Comment on “The effects of molecular weight and thermal decomposition on the sensitivity of a thermal desorption aerosol mass spectrometer”

Abstract

Copyright © American Association for Aerosol Research

EDITOR:

• Paul Ziemann

This article refers to:

The effects of molecular weight and thermal decomposition on the sensitivity of a thermal desorption aerosol mass spectrometer

This article is referred to by:

Reply to “Comment on the effects of molecular weight and thermal decomposition on the sensitivity of a thermal desorption aerosol mass spectrometer” by Jimenez et al.

Dear Editor,

Recently a manuscript was published in Aerosol Science and Technology (50:2, 118–125, 2016; doi: 10.1080/02786826.2015.1136403) by D. M. Murphy, entitled, “The effects of molecular weight and thermal decomposition on the sensitivity of a thermal desorption aerosol mass spectrometer.” That manuscript presents a simple model (hereafter “M16 model”) for the vaporization and ionization of molecules in a vacuum system, similar to the processes occurring in the Aerodyne Aerosol Mass Spectrometer (AMS). The idealized model is based on the dispersion of molecules in the free molecular regime and predicts an additional square root dependence of the ionization efficiency on the molecular weight (MW) of the molecule of interest. This model correctly accounts for only one of the many complex processes that affect AMS sensitivity to aerosol chemical components.

The AMS instrument is widely used in the international atmospheric sciences community. As experienced AMS users, we find the Murphy (2016) conclusions regarding AMS data quality to be inconsistent with our understanding of and findings on AMS performance. This inconsistency and the seriousness of the conclusions make it imperative to us to present a response providing a more balanced perspective to our atmospheric science colleagues inside and outside the AMS user community.

We present here a variety of evidence and reasoning demonstrating the inadequacy of the M16 model to represent important aspects of AMS performance. In the following, we use previously published results to address key issues raised in Murphy (2016) associated with basic AMS features (vaporization temperature, single-particle timescales, and ionization efficiencies), AMS calibration (relative ionization efficiency), and evaluation of possible biases on organic aerosol field data. Rather than being an exhaustive review, the evidence put forth in this letter is representative of the AMS performance experienced by the large and varied AMS users' community. We conclude that the generalizations and assertions based on the M16 model are not consistent with many experimental AMS results and, hence, this model cannot be used to draw conclusions about the sensitivity of the AMS as a function of MW and the accuracy of ambient AMS measurements.

1. Evidence that the M16 model is inconsistent with AMS measurements

1.1. The M16 model is not consistent with measured AMS vaporizer temperature dependences

AMS detection is based on impaction of the particles on a heated vaporizer (typically operating at a temperature, T_v, of 600°C) under vacuum followed by vaporization, electron ionization of the resulting gas-phase molecules, and mass spectrometric detection of the ions (Jayne et al. 2000; Jimenez et al. 2003; Canagaratna et al. 2007). Murphy (2016) proposes that the intensity of the ion signal (I) produced from an individual chemical species vaporizing from an aerosol particle in the AMS should follow the following equation (under a steady state assumption):[1] where C is a proportionality constant representing the efficiency of the analysis system, σ is the electron ionization cross section of the molecule, is the vaporizing mass flux, A is the area of the holes in the ionizer through which molecules can escape to the pumping system, k is the Boltzmann constant, T_v is the absolute temperature of the AMS vaporizer, and MW is the molecular weight of the vaporizing molecules.

According to Equation (1[1] ), the measured ion signal should follow a dependence as the vaporizer temperature is varied. However, published evidence shows that this dependence is not observed. Figure 1a shows a direct comparison of AMS data from two different field studies (Docherty et al. 2015) with the expected dependence from the M16 model for nitrate, sulfate, and ammonium. Data are only shown for the T_v range when the species size distribution and instrument background data indicate that vaporization was as fast as at 600°C (Docherty et al. 2015), consistent with the temperature range and the timescales in Drewnick et al. (2015) discussed later (Figures 1b and c). The signals of the three species stay constant or decrease slightly with decreasing T_v, in sharp contrast with the M16 model that predicts a signal increase of 20% over the range shown.

Figure 1. Examples of evidence that the M16 model does not capture important aspects of AMS detection. (a) Measured signals versus vaporizer temperature (relative to signals at T_v = 600°C) for ambient aerosols at Blodgett Forest, CA during BEARPEX-2007, and at an urban site in Riverside, CA during SOAR-1 in 2005 (Docherty et al. 2015). NO₃, SO₄, and NH₄ refer to total nitrate, sulfate, and ammonium as quantified by the AMS by adding all species-related ion signals (Allan et al. 2004a; Canagaratna et al. 2007). Vaporizer temperature was automatically cycled every 30 min for multiple days across the range shown. Data are only shown for the T_v range where the species size distribution data indicate that vaporization was as fast as at T_v = 600°C (see Figure 3 of Docherty et al. 2015). These T_v ranges are consistent with the ranges for fast single-particle vaporization timescales shown in panels b and c of this figure. Relative variation of timescales of detection of (b) ammonium nitrate and (c) ammonium sulfate ion fragments from single particles versus T_v, compared to the Murphy (2016) model predictions. Data from Drewnick et al. (2015). The very long timescales at lower temperatures are due to slower vaporization. (d) Comparison of absolute timescales of single-particle detection for ammonium nitrate (Drewnick et al. 2015), compared with the Murphy (2016) model results. Model results were interpolated from Figure 2 in Murphy's paper, assuming vaporization of NH₄NO₃ (MW = 80) as suggested by Murphy (2016), and for vaporization of HNO₃ thermal decomposition products such as NO₂ (MW = 46) (Drewnick et al. 2015). Digitization was also not simulated in the model and may account for approximately 6 μs in the Drewnick et al. (2015) results, which is much smaller than the observed model-measurement differences. (e) Timescales of single-particle vaporization for ammonium nitrate and (f) ammonium sulfate fragments vs. m/z, for the data of Drewnick et al. (2015), and the Murphy (2016) model using the intact salt MWs. Data points are sized so the area of each point is proportional to the fraction of the total signal at that ion. (g) Measured ammonium concentration during the SOAR-1 study compared to that predicted for cation/anion balance from the measured anions. Small corrections have been applied based on estimated organonitrate, organosulfate, and amine concentrations. See Figure 12 of Docherty et al. (2011) and associated discussion for further details. (h) Ratio of measured to predicted ammonium versus the ambient nitrate to sulfate mass ratio in the same study.

Figure 1. (Continued).

Experimental single-particle vaporization event lengths (Drewnick et al. 2015) also show a very different dependence on temperature than the predicted by the model (Figures 1b–c). Thus, Figures 1a through c indicate that other T_v-independent effects not accounted for in the M16 model must play a major role in the conversion of aerosol mass to ion signal in the AMS detection process. In particular, it appears that the vaporization process, and not the distribution of molecular speeds at T_v, controls the observed single-particle timescales.

One possible explanation of the discrepancy between the modeled and measured temperature dependence of the ion signal is that the effective translational temperature of a species after vaporization (T_eff) differs from T_v. These results suggest that T_eff < T_v and that T_eff is not proportional to T_v over very wide ranges of T_v. Murphy (2016) uses 500°C as the vapor temperature in the model to compare with data taken with T_v = 420°C. The use of (approximately) T_v in the model is justified based on limited evidence from a different instrument (Sykes et al. 2002), and is supported by the model result that many vaporizing molecules will experience one collision with the vaporizer before ionization (Murphy 2016). However, the data in Sykes et al. (2002) do not conclusively demonstrate that the molecules reached T_v in their instrument, as their measured pulses were longer than the modeled ones, and their model was tuned to the observations. In addition, more than one collision is needed to thermalize gas molecules to a surface temperature (Lahaye et al. 1995; Butler et al. 1997). Since boiling point increases with MW, it is possible that T_eff also increases with MW and this effect partially compensates the free molecular flow effect. The question arises whether quantifying and incorporating a species-dependent T_eff into the M16 model could result in better agreement with the observations. However, a T_eff < −200°C would be needed to explain the high ionization efficiencies for NH₄ and H₂O observed in the AMS and discussed later.

1.2. The M16 model is not consistent with measured single-particle timescales

The M16 model assumes that the single-particle vaporization pulse lengths measured with the AMS are solely determined by a species-independent particle vaporization time (4 μs in the model) and the time spent by vaporized molecules in the ionization region, assuming a Boltzmann velocity distribution at T_v. The only use of AMS data to validate the M16 model in Murphy (2016) is a comparison of calculated single-particle pulse length with measured single-particle pulse length from the experiments of Jayne et al. (2000). As reported by Jayne et al., the geometry of their vaporizer was a closed-tube furnace approximately 1 cm long with a 0.3 cm ID opening. Note that this geometry is not the same as the inverted cone vaporizer, which has been incorporated into all AMS instruments since 2000 and was used for the M16 model. The Jayne et al. (2000) ionizer was also smaller than the one in all current AMS instruments. These differences in geometry resulted in significantly shorter pulses measured in the Jayne et al. (2000) experiments compared to those from Drewnick et al. (2015). Hence, the timing and duration of single-particle pulses from Jayne et al. (2000) cannot be used to validate the M16 model.

Single-particle ammonium nitrate vaporization pulse lengths for the inverted-cone vaporizer from the Drewnick et al. (2015) experiments are compared directly with those calculated using the M16 model in Figure 1d, showing the measured pulses are at least twice as long as those predicted by the model, and three to four times in the case of m/z 30. In addition, the measured ion pulse at m/z 30 is substantially longer than that at m/z 46, which is inconsistent with both the predicted MW dependence in M16 (which predicts the opposite trend) as well as the proposed ionization of intact ammonium nitrate molecules (which predicts equal values, as described later in the article). This indicates that additional processes (e.g., vaporization at lower effective temperatures, vaporization and decomposition over times longer than 4 μs, adsorption and subsequent desorption onto and from ionizer surfaces, and/or other effects) have a significant influence on the measured particle vaporization timescales, none of which are included in the M16 model.

1.3. There is a large discrepancy between the observed sensitivity to ammonium and water and the M16 model predictions

1.3.1. Ammonium ionization efficiency

The M16 model predicts an AMS ionization efficiency for ammonia (relative to nitrate, RIE_NH3) of ∼0.75 compared to values of 3.5–6.2 that are observed for ammonium in the AMS (Salcedo et al. 2006; Canagaratna et al. 2007). In an attempt to explain this large discrepancy, Murphy (2016) proposes that ammonium nitrate and ammonium sulfate vaporize as molecular salt species, as follows:

Table

Ammonium nitrate	Ammonium sulfate
NH4NO3(p) → NH4NO3(g)	(NH4)2SO4(p) → (NH4)2SO4(g)
NH4NO3(g) + e^− → NH4NO3^++ 2e^−	(NH4)2SO4(g) + e^− → (NH4)2SO4^++ 2e^−
NH4NO3^+ → a * (NH3^+ + HNO3)	(NH4)2SO4^+ → c * (NH3^+ + NH4HSO4)
→ b * (NH3 + HNO3^+)	→ d * (2 NH3 + H2SO4^+)

where (p) and (g) indicate particle and gas-phase species, respectively, and a, b, c, and d indicate fractional yields in the respective reactions. Subsequent fragmentation of the ions shown would not change the ratios of total ions observed from each species. Mixed fragments (e.g., containing parts of both the NH₃ and H₂SO₄ molecules) are not observed in AMS mass spectra (Hogrefe et al. 2004a). In the M16 model, a vaporized molecular salt [e.g., NH₄NO₃(g)] would have a specific distribution of velocities in the gas phase, leading to the same detection timescales for different ion fragments of each pure salt (e.g., NH_x⁺ and NO_x⁺).

Figure 1e (1f) compares the measured single particle pulse widths for ammonium nitrate (sulfate)-related ions from Drewnick et al. (2015) with the results from the M16 model. Most apparent are the longer pulse widths for NO⁺ (m/z 30), SO⁺ (m/z 48), and SO₂⁺ (m/z 64) than for the other ions. The main ammonium and anion fragments produced from each salt do not have the same pulse widths, confirming that the ion fragments observed for ammonium nitrate and ammonium sulfate do not arise from complete neutral salt molecules reaching the ionization region. Rather, other processes such as ammonium nitrate decomposing into NH₃(g) and HNO₃(g) (and potentially other molecules such as NO and NO₂) at the vaporizer surface must be occurring in the AMS that causes the observed fragments to have different pulse lengths arising from separate decomposition products.

The robust fragmentation patterns for AMS ammonium detection (Allan et al. 2004a) are independent of the type of ammonium salt and very similar to the NIST EI spectrum for NH₃(g) (NIST, 2016). Moreover, AMS measurements consistently show that the sensitivity for ammonium from ammonium nitrate is the same as from ammonium sulfate. For example, field data such as those shown in Figures 1g and h show that the average ratio of the measured-to-predicted (based on ion balance) ammonium is constant within 5%, despite variations on the ambient sulfate/nitrate ratio of over a factor of 20 as well as substantial changes in the concentrations of other internally mixed species (Docherty et al. 2011). This would be very unlikely if the salts vaporized as molecular species. It is implausible that the result of the intact salt vaporization processes for pure and mixed salts would always produce the same ionization efficiency and fragmentation pattern for ammonium, independent of wide variations in mixture composition. These arguments strongly support the vaporization of NH₃(g) from the sampled ammonium salts.

Given the strong evidence for vaporization of NH₃(g) from ammonium nitrate and sulfate, not molecular salt vaporization, the large discrepancy with M16 model predictions of RIE_NH4 and single particle vaporization times indicate a major incompleteness in the representation of the AMS vaporization/ionization processes in the M16 model.

1.3.1.1. Sensitivity to ammonium in ammonium oxalate

In Murphy (2016), the understanding of AMS RIE_NH4 is questioned based on a report of lower than expected signal for ammonium from ammonium oxalate (AmOx) by Takegawa et al. (2007). These authors suggested that one of several possible interpretations of their results was that RIE_NH4 from AmOx was one third of the RIE_NH4 for NH₄NO₃, while noting that “the RIE_NH4 issue remains uncertain and requires additional laboratory experiments in future studies.” Mensah et al. (2011) performed similar laboratory experiments and report the same RIE_NH4 from AmOx and from NH₄NO₃, in contrast with the Takegawa et al. results. We note that the Mensah et al. (2011) results cannot be explained if RIE_NH4 from ammonium oxalate was lower than for ammonium nitrate, as it is implausible that their particles contained an amount of ammonium three times larger than required to neutralize oxalic acid. One possible explanation for the low ammonium signal in the Takegawa et al. experiments is preferential evaporation of NH₃(g) from the ammonium oxalate particles, likely in the aerosol sample lines or size classifiers. Indeed, thermodynamic model calculations using the extended aerosol inorganics model (E-AIM; Clegg and Seinfeld 2006a,b; Clegg et al. 2016) show that the saturation concentration of NH₃(g) over a solution of ammonium oxalate is of the order of 40,000 μg m⁻³, while that of oxalic acid is negligible (∼1 ng m⁻³), creating conditions that would favor preferential evaporation of NH₃(g) from AmOx. In contrast, the estimated vapor pressures of NH₃(g) and HNO₃(g) over NH₄NO₃ are between these values, which lead to evaporation of both NH₃(g) and HNO₃(g) in that case. Evaporation of NH₄NO₃ in laboratory setups has been reported (Zelenyuk et al. 2006) although it is not common during AMS calibrations (Canagaratna et al. 2007). The differences in estimated vapor pressures indicate that preferential evaporation of NH₃(g) is likely to be much more important for AmOx. Evaporation of AmOx in laboratory setups has been observed using size measurements (M. Petters, pers. comm., 2016). We thus hypothesize that the different experimental conditions in the two AmOx studies led to differences in NH₃(g) evaporation and thus detected ammonium. Further experimental study of this system is recommended, but taken together, these results suggest that the low ammonium signal in the Takegawa et al. experiments is most likely unrelated to RIE_NH4.

1.3.1.2. Potential conversion of ammonium to NO⁺ on the AMS vaporizer

The possibility of ammonium being converted to NO⁺ on the AMS vaporizer is discussed in Murphy (2016). While not explicitly stated in the table of sulfate ion fragments, early experiments of ammonium sulfate particles did not show evidence of conversion of ammonium to NO⁺ (Hogrefe et al. 2004a). Ammonium sulfate calibrations on several instruments in our groups show an NO⁺ signal of ∼1/2000 of the ammonium signal. This is likely due to cross contamination from ammonium nitrate because of alternating use of the same atomizer. Thus, we conclude that conversion of ammonium to NO⁺ is unlikely to play a role for AMS detection.

1.3.2. Water ionization efficiency

There is consensus on a high relative sensitivity to water from particles as they are sampled by the AMS (Matthew et al. 2008; Engelhart et al. 2011; Mensah et al., 2011). The first two studies did not directly report RIE for particulate water due to evaporation in the instrument. Mensah et al. (2011) used dried particles with known hydrates to estimate RIE_H2O = 2. Engelhart et al. (2011) reported that the product of RIE_H2O and the evaporated fraction of water in their system was close to unity during their ambient air study, and their data can be used to estimate RIE_H2O. There is substantial evaporation of water from ambient particles upon entering the AMS aerodynamic lens system, as supported from typical evaporation rates of water under ambient conditions (∼150 monolayers in ∼15 ms residence time in the aerodynamic lens). Zelenyuk et al. (2006) studied ammonium sulfate particles (similar in composition to those in Engelhart et al.) and estimated that ∼80% of the particle water evaporated after sampling. Even larger fractional evaporation has been published for ammonium sulfate by Matthew et al. (2008). Using an estimated 80% loss of water for the Engelhart results, we estimate RIE_H2O ∼5, similar to the values observed for RIE_NH4. Thus, the range of RIE_H2O estimated from experiments is 2–5, in strong disagreement with the value of 0.60 predicted by the M16 model.

2. Quantification of organic aerosol mass

2.1. Calibration of relative ionization efficiencies

In Murphy (2016), the following inaccurate and misleading statement is made about quantification of organic aerosol (OA) mass by the AMS: “… the calibration for ammonium nitrate is transferred to organic species using approximate electron ionization cross section (Canagaratna et al., 2007).” The connection between AMS sensitivities and electron ionization (EI) cross sections was originally made in Jimenez et al. (2003) in order to highlight that (a) the ion signals are approximately proportional to mass instead of to molar concentrations, providing the basis for mass being the primary data product of the AMS; and (b) sensitivities are of the same order for different species, so that differences of several orders-of-magnitude as seen in other techniques (e.g., single-particle laser ablation or chemical-ionization mass spectrometry) are not expected. However, Jimenez et al. (2003) included a species-specific response factor (later specified as CE_s × RIE_s) in the quantification equation to account for differences in ionization efficiency and other effects, and stated the need for calibration. In the standard AMS analysis, no aerosol species (including OA) is quantified based on relative EI cross sections.

AMS quantification of the mass concentration of a species s within a multicomponent aerosol particle is based on the following equation (Alfarra et al. 2004a; Canagaratna et al. 2007):[2] where C_s is the mass concentration of species s, MW_NO3 is the molecular weight of nitrate, CE_s is the collection efficiency of species s, RIE_s is the relative ionization efficiency of species s, IE_NO3 is the ionization efficiency of nitrate (from calibrations), Q is the volume flow rate into the AMS, N_A is Avogadro's number, I_s,i is the ion signal from ion i produced by species s, and the 10¹² factor accounts for unit conversions. We note that almost all ambient and laboratory experiments, including the calibrations discussed here, are carried out with a standard T_v ∼ 550–600°C.

Equation (2[2] ) relies on a calibrated relative ionization efficiency (RIE_s) and not on a single fundamental molecular property (such as the electron ionization cross section). The ionization efficiency of a species (IE_s) has an expected linear MW dependence since electron ionization cross sections increase approximately linearly with MW (Canagaratna et al. 2007):[3]

Murphy (2016) argues that there is an additional secondary dependence on for each component species that is not taken into account explicitly in Equation (3[3] ) and, based exclusively on theoretical arguments, asserts that substantial errors arise from its omission. While we agree with an expected theoretical dependence of RIE_s on the time spent by the vaporized molecules within the ionization region, we assert that this dependence (as well as other effects not explicitly included in Equations (2[2] ) and (3[3] )) is implicitly included in the experimentally determined RIE_s.

Since ambient OA consists of thousands of different molecules, an average RIE_OA must necessarily be used to calculate OA mass concentrations. As stated in Canagaratna et al. (2007), the average RIE_OA of 1.4 that is recommended for calculating ambient OA mass concentrations was obtained from laboratory calibrations. Figure 2a shows the data used to derive this value. RIE_OA is obtained from the slope of the linear regression of ionization efficiency (IE) versus MW for a calibration dataset obtained from a variety of commercially available, oxygenated organic compounds analogous to those observed in ambient aerosol particles (Zhang et al. 2007). The relatively large error bars in Figure 2a reflect the difficulties in accounting for factors, such as particle bounce, multiply charged particles, particle shape factors, and particle volatilization during these calibrations. Although Figure 2a has not been previously published in the peer-reviewed literature, it has been presented at major conferences in the United States and Europe (e.g., Silva et al. 2001; Jayne et al. 2002; Worsnop et al. 2002, 2004a,b) and has been publicly available for over a decade at the University of Colorado website archive of AMS Users' Meetings presentations (Worsnop 2007).

Figure 2. Calibration data for different species in the AMS. (a) Calibration data used to derive RIE_OA = 1.4 for ambient organic aerosol mass calculations (Canagaratna et al. 2007). The calibration dataset includes several types of organic species including methanesulfonic acid, malonic acid, maleic acid, succinic acid, glutaric acid, malic acid, nitrophenol, levoglucosan, phthalic acid, and azelaic acid (P. Silva, Aerodyne Research and Boston College, unpublished work). Particles were generated by atomization of aqueous or organic solutions of the species, size selected with a differential mobility analyzer and counted with a condensation particle counter. (b) Scatter plot of measured RIE_s vs. species MW for the organic species in Figure 2a as well as published RIE_s values for hydrocarbons (lubricating oil, decane, and 4 PAHs) from Slowik et al. (2004) and Dzepina et al. (2007), and unpublished values for diesel, oleic acid and hexadecanol (P. Silva, Aerodyne Research and Boston College). Values for inorganic species are also shown. The horizontal (green) line and light gray region denote RIE_OA = 1.4 with 2σ uncertainty of 20%. The region between the dashed green lines shows the range of estimated dependence for intact species resistant to thermal decomposition in Murphy (2016). Note that PAHs and lubricating oil would be expected to lie in between these lines according to the M16 model. “EB to 6.2” (error bar) is added to mark that RIE_NH4 up to 6.2 are often measured (Salcedo et al. 2006; Canagaratna et al. 2007).

The overwhelming majority of the signal from OA in ambient aerosol is detected at m/z < 100 even in urban areas (e.g., > 90%; DeCarlo et al. 2008). This is due to both thermal decomposition during/after vaporization and fragmentation after ionization with 70 eV electrons. Previous work utilizing AMS systems coupled to “soft” ionization sources indicates that an important cause for the dominance of low m/z ions in AMS spectra is thermal decomposition of organic species (both oxidized and hydrocarbon-like) prior to ionization (Northway et al. 2007; Kroll et al. 2009; Robinson et al. 2011; Canagaratna et al. 2015). Thus, for most organic molecules, the actual MW of the molecules being vaporized and travelling through the ionization region is smaller than the parent MW due to thermal decomposition on the AMS vaporizer (Canagaratna et al. 2015). Any additional MW-dependence of IE beyond that shown in Figure 2a is expected to be further reduced by thermal decomposition during vaporization.

Figure 2b shows published RIE_s values (in addition to those in Figure 2a) measured for organic and inorganic species as a function of MW. The organic data points in the figure are colored depending on whether they are more likely to thermally decompose (light green) or to vaporize intact or with limited thermal decomposition (dark green). The gray dashed lines show the dependence of RIE predicted for thermally stable intact molecules using the M16 model (see Figure 3b of Murphy [2016]). In Murphy (2016) thermal decomposition was also included as a potentially important process and decomposing molecules were shown to have RIE_s in the region below the lower dashed line. The possibility of formulating an RIE_s parameterization with an explicit dependence was originally considered with respect to high AMS RIE values (Figure 2b) measured for organic species, such as oleic acid (MW 282) and hexadecanol (MW 242). However, as shown in Figure 2b measured RIE values of several other hydrocarbon species that are expected to vaporize with limited thermal decomposition do not follow an obvious dependence. Lubricating oil, a high MW species (average MW of ∼375; Worton et al. 2014) that is representative of ambient primary organic aerosol (POA) from traffic emissions (Canagaratna et al. 2004) and is expected to be stable with respect to thermal decomposition, has an RIE value of 1.4 (Slowik et al. 2004) compared to the ∼4 predicted using the M16 model. Similarly, PAHs, which are good test cases for dependence of RIE since they tend to survive vaporization intact in the AMS (Dzepina et al. 2007), have measured RIE values of ∼1.7 (for PAHs for MW 202–248; Slowik et al. 2004; Dzepina et al. 2007) instead of ∼3.2 as predicted by the M16 model. Finally, RIE values of some inorganic species deviate strongly from the M16 model prediction, as discussed earlier. Taken together, there is significant variability in measured organic and inorganic RIE_s, but it cannot be explained with a simple term.

RIE_OA from the linear regression slope in Figure 2a is 1.42, while the average of all organic RIE_s in Figure 2b is ∼1.6 (13% higher). A 2σ uncertainty of 20% for RIE_OA was estimated previously (Bahreini et al. 2009), consistent with this variability. The complexity of ambient multicomponent OA mixtures will lead to significant averaging and hence it is unlikely that using an average RIE_OA value for ambient AMS analysis will generate large biases in OA concentration. Note that some of the higher RIE_s values shown in Figure 2b belong to species that may not be highly representative of ambient OA constituents. For example, PAHs are known to make insignificant contributions to ambient OA mass (Dzepina et al. 2007). Ambient OA is dominated by contributions from oxidized organic species similar to those shown in Figure 2a, and hydrocarbon-like organic species such as lubricating oil (Canagaratna et al. 2004; Zhang et al. 2007). The applicability of a single RIE_OA = 1.4 for ambient aerosols over a range of conditions (including with higher POA) is discussed further in Section 2.2.

For the record, the data referenced in Murphy (2016) from Alfarra's doctoral thesis work (2004b) is not included in Figure 2 because the experiments were primarily performed for the purpose of exploring qualitative trends in detection of organics as a function of vaporizer temperature and not for absolute mass quantification. The absolute accuracy of the ions-per-particle values, needed to derive a quantitative trend with MW, is highly uncertain in this specific instance, as certain aspects of the experiment, such as how particle collection efficiency and multiply charged particles were accounted for, were not optimal at that early stage of AMS development.

It is important to note that RIE_OA = 1.4 is not recommended for laboratory data (or when studying a specific OA source), other than to obtain an estimate of the mass concentration. It has been known since the first environmental chamber experiments using an AMS that there is significant variability of collection efficiency (CE) for laboratory organic particles (Bahreini et al. 2005; Docherty et al. 2013; Ye et al. 2016). There is also real variability in RIE for different species (Figure 2) that is more important for laboratory experiments due to the use of single species or simpler mixtures. For these reasons, measurements of the response factor (RIE × CE) relevant to the OA type under investigation should be obtained empirically, for example by comparison with aerosol concentrations from Scanning Mobility Particle Sizer or other measurements (Li et al. 2013; Docherty et al. 2013). Alternatively, in lieu of determining specific RIE × CE values for various types of laboratory OA, some AMS users have chosen to use ratios such as OA/SO₄ to track relative changes in OA abundance, for example, in chamber experiments (Hildebrandt et al. 2009; Wong et al. 2015).

2.2. Accuracy of ambient OA mass concentration quantification with AMS

In Murphy (2016), it is claimed that the AMS sensitivity to OA may vary by more than a factor of 3, depending on the fraction of species thermally decomposing at the AMS vaporizer and those being thermally stable, and that, as a consequence, AMS OA mass concentration measurements may have several-fold biases. We dispute these claims as they are made with no compelling evidence while previously published work has already shown evidence to the contrary. OA uncertainty for ambient measurements was estimated to be significantly smaller (±38%) with the largest source of uncertainty being attributed to collection efficiency (Bahreini et al. 2009; Middlebrook et al. 2012). There are many published intercomparisons involving high quality measurements of ambient aerosols that show that the AMS OA mass concentrations are consistent with independent measurements within the stated uncertainty (e.g., Takegawa et al. 2005; DeCarlo et al. 2008; Timonen et al. 2010; Docherty et al. 2011; Middlebrook et al. 2012; Setyan et al. 2012; Washenfelder et al. 2015).

An example of evaluating the accuracy of AMS mass concentration measurements and how it might vary as different types of aerosols are encountered is shown in Figures 3a and b. The ratio of total aerosol mass concentration from the sum of AMS species mass concentrations and black carbon mass concentration (not detected by the AMS) to the total mass concentration determined by independent measurements of the accumulation mode (M_fine) was found to distribute around a value of unity for two airborne campaigns, during which OA comprised about one half of the fine aerosol mass. The distributions of measured ratio values are not wider than the distributions expected based on the combined uncertainties of the measurements and, hence, support the current operational AMS uncertainty estimates. For Figure 3b, the observed distribution is narrower than expected, suggesting that actual AMS uncertainties are lower than estimated for that case.

Figure 3. Examples of evidence that the OA concentrations reported by the AMS are correct within the stated uncertainties. (a) Ratio of the AMS total mass (using the standard RIE_OA) plus black carbon mass to the mass calculated from the fine particle volume for airborne campaigns over Alaska with aged biomass burning influence and (b) over Texas with aged urban and biogenic influences, with the combined experimental uncertainties depicted as dotted lines (adapted from Middlebrook et al. 2012). OA comprised 43% and 47% of total submicron nonrefractory mass in these campaigns, respectively. (c) Values of the OA hygroscopicity parameter (Petters and Kreidenweis 2007) needed for hygroscopicity closure in the OA-dominated southeastern United States and a forest site in the Rocky Mountains under different assumptions. Black: values expected for pure laboratory SOA from multiple precursors for the O/C measured in the ambient studies (Lambe et al. 2011). Gray: values reported in the literature assuming the AMS composition is correct (Brock et al. 2016; Cerully et al. 2015; Levin et al. 2014). Other colors: values needed to explain the measured hygroscopicity if AMS OA was overestimated by factors of 2 (orange), 3 (red), and +38% (green; 2σ upper limit of the OA accuracy stated in the literature; Bahreini et al. 2009). (d) Ratio of OC measurements from the AMS and a Sunset Laboratories OC instrument vs. the fraction of primary OC (POC) in total OC, for a campaign in Riverside, CA (Docherty et al. 2011). Also shown is the combined instrumental uncertainty (gray area, ±38% for AMS OA, ±15% for Sunset OC) and two lines representing assumed primary OA detection twice and three times more sensitively than SOA, as suggested by Murphy (2016).

Constraints on the accuracy of AMS ambient OA mass concentration measurements can also be obtained from aerosol hygroscopicity data, where comparisons between AMS organic concentrations and growth factors under subsaturated conditions have long been found to be consistent (Gysel et al. 2007). Values of the hygroscopicity parameter (κ_OA, Petters and Kreidenweis 2007) obtained from laboratory and field studies for oxidized OA are ∼0.05–0.3 (Gunthe et al. 2009; Lambe et al. 2011; Brock et al. 2016; Cheung et al. 2015; Petters et al. 2016). Values of κ_OA needed for hygroscopicity closure and assuming that AMS OA mass concentrations were erroneously high have been estimated with the method described in the appendix. Figure 3c shows that unphysical negative values of κ_OA would be necessary to explain the bulk hygroscopicity if OA mass was overestimated by a factor of 2–3. In fact, negative values of κ_OA are needed if OA is overestimated by just +38%, the upper limit of the reported AMS accuracy, much less than a factor of 2–3, while reasonable κ_OA are obtained for the actually reported OA mass concentrations. Combined with the volume closure as, for example, shown in Figures 3a and b, this further demonstrates that the biases in OA mass concentrations suggested in Murphy (2016) are inconsistent with independent measurements. We note that some of the κ_OA measurements and submicron aerosol volume measurements used here as well as the assumption that the latter are correct, were performed in publications by Murphy and coworkers (Froyd et al. 2010; Liao et al. 2015; Brock et al. 2016). When only those two previous results and assumptions are combined with the Murphy (2016) suggestion of a large overestimate of OA, a strong inconsistency emerges as illustrated in Figure 3c.

Ambient AMS data also do not support the assumption of a several-fold bias on the quantification of POA (consisting of more thermally stable molecules) relative to secondary organic aerosol (SOA; consisting of less thermally stable molecules), as suggested in Murphy (2016). For example, Docherty et al. (2011) evaluated the possible difference in AMS sensitivity to primary and secondary organics. Their data replotted in Figure 3d show that although the data are always in agreement within the combined instrumental uncertainties, it is possible that the response factor for POA (=RIE_POA × CE_POA) might be larger than that for SOA as already published by Docherty et al. (2011). The possibility of using a different response factor for POA compared to SOA has been under active discussion in the AMS community for many years in the context of reducing uncertainties in both CE and RIE. Jimenez et al. (2003) showed that EI cross sections (molecular properties, unrelated to the MW effect suggested in Murphy [2016]) of hydrocarbon molecules are about 50% higher than those of O and N containing organics. However, as shown earlier, the ratios of AMS response factors for two species can be different from ratios of their EI cross sections. CE factors may also differ between SOA and POA. Different response factors have not been widely adopted in standard AMS analyses of ambient OA as laboratory calibrations (Figure 2b) do not show consistent evidence for them, and as intercomparisons such as in Figure 3d (which encompass further effects such as CE) typically do not show large deviations beyond the estimated uncertainties as POA fractions vary. Work is in progress to better characterize the AMS response factor for different types of POA, but as indicated by the example in Figure 3d, ambient data are not consistent with changes in response factors that would result in biases in the range of 100%–200% and larger as implied in Murphy (2016).

In summary, we find no evidence of very large systematic biases in AMS ambient OA mass concentrations contrary to what is suggested in Murphy (2016). If biases in ambient OA mass concentration measurements were as large as a factor of 2–3, they would have been identified over a decade ago through comparisons, such as those just discussed. We note that systematic biases in sulfate and nitrate mass quantification of about a factor of 2 were identified in the very first intercomparisons of AMS with other instruments, and ultimately led to CE quantification (Drewnick et al. 2003, Hogrefe et al. 2004b; Allan et al. 2004b; Quinn et al. 2006; Matthew et al. 2008; Middlebrook et al. 2012).

Conclusions

While the M16 model correctly reflects certain physical features of some processes in the AMS, it is unable to predict multiple observed dependencies and other detailed features of AMS measurements. This is likely because it is too simple and it does not include important processes within the AMS. Furthermore, there is ample evidence from ambient analyses that AMS reported uncertainties are not underestimated, and large systematic biases in OA mass determination are generally not observed, contrary to what is predicted by the M16 model. Thus, the M16 model is not suitable to predict or evaluate the uncertainty or other performance metrics of AMS instruments used for ambient OA analysis, or to conclude that salts vaporize as complete molecules in the AMS.

Given the complexity of the sampling, vaporization and ionization processes in the AMS, the AMS community decided early on to base mass quantification on empirical calibrations rather than explicit modeling of these processes. Empirical calibrations underlie the quantification of the vast majority of atmospheric chemistry instruments (and also of broader analytical chemistry instruments), and the chemical composition of atmospheric aerosols is difficult to measure accurately by many methods. In the vast majority of ambient field measurements where the AMS was properly calibrated, operated without problems, and analyzed using standard procedures, the results have been overwhelmingly consistent with other colocated measurements of high quality. This, we believe, provides confidence in the robustness of the methods developed by the AMS community over the past 15 years.

We agree with Murphy (2016) that additional experimental and theoretical characterization studies are important for a fuller understanding of the physical processes in the AMS. The M16 model that describes free molecular flow within the ion source would have to be a component of any future larger model that successfully represents the full complexity of aerosol analysis by the AMS. We believe that a more important effect in practice is the uncertainty in the collection efficiency due to particle bounce, rather than the effect described in Murphy (2016). Work is ongoing in our laboratories, including the development and characterization of a novel capture vaporizer to address particle bounce, as well as studies of the detection of organic compounds in the AMS. Finally, we reiterate the need for direct calibration in laboratory experiments utilizing specific organic compounds or mixtures.

Acknowledgments

The authors are grateful to Simon Clegg for his support on running and interpreting the output of the E-AIM model, and to Markus Petters and Chuck Brock for useful discussions on constraints on composition from hygroscopicity measurements. The authors thank Kenneth Docherty, Philip Silva, Pedro Campuzano-Jost, Douglas Day, and Weiwei Hu for useful comments and discussions. The authors thank the many Aerodyne scientists, particularly in the Center for Aerosol & Cloud Chemistry (CACC), who have contributed to the development of the AMS and this work. The authors also thank the AMS community of users for their input on this topic as well as for many past and ongoing contributions to the development, characterization, and application of the different AMS instruments.

Funding

The authors thank NASA, EPA, and NSF, for supporting the collection and analysis of some of the data used here. JLJ acknowledges support from a College Scholar Award from the University of Colorado.

Appendix

Estimation of κ_OA values that provide hygroscopicity closure if OA was overestimated

The hygroscopicity parameter for particles containing an internal mixture of different chemical species can be estimated with the Zdanovskii, Stokes, and Robinson (ZSR) volume-mixing rule (Petters and Kreidenweis 2007):[A1] where ϵ_i is the volume fraction of species i in the dry particles, and κ_i is the hygroscopicity parameter for species i. For submicron particles dominated by ammonium-sulfate salts (AS) and OA, Equation (A1[A1] ) can be written explicitly as:[A2]

Since κ_AS is well known (∼0.6; Petters and Kreidenweis 2007), and measurements of the mass fractions of AS and OA are available from, for example, AMS measurements, Equation (A2[A2] ) can be solved to allow the estimation of κ_OA as:[A3]

In practice, these estimates are reliable only if the OA mass fraction is large (see, e.g., Figure 5 of Mei et al. 2013, and associated discussion). The volume mass fractions can be estimated from composition data as:[A4] where f_i and ρ_i are the mass fraction and density of component i, respectively. We use a density of 1.78 g cm⁻³ for ammonium-sulfate salts (Lide 1991), and we estimate the OA density from the parameterization of Kuwata et al. (2012).

While multiple published results could be used as examples, we chose recently published studies in two different regions of the United States. κ_OA was estimated with this method using AMS data in three studies with high OA mass fractions in the Southeast US during Summer 2013 from aircraft and ground platforms, as part of the SOAS, SENEX, and SEAC4RS field studies. Cerully et al. (2015) report κ_OA ∼0.15 for the cloud activation regime, while Brock et al. (2016) report κ_OA ∼0.05 for the subsaturated regime, and we use the average of 0.10 here. The same method was applied to data from a forested site in Colorado during the summer of 2011, yielding κ_OA ∼0.13 for the BEACHON-RoMBAS field study (Levin et al. 2014). Submicron nitrate concentrations were very small and submicron chloride was negligible at both locations (Fry et al. 2013; Brock et al. 2016).

Here we recalculated κ_OA using Equation (A3[A3] ) and reducing the OA concentration reported from the AMS by factors of 2 and 3, both of which are quoted by Murphy (2016) as possible overestimation factors. We assume that the measured hygroscopicity is correct, as well as that the volume concentration reported by collocated instruments is correct as routinely done in many papers including those by Murphy's group (Froyd et al. 2010; Liao et al. 2015), and thus any decrease in the OA mass concentration has to be compensated by an error of equal magnitude and opposite sign in the ammonium-sulfate salt concentration.

Results for these calculations are shown in Figure 3c and discussed in the main article text.