Anton O. Oliynyk

Perspective: Web-based machine learning models for real-time screening of thermoelectric materials properties

Chemistries Michael W. Gaultois, a) Anton O. Oliynyk, Arthur Mar, Taylor D. Sparks, Gregory J. Mulholland, and Bryce Meredig b) Materials Research Laboratory and the Department of Chemistry and Biochemistry, University of California, Santa Barbara, California, 93106, USA Department of Chemistry, University of Alberta, Edmonton, Alberta, T6G 2G2, Canada Department of Materials Science and Engineering, University of Utah, Salt Lake City, Utah, 84112, USA Citrine Informatics, Redwood City, California, 94061, USAThe experimental search for new thermoelectric materials remains largely confined to a limited set of successful chemical and structural families, such as chalcogenides, skutterudites, and Zintl phases. In principle, computational tools such as density functional theory (DFT) offer the possibility of rationally guiding experimental synthesis efforts toward very different chemistries. However, in practice, predicting thermoelectric properties from first principles remains a challenging endeavor [J. Carrete et al., Phys. Rev. X 4, 011019 (2014)], and experimental researchers generally do not directly use computation to drive their own synthesis efforts. To bridge this practical gap between experimental needs and computational tools, we report an open machine learning-based recommendation engine (http://thermoelectrics.citrination.com) for materials researchers that suggests promising new thermoelectric compositions based on pre-screening about 25 000 known materials and also evaluates the feasibility of use...

Phase equilibria in the Mo-Fe-P system at 800 °C and structure of ternary phosphide (Mo(1-x)Fe(x))3P (0.10 ≤ x ≤ 0.15).

Construction of the isothermal section in the metal-rich portion (<67 atom % P) of the Mo-Fe-P phase diagram at 800 °C has led to the identification of two new ternary phases: (Mo(1-x)Fe(x))(2)P (x = 0.30-0.82) and (Mo(1-x)Fe(x))(3)P (x = 0.10-0.15). The occurrence of a Co(2)Si-type ternary phase (Mo(1-x)Fe(x))(2)P, which straddles the equiatomic composition MoFeP, is common to other ternary transition-metal phosphide systems. However, the ternary phase (Mo(1-x)Fe(x))(3)P is unusual because it is distinct from the binary phase Mo(3)P, notwithstanding their similar compositions and structures. The relationship has been clarified through single-crystal X-ray diffraction studies on Mo(3)P (α-V(3)S-type, space group I42m, a = 9.7925(11) Å, c = 4.8246(6) Å) and (Mo(0.85)Fe(0.15))(3)P (Ni(3)P-type, space group I4, a = 9.6982(8) Å, c = 4.7590(4) Å) at -100 °C. Representation in terms of nets containing fused triangles provides a pathway to transform these closely related structures through twisting. Band structure calculations support the adoption of these structure types and the site preference of Fe atoms. Electrical resistivity measurements on (Mo(0.85)Fe(0.15))(3)P reveal metallic behavior but no superconducting transition.

Quaternary germanides RE4Mn2InGe4 (RE = La-Nd, Sm, Gd-Tm, Lu).

The quaternary germanides RE4Mn2InGe4 (RE = La-Nd, Sm, Gd-Tm, Lu) have been prepared by arc-melting reactions of the elements and annealing at 800 °C and represent the second example of the RE4M2InGe4 series previously known only for M = Ni. Single-crystal X-ray diffraction studies conducted on the earlier RE members of RE4Mn2InGe4 confirmed that they adopt the monoclinic Ho4Ni2InGe4-type structure [space group C2/m, a = 16.646(2)-15.9808(9) Å, b = 4.4190(6)-4.2363(2) Å, c = 7.4834(10)-7.1590(4) Å, β = 106.893(2)-106.304(1)° in the progression of RE from La to Gd]. The covalent framework contains Mn-centered tetrahedra and Ge2 dimers that build up [Mn2Ge4] layers, which are held weakly together by four-coordinate In atoms and outline tunnels filled by the RE atoms. This bonding picture is supported by band-structure calculations. An alternative description based on Ge-centered trigonal prisms reveals that RE4Mn2InGe4 is closely related to RE2InGe2. The electrical resistivity behavior of Pr4Mn2InGe4 is similar to that of Pr2InGe2.

The Ti-Fe-P system: phase equilibria and crystal structure of phases

Phase equilibria was investigated in the Ti-Fe-P system at T = 1070 K in the region 0–67 at.% of P, employing X-ray powder diffraction. The two ternary compounds, namely Ti0.5–0.8Fe1.5−1.2P (Co2Si-type; space group Pnma; a = 0.5964(2)–0.6011(3), b = 0.3575(3)–0.3600(1), c = 0.6828(2)–0.6882(2) nm) and Ti0.85−1.25Fe1.15−0.75P (ZrNiAl-type; space group P-62m; a = 0.6071(4)–0.6117(1), c = 0.3510(9)–0.3506(1) nm) exist in the Ti-Fe-P system at this temperature. The crystal structure of the Ti0.85–1.25Fe1.15−0.75P compound was additionally determined by X-ray single crystal diffraction on the phase with stoichiometric composition. The substitutions of Ti by Fe were observed for Ti5P3.16, Ti3P and TiP phases, and Fe for Ti in the case of Fe3P, Fe2P binary compounds.Graphical abstract

Searching for Missing Binary Equiatomic Phases: Complex Crystal Chemistry in the Hf−In System

There remain 21 systems (out of over 3500 possible combinations of the elements) in which the existence of the simple binary equiatomic phases AB has not been established experimentally. Among these, the presumed binary phase HfIn is predicted to adopt the tetragonal CuAu-type structure (space group P4/ mmm) by a recently developed machine-learning model and by structure optimization through global energy minimization. To test this prediction, the Hf-In system was investigated experimentally by reacting the elements in a 1:1 stoichiometry at 1070 K. Under the conditions investigated, the bulk and surface of the sample correspond to different crystalline phases but have nearly the same equiatomic composition, as revealed by energy-dispersive X-ray analysis. The structure of the bulk sample, which was solved from powder X-ray diffraction data through simulated annealing, corresponds to the γ-brass (Cu5Zn8) type (space group I4̅3 m) with Hf and In atoms disordered over four sites. The structure of crystals selected from the surface, which was solved using single-crystal X-ray diffraction data, corresponds to the CuPt7 type (space group Fm3̅ m) with Hf and In atoms partially disordered over three sites. The discrepancy between the predicted CuAu-type structure and the two experimentally refined crystal structures is reconciled through close inspection of structural relationships, which reveal that the γ-brass-type structure of the bulk HfIn phase is indeed derived through small distortions and defect formation within the CuAu-type structure.

How To Optimize Materials and Devices via Design of Experiments and Machine Learning: Demonstration Using Organic Photovoltaics

Most discoveries in materials science have been made empirically, typically through one-variable-at-a-time (Edisonian) experimentation. The characteristics of materials-based systems are, however, neither simple nor uncorrelated. In a device such as an organic photovoltaic, for example, the level of complexity is high due to the sheer number of components and processing conditions, and thus, changing one variable can have multiple unforeseen effects due to their interconnectivity. Design of Experiments (DoE) is ideally suited for such multivariable analyses: by planning ones experiments as per the principles of DoE, one can test and optimize several variables simultaneously, thus accelerating the process of discovery and optimization while saving time and precious laboratory resources. When combined with machine learning, the consideration of ones data in this manner provides a different perspective for optimization and discovery, akin to climbing out of a narrow valley of serial (one-variable-at-a-time) experimentation, to a mountain ridge with a 360° view in all directions.

Many metals make the cut: quaternary rare-earth germanides RE4M2InGe4 (M = Fe, Co, Ni, Ru, Rh, Ir) and RE4RhInGe4 derived from excision of slabs in RE2InGe2.

The formation of quaternary rare-earth (RE) germanides containing transition metals (Ms) from groups 6 to 10 was investigated through arc-melting and annealing reactions at 800 °C; about 50 new compounds were obtained. These include several new series of quaternary germanides RE4M2InGe4 (M = Fe, Co, Ru, Rh, Ir), previously known only for M = Mn and Ni; additional members of RE4Ni2InGe4 extended to other RE substituents; and a different but closely related series RE4RhInGe4. Detailed crystal structures were determined by single-crystal X-ray diffraction studies for 20 compounds. Monoclinic structures in space group C2/m are adopted by RE4M2InGe4 (Ho4Ni2InGe4-type, a = 15.1-16.5 Å, b = 4.1-4.4 Å, c = 6.9-7.3 Å, β = 106.2-108.6°) and RE4RhInGe4 (own type, a = 20.0-20.2 Å, b = 4.2-4.3 Å, c = 10.1-10.2 Å, β = 105.0-105.3°). Both structures contain frameworks built from MGe4 tetrahedra, InGe4 square planes, and Ge2 dimers, delimiting tunnels occupied by RE atoms. These structures can also be derived by cutting slabs along different directions from the more symmetrical RE2InGe2 structure. Although the Ge2 dimers are relatively invariant, the InGe4 square planes can undergo distortion to form two sets of short versus long In-Ge distances. This distortion results from a competition between M-Ge bonding in the MGe4 tetrahedra and In-Ge bonding in the InGe4 square planes.

Identifying an efficient, thermally robust inorganic phosphor host via machine learning

Rare-earth substituted inorganic phosphors are critical for solid state lighting. New phosphors are traditionally identified through chemical intuition or trial and error synthesis, inhibiting the discovery of potential high-performance materials. Here, we merge a support vector machine regression model to predict a phosphor host crystal structure’s Debye temperature, which is a proxy for photoluminescent quantum yield, with high-throughput density functional theory calculations to evaluate the band gap. This platform allows the identification of phosphors that may have otherwise been overlooked. Among the compounds with the highest Debye temperature and largest band gap, NaBaB9O15 shows outstanding potential. Following its synthesis and structural characterization, the structural rigidity is confirmed to stem from a unique corner sharing [B3O7]5– polyanionic backbone. Substituting this material with Eu2+ yields UV excitation bands and a narrow violet emission at 416 nm with a full-width at half-maximum of 34.5 nm. More importantly, NaBaB9O15:Eu2+ possesses a quantum yield of 95% and excellent thermal stability.Identifying phosphors with good thermal stability and quantum efficiency is a prerequisite to improve the performance of white LED light sources. Here, a combined machine learning and density functional theory method is introduced to identify next generation inorganic phosphors.

Machine learning directed search for ultraincompressible, superhard materials

In the pursuit of materials with exceptional mechanical properties, a machine-learning model is developed to direct the synthetic efforts toward compounds with high hardness by predicting the elastic moduli as a proxy. This approach screens 118 287 compounds compiled in crystal structure databases for the materials with the highest bulk and shear moduli determined by support vector machine regression. Following these models, a ternary rhenium tungsten carbide and a quaternary molybdenum tungsten borocarbide are selected and synthesized at ambient pressure. High-pressure diamond anvil cell measurements corroborate the machine-learning prediction of the bulk modulus with less than 10% error, as well as confirm the ultraincompressible nature of both compounds. Subsequent Vickers microhardness measurements reveal that each compound also has an extremely high hardness exceeding the superhard threshold of 40 GPa at low loads (0.49 N). These results show the effectiveness of materials development through state-of-the-art machine-learning techniques by identifying functional inorganic materials.

Polyanionic Gold–Tin Bonding and Crystal Structure Preference in REAu1.5Sn0.5 (RE = La, Ce, Pr, Nd)

During a systematic search of the RE-Au-Sn (RE = La, Ce, Pr, Nd) ternary phase space, a series of compounds with the general formula REAu1.5Sn0.5 have been identified. These phases can be synthesized by arc melting the elemental metals, followed by annealing. The crystal structures were solved using single-crystal X-ray diffraction, with the composition confirmed by energy-dispersive X-ray spectroscopy. All four compounds crystallize in orthorhombic space group Imma with the CeCu2-type structure. Most notable in these compounds is the polyanionic backbone composed of a single statistically mixed Au/Sn position, which creates a puckered hexagonal bonding network separated by the rare-earth atoms. Electronic structure calculations indicate that the Au 5d bands are dominant in the density of states, while the crystal orbital Hamilton population (-COHP) curves demonstrate Au-Au and Au-Sn interactions, which stabilize the crystal structure. Likewise, a qualitative electron localization function analysis supports the existence of a polyanionic network, and a Bader charge analysis implies anionic character on Au and Sn. The preference for these compounds to adopt the simple CeCu2-type structure is also determined using density functional theory calculations and compared to related compounds to establish a better picture of the unusual behavior of Au in polar intermetallic compounds.