Machine Learning‐Aided Crystal Facet Rational Design with Ionic Liquid Controllable Synthesis

Abstract

Crystallographic facets in a crystal carry interior properties and proffer rich functionalities in a wide range of application areas. However, rational prediction, on-demand customization, and accurate synthesis of facets and facet junctions of a crystal are enormously desirable but still challenging. Herein, a framework of machine learning (ML)-aided crystal facet design with ionic liquid controllable synthesis is developed and then demonstrated with the star-material anatase TiO 2. Aided by employing ML to acquire surface energies from facet junction datasource, the relationships between surface energy and growth conditions based on the Langmuir adsorption isotherm are unveiled, enabling to develop controllable facet synthetic strategies. These strategies are successfully verified after applied for synthesizing TiO 2 crystals with custom crystal facets and facet junctions under tuning ionic liquid [bmim][BF 4 ] experimental conditions. Therefore, this innovative framework integrates data-intensive rational design and experimental controllable synthesis to develop and customize crystallographic facets and facet junctions. This proves the feasibility of an intelligent chemistry future to accelerate the discovery of facet-governed functional material candidates.

Nanoparticle Size Effects on Phase Stability for Molybdenum and Tungsten Carbides

“…We wanted to ensure that we calculated the energy of all stable facets for the Wulff constructions; however, to calculate the DFT energy of higher Miller index slabs and all possible terminations would be computationally expensive. Motivated by the need to reduce the volume of DFT calculations and machine learning models that have previously been demonstrated to predict surface energies with reasonable accuracy, − we trained a neural network (NN) force field for the rapid estimation of energies for new slabs at an accuracy comparable with DFT.…”

Machine learning accelerated calculation and design of electrocatalysts for CO2 reduction

“…Moreover, reliable and sufficient data sources are critical for ML assistance in CO 2 RR, and there is a current lack of sufficiently diverse and extensive experimental datasets for model training and validation. The experimental complexity grows exponentially with the number of variables, limiting most searches to narrow regions of the material space 49,60,61 . Therefore, designing materials directly by trial and error is costly.…”

“…The experimental complexity grows exponentially with the number of variables, limiting most searches to narrow regions of the material space. 49,60,61 Therefore, designing materials directly by trial and error is costly. Interestingly, the emergence of high-throughput experiments performed by robotic chemists, 49 chemical synthesis machines, 62 and a programmable system of materials synthesis 63 could pave a new way for the generation of high-quality datasets.…”

Electrocatalysis enabled transformation of earth-abundant water, nitrogen and carbon dioxide for a sustainable future

“…Due to advancements in machine learning and material genome databases, accelerating catalyst discovery by highthroughput assessment and non-supervised analytical techniques such as AI algorithms, aided with the identification of key synthetic parameters, is realistic. 453,454 Moreover, the state-of-the-art computer-aided robotic and automated facilities enables autonomous catalyst synthesis, characterization and performance evaluation, which could significantly boost discovery of the advanced catalysts for electrochemical conversion of water, nitrogen, carbon dioxide and the other molecules. 455 This critical review with more than 500 references along with groups of expertise helped to lay the foundations in this research field.…”