Classification and prediction of bulk densities of states and chemical attributes with machine learning techniques

Claudio Sánchez Pérez de Amézaga, Víctor M. García-Suárez and Juan L. Fernández-Martínez

Applied Mathematics and Computation, 2022, vol. 412, issue C

Abstract: The electronic structure encapsulated in the density of states (DOS) is key to explain several physical properties of any material, i.e. from the knowledge of the DOS and its dependence on certain parameters it is possible to obtain a lot of information on the type of material and its behaviour. We outline in this article a series of methods that can be employed to classify and predict bulk DOS in an easy and efficient manner. These methods are based on machine learning techniques that are used to classify the elements and extract information from the DOS curves. We focus in particular in the bulk DOS of d-elements with different crystal lattices. We use a clusterization algorithm based on information obtained from the DOS, which is able to properly classify the elements in groups that are electronically similar. We find four groups clearly differentiated, whose more representative elements are scandium, iron, gold and mercury, and which tend to crystallize in similar lattice structures in their ground state. We further reduce the dimensionality of the data and find a new basis of DOS that, along with chemical properties of each type of element and basic information encapsulated in the DOS, is able to predict with a high degree of accuracy most of the original curves. We apply such basis and information to predict with reasonably accuracy the DOS curve of different elements. We also use that data to predict the chemical properties and the correlations between them. In particular, we apply the algoritm to calculate the Pauling electronegativity and find a rather good agreement between the predicted and real values. Finally, from intrinsic parameters, the DOS of other elements and additional information that is needed to properly describe the electronic structure, we predict the particular DOS of a particular element with a good degree of accuracy, including the width and general shape of the central part associated to the d states. These calculations prove that, by using pure electronic information encapsulated in the bulk DOS, it is possible to univocally classify materials and predict in a fast and accurate manner different physical and chemical properties.

Keywords: Density of states; Machine learning; Electronic structure; Chemical fingerprints (search for similar items in EconPapers)
Date: 2022
References: View complete reference list from CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0096300321006718
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:apmaco:v:412:y:2022:i:c:s0096300321006718

DOI: 10.1016/j.amc.2021.126587

Access Statistics for this article

Applied Mathematics and Computation is currently edited by Theodore Simos

More articles in Applied Mathematics and Computation from Elsevier
Bibliographic data for series maintained by Catherine Liu ().