Article Cite This: J. Am. Chem. Soc. 2017, 139, 15668-15680
pubs.acs.org/JACS
Exploration of Stable Strontium Phosphide-Based Electrides: Theoretical Structure Prediction and Experimental Validation Junjie Wang,†,‡ Kota Hanzawa,§ Hidenori Hiramatsu,†,§ Junghwan Kim,† Naoto Umezawa,‡ Koki Iwanaka,§ Tomofumi Tada,*,† and Hideo Hosono*,†,§ †
Materials Research Center for Element Strategy, Tokyo Institute of Technology, 4259 Nagatsuta-cho, Midori-ku, Yokohama, Kanagawa 226-8503, Japan ‡ International Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Sciences, Ibaraki 305-0044, Japan § Laboratory for Materials and Structures, Institute of Innovative Research, Tokyo Institute of Technology, Mailbox R3-4, 4259 Nagatsuta-cho, Midori-ku, Yokohama 226-8503, Japan S Supporting Information *
ABSTRACT: Inspired by the successful synthesis of alkaline-earth-metalsbased electrides [Ca24Al28O64]4+(e−)4 (C12A7:e−) and [Ca2N]+:e− and highthroughput database screening results, we explore the potential for new electrides to emerge in the Sr−P system through a research approach combining ab initio evolutionary structure searches and experimental validation. Through employing an extensive evolutionary structure search and firstprinciples calculations, we first predict the new structures of a series of strontium phosphides: Sr5P3, Sr8P5, Sr3P2 and Sr4P3. Of these structures, we identify Sr5P3 and Sr8P5 as being potential electrides with quasi-one-dimensional (1D) and zero-dimensional (0D) character, respectively. Following these theoretical results, we present the successful synthesis of the new compound Sr5P3 and the experimental confirmation of its structure. Although density functional calculations with the generalized gradient approximation predict Sr5P3 to be a metal, electrical conductivity measurement reveal semiconducting properties characterized by a distinct band gap, which indicates that the newly synthesized Sr5P3 is an ideal onedimensional electride with the half-filled band by unpaired electrons. In addition to presenting the novel electride Sr5P3, we discuss the implications of its semiconducting nature for 1D electrides in general and propose a mechanism for the formation of electrides with an orbital level diagram based on first-principles calculations. Because of its low work function6,7 and high electron conductivity,6 C12A7:e− has shown promise as a low electron-injection barrier for organic light-emitting diodes (OLEDs)8,9 and an efficient catalyst for ammonia synthesis.10−13 Compared with the 0D electrides, the 1D and 2D electrides have higher degrees of dimensionality and a higher density of anionic electrons in the lattice channels, and therefore show superior promise for the catalysis and electronic applications. Ca2N, along with Y2C, is a representative example of 2D electrides.16,17 Ca2N has been theoretically predicted to have intrinsic nucleus-free two-dimensional electron gas in the interlayer spaces between cationic layers with states at the Fermi level (even without electron doping),16,18 which has been confirmed by recent experiments.19 The anionic electrons of the 2D electrides are directly exposed to the environment, which in turn makes 2D electrides very active materials for catalysis such as NH3 synthesis16,17 However, the high activity of 2D electrides also comes with the disadvantage of lower stability. To improve
1. INTRODUCTION Electrides represent a distinct class of ionic compounds, in which electrons distributed in lattice cavities or channels can form individual orbitals and serve anions individually rather than being attached to atoms. In these unique structures, the loosely bound electrons at interstitial sites form partially occupied shallow bands (interstitial bands) leading to a dramatically reduced work function and excellent conductive properties. Electrides were first realized in organic materials; however, their poor stabilities at room temperature and ambient atmosphere limited their potential for use in applications.1−4 In recent years, inorganic electrides, which combine the unique characteristics of traditional organic electrides with much improved stabilities, have received growing attention from the theoretical viewpoint and as a source of novel materials concepts.5−24 The inorganic electrides discovered so far can be classified as 0D, 1D, and 2D according to the dimensionality of the distribution of the anionic electrons. 0D electrides are exemplified by C12A7:e− ([Ca24Al28O64]4+(e−)4), which was synthesized by Matsuishi et al. through removing two oxygen ions from the crystallographic cages of the parent compound [(12CaO·7Al2O3)2].6 © 2017 American Chemical Society
Received: June 16, 2017 Published: October 12, 2017 15668
DOI: 10.1021/jacs.7b06279 J. Am. Chem. Soc. 2017, 139, 15668−15680
Article
Journal of the American Chemical Society
Figure 1. Scheme presenting calculation-aided approaches to the discovery of new electrides used in previous studies and the present work. In present study, an extensive structure search and experimental validation, which are indicated in red and black in this scheme, have been carried out to search new electrides in Sr−P system.
the stability while achieving the superior properties of 2D electrides, Hosono and co-workers 14,15 have developed [La8Sr2(SiO4)6]4+:4e− and Y5Si3 as possible members of a 1D or quasi-1D electride family. The quasi-1D electride Y5Si3 indeed shows excellent stability in water while exhibiting catalytic activity for ammonia synthesis.15 Despite the extensive interest in high-dimensional (1D and 2D) electrides for applications in catalysis and electronic devices,14−17,23 the discovery of new members of these families is still rare and difficult. The strategies used previously in the search for new electrides are summarized in Figure 1. The black arrows indicate that experiments designed on the basis of information in databases comprise the main avenue by which novel inorganic electrides have been explored.11−18 To accelerate the discovery of new electrides, state-of-the-art numerical methods, e.g., ab initio high-throughput screening20,21 and evolutionary structure searches,25−29 have started to play an important role in these investigations. The blue arrows in Figure 1 show that high-throughput screening based on materials databases can be employed to choose promising candidates for experimental validation, allowing for more focused experimental efforts. For example, at time when Ca2N was the only experimentally realized 2D electride, Tada et al.20 suggested a series of possible 2D electrides, including Sr2N, Ba2N, and Y2C, through an extensive database screening of approximately 34000 materials in the Materials Project.30 In addition, Inoshita et al.21 proposed six carbides including Y2C as potential 2D electrides through searching two large-scale inorganic crystal structure databases: MatNavi31 and the Inorganic Crystal Structure Database (ICSD).32 The 2D electride Y2C was subsequently synthesized in accordance with those theoretical predictions.17 However, high-throughput screening based on available databases cannot access the unknown space of materials to directly identify “missing compounds” in systems with chemically reasonable combinations of atoms but no reported compounds. To address this drawback, the advanced structure-search methodology25−29 has begun to be employed in the search for the “missing” electrides beyond database. Liu and co-workers24
suggested that Li−C compounds could be 2D electrides using results from the CALYPSO algorithm.29 Hoffmann and co-workers23 have discovered 23 dynamically stable structures as candidates of new inorganic electrides by employing USPEX25−28 and particle swarm optimization29 algorithms. In a very recent investigation, Ma and co-workers developed a new algorithm for identifying electrides by using the electron localization function as a criterion in global structure searches with CALYPSO. Through this approach, they suggested 36 stable structures as potential electrides in binary A2B and AB systems, where A and B denote elements with relatively low and high electronegativities, respectively.33 This abundance of predictions highlights how one cannot bring the full potential of the structure search methods into play without the feedback information from the experiment and highthroughput screening. To improve the efficiency of this theoretical framework, we proposed a revised strategy, as indicated in Figure 1, through combining high-throughput screening (blue part) with the evolutionary structure searching (red part) and experiment validation (black part). By employing this strategy, one can explore the unknown materials space by high-throughput evolutionary structure searches after narrowing down the possible compositions based on achieved data from highthroughput screening and experimental validation. From our previous experimental16 and high-throughput database screening20,21 studies, we learned that metal-rich alkaline earth nitrides and phosphides can be good candidates for inorganic electrides. The alkaline earth metals Ca, Sr and Ba possess very low electronegativity (0.89−1.00), therefore, can have stronger ability to donate electrons to form electride structure than the other metals, e.g., Yttium (Y). Meanwhile, the N and P atoms possess electronegativity of 3.04 and 2.19. Comparing with those atoms of group IV (2.55 and 1.90 for C and Si) and group VI (3.44 and 2.58 for O and S), the moderate electronegativity of N and P would enable them to attract the electrons away from group II atoms while still allowing some electrons to form individual orbitals, which is crucial for the formation of electrides. 15669
DOI: 10.1021/jacs.7b06279 J. Am. Chem. Soc. 2017, 139, 15668−15680
Article
Journal of the American Chemical Society
softmutation 10%, permutation 10% and lattice mutation 10%. However, the types and fractions of variation operations were changed on the fly from the initial setting during the evolutionary structure search. The structure initiation and variation performed by USPEX is the global optimization step. The global and local optimizations were repeated until convergence was reached, as indicated by the best structure remaining unchanged over the course of 40 generations. To study the influence of external pressure on the stability of the SrxPy compounds, calculations were performed at pressures of 1 atm, 5 GPa, 10 GPa, and 15 GPa. In the local optimization step, the nonspin-polarized first-principles calculations were performed using the generalized gradient approximation in the Perdew−Burke−Ernzerhof (PBE)36 form as implemented in VASP. Because thousands of structures must be relaxed by VASP in the local optimization step, we used “good-enough” settings to reduce computational costs: the energy cutoff of 400 eV was set for the plane-wave basis set, and a Monkhorst−Pack k-point mesh with spacing of around 2π × 0.05 Å−1 was adopted for all structures. The PAW potentials Sr_sv_GW and P_GW were used for Sr and P in the calculations as recommended by the developers of VASP. The valence orbitals mainly constituted by 4d and 5s orbitals of Sr and the 3s and 3p orbitals of P are represented with plane waves, while lower-energy orbitals were frozen as core states. Finally, spin-polarized calculations were performed for the most stable group of the converged structures, whose energies are not higher by 200 meV/atom than the convex hull, to avoid the underestimation of the stabilities of the spin-polarized structures. More details of the general searching procedure can be found in the USPEX literature25−28 and our previous work.37−39 2.2. Structure Relaxation and Electronic Structure Calculations. We adopted the cutoff energy of 600 eV and k-point mesh resolution of around 2π × 0.02 Å−1 for the further relaxation of the most stable SrxPy structures. The lattice constants and atomic coordinates were fully relaxed using VASP until the force acting on each ion became less than 0.01 eV/Å at external pressures of 1 atm, 5 GPa, 10 GPa, and 15 GPa. Detailed enthalpy calculations at different pressures for the most stable structures allowed us to construct the convex-hull diagrams of enthalpy of formation as a function of the compositional ratio SrP/(Sr + SrP). Phonon dispersion calculations were performed to confirm the dynamic stabilities of all relaxed structures using Phonopy.40 The electronic structures of the most stable structures were calculated using the same settings as the structure rerelaxation step, except denser Monkhorst−Pack grids with a spacing of around 2π × 0.01 Å−1 was used in the density of states (DOS) calculations. For generation of projected DOS (PDOS) plots, the appropriate Wigner-Seitz radii were calculated as rs = (3V/4π)1/3, where V is the atomic volume determined for each atom by Bader charge41 calculations, which led us to adopt 1.79 and 2.15 Å as the radii for the Sr and P atoms, respectively. To obtain better band structures and gaps of Sr5P3, Sr8P5, Sr3P2 and Sr4P3, the hybrid density functional HSE0642 was used in the electronic structure calculations, in which the mixing parameter of 25% and screening parameter of 0.2 were used. The work functions (WF) of Sr5P3 were calculated as WF = Evac − EF, where Evac is the vacuum level and EF is the Fermi level. The vacuum levels were estimated using slab models with 64 atoms for [001], [100] and [101] orientations and 80 atoms for [110] orientation, and a vacuum region of 40 Å thickness was kept in the slab models. A 7 × 7 × 1 Monkhorst−Pack k-point setting was used for the slab models. The atomic positions in the slab models were fully relaxed with a constrained lattice volume. The vacuum levels were estimated as the energy levels at which the electrostatic potential became constant. Spin polarization was applied for all of the above structural relaxations and electronic structure calculations. 2.3. Experimental Validation. Polycrystalline SrxPy were synthesized by solid-state reactions of powdered phosphorus (99.9999%, Kojundo Chemical Laboratory Co., Ltd., Japan) and dendritic pieces of strontium (99.99%, Sigma-Aldrich Co., LLC., USA), whose average size is approximately 1−2 mm. The phosphorus powders and strontium pieces were loaded with the nominal chemical composition of Sr5P3.1 (with the excess P being due to high vapor pressure of phosphorus) into
Indeed, Hoffmann and co-workers recently suggested that the existence of NAn (N is nitrogen, A is alkali or alkaline earth metal, n = 6−9) polyhedra in crystal lattices is a crucial indicator of highdimensional electride character.23 The bigger the n number, the more metal atoms are incorporated in the crystal lattice as electron sources. We expect that phosphorus (P) has the potential of forming PAn polyhedra akin those to NAn. However, until now, there is no other metal-rich group II-group V compounds experimentally reported except Ca2N. Therefore, we believe there is a high possibility to find the “missing” electrides in the metal-rich II−P compounds. In a recent study, Ming et al. carried out a theoretical search using CALYPSO29 and considered an Sr2P structure with R3m ̅ symmetry as the most stable electride in Sr−P system.34 However, their structure search was done just for the formula Sr2P without considering the possibility of competing phases with other compositions, and their results have not been validated by experiment. In other words, the global thermodynamic stability of the proposed Sr2P phase has not been confirmed. In this article, we present an extensive structure search and validation study of the Sr−P system, beyond the existing materials databases, along the pathway shown in Figure 1. Our approach includes three steps: (1) extensive structure searches (with variable compositions) beyond the database, performed with the ab initio evolutionary algorithm USPEX;25−28 (2) structural screening with first-principles calculations, including structure reoptimization, thermodynamic and dynamic stability confirmation, and electronic structure analysis; and (3) experimental validation, carried out to confirm the theoretical prediction by materials synthesis and property measurement. Through this approach, we have theoretically identified that two promising electrides, Sr5P3 and Sr8P5, and two nonelectridesa narrow-gap semiconductor Sr3P2 and a p-type degenerate semiconductor Sr4P3are predicted to be obtainable experimentally by modifying the composition and synthesis pressure. Among of the searched structures, the 2D electride Sr2P suggested by Ming et al.34 was found as a thermodynamically metastable phase from ambient to 15 GPa, with the new quasi-1D electride Sr5P3 out-competing it. Our prediction of the Sr5P3 compound is validated by the successful synthesis and property measurements in the experimental part of this study. Finally, a mechanism for the formation of the electrides and subgap states is proposed to explain the electronic structures across the SrxPy system.
2. CALCULATION AND EXPERIMENTAL METHODS 2.1. Ab Initio Evolutionary Structure Search. Previous studies show that the electrides are always metal-rich compounds. Therefore, in the present search, Sr and SrP were adopted as reference components in the variable composition search of the SrxPy system. Any combinations of Sr and SrP in the unit cell were allowed (with the limitation of total number of atoms being ≤40). The search approach to find “missing” stable SrxPy phases involves two iterative steps: global optimization and local optimization, which are performed with the USPEX25−28 and VASP35 codes, respectively. The process begins with USPEX producing a first generation of structures using a random-number generator. Then local optimization of the obtained structures, i.e., the geometrical relaxation, is carried out by VASP for the comparison of the enthalpies of the candidate crystal structures. USPEX combines a proportion of the candidates with lowest enthalpies to new structures generated through variation operations based on the evolutionary algorithm (such as heredity, permutation, mutation, soft mutation...) to form the structures of the next generation. In the present study, the initial types and fractions of variation operations used to generate new structures are heredity 50%, random 20%, 15670
DOI: 10.1021/jacs.7b06279 J. Am. Chem. Soc. 2017, 139, 15668−15680
Article
Journal of the American Chemical Society
Figure 2. Results of an ab initio evolutionary structure search of the Sr−P system. (a) Convex hull diagram along the Sr−SrP axis at selected pressures of 1 atm, 5, 10, and 15 GPa. Solid squares represent stable compounds; open squares denote metastable compounds. The possible electrides Sr5P3 and Sr8P5 are labeled in red. Crystal structures are shown for (b) the quasi-hexagonal Sr5P3, (c) Sr8P5, (d) Sr3P2 (MP) and (e) Sr4P3. The space groups of Sr, Sr5P3 and SrP at different pressure are indicated in (a). Sr3P2 (MP) represents the Sr3P2 (I4̅2d) structure that documented by Materials Project. The structures of Sr and SrP documented in Materials Project were used as reference in this study. The space groups of Sr2P, Sr8P5 and Sr3P2 (MP) keep unchanged at different pressures. As shown in (a), no metastable SrxPy phase was found between the compositions of Sr and Sr2P. In this and following figures, green and gray balls represent Sr and P atoms, respectively. stainless steel tubes, and were tightly sealed under Ar atmosphere at ambient pressure. The samples were heated at 750 °C for 24 h. The obtained materials were grounded into fine powders for structural analysis and pressed in pellet shapes for electrical transport and electronic properties measurements. To synthesize Sr5P3H, which was prepared to clarify the influence of hydrogen in samples, the obtained powders were heated at 150 °C for 1.5 h in 0.5 MPa hydrogen atmosphere. To avoid the oxidation of strontium metal and the degradation of products, all synthesis and preparations for measurements were carried out in an Ar-filled glovebox (made by Miwa MFG Co., Ltd., Japan) with dew point of
