1 Introduction

The accurate theoretical description of electronically excited states remains a very difficult but important task for many applications including photocatalysis, [1] light harvesting, [2] photostability, [3, 4] photosensitizers, [5,6,7,8,9] and many more [10]. Already the description of absorption spectra might represent a challenge, e.g., when Rydberg states or charge-transfer excitations are encountered [10]. The complexity is further enhanced outside the Franck–Condon (FC) region, where single-reference methods become insufficient due to state crossings and more demanding multi-reference methods are needed [10,11,12]. Accurate description of conical intersections (CIs) represents an important task for the understanding of photochemical processes [13,14,15,16]. Another layer of difficulty is added in the case of theoretical investigations of metal complexes due to the presence of near-degenerate electronic states and relativistic effects [17].

Catalysis on metals has become a large focus in research due to their importance in, e.g., ammonia and methanol synthesis, [18, 19] but also for photoactivation [1, 20]. Due to its atmospheric relevance, especially activation and transformation of carbon dioxide are addressed in an increasing number of recent studies [21,22,23]. The direct charge transfer of an electron onto CO2 forming an activated bent CO2– could be a key step in the activation process. However, the CO2– ion itself is metastable [24,25,26,27] and has to be hydrated or attached to a metal center to gain stability [28,29,30,31]. Only recently, we revealed that hydration of only three water molecules already leads to activation of carbon dioxide forming a CO2– ligand on a Mg2+ core [32, 33]. The combination of theory and UV–VIS spectroscopy provides a powerful tool for characterization of complicated processes in ionic metal complexes upon excitation [34]. With this approach, we investigated copper formate clusters relevant for carbon dioxide activation on copper centers in the ground state as well as electronically excited states [35,36,37].

Due to their intriguing charge-transfer chemistry in the ground state, hydrated magnesium ions Mg+(H2O)n have been studied extensively by theory and experiment [38,39,40,41,42,43,44,45,46]. Photodissociation spectroscopy of hydrated magnesium ions, Mg+(H2O)n, provides a suitable model system and benchmark tool for theoretical calculations to investigate the hydrogen production on metal centers [47,48,49,50,51]. Theoretical investigations of Mg+ complexes go back to 1991 [52, 53]. Already in 1993, the group of Duncan found partial CO loss in [Mg(CO2)]+ after excitation within one of the two separated 3s–3p excitation bands of Mg+, pointing toward activation of carbon dioxide within the excited state [54,55,56]. Whereas the character, vibrational assignment and splitting of these bands was well understood, the photochemical process leading to the activation of the CO2 ligand remained unclear due to computational limitations.

Here, we combine investigation of the excited state PES with photodissociation experiments in the gas phase to analyze the photochemical activation of CO2 on a Mg+ core with and without an additional water molecule.

2 Experimental and theoretical methods

The experimental gas-phase action spectra are obtained using FT-ICR mass spectrometry with the cell cooled to the temperature of about 80 K, see Supporting Information for details. For quantum chemical calculations, we used the molecular structures published in Ref. [32] as a starting point. Structures in the ground electronic state are modeled using the Coupled Cluster Singles and Doubles, CCSD/aug-cc-pVDZ, level of theory. For excited state calculations, Equation of Motion CCSD, EOM-CCSD/def2TZVP, and Multi-reference Configuration Interaction, MRCI/def2TZVP, single-point calculations are applied. The def2TZVP basis set is sufficient to describe the orbitals participating in the photochemistry of the system as no Rydberg states are observed among the low-lying excitations. An active space of seven electrons in nine orbitals (7,9) was employed. It includes the valence 3s electron of Mg+ and six electrons of the CO2 ligand which are important for description of the bending coordinate; nine orbitals allow for inclusion of up to six doublet electronic states. Relaxed excited state potential energy surface scans are performed on the Complete Active Space Self-Consistent Field, CASSCF/def2TZVP, level of theory. In comparison with EOM-CCSD, optimization on the CASSCF(7,9)/def2TZVP level of theory yields reasonable structures with similar minima. MRCI(7,9) single-point calculations are performed to include dynamic correlation.

For spectra modeling, we used Franck–Condon simulations [57, 58] as well as the linearized reflection principle within the harmonic approximation [59,60,61] at the EOM-CCSD/aug-cc-pVDZ level of theory. The Gaussian 16 software was employed for CCSD and EOM-CCSD calculations [62], the Molpro software package for CASSCF and MRCI calculations [63, 64].

3 Results and discussion

3.1 UV/VIS spectroscopy

We start our analysis with the spectrum of [Mg(CO2)]+ from the group of Duncan illustrated in Fig. 1a. At 3.66 eV, a vibrationally resolved absorption band was observed, along with the flank of a second peak at about 4.66 eV [54, 56]. With our tunable OPO system, we have spectral access to the full second absorption, for which we observe a band with a maximum at 4.88 eV, which is fitted very well with a single Gaussian. The resolved vibrational progression in the low-energy band indicates excitation into a bound state while the structureless high-energy band suggests no excited state minimum in the vicinity of the Franck–Condon point in this state. Based on these assumptions, we modeled the spectrum in Fig. 1a. The calculated bands are shifted to slightly lower energies by about 0.05 eV. The vibrational resolution due to the CO2 stretching vibration at 360 cm−1 in the low-energy band yields a very good agreement between experiment and theory in the population of states even within the harmonic approximation. The width of the high-energy peak is well reproduced within the linearized reflection principle approximation.

Fig. 1
figure 1

Experimental photodissociation and modeled absorption spectra for (a) [Mg(CO2)]+ and (b) [Mg(CO2)(H2O)]+. For modeling, Franck–Condon simulations shifted by 0.046 eV are used for the low-energy band in (a), the linearized reflection principle otherwise. The EOM-CCSD/aug-cc-pVDZ//CCSD/aug-cc-pVDZ approach was employed. Data for the blue curve in (a) are taken from Ref. [54]

Full size image

Figure 1b shows spectral changes upon hydration in the [Mg(CO2)(H2O)]+ ion. Weak fragment signals corresponding to loss of CO2 were obtained in the range from 2.8 to 4.2 eV, where two well-separated bands with similar intensities are observed with maxima around 3.2 and 3.7 eV. Their shape is rather broad, which does not suggest excitation into a bound state near the FC point. Modeling of the spectra based on the linearized reflection principle at T = 0 K yields good agreement in excitation energies, with discrepancies smaller than 0.1 eV. However, the experimental width is significantly larger, almost by a factor of two. This points toward thermal effects playing an important role, with a more floppy ground state minimum compared to the case of bare [Mg(CO2)]+. To improve spectrum modeling, path integral molecular dynamics simulation on the CCSD potential energy surface would be possibly needed, lying beyond the scope of the present study.

3.2 Photochemistry

Decomposition pathways in [Mg(CO2)]+ highly depend on the investigated band. While for the first band, only CO2 loss (reaction (1) in Table 1) was reported by the group of Duncan [54,55,56], CO loss was observed additionally (reaction (2)) in the flank of the high-energy band. Here, the branching ratio shifted in favor of [MgO]+ and reaction (2) toward higher energies [54]. In line with this observation, we only detected the CO loss channel for the high-energy band, suggesting it is the predominant decomposition channel. Presumably due to the poor signal-to-noise ratio in our experiment, we could not detect the competing Mg+ fragment reported by Duncan.

Table 1 Possible decomposition channels shown in Fig. 1 and Figure S1 with the calculated reaction energy ∆E
Full size table

Upon hydration, excitation within the first two absorption bands leads only to CO2 molecule loss, reaction (3). This is energetically the most favorable decomposition channel in the ground state, which suggests internal conversion. When the pulse energy is tripled and the number of laser pulses is doubled, [Mg(CO2)(OH)]+, [MgO(H2O)]+ and [Mg(OH)]+ fragments are observed in smaller amounts starting at about 3.6 eV, see Figure S1. The respective reactions (4–6) would be accessible in the ground state with the available energy after internal conversion. However, the energetically preferred water evaporation is not observed, reaction (7). The observed pulse energy dependence and missing water loss in the experiment thus suggest the involvement of multiphoton processes.

We start our theoretical photochemical investigation with the linear [Mg(CO2)]+ complex. In the FC point, the singly occupied molecular orbital in the ground state is the 3s orbital of Mg+ which is slightly perturbed by the CO2 ligand (Fig. 2a). The first excitations correspond to the excitation of the 3s electron into 3px,y orbitals of Mg+ with a vertical excitation energy of 3.70 eV (MRCI(7,9)/def2TZVP//CCSD/aug-cc-pVDZ) in the FC point. These are shifted from three 3s–3p transitions of bare Mg+ at 4.31 eV (MRCI(1,4)/def2TZVP) due to destabilization of the 3s orbital upon binding of CO2. The excitation at 4.93 eV (MRCI(7,9)/def2TZVP//CCSD/aug-cc-pVDZ) corresponds to excitation into the significantly perturbed 3pz orbital of Mg+, which is collinear with the CO2 ligand. This is consistent with previous interpretation and the similar case of [Mg(H2O)]+ [47, 49, 54,55,56].

Fig. 2
figure 2

Relaxed PES scan of the CO2 angle in [Mg(CO2)]+ for (a) D1 and (b) D3 states at the CASSCF(7,9)/def2-TZVP and MRCI(7,9)//CASSCF(7,9)/def2-TZVP levels. FC point transition energies are given at the MRCI(7,9)/def2-TZVP//CCSD/aug-cc-pVDZ level. EOM-CCSD/def2-TZVP//CASSCF(7,9)/def2-TZVP energies are provided around the minima in (a). The structures and the most important singly occupied orbitals according to CASSCF CI vectors are shown for selected points

Full size image

To investigate whether the observed dissociation pathway involving CO loss can be explained by electron transfer from Mg+ to the CO2 molecule, we performed a relaxed scan along the CO2 angle for the D1 and D3 states on the MRCI(7,9)/def2TZVP//CASSCF(7,9)/def2TZVP level of theory in Fig. 2a, b, respectively.

The D1 minimum in the direct vicinity of the FC point lies only 0.08 eV below the D1 excitation energy, still with a linear CO2 molecule and minimal structural changes. This minimum is well separated with a barrier of about 0.48 eV from an energetically lower-lying D1 minimum where CO2 is activated with an O–C–O angle of about 130°. Here, the electron is transferred from the Mg+ center to the antibonding π* orbital of the CO2 ligand. During the charge transfer, the originally degenerate 3px,y orbitals split. According to the vibrationally resolved experimental spectrum for D1, the D1 minimum with bound linear CO2 is the target state. However, fluorescence from this minimum can only provide an energy of 0.12 eV, which is insufficient for the CO2 loss observed according to reaction (1). Similarly, CO2 loss within the excited state via reaction (1*) lies too high in energy, requiring 4.97 eV in total. The ground and the excited state manifold are well separated, and no curve crossing seems to be accessible.

However, a D1 minimum can be reached after excitation into the D3 state without any barrier (Fig. 2b), which corresponds to a charge-transfer complex between an activated, bent CO2– radical anion and Mg2+. This observation is consistent with the experimental structureless absorption band suggesting a dissociative shape of the PES around the FC point for excitation into D3. Orbital analysis of the relaxed structure in Fig. 2b confirms that the singly occupied 3pz-orbital of Mg+ mixes with the antibonding π* orbital of the carbon dioxide ligand upon bending. In the D1 minimum, the electron is fully transferred from Mg+ to the antibonding π* orbital of the CO2 molecule, the C–O bonds are slightly weakened (the one next to Mg from 1.19 to 1.27 Å) while the Mg–O bond is significantly strengthened (from 2.12 to 1.73 Å). However, for dissociation of the CO molecule according to reaction (2), the electron still needs to be transferred to one of the two isoenergetic p orbitals of the oxygen ligand in [MgO]+. Optimization of a conical intersection toward these states suggests that this process costs about 5.7 eV (EOM-CCSD/def2TZVP), see Figure S2 for the respective interpolation. Thus, the electron transfer along this decomposition channel is not accessible with the energy available upon excitation into D3. The dissociation of CO2 within the excited state, reaction (1*), is likely inaccessible, given the thermal energy calculated as 0.03 eV at 80 K. Curve crossings into the ground state seem unattainable with the available energy as well.

In planar [Mg(CO2)(H2O)]+, the weakly bound CO2 ligand is oriented end-on on the same side of the Mg+ core as the more strongly bound water molecule. The 3s and two 3p orbitals are perturbed by the ligands (Fig. 3a), only the 3p orbital perpendicular to the molecular plane stays unperturbed: The excitations into the D1 and D3 states are shifted to lower energies of 3.28 and 4.70 eV, respectively, the excitation into the D2 state remains almost unshifted at 3.78 eV.

Fig. 3
figure 3

Relaxed PES scan of the CO2 angle in [Mg(CO2)(H2O)]+ for (a) D1 and (b) D2 at the CASSCF(7,9)/def2-TZVP and MRCI(5,8)//CASSCF(7,9)/def2-TZVP along with the FC point excitation energies at the MRCI(5,8)/def2-TZVP//CCSD/aug-cc-pVDZ level of theory. The structures and the most important singly occupied orbitals according to CASSCF CI vectors are shown for selected structure and states. Interpolation between the FC point and the conical intersection of (c) D3/D2 and (d) D2/D1

Full size image

To investigate how the hydration changes the photochemical activation of CO2 in the excited state, relaxed CO2 angle scans for the D1 and D2 states are shown in Fig. 3a, b. For D1, multiple minima are found again, with an almost linear and a bent CO2 molecule (note that another minimum with bent CO2 and a very similar energy in D1 exists with a flipped CO2 bending angle, see Figure S3 for the respective scan). The D1 minimum with linear CO2 has a dramatically different structure compared to the FC point. After promoting the electron from the 3s to the 3p orbital of Mg+, the ion linearizes analogously to [Mg(H2O)2]+ [47, 49] and the water molecule rotates by 90° to maximize the interaction of the positively charged hydrogen atoms with the electron in the 3p orbital of Mg+, gaining 0.53 eV of internal energy. Contrary to the case without the water ligand, the bent minimum with an activated CO2– ligand is now accessible with the energy available after excitation. CO loss from this minimum would only require 2.51 eV for the hydrated case, reaction (5). However, the charge transfer into the p orbital of the oxygen to access this decomposition channel from D1 is still very expensive with at least 5.45 eV (EOM-CCSD/def2TZVP), similar to the case without water, see Figure S4 for the respective interpolation. Likewise, decomposition within the excited state manifold is not accessible with the available energy of a single photon, see reaction (3*). Contrary to the case without the water ligand, the gained internal energy after fluorescence from any D1 minimum is enough for CO2 evaporation, see reaction (3).

After excitation into D2, the D1/D2 conical intersection can be reached barrierlessly, see Fig. 3b. Again, the antibonding π* orbital of the carbon dioxide ligand mixes with the occupied 3p orbital from Mg+ in the optimized state upon bending. Afterward, access to a similar bent minimum involving a CO2– ligand in the D1 state can be expected and similar access to decomposition channels with the available energy.

We also investigated the reaction pathways after excitation into the D3 state and found pathways for direct internal conversion into D2 and D1 states (see Fig. 3c, d for interpolations). In both cases, the PES leads monotonically downhill from the FC point to the conical intersections. The orbital analysis of the CIs in Fig. 3c, d shows that for the D3/D2 CI, the water molecule and CO2 rearrange to perturb the 3p orbital of Mg+ which is occupied in D2 while providing more space for the 3p orbital populated in D3. For the CI into D1, the two ligands rearrange to provide space for two 3p orbitals perpendicular to the molecular axis. However, the ground state is well separated in all scans. The excitation energy into D3 is sufficient to evaporate a CO2 molecule from Mg*+ via reaction (3*).

The PES scheme in Fig. 4 summarizes our findings. For [Mg(CO2)]+, excitation into D1,2 leads to a bound state where it can only absorb another photon in order to decompose. From the linear D1 minimum, several higher lying states are resonantly accessible with an excitation energy of 3.2–4.0 eV (EOM-CCSD/aug-cc-pVTZ). These transitions mostly correspond to excitation of the electron into the empty 3d orbitals of Mg+. Internal conversion finally leads to CO2 loss. Formation of MgO+ can happen through an intermediate step after excitation from the ground state into D3. Here, an electron can be transferred from Mg*+ to the CO2 molecule, weakening the C–O bond. However, the observed CO loss is hindered by a significant barrier. Decomposition after internal conversion into the ground state of [Mg(CO2)]+ can be ruled out as reaction (1) was not observed in the high-energy band in our experiment which should have a significant contribution otherwise. Furthermore, the states are well separated in calculations. Therefore, the CO loss rather takes place photochemically through absorption of an additional photon from the D1 minimum involving the CO2– unit. With the excitation energy, higher lying states are accessible around an excitation energy of about 4.6–4.9 eV (EOM-CCSD/aug-cc-pVTZ). Contrary to the multiphoton excitation in the D1 minimum with a linear CO2 molecule, we excite here directly a lone pair electron in the CO2– ligand to its out of the bending plane π* orbital while Mg2+ plays a minor role. This explains the different photochemical decomposition pathways in the two experimental bands. Fluorescence or multiphoton excitation into 3pz, which requires about 4.5 eV (EOM-CCSD/aug-cc-pVTZ) in the bent D1 minimum, can explain the partially observed CO2 loss in the flank of the high energy band previously observed by the group of Duncan [54].

Fig. 4
figure 4

Simplified reaction scheme illustrating the predominantly observed experimental dissociation channels in Fig. 1 for (a) [Mg(CO2)]+ and (b) [Mg(CO2)(H2O)]+

Full size image

In [Mg(CO2)(H2O)]+, CO2 loss can already happen after fluorescing a photon in the linear D1 minimum or the D1 minima with a bent CO2. Upon excitation into D2, a CI into D1 can be accessed allowing similar decomposition pathways. Direct decomposition via CO loss is hindered again by a barrier. Decomposition after internal conversion into the ground state can be ruled out because water loss (reaction (7)) is not observed in our cooled experiment and the ground state is separated from the excited state manifold. Therefore, the [Mg(CO2)(OH)]+, [MgO(H2O)]+ and [Mg(OH)]+ fragments observed only with significantly higher laser power and with shifting branching ratios, see Figure S1, occur via multiphoton processes. Excitations into higher lying states are accessible with resonant excitation energies of about 3.6 and 3.5–4.2 eV (EOM-CCSD/aug-cc-pVTZ) in the bent D1 minimum of Fig. 3a and its flipped version in Figure S3, respectively.

Furthermore, formation of [Mg(OH)]+ as the second most intense channel likely occurs sequentially by absorption of the predominant [Mg(H2O)]+ fragment as it has an intense absorption band in this range and loses an H atom [47, 49]. Additionally, H loss from the comparable [Mg(H2O)2]+ case happens via a multiphoton-process, in line with our interpretation [47, 49].

4 Conclusion

We investigated the photochemistry of [Mg(CO2)]+ and [Mg(CO2)(H2O)]+ as model systems for the role of metal/ligand interactions in the photochemical activation of CO2 by a combination of ab initio calculations and mass spectrometry experiments in the gas phase. The observed decomposition channels are highly state selective. [Mg(CO2)]+ loses the CO2 ligand in the low-energy 3s–3px,y band via a multiphoton excitation into the Mg 3d shell. With the provided energy of the high-energy 3s–3pz band, CO2 is activated through a charge transfer from Mg*+ in the excited state manifold and forms a bent CO2– ligand on a Mg2+ center. This leads to predominant CO loss after absorption of an additional photon by the CO2– ligand.

Upon hydration with one water molecule, CO2 is activated already in the low-energy 3s–3p band. The rearrangement in the D1 minima provides enough internal energy for CO2 loss through fluorescence in addition to the previous multiphoton process. The [Mg(CO2)(OH)]+, [MgO(H2O)]+ and [Mg(OH)]+ fragments in the D2 band arise from an additional excitation in D1 minima in combination with sequential fragmentation of [Mg(H2O)]+ to [Mg(OH)]+. For the third 3s–3p transitions, we predict direct funneling into the first excited state, providing enough energy to directly evaporate a CO2 molecule on the excited state PES.