WorksPhysics and Astronomy
Comparative study of magnetocaloric properties for Gd compounds with different frustrated lattice geometries
Abstract
As materials with suppressed ordering temperatures and enhanced ground state entropies, frustrated magnetic oxides are ideal candidates for cryogenic magnetocaloric refrigeration. While previous materials design has focused on tuning the magnetic moments, their interactions, and density of moments on the lattice, there has been relatively little attention to frustrated lattices. Prior theoretical work has shown that the magnetocaloric cooling rate at the saturation field is proportional to a macroscopic number of soft mode excitations that arise due to the classical ground state degeneracy. The number of these modes is directly determined by the geometry of the frustrating lattice. For corner-sharing geometries, the pyrochlore has 50% more modes than the garnet and kagome lattices, whereas the edge-sharing fcc has only a subextensive number of soft modes. Here, we study the role of soft modes in the magnetocaloric effect of four large-spin Gd (, ) Heisenberg antiferromagnets on a kagome, garnet, pyrochlore, and fcc lattice. By comparing measurements of the magnetic entropy change of these materials at fields up to ~T with predictions using mean-field theory and Monte Carlo simulations, we are able to understand the relative importance of spin correlations and quantization effects. We observe that tuning the value of the nearest neighbor coupling has a more dominant contribution to the magnetocaloric entropy change in the liquid-He cooling regime (-~K), rather than tuning the number of soft mode excitations. Our results inform future materials design in terms of dimensionality, degree of magnetic frustration, and lattice geometry.
I Introduction
The ever-increasing need for cooling in modern technologies, as well as the scarcity of helium, has motivated the search for sustainable cooling alternatives, including magnetocaloric materials [1]. The magnetocaloric effect describes the adiabatic temperature change of a magnetic material due to a change in applied field. Common early refrigerants were based on dilute paramagnetic salts such as cerous magnesium nitrate (CMN) and ferric ammonium alum (FAA) [2]. Research interests shifted later on from dilute paramagnetic salts and superparamagnets to dense magnetic lattices magnets, when it was shown that frustration can enable an enhanced magnetocaloric effect via: 1) a suppressed ordering temperature at the same magnetic density and 2) a large ground state entropy/soft magnon modes [3]. Perhaps the most well-known frustrated magnet, , exhibits a magnetic entropy difference of 13.0 J K mol in a field change from 7 T to zero field at 2 K, with cooling capabilities down to K [4, 5]. Other notable examples include the inorganic frameworks (15.3 J K mol) and (16.3 J K mol), the dense metal organic framework (16.3 J K mol), and the frustrated monazite antiferromagnet (15.6 J K mol), all again in a field change from 7 T to zero field at 2 K [4, 6, 7, 8].
In the case of Heisenberg , current magnetic research for cooling in the liquid He regime has focused on minimizing anisotropy of magnetic ions and their superexchange interactions to maximize spin polarizability [4, 7, 9]. However, using the subtle effects of magnetic frustration to enhance the magnetocaloric effect and to suppress the magnetic ordering temperature remains a relatively unexplored area of research [10, 11]. In frustrated magnetic oxides, the main tuning parameters are the magnetic ions and the atomic lattice. In this way, the superexchange and dipolar interactions, and the crystal electric field (CEF) environment of individual ions are manipulated.
In the case where dipolar and CEF contributions are negligible, prior theoretical work has shown that the lattice geometry can dictate an enhancement of the magnetocaloric cooling rate, , that scales with the macroscopic number of soft modes at the saturation field [3]. The soft modes are ultimately a result of the macroscopic ground state entropy that frustration introduces. These modes have been modeled for the three corner-sharing geometries shown in Figure 1: the pyrochlore lattice, with the number of pyrochlore modes scaling as , and the garnet and kagome lattices with soft modes, where is the number of lattice sites [3]. Figure 1 also shows an edge-sharing geometry: the lattice, for which the number of soft modes has not been reported.
Figure 1. Frustrated lattice geometries in 2D (kagome) and 3D (garnet, pyrochlore, ).
In this work, we seek to test experimentally the role of these fundamental soft modes in the measured magnetocaloric effect in four representative frustrated oxide materials in the liquid He regime (2-20 K). Of particular interest is answering the question of the optimal frustrating geometry to inform future magnetocaloric material design. Four Gd-based oxides were chosen as model systems since the contribution of crystal electric field effects to the magnetism are negligible ( and hence and ). The four compounds investigated include GdMgSbO, in which the Gd ions lie on kagome layers separated by triangular layers of nonmagnetic Mg ions [12]; GdGaO, in which Gd ions lie on two interpenetrating rings of triangles [13]; GdSnO, in which the Gd ions form a pyrochlore lattice (corner-sharing tetrahedra); and BaGdSbO, an lattice of Gd ions (edge-sharing tetrahedra).
We find that, upon normalizing for the differences in superexchange across the materials, the normalized magnetic entropy change in this temperature range qualitatively scales with the number of soft modes for the three-dimensional lattices as predicted in Ref. 3. However, we find that the lattice exhibits a 30-95% greater magnetic entropy change than the corner-sharing lattices, which can be attributed to its paramagnetic response [14]. We compare the experimental results with Monte Carlo simulations of classical Heisenberg spins as well as with mean-field superexchange calculations that account for the quantised nature of the
spins. These allow us to show that, in the temperature and field regimes of interest in this study, the effect of correlations is dominant over quantization in corner sharing compounds, whereas the opposite is true for the case. The fact that the magnetocaloric cooling rate of the corner-sharing geometries cannot be fully described using a mean-field superexchange model is consistent with a significant contribution due to soft modes to the total magnetic entropy change measured.
For the compounds studied, we find that the paramagnetic contribution to the magnetic entropy outweighs that of geometric frustration. In particular, the three-fold reduction in the superexchange of the garnet compound ( mK [14]) results in a 50% improvement in the magnetic entropy change extracted compared to the pyrochlore lattice. These results suggest that compounds with the smallest superexchange and which also make use of magnetic frustration are promising future magnetocaloric materials.
Soft Modes Model: Validity in Liquid He Regime
In prior work [3], one of us predicted a strongly enhanced magnetocaloric effect for geometrically frustrated magnets with high classical degeneracy in the ground state. The work also included classical Monte Carlo simulations, generally valid for large- magnetic materials, for Heisenberg antiferromagnets with three typical frustrated geometries based on corner-sharing plaquettes: the two-dimensional kagome lattice (a network of corner-sharing triangles), the garnet lattice (its three-dimensional analog), and the pyrochlore lattice (a network of corner-sharing tetrahedra), see Figure 1. Here we revisit these geometries and compare them with an lattice.
The spin Hamiltonian consists of a nearest neighbor () superexchange term, assumed to be uniform across all pairs, and a Zeeman term:
where is the superexchange and a factor has been absorbed into the definition of the field for convenience. Dipolar interactions are also present in the system but we shall disregard them in our simulations. Considering the system parameters reported later in Table II, we expect this to be a reasonable approximation for the three corner sharing lattices. In the system, superexchange and dipolar interactions at nearest-neighbour distance are of similar strength; however, they are much smaller than the temperature regime of interest in our study and therefore truncating them to nearest-neighbour distance is a valid approximation.
In the low-temperature cooperative paramagnetic state, a condensation of a macroscopic number of soft modes at the saturation field is predicted to occur,
| Compound | Lattice | (T) | (T) | |
| GdSnO | Pyrochlore | 5.8(2) | ||
| GdMgSbO | Kagome | 4.6(2) | ||
| GdGaO | Garnet | 2.0(2) | ||
| BaGdSbO | 1.4(2) |
Table 1. Predicted number of soft modes for a lattice with sites of Heisenberg spins and predicted saturation field [3] versus measured saturation field for GdGaO, GdMgSbO, GdSnO, and BaGdSbO. The observed saturation field, , was determined as the field at which the measured temperature gradient of the magnetization is maximized at K. We estimate the expected scaling for the number of soft modes for BaGdSbO based on the magnon spectrum at derived in Ref. 16. Error bars listed for the measured saturation field are based on the 0.2 T step size for the measurements.
due to the enhanced ground state entropy afforded by the underconstraint of frustration [3]. The saturation field, defined as the threshold beyond which the energy becomes dominated by the Zeeman term, has been predicted [3] to be for the kagome and garnet lattices, and for the pyrochlore lattice, see Table I. Using the same frustrated block decomposition [15], we obtain here the saturation field for the nearest neighbor antiferromagnet (see Appendix § 3.1).
The magnetocaloric rate, , at the saturation field is predicted to follow the scaling relation:
This is in addition to the conventional contribution to from the paramagnetic (non-frustrated) ordinary dispersive modes () of the system,
where the excitation energies are nonnegative and vanish for corresponding to the propagation vectors of degenerate classical ground states in zero field [3].
For non-frustrated, three-dimensional Heisenberg magnets above the ordering temperature, is the only contribution and there is no enhancement in the magnetocaloric cooling rate, which is a temperature-independent constant for [3]. In frustrated systems where is macroscopic (i.e., , where is the number of lattice sites), one expects to dominate at low enough temperatures (where it grows as ). However, when with , as is the case for BaGdSbO (discussed below), there will always be a system size beyond which dominates, for any given temperature: .
Using classical spin-wave calculations in the saturated (collinear) state, the number of soft modes in the pyrochlore lattice was found to scale with the number of lattice sites, , Table I [3]. The kagome and garnet lattices are predicted to have as many soft modes as the pyrochlore. Thus, it could be expected that the pyrochlore lattice may have a greater magnetocaloric effect than a comparable garnet or kagome lattice. Indeed prior work has shown that the pyrochlore GdTiO exhibits an increased cooling rate compared to the garnet GdGaO, but the work does not address the overall magnetic entropy change (and hence total cooling) available from each material nor their differing superexchange [17].
The number of soft modes for an lattice has not been reported, but is expected to be lower than that for the corner-sharing geometries, as its zero energy modes correspond to lines (rather than surfaces) in the Brillouin zone [18]. The appropriate scaling for the number of soft modes can be predicted from the magnon spectrum at the saturation field reported in Ref. 16:
The zeros of this equation correspond to soft modes and are given by and the “cubic related lines” [16]. Along a given dimension of a material with lattice sites, there are such states for which this is the case. Thus, the lattice has a subextensive number of soft modes compared to the corner-sharing lattices.
The soft modes theory described here is valid for classical (large-spin) Heisenberg magnets [3]. The lower limit of this temperature regime, , is approximated by , below which quantum statistics must be used to analyze spin excitations [19]. Among the compounds studied, the superexchange is largest for the pyrochlore and kagome compounds, K in Table II, corresponding to a limiting temperature of K. Thus the liquid-He temperature range investigated here, 2–20 K, should provide reasonable insight into the role of soft modes on the magnetocaloric effect, as classical statistics of magnon modes applies at these temperatures.
Experimental Results
Magnetic Characterization
The measured magnetic susceptibilities of GdMgSbO, GdSnO, GdGaO, and BaGdSbO from 1.8 K to 300 K are depicted in Figure 2. Details of sample preparation and structural characterization can be found in Appendix § 12. The negative Curie-Weiss temperatures, see Table II, confirm the antiferromagnetic superexchange required for frustration and are consistent with previous literature reports [20, 21, 22, 23]. Rearranging the Curie-Weiss law into the dimensionless form:

Figure 2. a) Measured static magnetic susceptibility of Gd-based kagome (GdMgSbO), pyrochlore (GdSnO), garnet (GdGaO), and fcc (BaGdSbO) compounds versus temperature (inset: inverse magnetic susceptibility). b) Dimensionless inverse magnetic susceptibility versus dimensionless temperature (inset: zoomed-in region near zero temperature). Both axes are scaled using the appropriate factors of the Curie-Weiss temperature and Curie constant . All compounds exhibit positive deviations from paramagnetic behavior (black line) indicative of possible antiferromagnetic short-range correlations. The fcc lattice is effectively paramagnetic with deviations of less than 5% from Curie-Weiss behavior [14]. All measurements were made over the same temperature range, 1.8 – 300 K, but exhibit different dimensionless temperature ranges due to different .
where is the Curie temperature and is the Curie constant, we can compare the strength of magnetic short-range correlations between spins across different compounds [24]. Positive deviations from the Curie-Weiss law indicate antiferromagnetic short-range correlations while negative deviations signify ferromagnetic correlations.
tributions. As shown in Figure 2b), all corner-sharing geometries exhibit antiferromagnetic short-range correlations, while the fcc compound is qualitatively paramagnetic down to K, consistent with the literature [13, 14, 21, 25, 26, 27]. A mean-field estimate for the superexchange can be computed from the Curie-Weiss law using:
where is the number of .
Table II lists the reported values for the superexchange and dipolar interaction for each compound which are in agreement with the values obtained from the Curie-Weiss fits of the measured magnetic susceptibility in Figure 2. Notably, is of similar magnitude for the kagome and pyrochlore compounds ( K), and about 3 and 30 times smaller for the garnet and the fcc compounds, respectively. The smaller magnitude of for the fcc lattice is consistent with the minimal antiferromagnetic deviations in the dimensionless susceptibility. The dipolar interaction in the kagome, garnet, and pyrochlore lattices is around 50 mK, and much smaller in the fcc lattice, mK, due to the larger distance between nearest neighbors. The pyrochlore and kagome compounds have the lowest ratio, around 0.2, while the garnet has and the fcc lattice has . Hence the measured coupling from Curie Weiss for BaGdSbO may contain contributions from dipolar interactions. For all subsequent analysis, we consider BaGdSbO to be a weakly interacting antiferromagnet as discussed in § 4.
The isothermal magnetization was measured at 2 K, see Figure 3. Within a field of 9 T, all compounds
Figure 3. Isothermal magnetization at 2 K versus applied field of Gd-based kagome (GdMgSbO), pyrochlore (GdSnO), garnet (GdGaO), and fcc (BaGdSbO) compounds. The black solid line gives the theoretically predicted behaviour for uncoupled Heisenberg spins with .
TABLE II. Estimated superexchange and dipolar interaction for select Gd-based frustrated magnetocaloric materials. The superexchange constant was determined from Curie-Weiss fits of the measured ZFC (1000 Oe) magnetic susceptibility from 8–50 K. Comparisons to the literature-reported values for each compound are also provided. and refer to the superexchange and dipolar term reported for the pyrochlore GdSnO [22, 26]. The lowest temperature of validity of the classical soft modes theory discussed in the main text is .
| Kagome | Pyrochlore | Garnet | |||||||||
| GdMgSbO | GdSnO | GdGaO | BaGdSbO | ||||||||
| () | () | () | () | ||||||||
| Ref. [20, 21] | Fit | Ref. [22, 26] | Fit | Ref. [23, 27] | Fit | Fit | |||||
| (K) | -6(1), -6.70 | -6.8(1) | -9.6(1) | -8.6(1) | -2.6(1) | -2.30(1) | -0.78(1) | ||||
| (K) | 0.3, 0.32 | 0.324(5) | 0.3 | 0.273(3) | 0.107 | 0.110(1) | 0.0124(2) | ||||
| 1 | 1.1 | 1 | 0.9 | 0.4 | 0.4 | 0.04 | |||||
| (K) | 0.0502 | 0.0496 | 0.0457 | 0.0116 | |||||||
| 1 | 1 | 0.9 | 0.2 | ||||||||
| 0.16–0.17 | 0.17 | 0.43 | 0.94 | ||||||||
| (K) | 1 | 1 | 0.4 | 0.04 |
Table 2.
reach a maximum value around the /Gd predicted for uncoupled Gd Heisenberg spins. The compound exhibits an isothermal magnetization that agrees well with the uncoupled Heisenberg spins prediction, owing to its small, mK superexchange and dipolar interactions, as shown in our prior work [14].
The saturation field of each compound was estimated from the field at which the temperature gradient of the magnetization, shown in Figure 4, is maximized [3, 17]. This method of estimating the saturation field should be reliable in the temperature range where spin correlations are relevant, i.e. for . Theoretical predictions for (at zero temperature) based on the large- Heisenberg Hamiltonian, Equation (1), are listed in Table I. The observed values for the corner-sharing geometries agree well with the predictions, indicating that the compounds are well described by the Heisenberg model and suggests that for K, superexchange plays a key role in the magnetocaloric cooling rate, . As described in § 4, the compound BaGdSbO can best be described as a weakly frustrated Heisenberg antiferromagnet. At 2 K, its magnetocaloric cooling rate can be modeled well using a mean-field superexchange model, in which spatial spin correlations are altogether neglected. Due to its small superexchange, K, lower temperature measurements are needed to measure the saturation field with our experimental protocol. This is beyond the scope of the present work and, for completeness we report here the estimate of the saturation field at K, the lowest temperature measured.
Figure 4. Approximate magnetocaloric cooling rate versus field for BaGdSbO, GdGaO, GdMgSbO, and GdSnO at 2 K. These were obtained from experimental measurements of isothermal magnetization with K. The resulting value of the saturation field for each material is indicated by the arrows.
Magnetocaloric Effect
From one of Maxwell’s relations, the isothermal field gradient of the magnetic entropy is related to the temperature gradient of the magnetization at constant field via . Thus, the magnetic entropy change,
can be measured using the isothermal magnetization via:
The magnetic entropy change of each Gd-based compound was measured from 2 to 20 K in fields up to 9 T from the isothermal magnetization, Figure 5 and Figure 6a. We note that the magnetization data were measured in temperature steps of 2 K, and thus we are only able to measure the finite difference ratio as an estimate for the magnetocaloric cooling rate , prior to computing the magnetic entropy change via Equation (7) (see Appendix § 10).

Figure 5. Measured temperature gradient of the magnetization (with K) and resulting magnetic entropy change , normalized by the maximum free-spin value, , from 2–20 K under applied fields of 0–9 T for the Gd-based frustrated magnetocaloric materials discussed in this work.
Figure 6. a) Magnetic entropy change per mol versus temperature at fields of 7 T and 2 T (inset: versus field at 2 K). b) Superexchange-normalized magnetic entropy change versus temperature at fields of 7 T and 2 T.
Per mol Gd, the maximum entropy change attainable for a paramagnetic salt is (17.28 J K mol). We find that the fcc BaGdSbO exhibits the greatest magnetic entropy change , reaching in a field of just 7 T at 2 K. On the other hand at 2 K and a larger field of 9 T, the corner-sharing geometries each exhibit entropy changes of 0.8, 0.7, and for the garnet, kagome, and pyrochlore lattices, respectively. At a low field of 2 T, the fcc and garnet compounds are still the best performing with of and , respectively, compared to for the kagome and pyrochlore, Figure 6a. Despite having the largest number of soft modes, the pyrochlore material achieves the smallest magnetic entropy change per Gd ion.
Discussion
The enhanced performance of the fcc BaGdSbO and garnet GdGaO compared to the pyrochlore GdSnO is at first surprising, given the reduced number of soft modes (Table I). However, from Equations (2) and (3) we see that the measured magnetocaloric cooling rate results from two different contributions, one due to the number of soft modes (frustrated response) and one due to ordinary dispersive modes (paramagnetic response). In order to investigate this in greater detail, we compare the experimental data with mean-field modelling [14] and classical Heisenberg Monte Carlo simulations of the finite difference ratio for each of the four systems (see Figure 7). For reference, we show in Figure 9 in the Appendix a comparison between the finite difference ratio and the exact magnetocaloric effect , for both mean-field and Monte Carlo results.
Figure 7. Approximate magnetocaloric cooling rate extracted from the measured isothermal magnetization with K (data points), compared to predictions based on the magnetization for Heisenberg spins in the mean-field superexchange model (MFT) (solid lines) and classical Monte Carlo simulations (MC) (dotted lines). These results show that the fcc compound is well described by mean-field quantum spins, which neglect spatial correlations. Conversely, the better agreement of classical Monte Carlo for the corner-sharing geometries highlights the importance of magnetic correlations in those compounds.
Figure 9. Comparison of the exact, , vs approximate, with K, magnetocaloric cooling rates for the four frustrating geometries, using classical Monte Carlo (MC) (blue) and mean-field theory (MFT) (red) at K. Exact magnetocaloric cooling rates are shown as solid lines while approximate rates are shown as dotted lines.
Our results show that the fcc compound is described quantitatively well by the mean-field superexchange model [14] across the full range of temperatures of interest in this study 4. This finding suggests that the paramagnetic response dominates for this compound in the temperature and field regime of interest. We recall indeed that the fcc lattice is predicted to only have a subextensive number of soft modes, and therefore the contribution at the saturation field becomes irrelevant in the thermodynamic limit. The paramagnetic term is expected to be the dominant contribution to the measured magnetic entropy change .
In contrast, for the other three (corner-sharing) compounds we clearly see an increasing discrepancy between mean-field and experiment at lower temperatures (2 and 4 K). This is consistent with the 10-30-fold larger values of in these systems, resulting in correspondingly stronger correlations. The latter are generally expected to have two effects: (i) reduce the contribution to the magnetocaloric effect (the spins have reduced ability to fluctuate independently of one another); and (ii) give rise to an extensive number of frustrated collective soft mode , which contribute with a prefactor scaling as to . The comparatively better agreement of the experimental curves with classical Monte Carlo simulations, with respect to mean-field, demonstrates how the latter are able to capture both the paramagnetic as well as the soft modes contribution, with corrections at large fields due to quantization effects 5.
Having established the importance of soft modes in the magnetocaloric performance of the corner-sharing geometries, it remains to be seen why the garnet outperforms both the pyrochlore and the kagome lattices. This can be explained by the fact that the measured entropy changes in Figure 5 and Figure 6a do not account for differences in the value of the superexchange coupling . To account for the role of superexchange in the predicted enhancement at the saturation field, Equation (2), the magnetic entropy maps were scaled by , Figure 6b and Figure 8. While the soft modes contribution to the magnetic entropy change is expected to vanish for the fcc lattice due to its subextensive , we include the BaGdSbO compound for ease of comparison.
Figure 8. Measured magnetic entropy change normalized by the superexchange, . Accounting for the differences in the superexchange term, the pyrochlore lattice becomes the highest performing among the 3D frustrated lattices.
After normalization, the pyrochlore lattice becomes the top-performer among the 3D geometries at 7 T, as predicted by Ref. 3, while the kagome sample is the top-performer overall.
The scaled magnetic entropy change, , agrees qualitatively with the soft modes model. From Equation (2), the maximum in at the saturation field should scale proportionally to the number of soft modes , once normalized by the square root of the superexchange . This qualitative agreement is striking as it reproduces the predicted ranking for the three-dimensional frustrated lattices of the pyrochlore (), followed by the garnet (), and then the fcc () for fields greater than or equal to saturation (i.e., T). At the low field of 2 T, the kagome and pyrochlore compounds are not fully saturated, explaining why, as might be expected, the garnet and fcc lattices exhibit the greatest , Figure 6b. A more quantitative assessment of the agreement with the model at lower fields is not possible, likely due to other minor contributions (e.g., dipole-dipole interactions and disorder), which were not accounted for in our study.
The implication of this experimental validation is that while the order of soft modes of frustrating lattices can be accurately captured in experiment, soft modes are not the leading contribution in the magnetocaloric performance of a frustrated magnet in the liquid He regime. Rather, as demonstrated in Figure 5 and Figure 6, materials with the smallest superexchange (i.e., the fcc and garnet) exhibit the greatest performance, as the paramagnetic contribution to the entropy change is not constrained by short-range correlations so that spins can fluctuate independently of one another. It is conceivable that at lower temperatures relative to the superexchange ( K), soft modes may become the determining factor in the magnetocaloric effect. However, such low temperatures would soon require a revision of the model to include quantum statistics to study spin excitations.
These results suggest that future research for cryogenic magnetic refrigeration should focus on frustrated magnets with the smallest superexchange. The strategy of solely reducing the exchange by increasing the distance between magnetic ions, as in the dilute magnetic salts FAA and CMN, is not effective. When the four materials studied here are normalized by volume and mass, see Appendix E, then the fcc lattice is the poorest performer despite having the weakest superexchange coupling. This highlights the importance of reducing while maintaining a dense magnetic lattice [14]. Furthermore, in these dense lattices, soft modes could provide an additional cooling mechanism at lower temperatures.
Another interesting result from this study is that, in the limit where superexchange determines the magnetocaloric cooling rate, differences between the pyrochlore and kagome lattices are not predicted by the model. In this case, they may arise due to the presence of site disorder (10.5(2)% in GdMgSbO (Appendix F) relieving frustration, or spin-anisotropies not included in this analysis, which only applies to isotropic Heisenberg spins.
This study focuses on four frustrated oxides. However, there are many other magnetic lattices where a high magnetocaloric effect is reported including the 2D triangular (GdBO [28]) and Shastry-Sutherland lattices (GdBeGeO [29]), and quasi-1D spin chains (Gd(HCOO), GdOHCO, CaGdO(BO) [7, 30, 31]). In many of these systems the presence of polyanions (BO, HCOO, etc.) or molecular ligands may complicate the soft mode analysis presented in Reference 3 due to changes in the hydrogen bonding or other molecular interactions [32, 33]. Such materials are beyond the scope of this initial study. Future work could incorporate the soft mode analysis of other geometrically frustrated magnetic lattices, e.g. edge-sharing motifs such as the edge-sharing triangles in a honeycomb arrangement in SrLnO [34].
The mean-field nature of the analysis conducted here is only applicable to Heisenberg spins. Anisotropic magnetic systems such as those based on Dy have been found as useful magnetocaloric candidates at low fields (e.g., T), compared to the Gd counterparts. For example, the magnetic entropy change of Ising garnet DyGaO reaches four times that in the Heisenberg GaGaO for a 1 T field at 2 K [35]. Adiabatic temperature measurements have shown that the Ising spin ice DyTiO can cool from K to 0.3 K in a field of just 0.8 T, while GdTiO requires a field of T to cool to 0.5 K from the same starting temperature [36, 37]. Extending this analysis to anisotropic systems is an interesting avenue for future research, likely requiring non-trivial microscopic modeling of the single-ion anisotropy.
V Conclusion
This work examined the role of lattice geometry in maximizing the magnetocaloric effect via soft mode spin excitations. Four representative Heisenberg (3D, Gd-based) spin systems were investigated including the three corner-sharing garnet (GdGaO), pyrochlore (GdSnO), kagome lattices (GdMgSbO), and the edge-sharing lattice (BaGdSbO). For the liquid-He temperature range investigated (2 − 20 K), magnetic entropy change measurements indicate that the smaller superexchange of the and garnet lattices allows for better magnetocaloric performance than the strongly-coupled pyrochlore analog, despite its larger number of soft modes. Our results show that a paramagnetic response dominates above geometric frustration in enhancing the magnetocaloric effect. However, the contribution to the magnetocaloric effect in systems with an extensive number of frustrated magnetic soft modes is expected to take over at lower temperatures, thus providing an additional channel for maximizing the magnetocaloric effect in highly frustrated magnetic lattices. Future magnetocaloric materials design should focus on both the superexchange and magnetic lattice to optimize the trade off between spin polarizability and frustrated soft mode enhancement.
Acknowledgments
For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising from this submission. This work was supported supported by the Engineering and Physical Sciences Research Council (EPSRC) grants (EP/P034616/1, EP/V062654/1 and EP/T028580/1) and the Winton Programme for the Physics of Sustainability. E.C.K. gratefully acknowledges the support of a Churchill Scholarship from the Winton Churchill Foundation of the United States. M.E.Z. acknowledges financial support from Agence Nationale de la Recherche, France, Grant No. ANR-18-CE05-0023. Magnetic measurements were made on the EPSRC Advanced Characterization Suite EP/M0005/24/1.
Saturation field of the model
One can use frustrated block decomposition to obtain the saturation field for the nearest neighbor antiferromagnet, above which spins become fully aligned with the applied field. We start with the spin Hamiltonian
where the prefactor in the Zeeman term has been absorbed into . An lattice can be represented as an edge-sharing arrangement of tetrahedra, which are labeled by an index . Introducing the tetrahedron magnetization , and using the fact that every site belongs to 8 tetrahedra, it can be shown that:
Similarly, since every bond of the lattice belongs to 2 tetrahedra, we find that:
The Hamiltonian (A1) can be now rewritten, up to an irrelevant constant, as:
For a single tetrahedron, the energy is minimized when . If this condition can be met for all tetrahedra simultaneously, then it is the lowest energy state of the system and its zero-temperature state.
Because , is upper bounded by the maximum possible value , if then the minimum energy condition cannot be achieved and the lowest energy state corresponds the uniformly polarized spin configuration. This value is referred to as the saturation field,
Mean-field Approximation
A mean-field approximation (random phase approximation) [14, 21] was used to estimate the paramagnetic contribution to the magnetocaloric cooling rate for each compound, truncating the interactions at nearest neighbour distance. The antiferromagnetic coupling between spins amounts to an exchange field, , which adds to the external field, , to produce the net field, , experienced by a single spin.
Note that dipolar interactions can be neglected for the corner sharing lattice compounds considered in this study since they are weak in comparison to the nearest neighbor exchange (see Table II). This is not the case for the material. However, the dipolar interactions contribution for a cubic Bravais lattice vanishes identically in the case of a (uniform) mean-field approximation, due to rotational lattice symmetry.
Since for , the exchange constant can be assumed to be isotropic, so that the exchange field is given by:
where is the bulk magnetization in units of the Bohr magneton and is the “field parameter” in units of magnetic field [21]. The bulk magnetization of the system at a given temperature and external field is given by the solutions of the transcendental self-consistency equation:
where is the Brillouin function given by:
with , and [21].
As a further consistency check we leave as a free parameter in our model. Global least-squares fits to the measured isothermal magnetization (2 − 20 K) using Equation (14) were used to determine the superexchange in , , and . The fit values, Table III, are in agreement with the literature reported values and Curie-Weiss fits, Table II. The free-spin magnetization was used as an initial guess to solve iteratively the mean-field self-consistency condition, Equation (14). The observed saturation value of the magnetization at the maximum field can vary due to experimental uncertainty, so all compounds were fit with a scaled fraction of , to match the observed value, see Table III.
| Compound | () | Fit (K) | |
| 1.0268 | 0.312(3) | 0.9982 | |
| 0.9744 | 0.273(2) | 0.9976 | |
| 0.9257 | 0.114(1) | 0.9997 | |
| 1.04 | 0.0113(3) | 1.0000 |
Table III. Mean-field saturation magnetization parameter determined by fitting the observed saturation magnetization, as well as the superexchange , for each compound, and of the fit.
The approximate magnetocaloric cooling rate was then calculated from the model predictions of using the fit superexchange constants , Equation (D1), at the same resolution as the experimental data (i.e., K). In the limit that , one recovers the exact magnetocaloric cooling rate, . A comparison between the exact and approximate magnetocaloric cooling rate obtained from mean-field is shown in Figure 9, for all 4 materials (computed via a converged numerical derivative with K).
Monte Carlo Simulations
The classical Monte Carlo (MC) simulations have been performed on periodic lattices with 2000–4000 spins. Since in the studied temperature range all materials remain in the paramagnetic state, the simulated system sizes are sufficient to obtain the bulk behavior without
additional finite-size scaling. We simulate the nearest neighbor antiferromagnetic model
where the spins are classical vectors of unit length, . A hybrid Monte Carlo algorithm has been employed with canonical Metropolis sweeps over the lattice followed by microcanonical over-relaxation steps. At each temperature/field point, Metropolis steps were used for equilibration with subsequent hybrid MC steps for measurements. In addition, we performed averaging over independent MC runs. Relative statistical errors do not exceed 0.5% for all obtained quantities. The approximate magnetocaloric cooling rate was computed from the magnetization
using the finite difference method, Equation (D1), at the resolution of the experimental data (i.e., temperature steps of 2 K). The exact magnetocaloric cooling rate was computed from the energy-magnetization cumulant:
where denotes statistical averaging. Figure 9 shows a comparison between the exact and approximate values, for reference. The figure also allows for a comparison between the classical Monte Carlo results and the mean-field modelling in Appendix § 3.2; the enhanced magnetocaloric effect in the latter is likely due to the quantised nature of the spins.
Rescaling between dimensionless values obtained from the classical Monte Carlo simulations (C1) and real physical properties of each material was done using the conversion factors:
Here, and are the spin and the factor of Gd ions, is the nearest neighbor exchange, and are the Boltzmann constant and Avogadro’s number.
Experimental Methods
Sample preparation
Polycrystalline samples were prepared using a solid-state method consistent with prior reports in the literature [14, 20, 23, 26]. Reactants oxides were ground using a mortar and pestle and then heated in air in an alumina crucible at for several days with intermittent grindings to ensure a complete reaction. As described in Appendix F, x-ray diffraction indicates phase pure samples ( weight % impurities of GdSbO in BaGdSbO) with crystal structures consistent with the literature reported values.
X-ray diffraction and crystal structure refinements
Room temperature powder x-ray diffraction (XRD) measurements were conducted using a Bruker D8 Advance diffractometer (Cu K radiation, Å). Data was collected with from , with an overall collection time of 2-3 hours. Rietveld refinements [38] of the powder XRD data were completed using the Diffrac.Suite TOPAS5 program [39]. The background was modeled using a 13-term Chebyshev polynomial and peak shapes were fit using a pseudo-Voigt function [40]. All Debye-Waller factors were set constant to the literature reported values.
Bulk magnetic measurements
Bulk magnetic measurements of the magnetic susceptibility were conducted using a Quantum Design Magnetic Properties Measurement System (MPMS) with a superconducting interference device (SQUID). Susceptibility measurements were made at T, in the low-field limit where in zero-field-cooled conditions from 1.8 K to 300 K. measurements (described below) are linear at this field, confirming that this linear approximation of is valid. Isothermal magnetization was measured using the ACMS option of a Quantum Design Physical Properties Measurement System (PPMS) for GdMgSbO, GdGaO, and GdSnO and a Quantum Design MPMS SQUID for BaGdSbO. measurements were made at temperatures of 2 K to 22 K in steps of 2 K over a field range of 0 T to 9 T in steps of 0.2 T (only up to 7 T for BaGdSbO).
Magnetocaloric effect calculations
The magnetic entropy change for a field relative to zero field was calculated from the measured by first computing the temperature derivative of the magnetization using the finite differences approximation:
and then integrating across fields as in Equation (7). The magnetization data were linearly interpolated along the field direction in steps of 0.1 T prior to extracting the magnetic entropy.
Magnetocaloric performance per mass and volume
The measured magnetocaloric effect, Table IV, at T = 2 K for a field change of 9 T to 0 T for GdMgSbO (kagome), GdSnO (pyrochlore), and GdGaO (garnet), and for field change of 7 T to 0 T for BaGdSbO (), normalized by volume and by mass.
Structural refinements of Gd compounds
Rietveld analysis of the XRD shows that the refined structures are consistent with those reported in the literature [20, 21, 23, 25, 26]. Recently, the family of kagome compounds MgSbO has been found to exhibit sample-dependent cation site disorder, in which the ion on the Wyckoff site of a kagome layer swaps places with a Mg ion on the interlayer site (see Figure 11) [42, 43]. We find a 10.5(2)% cation site-disordering in the GdMgSbO sample in this report. A comparable amount of site disorder has been found in other samples of MgSbO including the emergent-charge-ordered kagome Ising magnet DyMgSbO [42] and the dipolar kagome ice HoMgSbO [43].

Figure 11. Refined crystal structure of sample. The compound exhibits some site disorder, with 10.5(2)% of ions from the site lying on the kagome site, Table V. ions and corresponding O-polyhedra are shown in purple, ions in orange, ions in gray, and ions in red. This figure was generated from the refined crystal structure file using VESTA [44].
| (Å) | 7.3634(1) |
| (Å) | 17.4511(3) |
| Mg(1)/Gd(2) (3a) | Occupancy 0.686(6), 0.314(6) |
| Gd(1)/Mg(3) (9d) | Occupancy 0.895(2), 0.105(2) |
| O(1) | 0.113(1) |
| O(2) | 0.5261(7) |
| O(2) | 0.8974(6) |
| O(3) | 0.4714(8) |
| O(3) | 0.3591(5) |
| Overall (Å) | 0.76(2) |
| (wt %) | 0.83(4) |
| 9.69 | |
| 1.42 |
Table V. Structural parameters of . PXRD refinements were carried out in the space group , with Mg(1) on the sites (0, 0, 0), Mg(2) on the sites (0, 0, ), Gd(1) on the sites (, 0, ), Sb on the sites (, 0, 0), and O(1) on the sites (0, 0, ), and O(2) and O(3) on the sites (). Partial occupancy between the and sites was also refined according to the formula as in Reference 42.
| Compound | (J K mol) | Molar Mass (g mol) | (g cm) | (mJ K cm) | (mJ K kg) |
| 1109.63 | 6.75 | 220 | 33 | ||
| 663.92 | 7.72 | 250 | 32 | ||
| 1012.36 | 7.09 | 280 | 39 | ||
| 649.66 | 7.09 | 170 | 24 |
Table IV. Measured magnetocaloric effect at K and T for the corner-sharing geometries, and T for the fcc lattice, normalized by volume and by mass. Mass densities were determined from Rietveld refinement for the kagome, garnet, and fcc lattices (see Appendix § 12) and from the literature reported value for the pyrochlore [41].
Figure 10. Rietveld refinement of from PXRD.
Figure 12. Rietveld refinement of from PXRD.
Figure 13. Rietveld refinement of from PXRD.
Figure 14. Rietveld refinement of from PXRD.
References
References
- [1]X. Moya and N. D. Mathur, Caloric materials for cooling and heating, Science 370, 797 (2020).DOI
- [2]P. Wikus, E. Canavan, S. T. Heine, K. Matsumoto, and T. Numazawa, Magnetocaloric materials and the optimization of cooling power density, Cryogenics 62, 150 (2014).
- [3]M. Zhitomirsky, Enhanced magnetocaloric effect in frustrated magnets, Phys. Rev. B 67, 104421 (2003).DOI
- [4]Y.-C. Chen, J. Prokleška, W.-J. Xu, J.-L. Liu, J. Liu, W.-X. Zhang, J.-H. Jia, V. S. Jakšć, and M.-L. Tong, A brilliant cryogenic magnetic coolant: magnetic and magnetocaloric study of ferromagnetically coupled GdF₃, Journal of Materials Chemistry C 3, 12206 (2015).DOI
- [5]D. A. P. Brasiliano, Etude et réalisation d’une ADR spatiale 4 K - 50 mK. Instrumentation et méthodes pour l’astrophysique., Phd thesis, Université Grenoble Alpes (2017).DOI
- [6]Gd(OH)₂: A promising cryogenic magnetic refrigerant, Journal of the American Chemical Society 144, 13787 (2022).DOI
- [7]G. Lorusso, J. Sharples, E. Palacios, O. Roubeau, E. Brechin, R. Sessoli, A. Rossin, F. Tuna, E. McInnes, D. Collison, and M. Evangelisti, A dense metal–organic framework for enhanced magnetic refrigeration, Advanced Materials 25, 4653 (2013).DOI
- [8]E. Palacios, J. A. Rodríguez-Velamazán, M. Evangelisti, G. J. McIntyre, G. Lorusso, D. Visser, L. J. de Jongh, and L. A. Boatner, Magnetic structure and magnetocalorics of GdPO₄, Phys. Rev. B 90, 214423 (2014).DOI
- [9]M. S. Reis, Magnetocaloric and barocaloric effects of metal complexes for solid state cooling: Review, trends and perspectives, Coordination Chemistry Reviews 417, 213357 (2020).DOI
- [10]P. Mukherjee, Investigation of the magnetic and magnetocaloric properties of complex lanthanide oxides, Ph.D. thesis, University of Cambridge, Cambridge, Cambridgeshire, UK (2018), publisher: Apollo - University of Cambridge Repository.DOI
- [11]D. A. Paixao Brasiliano, J.-M. Duval, C. Marin, E. Bichaud, J.-P. Brison, M. Zhitomirsky, and N. Luchier, YbGG material for adiabatic demagnetization in the 100 mK–3 K range, Cryogenics 105, 103002 (2020).DOI
- [12]Z. L. Dun, J. Trinh, K. Li, M. Lee, K. W. Chen, R. Baumbach, Y. F. Hu, Y. X. Wang, E. S. Choi, B. S. Shastry, A. P. Ramirez, and H. D. Zhou, Magnetic ground states of the rare-earth tripod kagome lattice Mg₂RESb₃O₁₄, Phys. Rev. Lett. 116, 157201 (2016).DOI
- [13]O. A. Petrenko and D. McK. Paul, Classical heisenberg antiferromagnet on a garnet lattice: A monte carlo simulation, Phys. Rev. B 63, 024409 (2000).DOI
- [14]E. C. Koskelo, C. Liu, P. Mukherjee, N. D. Kelly, and S. E. Dutton, Free-spin dominated magnetocaloric effect in dense Gd³⁺ double perovskites, Chemistry of Materials 34, 3440 (2022).
- [15]M. Zhitomirsky, Real-space perturbation theory for frustrated magnets: application to magnetization plateaus, in Journal of Physics: Conference Series, Vol. 592 (IOP Publishing, 2015) p. 012110.DOI
- [16]G. Jackeli and M. E. Zhitomirsky, Frustrated Antiferromagnets at High Fields: Bose-Einstein Condensation in Degenerate Spectra, Physical Review Letters 93, 017201 (2004).DOI
- [17]S. S. Sosin, L. A. Prozorova, A. I. Smirnov, A. I. Golov, I. B. S. Kuz’minov, O. A. Petrenko, G. Balakrishnan, and M. E. Zhitomirsky, Magnetocaloric effect in pyrochlore antiferromagnet Gd₂Ti₂O₇, Phys. Rev. B 71, 094413 (2005).
- [18]M. Zhitomirsky and H. Tsunetsugu, High Field Properties of Geometrically Frustrated Magnets, Progress of Theoretical Physics Supplement 160, 361 (2005).DOI
- [19]R. Schick, T. Ziman, and M. E. Zhitomirsky, Quantum versus thermal fluctuations in the fcc antiferromagnet: Alternative routes to order by disorder, Physical Review B 102, 220405 (2020).DOI
- [20]Z. L. Dun, J. Trinh, M. Lee, E. S. Choi, K. Li, Y. F. Hu, Y. X. Wang, N. Blanc, A. P. Ramirez, and H. D. Zhou, Structural and magnetic properties of two branches of the tripod-kagome-lattice family A₂R₃Sb₃O₁₄ (A = Mg, Zn; R = Pr, Nd, Gd, Tb, Dy, Ho, Er, Yb), Phys. Rev. B 95, 104439 (2017).DOI
- [21]C. Wellm, J. Zeisner, A. Alfonson, M.-I. Sturza, G. Bastien, S. Gaß, S. Wurmehl, A. U. B. Wolter, B. Büchner, and V. Kataev, Magnetic interactions in the tripod kagome antiferromagnet Mg₂Gd₃Sb₃O₁₄ probed by static magnetometry and high-field ESR spectroscopy, Phys. Rev. B 102, 214414 (2020).DOI
- [22]V. Bondah-Jagalu and S. T. Bramwell, Magnetic susceptibility study of the heavy rare-earth stannate pyrochlores, Canadian Journal of Physics 79, 1381 (2001).
- [23]A. C. S. Hamilton, G. I. Lampronti, S. E. Rowley, and S. E. Dutton, Enhancement of the magnetocaloric effect driven by changes in the crystal structure of Al-doped GGG, Gd₃Ga₅₋ₓAlₓO₁₂ (0≤x≤5), Journal of Physics: Condensed Matter 26, 116001 (2014).DOI
- [24]B. Melot, J. Drewes, R. Seshadri, E. Stoudenmire, and A. Ramirez, Magnetic phase evolution in the spinel compounds Zn₁₋ₓCoₓCr₂O₄, Journal of Physics: Condensed Matter 21, 216007 (2009).
- [25]H. Karunadasa, Q. Huang, B. G. Ueland, P. Schiffer, and R. J. Cava, Ba₂LnSbO₆ and Sr₂LnSbO₆ (Ln = Dy, Ho, Gd) double perovskites: Lanthanides in the geometrically frustrating fcc lattice, Proceedings of the National Academy of Sciences 100, 8097 (2003).
- [26]J. Paddison, G. Ehlers, O. Petrenko, A. Wildes, J. Gardner, and J. Stewart, Spin correlations in the dipolar pyrochlore antiferromagnet Gd₂Sn₂O₇, Journal of Physics: Condensed Matter 29, 144001 (2017).DOI
- [27]J. Paddison, H. Jacobsen, O. Petrenko, M. Fernández-Díaz, P. Deen, and A. Goodwin, Hidden order in spin-liquid Gd₃Ga₅O₁₂, Science 350, 179 (2015).DOI
- [28]P. Mukherjee, Y. Wu, G. I. Lampronti, and S. E. Dutton, Magnetic properties of monoclinic lanthanide orthoborates, LnBO₃, Ln=Gd, Tb, Dy, Ho, Er, Yb, Materials Research Bulletin 98, 173 (2018).
- [29]M. Ashtar, Y. Bai, L. Xu, Z. Wan, Z. Wei, Y. Liu, M. Marwat, and Z. Tian, Structure and Magnetic Properties of Melilite-Type Compounds RE₂Be₂GeO₇ (RE = Pr, Nd, Gd–Yb) with Rare-Earth Ions on Shastry–Sutherland Lattice, Inorganic Chemistry 60, 3626 (2021).
- [30]Y.-C. Chen, L. Qin, Z.-S. Meng, D.-F. Yang, C. Wu, Z. Fu, Y.-Z. Zheng, J.-L. Liu, R. Tarasenko, M. Orendáč, J. Prokleška, V. Sechovský, and M.-L. Tong, Study of a magnetic-cooling material Gd(OH)CO₃, J. Mater. Chem. A 2, 9851 (2014).DOI
- [31]N. D. Kelly and S. E. Dutton, Magnetic properties of quasi-one-dimensional lanthanide calcium oxyborates Ca₄LnO(BO₃)₃, Inorganic Chemistry 59, 9188 (2020), pMID: 32525304.
- [32]P. J. Saines and N. C. Bristowe, Probing magnetic interactions in metal–organic frameworks and coordination polymers microscopically, Dalton Transactions 47, 13257 (2018).DOI
- [33]J. Hellsvik, R. D. Pérez, R. M. Geilhufe, M. Månsson, and A. V. Balatsky, Spin wave excitations of magnetic metal-organic materials, Physical Review Materials 4, 024409 (2020).
- [34]H. Karunadasa, Q. Huang, B. G. Ueland, J. W. Lynn, P. Schiffer, K. A. Regan, and R. J. Cava, Honeycombs of triangles and magnetic frustration in Sr₂BO₄ (L= Gd, Dy, Ho, Er, Tm, and Yb), Phys. Rev. B 71, 144414 (2005).
- [35]T. Numazawa, K. Kamiya, T. Okano, and K. Matsumoto, Magnetocaloric effect in (DyₓGd₁₋ₓ)₃Ga₅O₁₂ for adiabatic demagnetization refrigeration, Physica B: Condensed Matter 329-333, 1656 (2003).DOI
- [36]M. Orendáč, J. Hanko, E. Čižmár, A. Orendáčová, M. Shirai, and S. T. Bramwell, Magnetocaloric study of spin relaxation in dipolar spin ice Dy₂Ti₂O₇, Physical Review B 75, 104425 (2007).DOI
- [37]S. Sosin, L. Prozorova, A. Smirnov, A. Golov, I. Berkutov, O. Petrenko, G. Balakrishnan, and M. Zhitomirsky, Adiabatic demagnetization of a pyrochlore antiferromagnet Gd₂Ti₂O₇, Journal of Magnetism and Magnetic Materials 290-291, 709 (2005), proceedings of the Joint European Magnetic Symposia (JEMS’ 04).DOI
- [38]L. B. McCusker, R. B. Von Dreele, D. E. Cox, D. Louër, and P. Scardi, Rietveld refinement guidelines, Journal of Applied Crystallography 32, 36 (1999).
- [39]A. Coelho, TOPAS and TOPAS-Academic: an optimization program integrating computer algebra and crystallographic objects written in C++, Journal of Applied Crystallography 51, 210 (2018).DOI
- [40]P. Thompson, D. E. Cox, and J. B. Hastings, Rietveld refinement of Debye–Scherrer synchrotron X-ray data from Al₂O₃, Journal of Applied Crystallography 20, 79 (1987).DOI
- [41]B. J. Kennedy, B. A. Hunter, and C. J. Howard, Structural and Bonding Trends in Tin Pyrochlore Oxides, Journal of Solid State Chemistry 130, 58 (1997).DOI
- [42]J. M. Paddison, H. Ong, J. Hamp, P. Mukherjee, X. Bai, M. Tucker, N. Butch, C. Castelnovo, M. Mourigal, and S. Dutton, Emergent order in the kagome Ising magnet Dy₃Mg₂Sb₃O₁₄, Nature Communications 7, 13842 (2016).DOI
- [43]Z. Dun, X. Bai, J. A. M. Paddison, E. Hollingworth, N. P. Butch, C. D. Cruz, M. B. Stone, T. Hong, F. Demmel, M. Mourigal, and H. Zhou, Quantum versus classical spin fragmentation in dipolar kagome ice Ho₃Mg₂Sb₃O₁₄, Phys. Rev. X 10, 031069 (2020).DOI
- [44]K. Momma and F. Izumi, VESTA 3 for three-dimensional visualization of crystal, volumetric and morphology data, Journal of Applied Crystallography 44, 1272 (2011).DOI










