Introduction
Atmospheric aerosols are complex mixtures that can contain a multitude of
chemical species . While the inorganic fraction comprises
a relatively small number of compounds, the organic fraction (or organic
aerosol, OA) includes thousands of compounds with diverse molecular
structures . These compounds take part in multitude of
gas phase, aerosol phase, and heterogeneous transformation processes
e.g., that must be modeled
with sufficient fidelity to predict atmospheric concentrations and impacts
from various emission scenarios.
A mechanism central to these processes is the formation of semivolatile
organic compounds (SVOCs) through gas-phase oxidation of volatile organic
compound (VOC) precursors and their reaction products. α-pinene (APIN)
and 1,3,5-trimethylbenzene (TMB) are examples of biogenic and
anthropogenic VOC precursors, respectively, which have been studied for their
chemical reaction mechanisms and aerosol yields in environmentally controlled
chamber experiments and numerical simulation. APIN is a monoterpene compound
primarily emitted from coniferous vegetation
with high emission rate, reactivity, and secondary organic aerosol (SOA)
generation potential e.g.,. TMB is an aromatic compound
emitted from vehicular emissions and a major contributor to urban organic
aerosol e.g.,; its degradation mechanism has also
been subject of collective evaluation .
Gas-phase oxidation reactions are modeled with
chemically explicit or semi-explicit treatment, or alternatively using a
basis set approach based on simplified molecular or property descriptors; SOA
formation is commonly modeled by coupling these reactions with partitioning
of oxidation products to an absorptive organic phase
e.g.,. SVOCs produced by such reactions can in reality partition among
multiple phases (vapor, organic liquid, aqueous, solid), and participate in
additional functionalization, accretion, or fragmentation reactions in one of
many phases . These
processes are represented in models with varying degrees of detail;
simplifying or wholly omitting various mechanisms out of concerns for
computational feasibility or lack of sufficient knowledge. For instance, in a
work we follow closely in this manuscript, used a fully
explicit gas-phase reaction mechanism with absorptive organic partitioning
and evaluated the potential importance of missing heterogeneous and
condensed-phase mechanisms based on discrepancy of model simulation and experiments.
Our capability to simulate SOA formation is often evaluated against aerosol
mass yield, O : C, carbon oxidation state, mean carbon number, volatility, and
specific species or compound classes when available
e.g.,. These
properties can be measured using various forms of mass spectrometry
e.g., or through monitoring
changes in size distribution in combination with isothermal dilution or
thermal heating e.g.,. Functional group (FG) composition is a complementary
representation of organic molecules and complex organic mixtures that offers
a balance between parsimony and chemical fidelity for measurement and interpretation.
FGs represent structural units of molecules that play a central role in
chemical transformations and provide insight into evolution of complex
organic mixtures without monitoring all species explicitly . FG abundances have also
been associated with volatility e.g.,, hygroscopicity
e.g.,, and magnitude of nonideal
interactions in the condensed phase e.g.,. However, two impediments have been the likely cause of slow
adoption of this representation. Building quantitative calibration models of
FG abundance have posed analytical challenges, but rapid progress has been
made over the past decade with Fourier transform infrared spectroscopy (FTIR)
e.g.,, nuclear magnetic resonance
, spectrophotometry ,
and gas chromatography–mass spectrometry with derivatization . The second
challenge is computationally harvesting FG abundance from a large set of
known molecular structures. To this end, developed a set
of substructure definitions corresponding to FGs that can be queried against
arbitrary molecules specified by their molecular graphs.
In this work, we apply these new substructure definitions to describe the FG
composition of products simulated by gas-phase reactions prescribed with the
Master Chemical Mechanism (MCMv3.2)
and SOA constituents formed by their dynamic
absorptive partitioning . Three instances of APIN
photooxidation under varying initial concentrations of oxides of nitrogen (NOx),
and TMB oxidation in the presence of NOx are
studied in accordance with aerosol FG composition characterized by
and in chamber studies using FTIR. The
model results are analyzed through a suite of FG abundances and
model–measurement comparisons of measured FGs are presented to hypothesize
reasons (including unimplemented mechanisms) for discrepancies where they occur.
Methods
We target our model simulations to mimic SOA formation in environmentally
controlled chambers for which FG measurements are available.
Summary of the experimental conditions studied in this work. For
simplification, an ID has been given to each system.
ID
Publication
Precursor
Measurement conditions
APIN-lNOx
α-pinene: 300 ppb
low NOx: 240 ppb
RH: 61 %
seed: none
radical initiator: propene, 300 ppb
APIN-hNOx
α-pinene: 47 ppb reacted
high NOx: 847 ppb
RH: 5 %
seed: ammonium sulfate, 27 µg m-3
radical initiator: CH3ONO, 200–400 ppb
APIN-nNOx
α-pinene: 46 ppb reacted
no NOx
RH: 4 %
seed: ammonium sulfate, 24 µg m-3
radical initiator: H2O2,
TMB-lNOx
1,3,5-trimethylbenzene: 1312 ppb
low NOx: 320 ppb
RH: 60 %
seed: none
radical initiator: propene, 300 ppb
Systems studied
Photooxidation of APIN under “low-NOx” (NOx / APIN of 0.8),
“high-NOx” (NOx / APIN of 18), and no-NOx conditions (designated as lNOx, hNOx,
and nNOx, respectively), and TMB under “low-NOx”
(NOx / TMB ratio of 0.24; designated as lNOx)
conditions were simulated in this study to compare with available
measurements of aerosol FG composition in environmental chamber experiments.
Simulations were run at 298 K and with conditions closely following experimental
descriptions summarized in Table , with a few exceptions.
In the case of APIN degradation in high-NOx conditions, the
H2O2 was used as the OH radical initiator as CH3ONO is not
available in the MCMv3.2 degradation scheme. When the reacted instead of
initial precursor concentration is reported, this value is used as the
initial concentration for the simulations. This decision is supported by the
virtual observation that 99 % of the precursor is reacted after 4.5–6.5 h
in these cases (Fig. S1 in the Supplement) and specification of higher initial
concentrations leads to reacted quantities inconsistent with experimental specifications.
Model formulation
While differing in implementation, the model specification resembles the
MCM-SIMPOL framework described by . The chemical mechanism
prescribed by MCMv3.2
was used to simulate the gas-phase oxidation of volatile organic compounds (VOCs).
The Kinetic Pre-Processor KPP;
was used to generate the gas-phase chemistry code in Fortran 90. A
separate dynamic absorptive partitioning module was
added via sequential operator splitting to simulate gas–particle (G-P) partitioning after the reaction
operator. Pure component vapor pressures of organic compounds in the MCMv3.2
degradation schemes were calculated using SIMPOL.1 , and
non-ideal interactions were neglected in these simulations (i.e., activity
coefficients were set to unity for all species). Vapor pressures are
converted to equivalent mass concentrations C0 (Sect. S1 in the Supplement), and
normalized by a reference value for presentation in logarithmic units
such that the notation logC0
implies log10(C0/1 µg m-3). LSODE (Livermore Solver for
Ordinary Differential Equations; ) was used as the
numerical solver for each operation (reaction and G-P partitioning). A time
step of 60 s is used in this study, as it is in the order of magnitude
of the timescale of gas-phase oxidation and condensation/evaporation under
chamber conditions and leads to stable solutions.
Radiation intensities were fixed at their maximum throughout the simulations
to mimic conditions used in the chamber studies, with values corresponding to
clear-sky conditions at an altitude of 0.5 km, 1∘ solar zenith angle
in July, and a latitude of 45∘ N .
Absorptive partitioning to a purely organic phase is considered in this model
(Sect. S1). The relative humidity (RH) specified
in the experiments are converted to equivalent concentrations of H2O
for participation in the HO2 radical self reaction to form hydrogen
peroxide , but water uptake by the aerosol and its
influence on G-P partitioning of organic compounds
is not considered. As aerosol growth following homogeneous and
heterogeneous nucleation processes of the condensed organic phase in the
chamber experiments are not included in the model, we use a seed
COA,init of 1 µg m-3 to initiate G-P
partitioning (Sect. S2). We specify the bulk of
COA,init to be a generic, non-volatile organic solvent
that does not participate in reactions or partitioning and is in equilibrium
with the initial composition of the gas phase (Sect. S2). The relative
composition reported in this study is insensitive to this value after 1 h of simulation (Figs. S3 and S4). To differentiate between the SOA formed
in the simulation and the total organic aerosol phase involved in
partitioning, we denote the former quantity as CSOA and
the latter as COA = COA,init + CSOA. No condensed-phase reactions are included; as
with we consider them a potential source of
model–measurement discrepancies. While the particle diameter of the
monodisperse population is allowed to grow according to the organic aerosol
condensed (Sect. S2 in the Supplement), the number concentration of particles is kept fixed
during the simulation; losses of both particles and gases to chamber walls
e.g., are neglected. These
assumptions will affect calculations of total yield and rate of change in
aerosol mass; however, aerosol mass yields are in the range of physical
expectation (Fig. S5; mass concentrations represented in the volatility
basis set convention are also shown in Fig. S6 for reference). Relative
abundances of functional groups are robust with respect to many of these
assumptions and will be the primary focus of our presentation and
model–measurement comparisons. However, the impact of vapor losses to
chamber walls may require investigation in future work. An assumption of a
common wall loss parameter for all species e.g., would
mostly reduce the overall yield from simulation, but compound-dependent wall
losses may preferentially reduce the
concentration of the most condensible substances in the system and lead to a
different relative particle composition . The
magnitude of this effect also depends on the number of condensable species
formed, the range of saturation concentrations spanned, and their absolute abundance.
Simulation analysis
A chemoinformatic tool (APRL-SSP; ) described by
is used to harvest FG abundances (enumeration of the FG
fragments) from each molecule in the simulations. This tool consists of
scripts invoking OpenBabel and Pybel and
SMARTS patterns formulated and validated for these
chemical systems. Using this tool, molecular structure is mapped to input
parameters for SIMPOL.1, and FG abundances of the organic aerosol mixture are
obtained from molecular concentrations. Most importantly, we extract two
arrays with elements ϕip, the number of times FG p occurs in
molecule i, and ϕipa*, the number of times atom type a occurs
in FG p in molecule i. We combine these two coefficients with the
molecular or molar concentrations C of compound i in phase α
generated by our simulations to estimate several useful mixture properties
for time tj:
∑i∈MCiαtjϕip=abundanceofFGp
Ciαtjϕip/∑i∈MCiαtjϕip=fractionalcontributionofmoleculeitoabundanceofFGp
∑i∈MCiαtjϕipa*=apportionmentofatomsoftypeatoFGp.
The summation is taken for the set of all compounds (or molecule types) M.
The last quantity is used to separate the contributions of O : C
and N : C from various FGs. The set of patterns were constructed to meet
conditions of completeness and specificity (each atom is matched by one
and only one group) such that the sum of oxygen and nitrogen atoms in each FG
sums to the total number of atoms in the system .
Polyfunctional carbon atoms are not considered in the condition for
specificity (matches by multiple groups lead to overestimation of counts
in ϕipa*); therefore, the total number of carbon used in the
denominator of these atomic ratios is estimated using the SMARTS pattern [#6].
We additionally estimate integrated reaction rates
IRRs; to examine degradation rates relative to rates
of production in the gas phase (g) for selected systems. The IRR for reaction r
affecting compound i at time tj is calculated from the rate constant k
and the product of concentrations C:
IRRritj=Ci(g)tj-Ci(g)tj-Δt=∫tj-Δttjdtkr∏i′∈MrC′i(g)(t)≈Δtkr∏i′∈MrC′i(g)tj.
Mr is the set of compounds involved in reaction r. The
expression in parentheses is the conventional rate equation for reaction r.
To obtain the IRR for functional group p, we multiply by the factor ϕip described above:
IRRrptj=∑i∈MrIRRritjϕip.
IRR estimates were harvested from the LSODE solver, and the PERMM package
was used to associate compounds and FGs with each reaction.
Time series of the relative molar contribution of different FGs to
the O : C in the gas phase (top panels) and aerosol phase (bottom panels)
simulated in this work for APIN-lNOx, APIN-hNOx,
APIN-nNOx, and TMB-lNOx. The contribution of each FG to
the O : C ratio accounts for the number of oxygen atoms per FG.
Measurements
FTIR analysis reported by and quantified
the molar abundance of alkane CH (aCH), carboxylic acid (COOH), non-acid
(ketone and aldehyde) carbonyl (naCO), alcohol OH (aCOH), and organonitrate (CONO2)
FGs. Uncertainties in the FG quantification have been
reported to be between 5 and 30 % .
collect particles in the range of 86–343 nm onto zinc
selenide substrates by impaction, while sample generated
aerosol onto polytetrafluoroethylene (PTFE) filters for FTIR analysis.
Measurement artifacts can arise during time-integrated collection of aerosol
samples and can differ according to duration of sampling
or method of collection . The
primary driver for absorptive and evaporative artifacts which may impact bulk
mass estimation is the difference between the changing gas-phase composition
and equilibrium vapor composition with respect to the aerosol phase, but
model simulations suggest the relative gas-phase composition stabilizes after
the first few hours. Changes in particle composition due to condensed-phase
chemistry may perturb the equilibrium, but this phenomenon may be interpreted
together with condensed-phase processes not included in the model. In the
analysis of , samples transported off-site for analysis
were frozen to minimize evaporation and reaction artifacts during storage.
Additionally, evaporative losses in the analysis chamber of the FTIR (during
purging of headspace with dry nitrogen gas) were minimized by rapid scanning,
and report that the spectrum was stable even when repetitive
measurements are performed.
In this work, we limit our discussion to results based on molar rather than
mass concentrations of FG abundances. While mass concentrations are commonly
reported for FTIR measurements of ambient samples
e.g.,, estimates are based on fixed assumptions
regarding the apportionment of polyfunctional carbon atoms to associated FGs
e.g.,. These assumptions can affect both mass estimation and atomic
ratios (e.g., O : C). proposed a modification based on
assumed molecular structures in their chamber experiments, and mass estimates
using these values are shown in Fig. S7. Constraining the mapping of
measured bonds to atoms for estimation of these quantities in various
mixtures are planned for future work. For model–measurement comparison, we
select the subset of FGs that are reported by measurement and use relative
metrics normalized only by measured fractions of OA.
Comparison of the changes of the relative mole fraction compared to
the first sample for COOH, COH, CO, aCH, and CONO2 of the aerosol
phase measured by and modeled in this work for
APIN-lNOx and TMB-lNOx. For a chosen FG, the changes of
the relative mole fraction compared to the first sample are calculated as the
ratio between the relative mole fraction at the chosen time and the relative
mole fraction at 1 h. naCO includes ketone and aldehyde FGs, but the
change in relative ketone FG abundance is also shown separately for
illustration. The contribution of ketone and aldehyde to CO have been
reported separately in the model results. The x axis refers to the hours after the lights were turned on in the chamber for the bottom panel (Measurement) and the time after the start of the simulation in the top panel (Model). The dashed line corresponds to y = 1 and has been added for
visual reference.
Pie charts illustrating the time-integrated relative aerosol mole
fraction of aCH, CO, COOH, CONO2, and aCOH in model simulations and
experiments. The mole fractions reported in simulations are summed with
respect to the subset of FGs that are reported by measurement to facilitate
direct comparison. The time reported refers to the hours after the lights
were turned on in the chamber (Measurements), and the time after the start of
the simulation (Model). In the pie charts reporting the measurement conducted
by (APIN-hNOx and APIN-nNOx) the
CNH2 fraction has been omitted in order to obtain a direct comparison
between model and experiments. The sum of percentages combines to
100 ± 1 %, as individual values were rounded to the nearest whole number
for labeling.
Conclusions
In this study, the FG distribution of SOA generated in environmentally
controlled chamber experiments reported in the literature for APIN and TMB photooxidation
have been compared to explicit gas-phase chemistry and partitioning simulated
with MCMv3.2 and SIMPOL.1. Varying degrees of agreement between the model and
FTIR measurements of FG evolution in SOA generated in environmentally
controlled chambers are found.
In the APIN-lNOx simulations, the FG relative abundance is well
captured by the model in the first hours of simulation, and general trends in
the changes of the mole fraction compared to the first sample are captured
qualitatively by the model. However, the underestimation of the measured
oxidized groups (COOH, aCOH, and CO) are apparent after 20 h in our
simulations; this discrepancy may be explained by heterogeneous reactions
missing in the model. O : C is generally overestimated for
APIN-hNOx, APIN-nNOx, and TMB-lNOx on
account of large contributions from CONO2, peroxide, or hydroperoxide
groups, while the aCH is simulated consistently in larger proportion to some
of the measured oxygenated species (COOH and aCOH). These errors are largely
correlated, as CSOA mass and individual FGs are
dominated by a few polyfunctional compounds in these simulations. The
dependencies of aerosol composition on a limited number of compounds also
speaks as to the sensitivity of simulation results on a few kinetic or
partitioning parameters, which might otherwise be averaged out in systems
where the condensed phase is composed of a larger number of compounds.
In the APIN-hNOx simulations, the model predicts a higher
fractional abundance of CONO2 in the aerosol phase than what is
observed in the FTIR measurements. The CONO2 fraction comes to constitute 46 % of the total O : C ratio, which partly contributes to the
higher O : C of the aerosol phase during the simulation (0.78) compared to the
O : C observed (∼ 0.4) by . Only four
CONO2-containing polyfunctional compounds account for more than 80 %
of the organic mass. The uncertainties due to lack of kinetic data in the
total CONO2 yield in the primary oxidation sequence of APIN may play
an important role in the high-NOx regime and explain the
discrepancies between model and measurements in this scenario. For the
APIN-nNOx simulations, four polyfunctional compounds account for
over 80 % of the CSOA mass and a large bulk of ketone
CO and hydroperoxide FGs. The relative abundance of ketone CO is
overestimated compared to observations; the O : C is also overestimated,
possibly on account of the large (42 %) contribution from the hydroperoxide
FG which originates from the same set of molecules. For the
TMB-lNOx photooxidation simulations, general trends in the
changes in relative mole fractions compared to the first sample for COOH,
aCOH, naCO, aCH, and CONO2 also qualitatively follow observations, but
their magnitudes have more discrepancies with experiments than in the case of
APIN-lNOx. These discrepancies have also been hypothesized as a
sensitivity to reaction rates and vapor pressures of a few dominant products
that contribute significantly to the aCH mole fraction and peroxide fraction
of the aerosol O : C ratio. As for the APIN-lNOx simulations, the
agreement in abundances of aCH relative to the measured set of oxidized FGs
may also be explained by additional condensed-phase oxidation chemistry not
included in the model.
This work illustrates that concurrent measurement of FGs alongside common
techniques for atomic and molecular characterization of OA can provide an
opportunity for complementary evaluation and further guide detailed
understanding of chemical and physical transformations. Analysis of FG
abundance can supplement tracking of individual tracers and evaluate the
importance of mechanisms that lead to production of a class of compounds in
the overall molar (or mass) budget. FG abundances can also provide structural
interpretation to variations in elemental ratios (e.g., O : C, H : C, and N : C).
Looking forward, systematic model–measurement comparison of FGs under
controlled conditions may be able to provide constraints and aid development
of chemical mechanism generators e.g.,. While
we have uncovered only a fraction of the analysis capabilities that a FG
perspective provides, we anticipate that the tools and approaches introduced
in this work can encourage further comparisons between model simulations of
both gas- and aerosol-phase chemistry in conjunction with emerging methods for
FG quantification.