Ambient pressure recovery of the structurally unconventional hydride Y3Fe4H20
Authors: Maélie Caussé, Loïc Toraille, Grégory Geneste, Paul Loubeyre
Published: 2026

ABSTRACT
Reaching pressures in the 100 GPa range enables the synthesis of hydrogen-rich compounds with non-traditional hydrogen stoichiometries and hydrogen sublattices. The remarkable properties of such polyhydrides, including superconductivity, have attracted great interest. A crucial next step is their recovery under ambient conditions. Here we report the synthesis of Y3Fe4H20 at pressures starting from 60 GPa by compressing a hydrogenated Y-Fe compound, embedded in hydrogen in a laser-heated diamond anvil cell. Single-crystal X-ray diffraction resolves the Y3Fe4 lattice framework, while a constrained ab initio structural search is used to determine the hydrogen positions. FeH8 units form the building blocks and are connected edge-to-edge by sharing two hydrogen atoms, creating a framework that hosts Y cations. Y3Fe4H20 can be recovered at ambient conditions, where it stays metastable for tens of hours.

Introduction

The H atom bonding flexibility in the solid state leads to various types of hydrides at ambient pressure, classified as interstitial, ionic, complex, and covalent, with textbook examples being PdH, LiH, Mg2FeH6, and AlH3, respectively. Fifteen years ago, a new chemistry of hydrogen with metals was projected to take place in the 100 GPa range1 which has launched an extensive exploration for H-rich compounds formed by combining H with another element of the periodic table. Many hydrogen-rich compounds have been predicted to be stable and have been termed superhydrides (see reviews in refs. 2–4), although no universally accepted definition exists. A criterion based solely on the hydrogen-to-metal ratio would be too restrictive, particularly for ternary hydrides. Instead, superhydrides are more appropriately characterized by an unusual hydrogen sublattice. We therefore propose that superhydrides are defined by a dense hydrogen sublattice with nearest H–H distances shorter than 1.5 Å at around 100 GPa, excluding the presence of H2 molecular units (see the Supplementary Material). This criterion, based on the nearest H–H distance, is working well for all superhydrides published as such in the literature, and is conceptually analogous to the well-established Switendick rule, which imposes a lower bound of approximately 2.1 Å on H–H separations in interstitial hydrides at ambient pressure. FeH55 and LaH106 were the first superhydrides to be synthesized. In LaH10, the H atoms form an extended weakly bonded clathrate structure, with nearest H–H distances around 1.1 Å, whereas in FeH5, four plane slabs of atomic hydrogen are bonded to iron atoms. The various known functionalities of ambient pressure hydrides, such as superconductivity, hydrogen storage, and hydridic ionic conductivity, are magnified for superhydrides. This is exemplified in the case of LaH10, which holds a record superconductivity temperature of 262 K7,8 and exhibits a superionic hydridic state9. Unfortunately, none of these superhydrides have yet been stabilized under ambient conditions, preventing any potential societal applications.

The possibility of recovering these dense hydrogen arrangements, typical of superhydrides, in a compound under ambient conditions is an open question. Almost all possible binary superhydrides have been tried out, so the search is currently targeting ternary systems, which offer more structural prototypes and potential for stability at lower pressure. Computational ab initio random searches to systematically screen for stable ternary superhydrides are a formidable challenge. Various routes have been followed recently to reduce the phase space to be explored. One such way is to test for chemical doping by an electron donor element, as in the case of the prediction of Li2MgH1610. Another method is to look for substitutions of two elements in a given superhydride structure, as observed in (La,Y)H10 and (La,Ce)H911,12. Another approach is to assemble a combination of heavy X and light Y elements with disparate sizes, which makes a very effective chemical compression originating from the X-Y scaffolding. A ternary XYH8-type structure was predicted13,14 and recently synthesized in LaBeH8, which shows superconductivity above 100 K at sub-megabar pressures15. LaBeH8 can be viewed as a saltlike configuration of La2+ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{BeH}}}}_{8}^{2-}$$\end{document}BeH82−16, yet it is not stable at ambient pressure. Finally, two works have recently performed high-throughput computation for ternary hydrides, over many combinations in the periodic table, to explore the possibility of high-temperature ambient-pressure hydride superconductivity. Their most promising candidate was Mg2IrH6, with a critical temperature Tc of 80–160 K17,18. It is a complex hydride based on the [IrH6]3− cluster, not experimentally observed yet19. To keep this computation feasible, calculations were limited to high-symmetry crystal structures with fewer than 24 atoms per unit cell 18, which likely explains why no superhydride have been found under these constraints.

Here, we investigate the synthesis of ternary superhydrides in the Y–Fe–H system, without pre-guidance from calculations. We used the recently developed DAFi tool20 to perform single-crystal refinement of micron-sized crystal powder obtained after chemical reaction in a laser-heated diamond anvil cell (DAC). Some chemical ingredients have guided our selection of the Fe-Y-H system, in particular, the already observed high uptake of hydrogen under pressure by both Fe and Y elements and by their alloy, the YFe2 Laves phase. Fe and Y form binary superhydrides in the 100 GPa range, FeH521 and YH1022, with very different H sublattices: 2D atomic metal slabs and H clathrate, respectively. When compressing the stable compound YFe2H4.2 in hydrogen without heating, YFe2H7 was synthesized above 20 GPa, reaching then a limit of interstitial hydrogen uptake23. Finally, [FeH6]4− is a well-known anion which is a building block of ambient pressure complex hydrides. It has been recently shown that under application of hydrogen pressure in the 10 GPa range, a novel homoleptic pentagonal bipyramidal [FeH7]3− is formed, which can be the building block of novel complex transition metal hydrides24. Therefore, the question arises: why not stabilizing [FeH8]2− by going to higher pressure? Using a laser-heated DAC, we synthesized a ternary superhydride, Y3Fe4H20, at pressures as low as 60 GPa. Its structure reveals a novel hydride structure type, with chains of [FeH8] moieties encapsulating Y cations. Remarkably, this superhydride remains metastable at ambient conditions. Even though it is a metal, it unfortunately does not have superconducting properties.

Results

Y3Fe4 structure resolution by single-crystal diffraction

The high-pressure hydrogen environment on YFe2 was created by loading micron-sized samples of YFe2H4.2 in excess hydrogen into the sample chamber of DACs. Four DACs were loaded (run1 to run4), some with two samples, giving a total of six different syntheses which demonstrate excellent repeatability in the results. As shown in Fig. 1, the samples were laser-heated at different pressures, which enabled to overcome kinetic barriers in the transformation towards the most stable compounds in the Y–Fe–H system at a given pressure. Raman measurements performed before and after the laser heating yielded only the Raman signature of H2 in the chamber, with no Raman peaks on the sample. This indicates that there is no evidence for molecular H2 in the synthesized compound. The samples were characterized by single-crystal X-ray-diffraction (XRD) and angular-dispersive powder XRD at the ID15b and ID27 beamlines of the ESRF synchrotron. YFe2H7 was observed as the initial state of our high-pressure samples before the laser heating was performed, as expected based on previous work23. Under laser heating at 60 GPa (run2), 80 GPa (run3 and run4), and 83 GPa (run1), a clear sample recrystallization took place, as depicted in Fig. 1. The samples chemically reacted to form high-quality micron-sized crystals. This allowed us to perform a full single-crystal structure solving and refinement by combining the software CrysAlisPro25, DAFi20, and Jana200626. The refinement of the dominant phase yielded an orthorhombic unit cell Cmcm of 4 formula units, with the formula Y3Fe4 (a = 3.8312 Å, b = 9.9016 Å, c = 13.2019 Å and V = 500.8142 Å3 at 83 GPa) and the Wyckoff positions of yttrium and iron atoms are respectively 4c and 8f for Y, 4b, 8f and 4c for Fe (see the Supplementary Material). Due to the low scattering power of hydrogen atoms, it was not possible to experimentally determine the hydrogen positions in that structure, and consequently, the H content, so the complete formula of this new compound was an indeterminate Y3Fe4Hx at this stage. The DAFi analysis enabled us to identify multiple domains of this new compound with different orientations present in the sample, and X-ray mapping allowed us to reconstruct its volume extent. Minor phases were also present depending on the runs, including a Y2Fe3Hx phase, a FeHx phase, and other various YxFeyHz phases. The corresponding structures, solved by single crystal analysis, are given in the Supplementary Material. Since Y3Fe4Hx is the dominant phase and was subsequently recovered stable at ambient pressure, it is the focus for our more detailed characterization, as reported below.Fig. 1High-pressure Y–Fe–H transformation by laser heating.a Zoom on a XRD image plate of the precursor Y–Fe–H sample at 83 GPa before laser heating in run1. Continuous rings correspond to the diffraction of the gold liner of the sample chamber. The diffraction spots of the sample show a clear distortion of the original single crystal sample. b After laser heating, numerous diffraction spots are visible. Their respective identical 2θ values indicate the presence of multiple similar single crystals oriented in various directions. c X-ray diffraction pattern registered at 80 GPa (run3) after heating. The structure resolution through single crystal analysis leads to the indexation of all significant reflection peaks. Two different phases are observed, a dominant phase Y3Fe4Hx (red) and a minor phase Y2Fe3Hx (blue). d Top: optical image of the reacted sample in run3, about 15 μm in diameter. Bottom: corresponding XRD map analysis, selecting different 2θ values associated with Y3Fe4H20. The contour of the sample is outlined in red. The diffraction signature of Y3Fe4H20 is distributed almost everywhere in the sample, showing that the sample is quite homogeneous. e Pressure domain of the synthesis of Y3Fe4H20 under laser heating, different markers identify the different runs. When markers are filled (gray), the phase Y3Fe4H20 was observed after laser heating, whereas it was not observed when the markers were empty (white). The orange area indicates the stability domain, and the orange dashed-line shows the metastability region of the phase Y3Fe4H20 during decompression to ambient pressure.

High-pressure Y–Fe–H transformation by laser heating.

a Zoom on a XRD image plate of the precursor Y–Fe–H sample at 83 GPa before laser heating in run1. Continuous rings correspond to the diffraction of the gold liner of the sample chamber. The diffraction spots of the sample show a clear distortion of the original single crystal sample. b After laser heating, numerous diffraction spots are visible. Their respective identical 2θ values indicate the presence of multiple similar single crystals oriented in various directions. c X-ray diffraction pattern registered at 80 GPa (run3) after heating. The structure resolution through single crystal analysis leads to the indexation of all significant reflection peaks. Two different phases are observed, a dominant phase Y3Fe4Hx (red) and a minor phase Y2Fe3Hx (blue). d Top: optical image of the reacted sample in run3, about 15 μm in diameter. Bottom: corresponding XRD map analysis, selecting different 2θ values associated with Y3Fe4H20. The contour of the sample is outlined in red. The diffraction signature of Y3Fe4H20 is distributed almost everywhere in the sample, showing that the sample is quite homogeneous. e Pressure domain of the synthesis of Y3Fe4H20 under laser heating, different markers identify the different runs. When markers are filled (gray), the phase Y3Fe4H20 was observed after laser heating, whereas it was not observed when the markers were empty (white). The orange area indicates the stability domain, and the orange dashed-line shows the metastability region of the phase Y3Fe4H20 during decompression to ambient pressure.

Y3Fe4H20 structure resolution by constrained AIRSS

The number of H atoms and their positions in the Y3Fe4 lattice were obtained at 90 GPa using first-principles density-functional theory (DFT) calculations under a constrained random search based on the AIRSS algorithm27, as described in the methods section. Essentially, this involved fixing the experimentally resolved Cmcm space group, the atomic positions, and the lattice parameters of the Y3Fe4 skeleton (in the generation of the random structures, these positions and lattice parameters being subsequently structurally optimized). Using this Y3Fe4 skeleton effectively reduces the problem to that of finding a binary hydride, the alloy Y3Fe4 being the metal, thus simplifying the search for a ternary hydride. Furthermore, the measured volume of the unit cell, 500 Å3, should not be too different from the volume obtained by assuming additivity of the volume of the stable hydrides of the two end-member elements, YHx and FeHx, stable around the same pressure, hence a mixture of YH3 and FeH2, or YH3 and FeH3. Correspondingly, the random searches for hydrogen content were constrained to vary between 17 and 21 per formula unit. The formation enthalpies of the various stable compounds obtained for each stoichiometry were then plotted and compared to the convex-hull, which shows thermodynamically stable compounds regarding decomposition towards H2 and Y3Fe4H17, as shown in Fig. 2. Two stable compounds, Y3Fe4H18 and Y3Fe4H20, were good candidates. The hydrogen-excess experimental conditions ensure that the most stable compound to be synthesized should fall on the convex-hull with the highest H concentration, hence the selection of Y3Fe4H20. Additionally, we performed enthalpy calculations of all possible known simple compounds that Y3Fe4H20 could decompose into (Y, Fe, H2, YHx, FeHx, YFe2Hx), at pressures ranging from 90 to 0 GPa. The results, including a total of 48 possible reactions, are gathered in the Supplementary Material: for all pressures studied ≥15 GPa, the reactions of formation of Y3Fe4H20 from the simpler compounds listed above are all negative, indicating that Y3Fe4H20 is energetically more favorable with respect to its decomposition into simpler compounds. At 0 GPa, however, the calculations indicate that Y3Fe4H20 is unstable with respect to its decomposition for two cases: to form a mix of Fe, H2, and YH3, as well as a mix of FeH, H2, and YH3. This is expected since Fe, FeH, H2, and YH3 are all compounds observed to be stable at ambient (or near-ambient) pressure. Since Y3Fe4H18 is a stable compound and a potential candidate for the observed compound, we have also calculated the reaction of formation of Y3Fe4H20 from Y3Fe4H18 and H2 between 0 and 90 GPa (see the Supplementary Material). The corresponding reaction enthalpy is negative above 10 GPa, confirming that the formation of Y3Fe4H20 is more favorable at high pressure in excess of H2.Fig. 2Structure, thermodynamic, and dynamical stability of Y3Fe4H20.a Calculated convex hull at 90 GPa for the various Cmcm-Y3Fe4Hx compounds. The formation enthalpies are divided by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{n}}_{\rm{H}}+{\rm{n}}_{{\rm{Y}}_{3}}{\rm{Fe}}_{4}$$\end{document}nH+nY3Fe4H17. The phases that lie on the convex hull (in red) are thermodynamically stable. The phase Y3Fe4H20 is likely to be the experimentally synthesized phase since it was synthesized in excess of hydrogen. b Crystal structure of Y3Fe4H20 (primitive cell), with the Fe atoms in gold, the Y atoms in green, and the H atoms in light pink. Each Fe atom is surrounded by 8 H forming [FeH8] units. These units are connected edge-to-edge by two hydrogen atoms. c Results of the DFPT phonons calculations for Y3Fe4H20 at 90 GPa: phonon dispersion curves along high-symmetry directions in the Brillouin zone; the green, blue, and red density of states correspond to yttrium, iron, and hydrogen atoms, respectively. d GGA-PBE total and projected electronic densities of states of Y3Fe4H20 at 90 GPa. The projected 1s-H electrons do not contribute to the Fermi level. The projected Fe-3d electrons are the main contributors at the Fermi level.

Structure, thermodynamic, and dynamical stability of Y3Fe4H20.

a Calculated convex hull at 90 GPa for the various Cmcm-Y3Fe4Hx compounds. The formation enthalpies are divided by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{n}}_{\rm{H}}+{\rm{n}}_{{\rm{Y}}_{3}}{\rm{Fe}}_{4}$$\end{document}nH+nY3Fe4H17. The phases that lie on the convex hull (in red) are thermodynamically stable. The phase Y3Fe4H20 is likely to be the experimentally synthesized phase since it was synthesized in excess of hydrogen. b Crystal structure of Y3Fe4H20 (primitive cell), with the Fe atoms in gold, the Y atoms in green, and the H atoms in light pink. Each Fe atom is surrounded by 8 H forming [FeH8] units. These units are connected edge-to-edge by two hydrogen atoms. c Results of the DFPT phonons calculations for Y3Fe4H20 at 90 GPa: phonon dispersion curves along high-symmetry directions in the Brillouin zone; the green, blue, and red density of states correspond to yttrium, iron, and hydrogen atoms, respectively. d GGA-PBE total and projected electronic densities of states of Y3Fe4H20 at 90 GPa. The projected 1s-H electrons do not contribute to the Fermi level. The projected Fe-3d electrons are the main contributors at the Fermi level.

The dynamical stability of Y3Fe4H20 was verified at 90 GPa and 0 GPa by computing the phonon dispersion curves using the Density Functional Perturbation Theory (DFPT). Phonon calculations, including the phonon dispersion curves and corresponding vibrational densities of states, are shown in Fig. 2 and indicate that the Y3Fe4H20 structure is dynamically stable at 90 GPa, as evidenced by the absence of imaginary phonon modes. The phonon dispersion curves at 0 GPa, displayed in the Supplementary Material, indicate that the structure is dynamically stable, supporting the possibility for Y3Fe4H20 to be metastable at ambient pressure. Our DFPT analysis reveals a clear separation between the dynamics of the heavy Y and Fe atoms and that of the lighter H atoms (Fig. 2). Overlapping peaks in the vibrational density of states for the metallic atoms Y and Fe are observed. Additionally, the H-related phonon dispersion curves show a weak dispersion of the phonon modes, which is unusual. The electronic densities of states and electronic band structure of Y3Fe4H20 computed at 90 GPa are shown in Fig. 2 and in the Supplementary Material, respectively. From these, we can conclude that Y3Fe4H20 is metallic but unlikely to be superconducting, as it exhibits a very low H-projected electronic density of states at the Fermi level.

The crystal structure of Y3Fe4H20 is depicted in Fig. 2. It is remarkable to see that the structure is formed of [FeH8] entities, while not being a complex hydride. The [FeH8] entities are connected to form two-dimensional chains encapsulating the Y cation, which is a previously unobserved structure for an H compound.

Equation of state of Y3Fe4H20

Single-crystal analysis of the diffraction data revealed that the lowest pressure at which the Y3Fe4H20 structure emerges from laser-heating is 60 GPa. The equation of state of Y3Fe4H20 was measured by decreasing the pressure down to ambient pressure at room temperature (Fig. 3). Remarkably, Y3Fe4H20 did not decompose upon decompression. The phase remained orthorhombic with space group Cmcm from 90 to 0 GPa. The single-crystal diffraction data quality progressively decreased with decompression, as expected. In one case, a distortion of single crystal domains towards a lower-symmetry monoclinic phase occurred below 10 GPa, with a more apparent effect on the angle β and the lattice parameter c, as shown in Fig. 3 and Supplementary Material. We fitted the V(P) data points with a Vinet-type equation and obtained an ambient pressure volume V0 = 703.3 (7.7) Å3, a bulk modulus K0 = 93.0 (12.7) GPa, and its pressure derivative \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{0}^{{\prime} }$$\end{document}K0′= 5.8 (0.5) (numbers between parentheses represent 95% confidence intervals).Fig. 3Volume and lattice compression curve of Y3Fe4H20 between 0 and 90 GPa.a Experimental (in blue) and calculated (in red) equation of state of Y3Fe4H20. The curves are fitted with a Vinet equation of state. Data from both samples #S1 and #S2 from run1 (circle), obtained during decompression, are plotted along with samples #S3 and #S4 from run2 (square). b Evolution of the lattice parameters with pressure and comparison between experimental and calculated data. The square and circle symbols follow the same convention as (a). The dotted lines correspond to the lattice parameters calculated with the GGA-PBE functional.

Volume and lattice compression curve of Y3Fe4H20 between 0 and 90 GPa.

a Experimental (in blue) and calculated (in red) equation of state of Y3Fe4H20. The curves are fitted with a Vinet equation of state. Data from both samples #S1 and #S2 from run1 (circle), obtained during decompression, are plotted along with samples #S3 and #S4 from run2 (square). b Evolution of the lattice parameters with pressure and comparison between experimental and calculated data. The square and circle symbols follow the same convention as (a). The dotted lines correspond to the lattice parameters calculated with the GGA-PBE functional.

The compression curve and lattice parameters obtained from DFT structural optimizations for Y3Fe4H20 are in excellent agreement with our experimental data, as seen in Fig. 3. The relative error on the volume is 1.4%, corresponding to less than 0.5% on the lattice parameters. Since the structure is dynamically stable at 0 GPa, it seems likely that the experimentally observed distortion at 0 GPa arises from non-hydrostatic pressure conditions occurring during decompression of the DAC.

Structural analysis of the hydrogen arrangement

The electron density distribution (Fig. 4) and electron localization function (ELF, see the Supplementary Material) were calculated to infer the chemical bonding in Y3Fe4H20. They both indicate high electronic density on metallic atoms, especially suggesting an electronic localization around Fe. The isosurface of electronic density distribution at 0.1 is shown within the primitive cell in Fig. 4a. The Y atoms do not share much electron density with their neighbors. Most of the electron density is concentrated on Y and Fe (0.3). This suggests that Fe-H and Y-H are not covalently bonded but rather ionically bonded, with an attractive electrostatic interaction between metallic atoms and hydrogen, not visible on the plots. However, the hydrogen cages around the Y atoms differ from those in clathrate superhydrides, where the ELF values are very high. The Fe atoms are surrounded by 8 H atoms, forming [FeH8] complexes. In contrast to LaBeH8, which can be considered as a complex hydride since it is formed by a rocksalt structure of La2+ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{BeH}}}}_{8}^{2-}$$\end{document}BeH82− units, here the [FeH8] units are bonded together edge-to-edge by hydrogen atoms. These “chains” of [FeH8] are tied to Y by ionic bonding. Two types of hydrogen can be distinguished, non-bridging hydrogen that are bonded to a single Fe and pointing towards Y (Fig. 4b), indicating an ionic bonding between H and Y; and bridging hydrogen that are bonded with two Fe (Fe-H-Fe), and participate in forming the chains. The chains contain three distinct polyhedral of 8 H: a distorted cube, a bicapped trigonal prism, and a tetragonal antiprism. The chain motif is composed of three polyhedra between two distorted cubes, which is repeated in the lattice. This allows us to identify a new class of hydrogen-rich compounds: superhydrides with 2D-chains of polyhedral complexes and counterions.Fig. 4Electronic density distribution and chain structure of Y3Fe4H20.a Electronic density distribution for the orthorhombic primitive cell calculated at 90 GPa. The isovalue scale is plotted as a color scale. b Crystal structure of Y3Fe4H20, with the Fe atoms in blue, the Y atoms in orange, and the hydrogen atoms in red. [FeH8] units (blue polyhedron) are connected edge-to-edge by two bridging hydrogen atoms (black circles). Non-bridging hydrogen atoms (green circles) are pointing towards Y cations and do not participate in forming the chains.

Electronic density distribution and chain structure of Y3Fe4H20.

a Electronic density distribution for the orthorhombic primitive cell calculated at 90 GPa. The isovalue scale is plotted as a color scale. b Crystal structure of Y3Fe4H20, with the Fe atoms in blue, the Y atoms in orange, and the hydrogen atoms in red. [FeH8] units (blue polyhedron) are connected edge-to-edge by two bridging hydrogen atoms (black circles). Non-bridging hydrogen atoms (green circles) are pointing towards Y cations and do not participate in forming the chains.

Discussion

Y3Fe4H20 exhibits metastability at ambient pressure both in a hydrogen environment and in room air. To assess the compound’s metastability, we conducted a series of tests. Samples of run2 were observed through X-ray diffraction during decompression down to ambient pressure and for several hours at ambient conditions. Evidence of metastability was established after more than a day (~30 h), as shown in Fig. 5. Samples from run1 and run4 were retrieved from the DAC, retained in ambient air on a silicon wafer, and monitored through X-ray diffraction after a period of about 3 months and 1 month, respectively. In both cases, a volume reduction of about 7–8% was observed, which we attribute to hydrogen desorption, showing a limit of the metastability. This desorption is not surprising given the calculated relative stability of Y3Fe4H18 and Y3Fe4H20 at ambient pressure. The observed volume after desorption, ~650 Å3 is much lower than the one calculated for Y3Fe4H18 at ambient pressure, ~693 Å3, suggesting that more than 2 atoms of hydrogen were lost from the lattice cell. Notably, both orthorhombic and monoclinic crystallites are observed after 30 h at ambient pressure. The remarkable metastability of this superhydride at lower and ambient pressure is also confirmed by first-principles calculations. As stated earlier, the formation of Y3Fe4H20 is energetically favorable with respect to simpler compounds in all cases, except at 0 GPa in the case of certain specific reactions, which appears to be consistent with expectations (see the Supplementary Material).Fig. 5Metastability of the phase Y3Fe4H20: evolution of the experimental lattice volume at ambient pressure versus time.Both samples of run1 (in red) at ambient pressure at time zero contained only monoclinic crystallites; three and a half months after, only orthorhombic crystallites were observed with a 7% reduction of lattice volume. In run2, both monoclinic and orthorhombic crystallites were observed in the sample. No significant structure or lattice volume variation were observed during a 30-h time period. In run4, the structure after 1 month was found to be orthorhombic, with an 8% volume reduction. Each point corresponds to one measurement.

Metastability of the phase Y3Fe4H20: evolution of the experimental lattice volume at ambient pressure versus time.

Both samples of run1 (in red) at ambient pressure at time zero contained only monoclinic crystallites; three and a half months after, only orthorhombic crystallites were observed with a 7% reduction of lattice volume. In run2, both monoclinic and orthorhombic crystallites were observed in the sample. No significant structure or lattice volume variation were observed during a 30-h time period. In run4, the structure after 1 month was found to be orthorhombic, with an 8% volume reduction. Each point corresponds to one measurement.

We performed SQUID measurements on the sample synthesized in run4, following the magnetic moment of the sample with temperature. These measurements, included in the Supplementary Material, show no sign of superconductivity at any temperature down to 3 K. While the samples were small and not perfectly pure, this is further evidence that Y3Fe4H20 is unlikely to be superconducting.

The nearest H–H distance in this material is approximately 1.33 Å, which is similar to that observed in FeH55 at 90 GPa. We calculated the H–H distances in the Cmcm structure down to ambient pressure. The pressure evolution of the nearest H–H distance (shown in the Supplementary Material) follows a similar trend as in FeH5. The nearest neighbor superhydride criteria remain valid down to ambient pressure. However, unlike FeH5, this material remains metastable at ambient pressure. The remarkable metastability of the structure at ambient pressure can be attributed to the presence of the extended two-dimensional chains of [FeH8] units, which brings mechanical stability. This is reminiscent of the uranium superhydride UH7, characterized by its extended one-dimensional chains of cubic hydrogen units, and known for its exceptional stability, with the lowest experimentally observed pressure among known superhydrides—around 35 GPa28,29. The structural stability of the Y3Fe4H20 superhydride under ambient conditions for several hours is a clear indication that certain structural features of superhydrides, typically stabilized in the 100 GPa range, can persist in a metastable state at ambient pressure.

The XRD single-crystal structural refinement of the reaction products obtained in a laser-heated DAC is presented here as a promising approach for exploring ternary hydrides. This method was inspired by its success in the search for polynitrogen compounds, leading to the discovery of structures beyond the current capabilities of random ab initio searches30. The structure we predicted has a 54-atom primitive unit cell, meaning that finding Y3Fe4H20 through calculations alone would have required a computationally expensive, large-scale exploration with large unit cells. Simultaneously, high-pressure experimental XRD alone cannot provide the arrangement of the H-sublattice. Therefore, a dialog between experiment and computation is essential for the discovery and study of novel superhydrides. Within this combined experimental-theoretical framework, several assumptions are necessarily introduced, as is common in high-pressure studies of complex hydrides. Laser heating is used to overcome kinetic barriers and promote equilibration at least locally, while the thermodynamic analysis relies on convex hulls calculated at 0 K. The structural modeling neglects disorder and adopts an ordered description of the atomic sublattices, as refinements allowing for partial site occupancies do not provide unambiguous experimental evidence for intrinsic occupational disorder within the sensitivity of the present data. These approximations are experimentally constrained, internally consistent, and supported by the robustness of the observed phase under repeated laser heating as well as by its agreement with first-principles stability. As such, they provide a reasonable and reliable basis for the identification and analysis of Y3Fe4H20.

It has recently been noted that most prototype clathrate-like structures adopted by binary superhydrides can be found in zeolite databases31, as they share similar 3D pore frameworks. Following this logic, the results obtained here indicate that structures with similar chains, such as the aluminosilicates32–34, could also be used as a promising database for identifying potential ternary superhydrides. In a first step, the present structure can now serve as a template for optimizing compound properties through chemical substitution. As a starting example, substituting Fe by Ni results in a possible Y3Ni4H20 superhydride. DFT calculations on this compound show a slightly enhanced contribution of hydrogen-projected density of states at the Fermi level (see the Supplementary Material), making it a more promising candidate for superconductivity, although with a low likelihood of high Tc since the density of state at the Fermi level is still dominated by transition metal states.

Methods

Experimental details

The sample configuration and the synchrotron X-ray diffraction measurements were similar to those used in previous studies of compounds formed in the Fe-H5 or U-H29 systems at high pressures. Experiments were performed on 3–8 μm YFe2H4.2 crystalline samples35,36 loaded in DACs with hydrogen as pressure-transmitting medium (PTM). We used DACs with an effective angle opening of about 29∘. The YFe2H4.2 hydrogenated compound was chosen to perform hydrogenation under high-hydrogen pressure to prevent any hydrogen-induced amorphization or decomposition into YH3 and ϵ-Fe. Rhenium gaskets were used, protected with a 1.5–6 μm gold layer to prevent hydrogen diffusion. Pressure was measured using the equation of state of gold in the coating close to the sample in the experimental chamber of the DAC. The uncertainty in pressure using this gold pressure gauge is about 2%.

Four total runs of compression/decompression were performed on the sample in hydrogen PTM, designated as run1 to run4.

In run1, two samples #S1 and #S2 were loaded, both about 8 μm in diameter, in a DAC with 100 μm diamond culets. The DAC was compressed up to 83 GPa and offline laser heating was performed on both samples. After the synthesis, we followed its evolution with pressure by first increasing the pressure up to 90 GPa and then decreasing it down to 0 GPa, at ambient temperature. Both samples were followed in order to check the repeatability. One sample recovered at ambient pressure was kept for about 3 months to evaluate its long-term metastability.

In run2, two samples #S3 and #S4, about 6 μm and 10 μm in diameter, were loaded in a DAC with 300 μm diamond culets. The samples were compressed up to 60 GPa, accompanied by offline laser heating at three different pressures: 30, 45, and 60 GPa. The aim of this run was to determine the lowest synthesis pressure of the new compound, and then to evaluate the hydrogen desorption speed of Y3Fe4H20 at ambient conditions after pressure decompression. The lattice volume of the resulting samples after pressure decompression was monitored for 30 h without noticeable change.

In run3, a single bigger sample #S5 about 14 μm in diameter was loaded in a DAC with 100 μm diamond culets. Y3Fe4H20 was synthesized by laser heating at 80 GPa. XRD measurements, including mapping, were performed to check the composition and homogeneity of the synthesized sample. Further studies on the metastability and retrieval of the sample were prevented by the sudden closing of the gasket at 40 GPa during decompression.

In run4, a single big sample #S6 about 12 μm in diameter was loaded in a DAC with 150 μm diamond culets. Laser heating was performed at 80 GPa and led to the synthesis of a mix including Y3Fe4H20 and other Y–Fe–H compounds, as informed by XRD measurements. A progressive decompression was performed over the course of 19 h, then the DAC was opened. The sample was slightly crushed by the partial closing of the gasket and was retrieved in multiple pieces, which were used for SQUID measurements, as detailed in the Supplementary Material. Further XRD was performed a month after the SQUID measurements and included a signature of the Y3Fe4H20 compound in its lower volume, dehydrogenated shape.

Offline laser heating was performed with a ytterbium fiber laser with a 1070 nm wavelength. We maintained the heating temperature below 1500 K to prevent hydrogen diffusion in the diamond anvils. The XRD data were collected at the ESRF ID15B and ID27 beamlines on accepted proposals HC-5070 and HC-5453. XRD measurements were performed with a monochromatic wavelength of either 0.3738 Å or 0.411 Å and diffraction data were acquired on an EIGER2 X CdTe 9M flat-panel detector. The X-ray beam spot size was about 1.5 × 1.5 μm2. A single crystal of vanadinite Pb5(VO4)3Cl (space group P63/m, a = 10.299 Å, c = 7.308 Å) was used as a calibration standard for refinement of the instrument model of the diffractometer. Preliminary analysis was performed using Dioptas software37. Single crystal analysis was performed by combining the use of the CrysAlisPro25, DAFi20, and Jana200626 software. Magnetization measurements were performed using an MPMS3 SQUID-VSM (vibrating sample magnetometer) from Quantum Design.

Computational details

DFT calculations were performed with the ABINIT code38, within the Projector-Augmented Wave (PAW) formalism39. The exchange-correlation energy functional used throughout this work was the generalized gradient approximation as formulated by Perdew, Burke, and Ernzerhof (GGA-PBE)40. We took the PAW atomic datasets for Y and Fe from the Jollet–Torrent–Holzwarth table 41. For H, we used a PAW atomic dataset with a PAW radius of 0.8 bohr.

We explored different hydrogen content by randomly searching hydrogen distributions within the interstitial sites of the orthorhombic Cmcm structure of Y3Fe4Hx. Indeed, single crystal analysis enables to resolve the metallic structure of the compound, but not its hydrogen content. Experimental data analysis gave essential information to guide and constrain the random search: the space group Cmcm, the cell volume 500 Å3/4f.u., the lattice parameters, and the atomic positions of Y and Fe atoms at 83 GPa. Y3Fe4Hx can be seen as a mix of yttrium and iron superhydrides: YH3, FeH2, and FeH3 at 83 GPa. As a mix of YH3 and FeH2, the hydrogen content of the ternary compound Y3Fe4Hx would be 17 H per formula unit. As a mix of YH3 and FeH3, the hydrogen content of the ternary superhydrides Y3Fe4Hx would be 21 H per formula unit. Thus, we only explored hydrogen stoichiometries between Y3Fe4H17 and Y3Fe4H21.

We started from the conventional unit cell of Y3Fe4 found experimentally (4 formula units), and then randomly drew the hydrogen positions in this cell between 17 H/f.u. and 21 H/f.u., within the Wyckoff positions of the Cmcm space group. In all generated structures, iron and yttrium positions were fixed using the experimental positions, i.e., the following Wyckoff sites for Y: 4c and 8f, for Fe: 4b, 8f, and 4c. We used the conventional unit cell for the random generation of structures, rather than the primitive one, because Cmcm Wyckoff sites are traditionally described in the conventional cell.

For the random generation of hydrogen positions, we used the random number generator of R. Chandler and P. Northrup42. We proceeded as follows: within the conventional unit cell, and for a given hydrogen stoichiometry, a determined number of hydrogen positions had to be randomly drawn among the Wyckoff sites of the Cmcm space group. While this number was not reached, a Wyckoff site was randomly drawn among the possible ones, along with the necessary x, y, z positions. This step was repeated until the exact amount of hydrogen was reached (the multiplicity of each site was taken into account). Among the 8 possible Wyckoff sites of the Cmcm space group, 8d can be drawn at most once because it has determined positions. 4b also has determined positions, but is already occupied by a Fe. 4a is excluded since it implies an unphysical arrangement of atoms. The random configurations having atoms too close to each other were systematically rejected. The number of Wyckoff sites randomly drawn also depends on the explored hydrogen stoichiometries.

In order to gain computational time, the generated structures were then reduced to primitive cells with 2 f.u., and finally fully structurally optimized (all atomic positions including Fe and Y, and lattice parameters, within the Cmcm space group). The structural optimizations at the random searching stage were performed using the BFGS algorithm as implemented in ABINIT, with the following parameters: the Brillouin zone associated with the simulation cell was sampled by a 4 × 4 × 2 k-point mesh, a plane-wave cutoff of 15 Hartree was employed, and the optimization criterion on atomic forces was 2.0 × 10−3 Hartree/bohr, with a number of ionic steps typically limited to 75 steps. The 4 × 4 × 2 k-point mesh provides a numerical precision better than 1 meV/atom (typically a few 0.1 meV/atom) on the total energy of Y3Fe4H20, while the 15 Hartree plane-wave cut-off provides a precision of ~1 meV/atom on the formation energy of Y3Fe4H20. This method is inspired by the ab initio random searching (AIRSS) algorithm27, with the exception that the underlying crystal structure is imposed by the experimentally determined Cmcm-Y3Fe4Hx, and that only the positions of hydrogen atoms are randomly searched throughout available Wyckoff sites of the space group.

At the end of each structural optimization, the final enthalpy was computed, which allowed to identify the most stable structure by comparing all the enthalpies obtained for a given hydrogen stoichiometry. The number of structural optimizations performed for each considered hydrogen stoichiometry was typically varying between 1000 and 2000. Spin-polarization was not taken into account in this random search process.

Candidate structures were then selected as those having the lowest enthalpy. The identified candidates (around 5 per considered hydrogen stoichiometry) were then re-optimized using the BFGS algorithm with more stringent numerical parameters: a plane-wave cut-off of 25 Hartree, 8 × 8 × 4 k-point mesh to sample the Brillouin zone associated with the simulation cell, and an optimization criterion on the atomic forces of 1.0 × 10−5 Hartree/bohr (and 1.0 × 10−7 Hartree/Bohr3 on the stress tensor components). Remarkably, the final selected structure for Y3Fe4H20 is particularly stable relative to the other Y3Fe4H20 candidate structures, with a difference of about 2 eV/f.u.

The calculation of the formation enthalpies of Y3Fe4H17 requires to have the enthalpies of Y3Fe4 and H2 under pressure. We have thus structurally optimized these two systems under pressure, using the same criteria, and dense hydrogen in its C2/c-32 structure. However, to our knowledge, there is no evidence that the Y3Fe4 alloy exists at this pressure or below. Thus, we chose to use Y3Fe4H17 as the reference for plotting the convex-hull and not Y3Fe4. Experimental excess hydrogen ensures that the thermodynamically favored structure is the one lying on the convex-hull nearest H2, which in this case is Y3Fe4H20.

The dynamical stability of the predicted crystal structure Y3Fe4H20 superhydride was then tested at 90 GPa (and 0 GPa) by computing the phonon dispersion curves. These calculations were performed within the framework of the DFPT, as implemented in ABINIT38. For these calculations, the structure was first fully re-optimized. We used a 4 × 4 × 2 Γ-centered q-point mesh in DFPT calculations to sample the Brillouin zone.

We then calculated the formation enthalpies at several pressures from 90 to 0 GPa and compared it to that of the decomposition compounds to ensure Y3Fe4H20 stability. The selected decomposition compounds and the results are listed in the Supplementary Material.

Finally, we computed the electronic density distribution and the ELF within the norm-conserving formalism (ELF is not available in the PAW formalism in ABINIT) to evaluate the nature of hydrogen-hydrogen and metal-hydrogen bonds.

The study of the Laves phase hydrides YFe2H6 and YFe2H723 highlighted the importance of taking into account spin-polarization and Hubbard corrections on iron atoms. The compression curves of Y3Fe4H20 calculated under various approximations are compared to experiment and GGA-PBE calculations (see the Supplementary Material). Interestingly, we see that within the GGA-PBE, taking into account spin degrees of freedom initialized with a hypothetical ferromagnetic order in Y3Fe4H20 always reverts to the non-magnetic state. Thus, it gives exactly the same results as the calculated compression curve without spin degrees of freedom. Including electronic correlations, by adding a Hubbard correction on the Fe d-orbitals within the GGA-PBE+U framework (U = 2.6 eV and J = 0.9 eV43) provides close results, with however a slightly ferromagnetic system, the magnetic moment per Fe atom decreasing from ~0.75 μB at 0 GPa to ~0.25 μB at ~70 GPa, and then to zero at higher pressure.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information Reporting Summary Transparent Peer Review file

Supplementary Information

Reporting Summary

Transparent Peer Review file