Products and mechanism of the OH-initiated photo-oxidation of perfluoro ethyl vinyl ether, C 2 F 5 OCF QQQ CF 2

The OH-initiated photo-oxidation of perfluoro ethyl vinyl ether (C 2 F 5 OCF Q CF 2 , PEVE) in air (298 K, 50 and 750 Torr total pressure) was studied in a photochemical reactor using in situ detection of PEVE and its products by Fourier transform IR absorption spectroscopy. The relative rate technique was used to derive the rate coeﬃcient, k 1 , for the reaction of PEVE with OH as k 1 = (2.8 (cid:2) 0.3) (cid:3) 10 (cid:4) 12 cm 3 molecule (cid:4) 1 s (cid:4) 1 . The photo-oxidation of PEVE in the presence of NO x at 1 bar results in formation of C 2 F 5 OCFO, FC(O)C(O)F and CF 2 O in molar yields of 0.50 (cid:2) 0.07, 0.46 (cid:2) 0.07 and 1.50 (cid:2) 0.22, respectively. FC(O)C(O)F and CF 2 O are formed partially in secondary, most likely heterogeneous processes. At a reduced pressure of 50 Torr, the product distribution is shifted towards formation of FC(O)C(O)F, indicating the important role of collisional quenching of initially formed association complexes, and enabling details of the reaction mechanism to be elucidated. An atmospheric photo-oxidation mechanism for PEVE is presented and the environmental implications of PEVE release and degradation are discussed.


Introduction
The thermal and chemical resistance of fluoropolymers has led to their use in many industrial processes and to their production in large quantities 1,2 and thus to their release to the environment. We recently presented 3 the first detailed kinetic study on the reaction of perfluoro ethyl vinyl ether (C 2 F 5 OCFQCF 2 , henceforth referred to as PEVE) with OH radicals (R1), determining the rate coefficient, k 1 , using absolute and relative-rate techniques. k 1 was found to be independent of pressure but it has a significant negative temperature dependence, indicating that the first step is addition of OH to the CQC double bond. A room temperature value of k 1 close to 3 Â 10 À12 cm 3 molecule À1 s À1 implies an atmospheric lifetime of a few days, indicating that PEVE is not an important greenhouse gas. The main environmental concern about fluorinated vinyl ethers is whether their photo-oxidation leads to formation of persistent, fluorinated pollutants in the environment.
The products formed in the atmospheric photo-oxidation of organic trace gases depends on the fate of peroxy-radicals formed in the reaction between the initially generated organic radical fragments and O 2 . The chemistry is typically described as either ''high NO x '' where the peroxy radicals react with NO to form oxy radicals, which then propagate the radical chain, or ''low NO x '' where the dominating fate of the peroxy radical is reaction with HO 2 or other peroxy radicals. Most short lived anthropogenic pollutants are oxidized under high NO x conditions, but as the two limiting cases have different mechanisms, differences in their product distribution provide valuable information about key reaction steps.
Extending our kinetics study, this work is a detailed examination of the products formed in the OH-initiated atmospheric degradation of PEVE in air, with the aim of deriving a detailed reaction mechanism applicable to atmospheric conditions (both high and low NO x ). We also present a new measurement of the rate coefficient, k 1 , for the title reaction.

Experimental methods
The experimental set-up used to study the OH-initiated photooxidation of PEVE has been described in detail previously 4 and only a brief summary of its central components is given here. The reaction volume is a 44 l cylindrical quartz chamber equipped with a White-type multiple reflection mirror system resulting in a 43.7 m optical path length. Infrared spectra of the cell contents were recorded using a Bruker Vector FTIR instrument with a liquid N 2 -cooled MCT detector. The spectrometer was operated at a resolution of 0.5 cm À1 using boxcar apodization. The quartz reactor was at room temperature (298 AE 2 K) and for most experiments 1 bar total pressure. Six external, radially mounted, UV photolysis lamps provided a homogeneous light flux within the reactor for radical generation.
For the determination of k 1 and for the product study in high-NO x conditions, OH radicals were generated by photolysis of methyl nitrite (2-10 Â 10 14 molecule cm À3 ) 5 in the presence of up to 4 Â 10 14 molecule cm À3 NO using six Philips TL08 lamps (l max B 350 nm): CH 3  Under these conditions, the dominant fate of the organic peroxy radicals formed will be reaction with NO.
As the methyl nitrite based OH generation scheme is unsuitable for NO x free photo-oxidation, we also generated OH radicals by photolysis of ozone (1-2 Â 10 14 molecule cm À3 ) at 254 nm in a large excess of hydrogen (up to 2 Â 10 17 molecule cm À3 ). The large excess of H 2 ensures that O( 1 D) does not react with PEVE. As desribed previously, 3 This OH radical generation scheme generates both OH and HO 2 radicals, and under these conditions, the organic peroxy radicals formed in the PEVE photo-oxidation will either react with HO 2 radicals or undergo self-reaction.

Chemicals
Synthetic air (Westfalen), hydrogen (Westfalen 99.999%), ethylene (Matheson, Res. Pur.) and PEVE (Merck, 99%) were used as supplied without further purification. Ozone was generated by flowing air over a low-pressure Hg ''Penray'' lamp (thereby dissociating O 2 to ground state O-atoms, which recombine with O 2 to form O 3 ) before entering the reaction cell.
Methyl nitrite was prepared by the drop-wise addition of 50% sulphuric acid (Sigma-Aldrich) to a saturated solution of sodium nitrite in 50% methanol (Merck) in a three-necked round flask with a stirring magnet at 0 1C. To collect the resulting methyl nitrite, a slow flow of nitrogen was passed over the reaction mixture and into a trap at À78 1C. The resulting yellow liquid was then purified by vacuum distillation. Finally, a gas-phase sample of CH 3 ONO, diluted in nitrogen was prepared and stored in a blackened glass bulb at 298 K. The IR spectrum of a dilute gas-phase sample of methyl nitrite revealed minor impurities of methanol only (o0.3%).

Infrared spectrum of PEVE
The IR spectrum of PEVE is characterized by a progression of strong absorption bands in the range 1050-1400 cm À1 (shown in Fig. 1A), a weak CQC stretching mode at 1842 cm À1 and weaker absorption features at 749 to 840 cm À1 . The spectrum closely resembles that of perfluoro methyl vinyl ether (CF 3 OCFQCF 2 , henceforth referred to as PMVE) previously reported and assigned. 6 The reference spectrum of PEVE was obtained by recording spectra at different pressures of accurately diluted mixtures of PEVE (0.3 to 3.0 Â 10 13 molecule cm À3 ) and converting to absorption cross-sections after baseline correction and checking for Beer-Lambert behaviour. The PEVE and C 2 F 5 OCFO spectra were obtained in this study, those of CF 2 O and CF(O)CFO have been reported previously. 7 3.2 Infrared spectra of photo-oxidation products Using previously reported reference spectra ( Fig. 1) 7 CF 2 O and FC(O)C(O)F were readily identified as products of the OHinitiated oxidation of PEVE. The product spectrum obtained following PEVE oxidation showed strong absorption features at 1100-1300 cm À1 in addition to a progression of carbonyl absorption features at 1800-2000 cm À1 .
After subtracting features due to CF 2 O and FC(O)C(O)F ( Fig. 1 panels B and C) the resulting spectrum is characterised by three strong absorptions at 1128, 1173 and 1249 cm À1 , one at 1899 cm À1 and some weaker absorptions at 720, 757, 853, 950 and 1218 cm À1 (Fig. 1D). The spectrum has some similar spectral features to PEVE, but without the two vinylic C-F stretching modes at 1345 and 1296 cm À1 . The other features in the range 1000-1300 cm À1 are somewhat blue-shifted compared to the PEVE spectrum and the weak CQC-stretching absorption band at 1842 cm À1 is replaced by a much stronger absorption at 1899 cm À1 . The initial assignment of this spectrum to C 2 F 5 OCFO was based on similarities to the spectra of perfluoromethyl fluoroformate. 8 This assignment was independently confirmed by experiments in which the oxidation of PEVE was initiated by reaction with Cl-atoms rather than OH. In this case, only CF 2 O and C 2 F 5 OCFO are observed as products (see Fig. 2). This is in agreement with the results of Mashino et al. 7 who observed CF 3 OCFO as a photo-oxidation product of the Cl initiated oxidation of PMVE with a yield of 100%. The reasons for the simplified product distribution when using Cl rather than OH to initiate the oxidation of PEVE is discussed later when we examine details of the reaction mechanism.
As each PEVE lost by reaction with Cl forms one molecule of C 2 F 5 OCFO, we were able to put the C 2 F 5 OCFO spectrum on an absolute basis. Once spectral features corresponding to water and CF 2 O had been subtracted, the calibrated C 2 F 5 OCFO absorption spectrum was obtained, as displayed in Fig. 1D.

Relative-rate measurements of k 1
In our previous paper 3 on the kinetics of the reaction between OH and PEVE we reported a rate coefficient under NO x free conditions in 50 and 100 Torr N 2 and in 1 bar air. In this work we have measured the rate coefficient of PEVE with OH radicals relative to that of ethylene with NO x present.
In the absence of loss processes other than reaction with OH, the depletion factors, ln(initial concentration/concentration after time t), for reactant and reference are described by: [PEVE] 0 , [ethylene] 0 , [PEVE] t and [ethylene] t are the concentrations of PEVE and ethylene at times t 0 and t respectively, and k 1 and k 12 are the rate coefficients of reactions (1) and (12), respectively. PEVE + OH -Products (R1) The initial concentration ranges used for PEVE and ethylene were (1.3-2.9) Â 10 13 molecule cm À3 and (0.8-2.9) Â 10 14 molecule cm À3 , respectively. A potential interference in relative rate measurements of fluorocarbons is the reaction of CF 3 O radicals with organics like ethylene. 9 To minimise the impact of (R13), high concentrations ((1-4) Â 10 14 molecule cm À3 ) of NO were added to the mixture. NO reacts rapidly with CF 3 O (k 14 (298 K) = 5.4 Â 10 À11 cm 3 molecule À1 s À1 ) forming CF 2 O and FNO. 10,11 The relative concentrations of NO and C 2 H 4 used, and the rate coefficients for their reactions with CF 3 O, ensure that CF 3 O reacts predominantly (to 80%) with NO and variation of [NO] by a factor of four did not result in any significant changes in the measured rate coefficients. We thus conclude that our derivation of k 1 is not significantly influenced by undesired reactions of CF 3 O radicals.
A typical experiment lasted around one hour during which the photolysis lamps were switched on for 5-10 periods of 10-300 s. The intermittent acquisition of FTIR spectra (128 scans at 0.5 cm À1 resolution) took E5 min. The time dependent depletion factors for PEVE and C 2 H 4 were obtained by least squares fitting to reference spectra in the range 1080-1260 cm À1 for PEVE and 3030-3180 cm À1 for ethylene. Absorption by reaction products were accounted for by including reference spectra of CF 2 O, FC(O)C(O)F and C 2 F 5 OCFO as well as CH 3 ONO and its degradation products in the fit procedure (DOASIS 12 ). An example spectrum with fit residuals is shown in Fig. 3.
A total of six relative rate experiments were performed under different experimental conditions, which are summarised in Table 1 and Fig. 4. A linear, least squares fit to all data (solid line in Fig. 4) gives k 1 /k 12 = 0.361 AE 0.006 (2s). From this k 1 can be determined using an evaluated literature value 13 for k 11 (7.8 Â 10 À12 cm 3 molecule À1 s À1 at 298 K and 1 bar pressure) to give: k 1 = (2.8 AE 0.3) Â 10 À12 cm 3 molecule À1 s À1 where the uncertainty includes assessment of the accuracy of the reference rate coefficient. 13 The rate coefficient is in good agreement with the value (k 1 = (3.0 AE 0.3) Â 10 À12 cm 3 molecule À1 s À1 ) from our previous kinetic study on the title reaction 3 and for other perfluoro vinyl ethers for which the OH-rate coefficients fall in the range (2.2-3.4) Â 10 À12 cm 3 molecule À1 s À1 . 7,14,15 The rate coefficients and trends therein for different vinyl ethers are discussed in our previous publication 3 and the interested reader is referred to that publication for details.

Reaction products and yields (U NO x ) in air (with NO x )
Products of the OH-initiated photo-oxidation of PEVE were detected and quantified in a separate series of experiments.
In these experiments, NO is generated in the photolysis of CH 3 ONO and in some experiments was also added to the initial reaction mixture.
As mentioned above, C 2 F 5 OCFO FC(O)C(O)F and CF 2 O were observed as products. The experiments were conducted with several photolysis periods interrupted by spectrum acquisition so that the time dependence of PEVE loss and product formation could be analysed. Time dependent concentrations of PEVE were obtained by fitting post photolysis spectra to reference spectra between 1080 and 1360 cm À1 . Not only PEVE but also the fluorocarbon products (CF 2 O, FC(O)C(O)F and C 2 F 5 OCFO) absorb in this range, as do CH 3 ONO, CH 3 ONO 2 and HNO 3 and they were thus included in the spectral deconvolution procedure. Compared to PEVE, the fluorinated products absorb only weakly in this range (   is non-linear and has a significant induction period before a roughly constant slope is achieved (in this case after E1 Â 10 13 PEVE has been reacted). This is a clear indication that CF 2 O is formed in secondary reactions. The yield of a product which is formed directly (or very rapidly) will be independent of time, whereas one which is formed in slower, secondary processes will show more complex time dependence.
Time dependent yields of each of CF 2 O, FC(O)C(O)F and C 2 F 5 OCFO are displayed in Fig. 7. Whereas the yields of C 2 F 5 OCFO and FC(O)C(O)F are roughly constant, that of CF 2 O acquires a maximum value after several hundred seconds. The CF 2 O yields reported in Table 2 are averages obtained from the plateau after E240 s. Combining an estimated 10% uncertainty in absorption cross-sections with experimental scatter results in the following, average molar yields of products obtained in the presence of NO x at 1 bar pressure of air: The experimental CF 2 O yields are more variable than those of FC(O)C(O)F and C 2 F 5 OCFO, which is related to its production both in gas-phase and heterogeneous reactions (see below).
For FC(O)C(O)F and C 2 F 5 OCFO there is no significant difference in the yield between the different experiments, with or without added NO. In contrast, as illustrated in Fig. 7, the CF 2 O yields obtained at 1 bar pressure of air were largest when NO was added. The yields of all products were sensitive to the total pressure with values of Possible reasons and mechanistic implications of the [NO] and pressure dependence are discussed below along with the overall degradation mechanism.

Reaction products and yields (U) in air (without NO x )
The formation of products in the OH-initiated oxidation of PEVE in the absence of NO x was studied in a separate series of experiments where OH radicals were generated by the photolysis of ozone in a large excess of hydrogen (see Section 2). The initial concentration range used for PEVE was (1. The results and experimental conditions of these experiments are summarised in Table 3, the mechanistic implications of these observations are discussed below. The time profiles and product yields observed in the experiments with NO x present, indicate that CF 2 O, FC(O)C(O)F and CF 3 CF 2 OCOF are not depleted to any significant extent by reaction with OH radicals. This is consistent with the fact that perfluorinated, saturated organic trace gases do not react rapidly with OH. In the NO x free experiments with 253.65 nm photolysis of O 3 /H 2 /air as OH precursor, a lower yield of FC(O)C(O)F was observed and it decayed significantly during the measurement. As FC(O)C(O)F does absorb weakly at 254 nm, 16 some of this loss may be caused by photolysis. To examine whether other processes also play a role we stopped the photolysis after  generation of a significant amount of the products and left the mixture in the dark with 2-6 Â 10 14 molecule cm À3 ozone for one hour with spectra recorded at regular intervals. Even in the dark, the concentration of FC(O)C(O)F decreased (with a decay constant of 2.5 Â 10 À4 s À1 ) while no significant loss of CF 2 O or C 2 F 5 OCFO was observed. Potential causes of the dark loss of FC(O)C(O)F could be direct reaction with ozone in the gas phase, or on surfaces, the latter mediated by surface catalysed dissociation of ozone. 17 The scatter in the yields in the experiments without NO x most likely reflect variation in the removal rate and we therefore chose to report the product yield of

Mechanism for PEVE photo-oxidation in the presence of NO
From the observations described above, we can develop a reaction scheme for the OH-initiated oxidation of PEVE in air in the presence of NO x , which is summarised in Fig. 8.
The reaction of PEVE with OH radicals is initiated by addition of the OH radical to the double bond, which can in principle happen at both carbon atoms.
Computational studies on reaction of OH with PMVE 18 and perfluoro propyl vinyl ether (C 3 F 7 OCFQCF 2 , PPVE), 15 indicate that addition to the terminal C-atom (R1a) dominates. The large experimental yield of FC(O)C(O)F is not commensurate with a significant contribution of (R1b) and we conclude that the OH-addition is predominantly at the terminal carbon atom (R1a). The adduct formed in (R1a) may either dissociate via C-O bond fission (R15) or react with oxygen at the neighbouring site giving a hydroxy-peroxy radical (R16).
The conversion of FC(O)CF 2 OH to FC(O)C(O)F in (R17) is unlikely to be an elementary reaction, as discussed below.  The fate of the hydroxyl-peroxy radical formed in R16 depends on the experimental conditions. In the experiments with NO x present, the radical will react with NO, forming an oxy-radical (R18) which may then dissociate to form either HOCF 2 C(O)F and the C 2 F 5 O radical (R19a) or the observed product C 2 F 5 OC(O)F along with the CF 2 OH radical (R19b). C 2 F 5 OCF(OO)CF 2 OH + NO -C 2 F 5 OCF(O)CF 2 OH + NO 2 (R18) As HOCF 2 C(O)F may dissociate to FC(O)C(O)F and HF, both reactions fit the observations, but the calculations by Vereecken et al. 18 on PMVE and PPVE indicate (R19b) to be the dominant pathway. The fate of the CF 2 OH co-product of (R19b), is reaction with O 2 to form CF 2 O.
As indicated above, the product yields were found to be dependent on the total pressure of air, with a substantial decrease in F NO x (C 2 F 5 OCFO) when going from 750 Torr to 50 Torr, accompanied by increases in F NO x (FC(O)C(O)F) and F NO x (CF 2 O). These trends are consistent with the results of Mashino et al. 7 who (for PMVE) found that the CF 3 OCFO yield was reduced from 0.53 AE 0.04 at 700 Torr, to 0.08 AE 0.02 at 10 Torr.
They attributed this pressure dependence to competition between HF elimination and collisional stabilization of the chemically activated PMVE-OH adduct. However, Vereecken et al. 18 calculated that the barrier of this reaction is too high for it to be of any significance. Instead, they suggested that (for PMVE) the OH-adduct would preferentially dissociate to CF 3 + CF(O)CF 2 OH, with a barrier around 125 kJ mol À1 below the entrance energy and proposed the equivalent mechanism for PPVE. 15 Our observations of pressure dependent product yields can thus be understood in terms of competition between dissociation of an activated OH-PEVE complex ((R15), most important at low pressures) and thermalisation/reaction of the complex with O 2 ((R16), most important at high pressures).
As addition of chlorine to double bonds is significantly less exothermic than addition of OH, the PEVE-Cl adduct is formed at lower energies. One may therefore expect a different product distribution if the conclusions drawn above are valid. This is indeed the case as only C 2 F 5 OCFO is observed, indicating that the PEVE-Cl adduct is formed below the barrier for dissociation to C 2 F 5 + CF(O)CF 2 Cl and confirming that the product distribution is largely determined through the competition between dissociation and collisional stabilization of the initial PEVE-OH adduct.
The C 2 F 5 radical formed (R15) in the dissociation of the C 2 F 5 OCFCF 2 OH radical will react with oxygen to give the perfluoroethyl peroxy radical (C 2 F 5 O 2 ) which reacts with NO to form C 2 F 5 O ((R21) and (R22)). This radical is thermally unstable and will dissociate on a microsecond timescale to CF 3 and carbonyl fluoride (R23). 19,20 The CF 3 radical reacts further with O 2 and NO similar to the methyl radical giving another carbonyl fluoride molecule as final product ((R24), (R25) and (R14)).
According to the reaction scheme ((R14)-(R25)) describing the subsequent fate of the initially formed OH-PEVE adduct, there are two main reaction routes with CF 2 O yields of one and two respectively, the total yield of CF 2 O should approximately be equal to the FC(O)C(O)F yield plus one, which is consistent with the experimental results. Fig. 6 and 7 indicate that the formation of CF 2 O is delayed, whereas FC(O)C(O)F and C 2 F 5 OCFO are formed promptly. This is consistent with CF 2 O being formed from slow dissociation of CF 3 OH (R26).
Perfluorinated primary alcohols are known to dissociate to the corresponding perfluoroacyl halide and HF. 21,22 The mechanism for formation of FC(O)C(O)F in the photo-oxidation of a fluorinated vinyl ether has been discussed by Mashino et al. 7 who suggested direct HF loss from CFOCF 2 OH in the gas phase. Later, Vereecken et al. 18 calculated the barrier of this reaction to be 190 kJ mol À1 indicating that the reaction is too slow to be relevant at ambient temperatures and that FC(O)C(O)F is likely formed in a different process. The dissociation of CF 3 OH to CF 2 O + HF 23 has a barrier of similar magnitude to that of CFOCF 2 OH, and it is reported to take place rapidly in chamber experiments with a rate depending on the history of the reaction chamber, indicating that a heterogeneous reaction is taking place. 24 It has also been suggested that the dissociation of CF 3 OH is catalysed in the gas phase by water vapour, 23 HF, 25 CF 3 OH, or radicals. [26][27][28] The difference in the observed CF 2 O yields upon changing the NO concentration can thus be rationalized as follows: When there is enough NO available for scavenging CF 3 O, CF 2 O is formed directly in the gas phase via (R14). At lower NO concentrations, a fraction of the CF 3 O radicals react to CF 3 OH which will then convert to CF 2 O and HF (R26). The time dependence of CF 2 O formation is thus controlled by the rate of diffusion of CF 3 OH to and from surfaces and explains the induction period of minutes. Note that neither FNO nor CF 3 OH were observed, and HF only in low concentrations. This reflects their high reactivity of these molecules on the quartz and metal surfaces inside the reactor.

Mechanism for PEVE photo-oxidation in the absence of NO
In the experiments with ozone and H 2 , there is no NO to scavenge either the CF 3 O radicals or the peroxy radicals. The peroxy radicals may react with HO 2 to form either the hydroperoxide (R27) or with ozone to generate the oxy radical and oxygen (R28). C 2 F 5 OCF(OO)CF 2 OH + HO 2 -C 2 F 5 OCF(OOH)CF 2 OH + O 2 (R27) No evidence for formation of the hydroperoxide was found in the IR spectra and we assume that if it is formed it will either dissociate to the oxy radical and the OH radical (R29), be photolysed or be lost on surfaces.
The CF 3 O radicals react with hydrogen to give CF 3 OH and a hydrogen atom which rapidly react with O 2 to form an HO 2 radical CF 3 OH will decompose to CF 2 O and HF as described above, but as this most likely happens on surfaces, some of the resulting CF 2 O may then hydrolyse to CO and HF, explaining the lower yield of CF 2 O in the ozone experiments.
3.8 Theoretical calculations on the fate of CF 3 OH and HOCF 2 C(O)F As discussed above, the formation of CF 2 O from CF 3 OH and FC(O)C(O)F from HOCF 2 C(O)F, has been shown to be driven by heterogeneous reactions. While a detailed analysis of such reactions is outside the scope of the present paper, it is worthwhile to examine the intrinsic difference in reactivity between CF 3 OH and HOCF 2 C(O)F. We therefore examined the reactants and transition states (TS) for a set of representative reactions at the M06-2X/aug-cc-pVTZ level of theory. 29,30 The relative energies were refined by single point CCSD(T)-F12a/ aug-cc-pVDZ calculations. 31,32 All calculations were performed using the Gaussian-09 and Molpro 2015.1 program suites. 33,34 For both CF 3 OH and HOCF 2 C(O)F, reaction proceeds by conversion of the -CF 2 OH moiety to a carbonyl group, with HF as a co-product. Literature data indicates that direct HF elimination has a large energy barrier, 4160 kJ mol À1 , but that the reaction can be catalysed by many co-reactants. 23,[25][26][27][28] The experimental data presented in this work (see Fig. 6 and 7) suggest that CF 3 OH eliminates HF significantly slower than HOCF 2 C(O)F. To study this dependence on the molecular structure, we have calculated the rate-limiting TS for HF formation both for the un-catalysed reaction, as for catalysed processes, where we examined CF 3 OH, CH 3 OH, H 2 O, HF, HC(O)OH, HO 2 , and OH as potential catalysts (see Table 4). In agreement with literature data, we find that the barrier for the un-catalysed reaction is too high to allow a significant contribution at room temperature, but that the various catalysts reduce the barrier heights, by up to 160 kJ mol À1 . For nearly all catalysts, the barrier height for the HOCF 2 C(O)F substrate is lower than for the CF 3 OH compound, up to 13 kJ mol À1 . The lower TS energies afford a faster rate coefficient by over an order of magnitude, affirming the experimentally observed difference in conversion rates for HOCF 2 C(O)F and CF 3 OH. The difference in the reactivity trend is expected to remain valid for nearly all reactions including heterogeneous reactions, as the underlying cause is the difference in reaction enthalpy imposed by the -C(O)F moiety in HOCF 2 C(O)F, which is absent in CF 3 OH (see Table 4). Of the gas phase co-reactants considered, we find that HO 2 is the most efficient catalyst, pairing a radical reaction site with an -OO-molecular span that reduces the TS ring strain compared to the shorter -O-or -F-moieties in the other gas phase reactants. Intriguingly, we also find that CH 3 OH is a more efficient catalyst than CF 3 OH. This is related to the electron density on the -OH moiety; this aspect is not investigated in detail here. Larger catalysts consisting of complexes (e.g. HO 2 with water molecules) have been shown to be even more effective, but we do not expect significant concentrations of such gas-phase complexes under our reaction conditions. Formic acid, HC(O)OH, expected to be present on the reactor walls, affords the most drastic reduction in barrier height of the catalysts considered, essentially removing any energetic barrier beyond the intrinsic reaction endoergicity. The catalysis mechanism involves donating an H-atom from the acidic hydroxy group and receiving an H-atom with the carbonyl group while shifting the double bond, thus combining a mobile H-atom with a large -O-CQOmolecular span reducing TS ring strain, and strong H-bonding in the pre-and post-reaction complexes; this mechanism and its concomitant low energy barriers are likely to be applicable to all oxoacids including HNO 3 and H 2 SO 4 . An in-depth study of heterogeneous reactions, be it with chemical contaminants or exposed metal surfaces, is beyond the scope of this manuscript. While we cannot conclude that the reactions in Table 4 are those responsible for our observations of CF 2 O and FC(O)C(O)F formation, the selected catalysts are all expected to be present in our system, and combined allow for very rapid conversion of the fluoro-hydroxy moiety to a carbonyl moiety + HF, where especially low-volatility, surface-bound catalysts such as carboxylic acids or aqueous complexes have high catalytic activity. The -CFO moiety in HOCF 2 C(O)F lends itself to additional interactions (e.g. H-bonding), which could again enhance the interaction with a reactive surface relative to CF 3 OH.
Finally, we examined the reaction enthalpy of some other processes that could lead to carbonyl compounds indirectly, e.g. by formation of an -C FOH radical moiety that can react with O 2 to form -CFO + HO 2 . However, the available co-reactants NO, NO 2 , FNO and OH do not afford accessible pathways for CF 3 OH forming CF 2 OH, with endoergicities exceeding 175 kJ mol À1 in all cases. Note that such reactions would also not readily yield HF, observed as co-product in the reactions of fluorinated esters.

Persistence of the products of PEVE photo-oxidation in the atmosphere
As for other fluorinated VOCs the atmospheric photo-oxidation products of PEVE are fluorinated carbonyl compounds. Unlike their non-fluorinated analogues, fluorinated carbonyl compounds react too slowly with OH radicals for this reaction to constitute a significant atmospheric sink and their lifetime is limited by uptake on water droplets followed by hydrolysis. 35 Glyoxal, the non-fluorinated analogue of FC(O)C(O)F, has a very short atmospheric lifetime with respect to photolysis. 36 In contrast, the first electronic excitation in FC(O)C(O)F appears at 334 nm and is very weak. 16 Photolysis does therefore not constitute a significant sink of FC(O)C(O)F in the atmosphere.
To further investigate the fate of the C 2 F 5 OCFO, CF 2 O and FC(O)C(O)F products with respect to loss to aqueous atmospheric systems, an experiment was performed in which, following PEVE oxidation, the contents of the photochemical reactor was passed (using a chemically inert membrane pump at a flow of 5 SLM) through a wash bottle (volume = 300 cm 3 ) containing distilled water. The gas flow did not bubble through the volume of the water, but was directed via a glass tube (velocity 4 m s À1 in the tube) towards its surface (area 50 cm 2 ). The residence time of the gas mixture in the head-space of the wash-bottle was E2.4 s and the flow/geometry induced turbulent flow conditions in the head-space should ensure efficient gas-liquid contact.
Infrared spectra were recorded at 10 min intervals directly after the initial, gas-phase degradation of PEVE, then over a 60 minute period in which the contents of the cell were circulated through the membrane pump without exposure to water and then a further sixty minutes in which the water surface was available for reaction. During the first hour (no interaction with water), no significant loss of any products was observed, indicating no significant losses in the membrane pump. In contrast, the concentration of all three products decreased significantly upon contact with the water surface. Whereas CF 2 O was almost completely removed after E240 s (decay constant 6 Â 10 À4 s À1 ), the decay of C 2 F 5 OCFO (4 Â 10 À5 s À1 ) was an order of magnitude slower. FC(O)C(O)F decayed on a similar time scale to CF 2 O, but as the only absorptions not overlapping with the saturated water lines are very weak, we were unable to determine a decay constant.
The lifetime with respect to hydrolysis of trace gases in contact with aqueous surfaces is determined by their Henry's law solubility (H) and hydrolysis rate coefficient (k H 2 O ) as well as their rate of transport to the surface and diffusion in the bulk. George et al. 37 studied the uptake of CF 2 O by aqueous surfaces and report H(k H 2 O ) 1 2 = 350 M atm À1 s À1/2 at 273 K. A substantially lower values (at 278 K) of 4.3 M atm À1 s À1/2 was reported by de Bruyn et al. 38 Subsequently, Kanakidou et al. 39 modelled the atmospheric fate of CF 2 O based on the data of George et al. 37 and reported a lifetime with respect to in-cloud losses of the order of one week, allowing for a considerable uncertainty in the Henry law coefficients and hydrolysis rates. Given that the rate of turbulent transport to the water surface is expected to be identical for all the trace gases we can reasonably assume that the similar removal rates of FC(O)C(O)F and CF 2 O are due to similar values for H(k H 2 O ) 1 2 and that they have a similar timescales for removal to aqueous surfaces in the atmosphere. In contrast, the much lower rate of removal of C 2 F 5 OCFO compared to CF 2 O implies a large (factor of E10) reduction in the H(k H 2 O ) 1 2 term and thus potentially longer atmospheric lifetime. The hydrolysis of FC(O)C(O)F is likely to lead to formation of HF and CO, whereas the potential hydrolysis products of C 2 F 5 OCFO are HF, CO and CF 3 C(O)OH. As both HF and CF 3 C(O)OH are highly soluble it is expected that they would remain undetected in the gas-phase but remain in the bulk water sample in our experiments.
Our crude experiments to investigate the relative loss rates of CF 2 O, C 2 F 5 OCFO and FC(O)C(O)F to aqueous surfaces can only give a broad indication of the fate of these trace gases in the atmosphere and dedicated experiments to investigate the atmospheric fate of C 2 F 5 OCFO and FC(O)C(O)F with respect to gas-phase and heterogeneous processes are necessary to fully characterise the environmental impact of PEVE.

Conclusions
The main products formed in the OH-initiated photo-oxidation of PEVE are C 2 F 5 OCFO, FC(O)C(O)F and CF 2 O with respective yields (at 1 bar and in the presence of NO x ) of 0.50 AE 0.07, 0.46 AE 0.07 and 1.50 AE 0.22. Variation of product yields with pressure and concentration of NO enabled a detailed mechanism to be derived, in which product branching is controlled by competition between dissociation and reaction (with O 2 ) of the initially formed reaction complex. In the environment, CF 2 O formed from PEVE degradation will hydrolyse, forming HF and CO. The fate of C 2 F 5 OCFO and FC(O)C(O)F in the atmosphere remains unknown, though heterogeneous processes are likely to be significant. In addition to the mechanistic studies, we also derived a rate coefficient (2.8 AE 0.3 Â 10 À12 cm 3 molecule À1 s À1 ) which is in good agreement with the values from our previous kinetic study.

Conflicts of interest
There are no conflicts to declare.