based 2D HOIPs, enabling accurate single-point and MD relaxations across a broad compositional space. Coupled with a simple random structure search, the approach rediscoveres known ground-state structures and successfully predicts new perovskites, which are subsequently synthesized to validate the predictions. The combination enables high-throughput, scalable screening of thousands of organic cations and inorganic layers, revealing a delicate landscape of low-energy minima that informs experimental targeting. The framework thus provides a practical path toward rapid discovery and verification of novel 2D HOIPs with tunable optoelectronic properties.

Abstract

Low dimensional hybrid organic-inorganic perovskites (HOIPs) represent a promising class of electronically active materials for both light absorption and emission. The design space of HOIPs is extremely large, since a diverse space of organic cations can be combined with different inorganic frameworks. This immense design space allows for tunable electronic and mechanical properties, but also necessitates the development of new tools for in silico high throughput analysis of candidate structures. In this work, we present an accurate, efficient, transferable and widely applicable machine learning interatomic potential (MLIP) for predicting the structure of new 2D HOIPs. Using the MACE architecture, an MLIP is trained on 86 diverse experimentally reported HOIP structures. The model is tested on 73 unseen perovskite compositions, and achieves chemical accuracy with respect to the reference electronic structure method. Our model is then combined with a simple random structure search algorithm to predict the structure of hypothetical HOIPs given only the proposed composition. Success is demonstrated by correctly and reliably recovering the crystal structure of a set of experimentally known 2D perovskites. Such a random structure search is impossible with ab initio methods due to the associated computational cost, but is relatively inexpensive with the MACE potential. Finally, the procedure is used to predict the structure formed by a new organic cation with no previously known corresponding perovskite. Laboratory synthesis of the new hybrid perovskite confirms the accuracy of our prediction. This capability, applied at scale, enables efficient screening of thousands of combinations of organic cations and inorganic layers.

Figures (9)

• Figure 1: (a): Key properties of the perovskites in the compiled dataset. Note that some structures have multiple organic cations, but the upper right histogram shows only the size of the largest cation in each structure. (b): Examples of 2D HOIP structures in the dataset.
• Figure 2: The relative force uncertainty and actual force error for one HOIP as a function of time during an MD simulation. The unstable potential (dashed lines) occasionally exceeds the relative force uncertainty threshold (red solid line at 0.2) with actual force errors as large as 100 meV/Å, while the final potential remains far below the threshold with force errors fluctuating between 10 to 20 meV/Å.
• Figure 3: The parity plot of (a) forces, and (b) energy (per atom) for training and test set samples.
• Figure 4: Evaluating the MACE model for geometry relaxations of experimentally reported structures. (a) Histogram of the RMSD between the DFT relaxed and the MACE relaxed structures for the entire dataset. (b) Comparison of the total RDF for a test set structure after relaxing with DFT and MACE. (c) The distribution of the Wasserstein distance between the RDFs given by the DFT relaxed structure and MACE relaxed structure for 137 unique HOIPs in the training and test set.
• Figure 5: Overview of the structure generation algorithm for creating initial guesses for the random structure search process. To make the figure more readable, the unit cell is only shown for one of the four candidate structures in the lower panel.