Photodissociation of CF3CHO provides a new source of CHF3 (HFC-23) in the atmosphere: implications for new refrigerants

Throughout the last century, chlorofluorocarbons (CFCs) and hydrochlorofluorocarbons (HCFCs) were the most widely used refrigerants. However, it was discovered that they released reactive chlorine in the upper atmosphere causing the hole in the ozone layer and they were phased out by the 1987 Montreal Protocol [1]. They were replaced by hydrofluorocarbons (HFCs), which contain no chlorine and do not damage the ozone layer. However, HFCs are potent greenhouse gases and they were phased out in a 2016 amendment to the Montreal Protocol [2]. The latest evolution of these refrigerants is the hydrofluoroolefins (HFOs). These molecules have no chlorine and incorporate a carbon-carbon double bond to greatly reduce their atmospheric lifetime, and hence their contribution as a greenhouse gas. In this work, we demonstrate that one of the most important HFOs in current use ultimately decomposes partially into HFC-23 (CHF3) in the atmosphere. HFC-23 is one of the most potent greenhouse gases known, and the most potent HFC. Despite its phaseout, the observed emissions of HFC-23 have been increasing recently and were the largest in history in 2018 with no conclusive explanation [3]. This work suggests that the production of HFOs might be partially responsible.


Main text
Halogenated molecules were widely used as refrigerants, propellants, foam blowing agents, fire retardants and other useful chemicals throughout the last century. The chlorofluorocarbons (CFCs) were, by far, the most proliferated until it was realised that their emission was contributing reactive chlorine to the upper atmosphere and causing the hole in the ozone layer. This led to the 1987 Montreal Protocol [1] that governed the global phasedown of CFCs via the hydrochlorofluorocarbons (HCFCs) as interim replacements that still deplete ozone, but to a much lesser extent. Ultimately, CFCs and HCFCs were replaced by hydrofluorocarbons (HFCs), which contain no chlorine and have no ozone depletion potential (ODP). However, HFCs are long-lived and potent greenhouse gases. Global warming potentials (GWPs) are used to compare newlyemitted greenhouse gases according to their contribution to surface heating, typically over 100 years and relative (by mass) to carbon dioxide i.e. GWP(CO 2 ) = 1. Typical HFCs exhibit GWPs in the thousands and up to 12 690 in the case of fluoroform (CHF 3 or HFC-23) e.g. 1 tonne of fluoroform, emitted today, contributes as much to surface heating over 100 years as 12 690 tonnes of CO 2 . Concerns about offsetting the hard-won environmental gains brought about by the CFC and HCFC phasedown led to the addition of HFCs in the 2016 Kigali amendment to the Montreal Protocol [2].
The leading replacement candidates are a new class of compounds called hydrofluoroolefins (HFOs), which have already been emitted and detected in the atmosphere in significant quantities [4]. These compounds are derived from HFCs but incorporate a carbon-carbon double bond that increases their reactivity, leading to much shorter atmospheric lifetimes (weeks compared to decades/centuries) and near-zero GWPs. One important HFO, currently, is HFO-1234ze (CHF=CH-CF 3 ) with zero ODP and a GWP < 1 (i.e. less than CO 2 ) [5]. These GWP estimates rely on chemical knowledge concerning the decomposition of the HFO and of its onward degradation products. It is known that atmospheric decomposition of HFO-1234ze yields trifluoroacetaldehyde (CF 3 CHO) with 100% molar yield [6], which is then removed from the atmosphere 4× faster than its progenitor via photolysis [7]. Studies dating back to the 1950s suggest that photolysis leads to trifluoromethyl (CF 3 ) and formyl (CHO) radicals [8][9][10] that lead to atmospherically less hazardous CO 2 and HF [6].
The apparently simple photochemistry of CF 3 CHO is in contrast to similar carbonyl species. Acetaldehyde (CH 3 CHO), the non-fluorinated analogue of CF 3 CHO, is known to exhibit far more complex photochemistry than reported for CF 3 CHO with multiple decomposition product channels. Analogous to CF 3 CHO, the dominant photolysis pathway for CH 3 CHO at all tropospherically-important wavelengths yields CH 3 and CHO radicals [11][12][13][14]. However, following absorption of an ultraviolet photon, CH 3 CHO can also access a rich photochemistry on the ground electronic state (S 0 ). These many ground state photodissociation pathways are discussed in the following section and, most importantly, include the exoergic formation of methane (CH 4 ) and carbon monoxide (CO) [15][16][17]. Fluoroform (CHF 3 or HFC-23) is the product analogue of CH 4 for CF 3 CHO. This is concerning as HFC-23 is the most environmentally hazardous HFC with a GWP of 12 690 [5].
In this paper, we demonstrate that CF 3 CHO photolyses to form HFC-23. Although the quantum yield is small, this pathway could account for the fate of 11.0 ± 5.5 % of atmospheric CF 3 CHO. This will considerably increase the 'effective GWP' of any HFO that includes CF 3 CHO in its decomposition scheme. Fig. 1 presents a potential energy diagram for the various photolysis channels of both CH 3 CHO and CF 3 CHO. The minimum energy structures for the intact molecules are shown in the centre and various reaction pathways proceed to the left for CH 3 CHO and to the right for CF 3 CHO. The dashed horizontal line represents the energy of a 308 nm photon (388.4 kJ mol −1 ), which is the wavelength employed in both the current and previous studies [8]. The potential energy schemes are strikingly similar. Several photodissociation channels are energetically open following 308 nm excitation, and there are no compelling differences that should preclude CF 3 CHO from accessing the same rich ground state photochemistry as its non-fluorinated analogue.

Photodissociation pathways
Photolytic cleavage of the C-C bond via the well-known Norrish Type-I (NTI) reaction, produces CH 3 /CF 3 and CHO (R1, green lines). This reaction is barrierless on the ground electronic state [18], requiring 347 and 335 kJ mol -1 for CH 3 CHO and CF 3 CHO, respectively. On the lowest energy triplet state, T 1 , the NTI reaction occurs over a small barrier (375 and 367 kJ mol -1 respectively). This pathway was not the focus of this study but we carefully studied the dynamics nonetheless since all previous work agrees that it is the dominant photolysis pathway for both species at 308 nm and it has not yet been explored for CF 3 CHO. The details and results of these experiments are in the supplementary information Section S1 (including Figs. S2 & S3). The main conclusion is that the dynamics are the same for CH 3 CHO and CF 3 CHO and in accord with the predicted energetics in Fig. 1.
The decarbonylation pathways produce CO, accompanied by CH 4 or CHF 3 (R2, red lines) and are the focus of this study. These are the thermodynamically favoured products, lying 25 and 43 kJ mol -1 respectively below their precursors, but proceeding over high barriers of 347 and 325 kJ mol -1 , respectively. These pathways were investigated using resonance-enhanced multiphoton ionisation (REMPI) with velocitymapped ion imaging (VMI) to probe CO photoproducts with quantum state specificity. We focus on the CO fragment because VMI depends on the target being ionised for detection, and CHF + 3 is unstable and decomposes to CF + 3 [19], which is already observed due to the radical channel (R1). These results are presented and discussed throughout the remaining sections of this article.
The most apparent difference in the photochemistry of these molecules is in the minor pathway producing CH 2 CO + H 2 in CH 3 CHO [20]. The transition state for the analogous pathway in CF 3 CHO could not be located; instead a hydrogen-bonding interaction is formed between the formyl H-atom and a fluoromethyl fluorine atom leading to CF 2 , HF and CO products. This predicted pathway has not been reported previously and has no analogue in the photochemistry of CH 3 CHO, but is a (potential) competing source of CO photoproducts and needed to be investigated. We searched for this channel using carefully designed lowpressure Fourier-transform Infrared (FT-IR) experiments with in situ photolysis and found no evidence for it with a detection limit corresponding to a maximum possible quantum yield of <0.07%. These experiments and analyses are described in detail in Section S2 (including Figs. S4, S5 & S6) of the supplementary information.
Dynamics of the decarbonylation channel (R2) Fig. 2 summarises the key experimental results from this study. Fig. 2A shows the 2D REMPI spectrum of CO formed from photodissociation of CF 3 CHO at 308 nm. The abscissa records the wavenumber of the CO detection laser on the lower axis and equivalent CO rotational quantum number, J, on the upper axis, and the ordinate shows the speed of the nascent CO. Two components are evident. The first appears in the lower left of Fig. 2A, outlined by the yellow triangle, and arises from slow CO with a narrow distribution of low-J states and is the dominant channel. The second component spans the entirety of Fig. 2A and reveals CO photoproducts with a broad distribution of speed and J. The 2D data in Fig. 2A can be integrated over either speed or wavelength as shown in Figs. 2B and 2C, respectively. Fig. 2B shows the 2D data integrated over low-speed (within the yellow triangle), high-speed and all-speed CO. Fig. 2C shows the orthogonal integration for low-J (within the yellow triangle), high-J and all-J. The data in both panels accentuate the bimodality in CO dynamics with low speed correlated with low J and high speed with high J.
This 2D REMPI spectrum is strikingly similar to that published previously for CH 3 CHO [15]. The dissociation of CH 3 CHO via R2 has been reported to occur via up to three different dynamical channels. Although the interpretation of each channel is still not agreed upon in the literature [15][16][17], the key factor for this work is that all hypothesised channels yield CH 4 and CO as photoproducts following dynamics on S 0 after initial photo-excitation to S 1 . The two different CO dynamical signatures from CF 3 CHO dissociation in Fig. 2 are consistent with the multi-dynamical processes in CH 3 CHO.

Quantum yield of the decarbonylation channel (R2)
From the experiments above, we conclude that R1 and R2 are active unimolecular photochemical decomposition pathways of CF 3 CHO at 308 nm while there is no evidence for R3. We take advantage of the similar photochemistry of CF 3 CHO and CH 3 CHO, as demonstrated above, and the well-known photochemistry of the latter to determine the quantum yield (QY) of R2 under low pressure conditions. We utilise CH 3 CHO as an internal standard with which we benchmark CF 3 CHO. Three gas cylinders containing pure He, and either 2% CF 3 CHO or CH 3 CHO pre-mixed in He were prepared in a parallel gas manifold and placed immediately before the pulsed nozzle of the REMPI experiment described above. These gas mixtures could be switched in seconds while all other experimental parameters remained stable. Fig. 3 shows the relative measurements of nascent CO (J = 8) signal from photolysis of each aldehyde as a function of time. Firstly, the CF 3 CHO supply was switched on for 10 seconds and off for 10 seconds for 10 cycles. The feed gas was then switched to helium, showing the slightly lower background and verifying the removal of residual CF 3 CHO. Finally, the feed gas was switched to CH 3 CHO and a similar profile and reproducibility was observed. The relative intensity between CF 3 CHO and CH 3 CHO signal in Fig. 3, the absorption spectrum of each compound [21] and survey CO photoproduct REMPI spectra for each compound (Fig. S7 of the supplementary information) were then used to determine that the nascent (zero-pressure) quantum yield of R2 from CF 3 CHO at 308 nm is 2.5 ± 0.7 times that of CH 3 CHO. Further details about these calculations are in the supplementary information section S3.

Atmospheric fate of CF 3 CHO
The results above demonstrate unambiguously that CHF 3 + CO is a primary product channel from the photolysis of CF 3 CHO at 308 nm and that CF 3 CHO and CH 3 CHO exhibit analogous photochemistry. We utilise this to contextualise our results and to estimate the yield of CHF 3 under atmospheric conditions. Fig.  4 shows a summary of all previous measurements of the CH 3 CHO and CF 3 CHO quantum yields in air at atmospheric pressure. The most extensive data set for CH 3 CHO was published by Moortgat and co-workers (MMW and WM) [13,14], while earlier measurements by Calvert and co-workers (HC and HKC) [11,12] are in good agreement. Panel A shows total photolysis QY, Φ(tot), as a function of wavelength, while Panels B and C show Φ(R1) and Φ(R2) (i.e. the photolysis QYs for channels R1 and R2) separately.
The function fit through Φ(tot) is the wavelength-pressure dependent quantum yield for CH 3 CHO photodissociation recommended by IUPAC, calculated at atmospheric pressure [22]. There are no recommended functions for the individual quantum yields. IUPAC recommends Φ(R2) = 0 for λ > 300 nm. However, the data in the inset to Fig. 4C show that, although the measured quantum yield at atmospheric pressure is small, it is not zero. The data shown here are consistent with the original publications [13,14], where the authors determine an upper limit of Φ(R2) < 0.005. Although this small quantum yield has negligible implications for the atmospheric chemistry of CH 3 CHO (which possibly led to the IUPAC recommendation), we argue below that this is not the case for CF 3 CHO.
We fit a sigmoidal function to Φ(R2) in Fig. 4C, while the line for Φ(R1) in Fig. 4B is the difference between Φ(tot) and Φ(R2). We do not attribute any meaning to the functional form and the important aspect is the non-zero offset at λ > 300 nm, which is almost flat. Our sigmoidal fit places this at Φ(R2) = 0.004 ± 0.002.
Previous studies of the photochemistry of CF 3 CHO are scarce. The available wavelength-dependent QYs for CF 3 CHO are also shown in Fig. 4 [8,9]. We have demonstrated that the energetics and dynamics of these two carbonyl reactions are remarkably similar. Fig. 4 shows that this is translated through to equally similar QYs under atmospheric conditions. Fig. 4C includes an estimate of Φ(R2) at 308 nm from this work. Here, we assume the collision-free measurement, Φ(R2, CF 3 CHO) = 2.5 × Φ(R2, CH 3 CHO), also applies at atmospheric pressure, i.e. that the quenching kinetics for CH 3 CHO and CF 3 CHO are similar. This provides an estimate of Φ(R2) = 0.010 ± 0.005. The previous estimate of Φ(R2) = 0.021 at 313 nm [9] lies a little above our estimate, as shown, but was recorded at lower pressure (700 Torr), with correspondingly less quenching and is therefore likely consistent with the present measurement.
The photolysis rate coefficient, k ph , can be calculated as k ph = λ j(λ)σ(λ)Φ(λ) where j(λ) is the the actinic flux, σ(λ) the absorption cross section and Φ(λ) the photolysis quantum yield. Using the actinic flux for an overhead sun on a clear day from the US EPA [23], the published absorption spectrum of CF 3 CHO [10] and the quantum yields from Fig. 4 evaluates to k ph (tot) = 5.9 × 10 -6 s -1 and k ph (R2) = (7 ± 4) × 10 -7 s -1 . CF 3 CHO can also decompose in the atmosphere by reaction with OH with a reported rate coefficient, k OH = (4.8 ± 0.3) × 10 -13 cm 3 molecules -1 s -1 [10]. Using the global average concentration of [OH] = 1.1 × 10 6 molecules cm -3 [24] yields a pseudo-first order rate coefficient for reaction with OH of k OH = 5.3 × 10 -7 s -1 . These rate coefficients govern the atmospheric fate of CF 3 CHO resulting in a half-life of ∼4 days with the dominant radical channel (R1) accounting for the removal of 79% of atmospheric CF 3 CHO, 10% decomposing by reaction with OH and 11% of CF 3 CHO in the atmosphere undergoing the decarbonylation reaction, R2, leading to CO and CHF 3 (HFC-23). Details of the calculations can be found in supplementary information Section S4, including Tables S1 & S2. The photochemistry of CH 3 CHO and CF 3 CHO are fundamentally the same, however, the CH 4 pathway in CH 3 CHO comprises only about 1% of the atmospheric fate, in comparison to the 11% calculated above for HFC-23 from CF 3 CHO. The underlying difference is that the CH 3 CHO reaction with OH has a rate coefficient (k OH = 1.5 × 10 −11 s −1 [22]) that is over an order of magnitude greater than that for CF 3 CHO. While CF 3 CHO undergoes an average of 11 absorption cycles before reacting with OH and a 1% quantum yield becomes an 11% reaction fate, CH 3 CHO absorbs only ∼2 photons on average before reacting.

Implications for HFO use
We have determined that CHF 3 + CO is unambiguously a primary photochemical product channel of CF 3 CHO. Under collision-free conditions, the quantum yield is 2.5 times higher than for the equivalent channel in CH 3 CHO. The dynamics, energetics and QYs of CF 3 CHO are very similar to those of CH 3 CHO. Utilising the more extensive data for CH 3 CHO and the measurements in this work for CF 3 CHO, we estimate the quantum yield of CHF 3 under atmospheric conditions to be (1.0 ± 0.5)%. Because the rate coefficient for reaction with OH is very slow, this 1% quantum yield results in (11.0 ± 5.5)% of CF 3 CHO molecules in the atmosphere decomposing via this CHF 3 pathway, resulting in a secondary GWP = 1400 ± 700 for CF 3 CHO. CF 3 CHO is formed in 100% molar yield from the atmospheric decomposition of HFO-1234ze [6] as well as from the decomposition of other HFOs [25] and HFCs [26]. This introduces a concern that HFOs being considered as replacement gases for air conditioning and other industrial uses might have significant impact as climate change agents. We note that if the GWP for CF 3 CHO from this work withstands further scrutiny, the European Union would re-classify HFO-1234ze from a 'low or no GWP' compound into one that is forbidden [27].
At this stage, we refrain from recommending the withdrawal of HFO-1234ze from commercial use; there are too many uncertainties at present. The most significant uncertainty is using quantum yields from CH 3 CHO as a proxy for CF 3 CHO, despite the similarity shown in Fig. 4. More research is needed measuring the wavelength and pressure dependence of quantum yields for R1 and R2 for CF 3 CHO explicitly and we call for urgent experiments in this regard.
In closing, we note that atmospheric monitoring has revealed an unknown source of HFC-23 over the past 3 years [3]. After decreasing for several years, the emission of CHF 3 in 2018 was higher than measured any year in history. The authors of that work explicitly suggest that China might not have met abatement targets. However, this study could provide an alternative explanation whereby legal production of HFOs might produce HFC-23 as a secondary atmospheric product thereby accounting for some of the observed increase. Figure 1: Potential energy scheme for the photodissociation of CF 3 CHO and CH 3 CHO. The CH 3 CHO energies are experimental values from the references cited in the text and the CF 3 CHO energies were calculated using the G4(MP2)-6x method [28]. The product and transition state energies are relative to the groundstate energy of the equilibrium geometry of each respective species. The dashed horizontal line represents the energy of a photon at 308 nm. Figure 2: (A) 2D REMPI spectrum of the CO fragments formed following photolysis of CF 3 CHO at 308 nm. B) CO REMPI spectra from integrating (A) at low speed (within the yellow triangle, red), high speed (outside the yellow triangle, blue) and all speeds (green). C) Speed distributions determined by integrating (A) within and without the yellow triangle and across all J.  [MMW: Reference [13]; WM: Reference [14]; HC: Reference [11]; HKC: Reference [12]. A) Total QY; B) QY for R1; C) QY for R2.

Methods Experimental
In this work, we employed three experimental techniques to examine the three photochemical pathways, R1 to R3. Vacuum ultraviolet ionisation coupled with velocity-mapped ion imaging was used to investigate the CF 3 and CHO radical fragments from R1. Resonance-enhanced multiphoton ionization with VMI was used to probe CO photoproducts from R2. FT-IR spectroscopy was used to probe the HF products from R3.
The apparatus for all experiments has been described in detail previously [29,30,31] and the specific