Structural modulation and spin glassiness upon oxidation in oxygen storage material LnFeMnO_4+x for Ln = Y, Lu, and Yb Open Access

Hexagonal-layered LnM²⁺M³⁺O₄ (Ln = lanthanide, M = transitional metal) has been attracting great interest in materials chemistry and physics due to its unique crystal structure and high tunability in compositions.^1–4 The structure of LnM²⁺M³⁺O₄ consists of alternating layers of edge-sharing LnO octahedra with a double layer of edge-sharing MO trigonal bipyramids. The M²⁺ and M³⁺ cations occupy the same crystallographic sites. In the LnO octahedra layer, the Ln³⁺ cations arrange in a triangular lattice, whereas in the MO layer, the M²⁺/M³⁺ cations are found in a non-corrugated honeycomb lattice. Both the triangular⁵ and the honeycomb⁶ lattices can display geometric frustration relevant to magnetism. This potential spin frustration has attracted attention from the condensed matter community as this family of oxides could host quantum spin-liquid behavior. Such candidates include those where S = ½ as found in YbMgGaO₄,^7–10 YbZnGaO₄,^11,12 and LuCuGaO₄.¹³ The fact that M²⁺ and M³⁺ share the same crystallographic site also leads to a great variety of interesting chemical and physical phenomena. In the famous case of mixed-valence LnFe₂O₄, where Fe²⁺ and Fe³⁺ coexist at the same sites, ferrimagnetic ordering coincides with charge ordering between Fe²⁺ and Fe³⁺, leading to multiferroicity.^14–19 

Oxygen non-stoichiometry in LnM²⁺M³⁺O₄ has also attracted great interest since physical properties of LnM²⁺M³⁺O₄, especially LnFe₂O₄, are extremely sensitive to oxygen stoichiometry.^20–25 However, the oxygen-abundant form of LnFe₂O_4+δ was not realized until Hervieu et al. observed that LuFe₂O₄ could be oxidized to the hyper-stoichiometric LuFe₂O_4.5 at a relatively low temperature of 200 °C in air.²⁶ The oxygen uptake behavior of other LnFe₂O₄ (Ln = Yb, Y, and In) was later reported following this discovery.^27–31 The structural transition from LnFe₂O₄ to LnFe₂O_4.5 is also found to be highly reversible, implying a reversible oxygen uptake/release behavior in the LnFe₂O₄ system.^29,30,32 The oxygen uptake behavior of LnFe₂O₄ suggests its potential as an oxygen storage material and can also lead to interesting structure transitions and physical property changes.^26,29

Inspired by the interesting oxygen uptake behavior in LnFe₂O₄, we hypothesize that the substitution of Fe²⁺ by Mn²⁺ will control the oxygen uptake behavior and the nature of associated structural transitions and magnetism. In this study, we systematically explore the change before and after the oxidation of LnFe³⁺Mn²⁺O₄. We focus on the difference in the average structure, local structure, structural modulation, and magnetic properties between the reduced phase and corresponding oxidized phase. We investigate the possible influence of the Ln cations on oxygen uptake and compare them with their LnFe₂O₄ analogs.

LnMnFe₂O₄ (Ln = Yb, Lu, and Y) series, labeled as the reduced phases, were prepared using solid-state reactions.² Ln₂O₃ (Ln = Yb, Lu, and Y), Fe₂O₃, MnO, and Fe powder (99.9% pure) were ground in a 1:1:2 ratio, respectively, to prepare around 2 g of the target compound. The powder mixtures were pressed into pellets of 13 mm in diameter. The pellets were placed in a 2 ml alumina crucible and sealed inside evacuated 8 mm diameter quartz ampoules. The ampoules were then heated to 1180 °C (heating rate 10 °C/min) for at least 12 h and subsequently quenched in an ice water bath.

The AB₂O₄ phase is metastable; hence, quenching is required to kinetically lock it in at lower temperatures. Quenching limits the number of oxide impurities. The elemental ratio of the product was verified with ICP and scanning electron microscopy (SEM, Hitachi SU-70) in conjunction with energy dispersive spectroscopy (EDS, Bruker XFlash 6/60), SEM-EDS. To prepare the oxidized phases, LnMnFe₂O_4.5 (Ln = Yb, Lu, and Y), for ex situ studies, the as-synthesized samples were oxidized at 600 °C (heating rate 10 °C/min) under air for 12 h and then cooled to room temperature. These samples are labeled as oxidized phases.

TGA was conducted on reduced phase LnMnFe₂O₄ (Ln = Yb, Lu, Y). In total, 5–10 mg of sample was used for each measurement. TGA measurement with isotherm heating was performed. Samples of ∼10 mg were heated to 350 °C (ramping rate 10 °C/min) in air, and the temperature was held for 6 h.

LnMnFe₂O₄ (Ln = Yb, Lu, and Y) series as well as their oxidized phases LnMnFe₂O_4.5 were characterized with high-resolution synchrotron x-ray powder diffraction (SXPD). The experiments were performed on the 11-BM beamline at the Advanced Photon Source (APS) at Argonne National Laboratory. X-rays were of wavelength 0.457 895 Å. Room temperature data were collected. To compare with the SXPD patterns and explore the possible magnetic structures, we also took time-of-flight (TOF) neutron diffraction patterns for LnMnFe₂O₄ and LnMnFe₂O_4.5 at different temperatures between 5 and 300 K. The TOF patterns were collected on the powder diffractometer POWGEN beamline at the Spallation Neutron Source (Oak Ridge National Laboratory). Rietveld analysis was carried out using JANA^33,34 and GSASii.³⁵ Structures of oxidized phases were solved using the charge flipping method.^36,37 Structural modulation of the oxidized phase was refined with JANA.

The x-ray total scattering data used for x-ray (λ = 0.457 895 Å) pair distribution function (XPDF) analysis were collected on the 11-IDB beamline also at the APS. Data were collected at room temperature (∼300 K) and were reduced using GSASii.³⁵ Neutron total scattering data were obtained at the POWGEN beamline at the SNS. Data were collected at 5 and 300 K.

Raman spectroscopy was obtained with Yvon Jobin LabRam ARAMIS using a 532 nm laser source. Powders were used for the measurement. Because the samples strongly absorb the laser light and can be oxidized, a long exposure time (>5 min) and a small aperture were used to obtain the Raman scattering signal.

Transmission electron microscopy (TEM) images, as well as electron diffraction patterns, were taken using a JEOL 2100F field emission gun (scanning) TEM, (S)TEM, equipped with Gatan image filter (GIF, model of Tridiem 863), operated at 200 kV accelerating voltage.

Temperature-dependent magnetic susceptibility measurements were performed on both reduced and oxidized phases with a SQUID magnetic property measurement system (MPMS, Quantum Design). For each sample, a portion of powder with known mass was loaded into a gel capsule with cotton suspended inside a plastic straw. Field-cooled (FC) and zero field-cooled (ZFC) magnetization measurements were recorded in direct current mode from room temperature (300 K) to base temperature under an applied field of 100 Oe. Zero field AC susceptibility at 1, 10, 100, and 1000 Hz was also performed at the temperature range of interest.

Since crystal structure of LnMnFeO₄ (Ln = Lu, Yb, and Y) is isostructural to the parent phase LnFe₂O₄ (space group $R3̄m$⁠),^2,4,38 we used the latter’s structure as the starting model for our Rietveld analysis. The refinements were performed with combined neutron and synchrotron x-ray powder datasets. The neutron diffraction pattern with the Rietveld refinement fit for the YbFeMnO₄ sample is displayed in Fig. 1(a), and the corresponding plots for the Ln = Lu and Y samples are presented in Figs. S4 and S8. The refined lattice parameters and atom positions of LnMnFeO₄ (Ln = Lu, Yb, and Y) are shown in Tables S1, S3, and S6.

No significant difference was found between LnMnFeO₄ and LnFe₂O₄. The Ln site displays significant displacement along the c-direction, which is commonly observed in the LnM²⁺M³⁺O₄ phase.^29,39 We noted that some studies use a structure model, where the position of Ln (0, 0, and 0) is split into two partially occupied sites (0, 0, and 0 ± δ).⁴⁰ Both models suggest a static positional disorder on the Ln site.

We performed TGA measurement on LnMnFeO₄ under the oxygen flow, and the result for YbFeMnO₄ is displayed in Fig. 2 (Ln = Lu, Y in Figs. S4b and S8b). Similar to what was observed for LnFe₂O₄ (Ln = Lu, Yb, and Y),^26,29,30,32 LnFeMnO₄ also uptakes oxygen into its lattice at a relatively low temperature of 200 °C. The maximum weight gain was 2.51%, 2.21%, and 3.88%, corresponding to the change from LnFeMnO₄ to LnFeMnO_4.5. Assuming that Mn²⁺ is oxidized to Mn³⁺, the theoretical mass gain should be 2.29, 2.3, and 3.03% for Ln = Lu, Yb, and Y, respectively.

To verify the Mn oxidation state, we took electron energy-loss (EEL) spectra of Mn before and after oxidation (Fig. S13). By examining the electron energy-loss near-edge structure (ELNES) on the Mn L_2,3-edge, which corresponds to the transitions from the Mn 2p core orbitals to the unoccupied Mn 3d state (dipole selection rule ∆ℓ = ±1, where ℓ is an angular quantum number),⁴¹ one can obtain information on the chemical coordination of Mn. Our EELS results confirm that the Mn valence state transits from Mn²⁺ into Mn³⁺,⁴² and more details on the ELNES analysis can be found in the supplementary material.

The powder diffraction patterns of LnFeMnO_4.5 can be indexed with the $P3̄$ space group, the same that was found in YbFe₂O_4.5²⁹ and LuFe₂O_4.5.²⁶ However, we also found space group $P3̄m1$ to provide an identical fit. We then, therefore, applied the charge flipping method to the SXRD patterns to elucidate the crystal structure of the oxidized phase, LnFeMnO_4.5. Such a methodology helped determine the details of the Ln (Ln = Yb, Lu, and Yb), Fe/Mn, and one of the oxygen atom positions(O1). Afterward, we employed a Fourier difference map to locate the average positions of other oxygen atoms (O2 and O3) with both the NPD and SXRD datasets. Combined refinement was then performed based on the solved structure.

The neutron diffraction pattern with the Rietveld refinement fit of LnFeMnO_4.5 is shown in Fig. 1(b) for Ln = Yb and Figs. S4c and S8c for Ln = Lu and Y. The refined lattice parameters and atom positions for LuFeMnO_4.5, YbFeMnO_4.5, and YFeMnO_4.5 are included in Tables S2, S4, and S6. Not surprisingly, the solved average structure of LnFeMnO_4.5 (Ln = Lu, Yb, and Y) resembles what is reported on YbFe₂O_4.5²⁹ and is highly correlated with the original reduced phase LnFeMnO₄. Figures 1(c) and 1(d) show the structure of LnMnFeO₄ and LnFeMnO_4.5, respectively. Both $P3̄$ and $P3̄m1$ space groups can lead to the same structure for LnFeMnO_4.5 because the atoms only occupy special locations.

From LnMnFeO₄ to LnFeMnO_4.5, the abundant oxygen atoms are inserted in the middle of Fe/Mn oxide bipyramid double layers, which leads to the shift of the layers along the ab-directions.²⁹ Due to the layer shift, the newly generated unit cell is only 1/3 of the original unit cell (a_o = a_r, c_o = 1/3c_r). Among all three LnFeMnO_4.5 (Ln = Lu, Yb, and Y) samples, we noted that the diffraction peaks for LnFeMnO_4.5 show a broader and more asymmetric nature than LnFeMnO₄, which is an indication of stacking faults upon oxidation.^43,44 The appearance of the stacking faults also implies the layer-shift mechanism upon oxygen uptake.

We note that the refinement on the LnFeMnO_4.5 with the solved average structure gives relatively high R_wp values (16%). The misfit mainly arises from the unaccounted satellite peaks [Figs. 1(b), S4c, and S8c], which are similar in all three samples of Ln. The appearance of the satellite peaks indicates a structural modulation,^45,46 which has also been observed in YbFe₂O_4.5²⁹ and LuFe₂O_4.5.²⁶ Given the similarity of the average structures between LuFe₂O_4.5/YbFe₂O_4.5, structural modulation in LnFeMnO_4.5 is expected. However, the positions and the distributions of the satellite peaks in LnFeMnO_4.5 (Ln = Lu, Yb, and Y) appear to be completely different from those reported in YbFe₂O_4.5 and LuFe₂O_4.5. We will discuss the structural modulation of LnFeMnO_4.5 in detail in Sec. III C. For both LnMnFeO₄ and LnMnFeO_4.5 (Ln = Lu, Yb, and Y), no cation ordering was observed for the Mn and Fe positions.

The Rietveld refinement results indicate that the oxidation of LnFeMnO₄ to LnFeMnO_4.5 leads to a considerable contraction in the c-direction and a slight expansion along the a-direction when the unit cells are normalized to the same chemical formula (a_r = 1/3c_r, a_o = c_o). This trend is not influenced by the nature of Ln, as shown in Tables S1–S6. This c-direction contraction is also evidenced by the Raman measurements (Figs. 3, S4g, and S7g), which afford information on the local structure. The Raman spectrum of LnFe₂O₄ is well studied,^47–52 and the spectrum we recorded of YbFeMnO₄ (blue curve in Fig. 3) resembles that of LnFe₂O₄. The major phonon modes of YbFeMnO₄ are assigned based on the Raman spectra studies of LnFe₂O₄.⁵¹ 

For the Raman spectra of the oxidized phase YbFeMnO_4.5 (red curve), we were not able to assign the phonon modes exactly due to the new space group symmetry and structural modulations. However, the structures of YbFeMnO₄ and YbFeMnO_4.5 are highly correlated; hence, there must be some resemblance between their phonon modes. The most intense peak (assigned to $A1g3$ at 610 cm⁻¹) for the reduced phase YbFeMnO₄ corresponds to the stretching vibration of the lattice along the c-direction. After oxidation to YbFeMnO_4.5, the layered framework remains; thus, this $A1g3$ mode should still exist in the oxidized phase. We argue that the most intense peak blueshifts to 720 cm⁻¹ for YbFeMnO_4.5. Based on such an assignment, oxidation of these materials leads to stronger covalency and shorter bond distances along the c-direction.

Neutron PDF analysis also affords information on the local structure. The PDF curves before and after oxidation are presented in Fig. 4 for Ln = Yb (Fig. S4f for Ln = Lu). The comparison of neutron PDFs indicates that oxidation mainly changes the local environment of the Fe/Mn and O atoms. The Fe–O1 distance (along the c-direction) is shortened, and the Fe–O2 distance is elongated (along the horizontal direction) after oxygen uptake, which is also confirmed by the refined average structure. O–O distances around 2.5–3.0 Å are shortened after oxidation, implying that the crowding among the oxygens is due to oxygen insertion. It is also noted that the relative O–O pair intensities of LnFeMnO_4.5 are enhanced compared with LnFeMnO₄, confirming more oxygen atoms are accommodated into the structure after oxygen uptake. The x-ray PDFs of LnFeMnO₄ and LnFeMnO_4.5 (Ln = Yb and Lu) are shown in Figs. S3 and S7. The M-O and O–O pairs are much less obvious compared with the neutron PDFs due to the insensitivity of oxygen compared with neutron scattering. However, we do confirm from x-ray PDFs that the observable atomic distances along the c-direction are shortened after the oxygen uptake, consistent with the Raman and Rietveld results.

To better understand the structural differences between LnFeMnO₄ and LnFeMnO_4.5, high-resolution TEM (HRTEM) imaging and electron diffraction (ED) were performed on powder grains (Fig. 5 for Ln = Yb). The orientations of the ED patterns were determined based on the simulated electron diffraction (ED) patterns of the refined crystal structures (Figs. S11 and S12). For the reduced phase YbFeMnO₄, we observed that most grains prefer to orient either along the [001] direction [Figs. 5(a) and 5(b)] or the [010] direction [Figs. 5(c) and 5(d)], suggesting that (001) and (010) are the preferred exposed facets for YbFeMnO₄.

The fast Fourier-transform (FFT) pattern converted from the HRTEM images and ED patterns also shows a hexagonal arrangement of diffraction spots, confirming the rhombohedral cells. The HRTEM image [Fig. 5(c)] along the incident electron beam of [010] direction shows a layered stacking fringe, consistent with the refined crystal structure where LnO octahedra layers and Fe/MnO bipyramid double layers are alternatively stacked. It is reported that satellite peaks can be observed in the ED patterns for LnFe₂O₄ due to the charge ordering of Fe²⁺ and Fe³⁺.^15,17,53,54 However, in our case of YbFeMnO4, there is no charge ordering because the Mn²⁺/Fe³⁺ pair is much more energy-favored than the Fe²⁺/Mn³⁺ pair.^18,55

After oxygen uptake, the major Bragg diffraction spots along the [001] direction maintain the hexagonal arrangement, whereas those along the [110] direction remain vertically aligned, confirming the unchanged rhombohedral unit cell (a = b, α = β = 90°, and γ = 120°) after oxidation. HRTEM image [Fig. 5(g)] along the incident electron beam of [110] direction also shows a layered stacking fringe, indicating that the stacked layered framework persists after oxygen uptake. We also realized that most grains are now either oriented along the [001] direction [Figs. 5(e) and 5(f)] or the [110] direction [Figs. 5(g) and 5(h)] after the oxidation, which implies that the (110) planes observed in the oxidized phase are transformed from the (010) planes in the reduced phase. This transformation strengthens the argument that the structural transition from YbFeMnO₄ to YbFeMnO_4.5 is through the mechanism of layer shift along the ab directions.

Apart from the major reflections in the ED patterns for YbFeMnO_4.5, satellite reflections are noted in both [001] and [110] directions [Figs. 5(f) and 5(h)]. The existence of satellite reflections is clearly an indication of structural modulations, which is consistent with the observed satellites in both synchrotron x-ray and neutron diffraction. In the [001] direction [Fig. 5(f)], each main Bragg diffraction spot was surrounded by 12 satellite spots. We interpret this number of satellites as a pair of reflections at each hexagonal edge with some angle of splitting. The symmetric arrangement of the satellites implies a modulation vector q of high symmetry. In the [110] direction [Fig. 5(h)], the additional superlattice spots were observed within the two main Bragg diffraction spots and were horizontal to the main $(1̄01)$ Bragg diffraction spot with a modulation wave vectors q = (0, 1/3, and 0), suggesting that the modulation is 2D and propagates within the ab-plane [q= (a, b, and 0)]. Interestingly, the structural modulation seems so strong that it can be rendered in real-space HRTEM images. Both Figs. 5(e) and 5(g) show aperiodic structures with a nanometer scale apart from the atomic arrangement. The satellites are also visible in the FFT patterns [Figs. 5(e) and 5(g) inserts]. The observed HRTEM images and ED patterns are highly identical, regardless of the Ln site elements. Figures S14 and S15 show HRTEM images and ED patterns of LuFeMnO₄ and YFeMnO₄ before and after oxidation. The satellite Bragg diffraction spots appearing in the ED patterns of oxidized phases along [001] and [110] directions are highly identical to what is observed on YbFeMnO_4.5, showing a 12-direction and highly symmetric structural modulation that propagates within the ab-plane.

Structural modulation is reported on LuFe₂O_4.5 and YbFe₂O_4.5 as well. However, the observed propagation of structural modulation in LnFe₂O_4.5 shows a great deviation from what we observed on LnFeMnO_4.5. In the LnFe₂O_4.5 system (Ln = Yb and Lu), the structural modulation is reported to propagate along ac and bc planes [modulation wave vectors q = (a, 0, c) ].^26,27,29 In addition, only 6 instead of 12 propagation directions of the modulation are observed in their ED patterns.^26,27,29

The magnetic properties of reduced phase LnFeMnO₄ and oxidized phase LnFeMnO_4.5 are investigated through magnetic susceptibility and neutron diffraction measurements. Susceptibility measurement of YbFeMnO₄ has been previously reported, and our results on YbFeMnO₄ display similar behavior.¹⁸ The DC susceptibility from 2 to 300 K (Fig. S16a) shows that the reduced phase YbFeMnO₄ displays an antiferromagnetic transition around 65 K. The Currie-Weiss fit (250–300 K, Fig. S16b) provides an effective moment size of 4.02 u_B, per formula unit, which deviates significantly from the theoretical value of 11.8 u_B based on the addition of all three magnetic atoms (one Mn²⁺, one Fe³⁺, and one Yb³⁺). This deviation suggests that there are still strong magnetic correlations among the magnetic ions of YbFeMnO₄ even at room temperature.

The AC susceptibility measurements [Figs. 6(a) and 6(b)] show that the antiferromagnetic transition temperature varies when different frequencies are applied. We calculated the value of ΔT/T/Δ log f to be 0.007 (where ΔT = peak shift, T = peak temperature, and Δ log f = difference of the logarithm of frequency). This calculated value is close to the minimum one for spin glassy systems (∼0.004).⁵⁶ In the imaginary part (χ″) for YbFeMnO₄, a broad peak or shoulder was also observed below T_f (around 30 K) and was found to shift with changing the frequency, which may be related to the anisotropic spin freezing.^57,58

Time of flight powder neutron diffractions at 300, 100, 40, and 5 K on YbFeMnO₄ are presented in Fig. 6(c). We did not observe any structural transition from 300 to 5 K as the diffraction peaks from the nuclear component remain nearly identical across the whole temperature range. PDFs from 5 to 300 K do not show any major deviations either (Figs. S20 and S21). However, broad magnetic peaks at Q = 1.25 Å⁻¹ and Q = 2.5 Å⁻¹ grow upon cooling. The magnetic peaks appear even at room temperature, implying a strong magnetic correlation, which is consistent with the DC susceptibility analysis. We noted the magnetic peaks are considerably broad and highly asymmetric, suggesting the magnetic ordering is short range and two-dimensional.^59–61 We indexed the magnetic peaks as the (1/3, 1/3, l) and (2/3, 2/3, l).

After oxidation to YbFeMnO_4.5, the magnetic properties significantly altered. DC susceptibility measurements from 2 to 300 K (Fig. S16) indicate YbFeMnO_4.5 also displays an antiferromagnetic transition, however, at a much lower temperature (around 30.5 K compared with 62 K for YbFeMnO₄). The moment gained from the Currie-Weiss fit is 15.61 u_B per formula, still deviating from the theoretical value (15.3 u_B: one Mn³⁺, one Fe³⁺, one Yb³⁺). AC susceptibility measurements [Figs. 6(d) and 6(e)] show that this antiferromagnetic transition temperature varies when different frequencies are applied. We calculated ΔT/T/Δ log f to be 0.0066, still very close to the minimum value for spin glass systems. In the imaginary part (χ″) for YbFeMnO_4.5, a relatively broad peak except for the peak for T_f was also observed around 10 K and was found to shift with changing the frequency [Fig. 6(e)].

Time of flight powder neutron diffraction at 300, 10, 40, and 5 K on LuFeMnO_4.5 are presented in Fig. 6(f). Similar to YbFeMnO₄, no nuclear structural transition is observed from 300 to 5 K (also true for the PDFs at 5 and 300 K). A magnetic peak is still visible at Q = 1.25 Å⁻¹ even at room temperature, but more diffuse compared with YbFeMnO₄.

We found that the Ln site element has minimal influence on the magnetic properties of the reduced phase LnFeMnO₄ and the oxidized phase LnFeMnO_4.5. The same magnetic susceptibility and temperature-dependent neutron diffraction measurements were performed on the Ln = Lu and Y samples, and the results are provided in Figs. S16–S18. Important parameters gained from the magnetic measurements are listed in Table S7. All LnFeMnO₄ and LnFeMnO_4.5 show spin-glass behavior. The effective moments from Curie-Weiss fits never agree with the theoretical values based on isolated moments. In the neutron datasets, the reduced phases of LnFeMnO₄ all show broad and highly asymmetric magnetic peaks upon cooling. The positions of the magnetic peaks can be indexed into (1/3, 1/3, l) and (2/3, 2/3, l), regardless of the Ln cations. Upon oxidation, magnetic peaks become much less visible, and there is a decrease in temperature for the antiferromagnetic transition.

Table I compares the normalized lattice parameters and some of the special distances within the crystal structure of LnFeMnO₄ before and after oxygen uptake. The refined lattice parameters indicate that for all LnFeMnO₄ (Ln = Y, Lu, and Yb) samples, oxygen uptake leads to contraction along the c-direction. Measuring the thickness of the LnO octahedra layer and the Fe/MnO bipyramid double layer, we found out the Fe/Mn bipyramid double layer undergoes contraction, whereas the LnO octahedra layer expands after the oxygen uptake, as shown in Table I. Thus, the contraction of the c-lattice parameter of the oxidized phase is mainly a result of the contraction of Fe/MnO bipyramid double layers. Previous studies show the oxidation of LnFe₂O₄ (Ln = Y, Lu, and Yb) leads to the expansion along c-directions,^26,29,30 which is opposite to what we have observed on LnFeMnO₄, as presented in Fig. 7.

The oxygen uptake of LnFeMnO₄ correlates with the oxidation of Mn²⁺ (d⁵) into Mn³⁺ (d⁴) within the structure, whereas the oxygen uptake of LnFe₂O₄ correlates with the oxidation of Fe²⁺(d⁶) into Fe³⁺(d⁵). The different electron configuration changes upon oxidation (d⁵ to d⁴ for oxidation of LnFeMnO₄ and d⁶ to d⁵ for oxidation of LnFe₂O₄) can influence the crystal field effect in the Mn/FeO₅ bipyramid,^32,62 causing opposite lattice expansion behaviors for LnFeMnO₄ and LnFe₂O₄.

The oxygen uptake of LnFeMnO₄ into LnFeMnO_4.5 leads to the insertion of extra oxygen into the crystal lattice. The average structure of LnFeMnO_4.5 [Fig. 1(d)] indicates that the extra oxygens (O3) are located in between the Fe/Mn–O double layer. Based on such an observation, intuitively, we can infer that the oxygen insertion process on LnFeMnO_4.5 should happen in between the Fe/Mn–O bipyramid layers [in the middle of the Fe/Mn–O bipyramid double layer, close to the lattice plane (002)].²⁹ If this proposed mechanism is correct, at the early stage of oxygen uptake of LnFeMnO₄, where the oxygen uptake is far smaller than 0.5, the inserted oxygen atoms should exist between the Fe/Mn–O bipyramid layers as interstitial oxygen before the phase transition occurs. To further verify this proposed insertion mechanism, the bond valence summation approach^63,64 was applied to the reduced phase LnFeMnO₄ to estimate the relative energy of the interstitial oxygen ions at different locations. The program softBV is commonly used to estimate the migration pathways for ions based on the energy of the sites estimated from the bond valence summation.^65,66 The calculated energy of sites can also be used to estimate the preferred occupying sites of certain ions. Figure 8 displays the sites for interstitial oxygen (position coordinates are included in Table S8), which are calculated to have the lowest energy. We clearly observe that the possible low-energy interstitial oxygen sites are all concentrated in between the Fe/Mn–O bipyramid layers, suggesting that the oxygen insertion in between the Fe/Mn–O bipyramid layers is the most energy favorable.

Bond valence summation analysis also indicates that the Mn/Fe sites are under-bonded in LnFeMnO₄. The bond valence sum of the Mn/Fe site is calculated to be around +2.25 compared with the theoretic value of +2.5. This under-bonded environment in LnFeMnO₄ potentially allows the Mn/Fe site to bond extra oxygen atoms under oxidation conditions.

Structural modulation upon oxidation of LnFe₂O₄ into LnFe₂O_4.5 has been recently reported.^26,27,29 Also, it was shown that the propagation of modulation changes in LuFe₂O₄, depending on how much oxygen the material can uptake.^26,28 Our study only focuses on the fully oxidized phase LnFeMnO_4.5. As mentioned previously, YFeMnO_4.5, LuFeMnO_4.5, and YbFeMnO_4.5 possess similar modulation structures as they all show a similar arrangement of satellite peaks in the diffraction patterns. Figures 9(a) and 9(b) show the enlarged ED patterns of YbFeMnO_4.5 from Figs. 5(f) and 5(h), respectively. The 12 modulation vectors can be clearly seen in Fig. 9(a). Since direction does not change between the real space and reciprocal space, the direction of modulations can be visualized in the real crystal structure based on the direction of satellites in the electron diffraction. Figure 9(c) presents the top view (from the c-direction) of the YbFeMnO_4.5 average structure lattice, with the direction of modulation vectors indicated. We note that the directions of the 12 modulations are indeed identical in the average structure.

The angle between the wave vector q and a-axis is measured to be around 19°, aligned with the diagonal line of the two-unit cells [shown in Fig. 9(c)]. The modulation wave vector q is measured to be (−0.28, 0.14, and 0), close to the commensurate vector (−2/7, 1/7, 0). Assuming that it is indeed commensurate, the 12 modulation vectors are measured to be (−2/7, 1/7, 0), (1/7, −2/7, 0), (3/7, 1/7, 0), (3/7, 2/7, 0), (2/7, 3/7, 0), (1/7, 3/7, 0), (−3/7, −1/7, 0), (−3/7, −2/7, 0), (−2/7, −3/7, 0), (−1/7, −3/7, 0), (2/7, −1/7, 0), and (1/7, −2/7, 0).

Our observed ED patterns are homogeneous in the samples, as all tested crystalline grains show the same results. From these observations, we quickly realized that all 12 modulation wave vectors can be generated through symmetry operations from the $P3̄m1$ space group (symmetry operations of $P3̄m1$ are listed in Table S9)⁶⁷ on wave vector q. However, applying symmetry operations from $P3̄$ space group (symmetry operations of $P3̄$ are listed in Table S9) on wave vector q can only generate six identical modulation wave vectors, in which case, two independent modulation wave vectors q = (−2/7, 17, 0) and q′ = (−17, 2/7, 0) are necessary to describe the structural modulations. From the symmetry of the modulation vectors, $P3̄m1$ seems to be a better choice for describing the average structure of LnFeMnO_4.5. However, it is rare but might be possible that the modulation wave vectors q = (−2/7, 1/7, and 0) and q′ = (−1/7, 2/7, and 0) are twined, in which case, $P3̄$ will be a better choice for describing the average structure of LnFeMnO_4.5.

Although LnFe₂O_4.5 also shows structural modulations,^26,27,29 the observed modulation wave vectors of LnFe₂O_4.5 (Ln = Lu and Fe) are in great difference with our LnFeMnO_4.5 system. LuFe₂O_4.5 show a modulation wave vector (0.31, 0, and 0.33)²⁶ and YbFe₂O_4.5 show a modulation wave vector (0.31, 0, and 0.33),²⁹ both of which propagate within bc- and ac-planes rather than the ab-plane. The multiplicity of the modulation wave vector observed in LuFe₂O_4.5 and YbFe₂O_4.5 is less than that of the LnFeMnO_4.5 system, indicating a lower symmetry for the structural modulations. Since LnFe₂O_4.5 and LnFeMnO_4.5 show nearly identical average structures, the ultimate reason for their very different modulation vectors is not clear. We previously noted that the oxidation in LnFe₂O₄ leads to c-direction expansion,^26,29 whereas oxidation in LnFeMnO₄ to LnFeMnO_4.5 leads to c-direction contraction. Likely that different anisotropic lattice expansion behavior upon oxidation is correlated with the different structural modulations.

We further refined the structural modulation of LnFeMnO_4.5 with the powder diffraction data. Although the measured modulation wave vector from ED results seems to be a commensurate value, treating the structure as a commensurate case would have required building a 7 × 7 supercell, which can be extremely complex. Thus, here, we approximate the structural modulation as an incommensurate case and apply the superspace approach.^45,46 The modulation wave vectors are indexed and refined on both synchrotron and neutron diffractions. Figures 9(d) and 9(e) show the refinement of YbFeMnO_4.5 with SXRD and NPD data. The refinements of LuFeMnO_4.5 and YFeMnO_4.5 are displayed in Figs. S21–S24. Here, we only applied one modulation wave vector q= (1/7, 3/7, and 0) with $P1̄(ab0)$ group, and we can see that is already enough to account for all the positions of satellite peaks, confirming the symmetric nature of 12 modulation vectors. The indexed and refined modulation vectors from powder diffractions are listed in Table II, which are in good agreement with ED. We also note that those satellite peaks in the powder diffraction patterns display extreme Lorentzian broadening features [Figs. 9(d) and 9(e)], suggesting that the modulation is not continuous across whole crystal grains but has nano-size domains.

Due to the existence of stacking faults and small domains of structural modulation, solving the modulation structure of LnFeMnO_4.5 from only powder diffraction is extremely challenging. Thus, we aimed to qualitatively understand the modulation rather than fully solve the modulations. To alleviate the complexity modulation refinement, $P1̄ab0$ group was used, and only modulation along one q vector (1/7, 3/7, and 0) was refined. We also assumed the modulation along other modulation vectors should be symmetric or centrosymmetric to the modulation along one q vector (1/7, 3/7, and 0). We mainly used YbFeMnO_4.5 to perform analysis for understanding, as the quality of diffraction patterns of LuFeMnO_4.5 and YFeMnO_4.5 are lower due to much more severe stacking faults. For SXRD, we only refined the modulation of Ln and Fe/Mn sites as x-ray is not sensitive to oxygen. For NPD of YbFeMnO_4.5, the Yb, Fe/Mn, and all O sites were refined. The refinement on LuFeMnO_4.5 was also attempted. The displacive modulation along a, b, and c for Yb, Fe, and oxygen atoms are shown in Figs. S25–S39, which is refined from neutron diffraction of YbFeMnO_4.5. The refined modulation parameters are listed in Tables S10–S13.

Figure 10 shows a displacive modulation of O1, Fe/Mn, O3, and O2 sites along the c-direction. The Yb, Fe/Mn, and O1 sites all show strong periodic displacement along the modulation vector. Displacement of Fe and O1 along the c direction is relatively strong. We noted that the modulation curves for Fe/Mn and O1 along the c-direction have a phase difference of t/2, causing periodic breathing of Fe–O1 bonds, whereas O2 and O3 are partially occupied. Their occupancy periodicities also show a phase difference of roughly t/2, suggesting that O2 and O3 do not appear in the same unit cell. In the average structure of LnFeMnO_4.5, the distance between O2 and O3 is around 2–2.2 A (Table I), which is far smaller than two times the radius of the O²⁻ anion (2.4–2.8 A).⁶⁸ The alternate occupancy of O2 and O3 makes physical sense as the occupancy at the same unit cell of O2 and O3 would cause a covalent overlap. Since structure has an inversion center, inversion of the modulation of O3 shows that the periodicity of the occupancy of O2 and O3 inversion sites will have a t/2 phase shift. The phase shift of the inversion sites O3′ (or O2′) indicates that O3′ and O3 should probably not appear in the same unit cell either. Since modulation appears after oxygen insertion, it is likely that the ordering of oxygen vacancy ordering is one of the major drivers for the structural modulations.

The magnetic properties of YFeMnO₄ have been previously studied.^60,61,69,70 The AC susceptibility suggests that LnFeMnO₄ shows a weak spin glass-like state with the spin freezing transition. Our temperature-dependent neutron diffraction indicates that 2D short-range magnetic ordering exists in LnFeMnO₄, regardless of Ln. Following the order of Ln = Y, Yb, and Lu, as the c-lattice parameter increases, the spin-frozen transition temperature decreases slightly increases (Table S7), suggesting the dominant magnetic interaction might be along the ab-plane rather than the c-direction.

Magnetization measurement on single crystal YFeMnO₄ at 4.2 K shows an anisotropic ferrimagnetic behavior of YFeMnO₄. A larger magnetic hysteresis loop was observed on YFeMnO₄ parallel to c-direction than vertical to it. Two-dimensional ferrimagnetic ordering is also observed in the LnFe₂O₄ systems.^14,71,72 In the LnFe₂O₄ or LnFeMnO₄ systems, the Fe/Mn–O bipyramid double layers are separated by the Ln-O octahedral layers [shown in Fig. 1(c)], leading to extremely weak magnetic interactions between Fe/Mn–O bipyramid double layers in the c-direction.

Different from LnFeMnO₄, long-range 2D ferrimagnetic magnetic ordering develops upon cooling in LnFe₂O₄,^{14,24,71,72,76} although the crystal structure and the location of the magnetic moments in the structure are identical for LnFeMnO₄ and LnFe₂O₄. The disordering of Fe³⁺ and Mn²⁺ is believed to contribute to the lack of long-range magnetic ordering in the LnFeMnO₄ system.^60,61 Charge ordering exists in the LnFe₂O₄ system, where Fe²⁺(d⁶) and Fe³⁺(d⁵) are located at different layers within the Fe–O bipyramid double layer.^15,49,71 For LnFeMnO₄, Mn²⁺(d⁵) and Fe³⁺(d⁵) are randomly distributed Fe/Mn–O bipyramid double layers due to the lack of charge order. However, we must note that both Fe³⁺ and Mn²⁺ in LnFeMnO₄ have the same electron configuration d⁵ despite carrying different charges. Fe³⁺ and Mn²⁺ should possess the same magnetic dipole moment as they are in the same trigonal bipyramidal crystal field environment in the structure. The magnetic superexchange interaction thus should be the same on each Mn²⁺ or Fe³⁺ site despite charge disordering.

Figure 12 displays a qualitative analysis of the superexchange interaction in the LnFeMnO₄ system. Only the top and side views of the Fe/Mn–O bipyramid double layers are shown for simplicity. Two major superexchange pathways are indicated, J1 and J2. We write J1 between two Fe³⁺/Mn²⁺ ions that are connected through two O²⁻ anions (M-O-M angle = 97.803°). Fe³⁺/Mn²⁺ ions form the top and bottom bipyramid layers, respectively. We designate J2 between two Fe³⁺/Mn²⁺ ions within the same bipyramid layer; hence, the cations are connected by only one O²⁻ ion (M-O-M angle = 118.2°). It is worth noting that J1 exactly aligns with the (1/3, 1/3, l) lattice plane, which is where the magnetic peak appears in the neutron diffraction patterns. However, we are unclear how the charge disorder influences other magnetic interactions other than through M-O-M superexchange.

Oxidation of LnFeMnO₄ into LnFeMnO_4.5 causes the decrease of the T_f from around 60–30 K, implying a weakening of the magnetic interactions between moments. We note that there is a a-lattice parameter expansion upon oxygen uptake of LnFeMnO₄ into LnFeMnO_4.5. The weakening of the superexchange interaction along the ab-plane may contribute to the decrease of T_f upon oxidation. It is also evident that oxygen uptake greatly suppresses the short-range magnetic ordering. Figures 11(d)–11(f) present the Warren line shape fitting on the magnetic peak (located at Q ∼ 1.2 Å⁻¹) of YFeMnO_4.5, LuFeMnO_4.5, and YbFeMnO_4.5. The fitted magnetic correlation lengths are displayed in Table III. The magnetic correlation lengths of LnFeMnO_4.5 (Ln = Y, Lu, and Yb) are 1–2 nm, much shorter than that of LnFeMnO₄ (∼10 nm). We consider several factors that cause the suppression of short-range magnetic ordering upon oxygen uptake of LnFeMnO₄. First, oxygen uptake leads to Mn²⁺(d⁵) oxidation into Mn³⁺(d⁴). The disordering of Mn³⁺(d⁴) and Fe³⁺(d⁵) cause the disordering of the different magnetic dipole moments. Second, the insertion of the extra O²⁻ changes the existing magnetic superexchange interactions. The new O²⁻ sites will undoubtedly connect some magnetic ions, which disturb the original ordered superexchange interactions. Finally, the displacement of Mn/Fe sites due to strong structural modulation in the oxidized phase can also disturb the originally ordered superexchange geometry.

Interestingly, in both reduced phase YbMnFeO₄ and oxidized phase YbFeMnO_4.5, their magnetic behavior does not differ from LnMnFeO₄ and LnFeMnO_4.5 (Ln = Y and Lu), although Yb³⁺ carries magnetic moment (4.5 μ_B in octahedra field),⁷⁷ but Y³⁺ and Lu³⁺ do not. It seems the YbO octahedra layer does not influence the spin-frozen states or the nature of the short-range magnetic order. This “non-involvement” of Yb³⁺ might be the result of magnetic frustration of the YbO octahedra layer. The severe displacement of Yb³⁺ in the structure and triangular arrangement of Yb³⁺ leads to strong magnetic frustration. Indeed, the isostructural YbZnGaO₄ and YbMgGaO₄ (where only Yb³⁺ carries moment) do not display any magnetic transition even at 2 K due to the magnetic frustration in the YbO octahedra layer.^12,78

In summary, like LnFe₂O₄, LnFeMnO₄ displays an astonishing oxygen uptake behavior at low temperatures (∼200 °C). Despite the similarity of the average crystal structure of fully oxidized phases, LnFe₂O₄ and LnFeMnO₄, the anisotropic lattice expansion behavior and structural modulations show significant variations. The change of the electron configuration of the transition metal upon oxidation (d⁶ to d⁵ for LnFe₂O₄, d⁵ to d⁶ for LnFeMnO₄) seems to play an important role in those differences. We also found that the oxidation of LnFeMnO₄ greatly suppresses its magnetic ordering. Our studies expand the LnM²⁺M³⁺O₄ family to oxygen storage materials. The modifiable nature of LnM²⁺M³⁺O₄ shows its potential to be tuned for many applications involving oxygen carriers such as chemical looping, catalytic oxidation/reduction, and oxygen ion conductors. The dramatic change of the physical property after the oxidation of LnFeMnO₄ suggests it can be used for oxygen gas sensing as well.

See the supplementary material for additional characterizations of LnFe2+Fe3+O4 and LnFe2+Fe3+O4.5 (where Ln = Y, Lu, and Yb) including refinement of average structure with Synchrotron x-ray and Neutron diffraction, TGA, Pair distribution function, Raman, TEM and magnetic susceptibility measurement. Crystal structure Refinement parameters are included. Crystal structure files (cif files) are provided.

We acknowledge the NIST Cooperative Agreement Nos. 70NANB20H139 and 70NANB17H301 for support. Use of the Advanced Photon Source at Argonne National Laboratory was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357. This research used resources at Spallation Neutron Source, a DOE Office of Science User Facility operated by the Oak Ridge National Laboratory. We thank M. Kirkham at POWGEN, SNS, and ORNL for her help with the powder neutron diffraction experiment.

The authors have no conflicts to disclose.

Tianyu Li: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Sz-Chian Liou: Formal analysis (supporting); Investigation (supporting); Writing – review & editing (supporting). Stephanie J. Hong: Writing – review & editing (supporting). Qiang Zhang: Investigation (supporting); Writing – review & editing (supporting). H. Cein Mandujano: Investigation (supporting); Writing – review & editing (supporting). Efrain E. Rodriguez: Conceptualization (supporting); Funding acquisition (lead); Methodology (supporting); Resources (lead); Supervision (lead); Validation (lead); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.