Introduction
Secondary organic aerosols (SOAs) represent a major fraction of the fine
particulate matter mass (e.g., Jimenez et al., 2009), thus contributing to
the physicochemical properties of aerosols and to their impact on human
health, climate and visibility. SOAs are produced by condensation of low
volatile organic species formed during gaseous oxidation of emitted volatile
and intermediate volatility organic compounds (VOCs and IVOCs) (e.g., Kroll and
Seinfeld, 2008). Large uncertainties still exist in our understanding of
processes leading to SOA formation. To assess and improve our knowledge of
SOA formation, atmospheric chambers are widely used to perform controlled
experiments of SOA formation from various VOCs and IVOCs. These experiments
provide kinetic and thermodynamic data, i.e., kinetic constants, branching
ratios and partitioning coefficients (e.g., Atkinson and Arey, 2003; Aschmann et
al., 2011), needed to design deterministic SOA models. Experiments performed
in atmospheric chambers also provide an ideal data set for the evaluation of
deterministic SOA formation models (e.g., Camredon et al., 2010; Valorso et
al., 2011; Jenkin et al., 2015) and the development of SOA formation
parameterizations for chemical transport models (CTMs) at regional and global
scale (e.g., Zhang and Seinfeld, 2013; Cappa and Wilson, 2012; Donahue et
al., 2011; Santiago et al., 2012). The presence of walls in atmospheric
chambers constitutes, however, an artifact for the study of SOA formation.
Particles, and thus SOAs, formed during an experiment are known to be deposed
on chamber wall surfaces (e.g., McMurry and Grosjean, 1985). This loss of
particles has been shown to depend on the size and the concentration of the
particles (Bowman et al., 1997). SOA mass is usually corrected assuming that
the loss rate during an experiment is a first-order process (e.g., Presto and
Donahue, 2006; Pathak et al., 2007) possibly depending on particle size
distribution (e.g., Cocker et al., 2001). Recently, few studies have shown
that low volatility gaseous species can also be lost onto the chamber wall
surfaces (e.g., Kokkola et al., 2014; Loza et al., 2010; Matsunaga and
Ziemann, 2010; Yeh and Ziemann, 2014, 2015; Zhang et al., 2014a, 2015). This
loss of gaseous organic species on the chamber walls is, however, not yet well
characterized.
Matsunaga and Ziemann (2010) have investigated the gas-wall partitioning of
C8–C16 n-alkanes and 1-alkenes, C8–C13 2-alcohols
and 2-ketones in Teflon FEP chambers. They observed that the gas–wall
partitioning of organic species is a reversible process and that equilibrium
can be described by a condensation process. They proposed a parameterization
to represent the gas–wall partitioning in which (i) the sorption of gaseous
organic species to the wall and their desorption from the wall back to the
gas phase are represented by a first-order rate constant (called kgw
and kwg, respectively, hereafter, see Fig. 1) and (ii) the gas–wall
partitioning equilibrium follows the Raoult's law, i.e., walls being treated as a
phase into which the organic compounds can partition. The temporal variation
of the concentration of a semi-volatile organic compound (SVOC) in the gas phase due to the gas–wall
transfers is thus given by the following expression:
dSVOCgdtgas–wall=-kgwSVOCg+kwgSVOCw,
where [SVOC(g)] and [SVOC(w)] are the concentration of the SVOC in the gas phase
and on the wall at a given time respectively. The rate constants are thus
linked according to the following equation:
kgwkwg=RTPvap×CwMwγw,
where R is the
ideal gas constant, T is the temperature and Pvap is the vapor
pressure of the species. Cw and Mw are an equivalent
organic aerosol mass concentration and an equivalent organic aerosol mean
molar weight associated with the Teflon film, and γw is the
activity coefficient of the species in the Teflon film. Values for
Cw / (Mwγw) are empirically derived from
chamber observations and Matsunaga and Ziemann (2010) reported values of 9,
20, 50 and 120 µmole m-3 for n-alkanes, 1-alkenes, 2-alcohols
and 2-ketones, respectively. In these series of experiments, the
characteristic times to achieve gas–wall equilibrium
(τ*= 1/(kgw+kwg)) ranged from less than 8 min to
1 h. Additional measurements for this chamber showed values of τ* up
to 100 min with large variability owing to measurement uncertainties (e.g.,
Yeh and Ziemann, 2015). For the studied species, the fraction remaining in
the gas phase at equilibrium ranged from 0.4 to 0.9, meaning that
kwg and kgw are of the same order of magnitude (Eq. 2).
In the Matsunaga and Ziemann (2010) parameterization for gas–wall transfers,
it is considered that the gas transfers towards the walls are driven by the
turbulence inside the chamber and interfacial mass transport through the
walls; thus, kgw is a constant, independent of the compound
structure (e.g., Yeh and Ziemann, 2015).
Schematic representation of gas–particle and gas–wall mass transfers
of a semi-volatile organic produced during gaseous oxidation
of VOCs.
![]()
This wall loss is in competition with gas–particle partitioning (see Fig. 1)
and the distribution of organic compounds between the gas phase, the
particle phase and the walls depends also on the characteristic times
associated with the gas–particle mass transfer. Assuming this process is
described by two first-order rate constants, the loss rate constant of
gaseous organic species to the particle and the evaporation rate constant of
condensed species are called kgp and kpg, respectively. The
temporal variation of the concentration of a SVOC in the gas phase due to
the gas–particle transfers is thus given by the following expression:
dSVOCgdtgas–particle=-kgpSVOCg+kpgSVOCp,
where [SVOC(p)] is the concentration of the SVOC in the particle phase. Assuming
the sorption of gaseous organic species to the particle is to be limited by gas
phase diffusion, the first-order loss rate constant of gaseous organic
species to the particle, kgp, can be expressed as
kgp=4πDgrpCp,
where Dg is the species gas phase diffusivity, rp the particle
radius and Cp the number of particles per unit volume of air (e.g.,
Seinfeld and Pandis, 2006). This value for kgp is an upper limit if the
gas to particle transfers are limited by interfacial mass transfer and/or
particle phase diffusion (Shiraiwa and Seinfeld, 2012). The wall uptake
could then be underestimated. At thermodynamic equilibrium, the gas-particle
partitioning is expected to follow the Raoult's law (e.g., Pankow, 1994). The
rate constants are thus linked according to the following equation:
kgpkpg=RTPvap×CaerMaerγaer,
where Caer is the organic aerosol mass concentration, Maer is the
organic aerosol mean molar weight and γaer is the activity
coefficient of the species in the particle phase.
Figure 2 shows the molar fraction ξphase of organic compounds in
the gas phase, the particle phase and the walls as a function of their
volatility (with ξphase=nphase/ntot). A continuous
distribution of species was considered, having volatilities in the
10-12–10-2 atm range and assumed to be in the gas phase at
t= 0. The gas–wall transfer rate constants were calculated using the
parameterization provided by Matsunaga and Ziemann (2010) for ketones. Here
we used an intermediate value for the characteristic time for gas to wall
transfer (τgw=1/kgw) of 30 min to examine the partitioning
of the species reported in Fig. 2. Calculations were made using typical
conditions encountered in experiments performed in the chamber for which
Matsunaga and Ziemann (2010) designed the gas–wall parameterization (e.g.,
Lim and Ziemann, 2009a, b; Matsunaga et al., 2009):
T= 298 K, Caer= 2000 µg m-3,
rp= 0.1 µm and
Cp= 7 × 104 cm-3. The distributions were computed
using Dg= 0.03 cm2 s-1, Maer= 200 g mol-1 and
γaer= 1. These conditions lead to the following mass transfer
rate constants: kgp= 2.6 × 10-1 s-1,
kpg= 1.1 × 106 Pvap s-1 atm-1,
kgw= 5.6 × 10-4 s-1 and kwg= 1.9 × 10-4Pvap s-1 atm-1.
Results are shown at various times:
6 min (time length of the alkene experiments presented by Matsunaga et al., 2009), 1 h
(time length of the alkane experiments in Lim and Ziemann,
2009a, b), 20 h (upper value for the length of a chamber
experiment) and at equilibrium. At 6 min, species with volatility greater
than 10-4 atm remained in the gas phase and those having volatility
less than 10-8 atm totally partitioned into the particle
phase. Between these two values of saturation vapor pressure, the fraction
of organic species lost into the wall is greater than 10 %, with a
fraction in the wall reaching 15 % for organic species having a saturation
vapor pressure equal to 10-6 atm. At 1 h, wall losses increase
substantially. The fraction of species lost on the wall reaches at least
10 % in the 10-8–10-5 atm vapor pressure range. This fraction
peaks at 65 % for species with vapor pressures around 10-6 atm. At
20 h, the volatility range showing a substantial uptake by the wall spans 4
orders of magnitude, from 10-9 to 10-5 atm. At equilibrium, the
fraction lost to the wall exceeds 90 % for species having a vapor pressure
under 10-7 atm. For high volatility species, equilibrium is achieved
on a short timescale (below 6 min): the gas–particle–wall partitioning is
under a thermodynamic control. By contrast, for low volatility species, the
distributions change with time and the gas–particle–wall partitioning is
under kinetic control. Note that the range of SVOC+OH rate constants is
between 10-12 and 10-10 molec-1 cm3 s-1. For
typical experiments, OH concentrations lie between 106 and 107 molec cm-3.
The chemical loss rates due to OH oxidation therefore range
between 10-6 and 10-3 s-1. The gas to wall transfer rate of a
given SVOC could be of the same order of magnitude as the OH loss rates
(for τgw= 1 h, kgw= 2.8 × 10-4 s-1);
wall loss could thus be in competition with OH oxidation. Wall loss is
therefore expected to significantly impact SOA measurements, being in
competition with the gas–particle partitioning of SVOCs and acting as a sink
for gaseous intermediates that ultimately produce SOA contributors after
additional oxidation steps.
Time evolution of the fraction in the gas, particle and wall phases
as a function of the vapor pressure for a continuous distribution of species
considered to be initially only in the gas phase. Grey zones represent the
volatility range impacted by wall losses at a molar fraction higher than
10 %.
![]()
The objectives of this study are (i) to quantify the potential impact of
chamber walls on the loss of gaseous organics and therefore SOA formation,
(ii) to explore the structure of organic species mainly impacted by this
loss and (iii) to assess our current understanding of SOA formation from
various structures of VOCs and IVOCs. The methodology used here consists of
simulating a selected set of experiments using an explicit description of
SOA formation (i.e., gaseous oxidation and gas–particle partitioning). The
Generator for Explicit Chemistry and Kinetics of the Organics in the
Atmosphere (GECKO-A) (e.g., Aumont et al., 2005; Camredon et al., 2007;
Valorso et al., 2011) has been used to generate the explicit gaseous
chemical schemes and the properties required to describe the gas–particle
partitioning (i.e., the saturation vapor pressure of all species). Alkanes
and alkenes are an important class of compounds to study fundamental
chemical reactions involved in atmospheric oxidation and the effects of
molecular structure on the formation of low volatility compounds, which could
form SOAs. The set of smog chamber experiments performed in the Ziemann lab
at the UC Riverside Air Pollution Research Center (APRC) has been selected
as it includes a large set of alkane and alkene structures (of varying
carbon chain length and including cyclic, linear and branched species) (e.g.,
Lim and Ziemann, 2009a, b; Matsunaga et al., 2009).
Furthermore, the Matsunaga et al. (2009) parameterizations of gaseous organic
species wall loss have been fitted for this specific chamber (e.g., Matsunaga
and Ziemann, 2010). Experimental conditions are described in Sect. 2.
Section 3 is devoted to describing the modeling tool and setup. In Sect. 4,
the results of the comparison between the measured and simulated SOA yields
are presented and the GECKO-A chemical mechanisms are used to identify SOA
products likely affected the by gas–wall partitioning process.
Model description and simulation setup
Gaseous and condensed phase chemical schemes
An explicit description of long-chain hydrocarbon oxidation processes up to
the final production of CO2 involves millions of reactions (e.g.,
Aumont et al., 2005). Such explicit chemical schemes are too large to be
written manually. The GECKO-A computer program has been developed to
overcome this difficulty. This tool is used here to generate explicit
chemical schemes for the long-chain hydrocarbons.
The GECKO-A tool generates chemical schemes for a given precursor on the
basis of a prescribed protocol. Reaction pathways and kinetic data are
assigned from experimental data if available or structure activity
relationships (SARs) if not. The protocol implemented in GECKO-A is described
in Aumont et al. (2005), with chemistry updates performed by Aumont et
al. (2012, 2013). The evolution of alkoxy radicals through decomposition,
isomerization or stabilization chemical pathways characterizes the
functionalization or the fragmentation of an organic species during its
oxidation. Therefore, the fate of alkoxy radicals determines SOA formation.
The SAR developed by Atkinson (2007) is used in GECKO-A to estimate the rate
coefficients for alkoxy radical isomerization based on experimental data.
Vereecken and Peeters (2009, 2010) have recently developed SARs based on
quantum chemical calculations to estimate the rate coefficient of both
alkoxy radical decomposition and H-migration. These new SARs for alkoxy
radical decomposition and isomerization have also been implemented in
GECKO-A.
No reactions were previously assumed in the condensed phase. Dihydrofurans
(DHFs) have been found in equilibrium with 1,4-hydroxycarbonyls (1,4HCs) in
the gas phase (Cavalli et al., 2000; Martin et al., 2002). These oxidation
products of DHFs have been shown to contribute largely to the composition of
SOAs formed during alkane and alkene oxidation (e.g., Gong et al., 2005; Lim
and Ziemann, 2009a; Zhang et al., 2014b). DHFs are produced by the
heterogeneous cyclization of 1,4HCs into cyclic hemiacetals (CHAs) that will
dehydrate to form DHF (e.g., Martin et al., 2002; Atkinson et al., 2008). DHF
formation is expected to be significant during SOA formation from alkane and
alkene oxidation as 1,4HC are produced at high yields during the first
gaseous oxidation step of these precursors. The formation of DHF is therefore
expected to significantly influence the SOA mass and composition. The
heterogeneous formation of DHF has therefore been added to the GECKO-A
protocol.
The DHF formation protocol is designed based on the Lim and Ziemann (2009c)
mechanism, which considers (i) the cyclization of the condensed 1,4HCs into
CHAs and (ii) the dehydration of the CHAs forming the DHFs with (iii) hydration
allowed for the reverse reaction (see Fig. 3). Lim and Ziemman (2009c)
provided an effective uptake coefficient of 0.5 for the 1,4HC(g) → CHA(p)
cyclization meaning that the rate of the transformation is collision
limited. Here, we considered explicitly the 1,4HC(g) ↔ 1,4HC(p) mass transfers and assumed that the 1,4HC(p) → CHA(p)
cyclization is fast (see Fig. 3). This model configuration is an upper value
for the cyclization process of 1,4HCs. Furthermore, Lim and Ziemann (2009c)
optimized a value of 10-3 s-1 for the dehydration rate constant
and values between 0.002 and 0.15 s-1 for the hydration rate
constant. The rate constant for the hydration reaction is set in GECKO-A
protocol to the higher value obtained for the set of n-alkane experiments
(that for tetradecane under dry conditions), kH2O= 0.15 s-1. This
process has been shown experimentally to occur not only in the organic particle phase
but also on Teflon wall surfaces (Atkinson et al., 2008). The DHF formation
mechanism was thus considered in the GECKO-A protocol in the particle phase
as well as on the walls. Gas–particle and gas–wall mass transfers and gaseous
chemistry of 1,4HC, CHA and DHF were treated automatically with the GECKO-A
protocol (see Fig. 3).
Representation of DHF formation in the GECKO-A protocol.
![]()
Dynamic representation of mass transfers
In the previous version of GECKO-A, the partitioning between the gaseous and
the aerosol phases was described assuming equilibrium (e.g., Camredon et al.,
2007; Lee-Taylor et al., 2015). In order to take into account the
competition between the aerosols and walls for the condensation of gaseous
organic compounds, the partitioning has now been implemented dynamically.
The gas–particle transfer is considered limited by the gas phase diffusion.
Equation (4) is used to represent the rate constant of the gas–particle mass
transfer, kgp. The species gas phase diffusivity, Dg, is derived
using the kinetic theory of gases, Dg= < ci > λ/2,
where < ci > is the mean velocity of gaseous
species i and λ is the mean free path (Seinfeld and Pandis, 2006).
During the experiments, SOA mass formed from nucleation process accounts for
less than 10 % to the total measured SOA mass (see Lim and Ziemann, 2005).
We thus considered here that (i) the particle number Cp remains
unchanged during the course of an experiment and equal to the number of
initial DOS seeds and (ii) the particle radius rp was growing because
of the aerosol mass increase, rp, being calculated assuming spherical
particles and a particle density of 1 g cm-3 (Lim and Ziemann, 2009b;
Matsunaga et al., 2009). The particle–gas rate constant kpg was
calculated according to Eq. (5). Vapor pressures of organic species were
estimated using the Nannoolal et al. (2008) method combined with the
Nannoolal et al. (2004) method for boiling-point estimates. Particles were
considered as a homogeneous and ideal phase (i.e., γi= 1).
Gas–wall partitioning was implemented in the GECKO-A protocol according to
the parameterization of Matsunaga and Ziemann (2010) (see Eq. 2).
Sensitivity tests were performed on both τgw and
Cw/(Mwγw) values. Three sets of simulations were
therefore performed for τgw using 10, 30 and 60 min to
encompass the observed values. Similarly, two sets of simulations were
performed for the Cw/(Mwγw) parameter using either the
values given in Matsunaga and Ziemann (2010) for ketones (120 µmole m-3)
or 2-alcohol (50 µmole m-3) for all oxygenated
species. For the parent hydrocarbon, the values of 9 and 20 µmole m-3
were used for alkanes and alkenes, respectively.
Reduction of chemical schemes
The size of the explicit oxidation scheme generated by the GECKO-A tool
rises exponentially with the carbon number of the precursor (Aumont et al.,
2005). Solving fully explicit chemical schemes requires excessive computer
resources. The chemical schemes were therefore shrunk using the
simplifications discussed below.
Experiments were carried out under high NOx levels (∼ 10 ppm).
The reaction of peroxy radicals with NO is dominant compared to those
with HO2 and RO2. The reactions of RO2 with HO2 and with
other RO2 were therefore ignored in the chemical schemes. The length of
experiments was short (6 min for alkenes or 1 h for alkanes). The
contribution of organic species formed after several oxidation steps can
thus be neglected. The generated chemical schemes include the description of
the formation of organic species up to five generations. Moreover organic
species having a low vapor pressure can be considered as mainly present in
the condensed phases. The inclusion of their gaseous chemistry in the
chemical schemes is thus not needed. Gaseous reactions were not considered
in the chemical schemes for species having a vapor pressure lower than
10-13 atm, being negligible in the gaseous phase under these
experimental conditions (Caer varying between 100 and
6000 µg m-3). Gas–particle and gas–wall mass transfers were considered for
stable (i.e., non-radical) species only. Figure 4 shows the number of stable
and total organic species in these reduced chemical schemes, as a function of
the number of carbon atoms in the parent compound for the studied
precursor's families.
Number of stable (blue) and total organic species (red) in the
reduced chemical schemes generated with GECKO-A as a function of the number
of carbons in the parent compound for the studied precursor's families.
![]()
Simulation setup
The explicit chemical schemes generated by GECKO-A were implemented in a box
model. Time integration of chemical schemes and mass transfers is solved
using the two-step solver (Verwer et al., 1994, 1996). Photolysis frequencies
for inorganic and organic species for a black-light spectrum were calculated
using cross section and quantum yields described in Aumont et al. (2005). The
photolysis of methyl nitrite was calculated using a quantum yield equal to
0.33 (Heicklen, 1988) and the absorption cross section measured by Taylor et
al. (1980). All the photolysis frequencies
were scaled based on the measured JNO2. Simulations were
initialized with the measured concentrations listed in Table 1. Similar to
the experimental conditions, a 60 min period in the dark is first simulated
to enforce gas–wall equilibrium for the parent hydrocarbon before starting
the oxidation.
Simulated temporal evolutions of the decays of n-alkane in the gas phase are
represented in Fig. 5 with and without considering gas–wall partitioning.
Linear alkanes having more than 13 carbon atoms show a drop in the gaseous
concentration early in the simulation mainly due to the chamber wall
partitioning. As expected, this drop increases with the chain length due to
the decrease of vapor pressure. Similar trends are observed for the other
series of precursors. These behaviors indicate that a significant fraction
of the initial hydrocarbon load can be removed into the wall for IVOC
precursors. Furthermore, it is noteworthy that for these species the gas–wall
partitioning equilibrium is achieved within a few minutes, a timescale short
enough to not observe that process during most of the chamber experiments.
Simulated temporal evolutions of the decays of n-alkanes in the gas
phase without (solid line) and with (dashed line) gas–wall partitioning. The
dotted line at 60 min corresponds to the time at which the lights are turned
on.
![]()
The measured mass of ΔHC was used to assess the
simulation setup. Figure 6 shows the comparison between the measured and
simulated ΔHC mass. The values are displayed by chemical family and
as a function of carbon chain length or number of methyl groups on the
carbon backbone. Within each family, the amount of reacted compound
typically grows with the size of the carbon skeleton. This behavior is
consistent with the corresponding increase of the hydrocarbon+OH rate
constant with the size of the carbon skeleton (e.g., Kwok and Atkinson,
1995). For the whole set of simulations, ΔHCs are well reproduced by
the model. Hence, the simulation setups are considered suitable to represent
experimental conditions.
Comparison between the measured (black) and the simulated (red)
ΔHC mass for the different chemical families as a function of carbon
chain length or number of methyl groups of the carbon backbone.
![]()
Results and discussions
Comparison between measured and simulated SOA yields
SOA yield is defined as the ratio of SOA mass produced to the
mass of ΔHC. A comparison of the experimental
and the simulated final SOA yields is presented in Fig. 7 (i.e., after 1 h of
experiment for alkane and 6 min for alkene). Results are reported as a
function of the number of carbon atoms or branching in the parent
hydrocarbon. Experimental data are displayed as given in the experimental
papers by Lim and Ziemann (2009b), Matsunaga (2009) and Matsunaga et al. (2009).
These published SOA yields were corrected from aerosol wall loss,
considering a first-order process and no interaction between the gas and the
wall deposited particles.
Comparison between measured (black) and simulated SOA yields for the
different chemical families as a function of carbon chain length or presence
of methyl groups on the carbon backbone. Simulations are shown without wall
loss (red) and with wall loss using a value of
Cw/(Mwγw)= 120 µmole m-3 for
all secondary species and a τgw of 10 min (yellow), 30 min (blue)
or 60 min (green).
![]()
Simulations were performed with and without wall loss taken into account for
the organic vapors (see Fig. 7). Simulations reported in Fig. 7 were
obtained using Cw/(Mwγw)= 120 µmole m-3
for all secondary species and for τgw using 10, 30 and
60 min. Similar results were obtained for the simulations performed with
Cw/(Mwγw)= 50 µmole m-3 (see Fig. S1
in the Supplement). Simulations presented in Fig. 7 were carried out using
the SAR developed by Atkinson (2007) to describe the alkoxy chemistry
(hereafter denoted ATK simulations). A set of simulations was also performed
using the Vereecken and Peeters (2009, 2010) SARs for the estimation of the alkoxy
decomposition rates and the alkoxy isomerization rates
(hereafter denoted VER simulations). Figure S2 in the Supplement shows the
comparison of the results obtained with ATK and VER, which is discussed
below.
For the alkane series, the experimental trends are well captured by the
model. In particular, both the simulated and observed SOA yields (i)
increase with the carbon number of the precursor, (ii) are higher for a
cyclic compared to a linear structure and (iii) decrease with the number of
methyl groups on the carbon skeleton. This behavior is mostly driven by the
fate of alkoxy radicals. While decomposition (i.e., C–C bond breaking)
pathways of linear alkoxy radicals lead to molecules with a smaller carbon
backbone (see Fig. 9), the decomposition of cyclic alkoxy radicals preserves
the size of the carbon backbone. Similarly, decomposition is enhanced for
branched alkoxy radicals (e.g., Lim and Ziemann et al., 2009a). This effect
is well represented in the simulations, as discussed in Aumont et al. (2013).
Figure 7 shows that the SOA yields simulated without wall loss taken
into account are systematically overestimated. A better agreement is
obtained when the partitioning of organic species to the wall is considered
in the model. Decreasing characteristic time of the gas to wall mass
transfer decreases the simulated SOA yields (see Fig. 7). The best agreement is
obtained for the lower value of τgw (10 min) with SOA-yield
reductions ranging from a factor of 1.25 to 2. Note that a change in gas–wall
equilibrium value does not affect these results (see Fig. S1). Besides,
using the lower value of Cw/(Mwγw) (50 µmole m-3)
increases SOA yield by less than 15 % (see Fig. S1). As
discussed above, for a simulation lasting 1 h, the gas–particle partitioning
outweighs the gas–wall partitioning for low volatility species, leading to a
rather low sensitivity to the wall properties. Partitioning is therefore
rather under the kinetic control than the thermodynamic control.
Distribution of the top 15 species between the gas phase (purple),
the particle phase (yellow) and the walls (red) at the end of the simulations
for n-octane (C8), n-dodecane (C12) and n-hexadecane (C16).
Simulated results are reported in mass concentrations (in
µg m-3 of air or equivalent for the Teflon phase) for
simulations conducted without wall loss (left panel) and with partitioning to
the wall using τgw= 10 min and
Cw/(Mwγw)= 120 µmole m-3
(right panel).
![]()
The experimental trends are also well captured by the model for the alkene
series. Simulations and experimental observations show that SOA yields (i)
increase with the number of carbon atom in the parent hydrocarbon and (ii)
are higher for 1-alkenes than for internal alkenes and 2-methyl-1-alkene.
These trends are largely controlled by the fate of the β-hydroxyalkoxy radicals resulting from OH addition on the C=C bond. In
particular, typically one carbon atom is lost from the backbone during the
fragmentation of alkoxy radicals produced from 1-alkenes, while in the
case of internal alkenes the carbon skeleton is broken in the middle of the
molecule, ultimately leading to volatile products and therefore lower SOA
yields. For the simulations with C< 13 alkene precursors, the simulated
SOA yields are consistent with experimental data with a mean bias of less than
30 %. However, the model/observation agreements for larger parent
hydrocarbons depend on the considered alkene series: for 1-alkenes, the
simulated SOA yields are overestimated and the discrepancies rise with the
size of the precursors; for internal alkenes the simulated SOA yields are
relatively constant while the experimental yields show a significant growth
with the carbon chain length; and for 2-methyl-1-alkenes, the model
underestimates SOA yields. For these 6 min experiments, simulation results
appear to be rather insensitive to the wall loss of organic vapors (see
Fig. 7), even for simulations performed using the lowest value of τgw
(10 min) with SOA-yield reductions less than a factor of 1.1.
A comparison of the ATK and VER simulation results is shown in Fig. S2. The
VER configuration of the model produces lower SOA yields compared to the ATK
configuration. For the alkane simulations, the differences are small, with
both SAR predicting similar branching ratios for the simple alkoxy chemical
structures produced during the first generations. A larger effect is,
however,
observed in the simulated SOA yields for alkenes, reaching for instance a
factor of 2 for the C11 1-alkene. This effect is, in particular, due to
different predictions of the branching ratios of β-hydroxyalkoxy
radicals, a key chemical structure produced from OH addition to the C=C
bond. For the conditions examined in this study, the reaction of β-hydroxyalkoxy radicals with O2 is negligible compared to the
isomerization and decomposition routes (Atkinson, 2007). The branching
ratios predicted for these two pathways according to the ATK and VER
configurations are reported in Table 2 for selected generic structures
produced from the oxidation of terminal, internal and 2-methyl alkenes. The
VER configuration leads to a larger fraction of decomposition than the ATK
configuration, which is consistent with lower simulated SOA yields for the
VER configuration compared to the ATK one. Table 2 also reports
experimentally derived branching ratios from measurements of the oxidation
products (Matsunaga et al., 2009; Aschmann et al., 2010; Ziemann, 2011).
These experimental results suggest that the isomerization routes are
underestimated in both the ATK and VER configurations for OH+alkene
systems. The implementation of organic vapor wall loss or the use of a
different structure activity relationship to estimate the fate of the alkoxy
radicals does not fully explain the discrepancies between the model and the
observations for the alkene series. Some chemical pathways are clearly not
well represented or missing in the GECKO-A mechanism (e.g., autoxidation
processes, dimers formation and/or heterogeneous oxidation processes in the
particle phase). The measurement of final SOA yields does not
provide enough insights to identify the source of the model/measure
discrepancies, additional information on SOA composition would be needed.
Decomposition and isomerization branching ratios for selected
generic β-hydroxyalkoxy radicals produced from the oxidation of
terminal, internal and 2-methyl alkenes as estimated by the ATK and VER
configurations of the model and experimentally derived values by Matsunaga et
al. (2009), Aschmann et al. (2010) and Ziemann (2011).
Decomposition/isomerization branching ratio
ATK configuration
VER configuration
Experimental data*
1-alkenes
0.67/0.33
0.79/0.21
0.45/0.55
Internal alkenes
1.00/0.00
1.00/0.00
0.67/0.33
2-Methyl-1-alkenes
0.93/0.07
1.00/0.00
0.59/0.41
* Reported in Ziemann (2011).
Gas–wall partitioning impacts on SOA composition
Simulations performed with the GECKO-A mechanisms were used to examine which
species are prone to partition with the chamber walls. Figure 8 presents the
distribution of the top 15 species in the gas, particles and wall (i.e.,
Teflon) phases for various experiments. Simulated results are reported at
the end of the experiments in mass concentrations (in µg m-3
of air or equivalent for the Teflon phase) for both cases: without (left
panel) and with partitioning to the wall using τgw= 10 min and
Cw/(Mwγw)= 120 µmole m-3 (right
panel). Results in Fig. 8 are reported for n-octane (C8), n-dodecane
(C12) and n-hexadecane (C16) to examine the effects of chain
length on mass distribution.
In Fig. 8, species are categorized based on organic moieties (isomers having
identical functional groups are lumped together), and whether the species
are produced by functionalization (with number of generation) or by
decomposition routes. The letter code given in Fig. 9 for each category
denotes the functional groups on the carbon backbones (N: nitrate,
O: alcohol, K: ketone, D: aldehyde, E: ether, P: peroxyacetyl nitrate,
S: ester, U: carbon double bond) or characteristics of the carbon backbone
(T: cyclic structure, C: saturated aliphatic hydrocarbon – here the
parent compound). A generic chemical mechanism leading to this set of
species is presented in Fig. 9, with the codes corresponding to the various
chemical structures.
Generic GECKO-A chemical mechanism leading to the major products
formed during n-alkane gaseous oxidation under high-NOx conditions.
![]()
Saturation vapor pressure of the major products from
functionalization routes simulated during n-alkane gaseous oxidation under
high-NOx conditions as a function of the number of carbon atoms in the
backbone. Dashed segments link two species with increasing carbon numbers
within a chemical family to facilitate the reading of the figure. The grey
zone represents the volatility range likely affected by gas–wall partitioning
for a 1 h experiment (see text).
![]()
For these experimental conditions, a large fraction of the parent
hydrocarbon (denoted C in Fig. 8) persists after 1 h of oxidation (see
Fig. 5). The simulated fraction of n-octane and n-dodecane in the condensed
phase is negligible. However, the partitioning of n-hexadecane to the
particles and the walls is significant.
Major first generation products are alkylnitrates (N), 1,4-hydroxynitrates
(NO), cyclic hemiacetals (denoted TOE and TOOE in Fig. 8) and dihydrofurans
(denoted TUE in Fig. 8) produced from the fast cyclization of
1,4-hydroxyketones and 1,7-dihydroxy-4-ketones in the condensed phases
assumed in this model configuration. Reaction sequences leading to the first
generation products are repeated at other positions on the carbon backbones,
leading to many second generation products (see Fig. 9). For example, alkyl
nitrates react with OH producing dinitrate species (NN),
hydroxyketonitrates (NKO) and their corresponding cyclized form (TNOE).
Oxidation of the hydroxyl moiety is significant as well. For example, the
yields of the second generation ketonitrate (NK) are essentially explained
by the oxidation of the first generation hydroxynitrate (NO). Figure S3 in
the Supplement displays the fraction of DHF reacting with OH radical,
NO3 radical and O3 by addition to the C=C double bond, during
the course of a typical alkane simulation. As expected, DHF oxidation is
mainly driven by OH and leads to one of the major products, the
carbonylesters (DS). However, DHF oxidation by O3 and NO3
appears to be also substantial (see Fig. S3) and explains the formation of
acid carboxylic esters (AS) and cyclic dinitrate ethers (TNNE).
Interestingly, the GECKO-A mechanisms lead to the production of lactones
(TS) for short chain alkanes. These species are suspected to form oligomers
(Kjaergaard et al., 2012). The model suggests that these lactones are
produced in the gas phase from the OH radical oxidation of the cyclic
hemiacetal structures via the abstraction of the hydrogen atom in the beta
position relative to the hydroxyl group, ultimately leading to a specific
alkoxy radical that forms the lactone by decomposition.
Figure 8 shows that wall uptake affects particular categories of species,
depending on the chain length of the parent compound. As already discussed
in the introduction, the partitioning of a species in the various phases is
driven by its volatility. Similar chemical reactions are involved in the
oxidation of long-chain alkanes, leading to the same families of species but
with decreasing volatility when the size of the carbon skeleton increases.
The vapor pressures of the major chemical families simulated during the
alkane experiments are given as a function of the number of carbon atoms
(nC) in the backbone in Fig. 10. The shaded grey area corresponds to
the volatility domain where significant partitioning to the walls is
expected after 1 h (i.e., 10-8–10-5 atm; see Introduction).
The location of a given family in this volatility/nC framework largely
explains the wall effect highlighted in Fig. 8. For example, for the alkyl
nitrate family (N), a first generation product, (i) C8 species appear
to be too volatile to partition to the condensed particle phase (those are
located on the right side of the grey area), (ii) C12 species are
volatile enough to partition to the wall but not enough to substantially
partition to the particles (even if no wall is considered) and (iii)
C16 species are mostly found in the condensed phase but the volatility
is not low enough to prevent a substantial partitioning to the wall (those
are still on the volatility domain where the walls substantially impact the
species distributions). The mass reductions due to gas–wall partitioning for
C8, C12 and C16 alkyl nitrates are, respectively, -2,
-40 and -7 % in the gas phase, and -63 and -26 % in the particle
phase for C12 and C16 species, respectively. Compared to the
nitrate family, the volatility of the NO, also first
generation products, is shifted toward lower volatility (typically by a
factor of 30). For that NO family, C8 species partition to the walls to
more than 50 % (but not into the particles) affecting gas composition
while C16 species are almost exclusively found in the particle phase,
owing to their low volatility and the resulting kinetic control of the
partitioning (see introduction). For C12 hydroxynitrates, 50 % of the
mass is lost into the walls modifying both gas and particle concentrations.
The behavior of the first generation of cyclic hemiacetals (TOE) is similar
to the nitrate family but most second generation species behave rather like
the hydroxynitrate family, e.g., the dinitrates (NN) or the DHF oxidation
products (TNOE, DS). The deposition of vapor to the walls during the
experiments therefore splits a specific family of compounds and
substantially alters both the simulated SOA yields and compositions.
Conclusion
In this study, simulations performed with the GECKO-A chemical mechanisms
were assessed against 41 smog chamber experiments aimed at examining SOA
formation during the OH oxidation of alkanes and alkenes under high NOx
conditions. Various homologous series of species were examined containing
hydrocarbons with 7 to 17 carbon atoms. The set of parent hydrocarbons
includes linear, cyclic and C12-branched alkanes and terminal, internal
and 2-methyl alkenes. Wall loss of organic vapors was parameterized in
the GECKO-A modeling tool to investigate the sensitivity of SOA formation to
gas–wall partitioning.
Simulated trends match the experimental trends observed within and between
homologous series. In particular, (i) SOA yield increases with the carbon
skeleton size of the parent hydrocarbon, (ii) SOA yield increases from
branched to linear to cyclic chemical structures, (iii) SOA yield decreases
with the number of branches in the carbon backbone and (iv) SOA yield
decreases from internal to terminal alkene isomers. The loss of organic
vapors to the chamber wall is found to affect final SOA yields, with
reductions ranging from a factor of 1.1 to 2. The extent of this process
depends on the rate of gas–wall mass transfer and the vapor pressure of the
species. As a result, the simulations of the alkene series (lasting 6 min)
appear insensitive to wall uptake, unlike the simulations for alkane series
(lasting 1 h).
Accounting for wall losses in the simulations of the alkane series improves
significantly the agreement of the model results with the observations, the
overall best agreement being obtained when a fast gas to wall mass transfer
rate is assumed. The composition of the gas and the particle phases is also
impacted by wall losses. For the alkane series, both phases are mostly
composed of first generation species produced along functionalization routes
such as alkyl nitrates, 1,4-hydroxynitrates, cyclic hemiacetals and
dihydrofurans coming from the fast cyclization of 1,4-hydroxycarbonyls in
the particle. Second generation species are also found to be substantial
contributors to SOA. These species are produced from OH oxidation and lead
to additional functional groups (mostly nitrate, alcohol or carbonyl) on the
carbon backbone. The presence of acid carboxylic esters (AS) and cyclic
dinitrate ethers (TNNE) among the major simulated products suggests that DHF
oxidation by O3 and NO3 can compete with OH oxidation during these
experiments. Simulated distributions of the species in the various phases
suggest that substantial amounts of nitrates, hydroxynitrates and
carbonylesters could be lost onto the walls.
For the alkene series, the simulated SOA yields exhibit a strong dependence
on the structure activity relationship used to estimate the fate of the
alkoxy radicals. The gas–wall partitioning process cannot fully explain the
discrepancies between the model and the observations. Some chemical pathways
are clearly missing in the GECKO-A mechanism, e.g., dimers formation and/or
heterogeneous oxidation in the particle phase. Thus, further investigations
of SOA composition are required.
This work suggests that SOA yields inferred from chamber experiments could
be substantially underestimated due to the loss of organic vapors to the
wall of the chamber. This process also likely alters the inferred SOA
composition. The gas–wall partitioning of organic compounds needs to be
routinely characterized for smog chamber data and derived SOA yields. These
characterizations appear to be a critical issue to support the development
of SOA parameterizations for air quality and climate models based on
experimental observations in atmospheric chambers.