Million-scale data integrated deep neural network for phonon properties of heuslers spanning the periodic table

Abstract

Existing machine learning potentials for predicting phonon properties of crystals are typically limited on a material-to-material basis, primarily due to the exponential scaling of model complexity with the number of atomic species. We address this bottleneck with the developed Elemental Spatial Density Neural Network Force Field, namely Elemental-SDNNFF. The effectiveness and precision of our Elemental-SDNNFF approach are demonstrated on 11,866 full, half, and quaternary Heusler structures spanning 55 elements in the periodic table by prediction of complete phonon properties. Self-improvement schemes including active learning and data augmentation techniques provide an abundant 9.4 million atomic data for training. Deep insight into predicted ultralow lattice thermal conductivity (<1 Wm−1 K−1) of 774 Heusler structures is gained by p–d orbital hybridization analysis. Additionally, a class of two-band charge-2 Weyl points, referred to as “double Weyl points”, are found in 68% and 87% of 1662 half and 1550 quaternary Heuslers, respectively.

Anti-bonding mediated record low and comparable-to-air lattice thermal conductivity of two metallic crystals

“…The MSD increases with the increase in temperature, but k p-p p has an inverse relationship with MSD, i.e., the higher the MSD, the lower the k p-p p , as also evidenced by our recent big data analysis on B29 000 cubic structures. 22 The MSDs of all atoms in CsKNa and PbAuGa are much higher than those of Pd and Ag atoms in pure metals. The higher MSD indicates that the atoms can do a periodic movement with large displacement away from their equilibrium position, and also implies that atoms are loosely bonded with neighbors and indeed act as intrinsic rattlers.…”

“…Recent advances in artificial intelligence (AI) techniques have significantly accelerated the discovery of materials with optimal properties for various applications, including superconductivity, catalysis, and thermoelectricity. [19][20][21][22] In our recent high-throughput prediction of the phonon transport properties of large-scale inorganic crystals based on deep learning, 22,23 after thoroughly screening 80 000 cubic crystals from the Open Quantum Materials Database (OQMD), 24,25 two metallic compounds, namely PbAuGa and CsKNa, are identified with anomalously low phononic thermal conductivity, having the lowest phononic thermal conductivity among all crystalline materials we have known so far. Obviously, the phononic thermal conductivity of metallic PbAuGa and CsKNa is well below the reasonable range.…”

High-throughput computational discovery of 3218 ultralow thermal conductivity and dynamically stable materials by dual machine learning models

“…216), 83 quaternary Heuslers (ABCD type with space group no. 216), 84 single perovskites (ABC 3 type with space group no. 221), 85 layered materials (such as space groups no.…”

Lattice Dynamics and Thermal Transport in Semiconductors with Anti-Bonding Valence Bands

“…Indeed, the quests for new materials with extremely low thermal conductivity have been abundant. Previous studies have utilized high-throughput screening techniques to identify potential candidates based on physical characteristics such as large atomic mass, structural information of rattling atoms, and the combination of machine-learning algorithms and automatic ab initio calculations. , However, a comprehensive search focusing on the chemical bonding character and its impact on lattice dynamics has not been carried out due to the lack of an effective descriptor. In this work, we focus on a chemical bonding signature: highest-occupied valence bands with a strong anti-bonding character in a semiconductor.…”