Impact of ozone and inlet design on the detection of isoprene-derived organic nitrates by thermal dissociation cavity ring-down spectroscopy (TD-CRDS)

. We present measurements of isoprene-derived organic nitrates (ISOP-NITs) generated in the reaction of isoprene with the nitrate radical (NO 3 ) in a 1 m 3 Teflon reaction chamber. Detection of ISOP-NITs is achieved via their thermal dissociation to nitrogen dioxide (NO 2 ), which is monitored by cavity ring-down spectroscopy (TD-CRDS). Using thermal dissociation inlets (TDIs) made of quartz, the temperature-dependent dissociation profiles (thermograms) of ISOP-NITs 10 measured in the presence of ozone (O 3 ) are broad (350 to 700 K), which contrasts the narrower profiles previously observed for e.g. isopropyl nitrate (iPN) or peroxy acetyl nitrate (PAN) under the same conditions. The shape of the thermograms varied with the TDI’s surface to volume ratio and with material of the inlet walls, providing clear evidence that ozone and quartz surfaces catalyse the dissociation of unsaturated organic nitrates leading to formation of NO 2 at temperatures well below 475 K, impeding the separate detection of alkyl nitrates (ANs) and peroxy nitrates (PNs). We present a simple, viable solution to this problem and discuss the potential for interference by the thermolysis of nitric acid (HNO 3 ), nitrous acid (HONO) and O 3 . in the 575-650 K range, which may be related to changes in gas-flow and heat-transfer within the inlet caused by the glass beads. Neither TD-inlet type results in dissociation to NO 2 at temperatures < 500 K. In contrast, the ISOP-NIT thermograms (normalised to the signal at the plateau at 625 K of TDI-2) indicate formation of NO 2 over a much broader


Introduction
Understanding the atmospheric fate of nitrogen oxide (NO) and nitrogen dioxide (NO2) is critical as both trace-gases have a great impact on air quality and human health (Crutzen and Lelieveld, 2001;Lelieveld et al., 2015). Ambient measurement of trace-gases that function as NOx reservoirs or sinks (where NOX = NO + NO2) are thus needed to provide insight into NOx 20 removal and transport. Organic compounds with nitrate functionality can serve as NOX reservoirs in the troposphere (Thornton et al., 2002;Horowitz et al., 2007) and are generally categorised as peroxy nitrates (PNs, RO2NO2, with peroxy acetyl nitric anhydride (PAN) being its most abundant representative in the troposphere) and alkyl (aliphatic) nitrates (ANs, RONO2). PNs are formed via the reaction of organic peroxy radicals (RO2) with NO2 (R1); ANs are formed via the minor (termolecular) channel of the reaction of RO2 with NO (R2a). The competitive bimolecular process leads to alkoxy radicals (RO,R2b). During 25 the daytime, RO2 is formed mainly by the oxidation of volatile organic compounds (VOCs) by hydroxyl radicals (OH)  At nighttime, when OH radicals and NO are significantly less abundant, the NO3 radical initiates the oxidation of many VOCs (Ng et al., 2017). NO3, formed in the oxidation of NO2 by O3 (R5), is photolysed rapidly by sunlight (R6) and also reacts efficiently with NO (R7) so that it is generally of minor importance during the day (Wayne et al., 1991). 35 NO3 readily undergoes reactions with many unsaturated organic trace-gases of biogenic origin including isoprene, monoterpenes and sesqui-terpenes to form organic nitrates in high yields (Ng et al., 2017;Wennberg et al., 2018;Mellouki et al., 40 2020). The focus of this work is the formation of organic nitrates in its reaction (in air) with the C5-di-ene isoprene (ISOP, R8) where ISOP-NIT represents an isoprene-derived nitrate. Isoprene is the most abundant non-methane VOC in the atmosphere (Guenther et al., 2012) and a large fraction of the organic nitrates formed at night-time is attributed to the reaction between NO3 and isoprene (R8) (Carlton et al., 2009). The atmospheric oxidation of isoprene involving OH, O3 and NO3 as oxidizing 45 agents is complex and leads to a huge variety of products (Ng et al., 2017;Wennberg et al., 2018) including multifunctional, unsaturated nitrates such as O2NOCH2C(CH3)=CHCHO, O2NOCH2C(CH3)=CHCH2OH or O2NOCH2C(CH3)=CHCH2OOH among other secondary oxidation products like dinitrates or epoxides (Wu et al., 2020).
Studies of the NO3-induced oxidation of isoprene in air report AN yields between 60 and 100 % (Barnes et al., 1990;Berndt and Boge, 1997;Perring et al., 2009;Kwan et al., 2012;Schwantes et al., 2015;IUPAC, 2017;Wu et al., 2020;Brownwood 50 et al., 2021) and the NO3-induced oxidation of isoprene is responsible for a dominant fraction of organic nitrates observed in rural environments with strong biogenic emissions (Beaver et al., 2012). The major fate of isoprene-derived organic nitrates formed in the boundary layer is deposition onto particulate matter to form nitric acid (HNO3) or secondary organic aerosols (SOA) leading to largely irreversible removal of NOX from the gas phase (Ng et al., 2008;Rollins et al., 2009;Fry et al., 2018;Hamilton et al., 2021). 55 Individual isoprene nitrates have been measured selectively in the atmosphere by mass-spectrometric methods (Wolfe et al., 2007;Wu et al., 2020). An alternative detection scheme, in which the sum of all atmospheric PNs (ΣPNs) and ANs (ΣANs) are separately measured, takes advantage of their different C-N bond energy by combining thermal dissociation to NO2 (R9 and R10) with detection of the latter with means of laser-induced fluorescence (TD-LIF) or cavity ring-down spectroscopy (TD-CRDS). Several instruments using thermal dissociation inlets consisting of fused silica (quartz) with residence times between tens to hundreds of milliseconds have been described in the literature (Day et al., 2002;Paul et al., 2009;Wild et al., 2014;Sobanski et al., 2016;Thieser et al., 2016;Keehan et al., 2020). In these instruments, quantitative conversion of PAN is reported for 65 temperatures between 375 and 420 K. Generally, the temperature dependence of the instrument's response to ANs has been tested using mostly saturated organic nitrates such as isopropyl nitrate (iPN) and isobutyl nitrate, which are dissociated to NO2 at temperatures between 500 and 675 K. The temperatures at which PNs and ANs are quantitatively converted to NO2 thus differ by ~ 200 K and are largely independent of their organic backbone (Kirchner et al., 1999;Wild et al., 2014) allowing separate measurement of the sum of all alkyl nitrates (ΣANs) and of the sum of all peroxy nitrates (ΣPNs). These observations 70 have provided the basis for analysis of field data in which an unknown mixture of PNs and ANs are present. We note however, that most of the first generation ANs formed in the NO3 + isoprene system still contain a double-bond (Barnes et al., 1990;Skov et al., 1992;Schwantes et al., 2015) which renders them more reactive towards oxidizing agents than e.g. iPN. A well characterised thermogram for aliphatic nitrates derived from the oxidation of e.g. isoprene is thus a pre-requirement for extracting the mixing ratios of PNs and ANs from ambient measurements when using a TD-inlet. To date, only one such 75 thermogram has been presented (Brownwood et al., 2021) which appears to be the result of a single experiment (i.e. no variation of experimental conditions) using a sample that was not stable over time. The thermogram also features slopes before and after the ANs transition temperature which is consistent with the ideal behaviour of e.g. iPN.
In this study, we generated ISOP-NITs by reacting isoprene and NO3 in a Teflon simulation chamber and used a custom-built, five-channel Cavity-Ring-Down Spectrometer (CRDS) (Sobanski et al., 2016) to analyse the organic nitrates formed. In the 80 presence of O3 we find that ISOP-NIT does not behave like the saturated analogue iPN in our quartz TD-inlet and we characterised the processes (both gas-phase and surface-catalysed) that lead to the observed behaviour. We also examined the potential role of surface-catalysed dissociation of HNO3 and nitrous acid (HONO) to NO2 as well as the effect of humidity as a potential bias to measurements of PNs and ANs.

Simulation Chamber
In order to analyse organic nitrates formed from the NO3 + isoprene system under realistic operational conditions for the 5-Channel-CRDS (e.g. normal sample flow rates), we constructed a dynamic, flow-through simulation chamber SCHARK (Simulation CHamber for Atmospheric Reactions and Kinetics) of volume 1 m 3 (cubic, all sides ~1m long) made of PFA foil of 0.005 in. (~0.13 mm) thickness (Ingeniven). The chamber is operated at ambient pressure and temperature; a magnetically 90 coupled, Teflon coated propeller-type stirrer situated in the centre of the chamber floor ensures continuous mixing of the air.
The trace-gas inlets and sampling ports were located at opposite corners of the cubic chamber to reduce the potential of sampling gas that had not yet mixed. The PFA foil is surrounded by a 120  120  120 cm cube constructed of four Perspex and two steel walls, the interspace (0.7 m 3 ) is permanently flushed with 1 SLPM (L (STP) min -1 ) of dry synthetic "zero-air" in https://doi.org/10.5194/amt-2021-128 Preprint. Discussion started: 7 May 2021 c Author(s) 2021. CC BY 4.0 License. order to avoid contamination through permeation of trace-gases present in the laboratory air. The Perspex walls serve as 95 observation windows and were covered with light-tight material during the experiments described here.
Zero air was provided by passing pressurized air through a commercial air purifier (CAP 180,Fuhr GmbH). Humidification of the air was achieved with a permeation source (MH-110-24-F-4, Perma Pure LLC) filled with deionized water. Typical total flow rates of 15 or 23 SLPM zero air into the chamber result in exchange rates kexch of 2.7 or 4.2 x 10 -4 s -1 , i.e. lifetimes of gases in the chamber of ~ 40-60 minutes. Note that in "flow-through" operation, the concentrations of trace gases in the 100 chamber are controlled both by chemical processes and by the rate of flow into (and out of) the chamber so that "steady-state" is achieved on the order of hours.
Ozone mixing ratios in the SCHARK were measured by sampling 2 SLPM through a ~3 m long section of 0.25 in. (outer diameter, OD) PFA-tubing to a commercial ozone monitor (2B Technologies, model 205) with a detection limit of ~ 1 part per billion by volume (ppbv) and 5 % uncertainty. O3 measurements were also used to establish the time required (under standard 105 flow conditions) to acquire complete mixing within the chamber (< 1 minute) and to derive the exchange rate by monitoring the exponential rise or decay of O3 when its supply was switched on or off (Fig. S1 in the Supplement). O3 (up to 600 ppbv) was generated by passing a fraction of the air flowing into the chamber over a low-pressure Hg-lamp (PenRay) that dissociated O2 (to O atoms and thus O3) at 185 nm. A known flow of isoprene entered the chamber as a dilute sample from a 12 L stainless steel storage canister (Landefeld GmbH) which was prepared manometrically from evaporation of pure isoprene (Acros 110 Organics, 98 %) and mixing with helium (5.0, Westfalen). The isoprene concentration in the storage canister was quantified indirectly by measuring the NO3 reactivity via flowtube-CRDS (Liebmann et al., 2017;Dewald et al., 2020) and was found to be 46.5 ppmv, in agreement (within 15 %) with the manometrically derived mixing ratio. A gas sample of isopropyl nitrate (Sigma Aldrich, 58 ppmv in N2 5.0, Westfalen) was prepared in a similar fashion.
Two methods of in-situ NO3 generation were employed. In the first, NO3 was produced in the chamber via the reaction of NO2 115 with O3 (R5), whereby O3 was generated as described above and NO2 was taken from a bottled sample (Air Liquide, 1 ppmv in N2). Typical concentrations of NO2 and O3 were 6-10 and 100-160 ppbv, respectively. Alternatively, NO3 was generated in the thermal decomposition of N2O5 (R11) which was eluted into the chamber by passing a regulated flow of N2 over N2O5 crystals held at temperatures between -78 and -70°C.
N2O5 was synthesised by the sequential, gas-phase oxidation of NO (5% in N2, Westfalen) in an excess of O3 (Davidson et al., 1978) and trapped at -78°C (acetone and dry ice). In this case, O3 was obtained by electrical discharge through oxygen (5.0, Westfalen) using a commercial generator (Ozomat Com, Anseros). Note that, the latter method enables us to generate NO3 in the chamber in an O3-free environment.

Detection of organic nitrates by cavity ring-down spectroscopy (CRDS) 125
Simultaneous measurements of the mixing ratios of NO2, NO3, N2O5, ΣANs and ΣPNs in the SCHARK chamber were made using a five-channel cavity ring-down spectrometer (CRDS) that has been described in detail (Sobanski et al., 2016)  The volumes in front of the mirrors are purged with dry synthetic air, which results in a reduction of the effective optical path 130 length from 90 cm to 62.1 cm (d). The standard expression Eq. (1) is used to derive in-cavity concentrations [X] from the difference in ring-down constant in the absence (k0) and presence (k) of an absorber X: where c is the speed of light and σeff is the effective cross-section derived from the overlap of the laser emission and the NO2 (Vandaele et al., 1998) or NO3 absorption spectrum (Orphal et al., 2003). 135 Three of the cavities are operated at 409 nm for detection of NO2 whereby 409 nm light is provided by a square-wavemodulated (2500 Hz) laser-diode. The three 409 nm cavities, thermostated to 303 K and typically operated at a pressure of ~ 550 Torr (733 mbar), sampled from the SCHARK at a total flow rate of 6 SLPM, which initially passes through a 2.3 m long PFA inlet (1.5 m with OD 0.25 inch. and 0.8 m with OD 0.125 inch) before being split into three equal flows. One flow is directed to a cavity via an unheated, 60 cm long PFA tube (0.375 inch OD) to measure NO2. The other two flows are directed 140 through thermal dissociation inlets (TDIs) in which PNs and ANs are converted to NO2. A TDI at temperatures close to 448 K, results in conversion of PNs to NO2 so that the cavity sampling via this inlet measures the sum of PNs + NO2. A hotter TDI (~ 650 K) results in the additional conversion of ANs to NO2 so that the sum of ANs + PNs + NO2 can be measured as described in the literature cited in the introduction. The choice of material for these TD-inlets has a profound influence on the results obtained, as described below. 145 The standard deviation (2σ) of consecutive baseline measurements define the limits of detection (LOD) which are 38, 44 and 90 pptv for [NO2], [ΣPNs] and [ΣANs] respectively under laboratory conditions. The total uncertainty for the NO2 measurement is 9 % which includes uncertainty in the (effective) NO2 cross-sections. For the measurements of ANs and PNs the associated uncertainties are highly dependent on the concentrations of other trace gases and the corrective procedure accounting for radical recombination effects (Sobanski et al., 2016).. 150 For simplicity, we refer to the three cavities as the "NO2 cavity" (room-temperature inlet), the "PNs cavity" (TD-inlet at circa 473 K in which PNs + NO2 are measured) and the "ANs cavity" (TD-inlet at circa 673 K in which ANs + PNs + NO2 are measured).
The remaining two cavities of the CRDS were operated at 662 nm (laser modulation at 625 Hz) for detection of NO3. While one cavity is thermostated to 303 K (and detects NO3 only), the second one (as well as an FEP-coated glass reactor located 155 upstream) is thermostated to 373 K so that N2O5 is stoichiometrically converted NO3 and the summed mixing ratio of NO3 and N2O5 is obtained. The two 662 nm cavities sampled air from the SCHARK at a total flow rate of 15 SLPM through a ~1.5 m ¼ in. (OD) PFA tube. Corrections to the mixing ratios were made to account for loss of NO3 and N2O5 during transport to and through the cavities. Using the method described in Sobanski et al. (2016), NO3 transmission was found to be 89 % in both https://doi.org/10.5194/amt-2021-128 Preprint. Discussion started: 7 May 2021 c Author(s) 2021. CC BY 4.0 License.
cavities. The NO3 and N2O5 measurements are not central to this study, but allowed the quantitative surveillance of NO3 (and 160 indirect N2O5) consumption by isoprene. Figure 1 shows three types of thermal dissociation inlets (TDIs) used to convert organic nitrates to NO2. In the original version of this instrument (Sobanski et al., 2016) the ΣPNs and ΣANs cavities sampled via 12 mm ID quartz TD-inlets (TDI-1), with a length of 55 cm, the first ~10 cm of which was wrapped with heating wire. In order to reduce bias caused e.g. by the reformation of the organic nitrate after its thermal dissociation, this section was filled with glass beads (Sigma-Aldrich G9268, 165 ø ~ 0.5 mm) to provide a surface for heterogeneous loss of radicals. The glass beads were supported on a 2 cm thick glass frit.
Problems associated with temperature-dependent flow resistance through these small beads and the need for an extra filter (upstream) to prevent their transport into the inlet lines when flows were temporarily reversed (e.g. during instrument shutdown), led us to switch to larger beads (Merck, ø = 3 mm) and these were used throughout this study in TDI-1. TDI-2 is made of a quartz glass tube with the same dimensions as TDI-1, but features a longer heated section (20 cm) and is free of additional 170 surfaces like glass beads or frits. TDI-2 is thus similar to many other thermal dissociation inlets described in the literature (see above). TDI-3 is constructed from a 55 cm long PFA tube (0.375 inch OD), where the first 20 cm are heated. The melting point of PFA is lower than the temperature required to thermally dissociate ANs, so TDI-3 could only be used for the measurement of PNs + NO2. The temperature of the external wall of the TD-inlets were measured with a K-type thermocouple situated at the centre of the heated section, which was insulated with mineral wool. At a flow rate of 2 SLPM and an operating 175 pressure of 550 Torr, approximate residence times in the inlets without glass-beads are 0.20 s (in TDI-2 at 650 K) and 0.13 s (TDI-3 at 450 K) when assuming a homogeneous temperature distribution equal to that measured on the outer wall of the tubing. Figure 2 shows the result of an experiment in which 150 sccm NO2 (1 ppmv in air) was flowed into the SCHARK along with 180 isoprene (7 sccm of 46.5 ppmv in He) and 24 SLPM zero air of which 5 SLPM were passed over the low-pressure Hg-lamp zero-air to generate O3 (~ 96 ppbv). NO2 was sampled (as usual) via the room-temperature PFA inlet, the PNs (473 K) and ANs cavities (673 K) both sampled via TDI-1 (quartz tube with glass-beads). O3 was added at 09:30 and NO2 at 10:00 (all times are local times, LT).

Results and Discussion
Just prior to the addition of isoprene at 12:00 LT, the system is close to steady-state with ~ 5 ppbv NO2 and 92 ppbv O3. After 185 subtraction of the measured N2O5 mixing ratios, a residual signal of ~ 100 pptv is detected in both PNs and ANs channel, which may be caused, as discussed below, by interference of HNO3 in the ANs channel and a memory effect of the glass beads in the PNs channel (section 3.2). Note that after addition of isoprene, both NO3 and N2O5 are reduced drastically (NO3 ~3 pptv, N2O5 ~ 5 pptv) and the thermal dissociation of N2O5 no longer contributes to NO2 signals in the PNs and ANs channels (Sobanski et al., 2016;Thieser et al., 2016). At 14:00 LT, the cavity sampling from the 673 K TD-inlet indicated ~ 610 pptv for the summed mixing ratio of (ΣANs + PNs), whereas the cavity sampling from the 473 K TD-inlet (PNs) indicated ~ 400 pptv. Since the signal in the ΣANs channel includes both the contribution of peroxy and alkyl nitrates this implies that only 210 pptv (34 %) of the detected products can be attributed to alkyl nitrates, which is inconsistent with the high yields (60-100 %) of ANs that result from the reaction of NO3 with isoprene (Barnes et al., 1990;Berndt and Boge, 1997;Perring et al., 2009;Rollins et al., 2009;Kwan et 195 al., 2012;Schwantes et al., 2015;IUPAC, 2017;Brownwood et al., 2021). Compared to ANs, we expect the mixing ratios of PNs (e.g. PAN, O2NOCH2C(CH3)=CHC(O)O2NO2 or MPAN) in this system to be negligible as their precursors such as O2NOCH2C(CH3)=CHCHO (Jenkin et al., 2015) or methacrolein (Kwok et al., 1996;Berndt and Boge, 1997;Schwantes et al., 2015) are oxidized only inefficiently in the dark. The formation of PNs only takes place once isoprene has been reduced to very low concentrations and secondary oxidation of the above-mentioned aldehydes by OH or NO3 leads to further acyl-200 peroxy radicals which form PNs. This is however never the case in the present experiments as isoprene is continuously flowed into the chamber and remains at a level of  11.4 ppbv. Given the high abundance of O3 and isoprene in this system, ozonolysis of the latter together with the associated formation and decomposition of Criegée intermediates to acetylperoxy radicals CH3C(O)O2 (Nguyen et al., 2016;Vansco et al., 2020) should make PAN the most important, potential contributor to a signal in the PNs cavity. However, this reaction path is a minor one and CH3C(O)O2 (and thus PAN) should be formed in negligible 205 amounts.
In order to identify the origin of the unexpectedly high ΣPNs signal when NO3 and isoprene are mixed in the dark, thermograms of the NO3 + isoprene system were recorded in an experiment where ~ 2.8 ppbv of isoprene-derived nitrates (as measured with TDI-2 at 625 K) were generated by flowing NO2 (200 sccm of 1 ppmv) and isoprene (9.8 sccm of 46.5 ppmv) in 15 SLPM dry synthetic air (with 5 SLPM over the Hg lamp for generation of ~ 150 ppbv O3). Similar to the experiment in Fig. 2, N2O5 210 mixing ratios are expected to be suppressed to a few pptv under these conditions so that its thermal dissociation (to NO2) did not contribute to the PNs and ANs signals. Using this chemical system, we simultaneously measured ISOP-NIT thermograms once steady-state established using TDI-1 (quartz, glass beads, 10 cm heated section) and TDI-2 (quartz, no glass beads, 20 cm heated section) both initially held at 703 K. Subsequently, both TDIs were cooled to ambient temperature over a period of ~ 1.75 h. The ΣANs signals from this experiment are plotted against the inlet temperature in Fig. 3a to generate the 215 ISOP-NIT thermogram. This is displayed along with an isopropyl-nitrate thermogram (iPN, red data points) measured using the same inlets under the same flow-conditions but using iPN diluted to 5.5 ppbv in dry synthetic air sampled directly through a PFA line (together with 1.5 ppbv NO2 impurity) to the instrument.
For iPN, we observe a well-defined onset of thermal dissociation at ~ 525 K with a plateau (maximum conversion) at ~ 650 K as reported previously for this setup (Sobanski et al., 2016). When measuring iPN, TDI-1 results in a slightly steeper 220 thermogram than TDI-2 in the 575-650 K range, which may be related to changes in gas-flow and heat-transfer within the inlet caused by the glass beads. Neither TD-inlet type results in dissociation to NO2 at temperatures < 500 K. In contrast, the ISOP-NIT thermograms (normalised to the signal at the plateau at 625 K of TDI-2) indicate formation of NO2 over a much broader range of temperatures (350-700 K). The effect of humidifying the air was examined in an almost identical experiment conducted with NO2 (150 sccm of 1 ppmv) 225 and isoprene (7 sccm of 46.5 ppmv) in 15 SLPM synthetic air with relative humidity (in the SCHARK) of 33.5 % at 22°C. In this case, ~ 2.3 ppbv ISOP-NIT was formed. The thermograms obtained with TDI-1 and TDI-2 under these conditions are depicted in Fig. 3b. The broad thermogram measured with TDI-1 is very similar to that obtained under dry conditions (Fig.   3a) although even at room temperature an additional NO2 signal of 500 pptv is detected. Sampling via TDI-2, yields an ISOP-NIT thermogram that has similar features to that obtained under dry conditions, although the peak at ~400 K has well defined 230 minima on both flanks and is shifted to higher temperatures. In separate experiments, humidified synthetic air and NO2 were sampled through a PFA line directly to the instrument. The thermogram (in the absence of isoprene or ISOP-NIT) using TDI-2 was recorded and can be found in the Supplement (Fig. S2) revealing that the presence of water and NO2 in the inlet is sufficient to reproduce some features displayed in Fig. 3b with TDI-2. It is well known that H2O and NO2 can react on surfaces to form HONO and HNO3 (Pitts et al., 1984;Finlayson-Pitts et al., 2003) and their formation in the SCHARK was verified in 235 section 3.1. In the presence of H2O, the efficiency of conversion of ISOP-NIT to NO2 drops to about 5% at ~ 460 K. This is much less than under dry conditions whereby 20 % conversion of ISOP-NIT was observed between 375 and 475 K (Fig. 3a).
Within a framework for surface-catalysed conversion of ISOP-NIT to NO2 presented below, this observation can be interpreted as arising from the competitive adsorption to the surface of nitrated hydroperoxides and H2O, i.e. H2O (which is vastly more abundant) reduces the surface coverage of the organic nitrate at the surface. 240 The results in Fig.2 and Fig.3 show that separate detection of ANs and PNs based on their thermal dissociation can be problematic for the NO3 + isoprene system. Identifying the cause of this and and providing potential solutions to circumvent the problem is the aim of this work. To do this, we first focus on the "dry" experiment and highlight two regions of the thermograms in which large deviations from the expected behaviour are observed.

245
Region I (T > 648 K) Figure 3a indicates that, at temperatures above 648 K (shaded region I), the behaviour of the two TDIs diverges significantly: While use of TDI-1 (glass beads) results in an increase in NO2 with increasing temperature, the use of TDI-2, leads to a decrease in the NO2 signal in the same temperature range. The increase in NO2 continues at temperatures above that required to convert ANs to NO2, which implies the presence of a NO2-containing trace-gas where the NO2-moiety is more strongly 250 bound than in ANs.
In order to assess to which extent this behaviour is potentially caused by inorganic trace-gases that are not directly related to isoprene oxidation, an experiment with only NO2 (2.75 ppbv) and O3 (146 ppbv) in 23 SLPM dry synthetic air was performed.
The steady-state concentration of N2O5 + NO3 was measured as 78 pptv. The resulting thermograms using TDI-1 and TDI-2 and after subtraction of the signal from the NO2 cavity (i.e. unheated inlet) are depicted in Fig. 4. No significant additional 255 signal is observed below 475 K (region I) in either of the inlets. In region II (T > 675 K) on the other hand, we observed an increase (by ~500 pptv) in the signal to at 703 K with TDI-1, whereas ~50 pptv are lost in TDI-2. In order to identify the tracegas(es) responsible for the signals observed in the system without isoprene, an Iodide-Chemical-Ionization Mass Spectrometer https://doi.org/10.5194/amt-2021-128 Preprint. Discussion started: 7 May 2021 c Author(s) 2021. CC BY 4.0 License.
(I-CIMS (Eger et al., 2019)) was coupled to the experiment. As shown in the Supplement (Fig. S3) both HNO3 and nitrous acid (HONO) were observed as soon as O3 and NO2 were present in the chamber and their formation is enhanced in the 260 presence of water vapour, which is a common phenomenon in Teflon chambers (Pitts et al., 1984). We also found that reversing the flows and sampling the air into the CIMS after passing through the TDI-1 or -2 (at 475 K) resulted in removal of the HNO3.
Sampling through TDI-1 also led to quantitative loss of HONO.
Various TD-CRDS and TD-LIF instruments report the detection of HNO3 as NO2 following thermal dissociation at temperatures around 700 K (Day et al., 2002;Wild et al., 2014;Thieser et al., 2016). The sensitivity of the present set-up to 265 HNO3 was investigated by sampling nitric acid from a calibrated permeation source  via TDI-1 and TDI-2 simultaneously. In these experiments, 22 ppbv HNO3 (with 780 pptv NO2 impurity) in dry synthetic air was delivered to the TDIs along with 350 ppbv O3. Figure 5 shows the temperature-dependent conversion efficiency of HNO3 to NO2 in the presence of ozone (squares) with TDI-1 (black solid symbols) and TDI-2 (open symbols). Conversion of HNO3 to NO2 starts at ~550 K and increases as the temperature is ramped to 680 K. At 680 K, the conversion efficiency is 23 % for TDI-1 and 8 270 % for TDI-2. No significant conversion of HNO3 to NO2 was observed when ozone was absent (blue datapoints) and was drastically reduced when the synthetic air was humidified to RH = 55 % at room temperature (red datapoints). The effect of water vapour is consistent with previous observations on the effect of humidity (Sobanski et al., 2016;Thieser et al., 2016;Friedrich et al., 2020).
The decomposition of HNO3 to NO2 thus only occurs in the presence of ozone under dry conditions and its rate increases 275 greatly at T > 650 K. This is consistent with the observations in Fig. 3b for TDI-1 and represents a likely explanation for the increase in signal when sampling from the SCHARK to investigate the NO3 + isoprene system. The apparently more efficient (~ factor three) conversion of HNO3 to NO2 in TDI-2 is explained by the loss of NO2 at high temperatures in TDI-2 through the reaction with O-atoms (see section 3.3). In TDI-1 this is prevented by the scavenging of O-atoms by the glass beads.
The ozone-assisted conversion of HNO3 to NO2 cannot be explained by known gas-phase processes as the reaction between 280 HNO3 and O( 3 P) (R13) has a low rate coefficient (k13 < 3 x 10 -17 cm³molecule -1 s -1 at 298 K (Burkholder et al., 2016)) and results mainly in the formation of OH and NO3 (R13). The more efficient conversion of HNO3 to NO2 in TDI-1 (with glass beads) compared to TDI-2 indicates that a surface-catalysed process involving either ozone or O( 3 P) is involved (R14).
Assuming that, in a Langmuir-Hinshelwood type process, the first step in the surface catalysed reaction is physi-adsorption of HNO3 to the surface, the strong reduction in conversion of HNO3 to NO2 under humid conditions is explained by the competitive adsorption of HNO3 and H2O, the latter favoured by its much larger concentrations. i.e. H2O drives HNO3 from the surface and thus protects it from surface reactions.

Region II (T = 350-475 K)
In region II (350-475 K, shaded area in Fig.3), instead of the near-zero signal expected in the absence of significant amounts of PNs or N2O5 we observe a monotonic increase in NO2 which is a factor of ~ 2 steeper in TDI-1 than in TDI-2. The signal 295 in TDI-1 at 475 K, where only PNs are expected to dissociate, is ~50 % of the maximum signal at 650 K. There are several potential explanations for this behaviour which include: 1) the formation and detection of thermally less stable ANs (including e.g. di-nitrates), which dissociate at lower temperatures than e.g. iPN, 2) the formation of non-acyl, isoprene-derived peroxynitrates (RO2NO2) that are sufficiently long-lived to build up to appreciable concentrations in the SCHARK, 3) chemical processes taking place in the TD-inlets that convert ISOP-NIT to NO2. Scenario 1) appears unlikely as several studies have 300 shown that the C-N bond-strength in various alkyl-nitrates is very similar (Hao et al., 1994;Wild et al., 2014). We also note that the formation of di-nitrates (in the absence of NO) only takes place when isoprene levels are very low and the firstgeneration nitrates formed in the NO3 + isoprene reaction can react with a further NO3. This can be ruled out for the present experiments in which the isoprene mixing ratio is always much larger than that of the first generation nitrates formed, which in any case react much more slowly with NO3 than does isoprene. The second explanation requires that RO2 formed in the 305 initial reaction between NO3 and isoprene react with NO2 to form RO2NO2. Given our experimental conditions, we would indeed expect that the main fate of any RO2 formed in the reaction between NO3 and isoprene is reaction with NO2, which will dominate over self-reaction or reaction with NO3, other RO2 or HO2. Non-acyl RO2NO2 are however generally highly thermally unstable, with lifetimes (at room temperature) of seconds or minutes, with respect to re-dissociation to RO2 + NO2.
For isoprene-derived RO2NO2 to contribute to the signal observed in region II would require that the RO2 -NO2 bond strength 310 be comparable to those of acyl nitrates such as PAN. The dominant 1,4 peroxy radical formed when NO3 reacts with isoprene which has a nitrate group separated by two carbon atoms from the peroxy carbon. It seems unlikely that this could have a stabilising effect on the C-N bond in RO2NO2 in the same way that an -carbonyl group does. Indeed, chamber experiments investigating the products of the NO3 + isoprene reaction in detail (Barnes et al., 1990;Wu et al., 2020) have failed to identify neither acyl-nor non-acyl-RO2NO2 as stable or semi-stable products formed from primary oxidation unambiguously. 315 In support of scenario 3, sections 3.2 to 3.4 describe the evidence for chemical reactions leading to NO2 formation that bypass the thermodynamic barrier for direct NO2 formation but are surface-catalysed, require the presence of O3 in either the SCHARK or in the inlet. These processes are peculiar to alkyl nitrates with a C=C double bond and thus have not been observed in TDinlets tested only with saturated alkyl nitrates such as the frequently used isopropyl nitrate.

The role of O3 320
To further investigate the conversion of ISOP-NITs to NO2 at low temperatures in the TD-inlets, we generated NO3 via the room-temperature thermal decomposition of N2O5 thus ruling out chemical processes that were initiated or catalysed by O3.
In these experiments, N2O5 was transported into the SCHARK by passing a flow of 0.1 SLPM dry synthetic air over a crystalline N2O5 sample cooled to -78°C with further dilution in a 15 SLPM flow of zero-air. The combined concentration of N2O5 + NO3 in the presence of ~ 22 ppbv isoprene (7 sccm of 46.5 ppmv) was measured as 40.5 pptv. The use of high isoprene 325 https://doi.org/10.5194/amt-2021-128 Preprint. Discussion started: 7 May 2021 c Author(s) 2021. CC BY 4.0 License.
concentrations guarantees that the thermal decomposition of N2O5 does not contribute significantly to the thermograms. Under these conditions several ppbv of ISOP-NIT were formed. A constant flow (2 SLPM) of zero air was passed over a low-pressure Hg-lamp and added between the sampling port of the SCHARK and the inlets and TD-inlets of the NO2, PNs and ANs cavities.
This way the O3 mixing ratio in the TD-inlets could be varied without affecting the chemistry in the chamber. Figure 6 presents the results of one such experiment in which ISOP-NIT was sampled from the SCHARK using TDI-1 and 330 TDI-2, both initially held at 703 K with thermograms obtained by decreasing the temperature of both inlets in 25 K steps. At each step, after recording the signal under O3 free conditions (black squares), a low (40-54 ppbv, green triangles), medium (97-111 ppbv, blue triangles) and high (185-219 ppbv, orange circles) mixing ratio of O3 was added in front of the TD-inlets.
To enable comparison with "ideal" behaviour, the thermogram of isopropyl nitrate (iPN, red circles) recorded from an experiment while flowing 162 ppbv O3 (and 3 ppbv NO2 impurity) through the SCHARK is also plotted. 335 Figure 6a displays a thermogram obtained when sampling ISOP-NITs from the SCHARK via TDI-1 (glass beads). In the presence of O3, the thermograms are very broad with substantial NO2 formation between 350 and 475 K (shaded region II) and in this sense are comparable to those displayed in Fig. 3, where NO3 was obtained from the reaction of NO2 with O3. In region I, the effect of going from ~50 to ~200 ppb of ozone is to increase the NO2 generated drastically. This is the opposite of that observed when sampling via TDI-2 and thus in agreement with the results of the experiment depicted in Fig. 3, where 340 the increase in signal was assigned to detection of HNO3. For the experiments in which NO3 was generated in the roomtemperature thermal dissociation of N2O5, HNO3 can arise from reactions of N2O5 with moisture on the walls and is present as impurity in the N2O5 sample. As described above, the presence of glass beads has two effects which operate in the same direction in this temperature regime: The conversion of HNO3 to NO2 is catalysed by the surface provided by the glass-beads and at the same time the loss of NO2 (via reaction with O( 3 P)) is reduced as O( 3 P) is scavenged by the glass surface. 345 For TDI-1, the thermograms obtained without ozone (black squares) differ greatly to those in which ozone was present.
Without ozone, NO2 is not generated at temperatures lower than 550 K but its concentration increases rapidly at temperatures above ~ 600 K with no indication of a plateau being reached. The thermograms obtained in this NO3-isoprene system using TDI-1 in the absence of ozone bears little resemblance to that of iPN. Furthermore, during periods without ozone or heat in TDI-1, compounds of lower volatility like ISOP-NITs or HNO3 appear to deposit on the glass beads and frit. This would form 350 an explanation for a "memory effect" observed for TDI-1, whereby after exposure to HNO3 or organic nitrates during unheated periods, an increase in the NO2 signal followed by a slow decrease taking several hours is observed as soon as pure synthetic air and ozone were added to the flow through the heated inlet. An example of this phenomenon is shown in Fig. S4 in which (at peak signal) 60 ppbv of NO2 was detected just by heating TDI-1 to 703 K in the presence of O3 in synthetic air. In order to avoid bias in results caused by this effect, a cleaning procedure was adopted prior to all experiments whereby the inlet was 355 heated to 703 K and exposed to ozone in synthetic air until a constant, low residual signal, usually between 20 and 200 pptv, was established. This memory effect seen for TDI-1 is also observed when a thermogram of ISOP-NIT (generated in a system similar to the experiment in Fig. 3) is measured by going from cold to hot temperatures (Fig. S5). In Fig. 6b we present the results of the same experiment using TDI-2. The non-zero signal (~ 250 pptv) at temperatures between ~320 and 450 K) results from instability in the baseline. In the absence of ozone, the organic nitrates generated in the NO3-360 initiated oxidation of isoprene follow a well-defined thermogram (black squares and solid line) between 475 and 650 K, which is very similar to the thermogram measured for iPN (red circles). The addition of ozone does not result in the formation of NO2 in region II, but does induce NO2 losses for temperatures above 650 K (region I). The fact that ISOP-NIT was not converted to NO2 at temperatures < 475 K suggests that the observation of a large signal in Fig. 3a (region II, white squares) is linked to O3-induced chemistry in the SCHARK which will be discussed in more detail in section 3.3. The loss of NO2 in 365 region I increases with increasing amounts of O3 with ~35 % of the NO2 formed at 625 K lost when O3 was increased to ~ 200 ppbv. The same behaviour is observed in the thermogram of iPN which confirms that this process is independent of the nature of the nitrate but solely linked to thermal decomposition of O3 and subsequent reactions of O( 3 P) with NO2.

Thermal dissociation and gas-phase reactions of O3 and O( 3 P).
An important clue to the underlying chemical process that lead to the conversion of ANs to NO2 at temperatures lower than 370 those required to break the C-N bond is the fact that the thermogram of iPN (measured with TDI-1) is not significantly affected by the presence of 163 ppbv O3 whereas thermograms of ISOP-NIT, the vast majority of which are unsaturated, are greatly broadened when O3 is present.
It is well known that O( 3 P) reacts rapidly (via electrophilic addition) to C=C double bonds (Leonori et al., 2015) and we thus assessed the potential impact of NO2 formation via reactions of O3 or O( 3 P) (formed in the thermal dissociation of O3 in the 375 TDIs) with ISOP-NIT.
The concentration of O( 3 P) in the TDIs depends on the concentration and rate of thermal decomposition of O3 and thus on the gas-temperature as well as its rate of recombination with O2, reactions with O3, NO2, isoprene, isoprene nitrates and loss to the walls: The contribution (in addition to the thermal dissociation of ISOP-NIT) to NO2 formation via reactions (R20) and (R21) in TDI-2 were assessed via numerical simulation (FACSIMILE/CHEKMAT release H010 (Curtis and Sweetenham, 1987)). The 390 rate coefficients for the most important reactions are listed in Tab. 1; the complete reaction scheme is listed in the https://doi.org/10.5194/amt-2021-128 Preprint. Discussion started: 7 May 2021 c Author(s) 2021. CC BY 4.0 License.
supplementary information (S7). Reaction times in heated and unheated section of TDI-2 were calculated from the temperature, internal diameter of the quartz tube and the flow rate. The rate coefficients for the gas-phase reactions of isoprene (IUPAC, 2017) and 2-methyl-2-butene (Herron and Huie, 1973) with O( 3 P) and O3 were used as surrogates for the reactions of ISOP-NIT for which data is not available. The temperature-dependent dissociation rate coefficient of n-propyl nitrate was used to 395 account for NO2 from the thermal dissociation of isoprene-derived nitrates (Morin and Bedjanian, 2017). Wall loss of O( 3 P) in the instrument was estimated to be 90 s -1 using the method as described in Thieser et al. (2016) and implemented in the model run.
The initial conditions for the simulation were 1 ppbv ISOP-NIT, 10 ppbv isoprene and 5 ppbv NO2 at a cavity pressure of 543 Torr and a temperature of 298 K. The results obtained are shown by black curves in Fig.7. For temperatures up to 575 K the 400 simulated thermograms with and without O3 are almost identical, whereas at higher temperatures, the amount of NO2 exiting the inlet decreases because of its reaction with O( 3 P) (R18), in broad agreement with the experiments carried out using TDI-2 as shown in Fig. 3b. The model simulations show (Fig. 7), that almost no O( 3 P) is formed by the thermal dissociation of O3 at the lower temperatures of region II. Only at higher temperatures, is a significant fraction of O3 (27 % at 703 K) converted to O( 3 P) with the majority subsequently lost at the inlet walls. These calculations underline the observation that the low 405 temperature formation of NO2 from ISOP-NIT seen when using TDI-1 cannot be explained with known gas-phase chemistry.
In summary, the experimental observations and the numerical simulations indicate that the presence of O3 is required in the inlet for TDI-1 and in the chamber for TDI-2 to generate NO2 from isoprene-derived nitrates at temperatures less than 475 K.
We have shown that the generation of NO2 from alkyl-nitrates at low temperatures using TDI-1 requires that the organic nitrate has a double-bond and that, while gas-phase reactions of O( 3 P) are responsible for the loss of NO2 at high temperatures, they 410 are NOT responsible for the conversion of isoprene-derived nitrates at lower temperatures in neither of the TD-inlets. The presence of glass beads (large surface area) favours the formation of NO2 from ISOP-NIT at low temperatures. Altogether, these observations indicate that a surface-catalysed reaction involving ozone is the process most likely to be responsible for the conversion of ISOP-NIT to NO2 at temperatures below those required for the gas-phase thermolysis.

Surface catalysed reactions with ozone 415
Quartz tubes contain impurities and surface defects that can provide reactive sites (RS) at elevated temperatures and the surface catalysed chemistry of ozone on e.g. mineral silicates is well known (Bulanin et al., 1994;Hanisch and Crowley, 2003a,b;Usher et al., 2003). O3 can be surface-catalytically converted to O2 (R25) by the formation and loss of reactive, oxygenated surface sites (RSO) via (R23) and (R24).

RS + O3
 RSO + O2 (R23) 420 In order to test for ozone loss in our setup, O3 in synthetic air was passed through the TDIs. The O3 mixing ratio prior to entering the inlets was measured continuously using the ozone monitor (UV absorption). The ozone exiting the TD-inlets was converted to NO2 (by addition of 1 ppmv NO (R4) in a 1.5 m long PFA tubing with 0.5 inch OD and a residence time of 5.2 425 s) and then measured in the 409 nm cavities. Figure 8a shows the results of such an experiment for TDI-1. The concentration of ozone before entering the inlets was 20.5 ppbv (open dots) which was detected as 17.1 ppbv after passing through the inlet at 298 K (black squares, 11:15 to 11:55). This represents a conversion efficency of 0.83 which matches exactly that expected when considering the reaction time, NO concentration and the rate coefficient (k4 = 1.9 x 10 -14 cm³ molecule -1 s -1 , IUPAC, 2021). 430 At  12:00 LT, upon heating the inlet to 473 K the ozone exiting TDI-1 was depleted by up to 27 %, while heating to 573 K results in further loss (up to 40 %). An analogous experiment using TDI-2 (Fig. 8b) showed that while some O3 was also lost (e.g. 4 % at 573 K) this was much less than in TDI-1. As the gas-phase thermal decomposition of O3 is negligible under these conditions (0.6 % at 575 K, Fig. 7) the loss of O3 when passing through the TD-inlets indicates that surface catalysed ozone decomposition takes place (R23 and R24), especially in TDI-1 where larger surface areas are available. 435 We now consider the possibility that the conversion of isoprene-derive nitrates to NO2 can be catalysed by surfaces in the presence of O3. We note that previous work has shown that the heterogeneous ozonolysis of alkenes on glass or other surfaces can be more efficient than its analogous, gas-phase process (Dubowski et al., 2004;Stokes et al., 2008;Ray et al., 2013) and now consider the possibility that RSO is the mediating reactive species in TDI-1 in our experiments.
A hypothetical ISOP-NIT degradation scheme involving the initial attachment of RS-O to the remaining double bond is given in Fig. 9. We consider only the fate of the most stable surface-adducts, i.e. tertiary radical in case of δ-products and secondary radicals in case of β-products. Both radical-adducts will react with O2 to form organic peroxy radicals which may undergo Hshifts (via five-or six-membered rings) resulting in formation of a radical with its unpaired electron in the direct vicinity of 450 the nitrate functionality. Such unimolecular processes may become competitive to bimolecular reactions under atmospheric conditions (Møller et al., 2019). A possible fate of this radical is decomposition to form a carbonyl compound via NO2 elimination (Hjorth et al., 1990;Berndt and Boge, 1995;Vereecken et al., 2021). For δ-products, the tertiary product may also eliminate NO2 under the formation of an epoxide.
With TDI-2, the conversion of ISOP-NIT to NO2 at low temperatures (region I) is only observed when O3 is present in the 455 chamber (Fig. 3a), but not when it is added only to the inlet (Fig. 6b) implying that the presence of O3 as an oxidizing agent in the SCHARK is the main difference between these two reaction systems. This suggests that the ozonolysis of isoprene may play an important role. As the reaction between isoprene and O3 leads to the formation of OH and additional HO2 (Zhang et https://doi.org/10.5194/amt-2021-128 Preprint. Discussion started: 7 May 2021 c Author(s) 2021. CC BY 4.0 License. al., 2002;Malkin et al., 2010;Cox et al., 2020), the fraction of RO2 reacting with HO2 in this system is enhanced when compared to the experiments in which NO3 was generated from N2O5. The major product from the reaction between nitrated 460 -and β-peroxy radicals with HO2 are hydroperoxides, such as O2NOCH2C(CH3)=CHCH2OOH (Schwantes et al., 2015). With the intention of assessing the effect of O3 on hydroperoxide yields, further model calculations were performed using the Framework for 0-D Atmospheric Modelling (F0AM) (Wolfe et al., 2016). The NO3 isoprene oxidation scheme is still subject of current research and uncertain which is why the Master Chemical Mechanism (MCM, version 3.3.1, http://mcm.leeds.ac.uk/MCM) (Jenkin et al., 2015) as well as the Reduced Caltech Isoprene Mechanism Plus (RCIMP, version 465 5, https://data.caltech.edu/records/247) (Wennberg et al., 2018;Bates and Jacob, 2019) were applied. Figure 10 shows the calculated fraction of hydroperoxides to ISOP-NIT using both models and mixing 22 ppbv isoprene with either 10.8 ppbv NO2 and 150 ppbv O3 (runs 1 and 3) or 3 ppbv N2O5 (runs 2 and 4) to achieve NO3 in a flow-through chamber with an exchange rate of 2.73 x 10 -4 s -1 . Generally, the MCM predicts higher hydroperoxide yields than RCIMP. A possible reason might be that MCM has higher rates of HO2 reformation from reactions of isoprene-derived alkoxy radicals than RCIMP does. For that 470 reason, MCM only predicts an increase in the hydroperoxide yield by a factor of 1.33 when the NO3 source is changed from N2O5 to NO2 + O3 (model runs 1 and 2), whereas RCIMP predicts a factor of 2.56 (model runs 3 and 4). The calculations thus broadly support the hypothesis that unsaturated, nitrated hydroperoxides are involved in the low temperture (surface catalysed) formation of NO2 from ISOP-NIT. Indeed, hydroperoxides not only have high affinity for surfaces but also have a rather weak O-O bond, and several routes to surface catalysed elimination of NO2 appear feasible. Decomposition of isoprene-derived 475 hydroperoxides within the sampling line has been observed for other instruments (Rivera-Rios et al., 2014). A possible degradation mechanism for an isoprene-derived hydroperoxide which is abundant from the NO3 + isoprene system (Schwantes et al., 2015) is depicted in Fig.11. In this case, O2NOCH2C(CH3)=CHCH2OOH coordinates via hydrogen bonds to the SiO2 surface prior to cleavage of the O-O bond to form OH and an alkoxy radical. The latter may dissociate via unimolecular reactions to form closed-shell products under elimination of NO2 (Wennberg et al., 2018;Vereecken et al., 2021). 480

Elimination of O3 and surface catalysed conversion of ANs at low temperatures
Our findings clearly show that that the combination of heated quartz surfaces and ozone catalyse the decomposition of the unsaturated nitrates formed in the reaction between NO3 and isoprene at temperatures below 473 K. While the exemplary, surface catalysed processes depicted in Fig.9 and Fig.11 fulfil the requirement of conversion of ISOP-NIT to NO2 we stress that these are purely hypothetical and we cannot state with any certainty that they are the reactions responsible for our 485 observations. While it would be highly interesting to investigate such processes using different (e.g. surface sensitive) techniques this is clearly beyond the scope of this paper and of the experimental capabilities of this research project. Instead, we take the more pragmatic approach and indicate potential methods to eliminate such unwanted reactions when using TDinlets.
In principal, as the process that convert ISOP-NIT to NO2 are clearly surface catalysed (involving formation of RSO) their 490 impact can be reduced by using a surface that does not support formation of RSO. We therefore tested a third TD-inlet (TDI-https://doi.org/10.5194/amt-2021-128 Preprint. Discussion started: 7 May 2021 c Author(s) 2021. CC BY 4.0 License.
3), consisting of 55 cm PFA tubing (0.375 in. OD) with a 20 cm heated section. PFA-tubing has been routinely used as TDI for measurement of e.g. PAN, as it is relatively unreactive to the peroxy radicals formed (Phillips et al., 2013). The C-F-bond of PFA is very strong and nonpolar which should reduce the formation of RSO as well as adsorption of ISOP-NIT to the surface. The performance of a TDI made of PFA (TDI-3) was examined by performing an experiment with the SCHARK 495 analogous to the one shown in Fig. 2. The resulting time-series of NO2, ΣPNs (using TDI-3 heated to 448 K) and ΣANs (using TDI-2 heated to 648 K) are depicted in Fig.12. At first, O3 (5 SLPM synthetic air passed over a Hg lamp) and NO2 (200 sccm of 1 ppmv) in 25 SLPM dry synthetic air were constantly introduced into the SCHARK. After detectable amounts of NO3 and N2O5 had been formed, isoprene (9.8 sccm of 46.5 ppmv) was added leading (as expected) to almost quantitative depletion of NO3 and N2O5. 500 During the following 3 hours the signal in the ANs cavity (TDI-2) increased to ~1.2 ppbv, while the signal in the PNs channel (~ 40 pptv) was close to the detection limit. A CIMS measurement obtained during this experiment validates that PAN mixing ratios are lower than 50 pptv. This result already confirms that 1) the ANs derived from NO3 + isoprene reaction are not detected at T < 448 K when PFA is used instead of quartz and 2) as expected, there is no significant generation of peroxy nitrates in these experiments which would be detectable with both ovens at T = 448 K (see Fig. S6 in the Supplement). 505 This result not only confirms that the previous detection of ISOP-NITs at low TD-inlet temperatures when using TDI-2 and especially TDI-1 were caused by heterogeneous reactions at the quartz surface, but also provides a solution to the problem of the separate measurement of ANs and PNs using TD-inlets.
We recall however, that the reason for adding glass beads to the inlet was to suppress recombination reactions of NO2 with peroxy radicals by providing a surface to scavenge the latter. The use of a PFA tube rather than quartz will certainly exacerbate 510 this effect, as the rate of loss of RO2 to PFA surfaces is expected to be lower than on quartz (Wooldridge et al., 2010). This implies that corrective procedures based on numerical simulation may be necessary in some environments as shown by Thieser et al. (2016) and this may limit the useful deployment of the method to regions where NOX levels are sufficiently low that reaction of RO2 with NO and NO2 become insignificant.

Summary, Conclusions and implications for atmospheric measurements of PNs and ANs 515
We have shown that the detection of isoprene-derived organic nitrates via its thermal dissociation in quartz / glass inlets can (in the presence of O3) be accompanied by undesirable side reactions which broaden the thermograms and thus impede the separation of PNs and ANs signals by sampling through TD-inlets at different temperatures as is commonly practised. While our experiments deal with the nitrates formed in the NO3-initiated oxidation of isoprene, it is very likely that similar broadening of thermograms would also occur with nitrates formed from the oxidation of other terpenoids, as some organic nitrates derived 520 from nighttime oxidation of e.g. limonene still contain a double bond (Fry et al., 2011) and/or contain hydroperoxy groups.
Specifically, we find that the presence of O3 in either the quartz TD-inlet or in a Teflon simulation chamber results in the generation of NO2 from isoprene-derived nitrates in TD-inlets made of quartz at temperatures less than 475 K and that this https://doi.org/10.5194/amt-2021-128 Preprint. Discussion started: 7 May 2021 c Author(s) 2021. CC BY 4.0 License.
only occurs when the organic nitrate has a double-bond (or features a hydroperoxy group). The formation of NO2 from ISOP-NIT was accelerated in the presence of glass-beads which indicates that a surface-catalysed reaction involving ozone and 525 reactive surface sites is the process most likely to be responsible both for the conversion of ISOP-NIT to NO2 at low temperatures (375-475 K) and the conversion of HNO3 to NO2 at high temperatures (> 550 K). By avoiding the use of O3 or using a non-quartz TD-inlet, we were able to show that the ISOP-NIT thermogram is entirely consistent with those of saturated alkyl nitrates.
We show that surface-catalysed reactions on quartz TD-inlets involving O3 represent a potential source of bias in measurements 530 of ANs and PNs during field observations, especially when isoprene is abundant. For example, we previously reported results from two campaigns carried out using the TD-CRDS system described here on the same rural mountain site (Kleiner Feldberg) and season (but in different years). We found that the average, relative abundance of PNs and ANs was quite different with ANs ≈ PNs during the PARADE campaign in 2011 and PNs > ANs during the NOTOMO campaign in 2015 . In 2011, the TD-inlet was a quartz tube (i.e. similar to TDI-2) (Thieser et al., 2016), whereas in 535 2015 the inlet contained glass beads (i.e. similar to TDI-1) (Sobanski et al., 2016). With our present understanding of the role of surfaces and O3 in TD-inlets, we cannot rule out that the observations during 2015 were biased to lower values for ANs and higher values for PNs, although, as discussed by  there are other, meteorological factors which would have contributed.
For the detection of PNs we have shown that (at the lower temperature required to thermally dissociate PNs to NO2) surface 540 catalytic effects that convert ANs (or other species) to NO2 can be completely avoided by using a TD-inlet made of a nonreactive material like PFA (TDI-3). In this case, O3 does not appear to have any impact.
When using a quartz TD-inlet for conversion of ANs + PNs to NO2 at higher temperatures the surface reactions that shift thermograms to lower temperatures are of less significance as, in any case, the role of the TD-inlet is to convert all ANs and all PNs to NO2. However, in order to avoid detection of HNO3 or HONO other materials may be more suitable than quartz. 545 Sapphire, commonly used in microwave discharge generated plasmas owing to its high purity and non-reactive surface, may represent a useful alternative. Under humid conditions, some of the observed interferences become negligible: No HNO3 appears to interfere with the ANs measurement and the interference of ANs to the PNs measurement is reduced in TDI-2, but not in TDI-1.
This study emphasizes the importance of characterising thermal dissociation inlets under conditions which are similar to those 550 found in the atmosphere. We recognise that the impact of surface catalytic processes will vary from one inlet to the next (even if made from the same material) and all quartz-TDIs will not necessarily exhibit the same degree of conversion of BVOCderived ANs at low temperature. For this reason, thermograms should be measured using trace-gases that are abundant in the atmosphere and the effect of e.g. O3 and water vapour should be thoroughly investigated.
Data availability. The data underlying the figures is available on request from the corresponding author.
Author contributions. PD conducted the experiments, analysed the data and wrote the manuscript. RD was responsible for the CIMS measurements. JS designed and built the SCHARK. JNC designed the experiments and together with JL contributed to 560 the manuscript.
Competing interests. The authors declare that they have no conflict of interest.