Machine Learning Predicted Superhard Tungsten Nitride

The research team led by Professor Sun Jian and professor Wang Huitian of Nanjing University has developed machine-learning to accelerate crystal structure search and predicted a superhard tungsten nitride which is the hardest transition metal nitride so far. Related papers and cover articles were published in Science Bulletin, No. 13 of 2018.

Machine learning algorithms have made remarkable progress in many fields, but the application in crystal structure prediction still needs to be developed. Crystal structure search usually deals with a large number of alternative structures. If we adopt first-principles calculation for every alternative structure, it would consume a lot of computational resources.

The program is to use machine learning to fit a model. This model will first screen the alternative crystal structure, which can improve the efficiency of crystal structure search.

Transition metal light element compounds, particularly tungsten nitrides, have been extensively studied as hard materials because of their high incompressibility and bulk modulus. However, tungsten hard nitrogen compounds with super-hardness (Vickers hardness exceeding 40 GPa) have not been found yet. Because d electron of the transition metal atom in the tungsten nitrogen compound passes through the Fermi surface and making it metallic, thereby greatly reducing the hardness of the material. Designing non-metallic tungsten-nitrogen compounds may gain the new materials with superhard mechanical properties.

Based on the pioneers’ studies, the research team has summed up three clues to fins superhard transition metal light element compounds: high-pressure stable and ambient-pressure metastable crystal structure, non-metallic electronic structures, and a large ratio of light elements. These clues inspire them to design the nitrogen-rich tungsten nitrides containing special nitrogen-based basic configurations, such as rings, chains, networks and frameworks etc. 

They successfully designed a non-metallic nitrogen-rich tungsten nitride h-WN6 based on these design ideas and newly-developed machine-learning accelerated crystal structure search method. It owns a sandwich-like structure which is formed by nitrogen six-membered ring and tungsten atoms. The analysis of its electron localization distribution and bonding characteristics shows that h-WN6 is an ionic crystal and a strong covalent bond with good directionality. It is a semiconductor with a small and indirect band gap. The abnormal band gap broadening occurs under compression.

The result of theoretical calculations shows that h-WN6 can be synthesized under high pressure while being metastable under ambient pressure. Surprisingly, the theory predicts that the h-WN6 structure has a Vickers hardness of about 57 GPa and a melting point of 1,900 K, which is by far the hardest transition metal nitride. At the same time, it has a high mass energy density (3.1 kJ/g) and a volumetric energy density (28.0 kJ/cm3), which is a potential material with high energy and density.

This work developed a method of machine learning accelerated crystal structure search and summed up the design idea of transition metal light element hybrid superhard material. Based on these ideas, it predicted a super hard tungsten nitride with high energy, high density and good thermal stability. This may encourage people to theoretically explore and synthesize potentially valuable materials through experiments. In addition, it has an importantly scientific significance for developing more efficient crystal structure search methods, verifying the design of superhard materials, expanding the family of superhard materials and studying the origin of hardness. 

• < Prev
• Next >