Geophysical Research Letters

Liquid-liquid phase separation in aerosol particles: Dependence on O:C, organic functionalities, and compositional complexity


Corresponding author: M. Song, Institute for Atmospheric and Climate Science, ETH Zurich, CH-8092 Zurich, Switzerland. (


[1] Atmospheric aerosol particles may undergo liquid-liquid phase separation (LLPS) when exposed to varying relative humidity. In this study we investigated the occurrence of LLPS for mixtures consisting of up to ten organic compounds, ammonium sulfate, and water in relationship with the organic oxygen-to-carbon (O:C) ratio. LLPS always occurred for O:C < 0.56, never occurred for O:C > 0.80, and depended on the specific types and compositions of organic functional groups in the regime 0.56 < O:C < 0.80. In the intermediate regime, mixtures with a high share of aromatic compounds shifted the limit of occurrence of LLPS to lower O:C ratios. The number of mixture components and the spread of the O:C range did not notably influence the conditions for LLPS to occur. Since in ambient aerosols O:C range typically between 0.2 and 1.0, LLPS is expected to be a common feature of tropospheric aerosols.

1. Introduction

[2] Tropospheric aerosol particles consisting of inorganic salts and organic compounds undergo phase transitions such as deliquescence and efflorescence as a consequence of changes in ambient relative humidity (RH) [Martin, 2000; Marcolli and Krieger, 2006; Martin et al., 2008]. In addition, non-ideal interactions between dissolved inorganic ions, water, and organic compounds may lead to liquid-liquid phase separation (LLPS) [Marcolli and Krieger, 2006; Ciobanu et al., 2009; Zuend et al., 2010; Bertram et al., 2011; Song et al., 2012; Zuend and Seinfeld, 2012]. The physical state and the hygroscopicity of aerosol particles need to be considered for an accurate quantification of atmospheric and climate effects because they influence gas-particle partitioning of semivolatile compounds [e.g.,Zuend et al., 2010; Zuend and Seinfeld, 2012], heterogeneous chemistry [e.g., Anttila et al., 2007; Cosman et al., 2008] and the scattering and absorption of light [e.g., Martin et al., 2004]. The presence or absence of LLPS in aqueous mixtures can be determined in laboratory experiments with model mixtures representing tropospheric aerosols [e.g., Marcolli and Krieger, 2006; Bertram et al., 2011; Song et al., 2012] or computed using a liquid-liquid equilibrium model [Zuend et al., 2010; Zuend and Seinfeld, 2012].

[3] While ammonium sulfate (AS) is a main constituent of aerosol particles and well suited to represent the inorganic aerosol fraction in model mixtures, selecting appropriate surrogates for the organic fraction is more complex. The fractions of organics (18–70%), sulfate (10–67%) and ammonium (6.9–19%) have been measured by aerosol mass spectrometry at various locations [Zhang et al., 2007]. The organic aerosol fraction typically consists of up to thousands of different compounds and their characterization and quantification is a task that challenges common analytical techniques. Typically only 10–20% of the organic aerosol fraction can be identified on the level of individual substances using gas and liquid chromatography of filter extracts [Decesari et al., 2006; Hallquist et al., 2009]. With this approach, classes representing the organic aerosol have been identified. These comprise oxidized aliphatic substances such as mono- or multifunctional carboxylic acids, alcohols, polyols, sugars, and oxidized aromatic substances such as aromatic acids and aldehydes. Alternative methods are needed to gain a more quantitative overview of organic functionalities. Functional group analysis by proton NMR spectroscopy has identified aliphatic groups bound to an unsaturated carbon, alkoxyl, acetal, alkylic, and aromatic groups as main functionalities representing the organic aerosol fraction [Decesari et al., 2006]. FTIR spectroscopy is able to quantify alcohol, aromatic, aliphatic unsaturated, carbonyl, carboxylic acid, and amine groups [Gilardoni et al., 2009]. An even more general characterization of organic aerosols based on their degree of oxidation is achieved by high-resolution mass spectrometry that allows quantifying O:C and H:C ratios [e.g.,Heald et al., 2010; Ng et al., 2010]. Typical values of organic O:C in ambient samples range from 0.2 to 1.0 [Heald et al., 2010; Takahama et al., 2011]. Recent studies on a limited number of model systems have shown that LLPS in mixed organic/AS/H2O particles commonly occurs for O:C < 0.7 [Bertram et al., 2011; Song et al., 2012]. Hence, the O:C ratio of the organic aerosol fraction may be a good predictor for the presence of LLPS in tropospheric aerosol particles because it reflects the polarity of organic compounds and their miscibility with water and electrolytes. However, the investigated systems so far are small in number and most of them represent the organic fraction by one substance only. To determine whether the O:C ratio is indeed an accurate LLPS predictor for more realistic representations of the organic/inorganic aerosol, the following questions are addressed in this study: (i) Does in addition to the O:C ratio also the specific functional group composition influence the occurrence of LLPS? (ii) Does the number of components influence the occurrence of LLPS in organic/AS/H2O mixtures? (iii) Does the spread of O:C ratios of the individual organic components present in a mixture influence the occurrence of LLPS? To answer these questions, we investigated the presence or absence of LLPS in model systems of aerosol particles, consisting of various organic components, AS, and water.

2. Experimental Methods

[4] Experiments were performed on droplets (20–65 μm in diameter) at 293 K using an optical microscope equipped with a temperature- and humidity-controlled flow-cell. The detailed experimental procedure is given in theauxiliary material [see also Song et al., 2012]. All the compounds were purchased from Sigma-Aldrich with purities ≥ 98% and used without further purification. We mixed the organic fraction together with AS at organic-to-inorganic dry mass ratios OIR = 2:1, 1:2, 1:6. With these mixtures, we performed humidity cycles to scan the phase diagram for miscibility gaps.

3. Results and discussion

[5] In Table 1, we define different sets of components, which constitute the organic fraction of the investigated model mixtures. Mixtures O:C ratios given in the third and fourth column of Table 1 were obtained by varying dry mass fractions of the organic components (mfd(org)) in a range from 0–50 wt%. The exact composition of all investigated mixtures together with their O:C ratio and information on whether they exhibit LLPS or not is listed in Tables S1–S7 in the auxiliary material. The van Krevelen diagrams given in Figure 1 display the occurrence of LLPS in relation to the O:C and H:C ratios of the mixtures. In the cases where only one or two of the investigated OIR led to a LLPS, as shown in Figure 2a, LLPS was most persistent for OIR = 1:2. Figure 2b gives in addition the average onset RH of LLPS (SRH) over all investigated OIR as a function of O:C. In the following sections, we discuss in detail the aspects (i)–(iii) outlined in the introduction based on Figures 1 and 2 and Table 1.

Table 1. Organic Components of the Investigated Organic Mixtures, Together With the O:C Ratios of the Mixtures With LLPS and Without LLPSa
Mixture NameOrganic Components (With O:C in Brackets)O:C Ratios of Mixtures
With LLPSWithout LLPS
  • a

    “−” = no experiment performed. Note that not all individual mixtures contain all components. O:C ratios of mixtures are obtained by varying dry mass fractions of the components in a range from 0–50 wt%. The individual mixture compositions together with O:C and H:C ratios are given in Tables S1–S7 in the auxiliary material.

Polyolsorbitol (1.00); 1,2,7,8-octanetetrol (0.50)0.710.76
Dicarboxylic acids (DCA)glutaric acid (0.80); methylsuccinic acid (0.80); dimethylmalonic acid (0.80); 2-methylglutaric acid (0.67); 3-methylglutaric acid (0.67); 2,2-dimethylsuccinic acid (0.67); 3-methyladipic acid (0.57); 3,3-dimethylglutaric acid (0.57); diethylmalonic acid (0.57)0.57, 0.670.80
Oxidized aromatic compounds (OAC)2,5-dihydroxybenzoic acid (0.57); 3,5-dihydroxybenzoic acid (0.57); 2,4,5-trimethoxybenzoic acid (0.50)0.550.57
Multifunctional compounds (MFC)maleic acid (1.00); dehydroacetic acid (0.50); kojic acid (0.67); 3,4-dihydroxy-2,2-dimethyl-4-oxo-2H-pyran-6-carboxylic acid (0.50); itaconic acid (0.80); 2-oxoglutaric acid (1.00)0.67, 0.790.85
Polyol+DCAdiethylmalonic acid (0.57); levoglucosan (0.83); glutaric acid (0.80); sorbitol (1.00); 1,2,7,8-octanetetrol (0.50); 2-methylglutaric acid (0.67)0.72, 0.76-
OAC+DCA3-hydroxybenzoic acid (0.43); 4-hydroxybenzoic acid (0.43); 2,5-dihydroxybenzoic acid (0.57); 2,6-dihydroxybenzoic acid (0.57); 3,4-dihydroxybenzoic acid (0.57); 3,5-dihydroxybenzoic acid (0.57); malonic acid (1.33); malic acid (1.25); 2-methylglutaric acid (0.67); diethylmalonic acid (0.57)-0.56, 0.61, 0.68
Complex organic mixture (COM)2-methylglutaric acid (0.67); methylmalonic acid (1.00); malonic acid (1.33); glutaric acid (0.80); 3-methyladipic acid (0.57); diethylmalonic acid (0.57); malic acid (1.25); levoglucosan (0.83); pinonic acid (0.30); pinolic acid (0.30); 3-hydroxybenzoic acid (0.43); 3,5-dihydroxybenzoic acid (0.57); 1,2,7,8-octanetetrol (0.50)0.63, 0.68, 0.700.77
Carminic acidCarminic acid (0.59)0.59-
Tannic acidTannic acid (0.61)-0.61
Figure 1.

Van Krevelen diagram showing organic/AS/H2O systems with LLPS (circles) and without LLPS (stars) for OIR = 2:1, 1:1, 1:2 and 1:6. Triangles: Components of organic mixtures. (a) Mixtures of AS and organics with specific functionalities, namely: Polyol/AS/H2O (cyan), carboxylic acids DCA/AS/H2O including C5/AS/H2O, C6/AS/H2O, C7/AS/H2O (all green) and (C5+C6+C7)/AS/H2O (yellow), oxidized aromatic compounds OAC/AS/H2O (pink), and multifunctional compounds MFC/AS/H2O (gray). (b) Complex organic mixture COM/AS/H2O (orange) with COM consisting of the following components (triangles): DCA (green), OAC (pink), pinonic and pinolic acid (purple), polyols (cyan) and levoglucosan (yellow). Colored regions in Figures 1a and 1b indicate the O:C and H:C ranges covered by the individual organic components for each model system. (c) Collection of all investigated systems from this study (filled symbols) including those shown in Figures 1a and 1b and from the literature (open symbols). Additional systems from this study: Polyol+DCA/AS/H2O (blue), OAC+DCA/AS/H2O (dark green), carminic acid/AS/H2O (brown), and tannic acid/AS/H2O (black). Systems from the literature: polyol/AS/H2O (cyan [Marcolli and Krieger, 2006; Zuend et al., 2008; Ciobanu et al., 2009; Bertram et al., 2011]), DCA/AS/H2O (green [Bertram et al., 2011; Song et al., 2012]), OAC/AS/H2O (pink [Bertram et al., 2011]). Yellow-hatched region: LLPS observed in all investigated systems; gray-hatched region: no LLPS detected in any of the investigated systems.

Figure 2.

Onset RH of LLPS (SRH) in the investigated organics/AS/H2O droplets. (a) Systems showing LLPS with AS dry mass fraction (mfd(AS)) = 0.33, 0.50, 0.67 and 0.86 (lower abscissa) corresponding to organic-to-inorganic dry mass ratios (OIR) = 2:1, 1:1, 1:2 and 1:6 (upper abscissa). (b) Average SRH (circles, average of all OIR that showed LLPS) from this study (filled symbols) and literature (open symbols, seeFigure 1c). Mixtures showing no LLPS are indicated as stars at RH = 0%. The gray shaded region indicates the RH range in which efflorescence was observed in this study. The curves represent a second-order polynomial fit in black fromBertram et al. [2011] and a sigmoid fit in red from this study of all systems with LLPS (all circles). The dashed curve parts in the gray shaded region represent extrapolations and are not supported by experimental data (due to AS efflorescence). Error bars: standard deviation of the SRH from experiments with different particles.

3.1. Dependence of LLPS on Organic Functional Groups

[6] Figure 1a shows the occurrence range of LLPS for organics/AS/H2O droplets covering the same O:C ratios but consisting of different organic functional groups. The functional group composition also influences the hydrogen-to-carbon (H:C) ratio. H:C is higher for open chain hydrocarbons and alcohols, and lower for molecules with ring structures and carbon-carbon or carbon-oxygen double bonds. InFigure 1a various mixtures are compared, comprising hydroxyl functionalities (Polyol/AS/H2O, cyan), dicarboxylic acids containing 5 (C5), 6 (C6) or 7 (C7) carbon atoms (DCA/AS/H2O, green; DCA(C5+C6+C7)/AS/H2O, yellow), oxidized aromatic compounds (OAC/AS/H2O, pink) and a mixture with multifunctional compounds including ring structures and double bonds (MFC/AS/H2O, gray). The components of the mixtures are plotted as differently colored triangles. The range of observed LLPS (circles) and no LLPS (stars) is very similar for the Polyol/AS/H2O and DCA/AS/H2O mixtures. It slightly shifts to higher O:C for MFC/AS/H2O, but significantly shifts to lower O:C for OAC/AS/H2O. The comparison with MFC/AS/H2O indicates that differences in H:C are hardly responsible for the behavior of OAC/AS/H2O, rather the high share of aromatic rings. This is confirmed by results of additional mixtures investigated in this study and shown in Figure 1c. The occurrence range of LLPS for polyols and dicarboxylic acids (Polyol+DCA/AS/H2O) mixtures are in line with the results of Polyol/AS/H2O and DCA/AS/H2O whereas the occurrence range of LLPS of mixtures with oxidized aromatic compounds and dicarboxylic acids (OAC+DCA/AS/H2O) are shifted to lower O:C. Carminic acid and tannic acid are high molecular weight compounds of 492.4 g mol−1 and 1701.2 g mol−1, respectively, both containing aromatic ring structures. They may represent high molecular weight organic molecules as are found in humic like substances (HULIS) [Graber and Rudich, 2006]. Carminic acid (O:C = 0.59) showed LLPS in mixtures with AS while tannic acid (O:C = 0.61) did not, again confirming that high shares of aromatic rings shift the LLPS occurrence range to lower O:C. We attribute this shift to lower O:C of the occurrence range of LLPS in systems with high shares of aromatic rings to a smaller salting-out effect of AS for aromatic than for aliphatic systems, possibly due to cation–π interactions between the ammonium ion and the aromatic rings [e.g., Ma and Dougherty, 1997].

[7] It can be seen from Figure 2athat LLPS of systems with high SRH extend over all investigated OIR. This is consistent with miscibility gaps observed for PEG-400/AS/H2O (O:C ≈ 0.56) and C7/AS/H2O (O:C = 0.57) with SRH ∼90%, which extend over a broad composition range starting from mfd(AS) < 0.05 and ending at mfd(AS) > 0.95 [Ciobanu et al., 2009; Song et al., 2012]. For systems with lower SRH, the miscibility gap is most persistent for OIR = 1:2 and disappears when the organic fraction is high. Figure 2b presents O:C versus SRH of all mixtures studied in this work (filled circles, average of all OIR that showed LLPS) and literature (open circles) [Marcolli and Krieger, 2006; Ciobanu et al., 2009; Bertram et al., 2011; Song et al., 2012]. Mixtures which did not exhibit a miscibility gap are indicated by a star at 0% RH. SRH decreases with increasing O:C. The lowest SRH was observed at 59.0% in the MFC/AS/H2O system for O:C = 0.79. Note that observation of LLPS is limited to the RH range above efflorescence, which typically occurred between 40 and 50% RH for the systems investigated in this study. Figure 2balso presents a second-order polynomial proposed byBertram et al., [2011](black line) and a three-parameter sigmoid curve fit of all available data to parameterize SRH (red line) as a function of O:C. The sigmoid curve parameterization avoids unphysical SRH values for both low and high O:C ratios. At O:C ratios higher than 0.8, AS effloresces before LLPS is expected to occur. Therefore, in the O:C range > 0.8, the sigmoid curve represents our expectation of the occurrence range of LLPS if AS efflorescence would be suppressed (i.e., an extrapolation, dashed curve parts). In comparison to our data, the data and the polynomial proposed byBertram et al. [2011]shows a bias to low SRH. The direct comparison of SRH of the system 2,5-dihydroxybenzoic acid/AS/H2O from this study (68.5% RH, OIR = 1:2) with Bertram et al., [2011](average: 61.9% RH) shows a difference of ∼7% RH. With the same O:C ratio, the SRH of 2,6-dihydroxybenzoic acid/AS/H2O from this study presents at 91.5% RH. The parameterization of the sigmoid curve from all data is given in the auxiliary material.

3.2. Dependence of LLPS on Compositional Complexity

[8] Most previous studies on LLPS in organic/AS/H2O have been performed with a single organic substance representing the organic aerosol fraction [e.g., Ciobanu et al., 2009; Bertram et al., 2011]. Here we investigate the influence of the compositional complexity of the organic aerosol fraction on the occurrence of LLPS by comparison of mixtures with similar O:C ratios but different number of components and different spread of O:C ratios of the components.

[9] The most complex mixture investigated is COM/AS/H2O consisting of 10 organic components, with O:C of components ranging from 0.30–1.33, as shown in Figure 1b. This mixture comprises dicarboxylic acids (green), aromatic compounds (pink), pinolic and pinonic acids (purple), polyols (cyan), and levoglucosan (yellow). COM4/AS/H2O (see Table S1 in Text S1 for exact mixture composition) with O:C = 0.77 shows no LLPS while a LLPS occurs in case of the mixtures COM1, COM2, and COM3 with O:C ≤ 0.71. This range in O:C with regard to the occurrence of LLPS is in agreement with the ones of the simpler systems.

[10] We investigated three different mixtures with O:C = 0.63 and 0.67 and different degree of compositional complexity: C6/AS/H2O (O:C = 0.67) consisting of three dicarboxylic acids each with O:C = 0.67 [Song et al., 2012]; (C5+C6+C7)/AS/H2O (O:C = 0.67) consisting of nine dicarboxylic acids with a spread of O:C from 0.57–0.80; and COM1/AS/H2O (O:C = 0.63) containing 10 components with more variety of compounds and a spread of O:C from 0.30–1.25. These three mixtures exhibit SRH in a similar range, namely at 73–74% (C6/AS/H2O), 70–74% (C5+C6+C7/AS/H2O), and 72–85% (COM1/AS/H2O) for the investigated OIR (Figure 2a). This suggests that the complexity of the mixture in terms of component number and spread of O:C of the components is of minor importance regarding the occurrence of LLPS but may influence the SRH.

3.3. Summary of All Investigated Systems

[11] Figure 1csummarizes the occurrence range of LLPS from all systems investigated in this study and from measurements reported in the literature. The yellow-hatched region inFigure 1cmarks the range of O:C < 0.56, in which all mixtures showed LLPS while the gray-hatched range of O:C > 0.80 indicates the region where LLPS never occurred. In the white region with 0.56 < O:C < 0.8, the occurrence of LLPS depends on the functional group composition. If one constrains the functional group composition to typical mixtures found in tropospheric aerosols with high shares of carboxylic acid and hydroxyl groups, and low amounts of aromatic compounds (below 1%) [e.g.,Gilardoni et al., 2009; Russell et al., 2009; Takahama et al., 2011], LLPS is expected for O:C < 0.71. Considering that typical values of O:C and OIR in ambient samples range between 0.2 and 1.0 [Heald et al., 2010; Takahama et al., 2011] and between 4:1 and 1:5, respectively [Zhang et al., 2007], LLPS is expected to be a phenomenon occurring frequently in tropospheric aerosols.

4. Atmospheric Implications

[12] The presence of LLPS affects the gas-particle partitioning of semivolatile compounds and water and influences efflorescence and deliquescence of the inorganic components compared to assuming completely mixed particles [Zuend et al., 2010; Zuend and Seinfeld, 2012], with consequences for aerosol size distribution and particulate matter mass, and thus, the direct radiative effect of aerosols. Atmospheric aerosol particles with LLPS may show a core-shell or a partially engulfed morphology [e.g.,Kwamena et al., 2010; Song et al., 2012]. A core-shell morphology has more drastic consequences for the interaction of the particles with the surrounding gas phase, especially when the organic-rich phase is highly viscous [e.g.,Krieger et al., 2012]. Low diffusivities in the organic-rich coating may hinder gas-particle partitioning of semivolatile species and water, conserving thermodynamic non-equilibrium states. If the aqueous inorganic-rich phase is totally enclosed by an organic coating, heterogeneous chemical reactions such as the hydrolysis of N2O5 may be suppressed effectively [e.g., Anttila et al., 2007; Riemer et al., 2009]. An AS-rich core surrounded by an organic-rich coating was the prevalent configuration of the investigated aerosol particles in this study. A comprehensive analysis of the particle morphology and its consequences will be the subject of a further study.


[13] This work was supported by the Swiss National Foundation, project 200020-125151 and by the CCES project IMBALANCE funded by the ETH Domain.

[14] The Editor thanks an anonymous reviewer for assisting in the evaluation of this paper.