Synthesizability and Mechanical Properties of High-Entropy Borides: First-Principles and Machine Learning Studies

Luke Moore, Ethan Fox, Bria Storr, Jayden R. Palomino, Shane A. Catledge, Yogesh K. Vohra, Cheng-Chien Chen · 2026

Read Synthesizability and Mechanical Properties of High-Entropy Borides: First-Principles and Machine Learning Studies by Luke Moore, Ethan Fox, Bria Storr, Jayden R. Palomino, Shane A. Catledge, Yogesh K. Vohra, Cheng-Chien Chen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-30540.txt.

Introduction

Luke Moore and six colleagues screen hundreds of ceramic recipes in the computer to find which extreme-heat materials can actually be made. High-entropy borides mix five metals into one crystal structure. They promise hardness and stability under brutal conditions. Not every mix forms a single clean phase, though. The team studies all combinations of nine transition metals. That gives one hundred twenty-six candidate five-metal borides.

Density functional theory computes properties for each candidate. An entropy forming ability score predicts single-phase synthesizability. Those predictions agree well with experiments on selected borides. Mechanical properties come from special quasi-random structures mimicking disorder. Several compounds look mechanically unstable in the calculations. Chromium-bearing ones dominate that unstable group. They are also predicted to be harder to synthesize. Machine learning models analyze the computed results further. The study maps a route to mechanically superior single-phase borides.

Catalog record

CREATORLuke Moore, Ethan Fox, Bria Storr, Jayden R. Palomino, Shane A. Catledge, Yogesh K. Vohra, Cheng-Chien Chen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.30540

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Synthesizability and Mechanical Properties of High-Entropy Borides: First-Principles and Machine Learning Studies Luke Moore, Ethan Fox, Bria Storr, Jayden R. Palomino, Shane A. Catledge, Yogesh K. Vohra, Cheng-Chien Chen Abstract We perform density functional theory (DFT) calculations to investigate five-metal high-entropy borides (HEBs) in the hexagonal AlB 2 structure, considering all 126 possible elemental combinations among the nine group 4-6 transition metals (Ti, V, Cr, Zr, Nb, Mo, Hf, Ta, and W). The entropy forming ability (EFA) descriptor is employed to evaluate their single-phase synthesizability, and the resulting EFA predictions show good agreement with the experimental data for selected HEBs. Mechanical properties are computed using special quasi-random structures. Several mechanically unstable compounds – primarily those containing Cr – are also predicted to be less synthesizable. Machine learning (ML) models are developed to analyze the results. This combined ab initio and ML study provides a systematic roadmap for identifying mechanically superior single-phase HEBs. I Introduction Transition-metal (TM) borides are an important class of ultra-high-temperature ceramics known for their exceptional hardness, incompressibility, and chemical stability, making them attractive for applications under extreme conditions 1 ; 2 ; 3 . Their superior mechanical properties originate from strong covalent boron-boron bonds together with robust metal-boron bonding. More recently, high-entropy borides (HEBs) have emerged as a promising subclass of compositionally complex ceramics 4 ; 5 ; 6 ; 7 ; 8 ; 9 . High-entropy materials consist of five or more elements in near-equimolar ratios, and they can be stabilized by configurational entropy and exhibit properties beyond those of their binary or ternary counterparts 10 ; 11 ; 12 . Recent experiments have successfully synthesized a variety of HEBs and demonstrated their excellent hardness, thermal stability, and oxidation resistance 4 ; 13 ; 14 ; 15 ; 16 ; 17 . These advances suggest that HEBs represent a rich materials-design space for next-generation extreme applications. Despite their promise, not all elemental combinations can form stable single-phase HEBs. The vast compositional space makes experimental trial-and-error approaches prohibitively expensive for systematically identifying synthesizable HEBs. Consequently, reliable computational descriptors are highly desirable. Several metrics, including entropy forming ability (EFA) 18 , disordered enthalpy-entropy descriptor (DEED) 19 , mixed enthalpy-entropy descriptor (MEED) 20 , as well as designing principles based on lattice mismatch and valence electron count 21 ; 22 have shown success in predicting the synthesizability of high-entropy ceramics. Comprehensive assessments across large families of compositions can thereby accelerate material discovery. Beyond synthesizability, predicting the mechanical properties of HEBs and understanding the effects of disorder remain challenging. Accurate first-principles modeling typically requires large supercells to capture their disordered atomic arrangements. The special quasi-random structure (SQS) approach 23 ; 24 has emerged as an effective framework for representing substitutional disorder in periodic density functional theory (DFT) calculations and has successfully reproduced the structural and mechanical properties of various high-entropy ceramics 25 ; 26 ; 14 ; 27 ; 28 ; 16 ; 29 . However, comprehensive datasets for HEBs remain limited, motivating further systematic investigations and data-driven materials design efforts. Such SQS-based studies can provide valuable insights into the composition-structure-property relationships of HEBs. In this work, we combine EFA analysis, DFT-SQS calculations, and machine-learning (ML) techniques to systematically study all 126 five-metal HEBs formed from the nine group 4-6 transition metals (Ti, V, Cr, Zr, Nb, Mo, Hf, Ta, and W). We compute their synthesizability and mechanical properties, and compare the results with available experimental data. The calculations reveal that several Cr-containing compounds exhibit large discrepancies in the elastic moduli between the Voigt and Reuss approximations 30 ; 31 – indicative of strong elastic anisotropy or mechanical instability – and they also tend to exhibit low EFA values and reduced synthesizability. We further employ ML models to analyze the DFT dataset and identify key compositional and structural descriptors governing the HEB properties. In particular, lattice-mismatch-related features associated with bond-length distributions are found to be correlated with mechanical strength, highlighting the role of disorder. Together, these results provide a systematic roadmap for identifying synthesizable HEBs with superior mechanical properties for applications in extreme environments. II Computational Methods II.1 First-Principles Calculation First-principles calculations are performed using the Vienna Ab Initio Simulation Package (VASP) 32 ; 33 , a plane-wave pseudopotential implementation of density functional theory (DFT). The Perdew-Burke-Ernzerhof generalized gradient approximation (PBE-GGA) exchange-correlation functional is employed 34 . The electronic self-consistent field calculations are converged to within 10^{-6} eV, while ionic relaxations are performed until the residual forces on all atoms are smaller than 10^{-2} eV/Å. A plane-wave energy cutoff of 450 eV is used, corresponding to at least 1.4 times the maximum recommended cutoff among the constituent pseudopotentials. Brillouin-zone integrations are performed using \Gamma -centered k -point meshes with a k-points-per-reciprocal-atom (KPPRA) value of 6000, automatically generated by the AFLOW package 35 for each input structure. To evaluate the synthesizability of hexagonal HEBs in the AlB 2 structure, DFT calculations are carried out for the minimal 15-atom unit cells generated by the AFLOW-Partial OCCupation (AFLOW-POCC) algorithm 36 ; 37 . Each unit cell contains five transition-metal atoms and ten boron atoms. AFLOW-POCC generates a set of distinct 15-atom structures that represent the specified stoichiometry and crystallographic symmetries. The synthesizability can then be assessed from the resulting energy distribution of these distinct POCC structures through the EFA descriptor discussed later. To evaluate the mechanical properties, DFT calculations are performed using the special quasi-random structure (SQS) 23 ; 24 , which provides the best periodic-supercell approximation to a fully disordered state for a given number of atoms. For each five-metal HEB composition, a 60-atom SQS supercell is first fully relaxed, followed by calculations of the elastic constants using the VASP stress-strain method. The resulting elastic constants are then used to evaluate the mechanical properties based on the Voigt-Reuss-Hill approximation 30 ; 31 . Previous studies on HEBs have shown that larger SQS supercells up to 150 atoms do not lead to substantial differences in the calculated mechanical properties compared to those obtained using 60-atom supercells 16 . II.2 Machine Learning Simulation The mechanical properties computed by DFT are further utilized to develop machine-learning (ML) regression models. For each HEB composition, such as CrHfMoNbTaB 10 , a set of 60 composition descriptors is generated 38 ; 39 . These descriptors comprise four categories: stoichiometric, elemental-property-based, valence-orbital, and ionic-bonding attributes. The stoichiometric attributes (3 total) quantify compositional diversity through the L_{p} norms of the elemental fractions ( p=0,2,3 ). The elemental-property-based attributes (50 total) represent statistical descriptors (minimum, maximum, range, fractional weighted mean, and deviation) of ten elemental properties: atomic number, atomic mass, periodic-table row and column, atomic radius, electronegativity, and the numbers of valence s , p , d , and f electrons. The valence-orbital attributes (4 total) quantify the fractional occupations of the s , p , d , and f orbitals. The ionic-bonding attributes (3 total) characterize the potential ionic character through charge-balance and electronegativity-based metrics. These compositional features are obtained using the Python Materials Genomics (pymatgen) package 40 . Additional 19 structural features based on the fully relaxed 60-atom SQS supercells are further incorporated to improve the ML models. These structural features include nearest-neighbor (NN) distances extracted from the supercells and are grouped into six categories: metal-boron (M-B), metal-metal along

Read the complete text