In situ studies of a platform for metastable inorganic crystal growth and materials discovery

Rapid shifts in the energy, technological, and environmental demands of materials science call for focused and efficient expansion of the library of functional inorganic compounds. To achieve the requisite efficiency, we need a materials discovery and optimization paradigm that can rapidly reveal all possible compounds for a given reaction and composition space. Here we provide such a paradigm via in situ X-ray diffraction measurements spanning solid, liquid flux, and recrystallization processes. We identify four new ternary sulfides from reactive salt fluxes in a matter of hours, simultaneously revealing routes for ex situ synthesis and crystal growth. Changing the flux chemistry, here accomplished by increasing sulfur content, permits comparison of the allowable crystalline building blocks in each reaction space. The speed and structural information inherent to this method of in situ synthesis provide an experimental complement to computational efforts to predict new compounds and uncover routes to targeted materials by design.Discovering new materials is a crucial step to address large-scale problems of energy conversion, storage, and transmission and other technological needs whether seeking bulk phases or thin films. Dense inorganic materials are desired for their tunable transport, magnetism, optical absorption, and stability, but their existence in general cannot be predicted with the near certainty of that of metastable organic and organometallic compounds. Whereas the desire to efficiently locate and assemble inorganic materials is great, it is hindered by traditional solid-state synthetic methods—at high temperatures often only the energy-minimum thermodynamic product is obtained. To strive toward an arena where metastable compounds can be discovered rapidly and made systematically, here we conduct reactions within liquid fluxes and use in situ monitoring to capture signatures of new phases, even when they quickly dissolve in the melt.Convective liquid fluxes (salts, metals, or oxides) can serve as reaction media that aid diffusion and enable rapid formation of compounds at temperatures far below their melting points (1–6). The flux can be nonreactive or reactive; in the latter case the flux itself becomes incorporated into the product (7, 8). This well-established approach has demonstrated the prolific discovery of novel inorganic materials grown out of low-melting fluxes, from oxides and other chalcogenides (9–12), to pnictides (13, 14), to intermetallics (15), many of which cannot be attained by direct combinations of the elements. Despite the variety of metastable phases formed in these reactions, the classical approach is to predetermine a given set of reaction conditions (e.g., time, temperature, and heating and cooling rates) and wait for completion to isolate and identify the formed compounds. It is not possible to observe how the reaction system itself has arrived at the isolated compound, whether the crystalline material formed on heating, on cooling, or on soaking at the given high temperature, nor it is possible to know whether any intermediates were present and, if so, their influence on product formation. This lack of awareness (“blind synthesis”) hinders our ability to identify the new materials or to devise successful synthetic processes for desired and targeted materials. If we are to develop a predictive understanding of synthesis and to more quickly discover new materials, we will greatly benefit from the input of much higher levels of detail in how syntheses proceed.We show here that in situ synchrotron X-ray diffraction maps of metastable inorganic compound formation in inorganic fluxes reveal complex real-time phase relationships and permit rapid access to new inorganic materials that would be missed using classical approaches. Specifically, we have discovered heretofore unknown phases in systems with simple elemental compositions of Cu and Sn with molten polysulfide salts K₂S₃ and K₂S₅ (melting points 302 °C and 206 °C, respectively) as paradigmatic representatives. Complex copper sulfides have been identified as possible earth-abundant photovoltaics (16) and are a source of exotic charge-density-wave materials, whereas tin chalcogenides form the basis for Cu₂ZnSnS₄ (CZTS) semiconductors (17) and exhibit ion exchange properties useful for heavy metal waste capture (18). We observe a complex phase space: In all but one of our reactions we observe additional crystalline phases in situ that are not present by the end of the reaction (as would be recovered ex situ). Both families of ternary compounds can exhibit a variety of coordinations by sulfur, and the range of properties is accordingly large: In just one of our Cu-containing reactions we found a previously undiscovered 1D metal (K₃Cu₄S₄) similar to heavy-metal capture materials and observed a different 2D metal (K₃Cu₈S₆) and a layered semiconductor (KCu₃S₂).Our approach can be combined with previous work that probes the formation and stabilization of inorganic materials from liquid media, using structural and spectroscopic data (19, 20). Within a given reaction, we can use temperature as a variable to probe the relationships among phases with the goal of understanding how the structures may be constructed from similar building units. Additionally, in situ studies of crystallization have been shown to be effective at probing the kinetics of oxide and sulfide formation, typically from aqueous solutions (21–23).The in situ technique we describe here provides rapid diffraction signatures of all crystalline phases formed during processing, including metastable intermediates. Reactions are performed in under 2 h (compared with typical flux reactions on the order of days), but time resolution of the data can be on the order of seconds. This is a route to efficient exploration of composition–temperature spaces that might be predicted to house novel compounds with favorable properties by first-principles (24) or data-mining calculations (25). Because new structure types cannot be easily predicted by computational theory, the approach described here is a necessary experimental tool in now-budding efforts to understand, validate, and expand the materials genome.A schematic of the in situ capillary furnace used to investigate phase formation during flux reactions is shown in Fig. 1A. The sample tubes, 0.7 mm in diameter, were sealed under vacuum, heated using a resistive coil, and continuously rastered through the synchrotron X-ray beam to maintain uninterrupted X-ray exposure of the sample as it melted and flowed within the tube.Open in a separate windowFig. 1.Schematic of in situ X-ray diffraction experiment and data analysis. (A) The in situ setup for collecting X-ray diffraction data from melting polysulfide fluxes. The sample tube diameter is 0.7 mm. Two-dimensional patterns are integrated and the 1D data can be fitted using Rietveld refinements to give individual contributions to the diffraction patterns in B. Upon heating, diffraction data are collected continuously, giving the set of patterns in C. These data are divided into regions with distinct phase contributions. A single pattern from each region is shown in D–F, with contributions to each pattern identified.For each reaction, the accompanying series of raw diffraction patterns provide a real-time monitor of the reaction progress and, to a first approximation, the number of phase formation and dissolution events. The X-ray diffraction pattern from a combination of Cu and K₂S₃ before heating is shown in Fig. 1B and fitted with a Rietveld refinement that confirms the contributions from both reagents. The diffraction patterns collected continuously while heating and cooling this reaction mixture are shown in Fig. 1C. Wholesale changes in the diffraction patterns indicate that the reaction pathway proceeded in a series of steps (color coded in Fig. 1C). First, signatures of the reagent metal and polysulfide appeared (blue region), which all resemble the refined pattern in Fig. 1B. Upon heating, low-Q peaks appeared in the diffraction data (red region). This real-time information (before any analysis) revealed that more complex ternary K–Cu–S phases were forming.Continued heating led to the disappearance of all Bragg peaks (gray region in Fig. 1C). At this point the ternary sulfides dissolved completely into the molten polysulfide salt flux. After cooling, low-Q peaks again signified the presence of ternary phases (violet region in Fig. 1C). Random variations in peak heights in adjacent patterns provided a morphological clue: The product on cooling formed as large sulfide crystals, leading to sharp diffraction spots rather than powder-averaged rings.Systematic least-squares refinements account for all crystalline phases present. Phase transformations and amorphous regions (defined by the absence of Bragg peaks) can also be observed in real time by visual inspection. Representative diffraction patterns and their least-squares refinements from each of the four temperature regimes in Fig. 1C are shown in Fig. 1 D–F. Data in Fig. 1D were taken at 320 °C upon heating and the reaction contained two ternary phases: KCu₃S₂ and the previously unknown phase K₃Cu₄S₄.At the time of the experiment, contributions to the diffraction pattern that do not match known compounds are considered unknown phases to be solved. Three options to identify new phases are as follows:

• i)Deduce structure by chemical analogy to known compounds. This can be accomplished manually or aided by computational tools such as the “Structure Predictor” of the Materials Project (26, 27).
• ii)Grow larger single crystals of compounds that crystallize out of a melt. In the present case, elemental analysis and single-crystal diffraction were used in this study to identify K₄Sn₂S₆ from crystals grown ex situ based on information provided by the in situ experiments.
• iii)Solve structure from powder diffraction, using computational packages such as FOX (28).

In the case of the experiment with data shown in Fig. 1D, the unknown phase K₃Cu₄S₄ was identified by option i: comparison of the isostructural phase Na₃Cu₄S₄.The diffraction pattern at 600 °C in Fig. 1E, taken from the gray region in Fig. 1B, has no Bragg peaks and represents the amorphous polysulfide melt after dissolution of KCu₃S₂ and K₃Cu₄S₄. A diffraction pattern from cooling at 50 °C is shown in Fig. 1F to contain the 2D metallic compound K₃Cu₈S₆ and the recrystallized flux K₂S₃. Because K₃Cu₈S₆ grew as large crystals, not a powder, Le Bail full-profile fitting (disregarding peak heights but refining peak profiles and peak locations constrained by unit cell symmetry) is shown in Fig. 1F to account for all peak contributions rather than the Rietveld method (29).The last step in diffraction analysis is to refine every powder pattern sequentially. Using the four diffraction patterns in Fig. 1 D–F and their refined phase contributions as anchors, we performed automated least-squares refinements to the ∼200 diffraction patterns to record the reaction progression as a function of time and temperature and form a “reaction map.” In this manner we identified phase fractions and points of crystallization, melting, and dissolution of all crystalline phases present during in situ reactions of Cu and Sn in fluxes of K₂S₃ and K₂S₅. The resulting panoramic reaction maps are shown in Fig. 2 and discussed subsequently.Open in a separate windowFig. 2.Reaction maps obtained from sequential refinements to in situ diffraction data. (A) Reactions of Cu + K₂S₃ lead to formation of KCu₃S₂ and the new phase K₃Cu₄S₄ on heating. These dissolve into the melt upon heating, and K₃Cu₈S₆ crystallizes upon cooling. A similar phase progression occurs for Cu + 5K₂S₃ (B), with KCu₃S₂ also forming on cooling. Reactions of Cu + K₂S₅ (C) and Cu + 5K₂S₅ (D) produce the layered phase KCu₄S₃. Reactions of Sn + K₂S₃ (E) produce SnS and the new phase K₆Sn₂S₇ on heating. No ternary phases are observed for Sn + 5K₂S₃ (F). The lowest-melting reaction Sn + K₂S₅ (G) reveals formation of the new phase K₅Sn₂S₈ from a melt upon heating. K₄Sn₂S₆ and K₂Sn₂S₅ crystallize concurrently upon cooling. For Sn + 5K₂S₅ (H), the progression is similar but K₂Sn₂S₅ is absent.