Development of a novel particle mass spectrometer for online measurements of refractory sulfate aerosols

2.1. Instrument description

Figure 1 depicts a schematic diagram of the rTDMS prototype. The basic configuration is similar to that of the particle trap–laser desorption mass spectrometer (PT-LDMS) (Takegawa et al. Citation2012). The main components include an inlet assembly (critical orifice, inlet valve, and aerodynamic lens unit (ADL)), a vacuum chamber with differential pumping by three turbo molecular pumps (V301, Varian), a particle collection unit, a quadrupole mass spectrometer (QMS) equipped with a cross-beam type electron ionization (EI) source (QMG700, Pfeiffer Vacuum), and a continuous wave CO₂ laser (wavelength: 10.6 μm, ULR-25, Universal Laser Systems). Aerosol particles are introduced into the vacuum chamber via the ADL, the structure of which is the same as that presented by Zhang et al. (Citation2004). The sample flow rate was controlled at ∼110 cm³ min⁻¹ (293 K, 1013 hPa) by the critical orifice with a pinhole diameter of ∼0.1 mm. The pressure in the upstream part of the ADL was 3.3 hPa when the sample air was loaded. The adjustment of the particle beam was visually performed by collecting nigrosine particles on a quartz filter in the chamber. The particle collection unit is composed of a graphite collector, copper holder, and rotary motion feedthrough. The key point of the rTDMS is the structure of the graphite collector, which is described in detail later. During particle loadings with the inlet valve open, the graphite collector faces the direction toward the inlet. After particles are collected, the rotary motion feedthrough turns to the opposite direction so that the graphite collector faces the ionizer of the QMS. The collected particles are vaporized by the CO₂ laser (irradiated through the ionizer cage), and evolved gas molecules are detected using the QMS. The electron energy for ionization was set at 70 eV. The m/z dependency of the transmission efficiency of ions in the quadrupole was evaluated by introducing a mixture of nitrogen and noble gases, as was done by Uchida, Ide, and Takegawa (Citation2019). We used a zinc selenide (ZnSe) focusing lens to increase the laser power density. Similarly to the PT-LDMS, the vaporization process included both direct heating of particles by laser absorption and indirect heating by thermal conduction from the graphite collector. Although the absorption of far infrared radiation around 10 μm is commonly found in many compounds (especially for organics), the radiation absorption by small particles (<1 μm) may not be efficient. The blackbody equivalent temperature of the outer surface of the graphite collector was measured using a radiation thermometer (Impac IGA 140, LumaSense Technologies, Inc.). Note that the radiation temperature was measured separately from the detection of ion signals to avoid the effects of thermal emissions from the EI filament. The distance between the graphite collector and the QMS ionizer was fairly long (∼50 mm) due to the assembly of various components in the prototype. This could be shortened in future improvements.

Figure 1. (a) Schematic diagram of the rTDMS. The main components of the rTDMS include an aerodynamic lens (ADL), a graphite collector and a copper holder, a CO₂ laser, and a QMS. (b) Detailed geometry of the graphite collector and the QMS. A cross-beam type ionizer is located ∼50 mm from the graphite collector. The CO₂ laser is irradiated through the ionizer cage. (c) Detailed geometry of the graphite collector.

Display full size

The major difference between the PT-LDMS and rTDMS is the method of particle collection and vaporization. The PT-LDMS employs a mesh-structured particle trap for the collection of particles (Takegawa et al. Citation2012). Although the particle trap enables efficient collection of aerosol particles upon high-velocity impact, the laser irradiation can significantly damage the trap when the desorption temperature becomes very high (>∼800 K). We tested various substrates including quartz, silicon carbide, and graphite, and found that graphite was the most suitable for our purpose because of its thermal durability at higher temperatures. To reduce the loss of particles due to bouncing during particle collection, we used a cup-shaped structure, as illustrated in Figure 1c. The cup-shaped structure may be beneficial for restricting the direction of plume expansion, which could be an important factor for quantification (Uchida, Ide, and Takegawa Citation2019). It may also be beneficial for high-temperature desorption because the closed structure is expected to suppress the radiative cooling of the inner side of the graphite collector.

2.2. Laboratory evaluation

The performance of the instrument was evaluated in the laboratory. Details of the experimental apparatus are presented in the Supplemental Information (SI), and only the key points are described below. The particle generation system includes a compressor, a Collison atomizer (3076, TSI, Inc.) or glass nebulizer (0-9478-01, Ishiyama-rikagaku glass, Tokyo, Japan), a diffusion dryer (3062, TSI, Inc.), and a differential mobility analyzer (DMA; 3080, TSI, Inc.). A condensation particle counter (CPC; 3022 A, TSI, Inc.) and the rTDMS were connected downstream of the particle generation system to measure the instrument sensitivity to various types of sulfate particles. We used the atomizer for AS, PS, SS, and MS and the nebulizer for CS. The contribution of hydrated water molecules was considered for preparing the solutions of these compounds and also for calculating the mass of particles downstream of the diffusion dryer (see Section 2.4). The sample flow rate through the DMA was 0.4 L min⁻¹ (sum of the flow rates of the rTDMS and CPC). The sheath flow rate for the DMA was set at 4.0 L min⁻¹. The particle diameter was set at 200 nm. Particle-free air (zero air: ZA) was also introduced into the rTDMS and CPC to correct for blank levels and also for evaluating potential artifacts. The DMA and CPC are combined as a scanning mobility particle sizer (SMPS) when measuring particle size distributions.

The collection efficiency for solid particles was tested using an oil coating system (Figure S1 in the SI), as was done by Takegawa et al. (Citation2012). The system consists of a stainless steel chamber for a liquid reservoir (inner diameter of 25 mm and length of 100 mm; hereafter referred to as “oil bath”) and a stainless steel tube for particle growth (inner diameter of 7.5 mm and length of 100 mm). Oleic acid vapor is supplied from the liquid reservoir in the bottom of the oil bath. The temperature of the oil bath was set at 90 °C, the length of the tube between the oil bath and the rTDMS was set at ∼0.4 m, and that between the oil bath and the CPC was set at ∼2.5 m. These conditions were determined to remove homogeneously nucleated oleic acid nanoparticles by evaporation and/or Brownian diffusion before reaching the CPC (Table S1 in the SI).

The rTDMS was operated with total measurement cycles ranging from 6 to 10 min, including the time for particle collection (2–6 min), laser irradiation and ion detection (2 min), and cooling of the graphite collector (2 min). The particle collection time was adjusted to obtain a range of particle mass loadings. The sample was irradiated by the CO₂ laser for a duration of 60 s at a laser power of ∼20 W. The inlet valve was kept closed for 4 min during the laser desorption analysis and cooling. The temperature of the graphite collector decreased to the lowest detectable temperature of the radiation thermometer (523 K) within 60 s after the CO₂ laser was turned off. If the temperature does not decrease to the thermal desorption temperature of AS within 2 min, it may yield evaporative loss of AS particles. We varied the cooling time from 2 to 5 min and confirmed that the cooling time of 2 min was sufficient for the detection of AS.

The laboratory experiments were performed using various types of sulfate particles. The test particles included single-component solid sulfate particles, oil-coated sulfate particles, and multi-component sulfate particles with various mixing states. Although these experiments were not sequentially performed, we show the results in this order for clarity of presentation.

First, we measured the sensitivities to single-component, solid sulfate particles (AS, PS, SS, and MS). We obtained two datasets (ion signals over a range of mass loadings) for each of the four types of sulfate particles. We did not observe detectable signals from CS particles, probably because of insufficient desorption temperatures; therefore, those data are not presented in this study. We also tested the collection efficiency of solid sulfate particles by comparing with the ion signals for oil (oleic acid)-coated sulfate particles. This experiment was based on the assumption that oil-coated particles could be collected with 100% efficiency.

Second, we tested the effects of the mixing state (internal and external mixtures) of AS and SS particles on the sensitivities. Because AS and SS particles may coexist in coastal regions near megacities, this experiment would reflect, to some extent, realistic conditions. The experiments for internally mixed AS and SS particles were performed by atomizing a mixed solution of AS and SS, whereas those for externally mixed AS and SS particles were performed by sequentially introducing AS and SS particles.

Third, we measured the ion signals for various types of internally mixed, multi-component sulfate particles (12 combinations of AS, PS, SS, and MS). The benefit of using internally mixed particles is that the possible variability in the particle collection efficiency among different species is canceled out. This experiment may not be representative of real atmospheric conditions because these compounds are unlikely to simultaneously exist in the form of internal mixtures. Nevertheless, as an initial stage of the instrument development, it is useful to test the performance of the instrument in these extreme cases.

2.3. Ambient measurements

Ambient measurements were performed in August 2020 at the ground-level floor of a building on the campus of Tokyo Metropolitan University (TMU) to test the overall performance of the rTDMS under real atmospheric conditions. TMU is located ∼30 km west of the center of Tokyo and ∼35 km northwest of Tokyo Bay. Details of the sampling setup are presented in the SI. The rTDMS was operated with a total measurement cycle of 10 min, including particle collection (6 min), laser irradiation and ion detection (2 min), and cooling (2 min). The operation was not fully automated, and the ambient sampling was performed for only a short period of time (∼2 h). The zero levels of ion signals were measured every hour by replacing the sample air with ZA. Furthermore, we sequentially introduced ambient aerosol particles after introducing laboratory-generated SS particles. This experiment was aimed at investigating potential interferences due to the presence of unknown materials in ambient air.

2.4. Data analysis

For a single sulfate compound i (i = AS, PS, SS, and MS), the “nominal” mass loadings of sulfate introduced into the rTDMS, W_i, were calculated as follows: (1) $$W_i=\frac{\pi }{6}{d}_m^3{\rho }_i f_i N_{\text{CPC}}\mathit{Ft},$$(1) where d_m is the mobility diameter of particles, N_CPC is the number concentration of monodisperse particles measured by the CPC, F is the sample flow rate, and t is the particle collection time. ρ_i and f_i are the bulk material density and the mass fraction of SO₄^2-, respectively, for compound i. The relative humidity (RH) downstream of the diffusion dryer was ∼5%. Considering that the efflorescence RHs are 35%, 60%, and 56% for AS, PS, and SS, respectively (Freney, Martin, and Buseck Citation2009; Seinfeld and Pandis Citation2006), these sulfate compounds were likely effloresced before reaching the DMA. MS particles were probably in the hydrated form (MgSO₄·7H₂O) because it is stable below 50 °C (Isa and Nogawa Citation1983). We assumed that AS, PS, and SS particles were in the dehydrated form and MS particles was in the form of MgSO₄·7H₂O both for single-component and multi-component sulfate particles. The term “nominal” signifies that the mass loadings of particles are calculated based on the mobility diameter without considering the particle collection and vaporization efficiencies. The nominal mass loadings can be directly derived from the DMA and CPC data. The effects of multiply charged particles were corrected using the same method as presented by Takegawa and Sakurai (Citation2011), but the correction term was omitted in EquationEquation (1)(1) $$W_i=\frac{\pi }{6}{d}_m^3{\rho }_i f_i N_{\text{CPC}}\mathit{Ft},$$(1) for simplicity.

The temporal profile of ion signals at a specific m/z value is integrated over the laser irradiation time to obtain the integrated ion signal for each compound i, Q_m/z,i, as was done by Takegawa et al. (Citation2012) and Ozawa et al. (Citation2016). We measured the ion signals at m/z 48 (SO⁺), 64 (SO₂⁺), and 80 (SO₃⁺) for AS and MS; m/z 39 (C₃H₃⁺, K⁺), 48 (SO⁺), and 64 (SO₂⁺) for PS; and m/z 23 (Na⁺), 48 (SO⁺), and 64 (SO₂⁺) for SS. We did not routinely monitor the ion signals originating from metallic elements for MS particles (m/z 24: Mg⁺; m/z 40: MgO⁺) because no detectable signal was observed at these m/z peaks. Note that there might be interferences from other unknown compounds (mostly from organics) with the selected m/z peaks under real atmospheric conditions, especially for m/z 39 and 80. The interferences with m/z 23, 48, and 64, which are the major m/z peaks used in this study, would be small considering the possible elemental compositions of ion fragments from organic compounds. Variations in Q_m/z,i due to drifts in the detector sensitivity were corrected by using m/z 14 signals, but the transmission efficiency for ions in the quadrupole was not considered for the calculation of Q_m/z,i. The “nominal” sensitivity of the rTDMS at m/z to compound i, S_m/z,i, is defined as the ratio of Q_m/z,i to W_i. Here, we also define the “true” sensitivity, $S_{\mathit{m/z,i}}^0,$ as the ratio of the integrated ion signal to the “true” mass of the vaporized sulfur compounds. Let ζ_i (≤1) be the ratio of the effective density (ρ_eff) to the bulk material density of single-component sulfate particles, and let ε_i (≤1) be the product of the collection and vaporization efficiencies for single-component sulfate particles. The effective density is used as an empirical parameter to represent the mass of non-spherical particles based on the mobility diameter (DeCarlo et al. Citation2004): (2) $${\zeta }_i=\frac{{\rho }_{\text{eff}}}{{\rho }_i}={\left(\frac{d_v}{d_m}\right)}^3,$$(2) where d_v is the volume equivalent diameter of particles. The “true” mass of vaporized sulfur compounds is expressed as ε_iζ_iW_i, and therefore we obtain: (3) $$S_{m/z,i}={\epsilon }_i{\zeta }_i{S}_{m/z,i}^0.$$(3) EquationEquation (3)(3) $$S_{m/z,i}={\epsilon }_i{\zeta }_i{S}_{m/z,i}^0.$$(3) indicates that the nominal sensitivity is smaller than the true sensitivity. Note that ζ_i is nearly equivalent to the dynamic shape factor (at power of −3) except for highly non-spherical particles (DeCarlo et al. Citation2004). While the absolute values of $S_{m/z,i}^0$ do not depend on the physical and thermochemical properties of aerosol particles, they depend on the spatial distributions of evolved gas molecules, which could be complicated functions of the geometry of the ionizer and vaporizer and the molecular weights and dynamics near the vaporization region (Ide, Uchida, and Takegawa Citation2019; Murphy Citation2016; Uchida, Ide, and Takegawa Citation2019).

For multi-component sulfate particles, we used the following equations: (4) $$\frac{\pi }{6}{d}_m^3=\frac{n}{N_A}\underset{i}{\sum }\frac{a_i{M}_i}{{\rho }_i},$$(4) (5) $$W=N_{\mathrm{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{f}_i{a}_i{M}_i,$$(5) where n is the nominal number of total sulfate molecules in each particle, N_A is the Avogadro number, and W is the nominal mass loadings of multi-component sulfate compounds introduced into the rTDMS. a_i and M_i are the molar fraction and the molecular weight, respectively, for compound i (Σa_i = 1). The value of n derived from EquationEquation (4)(4) $$\frac{\pi }{6}{d}_m^3=\frac{n}{N_A}\underset{i}{\sum }\frac{a_i{M}_i}{{\rho }_i},$$(4) is used to calculate the value of W by EquationEquation (5)(5) $$W=N_{\mathrm{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{f}_i{a}_i{M}_i,$$(5) . Similarly to EquationEquation (1)(1) $$W_i=\frac{\pi }{6}{d}_m^3{\rho }_i f_i N_{\text{CPC}}\mathit{Ft},$$(1) , the correction term for multiply charged particles is omitted in EquationEquations (4)(4) $$\frac{\pi }{6}{d}_m^3=\frac{n}{N_A}\underset{i}{\sum }\frac{a_i{M}_i}{{\rho }_i},$$(4) and Equation(5)(5) $$W=N_{\mathrm{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{f}_i{a}_i{M}_i,$$(5) .

Here we define the “true” number of vaporized total sulfur containing molecules, n₀. Let ζ (≤1) be the ratio of the effective density to the bulk material density of multi-component sulfate particles, and let ε (≤1) be the product of the collection and vaporization efficiencies of multi-component sulfate particles. We obtain n₀ = εζn. The observed ion signals for multi-component sulfate particles are compared with the “predicted” ion signals at m/z 48 or 64, $Q_{m/z}^{\text{pred}},$ which can be expressed as the linear combination of the product of the true sensitivity and true number of sulfate for each compound: (6) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n_0}{N_A}\underset{i}{\sum }{S}_{m/z,i}^0{f}_{\mathit{\text{i}}}{a}_i{M}_i.$$(6)

EquationEquation (6)(6) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n_0}{N_A}\underset{i}{\sum }{S}_{m/z,i}^0{f}_{\mathit{\text{i}}}{a}_i{M}_i.$$(6) is rearranged as: (7) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{S}_{m/z,i}{f}_{\mathit{\text{i}}}{a}_i{M}_i\frac{\epsilon }{{\epsilon }_i}\frac{\zeta }{{\zeta }_i}.$$(7) If we assume ε = ε_i and ζ = ζ_i for all compounds for simplicity, we obtain the following: (8) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{S}_{m/z,i}{f}_{\mathit{\text{i}}}{a}_i{M}_i.$$(8) EquationEquation (8)(8) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{S}_{m/z,i}{f}_{\mathit{\text{i}}}{a}_i{M}_i.$$(8) is actually used for calculating $Q_{m/z}^{\text{pred}}$ because the values of ε, ε_i, ζ, and ζ_i are unknown. EquationEquation (7)(7) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{S}_{m/z,i}{f}_{\mathit{\text{i}}}{a}_i{M}_i\frac{\epsilon }{{\epsilon }_i}\frac{\zeta }{{\zeta }_i}.$$(7) is used for estimating potential uncertainties associated with the assumption of ε = ε_i and ζ = ζ_i.

For ambient measurements, the mass concentration of a sulfate compound i (C_i) is calculated as the difference in the integrated ion signal between particle-loaded and ZA (net integrated ion signal; ΔQ_m/z,i): (9) $$C_i=\frac{\Delta Q_{m/z,i}}{S_{m/z,i}V},$$(9) where V is the volume of air sampled while the inlet valve is open.

3.1. Sensitivity to single-component sulfate particles

Figure 2 shows temporal profiles of the ion signals at m/z 23 (Na⁺), 39 (C₃H₃⁺, K⁺), 48 (SO⁺), 64 (SO₂⁺), and 80 (SO₃⁺) originating from monodisperse solid AS, PS, SS, and MS particles with mobility diameters of 200 nm, along with temporal profiles of the radiation temperature of the graphite collector that were separately measured. The maximum temperature reached ∼1200 K at a laser power of ∼20 W. The ion signals at m/z 23, 48, 64, and 80 from ZA are not displayed in the plots because the enhancements above the baselines (i.e., the ion signal levels before the laser irradiation) were much smaller than those from particle loadings. The above m/z peaks were found to be the major fragments originating from these sulfate compounds. The ion signals at m/z 48 and 64 for AS exhibited a spiked increase immediately after the laser was turned on (hereafter referred to as a prompt peak), whereas those for PS, SS, and MS showed delayed, broad increases (hereafter referred to as a delayed peak). This feature is probably explained by the difference in the thermal decomposition temperatures between AS and PS/SS/MS. The ion signals at m/z 39 were detected around prompt peaks during both particle-loaded and ZA measurements, as shown in Figure 2c, suggesting that they were primarily caused by artifacts (i.e., contamination of organic compounds on the graphite collector). Ion signals that likely originated from metallic elements were detected around delayed peaks for PS and SS particles (m/z 39: K⁺ and m/z 23: Na⁺, respectively), but not for MS particles (m/z 24: Mg⁺; m/z 40: MgO⁺).

Figure 2. Temporal profiles of (a) radiation temperatures of the graphite collector as a function of elapsed time since the laser on. The temperature data were obtained independently from the ion signal measurements, and the ensemble of lines represent repeated measurements on different days. (b)-(e) Temporal profiles of ion signals originating from single-component solid sulfate particles as a function of elapsed time since the laser on. It should be noted that the elapsed time may contain an error of ∼1 s because the laser operation was performed manually.

Display full size

Figure 3 shows scatterplots of the integrated ion signals versus the mass of sulfate for single-component, solid AS, PS, SS, and MS particles. The time window for the signal integration was set at 0–30 s. A good linearity (r² > 0.97) was found for all of these compounds. The error bars shown in Figure 3 represent the systematic errors originating from the particle collection efficiencies (see Section 4.3). The sensitivity at m/z 48, as determined from the linear regression slope of ion signals versus mass loadings, was found to be 2.5, 3.2, 3.1, and 3.7 pC ng⁻¹ for AS, PS, SS, and MS particles, respectively. Note that the sensitivity at m/z 48 to MS particles might be 1.1 pC ng⁻¹ if we assume a dehydrated form instead of MgSO₄·7H₂O. As described in Section 2.2, we obtained another dataset for single-component sulfate particles, and the sensitivity at m/z 48 was found to be 2.8, 3.0, 3.7, and 3.6 pC ng⁻¹ for AS, PS, SS, and MS particles, respectively.

Figure 3. Scatterplots of integrated ion signals versus mass loadings (SO₄ for m/z 48, 64, 80; K for m/z 39; Na for m/z 23) for (a) AS, (b) PS, (c) SS, and (d) MS particles. The slope, intercept, and r² values for the linear regression are indicated in the legend. The systematic errors originating from the collection efficiencies are shown as the error bars for selected data points (see Section 4.3).

Display full size

Figure 4 shows scatterplots of the integrated ion signals at m/z 48 versus those at m/z 64 for AS, PS, SS, and MS particles, with the incorporation of the relative m/z dependency of the transmission efficiency of ions in the quadrupole. Here, the relative m/z dependency represents the ratio of the ion transmission efficiency at m/z 64 relative to that at m/z 48, as determined from the noble gas measurements. We applied this correction for a quantitative comparison of the fragment ion ratio with earlier studies. The mass spectrum database provided by the National Institute of Standards and Technology (NIST) indicates that the fragment ratios of m/z 48 to 64 at an electron energy of 70 eV are 0.5 and 1 for sulfur dioxide (SO₂) and sulfuric acid (H₂SO₄), respectively. The mass spectrum for SO₃ is not available in the NIST database. Briggs, Hudgins, and Silveston (Citation1976) reported that the fragment ratios of m/z 48 to 64 originating from SO₂ and SO₃ were 0.991 and 1.73, respectively, indicating that the fragment ratio originating from SO₃ was larger by a factor of 1.75 (=1.73/0.991) compared to that from SO₂. The absolute values of the fragment ratios may not be directly compared between the NIST database and Briggs, Hudgins, and Silveston (Citation1976). We estimated the fragment ratio for SO₃ to be 0.88 by multiplying the fragment ratio for SO₂ (= 0.5) by 1.75. The data points for AS particles fall between the fragment ratio lines expected from H₂SO₄ and SO₃, suggesting that the major thermal decomposition products of AS were H₂SO₄ and SO₃. The data points for PS and SS particles were slightly above the fragment ratio line expected from SO₂, suggesting that the major thermal decomposition products of PS and SS were SO₂. The data points for MS particles were slightly below the fragment ratio line expected from SO₃, suggesting that the major thermal decomposition product of MS was SO₃. The difference in the absolute values of the sensitivities at m/z 48 among AS, PS, SS, and MS particles might be due to the difference in the fragment patterns of these compounds.

Figure 4. Scatterplots of m/z 48 to 64 ion signals for AS, PS, SS, and MS particles. The ratio of the transmission efficiency of m/z 48 to 64 ions in the quadrupole, as measured by using the mass spectra of noble gases, is incorporated in the ion signals for m/z 64. Fragment ion ratios reported in the literature are shown for comparison.

Display full size

Although the integrated ion signals for AS particles suggest the overall dominance of the above mechanism, the temporal profiles of m/z 48, 64, and 80 for AS particles in Figure 2b exhibit different behaviors during the evolution of temperature. These results suggest that the thermal decomposition products may be somewhat complicated and are highly dependent on the desorption temperature. Possible thermal decomposition processes for AS, PS, SS, and MS particles are discussed in Section 4.1 in comparison with previous results.

The collection efficiency of solid sulfate particles was tested by a comparison with the sensitivity at m/z 48 for oil-coated sulfate particles. Assuming that oil-coated particles can be collected with 100% efficiency, the ratio of the linear regression slopes (solid to oil-coated particles) can be interpreted as the collection efficiency for solid AS particles. The collection efficiency for solid AS particles was estimated to be 0.75. Similarly, the collection efficiencies for PS, SS, and MS particles were estimated to be 0.63, 0.71, and 0.70, respectively. The average of these values was found to be 0.70. The variability in the collection efficiencies among these particles was relatively small (<∼0.07). These values did not change significantly if we use the sensitivity values at m/z 64 (0.77, 0.62, 0.70, and 0.71 for AS, PS, SS, and MS, respectively).

3.2. Effects of the mixing state of sulfate particles on the sensitivities

Figure 5 shows the temporal evolution of the ion signals at m/z 23 (Na⁺), 48 (SO⁺), 64 (SO₂⁺), and 80 (SO₃⁺) for internally mixed AS and SS particles with a molar ratio of (AS, SS) = (1, 1) and externally mixed AS and SS particles. The ion signals at m/z 48, 64, and 80 from ZA are not displayed in the plots because the enhancements above the baselines were much smaller than those from particle loadings. The ion signals exhibited both prompt and delayed peaks, as expected. These results suggest that the ion signals for the mixture of AS and SS particles could be qualitatively explained by the sum of ion signals originating from single-component sulfate particles. The timings of the prompt and delayed peaks did not exhibit significant differences between the internal and external mixtures, whereas the shape of the prompt peaks varied depending on the mixing state of AS and SS particles.

Figure 5. Temporal profiles of ion signals originating from (a) internal mixture of AS and SS, (b) external mixture of AS and SS (loading of AS followed by that of SS), and (c) external mixture of SS and AS (loading of SS followed by that of AS).

Display full size

Table 1 summarizes the sensitivities at m/z 48 for internally and externally mixed AS and SS particles. The time window for the signal integration was set at 0–12 s (approximately the prompt peak) and 12–30 s (the delayed peak) for AS and SS particles, respectively, when we derived the sensitivities to AS and SS individually. Note that the split point of the prompt and delayed peaks (12 s in this case) varied depending on the heating rate of the graphite collector, which was not actively controlled. We empirically determined the split point considering the temporal evolution of the ion signals at m/z 23. The variability in the sensitivity at m/z 48 to total sulfate (AS + SS) for the three types of the mixing states was <∼20%, which was comparable to the difference in the absolute values of the sensitivities at m/z 48 to single-component AS and SS particles (2.5 and 3.1 pC ng⁻¹ for AS and SS, respectively; see Section 3.1). On the other hand, the individual sensitivities to AS and SS derived from the empirically determined split point significantly depended on the mixing state. This is probably because the shape of the prompt peaks from AS particles was affected by the presence of internally or externally mixed SS particles.

3.3. Quantification of multi-component sulfate particles

We measured the ion signals for internally mixed, multi-component sulfate particles. Figure 6 shows the temporal evolution of the ion signals at m/z 23 (Na⁺), 39 (C₃H₃⁺, K⁺), 48 (SO⁺), 64 (SO₂⁺), and 80 (SO₃⁺) for multi-component sulfate particles with the molar ratios of (AS, PS, SS, MS) = (1, 1, 0, 0), (1, 0, 1, 0), (1, 0, 0, 1), and (1, 1, 1, 1). Similarly to the single-component experiments (Figure 2), the ion signals at m/z 23, 48, 64, and 80 from ZA are not displayed in the plots because the enhancements above the baselines were much smaller than those from particle loadings. The ion signals exhibited both prompt and delayed peaks. Note that the ion signals at m/z 48 originating from PS, SS, and MS were likely superimposed on each other in the delayed peak.

Figure 6. Temporal profiles of the major ion signals originating from multi-component sulfate particles: (a) AS:PS = 1:1, (b) AS:SS = 1:1, (c) AS:MS = 1:1, and (d) AS:PS:SS:MS = 1:1:1:1.

Display full size

We also tested multi-component sulfate particles with various molar ratios. Figure 7 shows a scatterplot of measured versus predicted ion signals at m/z 48 for multi-component sulfate particles. The error bars in the figure represent the estimated systematic errors originating from the particle collection efficiencies (see Section 4.4). The predicted ion signals were calculated using the sensitivities at m/z 48 to single-component sulfate particles (i.e., 2.5, 3.2, 3.1, and 3.7 pC ng⁻¹ for AS, PS, SS, and MS particles, respectively). We found the distributions of the data points systematically varied with the molar ratio of PS. The linear regression slope tended to be smaller with increasing the molar ratio of PS. This tendency did not change when we used another set of sensitivities at m/z 48 (2.8, 3.0, 3.7, and 3.6 pC ng⁻¹ for AS, PS, SS, and MS particles) or the sensitivities at m/z 64.

Figure 7. Comparison between measured and predicted m/z 48 ion signals originating from multi-component sulfate particles with various molar ratios. The solid and shaded lines represent the linear regression lines for (AS:PS:SS:SS) = (1:2:1:1) and (1:3:1:1), respectively. The red line represents the linear regression line for the other groups. The systematic errors originating from the collection efficiencies are shown as the error bars for a selected data point (see Section 4.4).

Display full size

3.4. Ambient measurements

Figure 8a shows the temporal profiles of the ion signals at m/z 48 (SO⁺), 64 (SO₂⁺), and 80 (SO₃⁺) evolved from ambient aerosol particles. Similarly to the laboratory experiments, we detected prompt peaks at m/z 48, 64, 80 for ambient particles. We did not observe significant delayed peaks exceeding the baseline fluctuations at m/z 23, 39, 48, 64, and 80. Considering the location of the observation site, it is reasonable to assume that the observed sulfate aerosols were dominated by AS. Figure 8b shows a time series of the mass concentrations of non-refractory sulfate and PM_2.5. The mass concentration of PM_2.5 was observed at a national monitoring station ∼5 km from our measurement site. The sensitivity at m/z 48 or 64 determined from laboratory-generated solid AS particles was used to calculate the mass concentration of non-refractory sulfate. The non-refractory sulfate mass concentrations were approximately 30% of the PM_2.5 mass concentrations. Previous studies showed that the fraction of sulfate mass to total PM₁ mass measured by Aerodyne AMSs were ∼16%, 24%, and 38% in summer in suburban areas of Beijing, Prague, and Greece, respectively (Chen et al. Citation2020; Kubelová et al. Citation2015; Tsiflikiotou et al. Citation2019). Considering the difference between PM_2.5 and PM₁, our data may exhibit relatively larger sulfate fractions compared with those reported by the previous studies. The calculated non-refractory sulfate mass concentrations might have been overestimated if the collection efficiency for ambient particles was higher than that for the laboratory-generated solid AS particles, as indicated by the error bar in Figure 8.

Figure 8. (a) Temporal profiles of ion signals at m/z 48 (SO⁺), 64 (SO₂⁺), and 80 (SO₃⁺) originating from ambient aerosol particles. (b) Time series of the mass concentration of sulfate measured at TMU and that of PM_2.5 obtained from the nearest monitoring station on August 6, 2020. Note that the PM_2.5 values are preliminary. The systematic errors originating from the collection efficiencies are shown as the error bar for a selected data point (see Section 4.3). The dashed line represents the LOD values for sulfate mass concentrations estimated from the m/z 48 signals (Table 2).

Display full size

As described in Section 2.3, we introduced ambient aerosol particles sequentially after introducing laboratory-generated SS particles. Figure 9a shows the temporal profiles of the ion signals at m/z 23 (Na⁺), 48 (SO⁺), and 64 (SO₂⁺) evolved from ambient aerosol particles and laboratory-generated SS particles that were sequentially loaded on the graphite collector. The data show that ambient non-refractory sulfate particles and laboratory-generated SS particles can be separated in a similar manner as the laboratory experiments. We found a good linearity between the integrated ion signals and the mass loading of the laboratory-generated SS particles, regardless of the mass loading of ambient aerosols (Figure 9b). These results suggest that the interference of unknown materials with the detection of SS particles was small, at least for the sampling conditions tested.

Figure 9. (a) Temporal profiles of ion signals at m/z 23 (Na⁺), 48 (SO⁺), and 64 (SO₂⁺) originating from ambient aerosol particles and laboratory-generated SS particles obtained on 5 August 2020. We introduced ambient aerosol particles sequentially after introducing laboratory-generated SS particles. (b) Scatterplots of integrated ion signals versus mass loadings (SO₄ for m/z 48 and 64; Na for m/z 23) for SS particles with and without the loading of ambient aerosols.

Display full size

4.1. Thermal decomposition mechanisms

The thermal decomposition products of sulfate compounds inferred from the fragment patterns (Figure 4) are discussed in comparison with examples from the literature. Previous studies on the thermochemical properties of bulk AS reported that the thermal decomposition mechanisms of AS are rather complicated and are highly dependent on the desorption temperature and heating rate (e.g., Halsted Citation1970a). Laboratory studies on various types of aerosol mass spectrometers reported that thermal decomposition products of AS particles are highly dependent on the vaporization method (Chen et al. Citation2019; Hu et al. Citation2017; Ozawa et al. Citation2016; Voisin et al. Citation2003). These mechanisms may potentially explain the complicated behaviors of m/z 48, 64, and 80 signals originating from AS particles.

Halsted (Citation1970b) reported that the thermal decomposition of bulk PS at temperatures of 1180–1668 K in N₂ may take place via the following reaction: (R1) $${\mathrm{K}}_2{\text{SO}}_4\left(\mathrm{s}\right) \to 2\mathrm{K} \left(\mathrm{g}\right) +{ \text{SO}}_2\left(\mathrm{g}\right) +{ \mathrm{O}}_2\left(\mathrm{g}\right)$$(R1) The above expression implicitly includes the vaporization of solid K₂SO₄ and subsequent thermal decomposition. Halle and Stern (Citation1980) reported that the thermal decomposition of bulk SS at a temperature of 1123 K under vacuum conditions may take place via the following reaction: (R2) $${\text{Na}}_2{\text{SO}}_4\left(\mathrm{s}\right) \to 2\text{Na} \left(\mathrm{g}\right) +{ \text{SO}}_2\left(\mathrm{g}\right) +{ \mathrm{O}}_2\left(\mathrm{g}\right)$$(R2) Considering that the fragment ratio of m/z 48 to 64 approximately agreed with that of SO₂ (Figure 4) and ion signals originating from K⁺ and Na⁺ were detected for PS and SS particles, respectively, the above mechanisms can potentially explain our observed features.

Hildenbrand (Citation1979) reported that the thermal decomposition of bulk MS at temperatures of 900–1000 K may take place via the following reaction: (R3) $${\text{MgSO}}_4(\mathrm{s}) \to \text{MgO} (\mathrm{s}) +{ \text{SO}}_3(\mathrm{g})$$(R3) Considering that the fragment ratio of m/z 48 to 64 approximately agreed with that of SO₃ (Figure 4) and ion signals originating from Mg⁺ and MgO⁺ were not detected for MS, the above mechanisms can potentially explain our observed features.

4.2. Random errors

The random errors in the integrated ion signals and mass loadings depend on the absolute values of the integrated ion signals and mass loadings, respectively. A possible approach for estimating these errors is to introduce a constant concentration of particles into the system and determine the variability in the integrated ion signals measured by the rTDMS and the particle number concentrations measured by the CPC, which was not performed in this study. Nevertheless, relatively tight correlations of the ion signals versus the mass loadings for single-component sulfate particles (r² > 0.97) suggest that the random errors were relatively minor, at least for the range of particle concentrations tested in this study.

4.3. Systematic errors for single-component sulfate particles

The systematic errors in S_m/z,i for single-component sulfate particles are discussed. EquationEquation (3)(3) $$S_{m/z,i}={\epsilon }_i{\zeta }_i{S}_{m/z,i}^0.$$(3) indicates that the systematic errors in S_m/z,i may originate from the collection and vaporization efficiencies of the sulfate particles (ε_i) and the effective density of the sulfate particles (ζ_i), both of which are related to the x-axis of Figure 3. Other systematic errors may originate from the accuracy of the CPC detection efficiency, the DMA sizing, and the flow rate calibrations. These factors are not considered here because they are not specific to the detection of sulfate aerosols by the rTDMS.

The collection efficiencies for solid AS, PS, SS, and MS particles were estimated to be 0.75, 0.63, 0.71, and 0.70, respectively (see Section 3.1). Conversely, the systematic errors originating from the collection efficiencies were estimated to be 25%, 37%, 29%, and 30%. The error bars shown in Figure 3 (laboratory experiments) and Figure 8b (ambient measurements) are based on the above estimates.

The temporal profiles of the ion signals for AS, PS, SS, and MS particles suggest that these particles were completely vaporized by the 30 s laser irradiation (the ion signals decreased to the background level before the laser was turned off in most cases). Therefore, we assume that the systematic errors originating from the vaporization efficiencies were relatively minor as compared to those due to the collection efficiencies when we integrated the ion signals over the time period of 0–30 s. To test the stability of the vaporization efficiencies, we investigated the dependency of the sensitivities at m/z 48 on the laser power. We measured the sensitivities at m/z 48 to AS, PS, SS, and MS particles at a laser power of ∼18 W, which resulted in a blackbody equivalent temperature of ∼1180 K. The changes in the sensitivities at a laser power of 18 W relative to those at 20 W were found to be −20%, +14%, −3%, and −5% for AS, PS, SS, and MS particles, respectively. These values did not significantly change when we uses the sensitivities at m/z 64 (–27%, +15%, −2%, and −7% for AS, PS, SS, and MS, respectively). The relatively large decrease in the sensitivity at m/z 48 to AS particles was somewhat unexpected given that the thermal decomposition temperature for AS was the lowest among the sulfate compounds tested. This may be related to the complicated time-dependent changes in the relative intensities of m/z 48, 64, and 80 signals originating from AS (see Figure 2b). The relative change in the sensitivity due to the decrease in the laser power was found to be positive for PS particles (i.e., the lower laser power yielded the higher sensitivity). The mechanism for this tendency has not been identified at present.

The systematic errors originating from the effective density of the sulfate particles were difficult to quantify with the current experimental apparatus. We observed the morphology of 500–800-nm monodisperse sulfate particles by using a scanning electron microscope (SEM; VE-9800, KEYENCE, Japan; see Section S2 in the SI). The results suggest that the AS and MS particles were nearly spherical and the PS and SS particles were non-spherical (see Figure S3 in the SI). Earlier studies showed that AS particles generated from a TSI Collison atomizer (the same model as used in this study) exhibited nearly spherical shapes (Kuwata and Kondo Citation2009), which is consistent with our observations. These results suggest that the systematic errors due to the effective density were small for AS and MS particles but non-negligible for PS and SS particles (ζ_i < 1 in EquationEquation (3)(3) $$S_{m/z,i}={\epsilon }_i{\zeta }_i{S}_{m/z,i}^0.$$(3) ).

Finally, the accuracy of the position setting for analysis, which does not explicitly appear in EquationEquation (3)(3) $$S_{m/z,i}={\epsilon }_i{\zeta }_i{S}_{m/z,i}^0.$$(3) , might be important for the reproducibility of the measurements. The graphite collector was manually rotated to set the position facing the QMS ionizer, and there could be uncertainties in the setting of the rotation angle. We measured the sensitivity at m/z 48 to AS particles by intentionally misaligning the rotation angle (5° and 10° from the center position). The sensitivities at m/z 48 were 12% and 18% lower for the misalignment angles of 5° and 10°, respectively. A misalignment of 5° might be possible during normal operations. This uncertainty could be improved by precisely controlling the rotation angle.

4.4. Systematic errors for multi-component sulfate particles

EquationEquation (7)(7) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{S}_{m/z,i}{f}_{\mathit{\text{i}}}{a}_i{M}_i\frac{\epsilon }{{\epsilon }_i}\frac{\zeta }{{\zeta }_i}.$$(7) indicates that the systematic errors in $Q_{m/z}^{\text{pred}}$ at m/z 48 were related to the uncertainties in the ε/ε_i and ζ/ζ_i ratios. Note that ε and ε_i are the collection/vaporization efficiencies of multi-component and single-component sulfate particles, respectively. As described in Section 3.1 (oil-coated experiments), the average collection efficiency for AS, PS, SS, and MS particles was 0.70, and the variability in the collection efficiencies among the different sulfate compounds was <∼0.07. Assuming that the collection efficiency for multi-component particles can be approximated as the average collection efficiency of 0.70, the systematic errors in the ε/ε_i ratios due to the particle collection efficiencies were estimated to be ∼10%. The error bars shown in Figure 7 are based on the above estimates. Similarly to the estimates for single-component sulfate particles, we assume that the systematic errors in the ε/ε_i ratios due to the vaporization efficiencies were relatively minor as compared to those due to the collection efficiencies when we integrated the ion signals over the time period of 0–30 s.

The systematic errors originating from the effective density of the sulfate particles for multi-component sulfate particles were more complicated than those from single-component sulfate particles. The SEM results suggest that PS-rich multi-component particles had significant non-spherical shapes, while non-PS-rich multi-component particles were nearly spherical (see Figure S3 in the SI). These results suggest that EquationEquation (8)(8) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{S}_{m/z,i}{f}_{\mathit{\text{i}}}{a}_i{M}_i.$$(8) (i.e., the approximation of ζ/ζ_i = 1 in EquationEquation (7)(7) $$Q_{m/z}^{\text{pred}}=N_{\text{CPC}}\mathit{Ft}\frac{n}{N_A}\underset{i}{\sum }{S}_{m/z,i}{f}_{\mathit{\text{i}}}{a}_i{M}_i\frac{\epsilon }{{\epsilon }_i}\frac{\zeta }{{\zeta }_i}.$$(7) ) may lead to an overestimation of the $Q_{m/z}^{\text{pred}}$ values for PS-rich particles (possibly ζ < 1 for multi-component particles, ζ_i ∼ 1 for single-component AS and MS particles, and ζ_i < 1 for PS and SS particles) and an underestimation of the $Q_{m/z}^{\text{pred}}$ values for non-PS-rich particles (possibly ζ ∼ 1 for multi-component particles, ζ_i ∼ 1 for single-component AS and MS particles, and ζ_i < 1 for PS and SS particles). These effects are qualitatively consistent with the observed tendency that the slope of the measured versus predicted ion signals for PS-rich particles was smaller than that for non-PS-rich particles (Figure 7).

The systematic errors also originate from the assumption for the hydrated forms of the sulfate particles (dehydrated form for AS, PS, and SS and hydrated form for MS). The effects would be more complicated for multi-component particles compared to single-component particles. We have not precisely estimated the effects.

4.5. Limit of detection

The limit of detection (LOD) for sulfate aerosols can be estimated as the equivalent concentration at three times the standard deviation (3σ) of the integrated ion signals for repeated ZA measurements (9–12 samples for each compound). The equivalent concentration is calculated as the ratio of the 3σ value to the product of the sensitivity at m/z 48 or 64 and sample air volume, as indicated by EquationEquation (9)(9) $$C_i=\frac{\Delta Q_{m/z,i}}{S_{m/z,i}V},$$(9) . The time window for the ion signal integration was set at 0–30 s. The LOD values estimated by this method are listed in Table 2. The LOD values depended on the m/z peaks selected, probably because the background ion signals varied with m/z. Based on the estimated LODs, the current system could be used for measuring relatively high concentrations of sulfate in urban air, as shown in Figure 8, whereas the sensitivities at m/z 48 or 64 need to be improved by approximately one order of magnitude for measuring low concentrations of sulfate in remote areas.

4.6. Implications for future improvements

There are several important issues to be addressed for ambient measurements. They include the improvement of the LODs by increasing the sensitivities and the quantitative separation of the sulfate compounds. As shown in Section 3.2, ion signals at m/z 48 and 64 originating from non-refractory and refractory sulfate compounds might overlap in some cases, depending on the mixing state of the particles. A possible method for better separation is an active control of the heating rate of the graphite collector. Furthermore, as shown in Section 3.3, the ion signals at m/z 48 and 64 originating from the tested refractory sulfate compounds were not separately detected. The ion signals from the metallic elements of PS and SS particles (m/z 39: K⁺ and m/z 23: Na⁺, respectively) might be used for estimating the relative contributions from the individual refractory sulfate compounds. Other possible compounds in atmospheric aerosols that can yield these alkali metal signals would be sodium chloride (NaCl), sodium nitrate (NaNO₃), and potassium nitrate (KNO₃). Characterizing the ion signals originating from these compounds is necessary to establish a practical method for quantitatively separating sulfate aerosols under ambient conditions. Finally, an alternative ADL for transmitting supermicron particles is desirable for measuring refractory sulfate aerosols in the coarse mode.