Introduction

Since the discovery of ferrocene in the 1950s by Kealy and Pauson, as well as Miller et al.1, and the Nobel Prize-winning work of Fischer and Wilkinson2,3, various derivatives of metallocenes have been synthesized and have played pivotal roles in important discoveries in a variety of fields, including catalysis4,5, materials6,7,8,9, energy10, and medical11,12 sciences (Fig. 1a). The versatility of metallocenes stems from the ability of the cyclopentadienyl ligand (Cp) and its derivatives to stabilize metals with a wide range of valence electron counts. To date, d-block metallocenes and their derivatives with a formal electron count in the range from 14 to 20, including recently synthesized 16-electron ferrocene dication13 and 20-electron cobaltocene anion14, have been isolated. Wilkinson and Hursthouse et al. reported a possible formal 21-electron manganocene derivative, MnCp2(dmpe) [dmpe: 1,2-bis(dimethylphosphino)ethane] (Fig. 1b)15. However, as noted by the authors, the formation of this complex does not follow the expected trend of other metallocenes16. Furthermore, as with other manganocene complexes, the coordination mode of the Cp ligands in this complex deviates from the ideal η5-coordination because of ring slippage15,17,18,19,20,21,22. This anomalous behavior of manganocene complexes is attributed to the primarily electrostatic character of the Mn–C(Cp) interactions, and the observed coordination modes are mainly controlled by the steric factor of the ligands22,23,24. The solid-state zigzag chain structure of MnCp2 is a vivid example of the anomalous coordination chemistry of manganocene complexes23. Therefore, according to the available literature data, the formal electron count of MnCp2(dmpe) cannot be conclusively described. Geiger et al.25 and later Bard and Strelets et al.26 reported the electrochemical reduction of NiCp2 and proposed the formation of formal 21-electron NiCp2– based on cyclic voltammetry (CV) (Fig. 1b). However, since this complex is thermally unstable and cannot be isolated, the possibilities for its characterization are limited.

Fig. 1: Preparation of 21-electron metallocene derivatives.
figure 1

a Selected examples of metallocenes and their applications. b Previously proposed 21-electron metallocene derivatives. c Synthetic route to the 21-electron cobaltocene derivative, complex 1. THF: tetrahydrofuran. d Oak Ridge Thermal Ellipsoid Plot (ORTEP) of 1 at the 80% probability level according to high-resolution single-crystal X-ray diffraction (SC-XRD). Green ellipsoids: hydrogen atoms. e Comparison of M–Cnt(Cp), Δ, and Ψ values. M: metal. DFT: density functional theory. dmpe: 1,2-bis(dimethylphosphino)ethane. Cp: cyclopentadienyl.

Full size image

The 18-electron rule works best for the complexes with strong π-accepting ligands27 and the existence of formal 14- to 20-electron metallocenes clearly shows that the 18-electron rule is only loosely applicable to metallocene complexes, especially paramagnetic ones. Furthermore, it is worth nothing to point out that a bonding model based on sdn hybridization and 3c–4e hypervalent bonding28,29,30 indicates the possible formation of formal 21-electron metallocene derivatives by exploiting weaker attractive interactions such as higher multi-center donor–acceptor interactions, long-range electrostatics, London dispersion, and/or entropically favorable interstitial vacancies31. Nevertheless, the isolation of well-defined d-block metallocenes or their derivatives with more than a formal 20-electron count has remained elusive. The difficulty in synthesizing formal 21-electron complexes lies in the fact that increasing deviation from the stable 18-electron configuration destabilizes metal–C(Cp) interactions and promotes the change of hapticity of Cp ligands32,33,34,35,36,37 or decomposition of the complexes25,26. Here, we report the synthesis, isolation, and detailed theoretical and experimental characterization of a new cobaltocene derivative with a formal electron count of 21 (Fig. 1c).

Results and Discussion

Synthesis and spectroscopic characterization

Inspired by the coordination chemistry of Cp2Ni32, we envisioned that the incorporation of two Cp ligands into a pyridine-based pincer ligand motif might allow the N-coordination of the pyridine ligand to the metal center while maintaining the η5-coordination mode of the two Cp groups. For this reason, the Na2CpNCp ligand (Fig. 1c)38 was chosen for the synthesis of metallocene derivatives with formal valence electron counts of more than 20. Prior to this work, the coordination chemistry of d-block metals and Na2CpNCp was known only with metals with a formal d-electron count of less than four39,40,41. Portionwise addition of Na2CpNCp to a tetrahydrofuran (THF) solution of CoBr2 at room temperature resulted in the formation of a NaBr precipitate. Following the workup, an orange crystalline solid of 2,6-bis(methylenecyclopentadienyl)pyridinecobalt (1) was isolated in 34% yield (average of two experiments) (Fig. 1c). Unlike NiCp2–, complex 1 was stable as a solid or as a solution under nitrogen at ambient temperature.

In order to unambiguously show the bonding situation and formal electron count of 1, the paramagnetic complex 1 was characterized by a number of methods, including high-resolution single-crystal X-ray diffraction (SC-XRD), 1H, 13C, and 2H nuclear magnetic resonance (NMR) spectroscopies, electron paramagnetic resonance (EPR) spectroscopy, vibrating sample magnetometry (VSM), electrospray ionization mass spectrometry (ESI-MS), elemental analysis, and density functional theory (DFT) calculations at the TPSSh-D4/def2-QZVPP//TPSS-D4/def2-TZVPP level of theory. The molecular structure of 1 in the crystal, obtained by precision high-resolution SC-XRD at 100 K with a reciprocal resolution \({{\sin}}({\theta}_{{{\max}}}/\lambda)\) of 1.26 Å–1 and following multipole refinement42, supports the N-coordination of a pyridine moiety and η5-coordination mode of two Cp groups (Fig. 1d). More specifically, the Co–N bond distance of 2.1998(3) Å is within the internuclear CoII–NPy distances in the neutral CoIIPNP pincer complexes, which are in the range from 1.974 to 2.343 Å with an average distance of 2.144 Å (Supplementary Table 2). The Co–C(Cp) distances are in the range from 2.2783(3) to 2.3292(2) Å with an average distance of 2.302 Å, and the Co–Cp centroid (Co–Cnt(Cp)) distances are 1.9607(2) and 1.9593(2) Å. These distances are significantly longer than those of cobaltocene (2.112 Å for the average of the libration-corrected Co–C(Cp) distances and ca. 1.722 Å for Co–Cnt(Cp))43 and may indicate a weakening of the Co–C(Cp) interactions. The differences between the maximum and minimum Co–C(Cp) distances of 0.051 and 0.035 Å for the two Cp groups are similar to those of cobaltocene (0.033 Å)43. The coordination mode of the Cp groups was further analyzed by the Cp ring slip parameters Δ and Ψ (Fig. 1e)44. The Δ and Ψ values of complex 1 are among the smallest reported for Δ and Ψ35,44 and strongly support the η5-coordination mode of the two Cp groups. In comparison, the proposed 21-electron manganocene derivative, MnCp2(dmpe), has a significantly larger deviation from the ideal η5-coordination mode (Fig. 1e). The Δ and Ψ values derived from the gas-phase DFT calculations are in agreement with the experimental values, supporting that crystal packing effects are not essential for the η5-coordination (Fig. 1e). The DFT optimized structure reproduces the equidistance of the Co–C(Cp) bonds with an average distance of 2.297(21) Å and a short Co–N distance of 2.2176 Å. Furthermore, the average Wiberg bond index for the Co–C(Cp) bonds of 0.15(2) is similar to that of cobaltocene (0.22(5)) as calculated at the same level of theory, and a practically uniform distribution of negative charges on the Cp rings was observed by the calculations (Supplementary Table 3).

In solution, the 1H NMR spectrum of 1 showed five signals between –260 and +144 ppm (Fig. 2a). All the 1H NMR signals were assigned based on selective deuteration of the Cp groups and two-dimensional NMR experiments (Supplementary Figs. 6, 7, and 12). The two broader signals at the higher field are from the Cp rings, and the three sharper signals at the lower field are from the pyridine ring and CH2 groups of the CpNCp ligand. The 13C NMR spectrum shows four signals between –504 and +410 ppm (Supplementary Fig. 5), and the signals are partially assigned based on two-dimensional NMR and the splitting pattern of the signals. Three 13C NMR signals are missing, most likely due to the close proximity of the Cp groups to the paramagnetic Co center. Magnetic measurements conducted by the Evans NMR method (298 K, in C6D6) and VSM (298 K, bulk solid) showed that 1 has an effective magnetic moment μeff of 4.3 and 3.99 μB (μB: Bohr magneton) in solution and solid state, respectively, which indicates that 1 has the S = 3/2 ground electron spin state. The S = 3/2 ground electron spin state was further supported by continuous wave (cw)-EPR, where a large deviation of effective g-values from that of a free electron was observed due to the sizable zero-field splitting of the S = 3/2 complex (Fig. 2b). The EPR spectrum also clearly shows the presence of large 59Co hyperfine splitting, which supports the presence of a cobalt-centered radical.

Fig. 2: Characterization of complex 1.
figure 2

a 1H NMR (400.15 MHz, 298 K, in C6D6) spectrum of 1 and signal assignments (a-e). *: signals from residual THF. #: a signal from residual C6D5H. b Experimental (4.2 K, toluene glass) and simulated X-band EPR spectra of 1 using fictitious S = 1/2. g: effective g-values. A(59Co): 59Co hyperfine coupling constants. c Four-pulse HYSCORE spectrum of 1 observed at 320 mT (4 K, toluene-d8 glass). d CV of 1 and cobaltocene recorded with a glassy carbon electrode (3 mm diameter) from –1.6 to –1 V range with a scan rate of 0.01 V s–1, in 0.1 M [Bu4N]PF6 THF at 25 °C, and reported vs. FeCp2+/0. The starting point (open circuit potential) and direction of scans are indicated by arrows.

Full size image

The presence of the Co–N bond in the frozen solution was supported by pulse-EPR measurements, which observed the electron spin echo envelope modulation (ESEEM) effects45. The two-dimensional field-swept two-pulse and three-pulse ESEEM spectra observed at 4 K (Supplementary Figs. 30 and 31) showed the nuclear modulation effects due to the N atom coupled with the Co atom, indicating the presence of bonding interaction between them. Furthermore, we applied four-pulse hyperfine sublevel correlation (HYSCORE) spectroscopy46 to measure the nuclear spins coupled with the Co atom. The HYSCORE spectrum (Fig. 2c) showed the correlation peaks due to the 1H and 14N atoms. The 1H-HYSCORE signals spreading over a frequency range of 3–4 MHz in the angular selective spectra were assigned to the protons belonging to the Cp groups. Based on the cw-EPR spectrum and the modulation effects, we determined spin-Hamiltonian parameters of 1 with S = 3/247(Supplementary Fig. 33, Supplementary Table 1) as well as fictitious S = 1/2 (Fig. 2b). The experimental tensors showed that the principal axis with the largest principal value of the 14N-hyperfine coupling (A) tensor coincides with that of the 59Co-A tensor. This observation indicates the presence of electronic interaction between the Co and N atoms in the frozen solution.

The cyclic voltammogram of 1 showed reversible Co3+/2+ half-wave potential at –1.23 V (vs. FeCp2+/0 in MeCN), which is 0.11 V higher or less negative than that of cobaltocene (Fig. 2d). This change in the Co3+/2+ half-wave potential is not due to the introduction of the pyridinylmethyl group on each Cp ring, since such a substitution is expected to decrease the reduction potential by about 0.02 V48. Thus, the N-coordination of the pyridine group unexpectedly increases the reduction potential of 1 despite the increase in formal electron count.

Examination of the interatomic interactions

The bonding situation in 1 was further investigated by the quantum topological analysis of the experimental electron density \(\rho ({{{{{\bf{r}}}}}})\)49 and electrostatic potential \({\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\)50 in the crystal. The analysis identified six bond critical points (BCPs) with negative near-zero values of the electronic energy density \(h({{{{{\bf{r}}}}}})\) corresponding to the Co–N and Co–C(Cp) coordinate bonds (Supplementary Fig. 34)51. In comparison, the analysis of \(\rho ({{{{{\bf{r}}}}}})\) derived from DFT calculations resulted in a different set of Co–C(Cp) coordinate bonds (Supplementary Figs. 34 and 35), while Co–N was still present. It is known that the location of BCPs alone is not sufficient to correctly describe metal⋅⋅⋅Cp hapticity52. Therefore, the distribution of the Laplacian of electron density \({\nabla }^{2}\rho ({{{{{\bf{r}}}}}})\) (Fig. 3a and Supplementary Fig. 36) was studied. Examination of the \({\nabla }^{2}\rho ({{{{{\bf{r}}}}}})\) maps reveals that the charge concentration lobes associated with the N atom and each C atom of the Cp rings (lobes of red contour in Fig. 3a) are directed toward the Co atom in the electron depletion region (blue contours) approximately along the internuclear lines. Moreover, the N-atom lobe is directed into the furcation region between the valence shell charge concentrations of the Co atom, supporting the donor–acceptor nature of the coordinate Co–N bond. These observations suggest the η5-coordination of both Cp groups as well as the N-coordination of the pyridine group. The experimental and theoretical values of \(\rho ({{{{{\bf{r}}}}}})\), \({\nabla }^{2}\rho ({{{{{\bf{r}}}}}})\), and \(h({{{{{\bf{r}}}}}})\) at the BCP for Co–N: 0.057, 0.220, and –0.006 a.u. and 0.055, 0.218, and –0.011 a.u., respectively, showed good agreement. An approximate Co–N delocalization index (DI) of 0.78 and an average Co–C(Cp) DI of 0.52(5), obtained from DFT calculations, further support the presence of interactions between Co and N, as well as all C(Cp) atoms52,53.

Fig. 3: Quantum chemical topology of 1.
figure 3

Experimental maps based on SC-XRD data (left) and theoretical maps based on gas-phase DFT calculations (right) are compared. Map planes pass through the atoms Co1, N1, and C2. a Contour maps of \({\nabla }^{2}\rho ({{{{{\bf{r}}}}}})\); the logarithmic scale in the form of ± 1, 2, 4, 8 ⋅ 10n (–2 ≤ n ≤ 3) e Å–5 is adopted; blue and red colors correspond to positive and negative function values, respectively. b Superposition of the zero-flux boundaries of \(\rho\)-basins (black) and \({\varphi }_{{{{{{\rm{es}}}}}}}\)-basins (blue) in the gradient vector fields \(\nabla \rho ({{{{{\bf{r}}}}}})\) and \(\nabla {\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\).

Full size image

The Coulombic attraction between Co and N as well as C(Cp) atoms was examined by superposing trajectory maps of the gradient vector fields of charge density and electrostatic potential, \(\nabla \rho ({{{{{\bf{r}}}}}})\) and \(\nabla {\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\) (Fig. 3b)50,54. The electrons within the volumetric overlapping area of \(\nabla \rho ({{{{{\bf{r}}}}}})\) and \(\nabla {\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\) (colored yellow in Fig. 3b) belong to the atomic \(\rho\)-basins of N and all C(Cp) but simultaneously fall into the neighboring pseudoatomic \({\varphi }_{{{{{{\rm{es}}}}}}}\)-basin of Co; therefore, these captured electrons are drawn toward the Co nucleus by the electrostatic force \({{{{{{\bf{F}}}}}}}_{{{{{{\rm{es}}}}}}}\left({{{{{\bf{r}}}}}}\right)=\nabla {\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\). This area represents the trace of the expansion of the ligand atoms of the first coordination sphere, capturing electrons, and the compression of the Co atom losing electrons as a result of the chemical bond formation starting from free neutral atoms. The electron transfer thus defined is estimated to be 0.90 and 0.98 e for 1 in the crystal according to experimental diffraction data and for the free molecule from DFT calculations, respectively. The action of the electric field generated by the cobalt atom inside the nearest 11 atoms of the ligand, as well as the joint total interatomic charge transfer from Co to N and C(Cp), indicates that all 10 C(Cp), as well as the N-atom, form a “ligand binding field” around the Co-coordination center. The experimental and computational studies strongly support the presence of the Co–N bond and two η5-coordinated Cp ligands, and together show that complex 1 is a formal 21-electron cobaltocene derivative.

Attempted synthesis of 22-, 20-, and 19-electron analogs

Aiming at the formation of a 22-electron analog, we attempted to synthesize the analogous Ni(CpNCp) complex. However, the reaction between Na2CpNCp and NiCl2(DME) (DME: 1,2-dimethoxyethane) resulted in the formation of a dimeric nickel complex 2 that lacked Ni–N bonds (Fig. 4a). Therefore, the addition of one more electron cancels out the stability of the complex gained by the formation of the metal–N bond. The one-electron oxidation of 1 was also investigated to elucidate if a formal 20-electron analog could be prepared. Oxidation of 1 by [FeCp2]BF4 resulted in the quantitative formation of a diamagnetic cobaltocenium complex 3 (Fig. 4b). The SC-XRD analysis of 3 revealed long Co⋅⋅⋅N internuclear distances of 2.745(4) and 2.790(5) Å, which vary significantly for two molecules in the asymmetric cell. Furthermore, the average Co–C(Cp) bond length is significantly shortened from 2.302 to 2.084 Å and close to that of [CoCp2]BF4 (2.041 Å)55. The variable temperature (from –40 to 80 °C) NMR measurements showed no sign of reversible coordination of the pyridine group, and the 15N NMR chemical shift of 3 is close to that of Na2CpNCp. Therefore, 3 is best described as a formal 18-electron complex without Co–N bonding. The absence of Co–N bonding is likely due to the formation of a stronger attractive interaction between the anionic Cp groups and the Co3+ center, as evidenced by the shortening of the Co–C(Cp) bonds, which increases the energy penalty required to form a Co–N bond. To further investigate the coordination chemistry of the CpNCp ligand, we synthesized the Mn complex 4 (Fig. 4c). A μeff of 6.2 (in C6D6, at 298 K) and the cw-EPR spectrum and its simulation with S = 5/2 spin state supports an S = 5/2 ground electron spin state of 4 (Supplementary Fig. 27). The SC-XRD analysis of 4 showed that the Mn−N distance is 2.2709(14) Å, which is within the Mn−N distance of other N-donor-ligated manganocene derivatives24. The Cp ring slip parameters Δ of 0.213 and 0.105 Å as well as Ψ of 5.7° and 2.8° are relatively small but significantly larger than those of 1 (Fig. 1e). Thus, like other manganocene derivatives, the coordination chemistry of 4 is more strongly influenced by the ligand steric factor24. These studies of the Ni, Co, and Mn complexes led us to further investigate the origin of the stability of complex 1 and the instability of the metal–N bonds in complexes 2 and 3 by computational methods.

Fig. 4: Attempted formation of formal 22-, 20-, and 19-electron analogs.
figure 4

a Synthetic route to the dimeric NiII complex 2. THF: tetrahydrofuran. DMF: dimethylformamide. DME: 1,2-dimethoxyethane. b Oxidation of 1 with [FeCp2]BF4 to generate CoIII complex 3. c Synthetic route to the MnII complex 4. Insert: Molecular structure of corresponding complexes at the 80 % probability level for non-hydrogen atoms according to SC-XRD.

Full size image

Origin of the stability

The origin of the stability of 1 and the unexpected increase in the Co3+/2+ reduction potential were investigated by computational methods. Using DFT calculations, we were able to locate an isomer of 1 with an S = 1/2 spin state in which the coordinate Co–N bond is broken, as a local minimum energy structure (1’, Fig. 5).

Fig. 5: Comparison of MO diagram of cobaltocene, 1’, and 1.
figure 5

Inset: canonical KS-QROs (isovalues at 0.05-0.13 a.u.), blue lobes: positive phases, red lobes: negative phases. For a detailed MO diagram and KS-QROs, please see Supplementary Figs. 43-46. LUMO: lowest unoccupied molecular orbital. SOMO: singly occupied molecular orbital. SOMO–1: a molecular orbital at one energy level below SOMO. SOMO–7: a molecular orbital at seven energy level below SOMO1. npy: nitrogen lone pair electrons.

Full size image

The relaxed potential energy surface scans of the Co−N distance reveal that upon decreasing the Co−N distance, the 21-electron configuration with a spin state of S = 3/2 is more stabilized than the 19-electron configuration with S = 1/2 (Supplementary Fig. 41), with 1 being 5 kcal mol–1 (Gibbs free energy at 298 K) more stable than 1’. Thus, the coordination of nitrogen significantly stabilizes complex 1. Figure 5 shows the frontier molecular orbital (MO) diagrams of cobaltocene, 1’, and 1, derived canonically from Kohn–Sham quasi-restricted orbitals (KS-QROs)56. As expected, the MO diagrams of cobaltocene and 1’ are similar, and no significant change in the energy of the singly occupied molecular orbital (SOMO) was observed. Examination of the canonical KS-QROs of 1 revealed that the Co−N bond in 1 results from the σ-type interaction between the metal d-orbital and the nitrogen lone pair. The formation of the Co−N bond has two consequences. First, it generates the Co−N antibonding orbital, which is singly occupied (Fig. 5, SOMO1). Thus, the Co−N bond formally has a bond order of 0.5. This situation is reminiscent of the formation of 19-electron adducts from the 17-electron complexes, where the formation of one-half of a metal-ligand bond stabilizes the 19-electron complex57. Second, the formation of the Co−N bond significantly lowers the energy level of the SOMO and the lowest unoccupied molecular orbital (LUMO) of 1’, and both can be singly occupied to form an S = 3/2 ground electron spin state (Fig. 5, SOMO2 and SOMO3). These two orbitals are Co⋅⋅⋅Cp antibonding, and their partial occupation explains the significant elongation of the Co−С(Cp) internuclear distances in 1. The decrease in the energy of the highest SOMO3 explains the observed increase in the Co3+/2+ reduction potential. These half-filled Co⋅⋅⋅Cp antibonding orbitals allow the Cp ligands to maintain the η5-coordination mode. Therefore, the stability of 1 is a product of the formation of one-half of a Co−N bond and the half-filled two Co⋅⋅⋅Cp antibonding orbitals, and the pincer ligand motif that holds all these weak Co−N and Co−C(Cp) bonds together. Indeed, a control experiment showed that coordination of pyridine to cobaltocene does not occur without the chelate effect provided by the pincer motif (see the Supplememntary Information for details). This type of complex stabilization by half-filled metal−ligand antibonding orbitals contrasts with the way formal 20-electron M(CO)8 (M = Zr, Hf)58 and [M(CO)8]– (M = Sc, Y, La)59 species are stabilized by the presence of a filled purely ligand based MO. Our MO analysis also qualitatively explains the formation of 2 and 3. The hypothetical formal 22-electron Ni(CpNCp) complex should have a fully occupied Ni−N antibonding orbital, leading to the cleavage of a Ni−N bond and the formation of 2. The removal of one electron from 1 should lead to the pairing of the remaining two unpaired electrons in 1 to achieve a more stable low-spin configuration, in which the Co−N bond does not exist.

In summary, we have synthesized the formal 21-electron cobaltocene derivative 1 by utilizing the chelate effect of the CpNCp pincer ligand, which enables the coordination of the nitrogen lone pair donor while maintaining the η5-coordination mode of two Cp groups. The structural analysis, quantum-chemical topological studies, and DFT calculations strongly support the presence of the Co–N bond and two η5-coordinated Cp ligands, and thus the formation of the 21-electron cobaltocene derivative. The stability and unexpected redox property of the 21-electron complex 1 originate from the half-filled Co−N and Co⋅⋅⋅Cp antibonding orbitals, and the chelation effect of the pincer ligand. The formal 21-electron metallocene presented here expands our understanding and perception of metallocene and organometallic chemistry, as well as chemical bond theory. It is anticipated that this discovery will open up unexplored possibilities in transition metal chemistry, especially in the field of catalysis and materials science.

Methods

General considerations and materials for the synthetic study

All reactions were carried out under an N2 atmosphere using an MBRAUN glovebox, UNILAB Plus SP, equipped with an MB-20-G gas purifier, an MB-LMF-2/40-REG regenerable solvent trap, and an MB-GS-35 -35 °C freezer. All glasswares were dried overnight at 170 °C and cooled down under vacuum in the glovebox antechamber. All solvents were reagent grade or higher. n-Pentane ( ≥ 99.0%), hexane ( ≥ 95.0%, n-hexane with minor amounts of isomers of methylpentane and methylcyclopentane), n-heptane ( ≥ 99.0%), dichloromethane ( ≥ 99.5%), benzene ( ≥ 99.7%), toluene ( ≥ 99.8%), acetonitrile ( ≥ 99.8%), methanol ( ≥ 99.8%), tetrahydrofuran (Sigma-Aldrich 401757-1 L, anhydrous, ≥99.9%, inhibitor-free), diethyl ether ( ≥ 99.8%, inhibitor-free), hexamethyldisiloxane (Sigma-Aldrich 205389-500 ML, ≥98%), and dimethylformamide (DMF, ≥99.0%) were dried over MS3A (dried overnight in a 200 °C oven and cooled overnight under vacuum in the glovebox antechamber) in the glovebox for more than 2 days and stored in the glovebox. Common chemicals were purchased, kept in the glovebox, and used as received unless stated otherwise. Anhydrous CoBr2 (green powder, >97%) was purchased from Fujifilm Wako Chemicals. NiCl2(DME) (yellow powder, 98%) was purchased from Sigma Aldrich. MnBr2 (pink powder, 99%) was purchased from Acros Organics. Celite filtration was carried out using a pipet, cotton wool, and Celite®545, which was dried overnight in a 170 °C oven, cooled down overnight under vacuum in the glovebox antechamber, and kept in the glovebox. The dimensions of the 20 mL vial are 60 mm in height, 28 mm in outer diameter, and the Teflon-coated stirring bar is 15 mm long and 5 mm in diameter.

Instrumental analysis methods

NMR spectroscopy

All deuterated solvents were purchased and dried over MS3A in the glovebox for more than 2 days and kept in the glovebox. NMR spectra were recorded using a Bruker Avance III-400N spectrometer and an Avance III NEO 500 spectrometer equipped with a cryoprobe. 1H and 13C NMR chemical shifts are reported in parts per million (δ) relative to TMS (0 ppm) with the residual solvent signal (CDCl3: 7.26 (1H) and 77.16 (13C) ppm, C6D6: 7.16 (1H) and 128.06 (13C) ppm, THF-d8: 1.72 (1H) and 25.31 (13C) ppm, toluene-d8: 2.08 (1H) and 20.43 (13C) ppm, DMSO-d6: 2.50 (1H) and 39.52 (13C) ppm, methanol-d4: 3.31 (1H) and 49.00 (13C) ppm, CD3CN: 1.94 (1H) and 118.26 (13C) ppm) as the internal references 2H NMR chemical shifts are reported in parts per million (δ) relative to residual solvent signal as the internal reference. 11B, 15N, and 19F NMR chemical shifts are reported in parts per million (δ) relative to BF3⋅OEt2 in CDCl3 (0 ppm), NH3(liquid) (0 ppm), and CFCl3 (0 ppm), respectively, as external references.

NMR peak assignments of diamagnetic compounds were made using 1H-1H-gCOSY, 1H-13C-HSQC, and 1H-13C-HMBC NMR experiments. Abbreviations for NMR spectra are s (singlet), d (doublet), t (triplet), q (quartet), quint (quintet), sep (septet), dd (doublet of doublet), td (doublet of triplet), dq (doublet of quartet), m (multiplet), and br (broad). 1H NMR signals of paramagnetic complexes are reported with chemical shift (δ) and line width at half-height (Δν1/2). Air-sensitive NMR samples were prepared in the nitrogen glovebox using a J. Young NMR tube or a standard NMR tube sealed with a septa and parafilm.

NMR measurement of paramagnetic complexes

1H NMR of paramagnetic complexes (typically 20 mM solution) was measured with delay time (d1) of 0.1 s, time domain data points (td) of 128k, and scan number (ns) of 1 to 128, and spectral width (sw) of 200 to 600 ppm. Proton-coupled 13C NMR experiments were carried out using an Avance III NEO 500 spectrometer equipped with a cryoprobe. For proton coupled 13C NMR experiments, a delay time (d1) of 0.1 s, a time domain data points (td) of 64k, a scan number (ns) of 64k, and a spectral width (sw) of 1100 ppm were used.

Measurement of effective magnetic moment by Evans’ method

Measurement of the effective magnetic moment by Evans’ method was carried out using glass capillaries containing C6D6 (measurement in C6D6) as external standards.

Single-crystal X-ray diffraction (SC-XRD)

The X-ray diffraction experiments for 1-3 were performed on a Bruker D8 Venture diffractometer equipped with a PHOTON II CPAD detector and an IµS 3.0 microfocus X-ray source (Mo Kα radiation). The X-ray diffraction data for the single crystal 4 were collected on a Rigaku XtaLab PRO instrument equipped with a PILATUS3 R 200 K hybrid pixel array detector and a MicroMaxTM-003 microfocus X-ray tube (Mo Kα). Data were collected at 100 K according to recommended strategies, then processed and corrected. All structures were solved using SHELXT60. Structures 2-4 were refined by the full-matrix least-squares using SHELXL61. Non-hydrogen atoms were refined anisotropically.

VSM measurement

The VSM measurement was carried out using the Quantum Design PPMS DynaCool VSM module. VSM powder sample holders (part #: 4096-388) were weighed outside the glovebox using a microbalance and placed inside in the glovebox. All powdered samples were packed and sealed in the sample holders in the nitrogen glovebox. The sealed samples were weighed outside the glovebox using microbalance to calculate the weight of the samples.

FTIR spectroscopy

IR spectra were recorded using a Nicolet iS5 FT-IR spectrophotometer and are reported in absorption frequency (cm–1). Abbreviations for FT-IR spectra are s (strong), m (medium), and w (weak).

High-resolution mass spectrometry (HRMS)

HRMS data were recorded on a Thermo Scientific LTQ-Orbitrap mass spectrometer, using electrospray ionization (ESI) mode.

Elemental analyses

Elemental analyses were conducted using an Exeter Analytical CE-440 elemental analyzer. Empty tin cups were weighed outside the glovebox using a microbalance and brought in the glovebox. All samples were weighed and sealed in the tin cups in the nitrogen glovebox. The sealed samples were weighed outside the glovebox using microbalance to calculate weight of the samples. The N2 gas in the tin cups was replaced by argon through three vacuum-argon refill cycles. All the samples were analyzed using an autosampler under a He atmosphere.

Cyclic voltammetry (CV)

CV was measured inside the nitrogen glovebox using an ECstat-301WL potentiostat equipped with a glassy carbon working electrode (3.0 mm diameter), a platinum wire counter electrode (0.5 mm diameter), and an Ag/Ag+ reference electrode (Ag wire in 0.01 M AgNO3 in 0.1 M [Bu4N]PF6 MeCN solution). The sample solution was prepared by dissolving an appropriate sample in 0.1 M [Bu4N]NPF6 THF solution. All potentials are reported using the FeCp20/+ couple (0 mV) as an external reference.

Computational Methods

DFT computations were performed with the ORCA software package (versions 5.0.0, 5.0.1, 5.0.3) unless otherwise noted62,63,64. For structure optimizations and harmonic vibrational frequencies, we employed the TPSS meta-GGA functional65,66,67. Additionally, we utilized Grimme’s latest additive dispersion correction D468,69 and an Ahlrichs-type basis set with triple-zeta quality (def2-TZVPP)70. Minima on the potential energy surface were confirmed by the absence of imaginary frequencies. For accurate electronic energies and quasi-restricted orbitals (QROs)56 we used the hybrid variant of this functional, i.e., TPSSh, together with the quadruple-zeta quality basis set (def2-QZVPP)70. We tested the influence of relativistic effects by incorporating the zeroth-order regular approximation (ZORA) at the scalar spin-free level of the theory71. As recommended, these calculations use a relativistically recontracted basis set (ZORA-def2-XVP) and a decontracted auxiliary basis set (SARC/J)72. Unless noted otherwise, the resolution of identity (RI) and chain-of-spheres (RIJCOSX) approximations, as implemented in ORCA, were used for (meta)-GGA and hybrid functionals, respectively, together with the appropriate auxiliary basis set (def2/J)72,73. All computations performed with ORCA utilized the tight SCF convergence criteria and the default integration grid (defgrid2).

Wiberg bond indices were determined in the framework of natural bond orbital (NBO 7.0.10) analysis74. The topological analysis of the theoretical electron density \(\rho ({{{{{\bf{r}}}}}})\) and electrostatic potential \({\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\) was performed with Multiwfn 3.8(dev)75. For this purpose, we carried out single-point energy computations at TPSSh/def2-QZVPP level using the Gaussian16 software package (revision C.01) to generate a Gaussian checkpoint file as input (tight SCF convergence criteria, ultrafine integration grid)76. The electrostatic potential \({\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\) was evaluated using the built-in code77. The protocol for generating superposition maps of the zero-flux surfaces defined in the gradient vector fields \(\nabla \rho ({{{{{\bf{r}}}}}})\) and \(\nabla {\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\) was described in detail earlier78.

Quantum crystallography

The X-ray diffraction data of 1 were collected at 100 K with a reciprocal resolution \({{\sin }}({\theta }_{{{\max }}}/\lambda )\) of 1.26 Å–1. The multipole refinement of 1 was performed within the Hansen–Coppens formalism as implemented in MoPro79. The anharmonic atomic motion of Co was modeled using the Gram–Charlier expansion of the temperature factors. A block refinement of the charge density parameters and the Gram–Charlier coefficients was applied analogously to the procedure described earlier80.

The analysis of the multipole-derived electron density \(\rho ({{{{{\bf{r}}}}}})\) and electrostatic potential \({\varphi }_{{{{{{\rm{es}}}}}}}({{{{{\bf{r}}}}}})\) was performed in WinXPRO81 analogous to the published procedures82. The energy density \(h({{{{{\bf{r}}}}}})\) was approximated according to Kirzhnits83.