Henry Rzepa's blog

Mention carbon dioxide (CO₂) to most chemists and its properties as a metal ligand are not the first aspect that springs to mind. Here thought I might take a look at how it might act as such.

There are up to five binding modes with one metal that one might envisage:

1. Bonded interaction with the metal via just one oxygen atom,
2. Bonded interaction via just the central carbon atom,
3. Bonded interaction via the π-face of one C=O double bond,
4. A weaker non-bonded interaction via carbon, or
5. via oxygen.

Search queries of the Cambridge structure database (CSD) for these five modes are illustrated below (dataDOI: 10.14469/hpc/2524), with the constraints being applied to how many bonds (of unspecified type) each atom carries, along with no disorder and no errors. Thus query 1 is constrained by 1-coordination on one oxygen, and two on the carbon and other oxygen. 

1. This query yields four hits: 10.5517/ccvcdq9, 10.5517/cc12nq6n, 10.5517/cc12nq5m, 10.5517/cc12nq4l. The angle subtended at the central carbon of the CO₂ ranges from 172-176°, a very modest bending of the linear CO₂. There are no examples where the metal is bonded to both oxygens.
2. The next category involves the metal binding just to the central carbon. Two examples are known, differentiated from O-coordination by a more acute angle at the central carbon of 121-132°.
3. The π-coordinated type requires a slightly more complex search query, shown below. The π-complex is defined as adding one coordination to each of one oxygen and the carbon. 

   

   This reveals 16 examples:
   

   The sine of the angle subtended at the centroid of one C-O bond shows that for most of the examples, the metal is close to perpendicular to this bond. The angle subtended at the central carbon ranges from 128-138, rather larger than the examples where the metal is bound just to the carbon. I have picked these two for illustration. The first (dataDOI: 10.5517/cc86r17) contains both CO₂ and CO coordinated to the metal.This one (dataDOI: 10.1021/ic101652e) contains a short metal-centroid distance of 1.78Å (as also does 10.5517/ccz34kr). 
   

   There are two examples where BOTH π-CO bonds are coordinated to a metal; 10.5517/ccqlv7c and 10.5517/ccqlv8d (Ni-centroid distance 1.9Å) but these are intriguing because the two π-complexes are co-planar and not orthogonal.
   

4. The final two cases are defined in the CSD database by having not so much bonds between metal and either C or O, as close intermolecular contacts typical of e.g. hydrogen bonds. This one (dataDOI: 10.5517/cc12nq9r) is to Fe, with a metal-C distance of 2.87Å which is significantly shorter than the anticipated sum of the van der Waals radii of the two atoms. The next (dataDOI: 10.5517/cc12npn2) has a close approach of Co to O of 2.23Å. The angles subtended at the carbon range from 174-180°. There are no convincing examples of close non-bonded approaches of the metal to both oxygen atoms simultaneously.

It is striking that the searches (as defined above) reveal relatively few examples. This might simply be a result of how the compounds are indexed in the CSD, reflected in the coordination constraints applied in the searches. Nevertheless, we see three quite different types of ligand-metal coordination in which bonds can be said to form and a more diffuse spectrum of weaker interactions to carbon dioxide. As a metal ligand, it is certainly interesting! Several deserve their wavefunctions looked at and I might report back on this aspect.