Molecular and supported Ti(iii)-alkyls: efficient ethylene polymerization driven by the π-character of metal–carbon bonds and back donation from a singly occupied molecular orbital

While Ti(iii) alkyl species are the proposed active sites in Ziegler–Natta ethylene polymerization catalysts, the corresponding well-defined homogeneous catalysts are not known. We report that well-defined neutral β-diiminato Ti(iii) alkyl species, namely [Ti(nacnac)(CH2tBu)2] and its alumina-grafted derivative [(AlsO)Ti(nacnac)(CH2tBu)], are active towards ethylene polymerization at moderate pressures and temperatures and possess an electron configuration well-adapted to insertion of ethylene. Advanced EPR spectroscopy showed that ethylene insertion into a Ti(iii)–C bond takes place during polymerization from Ti(nacnac)(CH2tBu)2. A combination of pulsed EPR spectroscopy and DFT calculations, based on a crystal structure of [Ti(nacnac)(CH2tBu)2], enabled us to reveal details about the structure and electronic configurations of both molecular and surface-grafted species. For both compounds, the α-agostic C–H interaction, which involves the singly occupied molecular orbital, indicates a π character of the metal–carbon bond; this π character is enhanced upon ethylene coordination, leading to a nearly barrier-less C2H4 insertion into Ti(iii)–C bonds after this first step. During coordination, back donation from the SOMO to the π*(C2H4) occurs, leading to stabilization of π-ethylene complexes and to a significant lowering of the overall energy of the C2H4 insertion transition state. In d1 alkyl complexes, ethylene insertion follows an original “augmented” Cossee–Arlman mechanism that involves the delocalization of unpaired electrons between the SOMO, π*(C2H4) and σ*(Ti–C) in the transition state, which further favors ethylene insertion. All these factors facilitate ethylene polymerization on Ti(iii) neutral alkyl species and make d1 alkyl complexes potentially more effective polymerization catalysts than their d0 analogues.


Introduction
Since the discovery of Ziegler-Natta ethylene polymerization catalysts in the early 1950s, 1,2 the nature of the active site(s) has been a matter of debate. Later, group IV transition-metal metallocenes were developed as efficient homogeneous, 3-7 as well as supported, 8,9 olen polymerization catalysts for which cationic M(IV) alkyl species have been proposed as the active sites. 10 Such species have been isolated in the form of Lewis base adducts and have been shown to be competent in olen polymerization; one noteworthy example is Ti(IV) amidinate species [(Cp*)Ti {NC(Ph)N i Pr 2 }(OPPh 3 )Me][BArF 4 ] (Scheme 1a). 11 Taking into account the strong ionic character of MgCl 2 , the key support of Ziegler-Natta catalysts, surface Ti(IV) cationic alkyl species are sometimes proposed as the active sites in these systems in analogy to their metallocene equivalents. 12 However, the evidence for Ti(IV) cationic species in Ziegler-Natta heterogeneous catalysts has remained elusive. In fact, the reaction with triethyl aluminum, a common cocatalyst used in the Ziegler-Natta process, undoubtedly leads to reduction and/ or alkylation of certain sites, and Ti(III) centers have previously been observed by XPS and EPR. [13][14][15][16][17][18] Since their role as active sites in the polymerization of ethylene has not been evidenced so far, it remains unclear whether or not titanium d 1 complexes can be efficient in ethylene polymerization. 19,20 In parallel, cationic monocyclopentadienyl Ti(III) compounds show high activities towards styrene polymerization and are catalytically comparable to or better than the corresponding Ti(IV) derivatives (Scheme 1b). 21 Furthermore, it has been recently shown by pulsed EPR spectroscopy, combined with DFT calculations, that Ti(III)-alkyl surface species are formed when silica-supported titanium(III) hydride comes into contact with ethylene. 22 This nding is consistent with previous calculations that show Ti(III) hydrides as competent for initiating ethylene polymerization; 23 it further implies that Ti(III) d 1 alkyl complexes could polymerize olens such as ethylene.
Among well-dened Ti(III) alkyl species, the neutral b-diiminato Ti(III) dialkyl species (Scheme 1c, 1) are noteworthy for their stability. 24,25 While reported to be unreactive towards ethylene polymerization without a cocatalyst at room temperature under 6 bar of ethylene, 24 the presence of two alkyl ligands in this compound makes it particularly attractive to generate a supported Ti(III) alkyl catalyst through surface organometallic chemistry (SOMC) 26 and to evaluate its polymerization activity. Indeed, SOMC has been shown to provide access to many active and stable catalysts through isolation of metal sites at the surface of oxide supports. [26][27][28][29] In addition, the presence of strong Lewis acid surface sites, such as in alumina, can also help promote the formation of more active species. [30][31][32] Alternatively, cationic surface sites can be stabilized using sulfated metal oxides as a support. [32][33][34] Herein, we show that the molecular b-diiminato Ti(III) alkyl complex (Scheme 1c, 1) efficiently promotes ethylene polymerization without the need for co-catalysts at 80 C and pressures higher than 4 bar and that its alumina-supported analogue, prepared via the SOMC 26 approach (Scheme 1c, 1@Al 2 O 3-700 ), shows signicantly improved productivity. Using pulsed EPR spectroscopy, combined with DFT calculations and polymerization tests, we demonstrate that neutral Ti(III) alkyl species are indeed able to initiate ethylene polymerization to produce ultrahigh molecular weight polyethylene via ethylene insertion into a Ti(III)-C bond. Detailed DFT calculations show that ethylene insertion into the Ti(III)-C bond, a key step of ethylene polymerization, is favored by a partial alkylidenic character of the metal-carbon bond, as in its d 0 analogues, 35,36 and the added possibility of a partial electron transfer to the coordinated olen in Ti(III) compounds, which can be viewed as the p-back donation from the SOMO into the coordinated ethylene. This "augmented" Cossee-Arlman mechanism of olen polymerization, possible for d 1 metal-alkyl complexes, involves the delocalization of unpaired electrons in the transition state of olen insertion into the metal-carbon bond with a strong alkylidenic character. Bu or its 13 C-labeled analogue, Li 13 CH 2 t Bu, respectively, according to literature procedures. 25 Complex 1 was characterized by NMR (see Fig. S1 ‡) and roomtemperature CW EPR (see Fig. S2 ‡), consistent with its previous characterization. 25  Bu)] (Al s ¼ surface aluminium). Further characterization of 1@Al 2 O 3-700 by Fourier transform infrared spectroscopy (FTIR) shows the disappearance of the initial isolated hydroxyl groups and the appearance of a broad band from 3420 to 3760 cm À1 , associated with hydroxyl groups interacting with the ligands of the graed Ti centres (e.g. nacnac ligand). 37 These fragments are also revealed via n(C-H) vibrations at 3081-2874 cm À1 (Fig. S3 ‡).

Results and discussion
Both materials 1 and 1@Al 2 O 3-700 were then characterized with CW EPR spectroscopy. The X-band CW EPR spectrum of complex 1 recorded at 10 K (Fig. 1a, blue) originates from an S ¼ 1/2 electron spin system with a nearly axial g tensor, having its principal values g x ¼ 1.898 AE 0.023, g y ¼ 1.981 AE 0.016, and g z ¼ 1.996 AE 0.012 (the given intervals indicate Gaussian distributions of g principal values; the simulation of the EPR spectrum is shown in Fig. 1a, red). For the surface-graed material 1@Al 2 O 3-700 , the X-band CW EPR spectrum (Fig. 1b blue) was measured at 10 K; the spectrum is consistent with the presence of an S ¼ 1/2 system, with g x ¼ 1.880 AE 0.087, g y ¼ 1.970 AE 0.049 and g z ¼ 1.984 AE 0.016 (the simulation is shown in Fig. 1b, red) associated with a paramagnetic Ti(III) surface species. Signicant line broadening is observed in the graed material 1@Al 2 O 3-700 compared to molecular complex 1 which may result from the presence of different Ti 3+ surface species, possibly due to small differences in local surface environments and thereby coordination geometry. In spite of replacing one Ti-coordinated carbon atom by a more electronegative oxygen atom, the g principal values for both 1 and 1@Al 2 O 3-700 agree with each other within the given line widths, thus indicating a similar electronic structure and symmetry of molecular complex 1 and alumina-supported species 1@Al 2 O 3-700 .
Polymerization activity. We further examined the polymerization activity of 1 and its supported analogues. Molecular complex 1 was found to be active in ethylene polymerization in the temperature range 80-100 C and at ethylene pressures higher than 4 bar (Fig. 2a). The reaction was carried out in either benzene or toluene solutions. The formation of white lms of polyethylene (PE) was observed aer 2 hours. Within the range tested, a maximal calculated productivity of 11 kg PE (mol Ti h) À1 was achieved at 80 C under 7 bars of ethylene in toluene solution (Fig. 2a, green). Note that heating a toluene solution of 1 under the same conditions (80 C), but in the absence of ethylene, showed neither a white lm nor the formation of any new species visible by EPR or NMR spectroscopies (Fig. S4 ‡). We can therefore propose that 1 is a direct precursor of the active species that are formed under ethylene pressure.
The material 1@Al 2 O 3-700 is signicantly more active towards ethylene polymerization than its molecular analogue 1, and the polymerization stops within minutes due to the formation of a dense PE layer that can be directly observed with scanning electron microscopy (SEM) (Fig. 2c and d). Overall, this material displays a productivity of ca. 36 kg PE (mol Ti h) À1 under 6 bar of ethylene at 50 C. Note that, in contrast to 1, 1@Al 2 O 3-700 initiates ethylene polymerization even under very mild conditions, i.e. room temperature, 80 mbar of ethylene.
The molecular weight distribution for PE, produced by 1 and 1@Al 2 O 3-700 , is obtained with size-exclusion chromatography (SEC, Fig. 2b). For both catalysts, the distribution is asymmetric with a main heavy fraction and a broad distribution of molecular weights of lighter PE fractions, possibly being the products of chain termination reactions. The main fraction of PE has a molecular weight of ca. 1130 kg mol À1 for 1 and ca. 5660 kg mol À1 for 1@Al 2 O 3-700 , which is typical for ultra-high molecular weight polyethylene (UHMWPE). 38 As the PE molecular weight distribution has similar character for both 1 and 1@Al 2 O 3-700 , and as UHMWPE is produced in both cases, we propose that the polymerization on both catalysts takes place via a similar mechanism. The higher molecular weight of the main fraction of PE, produced with 1@Al 2 O 3-700 , can be explained by the absence and/or the slower rate of possible chain termination reactions, one example being the a-H abstraction reaction.
Detailed EPR characterization and evaluation of the structures of 1 and 1@Al 2 O 3-700 . We further characterize complex 1 and the associated surface species in 1@Al 2 O 3-700 by pulse EPR spectroscopy, namely by HYSCORE (Hyperne Sublevel Correlation Spectroscopy). 39 This method was selected for its ability to observe weak hyperne couplings (e.g. weakly coupled 14 N and 1 H) that are usually not resolved in the CW EPR spectra. The X-band HYSCORE spectra of 1 and 1@Al 2 O 3-700 , shown in Fig. 3b and d, respectively, were measured at 10 K at the eld positions corresponding to the maxima of the echo-detected EPR spectra ( Fig. 3a and c; the eld positions are marked with arrows). Both spectra shown in Fig. 3b and d were measured with an interpulse delay of s ¼ 128 ns. The X-band HYSCORE spectra were also measured with three s values s ¼ 128 ns, 160 ns and 224 ns to avoid loss of spectral information due to blind spots (see Fig. S5 ‡ for the s-summation spectra); however, it appeared that the spectra with s ¼ 128 ns contained all peak patterns present in the s-summation spectra except for the 1 H matrix peak, whose suppression is favorable.
The HYSCORE spectrum of complex 1 (Fig. 3b) shows the presence of 1 H and 14 N nuclei in the Ti(III) coordination sphere, revealed by cross peaks along the 1 H antidiagonal in the weak coupling (+, +) quadrant, corresponding to 1 H hyperne couplings, and by peaks in the low-frequency region both in the weak coupling (+, +) and strong coupling (À, +) quadrants, corresponding to 14 N hyperne and quadrupole couplings. For the material 1@Al 2 O 3-700 , the HYSCORE spectrum (Fig. 3d) reveals the presence of 14 N hyperne and quadrupole couplings as well, which are close to the ones observed for molecular complex 1 before graing. This indicates that the (nacnac)  Bu ligands upon graing, as shown in Scheme 1. Furthermore, 27 Al couplings are observed for 1@Al 2 O 3-700 as a matrix peak on the 27 Al antidiagonal line (Fig. 3d) and are well-resolved in Q-band HYSCORE spectra (Fig. 5c). This is consistent with the presence of nearby surface Al atoms in the surroundings of Ti(III), as expected for a graed species.
Using the experimental HYSCORE spectra, we could further estimate the conformation of molecular complex 1 in toluene solution and provide detailed structural information regarding the surface species in 1@Al 2 O 3-700 by comparing experimental and calculated hyperne and quadrupole tensors. In order to nd the explicit structure of complexes 1 and 1@Al 2 O 3-700 , the X-ray crystal structures of 1 (ref. 25) (see Fig. S6 ‡) and the derived model for the species in 1@Al 2 O 3-700 were optimized with unrestricted Kohn-Sham density functional theory (DFT), using the functional PBE0 (ref. 40) in ORCA 3. 41,42 For these geometry optimizations, a polarized triple-z def2-TZVPP basis set 43 was used for all atoms, together with Becke's three-center dispersion correction. 44 The COSMO continuum solvation model 45 was applied for complex 1. Furthermore, the hyperne and quadrupole tensor parameters were calculated with the def2-TZVPP basis set for Ti and Al atoms and the EPR-II basis set 46 for all other atoms. Based on the calculated parameters, the simulations of HYSCORE spectra were carried out in Easy-Spin. 47 Hyperne and nuclear quadrupole couplings were previously found to be highly sensitive to small structural changes. 22 Here we nd that rotation of one of the CH 2 t Bu ligands of 1 has a tremendous effect on the calculated isotropic part of the 1  Aer such an evaluation, the explicit structures and the Kohn-Sham molecular orbital sets for 1 and 1@Al 2 O 3-700 were obtained simultaneously, both veried by a comparison of the experimental and simulated HYSCORE spectra. The crystal structure of 1, however, does not yield the correct set of 1 H hyperne coupling parameters, since it fails to fully predict the experimental HYSCORE spectrum (Fig. S6 ‡). This indicates that complex 1 possesses a different conformation in frozen toluene solution than in the solid state. However, aer the geometry optimization using the parameters indicated above, the obtained conformation ( Fig. 4a) generates hyperne and quadrupole tensors for 14 N and 1 H nuclei (Table 1), which simulate the entire X-band HYSCORE spectrum rather nicely (Fig. 4b). Both the isotropic and dipolar parts of 1 H hyperne tensors t well to the experimental spectrum in Fig. 4b, thus indicating the correct positions of the a-H atoms of CH 2 t Bu ligands in the optimized structure ( Fig. 4a). Furthermore, the calculated 14 N hyperne and quadrupole tensor parameters are found to simulate both the X-band (Fig. 4b) and Q-band (see Fig. S8 ‡) HYSCORE spectra. We therefore propose that in a frozen toluene solution, molecular compound 1 is present in the form of the conformer shown in Fig. 4a   t Bu)] model was performed as described above for complex 1, followed by a DFT-based calculation of EPR parameters. The complex nature of the Al 2 O 3 surface results in a number of different types of OH groups that can participate in graing. 31 Here, we represent the (OH)Al s surface atom by the simplest possible tetracoordinated neutral Al model, namely (HO)Al s ¼ (HO)Al(OH) 2 (H 2 O). Within this approximation, the symmetry of the {AlO 4 } cluster may be decreased compared to the structure of an alumina surface center due to different Al-OH 2 and Al-OH bond lengths; this may result in overestimation of quadrupole coupling for the 27 Al nucleus. Furthermore, it is worth noting that the spin density on Al nuclei tends to be underestimated even with basis sets that contain diffuse functions, which are expected to better describe spin density near the nucleus. 50 Taking this into account, it appeared to be better to rely on experimentally determined 27 Al hyperne and quadrupole couplings rather than to evaluate these from DFT calculations. The 1 H and 14 N couplings (Table 1) computed for the thusobtained model (Fig. 5a) simulate the experimental X-band HYSCORE spectrum quite well (Fig. 5b). Similar to molecular complex 1, this indicates the correct positions of a-H atoms of a CH 2 t Bu ligand in the optimized structure. Although 27 Al couplings are not detected in the X-band HYSCORE spectrum, Q-band HYSCORE (Fig. 5c) provides the necessary information to determine the 27 Al hyperne couplings by least squares tting (Fig. S10 ‡). Together with the calculated 14 N hyperne couplings, they t reasonably to the Q-band HYSCORE spectrum (Fig. 5c, green). This allows us to consider the obtained model ( Fig. 5a) 27 Al, obtained from a least squares fit of the experimental spectrum. The ridge on the anti-diagonal in the left quadrant is due to an echo crossing that was not fully suppressed by phase cycling. 13 C labelling as a tool to probe Ti-alkyl chains and the polymerization mechanism. In order to gain further insight into the polymerization mechanism and the active state of the catalyst, we performed EPR studies on the molecular complex [Ti(nacnac)(CH 2 t Bu) 2 ] aer reaction with C 2 H 4 , in combination with 13 C isotope labelling. Among two possible labelling schemes (Scheme S1 ‡), we decided to use the one that involves the reaction of non-labelled ethylene with selectively 13 Clabelled complex [Ti(nacnac)( 13 CH 2 t Bu) 2 ] (1*) and that should yield (Ti III -(CH 2 CH 2 ) nÀ1 ( 13 CH 2 t Bu)) with a labelled 13 CH 2 t Bu terminating group. This reaction should lead to reduction of the initial 13 C signal intensity in the EPR spectra of 1*, which may be observed with pulse hyperne EPR methods and interpreted by comparison of the spectra before and aer the reaction.
Using the alternative labelling scheme, which involves the reaction of 13 C labelled ethylene with the unlabelled complex [Ti(nacnac)(CH 2 t Bu) 2 ] (1), proved to be difficult and did not allow the detection of 13 C hyperne couplings (see the ESI ‡ for details), possibly due to a broad distribution of conformations associated with the exibility of PE ligands (( 13 CH 2 ) n (CH 2 t Bu)). Such distribution would result in a broad set of 13 C hyperne couplings and thereby broad spectral lines unobservable in our hyperne EPR experiments. 13 C-labeled complex 1* was characterized by pulse EPR spectroscopy. It exhibits the same echo-detected EPR spectrum and, consequently, the same g tensor parameters as nonlabelled complex 1 (Fig. 6a). Detection of the 13 C couplings of the coordinating ( 13 CH 2 t Bu) ligands proved difficult. According to DFT calculations for the optimized structure of 1 (Fig. 4a), the 13 C hyperne tensors for both a-C atoms of ( 13 CH 2 t Bu) ligands of complex 1* show large couplings that are mostly isotropic (a iso ¼ À20. 51  . This leads to a low probability of the forbidden electron-13 C-nuclear spin transitions, making direct observation of the 13 C signals with ESEEM-based techniques (e.g. HYSCORE) difficult (in the Q band) or impossible (in the X band). Furthermore, strong 14 N ESEEM modulations in both the X-and Q-band may suppress the 13 C modulations due to a crosssuppression effect. 51 For this reason, we used an alternative EPR methodology based on the recently developed hyperne technique CHEESY-detected NMR (CHEESY ¼ chirp echo EPR spectroscopy). 52 This method is based on long selective hole burning pulses that drive forbidden transitions, similar to ELDOR-detected NMR, but the detection is based on broadband chirp echoes and subsequent Fourier transform. This leads to a multiplex advantage and, consequently, to higher sensitivity (see ESI Part 2.4 ‡ for more details). Indeed, the 13 C signals were observed already in the 1D CHEESY-detected NMR spectrum of 1*, revealed in comparison with the same spectrum of nonlabelled complex 1 where the 13 C signals were not observed (Fig. 6b).
These 13 C signals, observed at the orientation corresponding to g ¼ 1.983 (Fig. 6a, marked with an arrow), are better resolved in two-dimensional HYSCORE-type CHEESY-detected NMR spectra ( Fig. 6c and d), which are obtained by applying a selective p pulse with variable frequency before the HTA pulse (see ESI Part 2.4 ‡ for the details of CHEESY-detected NMR experiments). The comparison of the spectra for 1 (Fig. 6c) and 1* (Fig. 6d) reveals the peaks at (21,5) MHz, corresponding to the signals of 13 C in [Ti(nacnac)( 13 CH 2 t Bu) 2 ]. Based on the obtained spectra ( Fig. 6b-d), the 13 C hyperne coupling was estimated to be a iso ( 13 C) ¼ 16 MHz. This allowed us to conrm the assignment by Q-band HYSCORE and Q-band Davies ENDOR 53 (see Fig. S11 ‡), where the weak spectral signals corresponding to 13 C hyperne couplings were identied by comparison to the 1Dand 2D-CHEESY detected NMR spectra.
Next, a benzene solution of 1* was brought into contact with C 2 H 4 (1000 equivalents) at 80 C for 2 hours. Aer the reaction, the excess C 2 H 4 was removed and EPR measurements of (1* + C 2 H 4 ) were performed. The similarities in the echo-detected Qband EPR spectra (Fig. 7a) of complex 1* before and aer polymerization indicate similar g tensor parameters consistent with conservation of the symmetry of Ti(III). Using the same methodology as before, 1D CHEESY-detected NMR spectra before and aer the polymerization were measured (Fig. 7b) at the same frequency and eld positions, with an identical microwave resonator prole (Fig. S12 ‡). Although the absolute echo intensities for both samples may still be slightly different, the CHEESY-detected NMR signals, being essentially the ratio of the spectra with and without a high turning-angle pulse, can be considered a quantitative tool to probe the amount of EPR active nuclei in the Ti(III) coordination sphere before and aer the reaction with C 2 H 4 .
An obvious decrease of the 13 C signal intensity is observed aer polymerization (Fig. 7b, top) for all previously observed 13 C Fig. 6 (a) Q-band echo-detected EPR spectra of 1 (blue) and 1* (red), intensities normalized; the position of further measurements is indicated with an arrow. (b) 1D CHEESY-detected NMR spectra of 1 (blue) and 1* (red); the 13 C signals are marked on the spectrum. (c) Q-band 2D CHEESY-detected NMR spectrum of 1; the 13 C nuclear Zeeman frequency is shown by a red antidiagonal line. (d) Q-band 2D CHEESYdetected NMR spectrum of 1*; the 13 C nuclear Zeeman frequency is shown by a red antidiagonal line along which cross peaks due to 13 C hyperfine coupling are observed (marked with arrows). The gray area has not been recorded during the experiment in order to optimize sensitivity and the spectra were recorded at slightly different fields and frequencies.
lines of 1*, corresponding to a iso ( 13 C) ¼ 16 MHz, as well as for the combination signal of ( 13 C + 14 N). At the same time, the spectral lines, determined by 14 N hyperne and quadrupole couplings (e.g. double-quantum 14 N signals around 14 MHz), are the same before and aer polymerization both regarding their frequencies and intensities. Since the frequencies of these 14 N signals are sensitive even to small changes in the structure and conformation of 1* (see Fig. S7 ‡) we can conclude that the structure of the Ti(III) coordination sphere experiences minimal change upon polymerization. These 14 N signals were reasonably simulated with the calculated values for the previously estimated conformation of 1 (Fig. 4a) in the spectra both before and aer polymerization of ethylene (Fig. 7b, bottom). Together with the observed decrease of the 13 C signal intensity, this indicates that ligand exchange of 13 CH 2 t Bu to (CH 2 CH 2 ) nÀ1 ( 13-CH 2 t Bu) occurs with preservation of the initial structure and conformation of 1*. Indeed, the experimentally observed decrease of 13 C signals is simulated well as a difference between 1D CHEESY-detected NMR simulations for labelled complex 1* and non-labelled complex 1 (Fig. 7b, bottom). To that end, the experimental a iso ( 13 C) ¼ À16 MHz, together with DFT computed a dip ( 13 C) parameters, was used for the simulation of the spectrum of 1* (see ESI Part 2.4 ‡ for the details of the simulation). The comparison of the simulated and experimentally observed decrease of the 13 C signal intensity implies that probably not all the complex 1* present is affected by the ligand exchange, but only a part of it. This indicates that only a part of molecules of 1* acts as active centres of ethylene polymerization under the aforementioned reaction conditions. This is consistent with the presence of an induction period at the beginning of polymerization, revealed by changes in ethylene consumption (Fig. 2a). It is also consistent with the calculated energy barrier for the rst olen insertion (vide infra). An exact quantication is difficult without precise knowledge of the full 13 C hyperne tensor, which also affects line intensities. Based on the discussed experimental results, we propose that olen polymerization takes place via C 2 H 4 insertion into the Ti(III)-C bond in the molecular system Ti(nacnac)(CH 2 R) 2 -1. Unfortunately, we were not able to study ethylene polymerization with 1@Al 2 O 3-700 due to T 2 electron spin relaxation times, which are ca. 6 times shorter for 1@Al 2 O 3-700 than for 1. This limits the observation window length and, consequently, the resolution of CHEESYdetected NMR 26 such that the separation of 13  Electronic structures and the polymerization mechanism for 1 and 1@Al 2 O 3-700 a-Agostic C-H interaction and p character of Ti-C bonds of 1 and 1@Al 2 O 3-700 . The estimated structure of molecular complex 1 (Fig. 4a)  Essentially, a-agostic C-H bonds are described 35,54 as the donation of electrons from the lled molecular orbital corresponding to the C-H bond to a metal d-orbital of appropriate symmetry that is empty for d 0 metals. This agostic interaction has been recently related to a metal-carbon bond acquiring a p (or alkylidene) character, 35,55 which favors the olen insertion process. The degree of this p character could be indirectly estimated from the deviation of the Ti-C-H angle from 109 towards ca. 90 . In order to estimate directly the p character of Ti(III)-C bonds of 1 and 1@Al 2 O 3-700 , a Natural Bond Orbital (NBO) analysis 56 was performed, using the program NBO 7.0. 57 The molecular orbital sets for NBO analyses were generated using ORCA 4 (ref. 42) with the same parameters for the DFT calculations as the ones used for the simulations of the HYS-CORE spectra (PBE0 functional together with the def2-TZVPP basis set for Ti and Al atoms and EPR-II basis set for all other atoms). Given the good agreement between the measured and calculated HYSCORE spectra, this computational method describes the electronic structures of 1 and 1@Al 2 O 3-700 with sufficient accuracy. Fig. 7 (a) Q-band echo-detected EPR spectra of 1* (red) and 1* + C 2 H 4 (black), intensities normalized; the position of further measurements is indicated by an arrow. (b) Top: experimental 1D CHEESYdetected NMR spectra of 1* (red) and 1* + C 2 H 4 (black); no normalization was applied. Bottom: simulated 1D CHEESY-detected NMR spectra of 1* (red) and 1 (black). 13 C signals, corresponding to A( 13 C) ¼ 16 MHz are marked with lines under the spectra; the combination signal of ( 13 C + 14 N) is marked with an asterisk. The difference between the spectra due to difference in 13 C signal intensities is colored in blue.
The NBO analysis revealed a natural orbital, related to a singly occupied molecular orbital (SOMO) of paramagnetic complex 1 (Fig. 8a, red and blue). Its shape correlates well with the calculated distribution of spin density in space (Fig. 8a,  green), thus conrming the close relation of this natural orbital to the SOMO. This orbital is nearly axially symmetric, which is consistent with the experimentally observed axial symmetry of the g tensor.
The spatial distribution of the SOMO-related natural orbital of 1 includes four lobes of the p-type on two nitrogen atoms and two carbon atoms of CH 2 t Bu ligands, all being in antiphase with the central d-type lobes (Fig. 8a and c). The part of this orbital, which includes the central d-type part and the two carbon p-type parts (Fig. 8c), can be understood as a product of interaction of the half-lled d xy -type Ti orbital and one of two degenerate lled p À orbitals of the (CH 2 t Bu) 2 fragment (Fig. 8b). This orbital features p* symmetry with respect to the Ti-C bond. Therefore, the presence of the p* orbital, as part of the SOMO and delocalized between two CH 2 t Bu ligands, reveals the existence of p bonding in Ti-(CH 2 t Bu) 2 . This p interaction, although being weakened by the unpaired electron in the antibonding p* orbital (Fig. 8b), stabilizes the structure with the a-agostic C-H bonds for 1. Compared to metal d 0 complexes 35 this a-H agostic interaction involves a half-lled metal d orbital instead of an empty one, as revealed by the NBO analysis for the rened structure of 1. Such an interaction brings a p character into both Ti-C bonds of 1. This p character is also evidenced by the deviation of the natural hybrid orbital (NHO) on carbon from the Ti-C axis (q NHO-C-Ti ¼ 15.0 and 14.4 for the two Ti-C bonds of 1)for a pure s-bond no deviation would be expected (0.0 ). 35 A natural orbital of similar type, including a p-type lobe on the carbon atom of the single CH 2 t Bu ligand, is also found for 1@Al 2 O 3-700 (Fig. 8d). This indicates the presence of a p* orbital and, consequently, a p interaction in the [Ti-(CH 2 t Bu)] system.
However, the p character acquired by the Ti-C bond of 1@Al 2 O 3-700 is less pronounced compared to that of complex 1 (q NHO-C-Ti ¼ 9.2 for 1@Al 2 O 3-700 ). The described p interaction, which involves half-lled metal d orbitals, might be present for all paramagnetic transition metal alkyl complexes, provided that the corresponding half-lled d orbitals have appropriate symmetry.
Olen polymerization pathways of 1 and 1@Al 2 O 3-700 . The presence of p character in the metal-carbon bonds has been found to play a crucial role in the reactivity of d 0 compounds, making them reactive towards olen insertion. 35 It was also used as an explanation for C-H activation pathways, including a-H abstraction in dialkyl compounds, that have been shown to be isolobal reactions. 58 Indeed, the a-H abstraction is a known synthetic pathway of Ti(IV) d 0 alkylidenes, prepared via oxidation of 1 by AgOTf. 25 However, the transition state (TS) energy for the a-H abstraction process for d 1 complex 1, calculated in ORCA 3 (ref. 41) with the same DFT parameters as the ones used for ground state optimizations, was found to be relatively high (DH ‡ 298 ¼ 31.5 kcal mol À1 ; DG ‡ 298 ¼ 31.5 kcal mol À1 , Fig. S13 ‡). This should make the process slow and indicate the relative stability of 1 even under the elevated temperatures used in ethylene polymerization. In contrast, the transition state for C 2 H 4 insertion into the Ti(III)-C bond of 1 (Fig. S13 ‡) appeared to have an overall energy barrier with respect to the initial reagents (1 + C 2 H 4 ) of DH ‡ 298 ¼ 22.5 kcal mol À1 and DG ‡ 298 ¼ 33.7 kcal mol À1 . The large difference of 9.0 kcal mol À1 in the TS enthalpies (DH ‡ 298 ) suggests that the reaction of C 2 H 4 insertion into the Ti(III)-C bond is more facile than a-H abstraction. Looking at the free energy, where entropy factors in solution are typically overestimated, 59 one would expect that both processes can be competitive. Overall, the calculated DH ‡ 298 and DG ‡ 298 values of the ethylene insertion for complex 1 are consistent with a slow polymerization reaction at 80 C as well as the need to use high pressure to conduct the reaction. In fact, similar calculated energetics are reported for the Ti(IV) homogeneous catalysts of ethylene polymerization (e.g. DH ‡ 298 and DG ‡ 298 ¼ 16 and 28 kcal mol À1 , respectively, for the [H 2 Si(C 5 -H 4 )( t BuN)]TiCH 3 + /H 3 CB(C 6 F 5 ) 3 À ion pair 60 ). This supports our experimental evidence of ethylene insertion into the Ti(III)-C bond as the mechanism of ethylene polymerization of molecular catalyst 1.
It is noteworthy that for the model Ti(IV) cationic analogue of complex 1, namely [Ti(nacnac)(CH 2 t Bu) 2 ] + (1 + ), the ethylene insertion reaction is predicted to be less favorable compared to the a-H abstraction process, as revealed by calculations on the optimized structure (Fig. 9a). The calculated energy barrier for the a-H abstraction in d 0 complex 1 + is DH ‡ 298 ¼ 29.7 kcal mol À1 and DG ‡ 298 ¼ 30.3 kcal mol À1 , being slightly less than the one calculated for d 1 complex 1. This is consistent with a stronger degree of p character in the Ti-C bonds (q NHO-C-Ti ¼ 23.5 and 17.5 for the two Ti-C bonds) of 1 + . For ethylene insertion involving 1 + , the TS barrier is DH ‡ 298 ¼ 28.3 kcal mol À1 and DG ‡ 298 ¼ 41.7 kcal mol À1 with respect to the initial reagents. Despite having a stronger degree of p character of the Ti-C bonds, the TS energy of ethylene insertion is strongly increased by DG ¼ + 8.0 kcal mol À1 for the d 0 complex compared to the analogous d 1 complex. This indicates that the unpaired electron in a singly occupied molecular orbital of complex 1 plays an important role in its reactivity towards ethylene insertion, signicantly lowering the TS energy for d 1 active species compared to similar d 0 species.
For the neutral supported alkyl species in 1@Al 2 O 3-700 , the overall energy barriers for C 2 H 4 insertion into the Ti-C bond are DH ‡ 298 ¼ 16.7 kcal mol À1 and DG ‡ 298 ¼ 29.7 kcal mol À1 and thus both are lower than those found for molecular complex 1. This is consistent with the high polymerization activity of 1@Al 2 O 3-700 .
Ethylene insertion into the Ti-C bonds for both 1 and and very close to those of the TS (the corresponding TS energies are DH ‡ 298 ¼ 1.1 kcal mol À1 and DG ‡ 298 ¼ À0.3 kcal mol À1 with respect to the 1/C 2 H 4 complex). The same is found for the 1@Al 2 O 3-700 /C 2 H 4 p complex, with its CH 2 t Bu ligand being trans to the Ti-N bond of the nacnac ligand. With DH 0 298 ¼ 16.4 kcal mol À1 and DG 0 298 ¼ 28.2 kcal mol À1 the energy barriers remaining to reach the TS are almost zero and calculated to be DH ‡ 298 ¼ 0.3 kcal mol À1 and DG ‡ 298 ¼ 1.5 kcal mol À1 . Therefore, the energy cost for ethylene polymerization is mostly due to the initial formation of the p complexes followed by an almost barrier-less insertion, which is again in agreement with the observed induction period. The relatively high barrier of formation of the pethylene complex 1/C 2 H 4 could be overcome by elevated ethylene pressure and increased temperature; this is consistent with the high temperatures and pressures (e.g. 80 C, 7 bar) required for efficient polymerization on 1 (see Fig. 2a).
For the calculated Ti(IV) model cationic complex [Ti(nacnac)(C 2 H 4 )(CH 2 t Bu) 2 ] + (1 + /C 2 H 4 , Fig. 9d), the TS energy of ethylene insertion with respect to this complex is close to that of its d 1 analogue 1/C 2 H 4 (DH ‡ 298 ¼ 0.9 kcal mol À1 and DG ‡ 298 ¼ 2.4 kcal mol À1 ). Again, as soon as a p-ethylene complex is formed, the ethylene insertion into the Ti-C bond is nearly barrier-less; this behavior occurs in both d 1 and d 0 complexes. At the same time, while the structure is similar to that of the 1/C 2 H 4 complex, the formation energy of 1 + /C 2 H 4 is much higher than that for 1/ C 2 H 4 (DH 0 298 ¼ 27.4 kcal mol À1 , DG 0 298 ¼ 39.3 kcal mol À1 ). We thus further examine the p-ethylene complexes 1/C 2 H 4 , 1 + /C 2 H 4 and 1@Al 2 O 3-700 /C 2 H 4 via NBO analyses.
Back donation from the unpaired electron orbital and the enhanced p character in Ti-C bonds of the p-ethylene complexes support the "augmented" Cossee-Arlman polymerization mechanism. For both 1/C 2 H 4 and 1@Al 2 O 3-700 /C 2 H 4 , the degree of the p character of the Ti-C bonds may be indirectly estimated through the corresponding Ti-C-H angles in the calculated structures of the complexes, which acts as a marker of the a-agostic C-H interaction. For the complex 1/  low calculated or reported 35 TS energies. This indicates that ethylene insertion in neutral Ti(III) alkyl species or cationic Ti(IV) and Zr(IV) alkyls depends on the extent of the p character in Ti-C bonds. Therefore, a strong p character is a general reason for the facile ethylene insertion in d 1 and d 0 metal alkyl complexes aer the coordination of C 2 H 4 .
It is noteworthy that for Ti(III) d 1 complexes 1 and 1@Al 2 O 3-700 "back donation" of the unpaired electron from a SOMO to the p* orbital of coordinated C 2 H 4 is also involved in the insertion process. These two orbitals have a constructive overlap, as revealed by the NBO analysis for complex 1/C 2 H 4 via the overlap of negative parts of the corresponding natural orbitals (Fig. 10a). This results in some weakening of the C]C double bond of the C 2 H 4 ligand (C-C distance is 1.344 A compared to the calculated value of 1.333 A for free ethylene), together with a population of the p*(C 2 H 4 ) orbital (see ESI Part 2.7 ‡). The NBO energetic analysis reveals the stabilization effect, caused by the presence of a p*(C 2 H 4 ) orbital in 1/C 2 H 4 , of 30.0 kcal mol À1 . At the same time, the stabilization effect, caused by the presence of the p*(C 2 H 4 ) orbital in the d 0 1 + / C 2 H 4 complex, is 23.1 kcal mol À1 . Therefore, the energy of the back donation of unpaired electron density in 1/C 2 H 4 could be estimated through the difference of these two as a stabilization by 6.9 kcal mol À1 , consistent with the difference of formation enthalpies DH 0 298 of 1/C 2 H 4 and 1 + /C 2 H 4 of 6.0 kcal mol À1 .
This indicates that the back donation of the unpaired electron from the SOMO to the p* orbital of C 2 H 4 is the main reason for relative stabilization of the p-ethylene complex 1/C 2 H 4 compared to its d 0 analogue 1 + /C 2 H 4 . As the TS energies of ethylene insertion aer the formation of p-ethylene complexes 1/C 2 H 4 and 1 + /C 2 H 4 differ only by 2.7 kcal mol À1 (DG ‡ 298 ¼ À0.3 kcal mol À1 for 1/C 2 H 4 , compared to DG ‡ 298 ¼ 2.4 kcal mol À1 for 1 + /C 2 H 4 ), the principal reason for the lowering of the overall TS energy for d 1 catalyst 1 by DG ¼ 8.0 kcal mol À1 is the back donation of the unpaired electron to the p* orbital of C 2 H 4 .
An even higher degree of back donation to p*(C 2 H 4 ) is observed for the 1@Al 2 O 3-700 /C 2 H 4 complex, since a strong spin density transfer from the initial SOMO to p*(C 2 H 4 ) is revealed by the calculated spin density distribution (Fig. 10b). Similar to the molecular 1/C 2 H 4 complex, this leads to a weakening of the C]C double bond of the C 2 H 4 molecule (C-C distance is 1.364 A) and to the appearance of a bonding interaction between Ti in 1@Al 2 O 3-700 and C 2 H 4 . In fact, NBO analysis, being one of the possible ways of representation of the electronic structure of 1@Al 2 O 3-700 /C 2 H 4 , shows a breaking of the p system of ethylene, followed by formation of a bonding set of natural orbitals, corresponding to a Ti-C(C 2 H 4 ) bond ( Fig. S14 and Table S2 ‡), and a partially occupied natural lone pair on the other carbon atom of C 2 H 4 , derived from p*(C 2 H 4 ) (Fig. 10b). This indicates more favorable coordination of C 2 H 4 to 1@Al 2 O 3-700 where the stronger back donation in 1@Al 2 O 3-700 /C 2 H 4 is caused by better orbital overlap between the initial SOMO and p*(C 2 H 4 ). The better overlap between the SOMO and p*(C 2 H 4 ) in the supported catalyst is mostly caused by the difference in geometry of the p-ethylene complexes induced by replacing one of the strong s-donor alkyl ligands (CH 2 t Bu) by a weaker O anionic surface ligand OAl s . This is consistent with the lower formation energy of 1@Al 2 O 3-700 /C 2 H 4 of DH 0 298 ¼ 16.4 kcal mol À1 compared to DH 0 298 ¼ 21.4 kcal mol À1 for complex 1 and with the ability of 1@Al 2 O 3-700 to catalyze ethylene polymerization under milder conditions.
In fact, the unpaired electron in the 1@Al 2 O 3-700 /C 2 H 4 complex appears to be strongly delocalized between the Ti d orbital, p*(C 2 H 4 ) and s*(Ti-C CH2 t Bu ), as revealed by the spin density distribution (Fig. 10b) and the occupancies of the related natural orbitals (see Table S2 ‡). The same delocalization is also found in the structure of the TS of ethylene insertion for both 1 and 1@Al 2 O 3-700 , as revealed by the calculated spin density distributions (Fig. 10c and d). This delocalization appears to be stronger for the TS for 1@Al 2 O 3-700 (Fig. 10d), while being weaker for the TS for 1 (Fig. 10c). It appears that the unpaired electron, while weakening the C]C double bond of C 2 H 4 and facilitating the formation of a Ti-C(C 2 H 4 ) bond due to its presence at p*(C 2 H 4 ), also favors the cleavage of the Ti-C bond of the CH 2 t Bu ligand by occupation of the s*(Ti-C) orbital before ethylene insertion. These factors together facilitate the C 2 H 4 insertion into the Ti-C bond, in addition to the previously mentioned factor of its p character, which is consistent with a lowered energy of the TS of ethylene insertion in 1/C 2 H 4 , compared to 1 + /C 2 H 4 , by DG ¼ 2.7 kcal mol À1 . We, therefore, propose that the mechanism of ethylene polymerization for 1 In general, the mechanism of ethylene insertion in d 1 Ti(III) alkyl complexes 1 and 1@Al 2 O 3-700 is determined by two key factors: the p character in Ti-C bonds of (CH 2 t Bu) ligands and the back donation of the unpaired electron. While the presence of the p character facilitates the insertion of ethylene into Ti-C bonds aer its coordination, making the insertion in p-ethylene complexes nearly barrier-less at the TS, the back donation signicantly lowers the formation energies of the p-ethylene complexes, which facilitates the overall reaction of C 2 H 4 insertion. The delocalization of the unpaired electron in the TS structure, being noticeable for the systems with a high degree of back donation (i.e. 1@Al 2 O 3-700 /C 2 H 4 ), also has an effect on this reaction, favoring the cleavage of Ti-C and C]C bonds and slightly lowering the TS barrier. We denote this process (Scheme 2) as an "augmented" Cossee-Arlman mechanism, being essentially a [2s + 2p + d 1 ] cycloaddition involving a partially alkylidenic s(Ti-C) bond and a p(C 2 H 4 ) bond together with a delocalized d 1 electron.
The described delocalization of the unpaired electron is likely an important feature in d 1 systems able to polymerize olens, with a degree of the back donation (electron density transfer) that depends on the overlap of the SOMO and p*(C 2 H 4 ), which in turn depends on the geometry of the system. For instance, a higher polymerization activity towards styrene polymerization was observed experimentally for Cp*Ti(OCH 3 ) 2 / MAO and Cp 2 TiCl/MAO catalytic systems compared to Cp*Ti(OCH 3 ) 3 /MAO and Cp 2 TiCl 2 /MAO, respectively. 21 It is likely that the Ti(III) species, active towards styrene polymerization, show better performance compared to similar Ti(IV) systems due to a strong back donation (electron transfer), favored by the aromatic system of styrene. Therefore, under the same polymerization conditions, a d 1 catalyst may be more active than a d 0 catalyst of a similar structure. This nding also further suggests that Ti(III) alkyl species have competent electronic structures to act as efficient polymerization catalysts and may indeed be active species in the classical Ziegler-Natta heterogeneous catalysts.

Conclusions
In this work, we report the polymerization activity of molecular and the corresponding alumina-supported well-dened Ti(III) neutral alkyl species prepared via surface organometallic chemistry. Both of them were characterized and studied in detail by pulse EPR spectroscopy, combined with DFT calculations. This approach enabled us to identify the prevalent conformation of the molecular complex [Ti(nacnac)(CH 2 t Bu) 2 ] in a frozen toluene solution and to reveal the structure of the alumina-supported species that correspond predominantly to a neutral Ti(III) alkyl compound, i.e. [(Al s O)Ti(nacnac)(CH 2 t Bu)]. To the best of our knowledge, these are the rst examples of well-dened Ti(III) alkyl species able to efficiently polymerize ethylene, producing ultra-high molecular weight polyethylene. The ethylene insertion into the Ti(III)-C bond of [Ti(nacnac)(CH 2 t Bu) 2 ] was further evidenced by EPR hyperne spectroscopy (CHEESY-detected NMR), using isotope-labeled Ti(nacnac)( 13 CH 2 t Bu) 2 in contact with C 2 H 4 . These Ti(III)-based polymerization pre-catalysts display aagostic C-H bonds in their (CH 2 t Bu) ligands, supporting the presence of p character in the corresponding metal-carbon bonds. 35 Such a p character was further supported by DFT calculations via NBO analysis. It is noteworthy that the presence of the half-lled d 1 Ti orbital does not prevent a-agostic C-H bonding. Aer coordination of C 2 H 4 , the degree of p character in the Ti-C bonds of (CH 2 t Bu) ligands is signicantly increased, which allows a nearly barrier-less insertion of C 2 H 4 into the Ti-C bonds. Hence, the slow step is olen coordination, consistent with the need for high pressure to carry out this reaction and with the observation of an induction period. The back donating interaction (electron transfer) between the SOMO and the p* orbital of C 2 H 4 results in a signicant lowering of the formation energies of p-ethylene complexes, which facilitates an overall reaction of ethylene insertion in these Ti(III) systems. Due to the back donation, the unpaired electron could be delocalized between the Ti d orbital and p*(C 2 H 4 ) and s*(Ti-C) orbitals in both p-ethylene complexes and transition states, which also lowers the energy barriers for ethylene insertion. All these factors, which combine to give an "augmented" Cossee-Arlman mechanism, facilitate the overall reaction of C 2 H 4 insertion into the Ti-C bond, making the ethylene polymerization in d 1 metal complexes potentially more efficient than that in d 0 complexes of a similar structure under the same conditions. This study shows that neutral d 1 Ti alkyl complexes are competent in ethylene polymerization, being favored by a combination of the p character in the Ti-C bonds and the back donation of the unpaired electron. These ndings lend further support to the notion that d 1 Ti-alkyls are possible active sites in the heterogeneous Ziegler-Natta polymerization catalysts. Bu) 2 complex as Scheme 2 "Augmented" Cossee-Arlman mechanism of ethylene polymerization for 1 and 1@Al 2 O 3-700 . A partial electron transfer process from the SOMO to p*(C 2 H 4 ) ("back donation") is shown in the molecular orbital picture. a stable Ti(III) alkyl compound to be tested for ethylene polymerization activity. G. J. and C. C. coordinated the project and provided guidance. All the authors revised the manuscript.

Conflicts of interest
There are no conicts to declare.