Does acetone react with HO 2 in the upper-troposphere ?

Recent theoretical calculations showed that reaction with HO2 could be an important sink for acetone (CH3C(O)CH3) and source of acetic acid (CH 3C(O)OH) in cold parts of the atmosphere (e.g. the tropopause region). This work details studies of HO 2 + CH3C(O)CH3 (CH3)2C(OH)OO (R1) in laboratory-based and theoretical chemistry experiments; the atmospheric significance of Reaction (R1) was assessed in a global 3-D chemical model. Pulsed laser-kinetic experiments were conducted, for the first time, at the low-temperatures representative of the tropopause. Reaction with NO converted HO 2 to OH for detection by laser induced fluorescence. Reduced yields of OH at T < 220 K provided indirect evidence for the sequestration of HO2 by CH3C(O)CH3 with a forward rate coefficient greater than 2×10−12 cm3 molecule−1 s−1. No evidence for Reaction (R1) was observed at T > 230 K, probably due to rapid thermal dissociation back to HO 2 + CH3C(O)CH3. Numerical simulations of the data indicate that these experiments were sensitive to only (R1a) HO 2-CH3C(O)CH3 complex formation, the first step in (R1). Rearrangement (R1b) of the complex to form peroxy radicals, and hence the atmospheric significance of (R1) has yet to be rigorously verified by experiment. Results from new quantum chemical calculations indicate thatK1 is characterised by large uncertainties of at least an order of magnitude at T < 220 K. The large predicted values from Hermans et al. lie at the top end of the range of values obtained from calculations at different (higher) levels of theory. Atmospheric modelling studies demonstrated that whilst (R1) chemistry may be a significant loss process for CH3C(O)CH3 near the tropopause, it cannot explain observations of CH3C(O)OH throughout the troposphere.


Introduction
Acetone, CH 3 C(O)CH 3 , is emitted by vegetation, biomass burning, the oceans and industry at ∼95 Tg yr −1 (Jacob et al., 2002).The long lifetime of CH 3 C(O)CH 3 near the surface allows for transport out of the boundary layer (Atkinson and Arey, 2003).In the cold, dry upper troposphere (UT), CH 3 C(O)CH 3 photolysis may be an important initiator of free-radical chemistry, leading to generation of O 3 and OH, the primary oxidant in the troposphere (Arnold et al., 1986;Singh et al., 1995;McKeen et al., 1997;Wennberg et al., 1998;Arnold et al., 2005).It was recently proposed (Hermans et al., 2004) that reaction with HO 2 (R1) initiates degradation of CH 3 C(O)CH 3 in the UT.CH 3 C(O)CH 3 + HO 2 + M (CH 3 ) 2 C(OH)OO + M (R1) Using quantum-chemical and statistical rate calculations, Hermans et al. (2004) studied the (R1) equilibrium (in reality an association and rearrangement), and derived an overall K 1 (T ) = 7.8 × 10 −28 exp(7180/T ).The finer mechanistic details of the two-step (R1) process are unimportant when assessing atmospheric impact, but can lead to complications in interpretation of laboratory data.The first HO 2 -CH 3 C(O)CH 3 complex formation step (here denoted R1a), was identified in the theoretical calculations of first Aloisio and Francisco (2000) and then Hermans et al. (2005); experimental evidence for the (R1a) equilibrium has recently been published by Grieman et al. (2011) CH 3 C(O)CH 3 + HO 2 HO 2 •CH 3 C(O)CH 3 (R1a) The complex is a 7-membered ring structure whereby HO 2 is bound between the carbonyl group and an H from a methyl Published by Copernicus Publications on behalf of the European Geosciences Union.
group of CH 3 C(O)CH 3 .Concerted addition of the terminal O of HO 2 to the carbonyl C, and H-atom transfer from HO 2 to the carbonyl O (R1b) completes (R1).Hermans et al. (2004) report that at T < 220 K, (R1) proceeds rapidly (k 1 > 1 × 10 −12 cm 3 molecule −1 s −1 ) with formation of significant quantities of (CH 3 ) 2 C(OH)OO as the reverse reaction is slow (k −1 < 20 s −1 ).CH 3 C(O)CH 3 degradation is completed by subsequent reactions of (CH 3 ) 2 C(OH)OO.In the UT this is mainly with NO (R2) and to a lesser extent HO 2 and other radicals.
The key features of (R1-R2) chemistry of interest in the atmosphere or in this work are depicted on Fig. 1.
Using typical tropopause conditions of T = 200 K, [OH] = 6 × 10 5 , [HO 2 ] = 2.5 × 10 7 and [NO] = 4 × 10 8 molecule cm −3 , Hermans et al. ( 2004) calculated an effective CH 3 C(O)CH 3 removal rate via Reactions (R1) and (R2) of 6 × 10 −7 s −1 .The predicted product of (R2), (CH 3 ) 2 C(OH)O, would decompose to form acetic acid (CH 3 C(O)OH), and CH 3 radicals which are converted to HO 2 and HCHO.Following the well-characterised degradation of HCHO into HO 2 the overall result of (R1) initiated acetone degradation is therefore HO x production (net +0.6 HO x per CH 3 C(O)CH 3 destroyed, assuming UT chemistry not NO limited).The loss of acetone via (R1-R2) was thus predicted to compete with photolysis (R3, net + 3.2 HO x ) and reaction with OH (R4, net +1.2 HO x ).Large uncertainties remain however, and it is unclear which of (R1), (R3) or (R4) is the principal process for CH 3 C(O)CH 3 degradation, and subsequent radical production in the UT.Whilst k 4 (T ) values are reasonably wellestablished (Wollenhaupt et al., 2000), the Reaction (R3) photolysis quantum yields recently reported by Blitz et al. (2004) are considerably smaller than earlier measurements (Gierczak et al., 1998); the discrepancy is particularly large (≈ an order of magnitude) at the low temperatures of the UT.Meanwhile CH 3 C(O)CH 3 degradation via (R1) has yet to be experimentally verified, and theoretical estimates for the equilibrium constant K 1 (the crucial parameter for assessing atmospheric impact) differ widely.In a second publication (Hermans et al., 2005) demonstrated that their calculations reproduced experimental results for the analogous reactions of HO 2 with HCHO (Veyret et al., 1989) and CH 3 CHO (Tomas et al., 2001).Hermans et al. (2005) nonetheless acknowledged that their predictions for (R1) still await experimental verification.From thermochemical calculations (Benson, 2001) had previously identified a stepwise addition and rearrangement mechanism for (R1), with the group additivity method used to estimate a   Gierczak and Ravishankara (2000), who reported k(298 K) < 8 × 10 −16 cm 3 molecule −1 s −1 .The work presented in this manuscript details the first laboratory studies to investigate HO 2 -CH 3 C(O)CH 3 interactions at low temperatures characteristic of the tropospause, together with new theoretical calculations on the (R1) equilibrium.The potential for (R1) to impact on the chemistry of the upper-troposphere was assessed in a global 3-D chemical model.

Methods
Sections 2.1 and 2.2 below present respectively the experimental and the atmospheric modeling methods used in this work.Details of the quantum chemical methodology used to assess the uncertainties in K 1 (T ) are presented in the Supplement.

Laboratory experiments
The experiments detailed in this work used Pulsed Laser Photolysis (PLP) reaction initiation coupled to pulsed Laser Induced Fluorescence (LIF) detection of OH.The experimental set-up has been published previously (Wollenhaupt et al., 2000;Dillon et al., 2006), and is described only briefly here.
Experiments were conducted in a 500 cm 3 jacketed quartz reactor.Temperature was controlled by the circulation of a cryogenic fluid through the outer jacket, and monitored with a J-type thermocouple to an estimated accuracy of ±2 K. Reaction was initiated by an excimer laser operating at 351 nm (XeF).The frequency-doubled emission from a Nd-YAG pumped dye laser was used to excite the A Reagent concentrations, typically [CH 3 OH] = 5 × 10 14 and [O 2 ] = 4 × 10 16 molecule cm −3 were chosen such that Cl was converted within 50 µs and at 97 % yield to HO 2 .These experiments did not employ direct HO 2 detection, but relied on its conversion to OH by the presence of NO at concentrations of (1-3) × 10 14 molecule cm −3 .
In the absence of CH 3 C(O)CH 3 , the observed LIF profiles were therefore characterised by the formation of OH (R8, ≈1 ms) followed by its slow decay via (R4) and (R9-R10).
CH 3 C(O)CH 3 was added in concentrations of up to 1.6 ×10 15 molecule cm −3 to this photolysis mixture.The competition (see Fig. 1) between CH 3 C(O)CH 3 and NO for reaction with HO 2 , and the impact upon observed yields of OH, was used to derive kinetic parameters for (R1).Analysis of these experimental OH profiles was assisted by use of numerical simulation using the FACSIMILE program (Curtis and Sweetenham, 1987) using an appropriate chemical scheme (see Sect. 3.1.2).
Chemicals: liquid samples of CH 3 C(O)CH 3 (Merck > 99.8 %) and CH 3 OH (LS Labor > 99.8 %) were subject to repeated freeze-pump-thaw cycles; NO (Linde) was distilled by repeatedly removing the light boiling fractions at 77 K and discarding the frozen residue as the sample warmed; N 2 and O 2 (Messer 5.0, 99.999 %), and Cl 2 (Linde 2.0 % in He) were used as supplied.

Atmospheric modelling
The general circulation model for atmospheric chemistry ECHAM5/MESSy1 (henceforth EMAC), (Jöckel et al., 2005(Jöckel et al., , 2006) ) was used to investigate the impact of (R1), with particular focus upon CH 3 C(O)CH 3 degradation and CH 3 C(O)O 2 H formation.EMAC was used with T42 spectral resolution, corresponding to 2.8 • × 2.8 • at the surface.The vertical resolution was 90 layers, of which about 25 were located in the troposphere.The model dynamics was weakly nudged (Jeuken et al., 1996;Lelieveld et al., 2007) towards the analysis data of the ECMWF operational model (up to 100 hPa) in order to realistically represent meteorology in the troposphere.Important changes were made to the model to account for recent improvements in the kinetic/photochemical database for reactions of CH 3 C(O)CH 3 and CH 3 C(O)OH.Quantum yields for CH 3 C(O)CH 3 photolysis were based on results from Blitz et al. (2004) as preferred by IUPAC (Atkinson et al., 2006).In addition, rate coefficients for CH 3 C(O)OH formation and loss processes were updated (see Table 1) to reflect recent laboratory data on the reaction of HO 2 with acetylperoxy radicals (Hasson et al., 2004;Dillon and Crowley, 2008;Jenkin et al., 2008) and several recent studies of the OH + CH 3 C(O)OH reaction, see IUPAC (2011) and references therein.

Results and discussion
Section 3.1 details the results from the laser-kinetic laboratory-based study of (R1), including derivations of a lower-limiting forward rate coefficient at T = 207 K.In Sect.3.2 we present a discussion of the sensitivity of these experiments and calculations.Section 3.3 details results Reactions only included in SR2: 4×10 −11 * k in units of cm 3 molecule −1 s −1 unless stated otherwise; values taken from the IUPAC evaluations of Atkinson et al. (2004Atkinson et al. ( , 2006) ) and (for SR2) Hermans et al. (2004).
from incorporation of the largest published (Hermans et al., 2004) values for k 1 (T ) and K 1 (T ) into a global 3-D chemical model.

Results from laboratory studies
Evidence for reaction was only observed at the lowest temperatures.Fairly complex chemistry controlled the OH production and loss depicted in the OH profiles.Numerical simulations (see Sects.3.1.2-3.1.3below) were therefore used to obtain best estimates for experimentally derived rate (k 1exp ) and equilibrium (K 1exp ) constants, and to identify these quantities with either HO 2 -CH 3 C(O)CH 3 complex formation (R1a) or the overall (R1) process..5×10 14 mlecule cm −3 , and are associated with three numerical simulations: simulation using published theoretical values of k 1 and K 1 from (Hermans et al., 2004) are depicted as the red dot-dashed line; the green dotted line represents k 1 = 0; the blue dashed line represents a simulation using k −1 = 0 and a forward rate coefficient

Evidence for reaction between HO
impact of (R1-R2) if we are to obtain a reasonable estimate for the efficiency of the HO 2 -CH 3 C(O)CH 3 interaction.
The effect of CH 3 C(O)CH 3 upon physical processes was assessed previously (Dillon and Crowley, 2008).No changes to LIF intensity were observed following addition of CH 3 C(O)CH 3 at concentrations up to 4 × 10 15 molecule cm −3 , demonstrating that changes to fluorescence yields and/or detection efficiency of OH were negligible.The changes in LIF signal observed in Fig. 2 therefore directly reflect changes to OH production and loss chemistry due to CH 3 C(O)CH 3 .Other parameters such as photolysis/probe laser fluence and reagent concentrations were unchanged.Data obtained in the absence of CH 3 C(O)CH 3 may thus be used to calibrate the experiment.
A semi-quantitative analysis of the data presented in Fig. 2 may be made by consideration of first-order rate constants for HO 2 removal (assumed equivalent to the observed OH formation rate) and ignoring the thermal decomposition of the product peroxy radical (setting k −1 = 0).In the absence of CH 3 C(O)CH 3 , HO 2 is removed by reaction with NO at a rate of k 8 [NO] = 2000 s −1 (uses k 8 (200-400 K) = 3.45 × 10 −12 exp(270/T ) cm 3 molecule −1 s −1 (Atkinson et al., 2004).As a first approximation we attribute all of the reduction in observed OH (≈40 %) to the impact of CH 3 C(O)CH 3 via (R1).To compete with (R8), the CH 3 C(O)CH 3 must sequester HO 2 at a rate given by k 1exp [CH 3 C(O)CH 3 ] of several hundreds persecond, implying that k 1exp (207 K) > 1 × 10 −12 cm 3 molecule −1 s −1 .A series of such back-to-back experiments was conducted, and the LIF signals observed were in inverse proportion to [CH 3 C(O)CH 3 ], indicating that this result is robust.Details of the experimental conditions used in these and other experiments are listed in Table 2.Note that an exact analytical treatment of these datasets was not appropriate, as OH production did not commence at t = 0 (there is a short delay whilst the precursor HO 2 is generated).
Experiments conducted at somewhat higher temperatures were characterised by much smaller changes in OH.For example, Fig. 3 displays a typical pair of OH profiles obtained at T = 228 K, wherein the addition of [CH 3 C(O)CH 3 ] = 1 × 10 15 molecule cm −3 resulted in an approximate 15 % reduction in LIF intensity.Whilst some of this change in OH could be due to (R1), the addition of CH 3 C(O)CH 3 to the reaction mixtures does impact on OH via other chemical processes, e.g. by removing OH directly (R4), or the precursor Cl radicals in (R11).
In summary, the observations presented in Figs.2-3 provide some evidence for the theoretical predictions of Hermans et al. ( 2004) of a significant forward rate coefficient with k 1 (200-210 K) ≈3 × 10 −12 cm 3 molecule −1 s −1 and a strongly temperature dependent reverse process (R1).The lower limiting forward rate coefficient is clearly not consistent with the much smaller k 1 ∼ 10 −16 cm 3 molecule −1 s −1 predicted by Cours et al. (2007).To isolate and quantify the effects on observed OH that were attributable to (R1), and hence better constrain our estimates of k and K, numerical simulations of the datasets were conducted.et al., 2007).As indicated by the blue dashed line, the data could be reasonably well reproduced using much smaller values of K 1 (see Sect. 3.1.2).

Numerical simulations of the (R1) data
Numerical simulation of kinetic data using the FACSIM-ILE program (Curtis and Sweetenham, 1987)    observed OH profile (the blue circles), suggesting a good experimental sensitivity, and that CH 3 C(O)CH 3 does indeed interact strongly with HO 2 at this temperature.Note that the small differences in the simulations with [CH 3 C(O)CH 3 ] = 0 (black solid line), and the k 1 = 0 green, dotted line were principally due to an increased OH removal rate via reaction with the added acetone (R4), and to a lesser extent scavenging of the Cl precursor (R11).
Data simulations incorporating (R1) chemistry (see Fig. 1) were less straightforward, with three crucial parameters k 1 (T ), K 1 (T ) and k 2 (T ) to be estimated for simulation at each experimental temperature (207-298 K).There are to date no theoretical or experimental studies of (R2), (CH 3 ) 2 C(OH)OO + NO, in the literature.In this work, we have therefore adopted the expression k 2 (200-300 K) = 2.7 × 10 −12 exp(360/T ), listed by Atkinson et al. (2004) for the reaction of NO with another C 3 peroxy radical, CH 3 CH(O 2 )CH 3 .For the simulations presented in Fig. 2 this expression gives k 2 (207 K) = 1.5 × 10 −11 cm 3 molecule −1 s −1 , which as Hermans et al. (2004) noted is the value reported for a number of peroxy + NO reactions at around T = 200 K.Initial values of k 1 (207 K) = 3.2 × 10 −12 cm 3 molecule −1 s −1 and K 1 (207 K) = 9.1 × 10 −13 cm 3 molecule −1 were taken from the calculations of Hermans et al. (2004).The resulting simulation of the data in Fig. 2 is displayed as the red, dot-dashed line.This simulation slightly overestimates the impact of CH 3 C(O)CH 3 on the observed OH (blue circle datapoints).A lower-limit k 1exp (207 K) > 2 × 10 −12 cm 3 molecule −1 s −1 was obtained from a series of similar such simulations of all the T = 207 K data in which the dissociation process (R1) was neglected.This and similar experimentally derived lower limiting values of k 1exp (T ) are listed in Table 2.
A similar methodology was used to simulate the T = 228 K data displayed in Fig. 3.In contrast to results obtained at lower T , there is little difference between the data obtained without CH 3 C(O)CH 3 (black squares), and at [CH 3 C(O)CH 3 ] = 1.0 × 10 15 molecule cm −3 (blue circles).
The black, solid line again represents a simulation with [CH 3 C(O)CH 3 ] = 0, which reproduced the experimental data well and was therefore used to fix the relative positions of the LIF and [OH] y-axes.The results obtained in the presence of CH 3 C(O)CH 3 at T = 228 K could be simulated reasonably well using k 1exp = 0 (e.g. the green dotted line on Fig. 3), demonstrating that within the experimental scatter, little or no interaction between HO 2 and CH 3 C(O)CH 3 was evident.Similar results were obtained at all higher temperatures (see Table 2), in line with the earlier experimental study by Gierczak and Ravishankara (2000) who observed no reactive loss of HO 2 in the presence of CH 3 C(O)CH 3 at 298 K.
In an investigation such as this, where numerical simulations were used to analyse the experimental data, there were several potential sources of error.Systematic uncertainties, both experimental and from the literature/estimated rate coefficients used for simulations almost certainly outweigh any experimental scatter (see Figs. 2-3).We estimate that important contributions derive from uncertainties (2σ ) in the following parameters: ±5 % in [CH 3 C(O)CH 3 ]; ±15 % in [NO]; ± 2 K in T ; ± 50 % in k 2 ; ±20 % in k 8 .Propagation of these potential sources of error does not however allow us to reconcile the observed OH profiles with the theoretical predictions of Hermans et al. for the overall (R1) process (see e.g. the red dot-dash line on Fig. 3).Different chemical scenarios were therefore considered in an attempt to explain the experimental observations, and identify the processes involved (R1 or just R1a).

Simulations varying k 1 (T ), K 1 (T ) and k 2 (T )
Simulations using FACSIMILE were constructed whereby different values of k 1 (T ), K 1 (T ) and k 2 (T ) could be tested alongside the basis set of reactions listed in Table 3.As regards k 1 , some constraint is provided by the lower-limit obtained at T = 207 K. Considerably smaller values of k 1 were required to adequately simulate the data at higher temperatures if Hermans K 1 (T ) was used.At T = 228 K for  example, a reasonable reproduction of the data was obtained using k 1exp = 0 (green, dotted line in Fig. 3), or up to 4 × 10 −13 cm 3 molecule −1 s −1 .Such a strong temperature dependence for the forward association process does not seem reasonable.The temperature dependence of an equilibrium constant for such an association reaction is usually dominated by the barrier to the dissociation process (e.g.R1).All experimental datasets were therefore simulated using k 1 (T ) = 3.3 × 10 −15 exp(1335/T ) cm 3 molecule −1 s −1 from Hermans et al. (2004) with the pre-exponential factor scaled to yield our lower limit value of k 1exp = 2×10 −12 cm 3 molecule −1 s −1 at 207 K. Simulations using these constrained values of k 1 , and varying k 2 within generous uncertainty limits of up to a factor of three were not able to reproduce the T = 228 K data.In practice therefore, orders-of-magnitude larger values of k −1 were required to adequately reproduce the experimental observations.
Experimentally constrained estimates of K 1exp (T ) were obtained by using the normalised upper-limit values of k 1 (T ), k 2 (T ) = 2.7 × 10 −12 exp(360/T ) as described above, and by manually adjusting k −1 until the data at T = 210, 215, 220 and 228 K was best produced by simulation.For example, the dashed blue line on Fig. 3 represents a reasonable reproduction of a T = 228 K dataset (the www.atmos-chem-phys.net/12/1339/2012/blue circles) using k 1 = 1.1 × 10 −12 cm 3 molecule −1 s −1 , k 2 = 1.3 × 10 −11 cm 3 molecule −1 s −1 and a value for k −1 of 15 000 s −1 .We can then calculate a value for the equilibrium constant of K 1exp (228 K) = k 1exp /k −1exp = 7.5 × 10 −17 cm 3 molecule −1 via this method.It was possible to obtain values for the equilibrium constant at four temperatures below 230 K, which are listed in Table 2 and displayed in van't Hoff format in Fig. 4. Also displayed on Fig. 4 are lines representing K 1 (T ) values from Hermans et al. (2004) and Cours et al. (2007), neither of which are well-reproduced by the results from this work.Striking however, is the coincidence of the equilibrium constants derived in this work, with the predictions for K 1a from Aloisio and Francisco (2004), and the recent experimental determinations from Grieman et al. (2011).The experiments presented in this work would be sensitive to K 1a if the forward reaction rate coefficient k 1a is similar to or greater than that used in the simulations, and if the resultant complexes can be irreversibly converted to other products.To address the first point, whilst there are no literature values for k 1a available, a rate coefficient of k 1a (200-230 K) ≥2 × 10 −12 cm 3 molecule −1 s −1 is reasonable for formation of such weakly bound complexes.By way of comparison Christensen et al. (2006) measured forward rate coefficients of around 1 × 10 −11 cm 3 molecule −1 s −1 for the equivalent HO 2 -CH 3 OH complex at such temperatures, whilst a similarly large capture rate for OH by CH 3 C(O)CH 3 was reported by Talukdar et al. (2003).Note also that at T = 207 K, a forward rate coefficient of at least 2 × 10 −12 cm 3 molecule −1 s −1 was determined in this work, and such processes are rarely, if ever, characterized by a strong temperature dependence.The complex is predicted to be short-lived, rapidly decomposing back to HO 2 + CH 3 C(O)CH 3 (or indeed undergoing rearrangement (R1b) to the peroxy radical) such that, in these experiments, the only mechanism available for irreversible product formation would be by bimolecular reaction with one of the excess reagents.The complex is unlikely to interact with closedshell molecules such as CH 3 OH or CH 3 C(O)CH 3 , but given the predictions of a weakly bound terminal O-atom may react (R12) with the excess radical species NO.
The Reaction (R12) thus provides a means of removing HO 2 from the system, though in order to deplete OH in line with observation, the HOONO product should not decompose (to OH + NO 2 ) on our experimental timescales of milliseconds.A barrier of 83 kJ mol −1 for thermal decomposition of HOONO reported by Fry et al. (2004) does imply a sufficiently long lifetime for the HOONO at the temperature range (207-228 K) of interest in this work.
Estimates of K 1a derived from simulated OH profiles are inversely proportional to the value of k 12 used to set up the simulation.Thus the values listed in Table 2 and  (the purple dot-dashed line); and recent experimental results (green triangles) from Grieman et al. (2011).The blue square datapoints depict results from this work, which appear to describe K 1a (T ).
(k = 1.3 × 10 −11 cm 3 molecule −1 s −1 at around 220 K).Whilst such simulations using (R12) adequately reproduced the size and shape of the observed OH profiles recorded at T > 207 K, we note that no literature data is available for k 12 .A necessarily detailed experimental or theoretical investigation of (R12) is beyond the scope of this manuscript.However, theoretical predictions of the geometry of the HO 2 -CH 3 C(O)CH 3 complex seem to indicate that access by NO to the radical oxygen site is not significantly sterically hindered.Furthermore, the dispersal of reaction energy of the HOONO formation across the entire acetone-HOONO complex can reduce the impact of prompt redissociation of the HOONO moiety.Hence k 12 can be tentatively estimated to be comparable to that for the C3-RO 2 + NO reaction.
A "3rd Law" fit of the approximate K 1a values listed in Table 2, constrained by the calculated reaction entropy (Aloisio and Francisco, 2000;Hermans et al., 2005) yields K 1a (215-228 K) = 1.253 × 10 −29 exp(6703/T ).A tentative estimate of the enthalpy for complex formation is thus H 1a = −55.7 kJ mol −1 .Taking into account merely systematic errors from the experiments (see Sect. 3.1.2above) and a conservative ±50 % in k 12 we may similarly derive values in the range of −57.3 to −54.2 kJ mol −1 for H 1a , indicating that the results derived here are in satisfactory agreement with the predictions of Aloisio and Francisco (2001) of H 1a = −56.5 kJ mol −1 .

Discussion -sensitivity of experiments and theory
As was detailed in Sect.3.1 above, the experimental observations from this work support predictions of Hermans et al. of an efficient HO 2 -CH 3 C(O)CH 3 interaction at low temperatures.Numerical simulations of the data however are consistent with complex formation (R1a) only, the first step in the postulated removal mechanism for atmospheric CH 3 C(O)CH 3 .No evidence was found in this work (or another recent experimental study from Grieman et al., 2011) for the subsequent rearrangement of the complex to form peroxy radicals (R1b).In this section we consider why sensitivity to (R1b) might have been lacking in these experiments.

Unknowns in the chemistry
Chemistry whereby HO 2 , the HO 2 -CH 3 C(O)CH 3 complex or indeed other radical products are converted to OH would render the experiment blind to the reactions of interest, but it is difficult to assess the impact of such hitherto unknown reactions.It is worth noting that the experiments conducted at T = 207 K were characterised by large OH reductions upon CH 3 C(O)CH 3 addition, indicating that such interferences were not dominant at this temperature.
Experimental concentrations of CH 3 C(O)CH 3 and NO were chosen such that the overall rate of (R1) given by k 1 [CH 3 C(O)CH 3 ] was similar to, and could therefore impact on, the rate of NO converting HO 2 to OH (R8).Potential scavenging of the intermediate complex by NO was not anticipated.Good experimental sensitivity to the overall (R1) can only be obtained if the rearrangement (R1b) proceeds at a rate fast enough to compete with (R12) in addition to the thermal decomposition processes.This is certainly open to question, not least because of large uncertainties associated with k 1a , k 1b and k 12 .The theoretical data allows estimation of the critical k 1b at the high pressure limit of ∼270 s −1 at 207 K and ∼860 s −1 at 220 K, using the CCSD(T)/aug-cc-pV∞Z//B3LYP data, while k 1 values from Hermans et al. (2004) may be combined with the experimental [CH 3 C(O)CH 3 ] of up to 1.7 × 10 15 molecule cm −3 to estimate an overall HO 2 loss of around 3000 s −1 for (R1) at T = 228 K (data displayed in Fig. 2).This latter is then similar to the estimated (see Sect. 3.1.3above) rate of scavenging of the complex by NO in (R12), which in turn is much faster than the thermal rearrangement of the H-bonded complex to the adduct.It is conceivable therefore that the experiments are only sensitive to K 1a because peroxy radical formation is too slow compared to competitive reactions.However, it is worth reiterating that the estimated rate constant (and products) of (R12) carry a large uncertainty also.Considerably lower NO concentrations characterise the UT, hence Reaction (R12) would not interfere with any adduct formation in the atmosphere.

Pressure effects
The published values of k 1 (T ) from Hermans et al. (2004Hermans et al. ( , 2005) ) are derived from transition state theory, and are consequently high-pressure limiting values.It is unclear whether the experiments from this work (e.g.P = 85 Torr (N 2 ) at 228 K) were conducted in conditions sufficiently close to the high-pressure limit.In the fall-off or low-pressure region, slower formation of the adduct leads to reduced experimental sensitivity.Again, this interference is important only in interpretation of laboratory data.The relevant parameter for atmospheric modelling is the equilibrium constant K 1 (T ), itself unaffected by, for example, low pressures that would constrain stabilization of the complex or adduct, as well as the k −1 back dissociation reaction rates.

Uncertainties on the theoretical predictions
A third possible reason for the experiments insensitivity to (R1), namely that K 1 was itself overestimated by theory, has therefore more important implications for atmospheric chemistry.In new theoretical work, detailed here in the Supplement, we have conducted quantum chemical calculations on the complex (R1a) and subsequent adduct formation (R1b) from HO 2 + CH 3 C(O)CH 3 .Important results include that reaction enthalpies for complex formation (R1a) were fairly similar across all methodologies, as were barrier heights for adduct decomposition (R1b).Divergence was found however across different levels of theory for the overall adduct formation exothermicity (R1), or for the adduct formation energy barrier from the complex (R1b).The relative energy of the adduct with respect to the separated reactants reduces quite significantly upon increasing the basis set from double to triple zeta quality, on average by 4.3 kJ mol −1 .This implies a reduction in the K 1 (220 K) of 1 order of magnitude; uncertainties of this order were already indicated by Hermans et al.Our current best level of theory available, CCSD(T)/aug-cc-pV∞Z//B3LYP/cc-pVTZ, suggests a well depth of 54.4 kJ mol −1 , 4.2 kJ mol −1 below the best G2Mc values of Hermans et al.Such an uncertainty in the adduct stability translates into a factor of ten uncertainty in K 1 (220 K).Hence, the values of K 1 reported by Hermans et al. lie at the upper-end of a large (∼ order of magnitude) spread of calculated values, and may therefore represent upper-limits.Consequently we have used Hermans et al. parameters to assess the likely maximum impact of HO 2 -mediated CH 3 C(O)CH 3 depletion on the chemistry of the atmosphere (see Sect. 3.3).The impact of the uncertainty on the theoretically derived rearrangement Reaction (R1b) is less significant; while the new theoretical data suggests a somewhat higher barrier with consequently slower adduct formation, this would not affect the impact of (R1) on the atmosphere.For experimental studies, a slower k 1b does increase interference of bimolecular reactions of the complex.N solstice initiated by: photolysis (R3) in green; reaction with OH (R4) in red; and in black the reaction of interest (R1) using k 1 and K 1 parameters from (Hermans et al., 2004).See Sect.3.3 for details.

Modelling studies
The modelling focused on the role of (R1) as an CH 3 C(O)CH 3 degradation process in the UT and its potential as a source of CH 3 C(O)OH, as postulated by Hermans et al. (2004).The simulation S0, which has been extensively described and evaluated previously (Jöckel et al., 2006;Pozzer et al., 2007) was used as a point of reference.A further two simulations were performed using augmented chemical mechanisms.In simulation SR1, the chemical mechanism (Sander et al., 2005) was updated to account for recent developments in peroxy radical chemistry (see Table 1 for details).In simulation SR2 the chemical mechanism was extended to include the reactions of interest from this work: (R1), (R2) and the reaction (R13) of HO 2 with (CH 3 ) 2 C(OH)OO, a minor additional pathway for trapping peroxy radicals formed in (R1).
Note that the HO 2 -CH 3 C(O)CH 3 complex was not explicitly treated in this work, as its reactions, e.g. the proposed (R12) do not degrade CH 3 C(O)CH 3 .As was discussed in Sect.3.2 above, large uncertainties remain in the magnitude of K 1 (T ), the crucial parameter for determining atmospheric impact.In this work the largest published values of K 1 (Hermans et al. 2004) were used in order to assess the likely maximum impact of (R1) on the chemistry of the troposphere.
Figure 5 displays the calculated removal rate for CH 3 C(O)CH 3 at 30 • North on 21 June 2000, based on the results of the SR2 simulation.The largest removal rates due to (R1) are located in the UT (where the lowest temperature conditions are found), with a maximum at around 100 hPa where a similar rate to that of photolysis (R3) and reaction with OH (R4) was calculated.This compares well with the Fig. 6.Vertical profiles of CH 3 C(O)OH (nmol mol −1 ) from simulations and observations.Asterisks and boxes represent the average and the standard deviation (with respect to space and time) of the measurements from a specific region (Tahiti) taken during the PEM Tropics-A campaign.On the right side the number of measurements for the specific region is listed.The simulated average is indicated by the line and the corresponding standard deviation with respect to time and space by the dashed lines.The blue, red and green lines represent model results from simulations SR1, SR2 and S0, respectively.
previous modelling study by Hermans et al. (2005) who calculated 30 % of CH 3 C(O)CH 3 degradation was due to (R1) at the tropopause.As was noted in Sect. 1 however, the yield of HO x from (R1-R2) is considerably smaller than that from either photolysis (R3) or OH + CH 3 C(O)CH 3 (R4); we therefore conclude that (R1) may make only a small contribution to radical production in the UT.
The contribution of (R1) to CH 3 C(O)OH formation, was similarly found to be greatest in the coolest regions.Updating the chemical mechanism with new kinetic data resulted in reduced production rates and enhanced loss rates of CH 3 C(O)OH, thus dramatically decreasing its mixing ratio and increasing the discrepancies between observation and simulation.This is illustrated in Fig. 6 where we show a comparison of simulated values obtained using S0, SR1 and SR2 from this work, with observations, taken from the PEM-Tropics-A campaign (Hoell et al., 1999).While simulation S0 is in reasonable agreement with the observations, updating the chemical mechanism (SR1) leads to a substantially smaller CH 3 C(O)OH mixing ratio throughout the troposphere.Figure 6 also displays the smaller (in absolute terms) but still significant change in modelled CH 3 C(O)OH upon inclusion of (R1) chemistry (the red line, SR2).The effects are insignificant at low altitudes, and CH 3 C(O)OH observations (e.g. at 6 and 8 km) are not reproduced despite the generous uncertainties quoted for both model and measurement output.Note however that above 10 km more than 50 % of modelled CH 3 C(O)OH mixing ratio may be attributed to reaction of CH 3 C(O)CH 3 with HO 2 .
The simulated rates of formation of CH 3 C(O)OH will depend critically on the model levels of NO x and organic precursors for the acetylperoxy (CH 3 C(O)O 2 ) radical.Thus, the model output for NO, the levels of which determine whether CH 3 C(O)O 2 reacts with NO or with HO 2 to form CH 3 C(O)OH, were compared with measurements in the same region.Likewise, PAN, formed from combination of CH 3 C(O)O 2 with NO 2 may be considered an indicator of rates of CH 3 C(O)O 2 formation and a comparison between observation and model was also conducted.In both cases, the model was able to reproduce the observed levels of NO and PAN reasonably well, suggesting no important omissions in the model code in this regard.Exemplary plots are provided in the Supplement to illustrate this.Given that the kinetic database on CH 3 C(O)OH removal by OH is now in good shape, the model-observation discrepancy is most likely due to missing source terms.Recent model calculations by Paulot et al. (2011)

Conclusions
The reaction HO 2 + CH 3 C(O)CH 3 (CH 3 ) 2 C(OH)OO (R1) was, for the first time, experimentally investigated in low-temperature conditions representative of the uppertroposphere.Evidence of reaction was observed only in the lowest temperature experiments, allowing a lower limiting value of k 1 (207 K) ≈2 × 10 −12 cm 3 molecule −1 s −1 to be estimated, in line with the theoretical predictions of Hermans et al. (2004).At temperatures above 230 K there was no observable interaction.Simulations were conducted which suggest these observations were consistent with HO 2 -CH 3 C(O)CH 3 complex formation (R1a), meaning that experimental evi-dence for the crucial second step of rearrangement to peroxy radical (R1b) is still lacking.Results from new theoretical work demonstrate the large uncertainties inherent in calculating K 1 at T < 220 K. Atmospheric modelling studies using the largest reported K 1 values demonstrated that whilst (R1) may play an role in CH 3 C(O)CH 3 removal and consequent CH 3 C(O)OH formation at the tropopause, it does not explain observations of CH 3 C(O)OH throughout the troposphere.
CH 3 C(O)CH 3 + hν → CH 3 + CH 3 CO (R3) OH + CH 3 C(O)CH 3 → CH 2 C(O)CH 3 + H 2 O (R4) Fig.1.Depiction of the chemistry described in this manuscript.Reaction with NO (R8, in red) was used to convert HO 2 to OH for detection by LIF in these laboratory experiments (Sect.2.1).Competition from HO 2 + CH 3 C(O)CH 3 leads to reduced OH yields, from which kinetic and thermodynamic parameters for (R1) were derived (see Sect. 3.1).Note: these experiments did not differentiate between the reactions of atmospheric interest (R1-R2, in black) and an alternative removal path for HO 2 (R1a followed by R12, in green), important only in the laboratory.

Fig. 2 .
Fig. 2. OH LIF profiles recorded following HO 2 generation (R5-R7) in the presence of NO (1.5×10 14 molecule cm −3 ) at 207 K and 25 Torr.Black squares are data with [CH 3 C(O)CH 3 ] = 0; the black solid line is the numerical simulation used to calibrate the LIF signal response.Blue circles are datapoints with [CH 3 C(O)CH 3 ] = 7.5×10 14 molecule cm −3 , and are associated with three numerical simulations: simulation using published theoretical values of k 1 and K 1 from(Hermans et al., 2004) are depicted as the red dot-dashed line; the green dotted line represents k 1 = 0; the blue dashed line represents a simulation using k −1 = 0 and a forward rate coefficient k 1 (207 K) = 2×10 −12 cm 3 molecule −1 s −1 .

Fig. 3 .
Fig. 3. OH LIF profiles recorded following HO 2 generation in the presence of NO (1.6 × 10 14 molecule cm −3 ) at 228 K and 85 Torr.Black square datapoints were obtained with [CH 3 C(O)CH 3 ] = 0; the black solid line is the associated numerical simulation used to calibrate the LIF signal response.Blue circles depict data obtained at [CH 3 C(O)CH 3 ] = 1.0 × 10 15 molecule cm −3 and are associated with three numerical simulations.Results from simulations using theoretical values of k 1 and k −1 from Hermans et al. are depicted as the red dot-dashed line.The green dotted line represents k 1 = 0 (indistinguishable from simulations using parameters from, Courset al., 2007).As indicated by the blue dashed line, the data could be reasonably well reproduced using much smaller values of K 1 (see Sect. 3.1.2).
also underestimated CH 3 C(O)OH compared to observations.The authors postulated that organic aerosols could balance the CH 3 C(O)OH budget, accounting for as much as 50 % of total CH 3 C(O)OH emissions.This hypothesis is backed-up by the ubiquitous model underestimation of CH 3 C(O)OH throughout the atmosphere which is indicative of a slow secondary CH 3 C(O)OH formation, rather than a source concentrated in the upper troposphere as postulated by Hermans et al.Further clarification of the atmospheric chemistry of CH 3 C(O)OH, whilst beyond the scope of this study, is clearly necessary.
transition of OH at 281.997 nm.The resulting fluorescence from OH was detected by a photomultiplier tube, which was shielded by 309 ± 5 nm interference and BG 26 (glass cut-off) filters.HO 2 was generated following 351 nm photolysis of Cl 2 using well-established Cl-atom chemistry (R5-R7).Laser fluences of ≈8 mJ cm −2 were used in conjunction with [Cl 2 ] = 1 × 10 14 molecule cm −3 to generate an estimated initial chlorine atom concentration of [Cl] ∼5 × 10 11 molecule cm −3 .

Table 1 .
Modifications to the chemical scheme used in the modelling studies.

Table 2 .
Experimental conditions and estimated k 1exp and K 1exp parameters.T /K P /Torr [CH 3 OH] a [NO] a [O 2 ] a [CH 3 C(O)CH 3 ] a n b k c all reagentconcentrations in units of 10 14 molecule cm −3 ; b n = number of experiments at different [CH 3 C(O)CH 3 ]; c lower limits to k 1exp /10 −12 cm 3 molecule −1 s −1 calculated assuming k −1 = 0; K 1exp in units of 10 −17 cm 3 molecule −1 estimates obtained by numerical simulation (see Sect. 3.1.3for details). a