Electronic absorption spectrum and low-lying electronic states of Co2 isolated in solid neon

We report electronic absorption spectra in the mid-, near infrared and visible regions for cobalt dimers isolated in neon matrices.

In addition to the already reported transition near 2.8 eV and above, ten states can be observed spanning a wide energy range from 0.068 to 2.49 eV.

For four excited states in the IR range and two states in the visible, analysis of the vibronic systems leads to estimates of electronic energies, vibrational parameters and bond length variations.

The data are compared to results of photodetachment experiments and recent theoretical predictions.


Within the first row transition metal (TM) dimer series, Co2 should occupy an intermediate position between mid-row elements where high bonding orders imply significant “3d–3d” contribution to the bonding, and Ni2 or Cu2, which are described as primarily “4s–4s” bonded.

Co2 is, however, certainly the least known of the first row series, as its electronic ground state has not yet been experimentally established, nor its bond length actually measured.

A mass spectrometric study1 has yielded an estimate for the dissociation energy (1.69 ± 0.26 eV), but this value relied on many assumptions for the electronic and vibrational partition function.

Important information has been provided by the Co2 photodetachment study of Leopold and Lineberger,2 the main points being ground state frequencies for Co2 and Co2 of 280 ± 20 cm−1 and 240 ± 15 cm−1, respectively, and features indicative of excited states of Co2 some 600 and 5000 cm−1 above the ground state.

Optical studies on Co2 isolated in cryogenic matrices have evidenced strong UV absorptions near 270–280 nm and 340 nm, as well as a much weaker feature centered around 442 nm.3

All these electronic transitions appeared to be broad and structureless.

Next, Dong et al4. using mass-selection/reneutralization of Co2+confirmed the assignment to Co2 isolated in solid argon of a broad, unstructured absorption extending from 425 to 485 nm.

More important, they obtained Raman progressions, using excitations in the 458–488 nm range, yielding ωe = 296.8 ± 5.4 cm−1, ωexe = 2.2 ± 0.8 cm−1 for the ground state, in agreement with the preceding gas phase measurement.2

In contrast, theoretical studies are numerous and their results conflicting.

An ab initio HF-CI study by Shim and Gingerich5 first predicted a 5Σ+g ground state, with a very long re = 2.56 Å, and a very low vibrational frequency, ωe = 162 cm−1.

Several DFT studies have been performed, and differ in their conclusions regarding the ground state nature, suggesting either a 5Δg or a 5Σ ground state.6–12

Barden et al. warn in their systematic study of TM dimers10 that many states with different electron configurations and spin multiplicities lie very close, thus the predicted ground state depends very much on the method used, which indicates the necessity for multi reference ab initio calculations for a correct approximation of the electronic wavefunction.

In absence of these, Gutsev et al11. and Gutsev and Bauschlicher12 have used various DFT methods to predict energies, bond lengths, magnetic moments and vibrational frequencies for neutral and charged Co2.

These DFT studies predict a 5Δg ground state with a relatively short bond length (1.98 Å), but the vibrational frequency (ωe = 386–364 cm−1, depending on the functional used), and the dissociation energies (D0 = 2.51–2.24 eV) are well above the experimental values. 5Σ−u and 5Σ+g states of the same multiplicities were calculated 0.23 and 0.84 eV above the ground state and five other states of different multiplicities were also predicted to lie within 1 eV of the ground state, which underlines the high density of states presumable for Co2.11

Ideally, spectroscopic investigations of transition metal diatomics are best carried out in the gas phase, where rotationally resolved fluorescence13 or two-photon resonant ionization schemes,14 as developed by Morse and coworkers, can bring precise knowledge concerning bond length electronic structure or even dissociation energy.

In absence of these, careful matrix-isolation studies can nevertheless help in providing an overview of energies and relative intensities for a number of electronic and vibrational transitions spanning a broad spectral range.

These studies are best performed in solid neon, a less perturbing and polarizing medium than the heavier rare gases, and where electronic transitions are shifted by less than 1% in most cases.15

In the course of investigations of the reactivity of cobalt atoms and dimers, we were able to isolate large quantities of Co2 molecules in solid neon, and correlate several unreported electronic transitions to this molecule.

The results of the present study allow for a more detailed insight into some of the low-lying electronic states of Co2.


The experimental set-up has been described previously in detail.16

Briefly, Co2 molecules were prepared by co-condensing Co vapor and 77 K-precooled, high purity neon or argon (Air Liquide, France; 99,995%), onto one of six flat, Rh-plated copper mirrors maintained at less than 3 K using a pulse tube closed-cycle cryogenerator (Cryomech PT405, USA).

The metal atom source was enclosed inside another liquid nitrogen trap minimizing greatly the thermal radiation and outgassing impurities.

A tungsten filament, wetted with Co (Alfa Aesar, Germany, 99.995%), was heated to 1400–1600 °C generating the metal vapor.

The metal deposition rate was monitored with the aid of a quartz microbalance and varied from 1 to 20 nmol min−1.

For most samples, this corresponded to 50–1000 ppm Co/Ne molar ratios.

Absorption spectra were recorded between 30 000 and 80 cm−1 on the same samples using a Bruker 120 FT spectrometer.

CaF2, CsI and polyethylene wedged windows were mounted on a rotatable flange separating the interferometer vacuum (10−3 mbar) from that of the cryostatic cell (10−8 mbar).

In the far IR, (80–500 cm−1) a globar (Si-C) was used, with a composite beamsplitter, a polyethylene window and a He-cooled Si–B bolometer.

For the mid-IR region (500–5000 cm−1), a Ge/KBr beamsplitter with a CsI window was used along with a liquid N2-cooled narrow band HgCdTe photoconductor.

For the near IR (4500–9000 cm−1), a tungsten filament source, Si/CaF2 beamsplitter and CaF2 window, liquid N2-cooled InSb or room temperature InGaAs photodiodes were used.

Finally, for the 9000–16 000 cm−1 and 15 000–30 000 cm−1 visible regions, spectra were acquired using TiO2/Quartz beamsplitters, blue-or red-enhanced Si photodiode and either a W-filament or a Xe short arc source in combination with Schott BG38 or VG9 colored glass filters.

The resolution was varied from 0.05 cm−1 to 1 cm−1.

Selective excitations were performed for 10 to 40 min with a 200 W HgXe arc lamp using narrow (10 nm band width) bandpass interference filters, and focussed on the 10 mm target.

Irradiances were measured with an optical power meter, and were of the order of 7, 3, 7, 18 and 14 mW cm−2 for 10 nm bands centered at 280, 365, 405, 435 and 546 nm, respectively.


More than fifteen samples were grown, co-depositing Co atoms and precooled Ne or Ar gas with varying metal/rare gas molar ratios, deposit temperature and thickness.

We optimized the conditions in order to grow half a millimetre-thick, yet highly diluted samples (from 50 to 500 ppm Co/Ne).

Under these conditions, samples are almost completely absorbing in the UV range and the reported weak optical absorption for Co2 in argon near 445 nm (22 500 cm−1)3,4 dominates the spectrum.

Even in solid neon, it remains very broad and without clear structure (Fig. 1).

At lower energies, however, several weaker, new band systems appear, labelled from A to J in order of increasing energy (Fig. 1).

Their intensities varied always consistently when changing either the Co concentration or the matrix temperature, as well as upon selective photoexcitations.

In neon the best samples were obtained depositing at relatively high concentration (250 to 500 ppm Co in Ne) at the lowest temperature achievable (2.8 to 3 K).

When performing careful concentration effects, it was found that all these absorptions grew with a second order dependence with respect to the cobalt concentration at low values (50 to 200 ppm), but the growth rate with respect to concentration leveled off at higher Co/Ne ratios.

Higher concentrations (500 ppm or more) or elevated deposition temperatures did not increase much the product yield, due to formation of higher clusters, as will be discussed below.

In the visible range, weaker absorptions can be detected in the green and red, which cannot be so distinctly observed in argon.

First, on the low energy side of the broad blue band appear two much weaker vibronic (I,J) systems with ≈10 cm−1 line widths and onsets near 19 640 and 20 077 cm−1.

Second, three relatively intense absorptions (F, G, H) appear to the red with drastically larger line widths (210 to 250 cm−1) compared to the I and J systems.

In the infrared, several band systems can be observed, C, D and E spanning from about 2800 to 9000 cm−1, with varying line widths (2 to 6 cm−1) and also rather different relative intensities.

Next, a relatively intense but short progression (B) is found at lower energy with three components near 894, 1171 and 1446 cm−1, and last, a very weak but quite narrow absorption (A) is isolated near 551 cm−1, at the far infrared edge.

No other absorption correlable to this species was found down to 80 cm−1.

In solid argon, only the most intense band systems can be observed, and at slightly differing frequencies.

In general, observation is more difficult in argon, the bands being broader and the matrix medium more scattering in the near infrared and visible ranges.

Increasing the deposition temperature above 18 K helped the transparency, but intensities of all features that we correlate to Co2 decreased.

In both media, increasing the concentration led to growth of additional features (such as a weak band near 995 cm−1) which we can relate to larger Con clusters.

In addition to systematic concentration studies, we checked that none of these absorptions could be due to reactions of Co with residual vacuum impurities (H2, H2O, N2, O2, CO2) by growing samples with added small amounts (500 ppm) with no other result than appearance of reaction products and decrease of the Co2 bands.

While adding 500 ppm of O2, the formation of Co2O2 was observed.

When changing the Co/Ne concentration at constant O2 content, we checked that the bands here reported for Co2 electronic absorptions kept the same second order Co dependence than known Co2O2 bands,17 thus confirming that both species contains two Co atoms.

Ozin and Hanlan3 observed the progressive “bleaching” of the Co2 UV bands upon 270 nm selective irradiation and concomitant growth of Co atomic absorptions, implying that the molecule reaches from the upper state a rapidly dissociating channel and produces back Co atoms.

Also, Dong et al4. failed to obtain resonance Raman signals when using 364 nm UV excitation.

This prompted us to perform UV excitation in these domains, while monitoring changes in the lower energy band systems, in order to check that all these absorptions are due to that same species.

Indeed, marked effects were observed, which are typified in Fig. 2.

All bands presented a partly resolved structure, which is likely due to well-known matrix trapping site effects, as the less stable matrix sites could be annealed out by cycling the solid neon temperature up to 10 K and back (Fig. 2).

Photoexcitations in the F band system at 580 nm or in the reported Co2 absorptions at 435 or 405 nm caused a rearrangement between trapping sites or a small decrease in all band systems.

All the reported band systems can be partially restored by cycling the matrix temperature up to 9 to 10 K to allow atomic diffusion.

More energetic excitation in the deeper UV range at 365, 280 or 265 nm caused progressive bleaching of all band systems reported here, which could again be partially restored, following temperature cycling and, presumably, atomic recombination.

All other species detected in our samples by minor absorption bands did not present that specific behavior, and this confirms assignment of the new bands to the same species, identified previously as Co2.

Table 1 reports frequencies observed for all the new band systems in neon, along with relative integrated intensities for each system.

For the sake of simplicity, we have only retained the values observed for the most stable trapping site, which remains after complete annealing at 11 K. Note that after careful annealing, all bands appear narrower, but present an additional more or less resolved finer triplet substructure (3 to 0.2 cm−1 splittings) for the stable trapping site.

This structure did not appear to change from one sample to the next, nor following further annealing.

Frequencies reported in Table 1 pertain to the central component of this structure.


The new transitions observed for Co2 differ widely in energy, vibrational parameters and relative intensities.

In absence of rotationally resolved, gas-phase data, assignment of the different electronic states can only rely on comparison with theoretical predictions.

To facilitate such comparisons, it is necessary to analyze each spectral region and extract energies and spectroscopic parameters providing clues for future state identification.

The narrow absorption at 551 cm−1 does not fit into any regular pattern with bands observed in the same energy range, and is much narrower than the absorptions near 893 cm−1 and above (about 0.15 versus some 3.5 cm−1 for each triplet component, as observed after annealing).

It thus seems likely to correspond to an isolated A–X (0, 0) vibronic band for an electronic transition of much lower probability (ΔΛ ≠ 0, 1 or even ΔS ≠ 0), with state A corresponding to a very low-lying state with an equilibrium distance very close to that of the ground state, as indicated by the lack of observation of (1, 0) component.

The next three bands fall into a regular progression, with monotonically decreasing intensities, and energies can be well reproduced with parameters Te = 901 ± 3 cm−1, ωe = 280.6 ± 1.3 and ωexe = 1.5 ± 0.5 cm−1 for upper state B. The harmonic frequency is about 15 cm−1 smaller than the ground state value, as reported by Dong et al. from their Raman study in solid argon,4 which is consistent with a slightly elongated bond length.

To quantify this effect, we have simulated the expected intensity pattern using well-known procedure for computation of the Franck–Condon factors,18 supposing Morse potentials for ground and excited states with the parameters derived from spectroscopic measurements, and varying stepwise the bond length variation (Fig. 3).

Following this approximation, one would estimate a bond lengthening of the order of 4.5 ± 1 pm.

The next band series spread between 2837 and 9000 cm−1.

A close inspection reveals however, that these 33 bands can be grouped in a straightforward manner into three interleaved vibronic progressions, presenting no obvious sign of mutual perturbation, and analyzed as due to three systems C–X (n,0) with n = 0 to 3, D–X (n,0) n = 0 to 8 and E–X (n,0) n = 0 to 19, and the transition energies reproduced with the spectroscopic parameters reported in Table 2.

Note that the electronic energies are gradually increasing from C to E, but the vibrational frequency for the upper state E is slightly larger (257 cm−1) than for states C or D (246.4 and 251.7 cm−1, respectively).

Nevertheless, differences in relative intensities within the progressions indicate substantial changes of the Franck–Condon factors, compatible with progressively increasing bond lengths.

Following the same procedure as described above, simulations based on Morse potentials are proposed in Fig. 4, with bond lengthenings (7, 15 and 22 ± 2 pm) giving the best overall agreement on relative intensities within each band system.

These estimates should be regarded as only indicative, as the molecular potentials depart obviously from Morse-type functions at large vibrational excitations (see for instance the larger intensity of bands with v′ > 12 for system E ← X).

For system E, bands with high vibrational quanta in the upper state give also sign of broadening (fwhm ≈ 6 cm−1), but nevertheless remain well-defined in comparison to absorptions in the visible range.

From relative intensity considerations and comparisons with results recently obtained with other transition metal diatomics,19 it appears likely that transitions from ground to B, C and E states are fully allowed, but that to state D seems less probable.

Recently, Jules and Lombardi20 have extended to transition dimers empirical rules correlating force constants and equilibrium internuclear distances, such as Badger’s or Pauling’s rules.

It is interesting to note that Pauling’s relationship, reported to be the most successful, would predict, for the excited state B bond lengthening, of 5.1 pm vs. 4.5 ± 1 estimated here, or 8 ± 2 pm for the state observed near 600 cm−1 in ref. 2 and here, in acceptable agreement.

For states C, D and E, however, the same rule would predict internuclear distance changes of 16.5, 14.6 and 12.7 pm, quite different from what is estimated here (7, 15 and 22 ± 2 pm, respectively).

Other cases of departure of Badger-type empirical rules have been discussed for excited states of TM metal dimers, such as for Re2.21

Absorptions corresponding to systems F, G and H in the visible are intense and almost two orders of magnitude broader than the IR systems, recorded in the same experimental conditions and even after careful annealing.

This is indicative of a strongly allowed character for these transitions and of very short lifetimes.

Due to their large homogeneous line widths and close proximity, detection of vibrational structure is difficult for F and G, although shoulders are detectable for both.

For H, the homogeneous line width is slightly smaller (fwhm ≈ 230 cm−1) and two shoulders are detectable, which might represent a vibronic structure with ωe ≈ 265 cm−1.

The two next transitions, with origins near 20 000 cm−1 are again much weaker but sharper, at least near their onsets.

Fig. 5 compares the experimental data with simulation based again on Morse potentials, derived from the parameters presented in Table 2.

The data are consistent with states with less elongated bond distances and relatively higher vibrational frequencies than states D and E. Note that after 20 750 cm−1, bands in both systems I and J (corresponding to v′ = 4 and 3, respectively) appear to broaden abruptly, and to progressively blend into the broad, intense absorption centered at 22 200 cm−1, observed previously.3,4

Two points seem interesting to discuss in relation to previous gas phase studies of transition metal diatomics.

First the change in line widths between the IR band systems A to D situated below 1.1 eV, and the systems in the visible range, for which the progressive disappearance of any vibronic structure is conspicuous.

Most of the accurate data on transition metal diatomics is obtained using two photon ionization schemes of jet-cooled molecules, these techniques yield rotationally resolved absorption data in bound excited states.

It also enables observation of abrupt changes in excited state lifetimes14 or drop in ionization signal22 as soon as the excitation energy exceeds the molecular binding energy.

The underlying basis is that, for this type of molecule, the density of excited states is very high, and, whenever above the binding energy, an excitation can no longer be considered as occurring to a single potential surface.

Mixing of bound and dissociative states leads to rapid predissociation as soon as excitation energy exceeds that of the separated atoms.

For molecules trapped in inert solids, this should translate into an abrupt decrease in excited state lifetimes and dramatic increase of the absorption widths blurring the vibronic structure, because of the large interaction with the matrix cage.

This is what has been observed for dimer absorptions in domain slightly above the binding energy for instance with Cu223,24 or Ti2.25

If excitation energies are much in excess, then it can no longer be dissipated in the neon matrix lattice and opening of the matrix cage may lead to dissociation as well.

This is clearly what sets in with 405 nm excitation or happens rapidly with 365 or 280 nm excitations, while part of the molecules can be reformed following atomic diffusion.

Given the estimated low binding energy of Co2 (1.7 ± 0.25 eV),1 the observation of band systems I and J with well-defined and relatively sharp vibronic structures at energies up to 2.56 eV is thus surprising.

This suggests either that some excited states still relatively inefficiently couple to dissociative channels, or that the binding energy is larger than previously thought.

Secondly, a parallel between the excited states at 600 and 5000 ± 50 cm−1 of Co2 observed in the Co2 photodetachment spectrum2 and those observed here in solid neon at 551 and 4450 cm−1 comes to mind.

Leopold and Lineberger26 discuss their results concerning electron photodetachment of Fe2 and Co2 in terms of detachment of the least bound electron from a (3d)n(4s)2 configuration of the anion toward two states of the neutral with (3d)n(4sσg)2 (4sσu) configurations, the 4sσu unpaired electron being either low or high spin–coupled to the d electrons.

For Co2, this would thus imply a (3d)15(4sσg)2(4sσu) ground state configuration (high spin) with a lower spin excited state 600 cm−1 above.

Based on intensity considerations, it was also advanced that the upper states near 5000 cm−1 should have a bond length differing substantially from that of the anion, and perhaps arise from electron photodetachment from orbitals of different atomic parentage, thus involving a different configuration of the bonding d electrons.

The very different overall intensities and Franck–Condon factors observed for A–X and E–X transitions are consistent with this picture.

The 551 cm−1 transition is very sharp and weak, indicative of a low-probability transition, consistent with a long-lived upper state of different spin multiplicity.

The E-X band system centered around 5700 cm−1 clearly corresponds to an upper state with an elongated bond length, and its intensity is suggestive of an allowed, ΔΛ = 0, ±1, ΔS = 0 transition.

Assignments proposed on the basis of DFT results by Gutsev and Khanna11 of the ground and first excited states to the 5Δg and 5Σu states (0.23 eV separation with 382 and 328 cm−1 harmonic frequencies and 198 and 212 pm bond length, respectively) do not seem consistent with our observations in solid neon.

Note that the first two states in these calculations correspond to (3d)16(4sσg)2 electronic configurations, different from those suggested on account of photodetachment spectra.

Many more states can be expected even at low energy.

In view of the great complexity of the electronic structure in transition metal molecules, and the disparity between experimental results and conclusions drawn from single determinant calculations,10–12 it seems that careful theoretical studies using multi reference configuration interaction methods such as those developed recently for Fe227 are very much needed to improve our understanding of the cobalt dimer.