The photoacoustic spectrometer used in this study has been described previously elsewhere . Briefly, it is a single-cell four-wavelength laser PAS. Four diode lasers (406, 532, 662, and 780 nm) are combined into a single beam with dichroic mirrors and turned into a multi-pass cell consisting of two highly reflective cylindrical mirrors (R>99 %); the front mirror has a 2 mm entrance hole drilled in the center . The PAS cell itself sits within the multi-pass cell and follows the design of . A calibrated photodiode behind the rear multi-pass mirror is used to monitor the power of each laser simultaneously. The system includes a cavity ring-down cell (CRD) operating at 662 nm (from the same 662 nm laser employed by the PAS) for direct calibration of the PAS. The four lasers in the PAS are operated simultaneously at frequencies spaced every 2 Hz around the resonant frequency of the cell. A fast Fourier transform (FFT) is performed on the microphone signal to deconvolve the signals at each wavelength. The resonant frequency of the PAS cell is measured by scanning the laser frequency across the resonant peak of the cell, typically filled with only the bath gas, and finding the best fit to the frequency sweep data. A frequency sweep was conducted prior to each set of measurements and anytime the gas type was changed. From these sweeps, the quality factor, Q, of the cell was determined to be 30. The lasers can be individually switched from digital modulation (as is used for PAS) to continuous-wave mode, which is helpful in conducting photolysis studies. The incident single-pass powers, which are representative of the powers experienced by each O3 molecule, were 61, 32, 44, and 77 mW for 406, 532, 662, and 780 nm, respectively. A diagram and more thorough description of the instrument can be found in .

Following our typical procedure, as described in , we calibrated the PAS by pushing a mixture of nitrogen dioxide in nitrogen through the instrument. A standard 10.29 ppm (±5 %) mixture of NO2 in N2 with a trace of O2 for stability (Airgas, Athens, Georgia) was diluted to various concentrations into N2 boil off from a liquid nitrogen dewar (Airgas, Athens, Georgia). The rotameter was used to measure the flow of NO2 while the N2 flow was controlled with a needle valve at approximately 200 SCCM (standard cubic centimeters per minute) and measured with an electronic flow meter (TSI, Shoreview, Minnesota). The flow rate through the instrument was the sum of the two flows and ranged from 225 to 400 SCCM depending on the NO2 flow rate. NO2 was introduced first to the PAS cell and then transported to the CRD via a short length (10 cm) of copper tubing. The outlet of the CRD was plugged and the gas was directed out of the purge inlets to avoid dead volume in the cell (no purge flow was used for NO2 measurements). CRD and PAS measurements were conducted simultaneously at 662 nm, and all other lasers were turned off during NO2 measurements. Figure a shows a block diagram of the setup used for NO2 measurements. The outlet of the PAS cell was open to atmospheric pressure, and as such the pressure inside the cell was free to fluctuate with the local ambient pressure. Likewise, the temperature was free to fluctuate with ambient temperature but was within in the range of 22 ± 2 ∘C for all experiments. Prior to all NO2 experiments, 10 ppm NO2 was flowed through the cell at 1–2 SLPM for several minutes to passivate all components of the system.

Ozone was generated using a commercial corona discharge ozone generator (Pacific Ozone, Benicia, California) with high-purity O2 (99.999 %, Airgas, Athens, Georgia). The ozone was trapped on silica gel in a glass trap held in a slurry of solid CO2 and ethanol at -73 ∘C. Prior to trapping, the silica gel and trap were heated to 100 ∘C while being held under vacuum for at least 1 h to remove contaminants. As with NO2, no purge flow was used during O3 measurements and the sample was pushed to the PAS first and transported to the CRD via a short length of copper tubing. The outlet of the CRD was plugged and the sample was directed out of the purge flow lines to minimize dead volume inside the cell. Figure a shows a block diagram of the setup used for ozone measurements. Mass flow controllers were used to control the ratio of oxygen to nitrogen (MKS Instruments). The outlet of the PAS cell was open to atmospheric pressure, and as such the pressure inside the cell was free to fluctuate with the local ambient pressure. Likewise, the temperature was free to fluctuate with ambient temperature but was within in the range of 22 ± 2 ∘C for all experiments. Ozone calibrations were performed at 532, 662, and 780 nm; 406 nm measurements were not conducted because of a very low signal-to-noise ratio at that wavelength for the relatively low ozone concentrations used.

Figure b shows a block diagram of the setup used for nigrosin measurements. Nigrosin aerosol was generated using a constant output atomizer (TSI 3076) with an aqueous solution of nigrosin (4 gL-1, Sigma Aldrich catalog number 198285, CAS# 8005-03-6, LOT MKBG7493V) and dried using a series of two silica gel diffusion dryers. The relative humidity was kept below 5 % and monitored with an inline relative humidity probe (HMP110, Vaisala Corporation, Helsinki, Finland). Atomized, dried particles were size selected at electrical mobility diameters of 500, 550, 600, and 650 nm using an electrostatic classifier (TSI 3080) and differential mobility analyzer with a 10 : 1 sheath flow-to-sample flow ratio and an 0.071 mm diameter impactor orifice to provide a cut point of approximately 1100 nm and reduce transmission of doubly charged particles (DMA, TSI 3085). Monodisperse aerosols were split in parallel to a condensation particle counter (CPC, TSI 3775) and the photoacoustic cell and delivered to each instrument through conductive silicone tubing. After particles passed through the PAS, they entered the CRD cell, which had a purge flow of 60 SCCM N2 (maintained by a critical orifice) over each mirror to prevent particle deposition. The aerosol sample was pulled through the instrument with a diaphragm pump (KNF Neuberger, Inc., Trenton, NJ) and the flow rate was maintained at 330 SCCM total flow with a critical orifice (Lenox Laser, Glen Arm, Maryland). All lasers were operated simultaneously. The refractive index from was used to calculate nigrosin absorption cross sections using Mie theory assuming a geometric standard deviation of 1.05. Mie theory calculations were performed in MATLAB.

We prefer to calibrate with NO2 by measuring the PAS signal at 662 nm and comparing to the absorption measured by the CRD at 662 nm. Because each of the instruments is illuminated by the same laser, the uncertainty is determined only by the uncertainty of the CRD and the precision of the PAS; all uncertainties associated with flow measurement and absorption cross sections are irrelevant. Further, because all of our wavelengths are contained in a single cell, the power-normalized calibration at 662 nm can be applied to all wavelengths (including 406 nm, at which wavelength NO2 photodissociates) . This approach, however, adds some additional uncertainty from the measurement of the effective power of each wavelength.

Performing the calibration with NO2 using the CRD to determine absorption yields a calibration slope of m=11.9×10-4 (V / W)/Mm-1. Because we used a 10 ppm calibrated mixture of NO2, we were also able to independently derive a calibration slope using the calculated NO2 absorption from the product of the concentration, N, and the absorption cross section, σ, as measured by . This slope of m=11.7×10-4 (V / W)/Mm-1 is within 1.5 % of the CRD method despite the larger uncertainty due to uncertainties in flow measurements. With nigrosin, we obtain a slope of m=10.7×10-4 (V / W)/Mm-1, within 10 % of the NO2 calibrations. The calibration curves for these methods can be seen in Fig. . Although the agreement here is not bad, there is some discrepancy between ozone and nigrosin. We speculate this is due to errors with the nigrosin calibration due to CPC errors (accuracy = 10 %) and/or errors or lot-to-lot differences in the refractive index of nigrosin, for example.

We have observed discrepancies between ozone and NO2 calibrations. In Fig. , which shows the calibration data and fits to all wavelengths (for ozone at 532, 662, and 780 nm), the most dramatic outlier is the dashed grey-green line obtained from ozone in pure N2, which yields a slope more than 50 % lower than the slopes obtained with NO2 and nigrosin at m=5.1×10-4 (V / W)/Mm-1. We hypothesized that this difference was due to photolysis of ozone under irradiation by visible light via O3+ hν(>1.1eV) →O2(3Σg-)+O(3P) . If this were the case, we further hypothesized that diluting ozone with oxygen instead of nitrogen would yield a larger calibration slope because the oxygen would promote recombination of O(3P) and O2 to form ozone. Indeed, as the dotted teal line in Fig.  indicates, the calibration slope fit to all three wavelengths under conditions of 100 % O2, m=9.8×10-4 (V / W)/Mm-1, was much closer to the slopes obtained using NO2 or nigrosin. The slopes derived from fits to the data of the individual wavelengths are similar, as expected, since the calibration should be independent of the wavelength of light: m=9.6,10.3,and9.6×10-4 (V / W)/Mm-1 for 532, 662, and 780 nm, respectively.

To search for evidence of O3 photolysis, we operated our 532 nm laser in continuous-wave mode. This mode prevented the laser from contributing to the PAS signal and yielded maximum continuous power available for photodissociation. The PAS signal due to ozone was monitored with the 662 nm PAS channel, and the concentration of ozone was monitored with the cavity ring-down spectrometer using the absorption cross section of . This approach allowed us to separate effects due to a lowering of the ozone concentration (which would be evident with the CRD) and any additional loss of PAS signal resulting from energy loss due to photodissociation. The green shaded regions in Fig.  indicate when the 532 nm laser was turned on to illuminate the ozone inside the PAS. An immediate decrease of 5 % in both the PAS signal and the ozone concentration measured with the CRD is noticed, consistent with a loss of ozone due to photodissociation. However, a simple photolysis calculation assuming a unit quantum yield for photodissociation indicates that nearly all of the O3 (more than 99.9 %) should photodissociate. Given the small 5 % loss observed, we conclude that a trace of O2 must have been present, thereby promoting reformation of O3; indeed, we estimate that only 4 ppm of O2, perhaps coming from the O3 trap or just a tiny leak of ambient air, would be sufficient to compete with the photolysis loss. The origin of the slight upward drift apparent in the PAS signal is not known, but it may indicate a shift in cell resonant frequency or temperature; nonetheless, the observed 5 % loss of signal is substantially smaller than the 50 % reduction in sensitivity observed in Fig. . Furthermore, the loss due to the 532 nm light was not observed when we added 5 % oxygen (of the total sample flow), as shown in Fig. b, suggesting that in the presence of oxygen ozone is rapidly reformed. But how much oxygen is sufficient to accurately perform a PAS calibration with ozone? For example, it can be convenient to calibrate in air (i.e., 20 % oxygen, for example in ) but is there a sufficient amount of oxygen to ensure the full sensitivity of the PAS?

The effects of oxygen on the PAS signal can be seen clearly in Fig. . In Fig. a, oxygen was added to the sample line such that it made up 5 % of the total flow. The red regions indicate when O2 was added to the sample stream, and the white regions indicate when it was removed. There is a clear difference upon addition of O2 to the sample flow, with it increasing the signal roughly 50 %–75 %, and a similar trend was observed at all four measurement wavelengths available in our PAS. This effect cannot be due to changes in the concentration of ozone, which were monitored with the CRD and actually decreased slightly when oxygen was added (due to the slight dilution of the sample flow). An alternative explanation would be a shift in the resonant frequency upon addition of oxygen. However, because the resonant frequency was measured in nitrogen, any shift in resonant frequency should only decrease the signal. Further, measurements of the resonant frequency showed negligible differences between nitrogen-only samples and those with 5 % oxygen added. Thus, the change in composition was not enough to have an appreciable effect on the resonant frequency of our low-Q (wide-bandwidth) PAS cell. We therefore conclude that the observed increase in signal upon addition of oxygen is indeed attributable to a change in sensitivity accompanying the change in composition of the bath gas. Finally, such a phenomenon was not observed when adding argon instead of oxygen (data not shown), implying that the effect is attributable to the presence of oxygen specifically.

Figure b shows the effect of adding oxygen in varying amounts from 0 % to 100 % of the bath gas. A clear trend is observed in relation to the oxygen concentration at 532, 662, and 780 nm; the effect likely exists at 406 nm as well, but that wavelength was not measured because of ozone's low absorption cross section at that wavelength. In the absence of O2, the sensitivity is about 4×10-4 (V / W)/Mm-1, more than a factor of 2 lower than the normal PAS cell sensitivity measured with either NO2 or nigrosin. The sensitivity increases quickly as oxygen is added up to about 20 % oxygen in nitrogen, at which point it begins to asymptotically approach an upper limit that is more in line with the sensitivities measured by other methods. Others have observed similar effects measuring HCN when adding water vapor into the cell and when adding oxygen into a mixture of NO2 and N2, although in that case adding oxygen caused a decrease in the signal .

We note that the observed sensitivity dependence on bath gas composition could partially explain the lower sensitivity to O3 compared to nigrosin particles observed by since the bath gas in that study contained only 10 % O2. Using the data in Fig. b, we estimate that the sensitivity would be 17 % low, which is in the right direction but cannot explain the entire difference. Likewise, we estimate the sensitivity to O3 in the work of , which used 25 % O2 in the bath gas, to be 12 % low. We conclude, then, that the different amounts of O2 in the bath gas for these two studies cannot fully explain the discrepancy between them.

An underlying assumption of PAS is that all the photon energy absorbed by the sample is transferred to the bath gas as thermal energy to create an acoustic wave. This process requires efficient transfer of energy from the excited state of the analyte (e.g., O3*) into translational, rotational, and/or vibrational modes of the bath gas and the further relaxation of the bath gas molecule. However, if the transfer of energy from the analyte to the bath gas is inefficient or if the excited state of the bath gas, analyte, or another intermediate is long-lived with respect to the modulation frequency of the light source, the photon energy will not be efficiently converted to acoustic energy, which is observed as a decreased sensitivity. The observed dependence on O2 concentration indicates that energy transfer is more efficient with O2 as the bath gas compared to N2, and the shape of the dependence on O2 concentration suggests a competitive kinetic model. Indeed, the data are fit reasonably well by a simple model in which O2 and N2 are each assumed to deactivate the O3* in one step but with different rate constants, kO2 and kN2: dO3*dt=-kN2O3*N2-kO2O3*O2. Following the derivation of , the sensitivity, m ((V / W)/Mm-1), can be expressed as m=m01+A1+rχ(1-χ)2, where m0 is the asymptotic sensitivity coefficient (i.e., with instantaneous relaxation), χ is the O2 mole fraction, r is the ratio of the quenching rate constants for oxygen and nitrogen (kO2/kN2), and A=2πfτN2, where f is the modulation frequency and τN2 is the deactivation lifetime of O3* in 100 % N2 =1kN2[N2]. For efficient conversion of the absorbed photon energy to acoustic energy, the deactivation rate must be significantly faster than the modulation frequency, meaning A must be ≪ 1.

Fitting Eq. () to each of the three data sets in Fig. b results in reasonable fits with R2 values of 0.96 or greater. The values of the A parameter are 1.6, 2.2, and 3.0 for 532, 662, and 780 nm, respectively, reflecting the fact that the energy transfer in 100 % N2 is inefficient for all three wavelengths. The values of r, the ratio of the deactivation rate constants in O2 and N2, are 22, 37, and 55 for 532, 662, and 780 nm, respectively, reflecting the increased sensitivity in the presence of O2. The differences in these values may reflect differences in the densities of states of the bath gas and the ozone when excited by the different wavelengths of light, though a more definitive interpretation is beyond the scope of this work. The values of m0 are 11.2, 10.8, and 10.9×10-4 (V / W)/Mm-1 for 532, 662, and 780 nm, respectively, which indicate similar sensitivities in the limit of 100 % O2 for all three wavelengths and are within 10 % of the sensitivity measured with NO2. Clearly, however, the data appear not to have reached an asymptote even at 100 % O2, which may reflect the limitations of using such a simple model in which deactivation of O3* by N2 and O2 is represented by single steps. Nonetheless, this model captures the general shape of the sensitivity dependence on O2 concentration and provides a guide for assessing the relative efficiencies of the two bath gases. In fact, the measured values of the sensitivities at 100 % O2 are within 3 % of the NO2 measurement, indicating that calibration with O3 is a viable option as long as it is performed with 100 % O2 as the bath gas. It may even be possible to perform such a calibration with smaller concentrations of O2 and use a correction based on a curve similar to that shown in Fig. b, though the additional uncertainty incurred with doing so may make such an approach undesirable. Finally, we note that since the A term is a function of f, the sensitivity of PAS measurements made at frequencies higher than those used here (1414 Hz in 100 % N2) will demonstrate an even more pronounced dependence on O2 concentrations.