Main

The discovery of ferrocene has undoubtedly revolutionized organometallic chemistry, as metallocenes have shaped various areas of chemistry and have become standard textbook knowledge, nowadays1,2,3. Not only has their unexpected bonding situation revolutionized bonding theory by introducing the concept of a ‘sandwich complex’, but their properties have also attracted much attention and made them an extremely important class of compounds for various fields, including catalysis, materials chemistry, bio-medical applications and beyond4,5,6,7,8,9. Over the years, metallocene-type compounds—species of the general formula ‘(η5-Cp)2[M]’ (Cp = cyclopentadienyl; [M] = metal centre)—have been described for many elements across the periodic table10,11,12,13,14,15,16. Unlike these monometallic derivatives, the term dimetallocene refers to very rare sandwich complexes in which two metal atoms are bonded between the η5-coordinated Cp ligands arranged in linear/coplanar fashion and interlinked by a metal–metal bond. The synthesis of homobimetallic decamethyldizincocene (Cp*2Zn2) by Carmona and co-workers in 2004 was a paradigm-shifting milestone of modern organometallic chemistry17,18, as preceding reports on dimetallocenes did not provide suitable evidence and were subsequently shown to be erroneous19,20,21. Additionally, bimetallic complexes in which two metal centres are bridged by halides, hydrides or hydroxy, carbonyl, aryl or alkyl groups are well known22,23. Nevertheless, different transition metals and main-group elements have been theoretically predicted to form stable dimetallocenes24,25,26,27,28, yet dizincocene remained the only experimentally characterized example until very recently, when the likewise homobimetallic diberyllocene (Cp2Be2) was described by Boronski and Aldridge (Fig. 1)29. On the other hand, dimetallocenes of p-block elements are still unknown, although numerous attempts to isolate a dimetallocene of silicon were made but were all unsuccessful due to disproportionation of the alleged decamethyldisilicocene into decamethylsilicocene (Cp*2Si) and silicon(0) (refs. 30,31). Notably, heterobimetallic dimetallocenes have remained elusive so far, although they have been theoretically studied since nearly two decades32,33,34. Combinations of group 1 and group 13 metals were proposed as intermediates, but—despite considerable efforts—have never been detected let alone isolated. For example, Timoshkin and Schaefer speculated that the experimental observation of CpBn5Li in the reaction of CpBn5Al and Cp*Li might be explained by the occurrence of a weakly bonded donor–acceptor complex of the type CpBn5Al→LiCpBn5 (refs. 32,35).

Fig. 1: Overview of dimetallocenes.
figure 1

Top: reported homobimetallic dimetallocenes. Bottom: synthetic strategy to the lithium–aluminium heterobimetallic dimetallocene (this work).

Full size image

Our synthetic strategy towards a heterobimetallic group 1 group 13 dimetallocene relied on the isolation of a cyclopentadienylaluminylene (Fig. 1), a species whose crystal structure has been elusive for almost three decades36. Schnöckel’s report of (pentamethylcyclopentadienyl)aluminium(I) in 1991 demonstrated the ability of cyclopentadienyl ligands to stabilize aluminium(I) centres, which have since become common in low-valent aluminium chemistry37,38. Noteworthily, this compound exists in tetrameric form in the solid state and in solution at room temperature, while monomeric cyclopentadienylaluminylenes are accessible at elevated temperatures, or with more bulky substitution patterns on the Cp moiety36. The isolation of a monomeric cyclopentadienylaluminylene was reported recently by Braunschweig and co-workers, but its monomeric nature was confirmed solely NMR spectroscopically from the crude product in solution39. Thus, structural authentication of a monomeric cyclopentadienylaluminylene has remained elusive, although a few monomeric aluminylenes with σ-bonded substituents have been structurally characterized40,41,42,43,44,45,46,47.

In this Article, we report the synthesis, isolation and structural characterization of a monomeric cyclopentadienylaluminylene taking advantage of the sterically very demanding pentaisopropylcyclopentadienyl (5Cp) ligand23. Even more importantly, reaction of (5Cp)aluminylene 1 with (5Cp)lithium, 5CpLi, results in the formation of the heterobimetallic dimetallocene 2 with a unique dative metal–metal bond.

Results and discussion

Aluminylene 1

Following the report of Schnöckel and co-workers, the (pentamethylcyclopentadienyl)aluminium(I) tetramer can be utilized as a precursor for the synthesis of cyclopentadienylaluminylenes38,48; we reacted it with (5Cp)lithium diethyletherate, 5CpLi∙OEt2, and obtained the corresponding (5Cp)aluminylene, 1 (Fig. 2a).

Fig. 2: (5Cp)aluminylene 1.
figure 2

a, Synthesis of aluminylene 1. b, The molecular structure of 1 in the crystal (displacement ellipsoids at 50% probability level, hydrogen atoms omitted for clarity, iPr groups drawn as ball-and-stick models). Selected experimental and theoretical [M06-2X/def2-SVP] bond lengths: Al1–Cpcent: 196.71(6) [199.1] pm. c, Selected Kohn–Sham frontier molecular orbital contours of 1 (M06-2X/def2-TZVPP//M06-2X/def2-SVP; isodensity 0.05 a.u.). ε, orbital energy.

Full size image

As shown by the groups of Schnöckel and Braunschweig, monomeric cyclopentadienylaluminylenes have 27Al NMR chemical shifts in the range of −150 to −170 ppm, while the tetrameric aggregates exhibit more downfield-shifted resonances, usually in the range of −60 to −110 ppm (refs. 36,39). This is due to increased π type bonding interactions between the aluminium atom and the Cp ligand in the monomeric species, which results in an energetically higher lying lowest unoccupied molecular orbital (LUMO) and, thus, a larger highest occupied molecular orbital (HOMO)–LUMO gap and smaller paramagnetic contribution to the 27Al NMR chemical shift49,50. Accordingly, 1 exhibits a 27Al NMR chemical shift of δ27Al{1H}(C6D6) = −154 (ω½ = 521 Hz) in solution and of δ27Al(SPE/MAS(13 kHz)) = −154 (SPE = single pulse excitation; MAS = magic-angle spinning) in the solid state at ambient temperatures, clearly indicating its monomeric nature both in the solid state and in solution (Supplementary Figs. 3 and 4). Crystals of 1, suitable for single-crystal X-ray diffraction (XRD), were obtained by sublimation of the compound in vacuum at 323 K. The crystal structure of 1 reveals well-separated, monomeric (5Cp)aluminium moieties (Fig. 2b and Supplementary Fig. 36). The closest contact from the aluminium centre to a neighbouring molecule is to an H atom of a methyl group, which is 316.45(5) pm. The aluminium atom is η5-coordinated by the cyclopentadienyl moiety leading to an overall pentagonal pyramidal structure. Following polyhedral skeletal electron pair theory, more commonly referred to as the Wade–Mingos rules51,52,53, 1 can be classified as a nido cluster. The Al–Cpcentroid and Al–CCp bond lengths in 1 are slightly shorter than those in {Cp*Al}4 (Table 1). This originates from the increased Al–Cp bonding interaction in 1 due to its monomeric nature, as also apparent from the 27Al NMR chemical shift (vide supra)49,50. We analysed the electronic structure of 1 within the density-functional theory (DFT) framework, whereby the equilibrium geometry is in very good agreement with the structure determined by single-crystal XRD (Fig. 2b), with the Al–Cpcentroid distance being slightly longer than those observed experimentally. Similar to former theoretical calculations54,55, the Kohn–Sham frontier molecular orbitals of 1 (Fig. 2c) consist of a lone pair at the Al atom (HOMO) and two degenerated 3p orbitals (LUMO) at the Al atom, respectively. Moreover, the natural population analysis (NPA) and Bader’s quantum theory of atoms in molecules (QTAIM)56 show that the aluminium atom is positively charged by +0.70 a.u. (NPA)/+0.81 a.u. (QTAIM) (Supplementary Fig. 44), indicating a relatively ionic bonding interaction between the aluminium atom and the Cp ligand.

Table 1 Selected bond length for {Cp*Al}4; 1; 1∙AlBr3; 1∙W(CO)5 and 2
Full size table

Aluminylene complexes 1∙AlBr3 and 1∙W(CO)5

Due to the lone pair of the aluminium atom, sterically less demanding cyclopentadienylaluminylenes are known to act as donors towards electron-deficient acceptors39,44,47,57,58,59,60. Thus, to explore the reactivity of the sterically very encumbered 1, we initially investigated the donor ability of 1 towards electrophiles, which are known to coordinate to Cp*Al and Cp′′′Al. Treatment of 1 with one equivalent of aluminium tribromide indeed affords the corresponding adduct 1∙AlBr3. With tungsten hexacarbonyl under ultraviolet irradiation, the corresponding aluminylene tungsten complex 1∙W(CO)5 was formed. Single crystals of both compounds were obtained and allowed for structural characterization in the solid state by XRD (Supplementary Figs. 37 and 38). Complex 1∙AlBr3 exhibits an Al–Al bond length of 255.5(1) pm, which is similar to the bond in Cp′′′Al→AlBr3 (255.4(1) pm)39, suggesting similar donor abilities of Cp′′′Al and 1. A cyclopentadienylaluminylene tungsten carbonyl complex had not been described previously, although the analogous Cp*Al→Cr(CO)5 complex and other organoaluminium(I) tungsten carbonyl complexes are known44,59. The Al–W bond length in 1∙W(CO)5 amounts to 258.5(1) pm, which is longer than in a carbazolylaluminylene tungsten pentacarbonyl complex (253.6(1) pm)44, but shorter than those reported for (TMEDA)(R)Al→W(CO)5-type compounds (TMEDA = tetramethylethylenediamine; R = Cl: 264.5(2) pm; R = Et: 267.0(1) pm; R = tBu: 274.1(4) pm)60. The carbonyl vibration bands in the infra-red spectrum of 1∙W(CO)5 are similar to those of the carbazolylaluminylene tungsten complex, hinting to equal donor strength (Supplementary Fig. 30). Interestingly, the Al–Cpcentroid distances in 1∙AlBr3 (178.3(9) pm) and 1∙W(CO)5 (183.5(6) pm) are substantially shortened compared with what is observed in uncomplexed 1 (196.7(6) pm), which is a result of the electron deficiency at the aluminium(I) centre influenced by the electron-withdrawal power of the coordinated metal fragment, and the corresponding compensation by the 5Cp ligand.

Dimetallocene 2

The clearly apparent ability of 1 to act as a donor ligand despite the bulky 5Cp group suggested it as an excellent candidate for the deliberate synthesis of a heterobimetallic dimetallocene, inspired by theoretical predictions, as well as reports of aluminylene lithium complexes with σ-bonded substituents32,33,34,61. Indeed, the reaction of 1 with one equivalent of 5CpLi (in the presence of Cp*Li as diethyl ether scavenger) resulted in the formation of lithium–aluminium heterobimetallic dimetallocene 2 (Fig. 3a). Single crystals of 2 were analysed by XRD unambiguously proving the structure of 2 in the solid state (Fig. 3b).

Fig. 3: Lithium–aluminium dimetallocene 2.
figure 3

a, Synthesis of dimetallocene 2. b, The molecular structure of 2 in the crystal (side view and top view, displacement ellipsoids at 50% probability level, hydrogen atoms omitted for clarity, iPr groups drawn as ball-and-stick models). Selected experimental and theoretical [M06-2X/def2-SVP] bond lengths: Li1–Cpcent: 176.2(3)–176.6(1) [170.8] pm, Al1–Cpcent: 188.6(1)–189.3(1) [190.9] pm, Al1–Li1: 261.5(2) [265.8] pm. c, Laplacian distribution ∇2ρ(r) of 2 (M06-2X/de2-TZVPP//M06-2X/def2-SVP). Dashed red lines indicate areas of charge concentration (∇2ρ(r) < 0); solid blue lines indicate areas of charge depletion (∇2ρ(r) > 0) (bond ellipticity: 0.0). DI, delocalization index; Q, partial charge. d, Molecular orbital interaction diagram in eV for the Al–Li σ-bond in 2 (M06-2X/de2-TZVPP//M06-2X/def2-SVP; isodensity 0.05). e, DID plot (LMP2/cc-pVTZ). Orange/yellow/green zones indicate strong dispersion interactions, and blue/turquoise zones indicate weaker/diffuse contributions.

Full size image

Two crystal structures of dimetallocene 2 could be obtained, co-crystalized with toluene or with 1,2-difluorobenzene. From toluene, 2 crystallizes in the triclinic space group P1 with one formula unit and two molecules of toluene per asymmetric unit. Due to the high symmetry of the molecule, a positional disorder of the Al and Li positions of 93:7 is observed. The 5Cp ligands are both bonded in η5 fashion and adopt a staggered conformation, interestingly unlike in dizincocene and diberyllocene, where eclipsed conformations are observed17,18,29. This might be caused by steric pressure and/or packing effects. The Al–Li bond length is 261.5(2) pm and, thus, substantially shorter than in ionic aluminyl lithium complexes, for example, (NON)Al→Li(Et2O)2/{(NON)Al→Li}2: 274.6(3) pm to 276.7(2) pm (NON = 4,5-bis(2,6-diisopropylanilido)-2,7-di-tert-butyl-9,9-dimethyl-xanthene) reported before. Noteworthily, these examples are not only not structurally related to 2 but are also ionic in nature (‘[RAl]−→[Li]+’), while 2 consists of two formally neutral fragments (‘RAl→LiR’) and is, therefore, essentially without precedents62. Nonetheless, the Al–Li bond length is in good agreement with the predicted sum of the covalent radii of Al and Li (∑rcov(Al + Li) = 259 pm)63. The 7Li and 27Al NMR chemical shifts of 2 in solution are δ7Li = −9.63 and δ27Al = −151 (ω½ = 1,139 Hz), which are only slightly different from the 27Al NMR chemical shift of 1 (δ27Al = −154) and the 7Li NMR chemical shift of 5CpLi∙OEt2 (δ7Li = −8.18), hinting at a weak Al–Li interaction. In the solid state, 2 reveals similar NMR chemical shifts (δ7Li(SPE/MAS(13 kHz)) = −8.9; δ27Al(SPE/MAS(13 kHz)) = −157), with the signal in the 7Li NMR spectrum split to a hexet with a coupling constant of \({}^{1}{\mathrm{J}}_{{}^{7}{\rm{L}}{\rm{i}}-{}^{27}{\rm{A}}{\rm{l}}}=102\,{\rm{Hz}}\), clearly reflecting the Al–Li bonding interaction (Supplementary Figs. 12 and 14).

To gain further insight into the electronic structure of 2, we performed DFT calculations. As the nature of the Al–Li bond is of particular interest, we analysed the topology of the electron density with QTAIM56. The Bader analysis reveals a high charge concentration on the aluminium basin, which agrees with a lone pair (Fig. 3c). The bond critical point (BCP) of the bond path, which connects the Al with the Li atom, reveals low electron density (ρ(r)BCP = 0.10 e Å−3) with a large positive Laplacian value (∇2ρ(r) = +1.10 e Å−5) and positively charged Al by +0.84 a.u. and Li by 0.81 a.u. Moreover, the delocalization index of the Al–Li pair is rather low (0.09), and NPA also reveals positively charged Al by +0.73 a.u. and Li by +0.78 a.u. These features are typically found in ionic interactions and are similar to other reported ionic aluminyl lithium complexes. For example, NBO-derived natural atomic charges of (NON)Al→Li(Et2O)2 are +0.69 for Al and +0.73 for Li (ref. 61). Additionally, QTAIM analysis of ionic (NON)Al→Li(Et2O)2 complexes reveals BCPs for the Al–Li bond with even lower electron densities of ρ(r)BCP = 0.019 to 0.018 e Å−3 (refs. 61,62), indicating a more covalent character of the Al–Li interaction in 2. The relatively higher stability of 2 compared with former predictions originates from attractive dispersion interactions of the 5Cp ligands, as shown by the energy decomposition analysis (EDA) method (Table 2 and Supplementary Table 3)64. For comparison, we also performed EDA of the theoretical Cp and Cp* derivatives, as well as for Carmona’s Cp*2Zn2 and Boronski–Aldridge’s Cp2Be2. The bond dissociation energy for the Al–Li bond in 2 (De = 93.4 kJ mol−1) is notably higher than for the Cp (De = 42.7 kJ mol−1) and Cp* (De = 55.8 kJ mol−1) analogues (Supplementary Table 3). An examination of the ΔEint components for 2 suggests that dispersion interactions (ΔEdisp = −65.3 kJ mol−1, 37.7%) and electrostatic interaction (ΔEelst = −64.5 kJ mol−1, 37.2%) are almost identical in magnitude, while the orbital interactions (ΔEorb = −43.6 kJ mol−1, 21.5%) are smaller. The orbital interaction primarily corresponds to the donation of the lone pair of the aluminium atom of the 5CpAl fragment into the formally vacant orbital of the lithium atom of the 5CpLi fragment (Fig. 3d). The attractive dispersion interactions between the 5Cp ligands in 2 apparently play a major role to stabilize the dimetallocene, and are larger than in the Cp (−10.7 kJ mol−1) and Cp* (−16.8 kJ mol−1) analogues (Supplementary Table 3). To examine the origins of these dispersion forces, we performed energy partitioning, using local correlation methods LMP2/cc-pVTZ65. Within this method, the dipole–dipole moment interactions are quantified as the amplitude of pair excitations on localized orbitals of each fragment65. The dispersion interaction density (DID) plot (Fig. 3e)66 reveals dominating interactions between the isopropyl groups, namely C–H/C–H contacts, while the π–π interactions between the Cp rings are rather weak. For comparison, in Carmona’s Cp*Zn–ZnCp* and Boronski–Aldridge’s CpBe–BeCp, the homolytic fragmentations disclose more than three times larger dissociation energy (Cp*2Zn2: De = 302.9 kJ mol−1; Cp2Be2: 292.3 kJ mol−1) than in 2 (Table 2 and Supplementary Table 3)25,67. Furthermore, EDA of the dizincocene and diberyllocene shows that the stabilization interactions in these compounds are mainly the orbital (Cp*2Zn2: 41.2%; Cp2Be2: 32.0%) and electrostatic (Cp*2Zn2: 54.5%; Cp2Be2: 64.4%) terms, while attractive dispersion interactions play almost no role (Cp*2Zn2: 4.3%; Cp2Be2: 3.4%). These results clearly highlight the importance of the isopropyl groups of the 5Cp ligand to stabilize 2. While 2 exhibits a polar dative bond, originating from the lone pair at the aluminium atom donating to a vacant orbital at the lithium atom (Supplementary Fig. 45), it is formally valence-isoelectronic to dizincocene and diberyllocene, which exhibit unpolar electron sharing bonds.

Table 2 EDA results (BP86-D3(BJ)/TZ2P//M06-2X/def2-SVP) for the Al–Li bonds in 2, CpAl→LiCp and Cp*Al→LiCp*, the Zn–Zn bond in Cp*Zn–ZnCp* and the Be–Be bond in CpBe–BeCp
Full size table

Reactivity studies of dimetallocene 2

The computational investigations of 2 suggested that the Al–Li bond in 2 is fairly weak, compared with the metal–metal bond in dizincocene. To investigate this experimentally, we reacted 2 with an N-heterocyclic carbene (NHC), as in the case of decamethyldizincocene coordination of the NHC to one of the zinc atoms is observed, without cleavage of the Zn–Zn bond68. In contrast, a cleavage of the Al–Li bond in 2 is observed and the 5CpLi∙NHC complex 3 was isolated (Fig. 4b), which agrees with the DFT calculations that suggested the Al–Li bond to be rather weak and enforced by attractive dispersion interactions (vide supra). 3 exhibits a δ7Li shift of −9.07 ppm, which is similar to other cyclopentadienyl lithium NHC complexes69,70, as well as a Li–C1 bond length of 216.6(2) pm and a Li–Cpcentroid distance of 184.9(2) pm, which are in the same range as in other cyclopentadienyl lithium NHC complexes69,70. Next, 2 was reacted with three different heteroallenes, namely phenylisocyanate, mesitylisothiocyanate and 1-azidoadamantane. These reactions did also result in cleavage of the Al–Li bond, as complexes 5CpLi∙CNPh, 4a, 5CpLi∙CNMes, 4b, and {5CpAlNAd}2, 5, were formed and crystals suitable for single-crystal XRD were obtained (Fig. 4c–e). The Li–C1 bond lengths of 210.4(2) pm (4a) and 210.0(8) pm (4b) are similar to other lithium isocyanide complexes71, and the bond lengths in 5 are relatively similar to an analogue complex reported by Braunschweig39. The chalcogen-transfer products in the formation of 4a and 4b eluded isolation and characterization, but based on DFT calculations (Supplementary Fig. 47) and previous reports of trimeric {Cp′′′AlO}3 (ref. 39), dimeric or trimeric compounds of the type {5CpAlCh}n might be formed. We also performed control experiments in which we reacted aluminylene 1 with phenylisocyanate and mesitylisothiocyanate, but these experiments only yielded large amounts of pentaisopropylcyclopentadiene (5CpH) yet no isolatable amount of any {5CpAlCh}n species. Interestingly, treatment of 1 with 1-azidoadamantane did not result in the formation of 5, but gave a mixture of products also containing large amounts of 5CpH, indicating that the reactivity of aluminylene 1 and dimetallocene 2 towards 1-azidoadamantane differs.

Fig. 4: Reactivity of dimetallocene 2.
figure 4

a, Reaction of 2 with an NHC to give 3, PhNCO and MesNCS to give 4a,b, and AdN3 to give 5. b, The molecular structure of 3 in the crystal. c, The molecular structure of 4a in the crystal. d, The molecular structure of 4b in the crystal. e, The molecular structure of 5 in the crystal (displacement ellipsoids at 50% probability level, hydrogen atoms omitted for clarity, iPr groups drawn as ball-and-stick models). Selected experimental bond lengths: 3: Li1–C1: 216.6(2) pm, Li1–Cpcent: 184.9(2) pm; 4a: C1–Li1: 210.4(2) pm, Li1–Cpcent: 170.7(5) pm; 4b: C1–Li1: 210.0(8) pm, Li1–Cpcent: 153.1(7) pm; 5: Al1/2–N1/2: 181.7(2)–182.3(2) pm, Al1/2–CCp: 202.5(3)/202.8(3) pm, N1/2–CAd: 146.5(3)/147.3(3) pm.

Full size image

Conclusion

With the isolation of a monomeric cyclopentadienylaluminylene, 1, the stage was set for the synthesis of a heterobimetallic dimetallocene. The lithium–aluminium dimetallocene 2, while formally valence-isoelectronic to dizincocene and diberyllocene, exhibits a highly polar Al–Li bond, which is enforced by attractive dispersion interactions. As the Al–Li bond is relatively weak, it can be cleaved easily by donor molecules such as an NHC or in reactions with heteroallenes, such as phenylisocyanate, mesitylisothiocyanate and 1-azidoadamantane. These reactions resulted in the formation of (5Cp)lithium complexes 3, 4 and dialumazine 5. The cleavage of the Al–Li bond is in sharp contrast to the related valence-isoelectronic dizincocene, in which the Zn–Zn bond is perpetuated upon coordination of an NHC.