Catalysts for CO2/epoxide ring-opening copolymerization

This article summarizes and reviews recent progress in the development of catalysts for the ring-opening copolymerization of carbon dioxide and epoxides. The copolymerization is an interesting method to add value to carbon dioxide, including from waste sources, and to reduce pollution associated with commodity polymer manufacture. The selection of the catalyst is of critical importance to control the composition, properties and applications of the resultant polymers. This review highlights and exemplifies some key recent findings and hypotheses, in particular using examples drawn from our own research.


Introduction
The ring-opening copolymerization (ROCOP) of carbon dioxide and epoxides is an interesting method to synthesize a range of aliphatic polycarbonates (figure 1) [1][2][3][4][5]. The reaction was discovered more than 40 years ago and has since continued to attract attention as a means to reduce pollution associated with polymer manufacture and to 'add value' to carbon dioxide [1]. It should be made clear that the use of CO 2 in any chemical manufacturing process is unable to make a large impact on overall CO 2 levels in the atmosphere. However, it remains important to pursue CO 2 utilization as a means to reduce emissions, particularly those associated with existing, large-scale industrial processes and as an economic driver to support carbon capture [6,7]. The ROCOP process is strongly dependent on the selection of the catalyst, with various homogeneous and heterogeneous catalysts having been reported [5,[8][9][10][11][12][13][14][15]. The focus for this review article will be to highlight and exemplify some of the key findings in this area of catalysis, in particular using examples drawn from our own research. The intention is not to provide a comprehensive review of all known catalysts; indeed, such reviews are already available [5,[8][9][10][11][12][13][14][15].
The ROCOP reaction is a rare example of a truly catalytic process with the potential to deliver large-scale quantities of product, which genuinely consumes carbon dioxide. The most commonly studied epoxides are cyclohexene oxide (CHO) and propylene oxide (PO). Depending on the epoxide and the selectivity of the catalyst, up to 31% (polycyclohexene carbonate) or 43% (polypropylene carbonate) of the polymer mass derives from CO 2 . The primary application for the polymer products is as low-molecular-weight (M n ), (poly)hydroxyl-terminated 'polyols', which are widely used in the manufacture of polyurethanes [16]. Polyurethanes themselves are applied as flexible/rigid foams, adhesives, coatings and elastomers, as well as in many other areas. It has been shown that the properties of CO 2 -derived polyols are suitable to replace polyether polyols in some applications [16,17]. Given that polyether polyols are prepared by epoxide homopolymerization, the ROCOP of CO 2 and epoxides can also be viewed as a means to 'replace' a substantial portion of petrochemically derived resource (epoxide) with a renewable one (CO 2 ). A recent detailed life-cycle analysis study compared these two types of polyols, showing significant reductions (approx. 20%) in both fossil resource depletion and greenhouse gas emissions for the CO 2 -derived polymers [18].
Furthermore, the replacement of fossil-derived epoxides is economically attractive and is stimulating a number of commercialization studies [16,19] [20]. Studies have shown that some homogeneous magnesium catalysts can be used in polymerizations where the carbon dioxide is captured at a CCS demonstrator plant attached to a UK power station. The catalysts showed near-equivalent performances using such captured gases compared to using 'pure' carbon dioxide. Furthermore, the catalyst showed a high tolerance to various impurities present in captured CO 2 , including water, N 2 , CO, thiols and amines [20].

Polymerization pathways and mechanisms
A range of different catalysts are known but all catalysts contain metals, with Zn(II), Co(III) and Cr(III) being particularly common [5,[8][9][10][11][12][13][14]. Prior to an examination of these catalysts, it is worth considering the series of reactions that are proposed to occur at the metal active site during polymerization, the major ones of which are illustrated in figure 2 [3,11]. The polymerization is initiated by coordination of an epoxide molecule and the subsequent ring opening by the nucleophilic attack of a carbonate group or ligand (X), so as to form a metal alkoxide intermediate.  carbonate intermediates. The polymerization is terminated by exposure to conditions/reagents that lead to hydrolysis of the growing polymer chain and formation of a polymer chain end-capped with a hydroxyl group.
There are also side reactions within this process, the proportions of which depend on the conditions, substrate and catalyst selected. The formation of ether linkages in the polymer chain can occur due to the occurrence of a metal alkoxide attack on an epoxide molecule instead of CO 2 insertion. Such linkages change the polymer properties, which may be beneficial depending on the application, but nevertheless reduce the CO 2 sequestered in the polymer backbone [16,17].
Five-membered ring cyclic carbonate by-products can also form; indeed, this is the thermodynamic product of the reaction between CO 2 and epoxides and thus favoured under forcing conditions. The cyclic carbonates can form by depolymerization or 'back-biting' reactions; the extent of these depend on the catalyst and the ceiling temperature for the given polymer [21,22]. The cyclic carbonates can also form 'off-metal' and are particularly common, and sometimes problematic, contaminants when co-catalysts or ionic additives are applied [5,13]. For some types of catalyst, most commonly with metal salen catalysts, it is proposed that the addition of ionic co-catalysts leads to the polymer chains being in equilibrium between metal coordination and 'free' anionic polymer chains [5]. Such 'off-metal' chains have been shown to undergo cyclization to form cyclic carbonate by-products [23].
Chain transfer reactions also need to be considered. These reactions occur when the polymerizations are conducted under so-called 'immortal' conditions in the presence of protic compounds, such as alcohols, amines and water, among others [24][25][26][27]. When such protic reagents are present, the metal-alkoxide-terminated polymer chain is proposed to be in rapid exchange with the protic reagent, generating metal alkoxides and 'free' hydroxyl-terminated polymer chains. The chain transfer processes are proposed to occur more rapidly than propagation, leading to highly controlled polymerizations where the M n of the polymer is determined by both the catalyst and chain transfer agent (protic reagent) concentrations. It is worth noting that in ROCOP the polycarbonate M n is commonly experimentally observed to be rather lower than those expected for living polymerizations (where the M n would depend only on the catalyst concentration) [25]. This is due to the presence or formation of chain transfer agents, such as diols,   in the epoxide monomers used [28][29][30][31][32]. In fact, the exploitation of catalysts able to operate under immortal polymerization conditions, i.e. where such chain transfer agents are added in larger excess compared to the catalyst, is essential to selectively prepare the polyols that are required in polyurethane manufacture [18,20,24]. The catalysts for ROCOP have some general features: the metals are Lewis acids; the metals' redox reactivity should be limited; the metal alkoxide and carbonate intermediates are labile; and initiating ligands (X) include alkoxides, carboxylates, halides and other anionic groups [5,[8][9][10][11][12][13][14]. It is common in this area of catalysis that dinuclear or bimetallic catalysts show good performances and bimetallic pathways are proposed to accelerate epoxide ring opening [10]. Finally, the polymerization catalysts should ideally be colourless, odourless, inexpensive and have low toxicity, as they may contaminate the polymer product. In this context, there have been a number of reports of strategies to remove and recycle homogeneous catalysts [33][34][35].
The nature of the co-catalyst is also important; as mentioned PPNX salts are widely applied. Furthermore, the amount of co-catalyst is generally optimum at 1 equivalent (versus metal complex), as greater quantities reduce activity. This is proposed to be due to competitive binding (versus epoxide) at the metal centre [99,100]. Using more than 1 equivalent of co-catalyst also increases the rate of back-biting side reactions to form cyclic carbonate products [5,57,100,101]. The co-catalyst serves a number of roles in the catalytic cycle. It is proposed that the cocatalyst binds to the metal complex in order to complete an octahedral coordination geometry and, by trans-coordination, enhances the labilization of the initiating or propagating group [5]. Additionally, the co-catalyst may act as an external nucleophile which initiates polymerization. The precise mechanisms by which such catalysts operate are rather complex and not yet fully defined, but it is generally proposed that the rate-limiting step involves epoxide ring opening rather than carbon dioxide insertion [5].
Recent kinetic studies have revealed that the polymerization rates, using metal salen catalysts, are typically dependent on catalyst concentration to a fractional order, usually between 1 and 2 [89]. This is indicative of pathways involving two metal complexes and/or dimerization in the rate-limiting step. Figure 5 illustrates the two different proposed pathways for epoxide ring opening. The fractional orders in catalyst concentration could be rationalized if both pathways are feasible and occurring concurrently. Significant work has also been carried out to investigate the coordination of epoxides to such catalysts [102][103][104]. The gas-phase binding of epoxides in Al(III) and Cr(III) salen and porphyrin complexes was studied by Chisholm and co-workers and found to follow metal Lewis acidity trends [103]. Darensbourg also showed that the ring opening of epoxides by a nucleophile was dependent on the metal-epoxide bond length and the binding enthalpies, not solely on the rate at which epoxide binding occurred [104]. It is worth noting that the coordination of co-catalysts changes the Lewis acidity of the metal centre and so would affect epoxide binding [105].
Recently, Nozaki and co-workers [89] demonstrated a theoretical model to predict the catalytic activity and selectivity in PO ROCOP using planar bicomponent catalysts. By comparing the difference between the dissociation energies for metal epoxide, carbonate and alkoxide intermediates, the preferred pathway at each stage of the catalytic cycle could be quickly estimated [89].
One challenge of exogenous co-catalysts is that, if the stoichiometry (versus catalyst) is not finely balanced, significant side reactions to produce cyclic carbonate can result. Furthermore, because the systems are bicomponent, low catalyst loadings are not generally feasible. Finally, the fractional orders in catalyst imply that dimerization may be necessary, thereby providing an entropic barrier to catalysis at low metal loading. Two strategies have been used to overcome these limitations: (i) the development of catalysts whereby the co-catalyst is attached to the ancillary ligand and (ii) the development of dinuclear salen catalysts.
Increased activities, tolerance and selectivities were observed by covalently bonding the cocatalyst moiety to the catalyst ligand, removing the need to add an exogenous co-catalyst source. Nozaki first reported this phenomenon for a Co(III) salen complex, substituted with piperidinium 'arms' (figure 4c) [67]. This catalyst showed very high selectivity for carbonate formation (more than 99%), a high turnover frequency (TOF) (250 h −1 ) and operated at room temperature, under 14 bar pressure of CO 2 [67]. Lee subsequently also developed various bifunctional catalysts (figure 4d) substituted with ionic groups, which have shown some of the highest activities (TOF = 26 000 h −1 ) ever reported and which are active under low catalyst loadings (1 : 25 000 catalyst : PO) [29,30,33]. Lu has also shown that a cobalt salen complex with a tethered quaternary ammonium salt is highly efficient (TOF up to 5160 h −1 ) for the terpolymerization of CHO with a range of aliphatic epoxides (such as PO) and CO 2 [106].

(b) Dinuclear or bimetallic catalysts
An important development for the field was the report from Coates, in 1998, of highly active zinc BDI catalysts [107]. These catalysts have shown good activity and selectivity for CO 2 /CHO ROCOP, at 50 • C and 7 bar pressure of CO 2 . A wide range of [(BDI)ZnX] complexes have since been synthesized where ligand substituents exert a strong effect on activity (figure 6a) [107,[111][112][113][114]. In a landmark paper investigating the mechanism, it was shown that the most active catalysts existed as 'loosely associated' dimers under the polymerization conditions [56]. Rate studies revealed that the order in zinc varied from 1.0 to 1.8, implicating dimeric active sites [56].
This study has inspired many others to develop dinuclear catalysts so as to increase activity and selectivity. Lee pioneered this with a series of bis(anilido-aldimine) Zn(II) complexes which showed high activities (TOF = 2860 h −1 ) in copolymerization, at 1 : 50 000 [Zn] : [epoxide] loadings (figure 6b) [108]. The activity is sensitive to the N-aryl ortho-substituents and the fluorination of the aromatic rings significantly increased the TOF values [108]. Many researchers have investigated methods to 'tether' Zn(II) BDI catalysts [109,[115][116][117]. The geometry, site and flexibility of the tether exert a significant influence over the activity, the best system for CO 2 /CHO being a dinuclear Zn(II) catalyst (figure 6c), which showed TOF values of 155 000 h −1 , at 100 • C and 30 bar [109,116,117].
monometallic counterparts). Additionally, these catalysts were able to operate at low catalyst loadings (up to 20 000, epoxide/catalyst), supporting the notion that two metal sites may be involved in the pathway [91,97].
Furthermore, the detailed analysis of the temperature dependence of the rate coefficients for both polymerization and cyclic carbonate formation enables a comparison of the relative barriers for polymerization (polycyclohexene carbonate, PCHC) versus cyclic carbonate formation (cyclohexene carbonate, CHC), the results of which are qualitatively illustrated in figure 7 [124]. It is apparent that the di-zinc catalyst shows a high selectivity for polymer formation due to the barrier to polymerization being approximately half the value for cyclic carbonate formation (E a (PCHC) of 96.8 kJ mol −1 and E a (CHC) of 137.5 kJ mol −1 ).
The rate law shows a zero-order dependence of carbon dioxide pressure (1-40 bar), which suggests that, under these conditions, the ring opening of the epoxide is rate-determining. The first-order catalyst dependence also supports the proposal of dinuclear propagation pathways. Although the catalyst is dinuclear, it also has two acetate co-ligands and therefore could potentially initiate two polymer chains per catalyst [123]. where Cl and MeIm are coordinated) and the concave face refers to the inside of the 'bowl' (i.e. where the bridging Cl ligand is coordinated). Adapted from [122].
first-order dependences on both catalyst and epoxide concentrations, which implies that only one acetate co-ligand initiates polymerization, although more complex pathways cannot be ruled out on the basis of the rate law. This led to further investigations into the possible roles for the co-ligands. A series of di-cobalt halide catalysts were prepared, with various neutral co-ligands being coordinated to them, including pyridine, methyl imidazole and DMAP (figures 8 and 9) [122]. The complexes all showed closely related solid-state structures, with the ancillary ligand adopting a 'bowl' conformation, whereby the NH substituents occupy the same 'face' of the molecule.
In the concave face of the molecule, a halide ligand bridges the two cobalt centres. On the opposite convex face, a halide and Lewis base donor molecule coordinate to the cobalt centres ( figure 8). It was discovered that the complexes showed rates strongly dependent on the nature of the donor ligand, with DMAP complexes having low activities and those coordinated by the weaker pyridine donor being equivalently active to catalysts without any co-ligand coordination. The finding that coordination of a single DMAP molecule could significantly slow the catalysis supported the notion that polymerization dominates from the convex face of the molecule and that two metals are involved. A detailed spectroscopic and density functional theory (DFT) study also shed light on the polymerization pathways and key intermediates [123]. Using in situ attenuated total reflectanceinfrared (ATR-IR) spectroscopy to characterize the reaction between di-zinc acetate complex and CHO showed that there were different environments for the acetate co-ligand. One of the acetate groups shows resonances consistent with attack at the CHO group, while the other has a resonance consistent with it maintaining a 'bridging' coordination mode between the two zinc centres.
The DFT study, carried out in solution, using appropriate functionals to model the polar environment, and using the complete catalyst structure, also substantiated the proposal that, although both metals are involved in propagation, only one of the two co-ligands initiates and propagates polymerization. The remaining co-ligand remains coordinated to the metal centres and mediates the polymerization process. Indeed, it was revealed by DFT that the polymer chain 'switches' coordination site between the two metal centres with each monomer insertion (epoxide or carbon dioxide). The chain shuttling is counter-balanced by an equal but opposite change in coordination site for the acetate co-ligand. It was shown that, by changing the co-ligand, it was possible to alter the rate of polymerization and DFT could be used to predict activity [123].
The chain shuttling mechanism implies that there are distinct roles for the two metal centres, as sketched in figure 10. Thus, one of the metal centres (illustrated in blue) coordinates the epoxides, while the other centre (illustrated in red) inserts carbon dioxide. The requirements for these processes are distinct, with epoxide coordination being accelerated by Lewis Figure 11. The structures of di-zinc (1), di-magnesium (2) and the heterodinuclear mixed catalyst system (3) which were compared for the ROCOP of CO 2 /CHO [120]. (Online version in colour.) carbonate groups. Such a mechanism would be expected to be improved by having distinct metals serving the two roles and thus provided an impetus to study heterodinuclear catalysts.
In 2014, our group reported the first example of such a system [120]. The catalyst is a mixture of compounds, including both homodinuclear complexes and the heterodinuclear complex. Figure 11 compares the activity and productivity of this catalyst mixture, the homodinuclear complexes alone and in a 50 : 50 combination. It is clear that the mixture containing the heterodinculear catalyst shows an activity that is greater than either of the homodinuclear complexes or combinations of them. This finding provides the first evidence, albeit of a mixture of components, that heterodinuclear complexes are worth investigation in this field of catalysis [120].

Conclusion
The ROCOP of carbon dioxide and epoxides provides a useful means to reduce pollution, to consume carbon dioxide and to produce aliphatic polycarbonates. Low-M n , hydroxyl-endcapped, polycarbonate polyols are emerging as useful materials for the production of higher polymers, notably polyurethanes. There remain significant opportunities and challenges to understand and optimize the material properties of the polymers depending on the catalyst and raw materials available. Catalysis plays a key role in the success and scope of this reaction. The selection of the catalyst is central to being able to control features such as the rate, productivity, selectivity, regio-/stereo-chemistry, polymer molecular weight, polymer end-groups and polymer composition and to produce block copolymers. So far, several successful heterogeneous and homogeneous catalyst types have been studied. A common theme is that the pathways are often proposed to occur via dinuclear or bimetallic routes, whereby one metal activates the epoxide, while the other metal provides the nucleophile (carbonate) to attack and ring-open the epoxide. Thus, recently, a number of improvements to catalyst rate and selectivity have been achieved by targeting structures that optimize the coordination chemistry of dinuclear catalysts. The discovery and further development of dinuclear and bimetallic homo-and heterogeneous catalysts is an important area for future research.