Band Gap Prediction for Large Organic Crystal Structures with Machine Learning

Abstract

Machine‐learning models are capable of capturing the structure–property relationship from a dataset of computationally demanding ab initio calculations. Over the past two years, the Organic Materials Database (OMDB) has hosted a growing number of calculated electronic properties of previously synthesized organic crystal structures. The complexity of the organic crystals contained within the OMDB, which have on average 82 atoms per unit cell, makes this database a challenging platform for machine learning applications. In this paper, the focus is on predicting the band gap which represents one of the basic properties of a crystalline material. With this aim, a consistent dataset of 12 500 crystal structures and their corresponding DFT band gap are released, freely available for download at https://omdb.mathub.io/dataset. An ensemble of two state‐of‐the‐art models reach a mean absolute error (MAE) of 0.388 eV, which corresponds to a percentage error of 13% for an average band gap of 3.05 eV. Finally, the trained models are employed to predict the band gap for 260 092 materials contained within the Crystallography Open Database (COD) and made available online so that the predictions can be obtained for any arbitrary crystal structure uploaded by a user.

Machine learning the quantum-chemical properties of metal–organic frameworks for accelerated materials discovery

“…As a point of reference, a trivial model that simply predicts the mean band gap for every MOF would have an MAE of 0.97 eV, suggesting that CGCNN captures much of the underlying chemistry. The 44,68,126,127 It is also worth noting that the experimentally measured band gaps of MOFs can vary by several tenths of an electronvolt depending on the synthesis and post-treatment conditions. 128 As such, an MAE less than 0.3 eV is promising for the identification of structure-property trends and for sorting material candidates by band gap, the latter of which is further justified by the CGCNN's high Spearman rank-order correlation coefficient of r = 0.93.…”

Spin wave excitations of magnetic metalorganic materials

“…With the database grown to a decent size we furthermore see great opportunities in training machine learning models. We have recently shown that such machine learning tools can be applied to predict basic materials properties for extremely complex organic materials hosting thousands of atoms in the unit cell and with that are outside the realm of materials which can be calculated straightforwardly using density functional theory [36]. In a similar manner, we Figure 5.…”

“…The vast increase of experimental and theoretical data obtained over the past century has opened a novel approach to materials research based on computer science methods and the construction of materials databases [18][19][20][21][22][23][24][25]. Such databases were successfully applied in mining for functional materials [21,[26][27][28][29][30][31][32] and as training * hellsvik@kth.se data for machine learning algorithms predicting complex materials properties [33][34][35][36], bypassing computationally demanding ab initio calculations. In this article we report about the implementation of a novel dataset for organic magnets into the organic materials database OMDB [19,37], available at https://omdb.mathub.io.…”

Artificial intelligence-enabled smart mechanical metamaterials: advent and future trends

“…Advantages of the novel materials and structures are obtained mainly due to the capability of AI models in finding the good balance between rationale microstructures and high accuracy, such that to effectively satisfy the performance requirements for those architected materials and structures [286][287][288][289]. Comparing with the applications in other fields, AI has not debuted in materials and structures until recent years [290][291][292][293][294][295][296][297][298][299][300]. Studies have been reported on using AI algorithms to emulate complex biological processes (e.g.…”