Two sets of tests were performed to verify the sensitivity of the LGR CRDS to NO2. Response at high concentrations was verified at concentrations of 250 to 1000 ppb using dilutions of NO2 in zero air. A 514.5 ppm calibrated mixture of NO2 in N2 (Airgas) was diluted with a zero-air source to generate the required mixing ratios. The response at low concentrations was compared to a thermo chemiluminescent NOx detector between 1.5 and 11.5 ppb. Low-concentration NO2 was obtained using ambient lab air diluted using zero air. The results of these experiments are shown in Fig. 3.

The fit line has a slope of close to 1 over both measured ranges, indicating good agreement under both high- and low-concentration conditions. Since this experiment was performed using dilutions of zero air, any interference from NOy species in the CL detector would also be expected to scale with dilution. The urban location of the lab would support the relatively low levels of NOy compared to NO2, explaining the very small 4 % difference in slope between the two detectors. These experiments suggest an upper limit error due to the NO2 detection of 10 %. The intercept offset of the low-concentration experiment is 0.64 ppb, which may be attributable to the interference of organic nitrates in the chemiluminescence measurement or a slight zero offset in the chemiluminescence detector. Thus, the CRDS is accurate under both atmospherically relevant and elevated laboratory experiment conditions, but regular calibration against a known source or comparison with another NO2 measurement is nevertheless recommended.

The 400 L Teflon bag Reed Environmental Chamber (REC; Draper et al., 2015) was used to generate VOC + NO3 reaction products that could be analyzed using the TD-CRDS. The REC chamber was operated with steady inlet flows to the top of the chamber of zero air (4.3 L min-1), O3 (200 sccm), NO2 (4.4 sccm of 515 ppm), and VOC (14.2 sccm zero air through chilled liquid source containing gas-phase VOC of ∼100 ppm), which mix and react (average residence time ∼90 min) and are sampled for analysis from the bottom of the chamber. Zero air was generated using a Sabio Model 1001 zero-air generator, which removes water, particulates, and reactive gases. Ozone was generated using a UV light source (Pen-Ray Hg lamp at 254 nm) inside the middle neck of a three-necked round bottom flask, and the concentration was altered by adjusting the depth of the light source in the flask. The constant NO2 source was a gas cylinder (Airgas, concentration analyzed 17 April 2013) with a concentration of 514.5 (±2 %) ppm NO2 in N2. Approximately 300 ppb VOCs (typical VOCs used are Δ-carene, limonene, α-pinene, or β-pinene) were generated by flowing zero air over a chilled liquid sample of VOC in a three-necked round bottom flask.

Ozone, zero-air, and NO2 flows were allowed to stabilize inside the chamber prior to introducing VOC flow to initiate the experiment. All flows were then continuous until the completion of the experiment. Particle number and size data were collected using a scanning electron mobility sizer (SEMS; Brechtel Manufacturing, Inc.), connected via conductive silicone tubing to minimize particle losses. Ozone concentration was measured using a Dasibi Model 1003-AH ozone monitor or Teledyne Model T400.

Temperature ramps were performed on different mixtures of known gases to determine the appropriate set point temperatures for each of the three ovens. Temperature ramp results were used to identify the correct set points for each oven to achieve complete dissociation for each species. Both HNO3 and AN measurements were performed by flowing zero air over a pure liquid analyte sample. The AN standard used was isobutyl nitrate (Aldrich 96 % purity). A pure liquid sample of PN could not be obtained, so a NO3+Δ-3-carene mixture containing PNs was synthesized in the chamber as described above. Because concentrations were very stable, ramps were performed at 5 ∘C min-1. Normalized measured NO2 concentrations are plotted against temperature in Fig. 4 because absolute concentrations were different for each class of nitrate.

Complete dissociation of PNs occurred at a thermocouple temperature reading of 130 ∘C, ANs at 385 ∘C, and HNO3 at 600 ∘C. The HNO3 oven set point was chosen to be 700 ∘C to allow the quantification of interference from NO3 dissociated from N2O5 in that channel. At 600 ∘C, HNO3 is completely dissociated, but there is only partial conversion of NO3 to NO2, creating an interference in the hot channel in the TD-CRDS. At 700 ∘C, HNO3 is completely dissociated and NO3 is completely converted to NO2. Note that the dissociation plateaus do not overlap with the beginning of the adjacent curve, confirming the ability to quantitatively separate nitrate species by temperature.

High-concentration Δ-3-carene nitrate oxidation experiments in the REC chamber typically had 650 ppb O3 and 400 ppb NO2. When high concentrations of O3 and NO2 are present, they react in the chamber to form N2O5. This was verified by performing a temperature ramp from a chamber at low and high oxidant concentrations (Fig. 5). N2O5 dissociates to produce two NO2 products (see Reactions R3 and R4 below) across a broad temperature range, in contrast to the sharp dissociation curves for peroxy- or alkyl-nitrates, such that the presence of N2O5 removes the clear plateau between PNs dissociation and ANs and gives an interference in both PN and AN channels. A chamber with low-NOx conditions (335 ppb O3 and ∼3 ppb NO2) that was left to equilibrate for 56 min after the addition of Δ-carene gave a maximum N2O5 concentration of 1.6 ppb. The resulting temperature ramp gives the expected dissociation curve, showing both PN and AN plateaus (Fig. 5). In this case, there is good separation between PNs and ANs because N2O5 is lower in concentration. This N2O5 interference has been previously observed by Womack et al. (2017). Note that given the gradual dissociation of N2O5 across this full temperature range, the extent of the interference depends on the exact temperature set points, so any similar TD-based organonitrate instrument that may be operated in high-N2O5 conditions should characterize its individual N2O5 interference.

To measure the N2O5 interference such that it can be corrected for, we ran an experiment with only oxidants in the chamber (Fig. 6). The TD-CRDS detects one NO2 molecule from the first dissociation of NO2 from N2O5 (Reaction R3) either in Oven 3 (the PN channel) or in Oven 2 (the AN channel), and another NO2 is observed when the released NO3 fragment further dissociates in the HNO3 channel (Reaction R4). We note that due to its high reactivity and wall losses (especially the NO3 fragment), as well as the likelihood that some N2O5 remained incompletely dissociated even at the AN oven temperature, the total N2O5 detection is substantially less than 100 % of the N2O5 concentration present in the chamber. We also emphasize that these percentages are specific to the configuration used in this characterization experiment: from the chamber containing the modeled N2O5 concentration used to determine these interference percentages, a 2 m Teflon inlet line led to the TD-CRDS instrument. A kinetic model paired with measurements of NO2 and O3 in order to predict N2O5 can be used to quantify the interferences in each channel for a given setup. R3N2O5→NO2+NO3R4NO3→NO2+O The result of N2O5 is an elevated baseline in each of the PN, AN, and HNO3 channels before the VOC is added. If an accurate N2O5 measurement is available, the interference from N2O5 (and NO3) can be subtracted from each channel, and these pre-VOC injection signals can be used to assess the likely lower inlet transmission of N2O5 and NO3 vs. the more stable PNs, ANs, and HNO3. In the absence of a separate N2O5 measurement, kinetic modeling can be used to predict how much N2O5 will be formed in each experiment, which can then be subtracted. For example, in chamber experiments, comparing modeled N2O5 amounts to the amounts of signal in the AN, PN, and HNO3 channels before VOC is added can quantify what fraction of N2O5 appears in each channel. If Reaction (R3) happens across the PN and AN temperature range (130–385 ∘C) and Reaction (R4) between the AN and HNO3 range (385–700 ∘C), the sum of the signals from the AN and PN channels before the addition of VOC should be equivalent to the N2O5 signal from the HNO3 channel (from the NO3 fragment of the N2O5 dissociating to NO2). For the instrument application shown here, operating at UC Irvine in September 2019, 7 % of the modeled N2O5 produced based on a model constrained to measured NO2 and O3 is detected in the PN channel at 150 ∘C and 28 % in the AN channel at 385 ∘C.

The activated carbon denuder was tested for efficient removal of gas-phase molecules by flowing gas mixtures of single molecules diluted in zero air through the denuder. Gas mixtures were tested at several concentrations to determine if efficiency was concentration-dependent. Transmission of the denuder is defined to be the percentage of gas-phase molecules that passed through the denuder and were detected downstream when all should have been removed.

NO2 transmission was tested in 2016 at two relatively low concentrations, to mimic atmospheric conditions, and one higher concentration to mimic chamber conditions. In all cases at this time, more than 96 % of the NO2 was absorbed (Table 1). NO2 concentrations ranged from 26 to 271 ppb. In 2019, NO2 transmission was again tested to assess changes in denuder performance over time, at 275 ppb. The larger observed NO2 transmission in 2018 suggests a drift in the gas-phase breakthrough over time.  Because transmission appears to change over time, we recommend making periodic measurements and updating correction factors accordingly. Denuders can be cleaned by gentle heating and zero-air flow.

The same process was used to determine the transmission of isobutyl nitrate (an alkyl nitrate) in 2016 (Table 2). The outlet of the denuder was connected to both Channel 1 (temporarily at 520 ∘C) and Channel 2 (385 ∘C). The transmission of the denuder was not dependent on the concentration of gas in the original gas mixture or on which oven was used. This AN transmission was also retested in 2019, and in this case, no significant change in breakthrough was observed. These measured fractions of gas-phase breakthrough can be used to correct the aerosol measurements made in the denuded channels of the instrument cycle (see discussion of corrections below).

A chamber experiment with Δ-carene was performed to generate organic aerosol particles in order to test the aerosol throughput of the denuder. Low-NOx chamber conditions (450 ppb O3, 3 ppb NO2) were used to minimize potential N2O5 interferences. First, the chamber was hooked directly to the SEMS in order to get a background measure for the number of particles in the bag. Then the chamber mixture was pulled through the TD-CRDS inlet tubing while bypassing the denuder in order to quantify particle losses to the tubing. Finally, the chamber mixture was sampled while flowing through the tubing and the denuder to get the total particle loss through the instrument. The time series of this experiment is shown in Fig. 7. In order to quantify the efficiency of the denuder and tubing inlet of the TD-CRDS, the particle volume was averaged over the sampling time under each condition. This assumes particle concentration in the chamber was constant over the course of the entire experiment. Since the chamber had been running for 24 h prior to measurements, it is reasonable to assume that all concentrations had reached equilibrium.

A total of 28 % of the aerosol particles (assessed by volume) that flow into the instrument was lost to the tubing and the denuder before detection. There does not appear to be any bias toward removing smaller or larger particles. The denuder is responsible for only 10 % of total particle loss. This suggests that every deployment of this instrument should carefully consider and, if possible, quantify inlet line losses.

The CRDS can be set to zero automatically at regular intervals, which is accomplished by diverting inlet air through an NO2 scrubber. The instrument is typically set to run its 3 min zero every 2 h. The re-zeroing procedure results in small changes to the baseline before and after zeroing events. On ambient measurements, these changes are typically less than 0.5 ppb (see Fig. S3), and on zero air, they are typically less than 0.2 ppb and are sometimes positive and sometimes negative. We determine the standard deviation of 4 h of zero measurements (0.16 ppb) to estimate our blank error, σzero.

From this observed blank error, the detection limit of the instrument (LOD=3σ) can be calculated for each channel. The error for the NO2 channel is based only on σzero alone, since no subtraction is required (3σzero=3×0.16ppb=0.48 ppb). For all other channels, the error in the subtracted value A-B is calculated as 3σA-B=σA2+σB2, where σA=σB=0.16 ppb are the errors in the pre-subtraction NO2 concentration measurements. Thus, the estimated detection limit for the subtracted channels (ANs, PNs, and HNO3) 3σA-B=3×0.22ppb=0.66 ppb.

Modeling of the ovens can be employed to simulate the dissociation and recombination of the detected species in any oven design. Pressure- and temperature-dependent rate constants for dissociation (Day et al., 2002) and recombination (NASA/JPL Data Evaluation, 2015) reactions of PNs, ANs, and HNO3 were used (see Table S1 in the Supplement), alongside an assumed (a) step function or (b) linear rate of cooling from the heated to the unheated portions of the oven (see Fig. S4). We also included the IUPAC rate constant for a representative RO+O2 (7.2×10-14 e-1080/T, IUPAC), and OH wall loss rate (calculated to be 46 s-1 for these conditions) from Knopf et al. (2015). Based on these rate constants and the assumption that recombination or wall losses are the only fates for dissociated radicals, we found that the PN measurement would be the most affected by recombination. We found an expected 10 % difference in the amount of PNs recombined by the end of the PN oven between assuming linear cooling and step function, so the more conservative step function assumption can be used to provide a lower-limit concentration.

All ovens were modeled at their set point temperature, which is maintained by the thermocouple relay. However, each oven surely has gradients in temperature along its length, resulting in this average oven temperature measured at its midpoint (see Fig. 1b). As an example, for HNO3 there was very little difference in dissociation based on small changes in oven temperature. HNO3 is 100 % dissociated at the end of the oven, so modeling at 30 ∘C hotter than the set point temperature just extends the cooling region slightly. This leads to approximately 0.3 % less recombination than the set point temperature model. The same concept applies to the PN oven, but the recombination difference is larger (1 %–2 % depending on model molecule) due to a 30 ∘C increase in temperature being a larger percentage of the total temperature.

The PN dissociation and recombination oven model results in Fig. S5 show a predicted 55.9 % dissociation in the step function model and 65.4 % dissociation in the linear function model with no background NO2 and 10 ppb initial PN concentration.

In the PN oven, the only important reactions modeled were the dissociation and recombination of PNs. The background concentration of NO2 was considered and was found to have a significant impact on the recombination rate, especially at high concentrations. Figure S6 shows the percent PNs that remain dissociated at the detector as a function of initial concentrations of PNs and NO2. Two separate types of PN were considered in the modeling, due to their slightly different rates of recombination. Methyl PN gave 58.1 % detection at 10 ppb initial concentration and no background NO2, while ethyl PN gave 55.9 % detection under the same conditions. Given the relatively small difference in recombination percentages, no effort was made to incorporate both species into the model. Ethyl PN was chosen as the representative species because it was assumed that most PNs being encountered would be two carbons or larger. Including reasonable atmospheric concentrations of OH (4×106 molecules cm-3) in the model made no difference to the percent recombination of ethyl PNs and was therefore left out. We note that the PN measurement will be most affected by recombination but that this recombination can in principle be corrected for.

In addition to the potential reduction in PN signal due to recombination reactions, there is the potential for a spurious overestimation of PN signal due to reactions of thermally dissociated peroxy or peroxy acetyl radicals with ambient NO in the presence of O2, producing additional NO2 (Thieser et al., 2016). This effect will be minimal in chamber simulations of nighttime chemistry, where the mixing ratio of NO is zero but should be considered in any daytime field deployments.

In the AN oven, in addition to the major reactions of AN dissociation and recombination, the reaction RO+O2 is also important. The RO+O2 reaction is extremely fast at high temperatures (see Table S1) like those found in the heated portion of the AN oven, and we assume the reaction to be irreversible. Because O2 is abundant, the reaction negligibly affects O2 concentration. As a result, these assumptions give a model prediction of 100 % detection of alkyl nitrates at all initial AN and NO2 concentrations.

In the HNO3 oven, in addition to the dissociation and recombination of HNO3, the loss of OH radical to the walls is significant, competing with recombination. The model assumes that any OH that hits the walls after the heating part of the oven is lost due to reactions with the walls; as a result, recombination is generally less of an effect on the HNO3 measurement. Figure S7 shows example model outputs for the HNO3 oven, predicting the percent dissociation of HNO3 at the point of detection over a large range of initial concentrations for both HNO3 and NO2. As expected, recombination is most important at larger NO2 and HNO3 concentrations; below 50 ppb of each, for this instrument configuration the detection efficiency is above 80 %.

One additional reaction that can affect the HNO3 measurement is the pyrolysis of O3. At high temperatures, some fraction of O3 dissociates, releasing atomic O which reacts with NO2 to form NO+O2, which results in NO2 being removed from the final measurement. Therefore, in background conditions of high O3 concentration, the NO2 concentrations measured after the HNO3 oven are biased low and can even cause the [HNO3] to appear negative upon subtraction. Day et al. (2002) noted that at or above 530 ∘C, all O3 will separate into O2 and O molecules, which will then react with NO2. This suggests that for this instrument, the pyrolysis of O3 will result in a lower signal only in Oven 1 (700 ∘C) due to the high temperatures.

In some experiments from the 2018 Simulation of Atmospheric Photochemistry in a Large Reaction Chamber (SAPHIR) NO3ISOP campaign, HNO3 measurements appeared negative due to lower signals from the hottest channel. Using other available instruments' measurements of O3 and HNO3, we determined that approximately 4 % of the O3 signal was converted to this apparent negative HNO3 signal during one experiment (on 8 August 2018). However, this fraction did not appear consistent across experiments, perhaps due to substantial HNO3 inlet losses, and we did not determine a robust and consistent correction factor for this effect. Given that this does not affect alkyl nitrate measurements and that there were other measurements of HNO3 available, we did not pursue this further. But in principle, this is a relatively modest effect that can be corrected for after experimentally determining the efficiency of ozone pyrolysis for a particular inlet oven build.

The above modeled 100 % efficiency in detecting ANs is fortunate, since the AN measurement has thus far been the output of greatest interest from this instrument. Should one wish to use such an instrument for accurate measurements of PNs and HNO3, this too is possible, but it requires the determination of correction factors to account for the recombination in those ovens.

Beyond the correction factors for radical recombination in the cooling region after each oven (1), additional corrections that can be applied are (2) an oven-specific denuder breakthrough, based on data such as that shown in Tables 1 and 2, (3) background corrections, to account for any background signal detected in each channel while sampling zero air (this could account for inlet and/or denuder off-gassing), (4) subtraction of N2O5 interference, as described in Sect. 3.4 above, and (5) correction of the HNO3 channel for O3 pyrolysis loss of NO2, as described in Sect. 3.8 above. The importance of each of these corrections will depend on the nature of the experiments conducted; some example applications are shown below to illustrate this.

We have implemented each of these corrections as optional to apply to any raw data collected in our Igor-based data workup routine, which also sorts the data from the various ovens, averages, and subtracts the relevant signals.

The TD-CRDS was compared to the CU Boulder Jimenez group aerosol mass spectrometer (AMS) during collaborative chamber experiments in summer 2015, using the data from the denuded AN channel of the TD-CRDS and the high-resolution AMS organic-nitrate (pRONO2) measurement to assess the correlation of these two aerosol-phase organic-nitrate measurements. The experiments plotted here are those which showed substantial aerosol nitrate formation using Δ-carene or α-pinene as a VOC precursor and NO3 from an N2O5 trap, both spanning the nominal range of 10–100 ppb, at varying relative humidity. The comparison of individual measurements across 2 weeks of experiments shows significant scatter, but an orthogonal distance regression (ODR) fit to the scatterplot of TD-CRDS data vs. AMS data shows a slope of about 0.88–0.94 (depending on intercept treatment), and R2=0.73 (Fig. 8).

The AMS organic-nitrate concentrations in Fig. 9 were calculated by apportioning the total nitrate concentration using the NOx+ ion ratio (NO2+/NO+) method (Farmer et al., 2010), where the relative ratios of organic to inorganic NOx+ ratios (“ratio of ratios”, Fry et al., 2013) were determined by the average of several dry, unseeded experiments and ammonium nitrate ratios from offline calibrations (3.12 for Δ-carene, 3.78 for α-pinene). The organic–inorganic separation was conducted in order to account for possible NH4NO3 or particle HNO3 formation as was suggested by substantial shifts in NOx+ ratios observed during wet, seeded experiments, as has been reported previously (Takeuchi and Ng, 2019). Figure S8 shows a comparison of the Fig. 8 results to a plot of the TD-CRDS measurements against the AMS total nitrate (unapportioned), the latter resulting in slightly lower slopes and correlation coefficients.

Previous comparisons between AMS and thermal-dissociation-based aerosol organic-nitrate instruments have found varying agreement for ambient measurements (Ng et al., 2017). Some of these differences could be due to the fact that the ambient atmosphere contains a mix of diverse products from the oxidation of monoterpenes and isoprene in the presence of other gases; the resulting differing mixes of alkyl nitrate structures could alter the sensitivity of one or both instruments.

During 1 week in November 2014, the TD-CRDS inlet was situated outside the south end of the Reed College Chemistry building. Simultaneous measurements of NO2, PNs, ANs, and HNO3 were made, and 1 representative day is shown in Fig. 9, illustrating typical measurable ambient variability and diurnal cycle.

The TD-CRDS was also used in the month-long SAPHIR NO3+ isoprene campaign in the summer of 2018. SAPHIR is a 270 m3 double-walled Teflon chamber with movable shutters allowing for the simulation of both daytime and nighttime chemistry. The experiments were run in batch mode with periodic injections of oxidants and reactants. The reactant concentrations were comparable to real atmospheric concentrations of NO2, O3, and isoprene. Some experiments were run under humid conditions and some had seed aerosol added to facilitate condensations of gas products into the particle phase.

The low, near-ambient concentrations of reactants used, the small degree of partitioning of isoprene nitrates to the aerosol phase, and the relatively long inlet line required resulted in the aerosol organonitrate products being lower than the limit of detection of the TD-CRDS for the particle-phase AN monitoring. The gas-phase AN measurements from the TD-CRDS ranged from below parts per billion up to 16 ppb of organic nitrates, with an observed alkyl nitrate molar yield for NO3+ isoprene of ∼100 % under all explored reaction conditions. In order to determine gas–aerosol partitioning of nitrates, the gas-phase ANs measured by TD-CRDS were compared to AMS organic-nitrate aerosol measurement. These results are the subject of a forthcoming paper (Brownwood et al., 2020).

The TD-CRDS was also used for chamber experiments throughout the 2018–2019 academic year at the Reed Environmental Chamber (REC), running experiments similar to those from SAPHIR but at substantially higher concentrations. These experiments aimed to determine whether gas–particle partitioning coefficients (Kp) for the NO3-initiated oxidation of isoprene would be similar in a 0.4 m3 chamber at much higher concentrations to those measured in the 270 m3 SAPHIR chamber at much lower, near-ambient concentrations.

The gas–particle partitioning coefficients calculated in these experiments used the aerosol and total gas + aerosol measurements from the TD-CRDS and a total mass measurement from a Brechtel SEMS (BMI Model 2002). The partitioning coefficients derived from these experiments were 5×10-4 and 4.4×10-3 m3µg-1 for background aerosol loadings of 230 and 20 µgm-3, respectively. One of these experiments is shown in Fig. 10. The aerosol (caer) and total AN concentrations and background aerosol loading (Mtot) were averaged over the shaded period; aerosol phase was subtracted from the total to obtain cgas, from which Kp was determined via Eq. (4): 4Kp=caerocgasMtot. The fact that the two experiments at different background aerosol mass loadings (Mt) did not give exactly the same Kp value could reflect the uncertainty of these measurements or that the aerosol partitioning is not perfectly described as absorptive partitioning or that wall losses change as aerosol loadings change. Most important, the range of Kp measured here falls exactly within the range of values observed over 1 month of NO3+ isoprene experiments conducted under much lower-concentration conditions at the SAPHIR chamber (5×10-4–6×10-3 m3µg-1, Brownwood et al., 2020)

These Kp values were compared to theoretical calculations of Kp predicted by the simplified pLo prediction (SIMPOL.1) group contribution method (Pankow and Asher, 2008), and we find this average volatility consistent with a trifunctional isoprene nitrates, such as isoprene hydroperoxy nitrate, which has a SIMPOL.1 predicted Kp value of 2.38×10-3 m3µg-1. This shows a promising consistency of equilibrium gas–aerosol partitioning of isoprene nitrate products measured in two dramatically different chambers and suggests the robustness of the TD-CRDS over a wide range of concentrations.