Efficient Exploratory Synthesis of Quaternary Cesium Chlorides Guided by In Silico Predictions

Abstract

Exploratory synthesis of solids is essential for the advancement of materials science but is also highly time- and resource-intensive. Here, we demonstrate an efficient strategy to explore solid-state synthesis of quaternary cesium chlorides in the search space of Cs_nA^IBCl₆ (n = 2 or 3, A = Li, Na or K, and B = d or p-block metal), where the target compositions are selected from a pool of candidates based on computationally predicted stabilities and availability of viable precursor powders. Synthesizability of the targets is assessed by observing the evolution of starting phases upon heating under in situ synchrotron X-ray diffraction. Laboratory synthesis is attempted for promising targets, and resulting materials are characterized by powder X-ray and neutron diffraction and subsequent Rietveld refinement. We focus on how computational predictions can be bridged to experimental characterizations in exploratory synthesis and report on successful and failed synthesis attempts for compounds of type Cs₂A^IB^IIICl₆, revealing underexplored variants including new polymorphs of Cs₂LiCrCl₆ and Cs₂LiRuCl₆, and a new compound Cs₂LiIrCl₆.

Introduction

Discovery of functional materials that lead to transformative technologies, from superconductors to secondary batteries, builds on the diversity of compounds accumulating from exploratory syntheses.¹⁻³ Automation of synthesis and characterization⁴⁻⁷ and revealing of reaction mechanisms in situ with X-ray diffraction (XRD) are increasing the pace of such explorations for new materials.⁸⁻¹² Recent large-scale, artificial intelligence (AI)-accelerated density functional theory (DFT) searches indicate that thousands of low-energy compounds could be awaiting synthesis.¹³⁻²⁰ These computational predictions provide an opportunity to rationally prioritize targets in silico prior to synthesis experiments and, in turn, more efficiently bridge experimental knowledge gaps that may exist even in broadly studied chemistries. For example, countering a large anion such as Cl^– with a similarly sized alkali cation like Cs⁺ offers a flexible close-packed framework composed of Cs-Cl₁₂ units with other cations occupying octahedral voids to form a wide array of ternary and quaternary compounds such as CsM²⁺Cl₃, Cs₂NaYCl₆ or other chloro-elpasolite or perovskite-derived structures in similar chemistries.²¹⁻²⁶ Such chlorides are attractive for many applications from optoelectronics to energy storage,^27,28 hence hundreds of such compounds have been studied over many decades.²¹ Still, a preliminary analysis of DFT-based structure predictions (Table 1) and cross-comparison with the solved experimental structures available in the Inorganic Crystal Structure Database (ICSD)²⁹ indicate numerous relatively stable quaternary Cs-chloride targets with stoichiometries and structures analogous to well-known families are either unexplored (absent from the ICSD and no obvious mentions in the literature) or underexplored (mentioned in the literature but either missing key structure information and/or not deposited to ICSD), revealing gaps in syntheses and accessible experimental structures despite the extent of the prior work.

Table 1. Quaternary Cs-Cl Targets Explored in the Present Study^a.

computational predictions			prior knowledge		synthesis				characterization
formula	predicted SG	Ed (eV/atom)	ICSD	literature	powder avail.	in-situ XRD	lab synthesis attempted ∼ 350 °C ambient	lab synthesis attempted 500 °C 5 GPa	powder XRD	powder ND
Cs2LiRhCl6	Pm1	–0.065	N	N	Y		Y		fail
Cs2LiIrCl6	Pm1b	–0.026	N	N	Y	Y (>350 °C)	Y		P63/mmc
Cs3KCdCl6	Rc	–0.024	N	N	Y	Y (liq.>430 °C)		Y	fail
Cs2NaDyCl6	Fmm	–0.019	N	Fmm(21,36)	Y	Y (liq.>500 °C)	Y		Fmm
Cs3NaCdCl6	Rc	–0.018	N	N	Y	Y (liq.>490 °C)			fail
Cs2LiCrCl6	Pm1b	–0.015	N	Pm135	Y	Y (>300 °C)	Y		P6322	P6322
Cs2KDyCl6	C2/m	–0.015	N	P21	Y	Y (unknown)		Y	fail
Cs3KSnCl6	Rc	–0.011	N	N	Y	Y (liq.>350 °C)		Y	fail
CsK2TlCl6	C2/c	–0.010	N	N	N
Cs2LiTiCl6	Pm1	–0.009	N	N	N
Cs2LiVCl6	Pm1	–0.004	N	Pm135	Y		Y		w/ impurity
Cs2LiRuCl6	Pm1b	0.005	N	P632237	Y		Y		P63/mmc
Cs2LiFeCl6	Pm1	0.019	N	N	Y		Y		w/ impurity

^aSpace groups (SG) shown are for the predicted structures. E_d is the decomposition energy with respect to the Materials Project (MP)¹⁹ convex hull, hence is a measure of the energetic tendency of formation of listed compounds wrt. phases in MP. Powder avail. is the commercial availability of starting binary chloride powders. XRD: X-ray diffraction, ND: neutron diffraction, Y: Yes, N: No.

^bThe space group of the corresponding Li/M-site mixing with half occupancy experimental structures can be assigned to P6₃/mmc. See the text for discussion of partial ordering.

Here, we explore the efficacy of computation-assisted exploratory synthesis in the space of quaternary cesium chlorides, where target compositions are prioritized by the zero-temperature DFT stabilities, assessments of experimental reports, and commercial availability of precursor powders. This is followed by high-temperature in situ synchrotron XRD to observe the reactions and seek formation of the target and attempts of laboratory synthesis and structural characterization by powder XRD, neutron diffraction (ND), and Rietveld refinement for a promising subset. We present both successful and failed attempts to synthesize the selected Cs_nABCl₆ targets and discuss the current and future role of AI, computation, and in situ experiments in improving exploratory synthesis in light of our results.

Results and Discussion

To find plausible Cs_nABCl₆ (n = 2 or 3, A = Li, Na or K, and B = d or p-block metal) compounds that may have been missed in prior experimental studies of quaternary cesium chlorides, we first search for Cs_nABCl₆ structures that are computationally stable or close to stable in the combined zero-temperature convex hull of crystals from Materials Project, OQMD, and GNoME databases^13,18,19 (hence higher likelihood of experimental realization³⁰). In this work, we selected the compounds that did not have a matching ICSD²⁹ entry (by composition) at the time of the work as targets. Absence in the ICSD implies the structure may not have been available to the computational community for further exploration, and identifying compounds not in the ICSD is commonly considered a benchmark for experimental novelty.^{11,17,29,31−34}

Nevertheless, literature has gaps in digitization and is ever-evolving, and there could be compounds not registered in the ICSD; therefore, we further perform a manual search to see if or to what extent the targets were reported experimentally outside the ICSD. We found mentions of certain compositions such as Cs₂NaDyCl₆ (as an elpasolite) and Cs₂LiCrCl₆ (as a 2L-type perovskite) and their lattice parameters in reports from decades ago^21,35,36 but could not locate them in the recent literature. For the latter, in particular, we could not find any mentions after the original report. We found that Cs₂LiRuCl₆ single crystal with partially ordered cation arrangement was reported recently,³⁷ but an ICSD entry was not currently available to us. In light of these computational, ICSD, and literature analyses, we conclude that the targets listed in Table 1 are worth pursuing solid-state exploratory synthesis, as their addition to structural databases upon potential synthesis and characterization would be valuable for downstream functional materials discovery research, as synthetic discovery often precedes the latter.¹

Next, we prioritize compositions based on the commercial availability of their binary chloride powders, preferably in the target valence state of the metals. For instance, anhydrous TlCl₃ and TiCl₃ powders were not readily available to us; hence, the respective candidates were removed from consideration. Accordingly, in the first batch, we choose Cs₂LiIrCl₆, Cs₃KCdCl₆, Cs₃NaCdCl₆, Cs₂LiCrCl₆, Cs₂NaDyCl₆, Cs₂KDyCl₆, and Cs₃KSnCl₆ as initial targets for in situ synchrotron XRD to screen reaction temperature and synthesizability in a wide temperature range, which reduces the time and effort required for initial exploration during synthesis significantly. Figure 1 shows the in situ synchrotron XRD results for six of the initial targets, where the mixture of starting binary chloride powders is heated at the stoichiometric ratio. Each measurement takes approximately 20 min, including heating, cooling, and changing of the sample in the quartz capillary by a robot. This process allows fast screening of the structural evolution and synthesis temperature at the synchrotron facility. Based on in situ XRD, Cs₂LiIrCl₆ and Cs₂LiCrCl₆ appear to crystallize into a structure similar to the hexagonal 2L-type perovskites.³⁸ Interestingly, Cs₂LiCrCl₆ shows a weak peak around 3.8° above ∼ 400 °C and Cs₂LiIrCl₆ shows a broad peak at a similar angle above ∼ 380 °C, which may have relevance for Cr and Ir ordering in these systems as described later. The XRD pattern for Cs₂NaDyCl₆ matches the cubic elpasolite structure (Figure S1) reported previously.²¹ Precursor mixtures targeting Cs₃KCdCl₆, Cs₃NaCdCl₆, and Cs₃KSnCl₆ show phase transformations, including melting (appearing as diffuse peaks), but the final crystalline phase(s) could not be identified. Synthesis attempts at a high pressure (5 GPa) to suppress melting and stabilize these three solids were not successful (Figure S2). Hence, realization of Cs₃ABCl₆ (A = Li, Na or K) type compounds, for which we could not find any experimental reports on, remains an open challenge.

Figure 1.

In-situ hot-stage synchrotron XRD heating profiles of the mixtures of binary starting chloride powders. The heating rate is 60 K/min. Labels next to the XRD profiles highlight approximate ranges of the presence of major peaks, where P is precursor powders, T is target, C is other crystalline phase(s) forming, and L is the liquid phase.

The in situ XRD realization of Cs₂LiCrCl₆ and Cs₂LiIrCl₆ motivated us to attempt laboratory synthesis of not only these phases but also reasonably stable variants with other d-block cations as shown in Table 1, namely Cs₂LiVCl₆, Cs₂LiFeCl₆, Cs₂LiRhCl₆, and Cs₂LiRuCl₆. For laboratory synthesis, we target 350–450 °C, which is the temperature range in which Cs₂LiCrCl₆ and Cs₂LiIrCl₆ start to form during the synchrotron XRD experiments (Figure 1). Among the mentioned compositions, laboratory synthesis was achieved for Cs₂LiM^IIICl₆ with M = Cr, Ir, and Ru and confirmed with powder XRD and ND (Figure 2a). The details of Rietveld refinement are summarized in Table 2 and the crystallographic information files (CIFs). The Cs:Cr:Cl, Cs:Ir:Cl, and Cs:Ru:Cl ratios of resulting phases are semiquantitatively confirmed to be close to the ideal values of the stoichiometric Cs₂LiM^IIICl₆ (Table S1). For M = Ru, 10% LiCl excess was used to suppress Cs₂RuCl₆ and isolate Cs₂LiRuCl₆ as the main phase (Figure S3), which is a strategy to drive certain reactions forward.³⁹ Despite M = Cr, Fe, and V forming similar hexagonal CsMCl₃ perovskites, we were able to isolate only the M = Cr variant in the quaternary form. For M = Fe, the quaternary phase can be present with impurity phases. For M = V and Rh, structural characterization was difficult as the target phase was not formed as a major phase (Figure S4).

Figure 2.

(a) Rietveld refinement of XRD and neutron diffraction (ND) profiles of quaternary Cs₂LiMCl₆ chlorides synthesized at 350–400 °C: M = Cr, Ru, and Ir. Predicted (ordered), experimental disordered, and experimental partially disordered structures are shown in (b), (c), and (d), respectively. Cs, Li, and M atoms are shown as gray, blue, and red spheres, respectively. Partially ordered structure is displayed also along the <110> direction. (Li/M)-Cl octahedra are shown, but Cl atoms are not shown for simplicity. For Li/Cr, synchrotron XRD is shown. Structures in (c) and (d) were further corroborated in cluster expansion (CE) driven Monte Carlo simulations.

Table 2. Summary of the Rietveld Refinement Results of Cs₂LiCrCl₆, Cs₂LiIrCl₆, and Cs₂LiRuCl₆ Synthesized at 350-400 °C with the P6₃/mmc or P6₃22 Model^a.

formula	Cs2LiCrCl6b		Cs2LiIrCl6	Cs2LiRuCl6
radiation source	synchrotron X-ray	neutron	X-ray (CuKα)	X-ray (CuKα)
wavelength (Å)	0.354233	1.34171	1.5418	1.5418
Rwp (%)	5.27 [20.86]	6.38 [9.71]	12.9 [25.6]	11.7 [14.5]
S = Rwp/Re	3.61 [14.26]	1.87 [2.84]	2.68 [5.31]	2.35 [2.91]
crystal system	hexagonal		hexagonal	hexagonal
space group	P6322		P63/mmc	P63/mmc
Z	3		1	1
a (Å)	12.4403(2)		7.15306(10)	7.16299(8)
c (Å)	6.06772(12)		6.05619(10)	6.05324(11)
V (Å3)	813.24(3)		268.358(9)	268.972(8)
g (M3+)c	0.5/0.711(8)		0.491(3)	0.52(2)

^aThe numbers inside the brackets in R_wp, R_e, and S are those with the Pm1 ordered model.

^bCombined Rietveld refinement of the X-ray diffraction and the neutron diffraction.

^cRefined occupancy of M³⁺ with the restriction of g(Li) + g(M) = 1. Final refinements for CIF were performed with the fixed occupancy of 0.5, except for the Li/Cr case, where the occupancy in partially ordered chains was also refined.

By definition, DFT stability predictions are done using a zero-temperature fully-ordered guess for the structure. For instance, for M = Cr, Ru, and Ir, the respective structure has the space group of Pm1 (Figure 2b). Unless treated separately (external to DFT),⁴⁰ finite temperature (or entropic) effects, such as site mixing, are not captured. In the mentioned cases, site mixing on Li and M sites with half occupancies would alter the space group from Pm1 to P6₃/mmc and partial mixing can lead to an intermediate P6₃22 superlattice structure with (Z = 3) unit cell³⁷ (Figure 2c,d). While the XRD patterns are close to P6₃/mmc with half occupation of Li and M cations, for M = Cr, the small peak at Q ∼ 1.16 Å^–1 supports the presence of the P6₃22 superlattice, where the Rietveld refinement indicates 1/3 of Li/Cr chains remain disordered and 2/3 remain partially ordered with alternating Li and Cr-rich sites (Table 2 and Figure S5). This superstructure is different from the previous report with fully ordered Pm1 form (Table 1), where only the lattice parameter was reported.³⁵ In contrast, we found Cs₂LiRuCl₆ crystallizes into a site mixing Li/Ru variant with the space group of P6₃/mmc (without a clear superlattice peak), suggesting no long-range ordering of Li/Ru, which is different from the partially ordered Cs₂LiRuCl₆ reported before (Figure S3).³⁷ Due to the higher atomic numbers of Ru and Ir, Li/Ru and Li/Ir orderings are more detectable under XRD compared to Li/Cr. In the case of Cs₂LiIrCl_6, which is reported for the first time here, a broad hump is observed near the superlattice peak at Q ∼ 1.16 Å^–1 as shown in Figures 1 and 2. This broad hump suggests no long-range ordering of the Ir site, but its short-range ordering close to the P6₃22 superlattice structure is likely. TEM electron diffraction images of Cs₂LiCrCl₆ and Cs₂LiIrCl₆ (Figure S6) show the corresponding superlattice 10 spots along the diffuse streak along <110>. Therefore, a short-range ordered superlattice structure is likely for both, but for the latter, given also the XRD data with the absence of long-range order of Li/Ir, the average structure can be represented as a simple disordered structure with the space group of P6₃/mmc.

The 2L-type quaternary Cs₂LiM^IIICl₆ (Pm1) can be viewed as an ordered variant of the known hexagonal ternary perovskites CsM^IICl₃ (M = Cr, Fe, and V), where the M site is occupied by an alternating pattern of Li and M. While the electrostatic interactions should favor Li/M cation ordering due to the valence difference, varying degrees of site mixing between different cations on the B-sublattices of double perovskites (mostly in oxides,⁴¹ but also halides^42,43) or Li/metal sublattices in other compounds⁴⁴⁻⁴⁷ are reported even for dissimilar cation oxidation states, hence some degree of (full or partial) mixing of Li⁺ and similarly sized d-metal M³⁺ cations here in Cs₂LiM^IIICl₆ at finite temperatures is not unexpected. In cases where there is negligible Li⁺/M³⁺ disordering in experiments, cluster expansion (CE) and Monte Carlo (MC) simulations were shown to correctly estimate the Li⁺/M³⁺ disordering temperature to be too high to be relevant in practice.⁴⁸ Using similar CE-driven MC simulations, we estimate a notable degree of Li⁺/Cr³⁺ mixing to be possible in temperature ranges comparable to that accessed in our experiments (Figure S7), indicating that the extent of order/disorder for Li⁺/Cr³⁺ (or the similar Li⁺/M³⁺ pairs here) would strongly depend on the experimental conditions and thermal history. The CE-driven MC predicted stable configuration at 300 K is similar to Watanabe et al.’s (Z = 3) unit cell,³⁷ but has all chains ordered, yielding a space group of again Pm1. We find that at higher temperatures where there is partial long-range order (e.g., 1000 K), this structure partially disorders along (011) in a planar stacking similar to Watanabe et al.’s P6₃22, where chains that extend along the c-direction across two adjacent (011) planes remain mostly ordered and shifted by c/2 with respect to each other, and the third (011) plane in sequence loses the long-range order while maintaining short-range order. At more elevated temperatures (e.g., 1500 K), the Li/Cr long-range order mostly vanishes, and most chains/planes have antisite defects that reduce the structure on average to P6₃/mmc. These findings reconcile our results with the results in ref (37) and show that the low to moderately high-temperature sequence for these materials likely follows Pm1 (Z = 3 phase) → P6₃22 →P6₃/mmc. Hence, a spectrum from partial or full disordering of Li and M may exist in the present samples depending on the thermal history.

The present results summarized in Table 1 show that DFT-predicted stabilities can play a role in prioritizing targets for efficiently finding synthesizable compounds in experiments, where four out of ten laboratory-synthesis attempts yielded a phase close to what was targeted. There is no question that experts can use the intuition and geometrical heuristics such as tolerance factors^49,50 to arrive at most of the formulations in Table 1, but the present results indicate that AI and computations are becoming effective in reliably finding such gaps in combinatorially large chemical spaces that are otherwise impractical to navigate experimentally in a single study, and in turn, making exploratory synthesis research more productive. More concretely, there are ∼280 charge-neutral Cs_nABCl₆ (n = 2 or 3, A = Li, Na or K), ∼20 of which were in the ICSD at the time of writing, hence leaving a large space to manually navigate. On the other hand, of the ∼120 such charge-neutral candidates in our database of low-energy compounds with no match in the ICSD, 25 were on the convex hull, indicating higher synthesis likelihood. Further prioritization with decomposition energies yielded our initial short-list (Table 1), increasing the efficiency of exploration by roughly an order of magnitude. Besides, due to its dependence on the knowledge of other materials in the chemical space, stability is a highly informative metric that is difficult to match with intuition or heuristics. Stability assessment naturally improves with more data, for instance, our convex hull combines GNoME¹³ (the largest database of known stable materials) with other DFT databases,^13,18,19 and therefore likely provides the most comprehensive stability assessment possible today. In our analysis above, using the MP hull alone would predict 40% more candidates as stable, adding more potential false positives to the short-list. We observe that using a more comprehensive database in computational guidance also enabled us to get the initial crystal structures accurately enough for downstream refinement in experiments. Lastly, computational guidance helps reduce the bias and improves confidence toward less usual targets, e.g., Cs₂LiIrCl₆ is less likely to be attempted in a more traditional exploratory synthesis setting considering the cost of Ir precursors and limited examples of (qua)ternary Ir halides.

Both computational and experimental steps presented here can be improved in the future. Automation of expensive computational treatments of temperature (e.g., site-mixing, vibrations, melting, etc.)^11,31,40,51 or pressure effects,⁵² predictive methods for precursor selection,⁵³⁻⁵⁶ or determination of excess precursor requirements,^39,57 language model assisted extraction of information buried in the chemistry literature,⁵⁸⁻⁶⁰ and utilization of a broader set of experimental databases^61,62 can all provide significantly better guidance for synthesis and reduce time-consuming trial/error in synthesis (or single crystal growth) or expensive characterization (e.g., ND or single-crystal XRD for resolving occupancy of light elements). Lack of analogous compositions for a target in experimental databases (as in Cs₃ABCl₆) or ease of access to precursors (inferable from reaction data sets⁵⁸) could be indicators of challenging synthesis and used to augment the computation-based rankings. The present in situ XRD experiments are geared toward rapid screening, but a robust solution of diffraction patterns, combination with pair functional distribution analysis, longer heating durations, and ability to robustly control chemical potentials, pressure or atmosphere would produce fewer false negative or false positive results.^31,63−67 Finally, we emphasize that findability and accessibility of structural data play a crucial role in AI-accelerated exploratory synthesis.

Conclusions

In this work, we explored the gaps of experimental synthesis and characterization in the model space of Cs-based quaternary chlorides by prioritizing targets using DFT-based stability predictions, assessing synthesizability by tracking reactions in situ with synchrotron XRD, and subsequently performing laboratory synthesis of the products. We found that even in broadly studied chemistries like Cs-based chlorides, the compositional guidance provided by computational stabilities from large-scale DFT databases makes the synthesis attempts more efficient. The computationally predicted ordered structures for Cs₂LiMCl₆ (M: Cr, Ru, and Ir) served as a starting point for refinement and subsequent proposal of Li/M mixing in experimental structures due to entropic effects. Detailed diffraction experiments (X-ray and neutron) and subsequent careful refinements revealed that the extent of site mixing was different from prior reports for M = Cr and Ru, highlighting the sensitivity of ordering of monovalent and trivalent cations to the thermal history in the synthesized structures. Finally, the crystal structure data for previously unexplored (e.g., Cs₂LiIrCl₆) or underexplored Cs-chlorides synthesized in this work are made digitally available for downstream functional discovery searches by deposition to structure databases.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacs.4c10294.

• Experimental and computational methods, EDX measurements, supplementary XRD patterns, TEM images, Monte Carlo simulation results, and experimental setup at synchrotron and supplementary references (PDF)

Author Contributions

^∇ A.M. and M.A. contributed equally to this work.

The authors declare no competing financial interest.