Introduction

Mixed valency, the presence of more than one oxidation state for a particular chemical element, is associated with very rich physics, in both molecular and solid-state systems. It can be intrinsic, as for example in stoichiometric magnetite Fe3O41, or obtained through doping as in the colossal magneto-resistance manganese oxides (R1-xAx)MnO3 (R = rare-earth cation, A = alkali or alkaline-earth cation)2. It can also imply quasi-localized ligand-hole states, i. e., ligand mixed valency, as shown for underdoped cuprates3,4 or for the spin-Peierls compound NaV2O55,6. Here we explore the mixed valence landscape in the molybdates GaMo4S8 and GaMo4Se8, lacunar-spinel quantum magnets displaying remarkable properties, e. g., modulated cycloidal and Néel-type skyrmion-lattice states with sub-10 nm periodicity that are coupled to ferroelectric polarization7, a plethora of ferroelectric domain boundaries and junctions8, and a cluster spin-glass phase9. Using advanced wave-function-based quantum chemical computational schemes, we show that the peculiar interplay of electronic correlations and spin–orbit interactions in these materials sets the stage for interesting physics and new insights: the dressing of a jeff = 1/2 spin–orbit-coupled state10 with virtual valence space excitations. This dressing, or correlation cloud, is evidenced with the help of wave-function analysis and of observables such as the ground-state g factors and branching ratios. The use of the latter as indicators for ground-state correlations is possible due to the presence of sizable spin–orbit interactions: in their absence, information can be obtained by comparing ab initio and experimental data for e. g. the cohesive energy11,12. The outcome of magnetization measurements performed on GaMo4S8 supports the computational findings and calls for spectroscopic verification of other specific predictions we make, e. g., of N-particle excitation energies as probed by resonant inelastic x-ray scattering experiments.

Results

High-temperature electronic structure, dressed j eff = 1/2 states

Spinels are generically described by the chemical formula AM2X4. The M sites define a pyrochlore lattice, each of those ions being encapsulated within an X6 octahedral cage of nearest-neighbor chalcogenide ligands. Magnetite, for example, belongs to this class of systems, with Fe ions at both A and M sites1. In lacunar spinels, the A sites have half-integer occupation. With every second A cation removed, nearest-neighbor M4 tetrahedra of the initial pyrochlore structure experience different surroundings; structural relaxation in the new setting yields a so called “breathing” pyrochlore lattice. The simplest way of representing that is as a fcc network of M4 clusters (or of M4X16 units), see Fig. 1a; adjacent M4 entities share no M site.

Figure 1
figure 1

GaMo4X8 cluster model and molecular orbital schemes. (a) GaMo4S8/GaMo4Se8 cluster model. Mo4 tetrahedra are shown in green. Small atomic spheres indicate capped effective core potentials (cECPs). (Ga: blue, Mo: green, S/Se: yellow). (b) Molecular-orbital diagram for the Mo4 cluster in high-temperature, cubic Td point group with spin–orbit coupling (SOC) and low-temperature, rhombohedral C3v symmetry as a result of Jahn–Teller (JT) distortion.

Prior electronic-structure calculations based on density functional theory13,14 point to a single-tetramer molecular-orbital-like diagram with a set of a1, e, and t2 levels at the top of the occupied orbital space, where the notations correspond to Td point-group symmetry (see Fig. 1b). In GaMo4S8 and GaMo4Se8 there are formally 2.75 electrons per transition-metal (TM) site (i. e., 4 × 2.75 = 11 d electrons per Mo4 cluster). Since the t2 levels come at higher energy as compared to the a1 and e components, the basic single-tetrahedron electronic structure in GaMo4X8 would then imply filled a1 and e states and one hole within the t2 sector. However, given the results of state-of-the-art quantum chemical calculations on related “clustered” compounds, both d-15,16,17,18 and p-electron19 based, there are reasons to believe that the picture presently put forward for the single-tetramer electronic structure in GaMo4X8 is somewhat too sketchy.

For insights into the correlated electronic structure of Mo413+ units in GaMo4S8 and GaMo4Se8, we adopt the embedded-cluster quantum chemical approach described in Ref.18 and the lattice parameters and atomic positions determined by Powell et al. for GaMo4S820 and Routledge et al. for GaMo4Se821. An [Mo4X16]19- cluster was in each case embedded within a large array of point charges created with the Ewald program22,23; capped effective core potentials (cECPs) were assigned to the 98 closest atoms. Further details on the embedding procedure are provided in the Supplemental Material (SM-1). The more general theoretical frame is mapped in Refs.11,24.

The nature of the ground state and the low-energy excitation spectrum of a Mo4 tetrahedron were first analyzed for the high-temperature (HT) structure, in both GaMo4S8 and GaMo4Se8. The leading ground-state configuration is a12 e4 t25 t10 t20 (c.f. Figure 1b), but for a “converged” CAS(11e,12o) orbital space it comes with a weight of only 62% in the CASSCF ground-state wave-function (see Supplemental Material SM-3 for the active space orbitals and Supplementary Tables S1 and S2 in Supplemental Material SM-4 for the excited state configurations). The remaining configurations are mostly double excitations, e. g., a12 e3 t24 t11 t21; there is a multitude of those, each contributing to the overall wave-function with a weight of less than 5%.

These findings do not seem unusual for this class of materials: important multiconfigurational effects, i. e., a weight of the a12 e4 t21 leading electron configuration of less than 70%, were also found in the related 4d lacunar spinel GaNb4Se818. The competition between different electronic configurations seems to be even sharper in the vanadium analogue17,25.

Spin–orbit coupling (SOC) is effective for the multiconfigurational 2T2 ground-state, giving rise to a low-lying spin–orbit doublet and a spin–orbit quartet with an excitation energy of about 0.12 eV (see Figs. 1b, 2). Yet, different from the (…a12 e4) t25 leading ground state configuration, unambiguously described in terms of S = 1, Leff = 1/2, and jeff = 1/2 quantum numbers10, the multiconfigurational case entails more subtle nuances. This is discussed in the following paragraphs.

Figure 2
figure 2

Low-energy excitation energies for high-temperature GaMo4S8. The spectrum was calculated using an [Mo4S16]19- embedded cluster model (CAS(11e,12o)). Both CASSCF and NEVPT2 values (also with SOC for the latter) are given in eV. Four quartets and six doublets were included in the state-averaging procedure. Notations according to Td point group symmetry are used.

As depicted in Fig. 2, further excited states follow at energies exceeding 1 eV. Different from the ground 2T2 term, spin–orbit interactions play a less important role for the higher-lying multiplets. But noticeable modifications occur post-CASSCF for the latter, at the NEVPT2 level; the sequence of the 4A2 and 2A1 terms, for instance, is different in NEVPT2 as compared to CASSCF. Comparing the NEVPT2 energies in GaMo4S8 and GaMo4Se8 (see Supplementary Material SM-4), it is seen that the excitation spectra are rather similar, somewhat scaled down in the latter compound. This can be attributed to having longer Mo-Se bonds in GaMo4Se8, i. e., weaker ligand fields.

To quantify the strength of spin–orbit interactions, the SOC constant λ was determined from minimal active space calculations. For such a CAS(5e,3o), which only incorporates the (…a12 e4) t25 configuration, the jeff = 3/2 quartet is split by 100 meV from the jeff = 1/2 ground-state term, yielding a SOC constant \(\lambda_{{{\text{SOC}}}}^{{{\text{Mo}}_{4} }} = 67\; {\text{meV}} = 538\;{\text{ cm}}^{ - 1}\). This value lies close to the SOC constant estimated for the (…a12 e4) t21 compound GaNb4Se8, \(\lambda_{{{\text{SOC}}}}^{{{\text{Nb}}_{4} }} = 516 \;{\text{cm}}^{ - 1}\)18. Compared to a Mo+ single-ion value \(\lambda_{{{\text{SOC}}}}^{{{\text{Mo}}^{ + } }} = 630 \;{\text{cm}}^{ - 1}\)10, the SOC parameter is lower in GaMo4Se8, due to the more delocalized character of lacunar-spinel t2 electrons – those orbit in lacunar spinels a four-site unit, not a single TM ion. The fact that the splitting between the lowest spin–orbit states increases when using larger active spaces (from 0.1 to 0.12 eV, see Fig. 2) has to do with electron correlations that bring in multiconfigurational physics; such effects were not addressed in earlier computations based on density functional theory14,26.

A peculiar feature of nominal jeff = 1/2 ground states is free-electron-like g factors10, even if Leff = 1. On the basis of the active space employed for the calculations [i. e., CAS(11e,12o)], ground-state g factors of 2.18 are obtained. The sizable deviation from 2 has to do with contributions of other electron configurations to the multiconfigurational ground-state wave-function, i. e., dressing with intra-tetramer excitations. For comparison, magnetic susceptibility measurements indicate a g value of 2.37 (see Supplementary Material SM-5). The somewhat larger experimental estimations hint towards additional vibronic effects27,28 not included in the rigid cluster model.

Another quantity that can provide evidence for such dressing is the branching ratio (BR) in L-edge x-ray absorption spectroscopy (XAS), i. e., the ratio of the integrated intensities of the L3 and L2 absorption edges. It is directly related to the ground-state expectation value \({\mathcal{Z}}\) of the spin–orbit Hamiltonian and can be compactly expressed as \({\text{BR}} = \left( {2 + r} \right)/\left( {1 - r} \right)\)29,30,31, where \(r = {\mathcal{Z}}/n_{{\text{h}}}\), \(n_{{\text{h}}}\) is the number of holes in the valence shell, and for a given LS term \({\mathcal{Z}} = \lambda^{ - 1} {\mathcal{H}}_{{{\text{SO}}}} = {\mathbf{L}} \cdot {\mathbf{S}}\). \({\mathcal{Z}}\) can be alternatively written as30,31,32

$${\mathcal{Z}} = \lambda^{ - 1} \left( {E_{{{\text{LS}}}}^{0} - E_{{{\text{LSJ}}}}^{0} } \right),$$
(1)

where \(E_{{{\text{LS}}}}^{0}\) and \(E_{{{\text{LSJ}}}}^{0}\) are ground-state eigenvalues without and with SOC, respectively. \(E_{{{\text{LSJ}}}}^{0}\) may depend on both first- and second-order SOCs.

As an example, for a nominal jeff = 1/2 spin–orbit ground state in t2g5 iridates33,34 (\(n_{{\text{h}}} = 5\)), values \({\mathcal{Z}}\) = 1 and BR = 2.75 are expected35,36; in contrast, for vanishing valence-shell spin–orbit interactions, \({\mathcal{Z}}\) → 0 and BR → 2 are obtained. For the lacunar-spinel molybdates, using the minimal active space in the CASSCF calculation (five electrons within the three t2 cluster orbitals), we find indeed \({\mathcal{Z}}\) = 1 (see Table 1, first line). Slight reduction of the NEVPT2 minimal-CAS \({\mathcal{Z}}\) to 0.99 indicates the admixture of other electron configurations in the NEVPT2 treatment. Proper description of multiconfigurational physics requires however larger active spaces: enlarging the active space to CAS(11e,12o), \({\mathcal{Z}}\) increases to 1.2–1.3, even when only the lowest-lying 2T2 term is included in the spin–orbit part of the computation. The difference between the large-CAS \({\mathcal{Z}}\), \(\approx\) 1.2, and the minimal-CAS \({\mathcal{Z}}\) of 1 is an indicator for the amount of dressing of the jeff = 1/2 object with intra-tetramer electronic excitations. Interestingly, while in e. g. t2g5 iridates or rhodates this difference mainly arises from strong second-order SOCs32,35,36, here those are inactive at HTs. \({\mathcal{Z}}\) being smaller and \(n_{{\text{h}}}\) larger than in t2g5 iridates, the BR is expected to be less than \(\approx\) 2.5 in the molybdate lacunar spinels, which can be verified by XAS measurements.

Table 1 \({\mathbf{\mathcal{Z}}}\) values in GaMo4S8 and GaMo4Se8.

Low-temperature electronic structure, \(\tilde{S}\) = 1/2 pseudo-spins

At temperatures of about 50 K, the GaMo4X8 lacunar spinels undergo a phase transition from cubic (F\(\overline{4}3\)m) to rhombohedral (R3m) space group13, with compression of the [Mo4X4]5+ tetrahedra along the cubic [111] direction, i. e., along a line connecting opposite Mo and X vertices13,37,38. This polar distortion gives rise to axial magnetic anisotropy along [111]. The effects of structural distortions as reported by Powell et al. for GaMo4S820 and by Routledge et al. for GaMo4Se821 were analyzed on the basis of CASSCF + NEVPT2 computations and orbital active spaces as employed for the HT phase (see Fig. 3 for GaMo4S8 and Supplementary Material SM-4 for GaMo4Se8).

Figure 3
figure 3

Influence of Jahn–Teller distortion on the low-energy excitation spectrum of GaMo4S8. Both high-temperature (HT) and low-temperature (LT) energies are given for the [Mo4S16]19- cluster model (NEVPT2(11e,12o)), including SOC for the latter. Six quartets and ten doublets were included in the state-averaging procedure.

Trigonal crystal-field splittings as large as 0.33 eV and 0.24 eV for GaMo4S8 and GaMo4Se8, respectively, are found to occur within the t2 manifold, as illustrated by the splitting of the HT ground 2T2 state into 2A1 and 2E terms. Obviously, this renders the jeff = 1/2 picture inadequate. Yet, the effect of SOC is not negligible—three distinct Kramers doublets are found in the spin–orbit calculations, with relative NEVPT2 energies of 0, 0.32/0.23, and 0.38/0.29 eV. From a magnetic point of view, the pertinent effective picture for the Mo4 tetrahedra is therefore that of \(\tilde{S}\) = 1/2 moments with significant angular-momentum contribution, in accordance with susceptibility measurements39. This is further seen in the structure of the ab initio low-T (LT) g factors (Table 2), where \(g_{||}\) and \(g_{ \bot }\) refer to directions parallel and perpendicular to the rhombohedral axis (the [111] axis in cubic setting), respectively. A similar relation, \(g_{||} < g_{ \bot }\), was found for positive trigonal/tetragonal splittings in RuCl340 and Ba2IrO432,41.

Table 2 Calculated g factors for LT [Mo4X16]19- (X = S, Se) cluster models (CAS(11e,12o)).

In order to experimentally verify the predicted type of axial anisotropy, characterized by \(g_{||} /{\text{g}}_{ \bot } < 1\), we measured the magnetization of LT GaMo4S8 along two non-equivalent directions, the [110] and [100] axes shown in Fig. 4. The [110] direction is perpendicular to the [111] rhombohedral axis, hence the corresponding saturation magnetization directly yields \({g}_{ \bot }^{{{\text{exp}}}}\). Measurements with field applied along the [100] axis yield a combination of g factors, namely \(\sqrt {g_{||}^{2} \cos^{2} \left( {54.7^\circ } \right) + g_{ \bot }^{2} \sin^{2} \left( {54.7^\circ } \right)}\). Below 2 T, the low-temperature phase of GaMo4S8 shows a sequence of meta-magnetic transitions between different modulated magnetic states, including the skyrmionic lattice state7, which is evidenced by anomalies in this region. These magnetic states are not the focus of the present work; instead, we study the high-field magnetization, with special emphasis on the anisotropy of the saturation magnetization μsat. That allows us to determine the anisotropy of the g tensor. For both field directions, the measured magnetization is clearly saturated above ≈6 T. We find a tiny (≈1%) decrease of the magnetization at higher fields, originating from a small diamagnetic contribution of the sample holder. This is visible only when the magnetization of the crystal is fully saturated.

Figure 4
figure 4

Magnetization curves for LT GaMo4S8. Measured (thick full lines) and simulated (dashed lines) curves were obtained at a temperature of 2.5 K. [110] (red), [100] (blue), and [111] (black) directions are given in the cubic setting. The simulated curve is obtained for a [Mo4S16]19- embedded cluster (CASSCF(11e,12o)).

For comparison, magnetization curves computed on the basis of the CASSCF(11e,12o) g factors42 are also shown. At low fields between 1 and 4 T the simulated curves underestimate the measured magnetization, likely because of the single-tetrahedron model for which inter-tetrahedral magnetic interactions are absent. However, the experimental and simulated saturation magnetization values are in good agreement above 6 T, with \(\mu_{{{\text{sat}}}}^{{\left[ {110} \right]}} = 1.27 \mu_{{\text{B}}}\) and \(\mu_{{{\text{sat}}}}^{{\left[ {100} \right]}} = 1.12 \mu_{{\text{B}}}\). Since our calculated values are fully ab initio (with no ad hoc parameters in the underlying theory), small deviations can be assigned to structural differences present in experiment and embedded-cluster computations (see Supplemental Material SM-6).

Conclusions

With the help of ab initio quantum chemical calculations for both high- and low-temperature structural configurations, we shed light onto the nature of magnetic moments in mixed valence GaMo4S8 and GaMo4Se8. The quantum chemical computational data is referenced against low-temperature magnetization measurements. Our study documents the dressing of a bare jeff = 1/2 object10 with valence space excitations. We analyzed in this regard the wave-function itself and observables such as the ground-state g factors and the so-called branching ratios. The essential physics refers to genuine electron correlation effects, different from the case of layered t2g5 iridium oxides33,34 where the branching ratios and ground-state g factors deviate from values corresponding to nominal jeff = 1/2 moments due to jeff 1/2–3/2 mixing and second-order spin–orbit couplings in the presence of lower-symmetry crystal/ligand fields32,41. The dressing, or correlation cloud, evidenced here has to do with the ground state of a Mo4 tetramer, different from the quasiparticle concept utilized for excitations in solids. Its internal degrees of freedom determine the polar properties of the material, although a comprehensive description requires additionally an adequate treatment of vibronic couplings and inter-tetramer interactions.

Methods

All calculations were performed with the Orca program package, v5.043. Near-degeneracy correlation effects were accounted for by complete active space self-consistent field (CASSCF) theory44,45 while additional dynamical correlations were described by N-electron valence second-order perturbation theory (NEVPT2)46. In the framework of the Douglas-Kroll-Hess (DKH) approximation47, the all-electron SARC-DKH-TZVPP48 basis set for Mo and the DKH-DEF2-TZVPP49 basis set for S/Se were used to treat SOC. To the cECPs, pseudopotentials of Andrae et al.50 for Mo, Bergner et al.51 and Leininger et al.52 for Ga and S/Se were assigned. Additional computational details are given in the Supplemental Material (SM-2). For direct connection with experimental results, the magnetization of a GaMo4S8 single crystal grown by the flux method53 was measured at 2.5 K in fields up to 14 T using a Quantum Design MPMS3.