Untangling the formation of the cyclic carbon trioxide isomer in low temperature carbon dioxide ices

The formation of the cyclic carbon trioxide isomer, CO3(X 1A1), in carbon-dioxide-rich extraterrestrial ices and in the atmospheres of Earth and Mars were investigated experimentally and theoretically.

Carbon dioxide ices were deposited at 10 K onto a silver (111) single crystal and irradiated with 5 keV electrons.

Upon completion of the electron bombardment, the samples were kept at 10 K and were then annealed to 293 K to release the reactants and newly formed molecules into the gas phase.

The experiment was monitored via a Fourier transform infrared spectrometer in absorption-reflection-absorption (solid state) and through a quadruple mass spectrometer (gas phase) on-line and in situ.

Our investigations indicate that the interaction of an electron with a carbon dioxide molecule is dictated by a carbon–oxygen bond cleavage to form electronically excited (1D) and/or ground state (3P) oxygen atoms plus a carbon monoxide molecule.

About 2% of the oxygen atoms react with carbon dioxide molecules to form the C2v symmetric, cyclic CO3 structure via addition to the carbon–oxygen double bond of the carbon dioxide species; neither the Cs nor the D3h symmetric isomers of carbon trioxide were detected.

Since the addition of O(1D) involves a barrier of a 4–8 kJ mol−1 and the reaction of O(3P) with carbon dioxide to form the carbon trioxide molecule via triplet-singlet intersystem crossing is endoergic by 2 kJ mol−1, the oxygen reactant(s) must have excess kinetic energy (suprathermal oxygen atoms which are not in thermal equilibrium with the surrounding 10 K matrix).

A second reaction pathway of the oxygen atoms involves the formation of ozone via molecular oxygen.

After the irradiation, the carbon dioxide matrix still stores ground state oxygen atoms; these species diffuse even at 10 K and form additional ozone molecules.

Summarized, our investigations show that the cyclic carbon trioxide isomer, CO3(X 1A1), can be formed in low temperature carbon dioxide matrix via addition of suprathermal oxygen atoms to carbon dioxide.

In the solid state, CO3(X 1A1) is being stabilized by phonon interactions.

In the gas phase, however, the initially formed C2v structure is rovibrationally excited and can ring-open to the D3h isomer which in turn rearranges back to the C2v structure and then loses an oxygen atom to ‘recycle’ carbon dioxide.

This process might be of fundamental importance to account for an 18O enrichment in carbon dioxide in the atmospheres of Earth and Mars.


Ever since the first tentative characterization of the carbon trioxide molecule in photolyzed ozone–carbon dioxide ices at 77 K,1 the CO3 species has been a subject of various spectroscopic and theoretical studies.

Moll et al1. and Jacox et al2. assigned four fundamentals at 2045 cm−1 (CO stretch), 1073 cm−1 (O–O stretch), 972 cm−1 (C–O stretch), 593 cm−1 (C–O stretch), and 568 cm−1 (O–CO stretch) in low temperature carbon dioxide matrices; in argon matrices, these absorptions were shifted to 2053 cm−1, 1070 cm−1, 975 cm−1, and 564 cm−1;3 no feature around 593 cm−1 was identified in solid argon.

Absorptions at 1894 cm−1 (argon matrix) and 1880 cm−1 (carbon dioxide matrix) were tentatively assigned as a Fermi resonance of the 2045 cm−1 band with an overtone of the 972 cm−1 fundamental.

Jacox et al. conducted also a normal coordinate analysis and suggested a C2v bridged structure (Fig. 1(1)); in strong contrast, LaBonville et al. allocated a Cs symmetric structure of the carbon trioxide molecule (Fig. 1(2)).4

The interest in the carbon trioxide molecule has been also fueled by the complex reaction mechanisms of carbon oxides (carbon monoxide and carbon dioxide) with atomic oxygen in the Martian atmosphere.5–8

Carbon dioxide, CO2(X 1Σ+g), presents the major constituent (95.3% by volume); nitrogen (2.7%), argon (1.6%), carbon monoxide (0.7%), molecular oxygen (0.13%), water (150–200 ppm), and ozone (0.03 ppm) make up the rest.9

It has been suggested that the photodissciation of carbon dioxide by solar photons (λ < 2050 Å) produces carbon monoxide and atomic oxygen.

Near the threshold, only ground state O(3P) atoms are produced; shorter wavelengths supply also O(1D).5

The primary fate of electronically excited oxygen atoms is thought to be quenching to form O(3P); the detailed process is not known and has been postulated to proceed via a carbon trioxide molecule.

Once O(3P) and carbon monoxide has been formed, it is difficult to restore carbon dioxide, since the reversed reaction is spin forbidden.

Detailed photochemical models suggest that the oxygen atoms rather react to molecular oxygen and ultimately to ozone.5

The CO3 molecule has been also implied as an important intermediate in the 18O isotope enrichment of carbon dioxide in the atmospheres of Earth and Mars.10,11

Computations and laboratory experiments indicate that the 18O enrichment in ozone might be transferable to carbon dioxide,12,13 possibly via a CO3 intermediate.

On Earth, photolysis of stratospheric ozone generates O(1D), which in turn might react with carbon dioxide to form a carbon trioxide molecule.

In the gas phase, the latter was postulated to fragment to carbon dioxide and atomic oxygen, possibly inducing an isotopic enrichment in carbon dioxide via isotopic scrambling.14

However, the explicit structure of the CO3 intermediate has not been unraveled yet.

Various kinetic measurements have been carried out to determine the temperature-dependent rate constants of the reaction of electronically excited oxygen atoms, O(1D), with carbon dioxide.

At room temperature, rate constants of a few 10−10 cm3 s−1 have been derived.15–17

This order of magnitude suggest that the reaction has no or only little activation energy, proceeds with almost unit efficiency, and most likely involves a reaction intermediate.

However, neither reaction products nor the nature of the intermediate were determined.

On the other hand, a CO(X 1Σ+) + O2(X 3Σ−g) exit channel was found to have an activation energy between 15 and 28 kJ mol−1 in the range of 300–2500 K.18

This finding correlates also with a theoretical investigation of the singlet and triplet potential energy surfaces of the CO3 system.19

Froese and Goddard suggested that the barrier-less, spin-forbidden quenching pathway to form ground state oxygen atoms and carbon dioxide (ΔRG = −190.0 kJ mol−1) dominates over the formation of CO(X 1Σ+) plus O2(X 3Σ−g) (ΔRG = −157.5 kJ mol−1) and CO(X 1Σ+) plus O2(a 1Δg) (ΔRG = −63.3 kJ mol−1).

Further theoretical calculations indicated that the CO3 isomer identified in the carbon dioxide and argon matrices might be the C2v symmetric bridged molecule.

A D3h structure was identified as a local minimum, too, but lies 16.8 kJ mol−1 higher in energy than the cyclic isomer (Fig. 1, (3)); according to calculations, both structures are connected via a transition state located 36 kJ mol−1 above the cyclic molecule.20

At temperatures higher than 100 K, the cyclic CO3 isomer was predicted to decay to carbon dioxide and ground state oxygen atoms via singlet-triplet transitions.

Despite this information on the reaction of carbon dioxide with atomic oxygen, an incorporation of these data into homogeneous gas phase models still fails to reproduce the observed abundances of carbon dioxide, carbon monoxide, oxygen, and ozone in the Martian atmosphere quantitatively.9

Atreya et al. pointed out the necessity to include heterogeneous reactions on aerosols or carbon dioxide ice particles in the Martian air.21–23

However, these processes have not been investigated in the laboratory so far.

Also, the explicit structure and the actual formation mechanism of the CO3 isomer and its role in the 18O isotopic enrichment in stratospheric carbon dioxide remain to be solved.

In this paper, we present a detailed experimental and theoretical investigation on the formation mechanism of carbon trioxide in low temperature carbon dioxide ices and the implications for gas phase chemistry.

Reactive oxygen atoms are generated via electronic energy loss of high energy electrons to the carbon dioxide molecule in the solid sample.

Our first goal is to identify the infrared absorption features of the carbon trioxide molecule, to resolve the true nature of the 1880 cm−1 absorption of the carbon trioxide molecule unambiguously, and to assign the structure of the newly formed species.

Secondly, reaction mechanisms to synthesize the carbon trioxide molecule together with other newly formed species will be derived combining our experimental data with electronic structure calculations.

Finally, important implications of these results to planetary and atmospheric chemistry are addressed.

Note that these studies also have important implications to planetary, cometary, and interstellar chemistry since carbon dioxide has been identified as a major component of ices on Mars, in comets such as Halley,24 and of low temperature grain mantles in cold molecular clouds like the Taurus Molecular Cloud (TMC-1)25.

Theoretical calculations

The geometries of various local minima and transition states on three potential energy surfaces (PESs) of carbon trioxide, CO3, including the lowest triplet and two lowest singlet electronic states, have been optimized using the multireference complete active space self-consistent field (CASSCF) method26,27 with the 6-311G(d) basis set.

The active space in the CASSCF calculations included 16 electrons distributed over 13 orbitals, i.e., this was the full-valence active space excluding 2s lone pairs on three oxygen atoms.

Vibrational frequencies and infrared (IR) intensities have been also computed at the CASSCF(16,13)/6-311G(d) level of theory.

Single-point energies for various species have been subsequently refined employing internally-contracted multireference configuration interaction MRCI method28,29 with the same (16,13) active space and the larger 6-311+G(3df) basis set.

All calculations were carried out using the MOLPRO 200230 and DALTON31 programs.


The experiments were carried out in a contamination-free ultrahigh vacuum (UHV) chamber; the top view of this machine is shown in Fig. 2.

This setup consists of a 15 l cylindrical stainless steel chamber of 250 mm diameter and 300 mm height which can be evacuated down to 2 × 10−10 torr by a magnetically suspended turbopump backed by an oil-free scroll pump.

A two stage closed cycle helium refrigerator-interfaced to a differentially pumped rotary feedthrough is attached to the lid of the machine and holds a polished silver (111) single crystal.

This crystal is cooled to 10.4 ± 0.3 K, serves as a substrate for the ice condensate, and conducts the heat generated from the impinging electrons to the cold head.

To minimize the radiative heat transfer from the chamber walls to the target, a 40 K aluminum radiation shield is connected to the second stage of the cold head and surrounds the crystal.

The ice condensation is assisted by a precision leak valve.

During the actual gas condensation, the deposition system can be moved 5 mm in front of the silver target.

This setup guarantees a reproducible thickness and composition of the frosts.

To allow a selection of the target temperature, a temperature sensor, cartridge heater, and a programmable controller are interfaced to the target.

The carbon dioxide ices were prepared at 10 K by depositing carbon dioxide gas onto the cooled silver crystal.

Blank checks of the pure gas (BOC Gases, 99.999%) via a quadrupole mass spectrometer and of the frosts via a Fourier transform infrared spectrometer were also carried out.

Fig. 3 depicts a typical infrared spectrum of the frost; the absorptions are compiled in Table 1.

To determine the ice composition quantitatively, we integrated numerous absorption features and calculated the column density, i.e. the numbers of absorbing molecules per cm2, n, via the Lambert–Beer relationship (1) and eqns. (2)–(3).

The integrated absorption features, the corresponding integral absorption coefficients, and the column densities are summarized in Table 2.

These data suggest a column density of (1.1 ± 0.3) × 1018 molecules cm−2.

Considering a density of 1.7 g cm−3 at 10 K,32 this translates into an averaged target thickness of 0.48 ± 0.11 μm.

We would like to stress that the integrated absorption coefficients have been taken in transmission experiments,47 but the experiment has been carried out in an absorption–reflection–absorption mode.

This probably causes the large variations in the film thicknesses estimated from different absorption features.

I() = I0()eε()nwith the intensity of the IR beam after, I(), and before absorption, I0(), at a wavenumber , the wavenumber dependent absorption coefficient ε() in units of cm−2 and the number of absorbing species per cm2, n.

Reformulating eqn. (1) with A() = lg(I0()/I()) gives A() = ε()n/ln 10.Integrating from 1 to 2 yields with the integrated absorption 12A()d in cm−1 and the integral absorption coefficient Aexp = 12ε()d in cm.

The factor cos(75°) accounts for angle between the surface normal of the silver wafer and the infrared beam, whereas the division by two corrects for the ingoing and outgoing IR beams.

These ices were irradiated isothermally at 10 K with electrons of 5 keV kinetic energy generated in an electron gun at beam currents of 100 nA (60 min) by scanning the electron beam over an area of 3.0 ± 0.4 cm2.

Accounting for the extraction efficiency of 78.8% and the irradiation time, this exposes the target to 1.77 × 1015 electrons.

Higher beam currents, which increase the temperature of the frost surface, should be avoided.

After the actual irradiation, the sample was kept isothermally at 10 K and heated then by 0.5 K min−1 to 293 K.

To guarantee an identification of the reaction products in the ices and those subliming into the gas phase on line and in situ, two detection schemes are incorporated: (i) a Fourier transform infrared spectrometer (FTIR), and (ii) a quadrupole mass spectrometer (QMS).

The chemical modification of the ice targets is monitored during the experiments to extract time-dependent concentration profiles and hence production rates of newly formed molecules and radicals in the solid state.

The latter is sampled via a Nicolet 510 DX FTIR spectrometer (6000–500 cm−1) operating in an absorption–reflection–absorption mode (reflection angle α = 75°; Fig. 2); spectra were accumulated for 2.5 min at a resolution of 2 cm−1.

The infrared beam is coupled via a mirror flipper outside the spectrometer, passes through a differentially pumped potassium bromide (KBr) window, is attenuated in the ice sample prior and after reflection at a polished silver waver, and exits the main chamber through a second differentially pumped KBr window before being monitored via a liquid nitrogen cooled detector (MCTB).

The gas phase is monitored by a quadrupole mass spectrometer (Balzer QMG 420) with electron impact ionization at 90 eV electron energy of the neutral molecules in the residual gas analyzer mode.

The raw data, i.e. the temporal development of the ion currents of distinct mass-to-charge ratios, are processed via matrix interval algebra to compute absolute partial pressures of the gas phase molecules.33

Since, for example, carbon dioxide can fragment to molecular oxygen and also to carbon monoxide in the ionizer of the quadrupole mass spectrometer, different molecular species add to one mass to charge ratio (m/z) of, e.g. 32 (O2).

Therefore, we must perform the raw data processing via matrix interval algebra to calculate the actual partial pressures of the molecules in the gas phase.

Briefly, m/z ratios are chosen to result in an inhomogeneous system of linear equations including the measured ion current (right hand vector), partial pressures (unknown quantity), and calibration factors of fragments of individual gaseous species determined in separate experiments.

Since all quantities are provided with experimental errors, matrix interval arithmetic, i.e. an IBM high accuracy arithmetic subroutine defining experimental uncertainties as intervals, is incorporated in the computations to extract individual, calibrated components of gas mixtures.


Computational results

Our calculations suggest that two minima exist on the lowest singlet CO3 potential energy surface (PES): a C2v-symmetric three-member cyclic structure s1 and a D3h-symmetric isomer s2 (Fig. 4). s1 and s2 have similar energies and reside 197.5 and 197.1 kJ mol−1 lower than the O(1D) + CO2(X 1Σ+g) asymptote, respectively.

These two singlet isomers can rearrange to each other by ring opening/ring closure and are separated by a low barrier of 18.4 kJ mol−1 with respect to s1 occurring at transition state s-TS1.

The cyclic structure s1 can be produced in a reaction between O(1D) and CO2(X 1Σ+g).

The calculations suggest that the reactants first form a weakly bound complex (∼2.9 kJ mol−1) s5, which then rearranges to s1 with a barrier vias-TS2 of 5.6–7.6 kJ mol−1 relative to O(1D) + CO2.

In the triplet electronic state, separated O(3P) and CO2 have the lowest energy, while the C2v isomer t1 (3B2) resides 96.3 kJ mol−1 higher. t1 has a structure rather similar to that of s2, except that the three C–O bond lengths are not equal; there are one double (1.201 Å) and two single (1.343 Å) bonds.

Isomer t1 can decompose to O(3P) + CO2 overcoming a barrier 51.5 kJ mol−1 at transition state t-TS1.

In the reverse direction, the barrier for the O(3P) + CO2(X 1Σ+g) → t1 reaction is calculated to be as high as 147.7 kJ mol−1.

Interestingly, we were able to locate a minimal energy crossing point (MSX) between the lowest triplet and singlet electronic states in a close vicinity of t-TS1, both the geometry and energy of MSX are similar to those for the triplet transition state.

Another possible isomer of triplet CO3, Cs-symmetric OCOO t2 (3A″) is much less favorable and lies 260.7 kJ mol−1 higher in energy than O(3P) + CO2(X 1Σ+g).

In addition, we found two local minima on PES of the first excited singlet electronic state, which have lower energies than O(1D) + CO2(X 1Σ+g).

For instance, s3 (C2v, 1A2) is a complex of O(1D) with carbon dioxide and stabilized by 27.6 kJ mol−1 relative to the separated species.

The structure of s4 (Cs, 1A″) is similar to that of s2, however, all three C–O bonds and OCO angles are slightly unequal.

Transition state s-TS2 separating s3 and s4 lies 46.4 and 159.8 kJ mol−1 above O(1D) + CO2(X 1Σ+g) and s4, respectively.

Table 3 summarizes the infrared absorptions of s1, s2, and t2.

Infrared spectroscopy

The FTIR spectra are analyzed in three steps.

First, we investigate the new absorptions qualitatively and assign their carriers.

Hereafter, the temporal developments of these absorptions upon electron irradiation are investigated quantitatively as outlined in Section 3.

Finally, these data are fitted to calculate production rates of synthesized molecules in units of molecules cm−2 (column density), molecules per impinging electron, and absorbed electron volt (eV) per target molecule (dose).

The integration routine of the absorption features is accurate to ±10%.34

Qualitative analysis

The effects of the electron irradiation of the carbon dioxide target are displayed in Figs. 5–11.

A comparison of the pristine sample (Fig. 3) with the irradiated ice at 10 K clearly shows new absorption features of carbon monoxide at 2139 cm−1 (ν1(CO stretching); Fig. 5; Table 4)) and the corresponding isotopic pattern of 13CO at 2092 cm−1.

These data are in close agreement to matrix isolation studies of the carbon monoxide molecule.35

We were also able to detect four fundamentals of the C2v symmetric, cyclic CO3 structure at 2045 cm−1 (ν1), 1068 cm−1 (ν2), 973 cm−1 (ν5), and 565 cm−1 (ν6) (Fig. 6).

The position of all peaks is in excellent agreement with earlier matrix isolation studies (Section 1) and with our calculated, scaled frequencies (Table 3).

Note that although the unobserved ν3 and ν4 modes have larger absorption coefficients than the detected ν6 transition, the ν4 absorption overlaps with the broad ν2 band of the carbon dioxide reactant; the ν3 band of carbon trioxide is too close to the cut-off of the MCTB detector to be observable.

Finally, we detected also a transition at 1879 cm−1, which was assigned tentatively as a Fermi resonance of the 2044 cm−1 band with an overtone of the 973 cm−1 fundamental.

The calculated symmetry of the carbon trioxide modes (Table 3) confirm this tentative assignment.

Since the ν5 at 973 cm−1 has b1 symmetry, the overtone (2ν5) holds an a1 symmetry (b1 ⊗ b1 = a1), the latter has the same symmetry as the ν1 fundamental; this can give rise to the Fermi resonance as observed at 1879 cm−1.

These data make it exceptionally clear that the observed carbon trioxide molecule has a cyclic, C2v symmetric structure.

Neither the Cs nor the D3h symmetric structures of carbon trioxide were observed.

Besides the carbon monoxide and the carbon trioxide molecules, we were also able to identify the ozone molecule (Fig. 7).

Two absorptions at 1042 cm−1 (ν3, anti symmetric stretch) and the weaker bending mode at 704 cm−1 (ν2) were identified.

These data agree very well with previous assignments36.

Quantitative analysis

Figs. 8–11 compile the temporal development of the column densities of the carbon dioxide reactant and of the products (carbon monoxide, carbon trioxide, and ozone) during the irradiation at 10 K, the consecutive equilibration period at 10 K, and the heating phase.

During the irradiation of the carbon dioxide ice, the column density of the CO2 molecules decreases only slightly from 1.188 × 1018 cm−2 to 1.158 × 1018 cm−2 (Fig. 8); note that these data are afflicted with an error of ±25% (Table 2).

This means that only 3.01 × 1016 cm−2, i.e. 2.5%, of the carbon dioxide molecules are destroyed at the end of the irradiation (Table 5).

Accounting for the target surface and the electron beam current, we can conclude that each implanted electron destroys 17 ± 4 CO2 cm−2, i.e. 51 ± 13 carbon dioxide molecules.

As expected, the carbon dioxide column density stays constant during the isothermal phase.

With increasing temperature the CO2 molecules sublime; as the temperature is raised from 10 K to 20 K, strengths of all carbon dioxide absorptions start to diminish; at 94 K, no solid carbon dioxide is left on the silver waver.

With decreasing carbon dioxide column density, new absorptions arise from carbon monoxide (Fig. 9) and carbon trioxide (Fig. 10).

The carbon monoxide column density rises almost linearly with increasing irradiation time to 3.5 ± 0.4 × 1016 cm−2, i.e. an average production rate of 60 ± 6 carbon monoxide molecules per implant, i.e. 20 ± 2 CO cm−2; the integral absorption coefficient for the 2139 cm−1 band of 1.1 × 10−17 molecules cm−1 is accurate to ±10%.

On the other hand, the carbon trioxide column density rises quickly but starts to saturate toward the end of the experiment at 1.8 × 1015 cm−2 (2044 cm−1), 1.5 × 1015 cm−2 (1067 cm−1), and 1.1 × 1015 cm−2 (972 cm−1).

These data translate to a synthesis of 2.5 ± 0.5 carbon trioxide molecules per electron (0.8 ± 0.2 CO3 cm−2 per electron; production rates are averaged over those obtained from three CO3 fundamentals).

Based on these information, we can now investigate the carbon balance of the target.

Based on our integration routine, 3.0 ± 0.8 × 1016 CO2 cm−2 lead to the formation of 3.5 ± 0.4 × 1016 CO cm−2 and 1.5 ± 0.3 × 1015 CO3 cm−2, i.e. destruction of 3.0 ± 0.8 × 1016 cm−2versus formation of 3.7 ± 0.4 × 1016 cm−2; within the error limits, we can conclude that the carbon budget is conserved in the experiment.

This strongly correlates with our experimental findings that carbon dioxide and carbon trioxide are the only newly formed carbon-bearing species in our experiment.

Note that whereas the carbon monoxide column density stays constant during the isothermal phase at 10 K, the carbon trioxide column density seems to decrease slightly by 5%.

However, since the integration is accurate only to 10%, the drop of the 2044 cm−1 absorption might be within the experimental error limits: similar the increasing carbon monoxide column density upon warming the matrix to 20 K; alternatively we might conclude that the carbon trioxide molecule starts to decompose even at 10 K. Upon heating the target, both the carbon monoxide and carbon trioxide column densities decrease.

Note, however, that whereas the carbon dioxide absorptions disappear at 94 K, no carbon monoxide and carbon trioxide bands were observed at temperatures higher than 91 K. At 91 K, a column density of 5.6 ± 1.2 × 1017 CO2 cm−2 remains (0.21 ± 0.04 μm CO2 ice).

Since the carbon dioxide matrix sublimes in layers, those layers exposed to the vacuum sublime first.

Considering an initial column density of 1.188 × 1018 cm−2 CO2 cm−2 (0.48 ± 0.11 μm CO2 ice), we can conclude that the newly synthesized molecules are formed within the first 0.28 ± 0.09 μm of the sample, i.e. those layers which are subliming first into the vacuum.

Once these layers have been released, the remaining carbon dioxide ice of 0.20 ± 0.04 μm does not contain any newly formed molecules.

This in turn indicates that the 5 keV electrons are absorbed and induce radiation damage within the first 0.28 ± 0.09 μm of the carbon dioxide sample.

Compared to the carbon oxides, the temporal development of ozone depicts striking differences (Fig. 11).

The ozone column density increases after 60 min irradiation time to 9.3 ± 1.3 × 1015 cm−2; data has been calculated with an integral absorption coefficient of 1.4 ± 0.2 × 10−17 molecules cm−1.

This requires a destruction of 2.8 ± 0.4 × 1016 CO2 cm−2.

Statistically, each electron generates 15 ± 3 O3 molecules in the sample.

Since the formation of a single ozone molecule requires the destruction of three carbon dioxide molecules to initiate three oxygen atoms, 45 ± 9 carbon dioxide molecules have to be destroyed to account for the experimentally derived ozone production rate.

To account for the total oxygen balance after the irradiation, we have to include the newly synthesized carbon trioxide molecules (1.5 ± 0.3 × 1015 cm−2), too.

Hence, the total oxygen column density of the freshly formed molecules calculates as the sum of the carbon trioxide column density plus three times the ozone column density; based on these considerations, a column density of 2.9 ± 0.4 × 1016 cm−2 has to be generated in the carbon dioxide ice by the electrons.

On the other hand, the decay profile of the carbon dioxide absorptions suggest that 3.0 ± 0.8 × 1016 CO2 cm−2 have been destroyed after the irradiation.

Since each carbon dioxide molecule fragments to carbon monoxide and atomic oxygen upon interaction with an electron, we would expect an identical number of oxygen atoms to be incorporated inside the newly formed molecules, i.e. ozone and carbon trioxide.

Within the error limits, the oxygen balance seems to hold.

During the isothermal phase, the ozone column density slightly increases from 9.3 ± 1.3 × 1015 cm−2 to 1.0 ± 0.1 × 1016 cm−2 during the 10 K equilibration period; this suggest that oxygen atoms, which are mobile at 10 K, are present in the carbon dioxide matrix and may react with molecular oxygen to form additional ozone.

We have to keep in mind that this analysis only comprises the oxygen balance of the infrared active molecules, but not of infrared inactive species such as molecular and atomic oxygen.

In the solid state, the O2 absorptions at 1591 cm−1 and 1617 cm−1 hold absorption coefficients of about 10−21 cm molecule−1 (ref. 37).

It is not surprising that we were unable to detect these transitions in our experiments.

Therefore, we can conclude that molecular oxygen and oxygen atoms reside inside the carbon dioxide matrix as well.

This effect is even more pronounced while the sample is heated to 60 K. Here, the ozone column density rises significantly by about 30% reaching a maximum at 1.3 ± 0.2 × 1016 cm−2 before the column density drops sharply due to the subliming carbon dioxide matrix.

This suggests that at least 1.1 ± 0.3 × 1016 cm−2 of the oxygen atoms exist in the form of molecular or atomic oxygen.

Including the enhanced ozone column density in the oxygen balance gives a column density of generated oxygen atoms of 4.0 ± 0.6 × 1016 cm−2versus destruction of the carbon dioxide molecules of 3.0 ± 0.8 × 1016 cm−2.

At 92 K, the ozone absorption disappears completely.

It is interesting to compare this temperature with those where the bands of carbon monoxide 91 K, carbon trioxide 91 K, and carbon dioxide 94 K vanish.

As indicated in the previous section, carbon monoxide, carbon trioxide, and the oxygen atoms are formed initially in the first 0.28 ± 0.09 μm of the carbon dioxide sample.

As the temperature rises, oxygen atoms diffuse and could penetrate also those regions of the carbon dioxide ice which has not been penetrated by the electrons; here, the oxygen atoms could recombine to ozone.

This could explain the presence of ozone absorptions at the temperature of 92 K where absorptions of carbon monoxide and carbon trioxide are absent due to their sublimation with the outer layers of the carbon dioxide matrix into the vacuum.

Mass spectrometry

It is interesting to correlate the infrared observations with a mass spectrometric analysis of the gas phase.

During the annealing phase of the sample, the carbon dioxide matrix and the newly formed molecules (CO3, CO, O3) sublime into the gas phase.

Upon heating the sample to 25 K (180 min experimental time), the carbon dioxide matrix start to sublime (Fig. 8); similarly, the embedded carbon monoxide molecules are being released into the gas phase (Fig. 9).

Note that the ozone column density increases upon warming the sample due to reactions of atomic oxygen with molecular oxygen (section 4.2); at 60 K (240 min experimental time), the ozone column density decreases, too.

However, the mass spectrometric results do not correlate entirely with the infrared data.

Although the partial pressure of carbon monoxide begins to increase at 25 K (as expected from the infrared data), the temporal development of the partial pressure of ozone shows two distinct peaks: a small hump starting at 240 min experimental time (60 K), and a second intense peak at 291 min experimental time (Fig. 12).

A similar pattern has been observed for the carbon dioxide molecule, too.

To interpret the discrepancy between the infrared and mass spectrometric data, we have to keep in mind that the silver target (first stage) is annealed while the cold head is still in operation; this means that the outer aluminum cold shield, which is mounted to the second stage of the closed cycle helium refrigerator, is still cooled down.

Therefore, the molecules subliming during the annealing phase of the silver target (first ozone peak) can actually re-condense onto the aluminum cold shield.

Once the heat load from the cartridge heater increase also the temperature of the second cold head stage, those molecules condensed on the aluminum shield can sublime, too (second ozone peak).

Note that two peaks were observed only for ozone, carbon dioxide, and oxygen, but not for carbon monoxide.

Here, carbon monoxide does not re-condense at the 50 K aluminum cold shield since carbon monoxide ice is unstable at temperatures higher than 30 K.


Our investigations indicate that the response of the carbon dioxide system upon the keV electron bombardment is governed by an initial formation of carbon monoxide and atomic oxygen, eqns. (3)–(4).

Depending on the energy transfer form the impinging electron to the carbon dioxide molecule, the oxygen atom can be generated either in its electronic ground (3P) via inter system crossing to the triplet manifold and/or excited state (1D) on the singlet surface.

CO2(X 1Σ+g) → CO(X 1Σ+) + O(3P),CO2(X 1Σ+g) → CO(X 1Σ+) + O(1D).

These reactions are endoergic by 532 kJ mol−1 (5.51 eV) and 732 kJ mol−1 (7.59 eV), respectively.

Our experiments indicate that each electron initiates 60 ± 6 CO molecules within the carbon dioxide ice (Table 5); this translates to 331 ± 33 eV and 455 ± 46 eV to form O(3P) and O(1D), respectively.

Considering the energy of the electron of 5 keV, about 10% of its kinetic energy is utilized to cleave the carbon–oxygen bonds of the carbon dioxide molecules; note that this calculation assumes all the carbon monoxide molecules are formed in their vibrational ground states; also, the oxygen atoms have no excess translational energy.

However, to escape the matrix cage, the oxygen atom must have at least 0.5 eV excess kinetic energy; if its kinetic energy is less than the lattice bonding energy, at least O(1D) can react back without an entrance barrier to recycle the carbon dioxide molecule.

To fit the experimentally obtained profile of the carbon dioxide and carbon monoxide column densities, we employed the following model.

The carbon monoxide molecule was assumed to ‘decay’ first order upon electron bombardment similar to a radioactive decay, i.e. it follows the velocity law (5) (the square brackets indicate the column density in cm−2; Ie = 4.92 × 1011 s−1 presents the electron current in electrons s−1): −d[CO2]/dt = k1Ie[CO2] = k′1[CO2],d[CO]/dt = k2Ie[CO2] = k′2[CO2].This translates to the following temporal evolutions of the column density for carbon dioxide (7) and carbon monoxide (8) with the constant a: [CO2](t) = [CO2](t = 0)ek′1t,[CO](t) = a(1 − ek2t).The best fits of the carbon dioxide and carbon monoxide profiles are shown in Figs. 13 and 14, respectively, with [CO2](t = 0) = 1.19 ± 0.30 × 1018 cm−2, k′1 = 7.35 ± 0.20 × 10−6 s−1, a = 7.1 ± 0.7 × 1016 cm−2, and k′2 = 1.93 ± 0.1 × 10−4 s−1.

Accounting for the electron current, this yields k1 = 1.5 ± 0.2 × 10−17 and k2 = 3.9 ± 0.2 × 10−16.

We now investigate the fate of the generated oxygen atoms quantitatively.

Our data indicate that the carbon trioxide molecule is formed via reaction of atomic oxygen with carbon dioxide.

We were able to fit the temporal development of the column density assuming that a carbon dioxide dimer decomposed to carbon dioxide, atomic oxygen, and carbon monoxide; within the matrix cage, the generated oxygen atom reacts with the non-reacted carbon dioxide to form carbon trioxide via eqns. (9) and (10): −d[(CO2)2]/dt = k3Ie[(CO2)2] = k′3[(CO2)2],d[CO]/dt = k4Ie[(CO2)2] = k′4[(CO2)2].This leads to the temporal evolution of the column density for carbon trioxide via eqn. (11):[CO3](t) = b(1 − ek′4t).Fig. 15 depicts the best fit of the carbon trioxide profile respectively, with b = 1.5 ± 0.3 × 1015 cm−2, and k′4 = 1.1 ± 0.1 × 10−3 s−1.

Accounting for the electron current, this yields k4 = 2.2 ± 0.2 × 10−15.

Extrapolating eqns. (8) and (11) for t → ∞ and extracting the ratio of [CO](t = ∞)/[CO3](t = ∞) calculates the fraction of released oxygen atoms reacting with carbon dioxide to 47 ± 14.

This means that only 2.1 ± 0.5% of the generated oxygen atoms react to carbon trioxide.

What might be the reason for the low fraction of oxygen atoms reacting to carbon dioxide?

The potential energy surface (Fig. 4) helps to understand this scenario.

First, the experiments clearly indicated the formation of the cyclic s1 isomer; neither s2 nor t2 have been detected.

Our calculations show that in the gas phase, only the addition of O(1D) can form s1vias5 and s-TS3.

This is a clear indication that the interaction of energetic electrons with the carbon dioxide molecules generates reactive oxygen atoms.

However, this reaction has to pass s-TS3 which is located 5.6–7.6 kJ mol−1 above the separated reactants.

This requires that the O(1D) reactant has at least 5.6–7.6 kJ mol−1 excess kinetic energy to overcome the barrier.

Once s1 has been formed in the solid state, the surrounding matrix can divert the internal energy (197.5 kJ mol−1) of the carbon trioxide species; this stabilizes the latter and prevents an isomerization vias-TS1 to s2.

Non-reactive O(1D) can be quenched in the matrix easily via intersystem crossing to the 3P ground state.38

On the triplet surface, ground state oxygen atoms can react solely to yield t1; followed by intersystem crossing, the latter can form s2 in the carbon dioxide matrix.

However, since only the cyclic carbon trioxide isomer was detected, this pathway can be likely ruled out.

Similarly, the electronically excited 1A″ surface can be excluded to contribute to the formation of the cyclic carbon trioxide isomer since only s4 can be formed vias3 and s-TS2.

Instead, we located the seam of crossing (MSX) which connects the triplet to the singlet surface.

In the gas phase, this crossing is located close to t-TS1, and O(3P) with a sufficient amount of kinetic energy can surpass this barrier to form also s1.

In the solid matrix, however, the energy of MSX is likely lowered; the exact barrier height is currently under investigation.

Actually, the calculated equilibrium constants for the isomerization s2 ↔ s1 correlate with the failed observation of the D3h symmetric CO3 isomer in the ice matrix.

Here, K(10 K) = 0.002176482 means that at 10 K the concentration of s1 should be 500 times higher than that of s2, if they are in equilibrium.

The spectroscopic data suggest that the remaining oxygen atoms rather react via molecular oxygen to form ozone via eqns. (12)–(13).

Note that only ground state reactants have been considered; this is certainly true for the reaction during the equilibrating phase at 10 K and the annealing program to 60 K. However, during the electron bombardment, O(1D) atoms could react, too.

Both reactions are exoergic by 498.5 kJ mol−1 and 106.5 kJ mol−1, respectively and involve no entrance barrier except the diffusion energy of the oxygen atoms to migrate to the reaction site.

The ability of the oxygen atoms to diffuse even at 10 K and in particular at elevated temperatures has been established previously (section 4.2. and Fig. 11).

The detailed formation mechanism of ozone via eqn. (13) and/or electronically excited species is currently under investigation and might involve a short-lived cyclic ozone molecule.

So far, we were not able to fit the temporal evolution of the ozone column density; this fit requires knowledge of the diffusion coefficient of the oxygen atom which is currently under study.

A simple exponential fit fails as expected since the ozone is clearly a higher-order reaction product.

O(3P) + O(3P) → O2(X 3Σ−g),O2(X 3Σ−g) + O(3P) → O3(X 1A1).

Finally, we would like to comment on the possibility to generate reactive carbon atoms via the interaction of energetic electrons with carbon dioxide.

Recall that the slight increase of the carbon monoxide column density in the equilibration phase (Fig. 9) could indicate a recombination of diffusive carbon atoms with mobile oxygen atoms, eqn. (14), a process similar to the formation of molecular oxygen according to eqn. (12).

C(3P) + O(3P) → CO(X 1Σ+).Detailed electronic structure calculations depicted that a release of carbon atoms from a linear carbon dioxide molecule does not occur.39

Instead, the linear carbon dioxide species has to isomerize first to a cyclic structure which lies 582 kJ mol−1 higher in energy than the linear structure.

The cyclic isomer ring opens and forms a linear COO(X 1Σ+) molecule which then loses a carbon atom in its excited 1D state.

However, our experiments identify neither the cyclic carbon dioxide nor the linear COO(X 1Σ+) molecule as a reactive intermediate in our matrix.

Since the carbon budget is conserved in our experiment (section 4.2.2) we suggest that the contribution of reactive carbon atoms, if any, is only minor.

Astrophysical and planetary implications

Carbon dioxide ice presents also a major constituent of ices as condensed on sub-micrometre sized grain particles in cold interstellar clouds.

Although the dust component embodies only 1% of the interstellar matter, these nuclei play a key role in the formation of new molecules.

Deep in the interior of dense clouds, grain particles effectively shield newly synthesized molecules in the gas phase from the destructive external UV radiation field.

Most important, these sub micrometer sized particles present valuable nurseries to synthesize new molecules.

In dense clouds, these grains have typical temperatures of 10 K.40,41

Once molecules, radicals, or atoms from the gas phase collide with the solid particle, they are accreted on the grain surface resulting in an icy mantle up to 0.1 μm thick.

At ultralow temperatures, all species except H, H2, and He hold sticking coefficients of unity.

This means that each collision of a gas phase species with a cold surface leads to an absorption and hence thickening of the ice layer.

Here, solid mixtures containing H2O (100), CO (7–27), CH3OH (<3.4), NH3 (<6), CH4 (<2), and CO2 (15) were identified unambiguously via infrared spectroscopy towards the dense cloud TMC-1 employing the field star Elias 16 as a black body source;42 the numbers in parentheses indicate the relative abundances compared to water ice.

These molecular clouds are interspersed with ultraviolet photons (<13.2 eV) and energetic particles from T-Tauri winds, stellar jets, carbon stars, and galactic cosmic ray particles.

Therefore, pristine ice mantles are processed chemically by the cosmic ray induced internal ultraviolet radiation present even in the deep interior of dense clouds (ϕ = 103 photons cm−2 s−1) and in particular through particles of the galactic cosmic radiation field.

This can lead to the formation of new molecules in the solid state via non-equilibrium (non-thermal) chemistry even at temperatures as low as 10 K. The particle component of the cosmic ray radiation field consists of 97–98% protons (p, H+) and 2–3% helium nuclei (α-particles, He2+) in the low energy range of 1–10 MeV (1 MeV = 106 eV) with ϕ = 10 particles cm−2 s−1.43

It has been well established that cosmic ray MeV particles penetrate the ice mantles and the grain core and deposit parts of their energy inside the ices via interaction of the MeV implant and the electronic system of the ice molecules.44

Detailed dynamics simulation showed that 99.9% of the transferred energy leads predominantly to ionization and hence release of energetic (keV) electrons perpendicularly to the trajectory of the MeV implant (ultra track).

Therefore, our experiments simulate also the energetic processes in the ultra track of MeV particles inside interstellar and also cometary ices.

Based on these considerations, we can conclude that the cyclic carbon trioxide molecule should also be present in carbon-dioxide-rich extraterrestrial ices which have been identified in interstellar clouds, comets, and also on Mars.

Most importantly, our experiments indicated the initial formation of suprathermal (electronically excited) oxygen atoms.

In extraterrestrial ices, the species do not react solely with the carbon dioxide ice, but with the remaining ice components.

In particular the O(1D) atoms are very reactive and may destroy newly formed, astrobiologically important molecules such as sugars and amino acids even deep inside ices.

Potential organics inside the Martian soil might be degraded easily by energetic oxygen atoms.

Also, the interaction of carbon dioxide ices on Mars (and in comets as well as in the interstellar medium) with MeV particles produces significant amounts of ozone.

Upon warming the ices, molecules would sublime and could contribute considerably to the ozone budget in the Martian atmosphere.

Our investigations help to understand a potential 18O enrichment in the terrestrial and Martian atmospheres.

Whereas the cyclic carbon trioxide isomer is being stabilized upon reaction of electronically excited oxygen atoms with carbon dioxide in the matrix via phonon interaction, in the gas phase the cyclic carbon trioxide molecule can isomerizes to the D3h isomer.

This could scramble the incorporated 18O and hence lead to an enrichment of 18O within the carbon dioxide once the carbon trioxide molecule decomposes too atomic oxygen and carbon dioxide (Fig. 16).


We investigated the synthetic routes to form the cyclic carbon trioxide isomer, CO3(X 1A1), in carbon-dioxide-rich extraterrestrial ices and in the atmospheres of Earth and Mars experimentally and theoretically.

The studies indicate that the interaction of an electron with a carbon dioxide molecule is dictated by a carbon–oxygen bond cleavage to form ground state (3P) and/or electronically excited (1D) oxygen atoms plus a carbon monoxide molecule.

About 2% of the oxygen atoms react with carbon dioxide molecules to form the C2v symmetric, cyclic CO3 structure, via addition to the carbon–oxygen double bond of the carbon dioxide species; neither the Cs nor the D3h symmetric isomers of carbon trioxide were detected.

Since the addition of O(1D) involves a barrier of 5.6–7.6 kJ mol−1 and the reaction of O(3P) with carbon dioxide to form the carbon trioxide molecule via triplet-singlet intersystem crossing is endoergic by 2 kJ mol−1, the atomic oxygen reactant(s) must have also access kinetic energy.

The remaining oxygen atoms react barrier-less to form ozone via molecular oxygen.

In our matrix, CO3(X 1A1) is being stabilized by phonon interactions to the surrounding matrix.

In the gas phase, however, the initially formed C2v structure can ring-open to the D3h isomer which in turn loses an oxygen atom to ‘recycle’ carbon dioxide.

The atomic oxygen exchange pathway has been confirmed in a recent crossed molecular beams experiment; however, the authors were not able to assign the nature if the CO3 intermediate explicitly.45

This process could contribute significantly to an 18O enrichment in carbon dioxide in the atmospheres of Earth and Mars.