Discovery of a Low Thermal Conductivity Oxide Guided by Probe Structure Prediction and Machine Learning

Abstract

We report the aperiodic titanate Ba 10 Y 6 Ti 4 O 27 with aroom-temperature thermal conductivity that equals the lowest reported for an oxide.T he structure is characterised by discontinuous occupancy modulation of each of the sites and can be considered as aq uasicrystal. The resulting localisation of lattice vibrations suppresses phonon transport of heat. This new lead material for low-thermal-conductivity oxides is metastable and located within aq uaternary phase field that has been previously explored. Its isolation thus requires ap recisely defined synthetic protocol. The necessary narrowing of the search space for experimental investigation was achieved by evaluation of titanate crystal chemistry,prediction of unexplored structural motifs that would favour synthetically accessible new compositions,and assessment of their properties with machine-learning models.

Complex amorphous oxides: property prediction from high throughput DFT and AI for new material search

“…By collecting large amounts of structural and electronic properties, either being calculated or experimentally obtained, and combining these with machine learning, new materials and the mapping of known materials to new applications are being discovered. [75][76][77][78][79][80][81][82][83][84][85][86][87][88][89][90][91][92][93] Most of these works, however, mainly focused on crystalline materials and around variations of elements and sites in known crystalline materials. Although systems of off-stoichiometic compositions and materials of unknown structure are also getting attention, 94,95 the of amorphous materials, which is the focus of our interest, adds even more additional difficulties.…”

Machine Learning for Harnessing Thermal Energy: From Materials Discovery to System Optimization

“…120,159 The thermal conductivity of alloys can also be predicted by adding composition as another dimension for ML training. 160,161 In addition, ML can also assist the ab initio calculations of thermal conductivity, especially for high-temperature calculations, which are usually much more computationally expensive than ab initio calculation at 0 K. By performing principal component analysis and regression analysis, a correlation between 0 K force constants and 1000 K force constants can be built to accelerate the phonon scattering calculations. 48 In addition to homogeneous materials, the effects of compositional and structural factors on thermal conductivity can also be efficiently predicted with ML algorithms, mainly for nanostructures, composites, and porous materials.…”

“…A similar approach was also performed using different input data such as entries in Materials Projects and different ML algorithms like Gaussian process regression, random forest, transfer learning, and principal component analysis to map thermal conductivity with different descriptor sets. ,,,, Different from inorganic crystals, the descriptors for ML training of the thermal conductivity of polymers are more complicated, for example, the vectors of binary digits representing the chemical units. The search for high-thermal-conductivity polymers is underway but far from satisfactory considering the current progress. , The thermal conductivity of alloys can also be predicted by adding composition as another dimension for ML training. , In addition, ML can also assist the ab initio calculations of thermal conductivity, especially for high-temperature calculations, which are usually much more computationally expensive than ab initio calculation at 0 K. By performing principal component analysis and regression analysis, a correlation between 0 K force constants and 1000 K force constants can be built to accelerate the phonon scattering calculations …”

Phonon anharmonicity in binary chalcogenides for efficient energy harvesting

“…211 ML algorithms have also been used for screening various TE properties such as Seebeck coefficients and power factors for various traditional bulk (3D) and 2D TE materials such as oxides, nitrides, and half-Heuslers. 205,[212][213][214][215][216][217] Some of the databases created from these studies have been publicly available through the internet (e.g., Jarvis, 218,219 AFLOW-LIB, 220 ThermoE toolkit, 216 the Materials Project 221 ).…”