INTRODUCTION

Chemical bonds are an efficient way to store energy, and chemical fuels are one of the most convenient sources of energy on demand through combustion or fuel cells. As a result, our civilization relies on the combustion of fuels, most of which are carbon-based and come from fossil deposits. Carbon dioxide (CO 2 ) is an important combustion product, and the continued use of fossil fuels increases atmospheric CO 2 concentrations. Although our planetary ecosystem as we know it relies on the presence of CO 2 , current and projected concentrations of CO 2 will lead to a continued increase in global average temperatures and sea levels, with immense negative impact. Similarly, CO 2 is an important part of the acid-base chemistry in oceans and lakes, but its current high concentrations are leading to overacidification of the oceans. However, CO 2 could be harnessed as feedstock to produce chemical fuels on an industrial scale, using renewable energy sources to drive the conversion process. A carbon-neutral fuel cycle could be developed along these lines, taking advantage of the unsurpassed energy storage capacity of chemical bonds while also lessening the environmental impact of anthropogenic CO 2 emissions.

At first glance, CO 2 activation seems difficult and costly. It is a very stable molecule, with high bond dissociation energy [525.9 kJ mol -1 (1)] and ionization potential [13.777 eV (2)], and it has a negative electron affinity (see Section 2). Nonetheless, it can weakly donate electron density and has a remarkable ability to accept electron density where it occurs as a ligand or functional group, or as a part thereof. Reducing CO 2 provides the most convenient avenue to its chemical conversion. One-electron reduction of CO 2 is assumed to be the first step in most processes converting CO 2 to chemical fuels or to feedstock for other industrial use (3), although hydride transfer is considered to be an alternative with lower barriers (4). Reduction of CO 2 to more useful molecules requires multiple electrons and protons (3,5), and different reduction processes can yield a number of products, such as carbon monoxide, formate, methanol, or oxalate (which is often invoked in the context of CO 2 fixation rather than conversion to fuel) (6). Carbon monoxide can be used to generate fuels by means of Fischer-Tropsch chemistry (7). Similarly, methanol, another possible product from CO 2 reduction, can be used as feedstock in methanol-to-olefin conversion to produce fuels (8).

Unfortunately, the reduction of CO 2 is energetically unfavorable for both free and solvated CO 2 . Not only is this due to the obvious difference in free energy between CO 2 and final products, but there are also significant reaction barriers to be overcome during the conversion processes (3,9). For CO 2 reduction to be economically viable, suitable catalysts need to be developed and adapted to the process used for the conversion (e.g., electrochemistry, photochemistry, or a combination thereof ).

As in many other areas of chemistry, the simple concept of integer oxidation states does not hold particularly well when it comes to the description of the relevant reactions in CO 2 reduction, and intermediate oxidation states of CO 2 and related species must be considered to describe CO 2 conversion processes. There is an immense body of literature on the catalysis of CO 2 reduction, and this review is not intended to reiterate the many excellent recent reviews in this field (6,(10)(11)(12)(13)(14) or to be comprehensive. Instead, we focus on the characterization of CO 2 in intermediate oxidation states. In this context, we explore CO 2 as a molecule, but more importantly, CO 2 as a functional group or even as a subunit of a larger CO 2 -based unit such as oxalate, characterizing molecular-level properties (for example, electron binding energies and vibrational signatures).

RELEVANT MOLECULAR PROPERTIES OF FREE CO 2 AND ITS IONS

To gain insight into possible avenues to characterize intermediate oxidation states, it is instructive to first look at the molecular properties of CO 2 in different charge states. Its electronic structure can be summarized in its Walsh diagram (15,16) (see Figure 1), which is also a convenient tool to illustrate some of the properties of its ions.

OCO angle

Energy

In the ground state of neutral CO 2 , the highest occupied molecular orbital (HOMO) is the fully occupied 1π g orbital. An excess electron will be accommodated in the 2π u orbital, which is stabilized by bending the molecule, leading to a deviation of the molecular symmetry from D ∞h to C 2v . The singly occupied molecular orbital (SOMO) in this radical anion in its electronic ground state is of a 1 symmetry, with an OCO angle calculated to be 138 • (17), and it can be described as pseudo-antibonding (18). At the same time, the bonding 1π g orbital transforms into a 2 and b 2 orbitals that have been characterized as largely nonbonding (18). Consistent with the pseudoantibonding nature of the HOMO of CO 2 -, its C-O bond length (124 pm) is greater than that of neutral CO 2 (117 pm) (17).

The adiabatic electron affinity of CO 2 is not precisely known but is clearly negative. It is estimated to be -0.6 eV (19); in other words, the free CO 2 -radical anion is unstable against electron emission and is better described as a temporary negative ion state or electron scattering resonance. Quantum chemistry calculations recover this fact (17,20). However, the vertical electron detachment energy (VDE) of CO 2 -is about 1.4 eV (21), indicating that CO 2 -is metastable, and it has been observed in mass spectrometry experiments, with measured lifetimes up to milliseconds (22)(23)(24). In addition, the radical anion can be stabilized by interaction with a matrix [e.g., a rare gas matrix (25)(26)(27)] or by solvation. The solvated CO 2 -radical anion has been observed in bulk solutions (28)(29)(30) as well as in (CO 2 ) n -(n = 6-13) cluster ions (31)(32)(33). Since the SOMO in CO 2

-is (pseudo-)antibonding, its antisymmetric CO stretching vibrational frequency shifts to the red compared to neutral CO 2 , from 2,349 cm -1 to about 1,660 cm -1 (26,34). At the same time, the bending frequency shifts to the blue, from 667 cm -1 to about 714 cm -1

OSCILLATOR COUPLING AND CO STRETCHING MODES IN CO 2

The changes in the symmetric (ν s ) and antisymmetric (ν as ) stretching frequencies upon addition of an excess electron are based on two effects: (a) changes in the mechanical coupling of the two CO oscillators by bending the molecule and (b) weakening of the CO bonds. By itself, the former diminishes the splitting between ν s and ν as , shifting ν as to the red and ν s to the blue, while the latter leads to a redshift for both modes. For ν as , both effects add, exacerbating the overall shift. For ν s , the two effects partially cancel each other. Similar effects are observed for -COO -groups, where a divalent metal ion forces a less obtuse OCO bond angle than would be found in an uncoordinated -COO - group (35). Simple calculations (36) comparing the shifts of ν as and ν s from CO 2 to CO 2 -qualitatively show the effect of weakening the CO bonds separately. For a CO 2 molecule in which the CO oscillators are mechanically decoupled by a very heavy central carbon atom (e.g., 10 4 u), both ν as and ν s start at similar frequencies in the neutral molecule, and both shift to the red by similar amounts upon electron attachment. This is in contrast to regular CO 2 , as described above. (26,34). Also, the bent CO 2

-geometry leads to a decrease in the kinetic coupling of the CO oscillators; the combination of the decreased coupling and the (pseudo-)antibonding SOMO results in a strong redshift of the antisymmetric CO stretching mode but only a weak effect on the symmetric stretching mode, although the latter becomes infrared-allowed in the anion (see the sidebar titled Oscillator Coupling and CO Stretching Modes in CO 2 ). As a consequence of the changes in the vibrational spectrum, the pronounced Fermi resonance between the fundamental of the symmetric CO stretching mode and the bending mode overtone in neutral CO 2 does not exist in CO 2

-. The shifts of the vibrational modes, particularly the redshift of the CO stretching vibrational modes, offer an experimental way to probe the charge density on a CO 2 molecule or even a COO functional group in an R-COO molecule, as we discuss in Section 5.

The electronic structures of CO 2 and CO 2 -containing functional groups also provide an avenue for experimental characterization via, for example, photoelectron spectroscopy. The electron binding energy is a marker for the chemical identity of CO 2 -based species. While the first electronic excited state of neutral CO 2 is in the deep ultraviolet (UV), the radical anion has its lowest excited state in the near UV (37)(38)(39)(40). The electronic absorption band of CO 2 -at about 235 nm has been used for resonance Raman spectroscopy (30). However, CO 2 -can dissociate upon excitation (38)(39)(40), and its rather broad absorption bands limit the use of CO 2 -electronic excited states for characterization of CO 2 -and related systems. In addition, the radical anion can lose its excess electron by detachment or charge transfer to a solvent (30).

The CO 2 + radical cation, i.e., the result of CO 2 oxidation, is a stable ion, in contrast to the radical anion. It has an appearance energy of 13.777 eV (2), highlighting the exceptional stability of the closed-shell neutral CO 2 molecule against ionization. The cation is a linear molecule, and the 1π g orbital is not completely filled. Since this orbital is bonding, the vacancy causes a weakening of the CO bonds compared to the neutral molecule and a concomitant redshift of the CO stretching frequencies and of the bending frequency (41). The antisymmetric stretching frequency is more strongly shifted (to about 1,423 cm -1 ) than the symmetric stretching mode (to about 1,242 cm -1 ). This anomalously strong shift of the antisymmetric stretching mode has been explained by vibronic interaction between the X 2 g and Ã2 u states through this vibrational mode (see, e.g., 42 and references therein). Because of the high ionization potential of neutral CO 2 , the cation is not often observed in the condensed phase, since it is likely to abstract an electron from most solvent molecules, as data on cationic clusters containing CO 2 show (43).

STRUCTURAL AND ELECTRONIC PROPERTIES OF REDUCED DIMERIC CO 2 : THE C 2 O 4 -ION

The C 2 O 4 -ion is particularly interesting for the characterization of CO 2 at intermediate oxidation states because, in a zeroth-order picture, the singly charged anions represent CO 2 moieties at oxidation states of -0.5. In contrast to the CO 2 -monomer anion, the dimer anion is a stable species (21,31,32,44). The CO 2 dimer anion, C 2 O 4 -, has D 2d symmetry (45), and the charge in the dimer is symmetrically distributed over two bent CO 2 units (see Figure 2). The dimer anion structure can be qualitatively understood as a charge-resonance structure mixed between CO 2 -• CO 2 and CO 2 • CO 2 -charge-localized states, giving rise to an intermediate oxidation state of -0.5 for each CO 2 subunit. In line with this picture, the HOMO of the dimer anion can be viewed as a linear combination of the antibonding 6a 1 orbitals of the two bent CO 2 units constituting the dimer (see Figure 2), forming a C-C σ-bond. Coulomb repulsion between the oxygen atoms in the dimer anion stabilizes the D 2d structure. The vertical detachment energy of C 2 O 4 -is 2.79 eV (31), and the adiabatic electron affinity can be estimated to be about 1.7 eV, but photoelectron spectroscopy experiments have not provided a precise value because of the highly congested photoelectron spectrum (21,31,32,44), and the actual value remains unknown. Interestingly, coupled-cluster calculations at the CCSD(T) level ( 45 -. The reaction coordinate corresponds to the localization of the charge on one versus the other CO 2 monomer and can be parametrized using the difference in the OCO bond angles of the two CO 2 units and the C-C distance (see also 33).

122 pm; the C-O bond length is slightly greater than that in a neutral CO 2 dimer (117 pm). The greater C-O bond length reflects the population of an orbital that is net-antibonding for the C-O interaction, similar to the CO 2 -monomer anion (124 pm) (17). The C 2 O 4 -anion has five infrared active normal modes, and some have been observed in rare gas matrix isolation experiments (26,27,46) and in (CO 2 ) n -clusters (33,47,48), where they depend on cluster size. The most intense signature is due to a (degenerate) antisymmetric stretching vibration of e symmetry in the range 1,852-1,900 cm -1 , depending on the matrix. This feature consists of at least four partially resolved components, with the two most prominent peaks at about 1,865 cm -1 and 1,890 cm -1 . Unfortunately, the individual components of this feature have never been rigorously assigned by anharmonic calculations. The symmetric stretching motion of the two CO 2 units gives rise to a strong b 2 band that appears at 1,180-1,190 cm -1 . The similarly strong b 2 band corresponding to the out-of-phase combination of the bending vibrations of the two CO 2 units has been found at 679.9 cm -1 (26) in a Ne matrix and at 674 cm -1 in (CO 2 ) n clusters (48). Two degenerate e symmetry wagging modes are predicted at lower frequencies, but have not been observed experimentally.

The C 2 O 4 -ion has been observed as a core ion in neat CO 2 cluster anions (CO 2 ) n -in the size regimes n ≤ 6 and n = 14-17, whereas the core ion for clusters between those size regimes is a CO 2 -anion (31)(32)(33)44). This curious behavior reflects a subtle balance between intrinsic electronic stability (which favors the C 2 O 4 -ion at the smallest cluster sizes) and the size-dependent solvation energy that a given core ion in a (CO 2 ) n -cluster gains, which cannot be described using dielectric continuum models for the solvent. Unlike solvation by CO 2 molecules, hydration of

-leads to a localization of the excess electron on a single CO 2 moiety for as few as two water molecules (38,49,50). Solvation by methanol also shows preferential localization on a single CO 2 unit (51). The response of C 2 O 4

-to solvation reflects the delocalized nature of the dimer anion and its high inherent polarizability.

CO 2 -BASED CONJUGATE BASE ANIONS: FORMATE, OXALATE, AND CARBONATE

Formate (HCO 2 -)

From the viewpoint of intermediate oxidation states of CO 2 , in the formate anion (HCO 2 -), the CO 2 unit is bent, bearing significant negative charge (52), with the hydrogen atom binding to the carbon atom, resulting in overall C 2v symmetry. Negative ion photoelectron spectroscopy experiments determined the adiabatic electron binding energy to be 3.498 ± 0.015 eV (53). The OCO bond angle for gas-phase HCO 2 -was determined in CCSD(T) calculations to be 130 • , with C-O bond lengths of 125-126 pm (54). The symmetric and antisymmetric C-O stretching modes appear at ν s = 1,314 and ν as = 1,622 cm -1 , respectively, measured for argon-tagged gas-phase HCO 2 -(52, 55); these values are within 10 cm -1 of the observed values for matrix-isolated HCO 2 - (56). These modes shift slightly more in water, with a predicted change in OCO angle of less than one degree, and observed vibrations lie within 30 cm -1 of solution values (see 55 and references therein). Gerardi et al. ( 55) observed the ν s + ν as combination band at 2,928 cm -1 in the gas phase.

The C-H stretching mode in HCO 2 -is highly anharmonic, with a fundamental energy that is anomalously low in the gas phase and in rare gas matrices, with a calculated C-H bond distance of 113 pm and a measured gas-phase vibration at 2,449 cm -1 . Upon solvation with just one water molecule, which binds to the COO -group, ν CH shifts by more than 100 cm -1 to the blue, bringing this mode much closer to the bulk value. In both aqueous and polycrystalline NaHCO 2 , ν CH is in the range 2,800-2,830 cm -1 (57, 58).

Oxalate ([C 2 O 4 ] 2-) and Hydrogen Oxalate ([C 2 O 4 H] -)

The oxalate ion, [C 2 O 4 ] 2-, is a product of CO 2 reduction under aprotic conditions (59), and it also appears as a ligand in metal-CO 2 cluster ions (see Section 5), making it relevant for the present discussion. It can be thought of as two CO 2 -monomer ions with an overall charge of -2 e symmetrically distributed between the two fully reduced CO 2 subunits. Not surprisingly, the bare singlet dianion species [C 2 O 4 ] 2-has negative vertical and adiabatic detachment energies and is unstable with respect to autodetachment (60,61). For the (hypothetical) free dianion, both the staggered structure with D 2d symmetry and the planar structure with D 2h symmetry have been evaluated computationally, with the D 2d form being consistently lower in energy (see, e.g., [61][62][63][64][65]. Rotation about the long C-C single bond between the two CO 2 units in the dimer (there is essentially no C-C π-bonding) connects the two structures, with the D 2h energy lying 18.0 kJ mol -1 above the D 2d energy (65). Because of Coulomb repulsion, the C-C single bond between the two CO 2 units is quite long; it is typically predicted to be about 155 pm for the structure with D 2d symmetry and about 106 pm in the D 2h structure.

Although the free oxalate dianion is unstable with respect to electron loss, it is stabilized by

) n clusters starting at n = 3 (61, 66), and the vertical and adiabatic electron binding energies increase systematically with increasing solvation up to cluster sizes of n = 40. Extrapolation from cluster values to bulk (67) predicts a VDE of 6.92 eV, close to the experimental value in bulk aqueous solution, VDE = 7.32 eV (68).

For small clusters, the hydrated [C 2 O 4 ] 2-ion has a reduced O-C-C-O torsion angle compared to the structure with D 2d symmetry, but it remains nonplanar. The first four water molecules interact strongly with the [C 2 O 4 ] 2-ion through two hydrogen bonds each, lacking any water-water interactions; with additional water molecules, however, a water network begins to form around the ion. In bulk solution, infrared and Raman spectra of aqueous oxalate ion showed that the [C 2 O 4 ] 2- ion takes a D 2d form (69)(70)(71). However, thermal fluctuations in aqueous solutions are likely to lead to sampling of torsion angles of up to ±30 • about the global minimum torsion angle at 90 • (65).

The vibrational modes of aqueous [C 2 O 4 ] 2-have been assigned, after some debate in the literature, using isotopic labeling experiments and computational analysis (see 70-72 and references therein). Infrared signatures assigned to the D 2d symmetric stretching modes are observed at 905 cm -1 and 1,490 cm -1 , and antisymmetric stretching modes appear at 1,310 cm -1 and 1,575 cm -1 , respectively. Compared to CO 2 -, the calculated C-O bond length increases from 123 pm to 128 pm, and the O-C-O bond angles decrease from 138 • to 126 • (27,65).

Addition of a proton to oxalate results in the formation of hydrogen oxalate, [C 2 O 4 H] -, a singly charged ion that can be thought of as either protonated oxalate or deprotonated oxalic acid, which is stable with respect to autodetachment (65,73). In the gas phase, the proton binds to one of the four oxygen atoms and forms an intramolecular hydrogen bond to an oxygen atom in the CO 2 group, resulting in a planar structure. The presence of the proton causes significant decoupling of the four CO stretching oscillators. Vibrational spectroscopy of cryogenically cooled [C 2 O 4 H] - monoanions (73) shows the free carbonyl stretch (the C= =O stretch on the COOH side of the ion) at 1,766 cm -1 and the symmetric and antisymmetric C-O stretching modes of the COO - group at about 1,320 cm -1 and 1,700 cm -1 , respectively. The antisymmetric stretching frequency of the COO -moiety in the hydrogen oxalate anion falls between the values of the CO 2 -anion and neutral free CO 2 (2,349 cm -1 ) (34), which is consistent with the idea that the COO -group in hydrogen oxalate is in an intermediate oxidation state, i.e., not fully reduced (16).

Carbonate (CO 2 2-) and Bicarbonate (HCO 3 -)

In the context of CO 2 at intermediate oxidation states, carbonate and related species fill an interesting role. It is not straightforward to talk about carbonate as CO 2 in a reduced or oxidized form, since it is a product of acid-base chemistry rather than redox chemistry. It is instructive, however, to consider a hypothetical, free CO 3 2-dianion in its D 3h structure, viewing it (somewhat artificially) as a CO 2 moiety with an added oxygen atom, where the formal charge on each oxygen atom is -2/3 e. From this point of view, the CO 2 moiety is bent and negatively charged, and many of its properties (particularly its vibrational behavior) are close to those of a negatively charged, CO 2 -based functional group or ion.

The near-synonymous use of CO 2 and carbonate in (aqueous) solution is based on the set of equilibria

The delicate balance of these equilibria has a significant impact in a range of fields, including ocean acidification, carbonate dissolution in mineralogy and geology, and regulation of blood pH (74,75). When CO 2 is dissolved in water, the equilibrium between carbonic acid and bicarbonate so heavily favors the conjugate base (and the reaction of CO 2 with water is so slow) that less than 1% of the solution exists as H 2 CO 3 . The balance is complicated by factors such as solvation-catalyzed CO 2 + H 2 O reactions, the equilibrium between CO 2 (g) and CO 2 (aq), and autoprotolysis (76).

The carbonate dianion, CO 3 2-, is predicted to have D 3h symmetry in the gas phase, although it has never been isolated and is unstable against electron loss (77)(78)(79). In aqueous solution, infrared and Raman spectra demonstrate that the D 3h structure is never present, even in very dilute solutions (76,(80)(81)(82)(83)(84)(85)(86). Notably, the nominally infrared-forbidden symmetric C-O stretching mode at about 1,064 cm -1 appears strongly in both infrared and Raman spectra. Similarly, the out-of-plane deformation mode, ν 2 , appears in both infrared and Raman spectra near 880 cm -1 , even though it is nominally Raman-forbidden (assuming D 3h symmetry). Asymmetric solvation and ion pairing have been implicated in breaking the symmetry of CO 3 2-in aqueous solution. The degeneracy of the antisymmetric C-O stretching mode is lifted, giving rise to two modes around 1,400 cm -1 , split by 80-370 cm -1 (87). Their splitting depends on whether the interaction with the environment (e.g., ions or surfaces) is predominantly monodentate or bidentate (81,(87)(88)(89).

Gas-phase bicarbonate, HCO 3 -, is expected to have C s (planar) symmetry, while it has C 1 symmetry in dilute aqueous solutions, with the hydrogen atom protruding from the σ h plane of the CO 3 scaffold (76,86). Notable features include the C-O symmetric and antisymmetric stretching modes at 1,364 cm -1 and 1,634 cm -1 , respectively. The most intense band is the CO stretching mode involving the hydroxyl group, which appears at 1,014 cm -1 and can be used in conjunction with nearby spectral features from other chemical species (CO 3 2-or H 2 O network vibrations) to quantify bicarbonate concentrations in solution (76,90). The OH stretching mode has been assigned in experimental spectra at 2,615 cm -1 , but this value disagrees with harmonic approximation calculations by ∼1,000 cm -1 , suggesting that it is highly anharmonic (86).

In order to assess the microsolvation environment of HCO 3 -in more detail, Garand et al. ( 91) probed small clusters of HCO 3 -• (H 2 O) n (n = 1-10) using infrared multiphoton photodissocia- tion spectroscopy. Their work shows that water preferentially solvates the side of the HCO 3 -ion opposite from the OH group, in which the negative charge is shared between the two free oxygen

Structural motifs of metal-CO 2 interactions.

atoms, and which is reminiscent of a formate functional group. Solvation in this position strengthens the alcoholic C-O bond while weakening the others. The symmetric and antisymmetric CO stretching vibrational modes were found at about 1,300-1,350 cm -1 and 1,650-1,700 cm -1 , respectively. Unfortunately, Garand et al. did not address the OH stretching mode, and a better assessment of its anharmonicity is still needed.

CO 2 -BASED LIGANDS

The interaction of CO 2 with metal surfaces, metal atoms, and metal centers in metal-organic complexes is of great interest in the context of CO 2 reduction catalysis. The interaction of CO 2 with metal atoms has been investigated in solutions and on surfaces, as well as by studying matrixisolated species and mass-selected cluster ions. There are several binding motifs of CO 2 to single metal centers. In the limiting case of a single CO 2 ligand binding to a metal atom, four binding motifs are currently known. In line with the nomenclature of metal-organic chemistry, they can be abbreviated as η 1 -C, η 1 -O, η 2 -(C,O), and η 2 -(O,O), where superscripts denote the number of bonds between the metal atom and the CO 2 ligand and chemical element symbols describe the atoms directly interacting with the metal (see Figure 3). Carbonate and oxalate represent more complicated ligands, with more than just a CO 2 subunit. There are additional binding scenarios for complexes with multiple metal atoms (92,93), but we restrict our discussion to complexes with single metal centers.

One characterization method that is common to most chemical environments is vibrational spectroscopy, and much of our discussion in this section is therefore concerned with the vibrational spectra of CO 2 and related species bound as ligands to metals. Figure 4 summarizes the spectral regions of interest for the CO stretching vibrations of these CO 2 -based ligands (see the discussion below).

Cationic metal-CO 2 complexes have received much attention over the past decades because of interest both in the reactivity of CO 2 with metal surfaces and in the bonding of closed-shell ligand systems to a metal center. Gas-phase experiments, particularly work on mass-selected ions, can give detailed insight into fundamental aspects of metal-CO 2 interactions. Cationic metal-CO 2 complexes and clusters are largely dominated by electrostatic interactions. The thermochemistry of metal-CO 2 interactions in such systems has been examined by the Armentrout and Schwarz groups for a number of metal species (see, e.g., 94, 95 and references therein), and their

Range of wavenumbers for signatures in metal-ligand complexes involving CO 2 -based ligands discussed here (for nomenclature, see Figure 3). Symmetric (s) and antisymmetric (as) characters for the vibrational modes are indicated, except for the oxalate core ion, where the labels of free (f ) and metal-bound (m) are more appropriate. Dashed vertical lines correspond to signatures of neutral (black) and anionic (red ) CO 2 . The line representing the symmetric stretching mode in neutral CO 2 shows an average of the two components of its Fermi resonance with the bending overtone. metal-CO 2 bond strengths typically measure a few hundred millielectronvolts. The Duncan (48,(96)(97)(98)(99)(100)(101)(102) and Mackenzie (103) groups have examined a number of such systems using infrared photodissociation spectroscopy, including V + , Si + , Ni + , Mg + , Fe + , Al + , Ca + , Co + , Rh + , and Ir + . In most [M(CO 2 ) n ] + cluster cations, the CO 2 molecule interacts end-on with the metal ion through one of its oxygen atoms (η 1 -O), dominated by charge-quadrupole interactions and a linear M + -O= =C= =O structure. Typically, M-O distances are calculated to be about 210-230 pm, depending on the metal and the size of the cluster. The CO 2 molecules can be largely characterized as solvent molecules, solvating a metal cation, and each solvent molecule in the first solvation shell has roughly the same type of interaction with the metal center. Spectroscopically, these metal-solvating CO 2 complexes exhibit antisymmetric CO stretching signatures that are typically blueshifted by 20-30 cm -1 compared to free CO 2 . Duncan and coworkers (98) explained this blueshift with the assumption that the CO stretching motion involving the oxygen atom near the metal will encounter an additional repulsive force close to the outer turning point of the vibration as it comes closer to the metal atom. A similar interaction motif can be at play in metal-organic complexes with bulky organic ligands, as has been shown in a recent example in [Ru(bpy)(tpy)(CO 2 )] 2+ (bpy = 2,2 -bipyridine; tpy = terpyridine), where the CO 2 ligand binds to the Ru II metal center in the η 1 -O binding motif (104), according to density functional theory. It is important to note that the CO 2 ligand in this case was added in the gas phase, in contrast to other metal-organic complexes synthesized with CO 2 ligands in solution, where other interaction motifs have been observed (see below). In all these cases, the CO 2 molecules in direct contact with the metal center are calculated to be slightly positively charged, with the CO bond on the metal side elongated by about 1% and the other CO bond shortened by roughly the same amount.

If the combination of the overall charge in a complex or cluster and the oxidation state of the metal favors electron transfer to one or more CO 2 ligands, the resulting metal-CO 2 interactions can differ significantly from the simple electrostatic η 1 -O motif. Vanadium-CO 2 cluster cations present a striking example. Whereas one might expect these clusters to behave similarly to other [M(CO 2 ) n ] + cluster ions, they show vibrational signatures indicating charge transfer to the surrounding CO 2 molecules, resulting in the formation of a CO 2 dimer anion and a V 2+ ion (48). Specifically, a band at 1,140 cm -1 was interpreted as the b 2 band of the CO 2 dimer anion, shifted by about 50 cm -1 toward lower wavenumbers compared to (CO 2 ) n -clusters. An additional band at 1,800 cm -1 was assigned to be the remnant of a group of bands associated with the C 2 O 4 -ion at 1,865 cm -1 and 1,890 cm -1 (see also Section 3). This intracluster reaction is observed only for larger clusters (n ≥ 7), indicating that it is driven by solvation. Together with the emergence of the putative CO 2 dimer anion signature at 1,800 cm -1 , the antisymmetric stretching modes for the solvent molecules in the first shell of the vanadium ion are more strongly shifted to higher wavenumbers (about 2,400 cm -1 ) compared to those associated with first-shell solvent molecules at smaller cluster sizes (2,380 cm -1 ). These blueshifts constitute additional evidence for the intracluster reaction, since they point toward the presence of a more highly charged metal atom. It is interesting to note that similar blueshifted signatures were found for large [Ni(CO 2 ) n ] + and [Si(CO 2 ) n ] + clusters (96,97). This observation suggests that intracluster reactions may occur similarly for those ions, but the lower wavenumber regions that would indicate the presence of CO 2 dimer anion products were not studied for these clusters.

Metal-CO 2 interaction in the presence of an excess electron generally leads to the reduction of one or more CO 2 ligands, as has been observed in metal-CO 2 cluster anions, [M(CO 2 ) n ] -. These clusters can be seen as simplified model systems for the interaction of atomically rough metal surfaces with strongly undercoordinated metal atoms. It is widely believed that on supported nanoparticles or atomically rough surfaces used as cathodes in electrochemical CO 2 reduction, corner and edge sites are the main catalytically active structural moieties (105)(106)(107)(108). Such sites may promote CO 2 reduction via a number of different motifs. Specifically, the electrophilic carbon atom is attracted to the high charge densities at such undercoordinated sites. Depending on the metal and on the cluster size, metal-CO 2 cluster anions can exhibit an astonishing degree of complexity, with many different possible core ion isomers.

Anionic clusters of CO 2 with coinage metal and bismuth atoms exhibit the metalloformate motif (η 1 -C), with a single bond between the metal and the electrophilic C atom of a CO 2 molecule. Ab initio calculations show that there is significant overlap between an occupied s-orbital (in the case of Ag, Au, Cu) (109)(110)(111) or an occupied p-orbital (in the case of Bi) (112) and the empty π * orbital of the CO 2 ligand. This results in a partial reduction of the CO 2 ligand, as the excess electron density is shared between the metal and the CO 2 unit. The presence of excess electron density in the π * orbital of the COO -moiety is encoded in the CO stretching modes of the [M-COO] - complex, as the CO bond strength decreases compared to neutral CO 2 . As a consequence, the average of the frequencies of the symmetric and antisymmetric CO stretching modes shifts to the red by several hundred cm -1 compared to neutral CO 2 (109)(110)(111)(112). The redshift can be quite large, comparable with the difference for these modes between CO 2 -and CO 2 (see Section 2). The excess electron in metalloformate complexes is highly polarizable. As solvation in these complexes occurs preferentially at the COO -functional group, the presence of solvent molecules polarizes the excess charge toward the COO -moiety, and the extent of reduction depends on the level of solvation of the metalloformate anion. As a result, the antisymmetric CO stretching mode of the metalloformates generally shifts to the red with increasing solvation on the COO - group (109)(110)(111)(112). At the same time, an increase in excess electron density on the carboxylic group results in the OCO bond angle becoming more acute. This geometry change leads to a decrease in kinetic coupling of the two CO oscillators that constitute the COO -group. As a consequence, the symmetric CO stretching mode shifts to the blue with increasing solvation, although the average of the symmetric and antisymmetric CO stretching modes continues to shift to the red with increasing charge on the carboxylic group, reflecting the expected weakening of the CO bonds (112). In the limit of a very weak interaction of CO 2 with an atomic anion, a remnant of the η 1 -C motif has also been observed in clusters of CO 2 with the heavier halides (I, Br, Cl) (113)(114)(115)(116). In these cases, only little charge transfer to the CO 2 moiety occurs, and the deviation of the OCO bond angle from a linear configuration is less than 20 • . In qualitative terms, the antisymmetric stretching vibration of a CO 2 ligand bound in an η 1 -C motif will shift to the red with increasing charge on the COO -functional group, and its position is a sensitive probe for the charge distribution (see the sidebar titled Oscillator Coupling and CO Stretching Modes in CO 2 ). Auroformate [AuCOO - (109)] and bismuthoformate [BiCOO - (112)] are particularly clear examples of this behavior (see Figure 5). There is a pseudolinear correlation between the charge on the COO -group and the position of the antisymmetric stretch that is common to the coinage metals (Au, Ag, Cu) in the η 1 -C motif. This curve even extrapolates to the CO 2 -anion in the limit of complete one-electron reduction (16). However, this simple connection does not hold universally for other systems, e.g., for BiCOO - (112).

The η 1 -C binding motif has also been invoked in a variety of metal-organic coordination complexes where one coordination position is occupied by a CO 2 ligand but where bulky ligands restrict access to the metal center (92,93). Specifically, some metal-CO 2 complexes with Rh and Ir centers have been characterized that feature this binding motif, with symmetric and antisymmetric stretching modes assigned to infrared absorptions at 1,210 cm -1 and 1,610 cm -1 for the Rh complex (117) and at 1,230 cm -1 and 1,550 cm -1 for the Ir complex (118), the latter suggesting significant charge transfer to the CO 2 ligand. Tanaka & Ooyama identified the η 1 -C binding motif for CO 2 in a ruthenium bipyridine-based complex via X-ray diffractometry (119). The η 1 -C binding motif has also been proposed as the binding motif for CO 2 in a number of proposed catalytic mechanisms for polypyridine-based CO 2 reduction catalysts with Ru, Mn, Re, Fe, and Co centers (see, e.g., 119-121 and references therein). Other molecular CO 2 reduction catalysts in which the η 1 -C motif has been invoked consist of macrocycles, such as porphyrins or cyclam (10,(122)(123)(124)(125) with metal centers (particularly Fe, Ni and Co). However, while this binding motif has been frequently inferred, its presence often remains an unconfirmed hypothesis.

Another interaction motif observed in the interaction of CO 2 with transition metals is the η 2 -(C,O) motif, where the CO 2 ligand forms both metal-oxygen and metal-carbon bonds. In this binding motif, a significant amount of charge (up to -1 e) is transferred into the π * orbital of the CO 2 ligand, resulting in the complexes having a cationic or neutral metal center and an almost fully reduced CO 2 ligand. The vibrational spectra of gas-phase metal-CO 2 anionic clusters, [M(CO 2 ) n ] -, with M = Fe, Co, Ni, or Cu, show evidence for this binding motif (111,(126)(127)(128), with characteristic vibrational features at 1,650-1,750 cm -1 . When two CO 2 ligands bind to the same metal center in this configuration, the metal exists as a monocation. The infrared signatures of these butterfly complexes can be distinguished from complexes with a single CO 2 ligand by the appearance of strongly coupled in-phase and out-of-phase combinations of symmetric and antisymmetric CO stretches in the individual ligands, which appear around 1,100 cm -1 and 1,700 cm -1 , respectively.

Vibrational spectra of gas-phase metal-oxide CO 2 anionic clusters, [NiO(CO 2 ) n ] -and [CoO(CO 2 ) n ] -(129), also exhibit the η 2 -(C,O) motif, with similar vibrational frequencies. Calculations by the Jarrold group to interpret their photoelectron spectroscopy work suggest the formation of an η 2 -(C,O) connected CO 2 molecule as the transition state in the reaction of CO 2 with W x O y -cluster anions (130,131). The group concluded that reaction of tungsten oxide anions with CO 2 results in the oxidation of the cluster and the formation of unbound CO molecules.

While quantum chemical calculations clearly recover η 2 -(C,O) connectivity for many of the metal-CO 2 cluster anion systems discussed above, the literature for metal-cyclam and other metalorganic complexes in CO 2 reduction catalysis has been ambiguous on whether the metal-CO 2 interaction follows the η 1 -C or the η 2 -(C,O) motif (125,(132)(133)(134)(135). Using isotopic substitution of the CO 2 ligand in Ni(I)cyclam derivatives, Johnson and coworkers (136) have observed evidence for η 2 -(C,O) binding. Overall, the η 2 -(C,O) motif is quite common in neutral and cationic metalorganic complexes for a number of metals (92,93) coordinated to bulky organic ligands.

Interactions of CO 2 with metal or semiconductor surfaces have been a topic of continued research in the past decades due to the interest in electrochemical conversion of CO 2 into hydrocarbon fuels (3). However, the nature of these interactions is difficult to elucidate in operando due to the complex speciation in the condensed-phase environment of an active electrochemical device. Computational work on the interaction of CO 2 with several different metal surfaces has shown a propensity toward metal-carbon and metal-oxygen interactions straddling neighboring surface atoms prior to CO bond cleavage (137)(138)(139).

The formation of two M-O bonds with a single CO 2 ligand in an η 2 -(O,O) motif has been identified in several studies involving neutral alkali metal-CO 2 complexes studied by matrix isolation spectroscopy (140)(141)(142). These metal carbene species are proposed to be highly reactive, due to the lone pair on the carbon atom. Such a species could be an intermediate state in the formation of C-C bonds. In gas-phase ions, this η 2 -(O,O) motif was recently observed by the Uggerud and Asmis groups in Mg(CO 2 )Cl - (143). The infrared signatures of η 2 -(O,O) complexes span a wide range (see Figure 4) and are less characteristic than those of other ligands discussed here.

There are several other CO 2 -containing ligand motifs that involve more than a single CO 2 moiety. One example of more complicated ligand structure is C 2 O 4 in various forms (see also Section 4.2). A C 2 O 4 ligand can take a planar or a twisted configuration (63), depending on the size of the cation and on whether the metal and the C 2 O 4 ligand interact through one or two metal-oxygen bonds. This ligand has been observed in several anionic metal-CO 2 cluster systems, including Bi, Fe, Mn, Ni, and Cu (111,112,126,128,144). A charge analysis reveals that it has a typical partial charge of about -1.5 e and can therefore be most closely described as oxalate in these cases. Formation of oxalato complexes involving single first-row transition metal centers typically results in planar structures (111,126,128,144). The in-phase combinations of the free CO stretches, along with the in-phase combination of the metal-bound CO stretches, account for the main spectroscopic signatures, occurring in the spectral intervals 1,650-1,700 cm -1 and 1,300-1,350 cm -1 , respectively. The out-of-phase combinations of these modes usually do not have appreciable infrared activity, distinguishing them from butterfly complexes, where both in-phase and out-of-phase motions can have significant infrared signatures.

Another example of a more complicated CO 2 -based ligand is carbonate, where the CO 2 molecule interacts with a metal-oxide species. In the context of the present discussion, it is noteworthy that-apart from genuine metal oxides-the metal oxo functionality can be formed in reactions of metal vapor with CO 2 in a laser vaporization plasma by insertion of a metal atom into a CO bond, resulting in the concomitant formation of a carbonyl ligand (126,145). Carbonato ligands bind to the metal with one or two metal-oxygen bonds, and such species have been observed in gas-phase [M(CO 2 ) n ] -complexes with M = Fe, CoO, and NiO (126,129). In bidentate η 2 -carbonato ligands, two of the CO bonds are strongly influenced by binding to the metal center, resulting in an increase of the CO bond length by ∼12 pm and a decrease of the metal-facing OCO bond angle by ∼10 • compared to the 120 • in a (hypothetical) free dianion. Work on neutral carbonate-containing complexes of Sc, ScO, and TiO has identified spectral bands that can be assigned to vibrations of the carbonato ligands (146,147). Typical vibrational signatures of the carbonato ligands can be found at ∼1,600 cm -1 (free CO stretch) and ∼1,400 cm -1 (metalbound antisymmetric CO 2 stretch), with a very weak transition occurring at about 1,000 cm -1 (totally symmetric CO 3 stretch). Infrared signatures in the range 1,600-1,670 cm -1 in spectroelectrochemistry experiments on [Ni(II)cyclam] 2+ solutions under CO 2 atmosphere at sufficiently negative voltages (<1.2 V) were attributed to bicarbonate ligands coordinated to the Ni(II) center (132).

CLOSING REMARKS

Carbon dioxide in intermediate oxidation states is of high importance for CO 2 activation, particularly in the context of CO 2 fixation and solar fuels. Although CO 2 reduction catalysis has been at the center of much work in the past decades, the molecular-level mechanisms involving the different steps in solar fuel catalysis are often poorly understood. In particular, the catalytic cycles themselves, while often built on excellent chemical intuition, need to be confirmed by experiments, and their key intermediate species need to be found and characterized. There are still many open questions in this context, and their answers have direct impact on the validity of models describing the catalytic cycle: What are the active species in molecular CO 2 reduction catalysts (homogeneous or surface immobilized)? What are the active sites in heterogeneous catalyst materials? What is the role of the support for surface-immobilized catalysts? What is the role of the solvent? How does CO 2 bind to the active site?

One of the problems in constructing catalytic cycles and understanding mechanisms is that the analysis of experiments in solution suffers from speciation, particularly under turnover conditions, where multiple species will be present. As a result, although methods such as infrared spectroelectrochemistry can contain detailed spectroscopic information encoding the nature and state of CO 2 and related species in solution, identification and characterization of reaction intermediates in such systems are very difficult.

An alternative route to the detailed characterization of relevant species is to use mass spectrometric preparation of ionic complexes in tandem with laser spectroscopy, and much of this review has been concerned with the results of such experiments. In the characterization of CO 2 in its many possible roles, such a strategy circumvents many of the complications of speciation in solutions, particularly by employing electrospray ionization to bring the relevant species from solutions into the gas phase. This is a powerful approach, and we can learn about the properties of ionic complexes in vacuo and in clusters in unprecedented detail while maintaining relevant ties to their condensed-phase chemistry. The combination of cryogenic ion traps with electrospray ionization and laser spectroscopy (see, e.g., 148-151 and references therein) has added another set of tools that can now yield spectroscopic information on key species in catalytic cycles (see, e.g., 136) and also prepare solvated metal-organic complexes (152).

Of course, none of these experiments, whether in solution or in vacuo, can be exploited to their full potential without quantum chemical calculations, and conversely, experiments will continue to serve as benchmarks for improving the quality of quantum chemical predictions.

Interleaving these three approaches-spectroelectrochemistry, spectroscopy in vacuo, and quantum chemistry-has the potential to answer the many questions concerning CO 2 activation in the future.

SUMMARY POINTS

1. Negative charge on CO 2 or CO 2 -containing functional groups results in lengthening CO bonds and less obtuse OCO bond angles compared to free CO 2 , yielding characteristic infrared signatures.

2. The spectral region around 1,600-1,800 cm -1 is often highly congested and consistent with many ligand motifs, which can greatly complicate the interpretation of infrared spectra.

3. Clusters and complexes in vacuo, alongside quantum chemical calculations, can serve as model systems to obtain detailed structural information on the interaction of CO 2 with metal centers.